Reforma da Atenção Primária e Internações Evitáveis em Portugal 2000-2015 - Análise de Diferenças em Diferenças

Contents lists available at ScienceDirect Social Science Medicine journal homepage wwwelseviercomlocatesocscimed Effect of a national primary care reform on avoidable hospital admissions 20002015 A differenceindifference analysis Klára Dimitrovováab Julian Perelmanab Manuel SerranoAlarcónab a NOVA National School of Public Health Public Health Research Centre Universidade NOVA de Lisboa Avenida Padre Cruz 1600560 Lisboa Portugal b Comprehensive Health Research Center CHRC Campo Mártires da Pátria 130 1169056 Lisboa Portugal A R T I C L E I N F O Keywords Portugal Primary care Family health units Payforperformance Incentives Ambulatory care sensitive conditions Differenceindifference Event study A B S T R A C T In 2006 a major primary care reform was initiated in Portugal The most significant aspect of this reform was the creation of a new organizational model of primary care provision Family Health Units FHUs consisting of small voluntarily constituted multidisciplinary teams that have functional autonomy and are partly financed through capitation and payforperformance The creation of FHUs sought to increase access to care and to chronic disease management by improving the longterm relationship between health professionals and patients The objectives of this study are to evaluate the impact of the FHUs implementation on population health out comes measured by the rate of hospitalizations for ambulatory care sensitive conditions ACSC ie avoidable hospital inpatient admissions and to explore the effectiveness of the payforperformance in primary care by analysing the subset of disease specific hospitalizations for ACSC related to the financial incentives Using data from 276 Portuguese municipalities from 2000 to 2015 n 4416 and exploiting the gradual introduction of the FHUs over time we used a differenceindifferences approach contrasting the evolution of the hospitalization rate for ACSC in municipalities that implemented or not the FHUs We then explored heterogeneous effects by incentivized diabetes and hypertension and nonincentivized diseasespecific rates of hospitalizations for ACSC During the period under analysis 448 FHUs were created in 126 municipalities No significant impact of the FHUs implementation on the reduction of the hospitalization rate for ACSC was found This result also held for the incentivized hospitalizations for ACSC We only found a statistically significant effect of the FHUs im plementation in the reduction of one nonincentivized area the rate of urinary tract infection ACSC Our results question the capacity of this payment mechanism to achieve better health outcomes and invites a more careful and evidencebased action toward its wider diffusion 1 Introduction The quality of primary care is one of the key factors of an effective and efficient health system Studies show that stronger primary care services help to reduce major causes of death and disorders and are essential to ensure disease prevention early diagnosis and adequate referral to secondary care Kringos et al 2013 Macinko et al 2003 Health policies that strengthen primary care have therefore been encouraged worldwide and consequently many countries have un dergone recent primary care reforms in order to reduce the number of hospital emergency visits and avoidable inpatient admissions Basu and Phillips 2016 van Loenen et al 2014 Increasing the hours of primary care service creating or expanding the availability of urgent primary care services establishing General Practitioners GPs home visits and improving the coordination of primary care and emergency care are some of the measures adopted in many European countries Baier et al 2018 Also the introduction of payforperformance P4P in primary care has been widely implemented even though the overall evidence of its effectiveness is mixed and often conflicting Ammi and Fortier 2017 Mendelson et al 2017 Wilson 2013 The financial incentives linked to quality of care are generally de signed for improving processes of care eg blood pressure checks and intermediate outcomes eg cholesterol control in people with diabetes rather than for improving patient health outcomes since usually these are more difficult to influence by providers Vlaanderen et al 2019 Studies found that the impact of P4P on processes of care and inter mediate outcomes was often modest Campbell et al 2009 Harrison et al 2014 and can even have unintended detrimental consequences on quality of care for patients with nontargeted conditions Doran et al 2011 Studies have rarely focused on the improvements in health httpsdoiorg101016jsocscimed2020112908 Received 7 May 2019 Received in revised form 2 March 2020 Accepted 4 March 2020 Corresponding author Email address kdimitrovovaenspunlpt K Dimitrovová Social Science M edicine 252 2020 112908 Available online 10 M arch 2020 02779536 2020 Elsevier Ltd All rights reserved T outcomes and those that did have found little or no impact in this in dicator Harrison et al 2014 Houle et al 2012 Mendelson et al 2017 The largest and most evaluated P4P scheme in primary care is the Quality and Outcomes Framework implemented at a national level in the United Kingdom in 2004 which rewards GPs based on the quality in the delivery of care Campbell et al 2009 Ryan et al 2016 Several studies showed that this payment mechanism was accompanied by a rapid improvement in the indicators associated with financial in centives Gulliford et al 2007 Vaghela et al 2009 but also showed that this improvement was limited in time Campbell et al 2009 Gillam et al 2012 More recently Ryan et al 2016 using a differ enceindifference analysis and synthetic control methods found that this reform was not associated with significant changes in population health outcomes as measured by the mortality rate between 1994 and 2010 Ryan et al 2016 However in another study Harrison et al 2014 reported a decrease in the rate of emergency hospitalizations for incentivized ambulatory care sensitive conditions ACSC as compared with conditions that were not incentivized suggesting a positive effect of the P4P in the reduction of avoidable hospitalizations Harrison et al 2014 In a nutshell there is a lack of knowledge of whether P4P in primary care enables real improvements in populations health out comes while surprisingly these models are increasingly implemented worldwide In Portugal a primary care reform with characteristics and rationale close to those in the United Kingdom was implemented in 2006 and is still ongoing in 2020 11 Primary care reform in Portugal Portugal has a National Health Service Serviço Nacional de Saúde with a strong gatekeeping system Primary care was traditionally pro vided in primary care centres in which GPs worked in solo practice and were paid fixed salaries In 2006 a major primary care reform was initiated in order to improve access quality and satisfaction and to strengthen this level of care The organization of primary care centres was redefined and several models of primary care provision were created OECD 2015 The most significant aspect of the primary care reform was the creation of Family Health Units FHUs Unidades de Saúde Familiar which consisted of voluntarily constituted multidisciplinary teams of on average 20 health professionals GPs nurses and administrative technicians enjoying functional and technical autonomy and partly financed through capitation and P4P This P4P is based on a series of performance indicators mainly related to child and maternal health cancer screening vaccination and diabetes and hypertension man agement Perelman et al 2016 and can take the form of teambased institutional incentives or individual financial incentives which depend on the achievement of specific incentivized performance indicators There are two models of FHUs model A and model B All FHUs start as model A and must prove a specific level of quality and clinical and functional targets before they are allowed to apply for transition to model B Both models have teambased institutional incentives that correspond to monetary incentives but can only be used for example for the development of key infrastructure purchase of equipment or for the completion of specific professional training Ministérios das Finanças e da Administração Pública e da Saúde 2008 Additionally in FHUs model B there are individual financial incentives for all staff supplementary payments which are a variable component of the re muneration process the rest is a fixed legislated salary with a capita tion component Note that these incentives can reach up to 30 of total physician remuneration and up to 10 for nurses OECD 2015 All patients covered by FHUs are entitled to a designated GP named family physician which should allow for better access and better continuity of care due to a longerterm relationship with the patient Quality was also expected to be enhanced by the multidisciplinary nature of the practice its longer opening hours and possibility to schedule visits more easily This new provision model was expected to ultimately improve health outcomes through better prevention and followup and also to reduce the use of secondary care OECD 2015 Since 2006 there has been a progressive expansion of FHUs across Portugal which because it was based on selfselection of health pro fessionals was unrelated to any specific geographic criteria or popu lation needs assessment Consequently whether a population was covered or not by an FHU was dependent on whether a FHUs was created in their area or residence Also since 2006 the term primary care centre was discontinued and the health professionals that did not join the FHUs model automatically became part of the Personalized Healthcare Units Unidades de Cuidados de Saúde Personalizados which differed from FHUs in staff size facilities autonomy and payment mechanisms without P4P thus remaining very similar to the tradi tional primary care centres OECD 2015 Nonetheless since 2010 these Personalized Healthcare Units were also required to report vir tually the same performance indicators as FHUs model A even though they were not eligible for any teambased institutional incentives ACSS nd Despite the progressive FHUs implementation at the beginning of 2019 full coverage of the Portuguese population by a GP still did not exist with 707283 697 persons without a GP Ministério da Saúde 2019 All of these patients were assigned to the Personalized Health care Units This current way of primary care organization possibly causes many asymmetries in access and in the quality of services pro vided raising concerns about the FHUs induced disparities in the provision of primary care Some available data suggest that FHUs are performing better than the Personalized Healthcare Units in the quality of care delivered as measured by the improvement in processes of care eg blood pres sure checks and diabetes checks Perelman et al 2016 However to date no study has investigated the impact of the FHUs on patient health outcomes In this paper we analyse if the FHUs implementation ie its unique features in terms of multidisciplinarity and P4P affected patient health outcomes as measured by the rate of hospitalizations for ambulatory care sensitive conditions ACSC Specifically by exploiting the fact that FHUs were created in different municipalities and years we seek to assess their impact on the hospitalization rate for ACSC over the 20002015 year period ACSC are defined as specific conditions for which hospitalization is thought to be avoidable through patient education health promotion initiatives early diagnosis early treatment and appropriate chronic disease management ie timely and effective primary care Caminal et al 2004 Hospitalizations for ACSC are largely studied as an in direct measure of access to effective primary care and therefore we would expect that the implementation of FHUs should have a direct effect on this indicator Furthermore we aim to contribute to the in ternational literature on the effectiveness of P4P in primary care by analysing the impact of the FHUs implementation in the subset of hospitalizations for ACSC related to the incentivized indicators This approach has been used by other authors Chen et al 2010 Fiorentini et al 2011 Harrison et al 2014 Islam and Kjerstad 2018 2 Methods 21 Data We used data on all inpatient stays at all public nonspecialized Portuguese NHS hospitals for the years 20002015 from which we selected all hospitalizations for ACSC n 1278601 and data on the number and geographical location municipality of all FHUs that opened in Portugal during the 20062015 period made available by the Portuguese Central Administration of the Health System We also used aggregate socioeconomic data from the National Institute for Statistics for the 20002015 period INE nd Our total K Dimitrovová et al Social Science M edicine 252 2020 112908 2 number of observations was 4416 which corresponds to 276 of the 278 municipalities in mainland Portugal from 2000 to 2015 We excluded two municipalities due to lack of data on the hospitalizations for ACSC 072 of the sample 22 Variables We first used the overall hospitalization rate for ACSC per 1000 adult inhabitants 18 years old as our dependent variable This rate was calculated as follows first we selected from all inpatient stays the episodes classified as ACSC and valid for adult population using the set of guidelines defined by the US Agency for Healthcare Research and Quality AHRQ 2015 Then using information from the patients re sidence geographic code we calculated the total number of hospitali zations for ACSC per municipality and year and then calculated the hospitalization rate for ACSC per 1000 adult inhabitants There are 12 separate hospitalizations for ACSC in this overall rate Table 1 For the second part of the analysis we used as dependent variables diseasespecific hospitalization rates for ACSC related to the in centivized clinical areas of the FHUs As mentioned above between 2006 and 2015 in both FHUs model A and model B the teambased institutional incentives and individual financial incentives were based on a series of performance indicators mainly related to child and ma ternal health cancer screening vaccination and diabetes and hy pertension management defined at a national level and equal for all FHUs Table A1 and A2 in Appendix A ACSS nd We focused our analysis on the incentivized conditions that are ambulatory care sen sitive ie diabetes and hypertension management Thus from the 12 individual ACSC defined by AHRQ we created four groups of diseasespecific hospitalizations for ACSC two of them related to the incentivized areas of the P4P diabetesrelated ACSC and circulatoryrelated ACSC and two related to the nonincentivized areas of the P4P respiratoryrelated ACSC and urinary tract infection ACSC as illustrated in Table 1 Note that even though the performance indicators that are targeted in the P4P in primary care are related to hypertension management we added to this group the cardiovascular conditions which ultimate aim is to decrease by controlling the blood pressure such as heart failure and angina and therefore are also indirectly targeted in the P4P This grouping was also necessary because the number of hospitalizations for ACSC due only to hypertension was too low in the period under analysis to be analysed as a separate category Finally the hospitalizations for ACSC due to dehydration were not included in the analysis since the number of hospitalizations due to this condition was too low in the period under analysis To control for population characteristics we used the municipality yearlevel purchasing power and the proportion of elderly 65 years or older since studies show that hospitalizations for ACSC are mostly prevalent at older ages especially after the age of 65 years Purdy 2010 and that people from lowincome areas have a much higher risk of being hospitalized for these conditions Agabiti et al 2009 Dimitrovová et al 2017 Magán et al 2011 In Portugal the munici pality purchasing power which is a compound indicator that measures the relative purchasing power per capita based on a series of indicators such as the gross income per capita is calculated biannually so for the inbetween years we conducted linear interpolation We also controlled for the number of inhabitants in each municipality and year to account for the likelihood of an FHUs being less likely to be opened in very small municipalities Additionally we controlled for regional differ ences ie for the five Regional Health Administrations since these administrations are responsible for the management of some dimen sions of primary care and for regional health policies implementation Finally our treatment variable is a dummy indicating whether a municipality will have at least one FHU open during the period 20062015 adopting municipality vs the nonadopting munici palities in which none FHU open during the period 20062015 23 Empirical strategy 231 Overall impact of the FHUs implementation To estimate an average overall effect of the FHUs implementation on health outcomes at municipality level we used a differenceindif ferences DiD analysis by contrasting the evolution of the hospitali zation rate for ACSC in adopting and nonadopting municipalities as follows y FHUs implementation x After Year Municipality Year x RHA X mrt m mt t m t r mrt mrt 0 1 2 3 4 5 1 where m stands for municipality r for Regional Health Administrations and t for time period year ymrt is each of our outcome variables hospitalization rate for ACSC per 1000 inhabitants or diseasespe cific hospitalization rate for ACSC per 1000 inhabitants in each municipality m of each region r in year t equals one if the munici pality will have at least one FHU during the 20062015 period ie adopting municipalities Aftermrt equals one if the municipality has at least one FHU at year t Yeart are year fixed effects Municipalitym are municipality fixed effects RHAr are Regional Health Administrations so Year x RHA t r control for regionalspecific time trends Xmrt are covariates representing the characteristics of each municipality that vary over year purchasing power proportion of elderly and number of inhabitants and mrt is the random error term 1 aims to measure the overall effect of the FHUs im plementation on the average hospitalization rate for ACSC There are two main concerns in our analysis The first is that the opening of FHUs was not random over municipalities and time Note that the FHUs were voluntarily created by groups of health profes sionals so we would expect that their opening may depend on some preexisting characteristics of the municipality Actually descriptive analysis shows that FHUs have opened mainly in urban municipalities Table 1 Hospitalizations for incentivized and nonincentivized ambulatory care sensi tive conditions under the P4P scheme in primary care in Portugal 20062015 Diseasespecific ACSC Individual ACSC defined by AHRQa Incentivized ACSC Diabetesrelated ACSC diabetes shortterm complications diabetes longterm complications uncontrolled diabetes lowerextremity amputation diabetes Circulatoryrelated ACSC hypertension heart failure angina without procedure Nonincentivized ACSC Respiratoryrelated ACSC COPD or asthma in older adults asthma in younger adults bacterial pneumonia Urinary tract infection ACSC urinary tract infection dehydrationb Notes a These 12 conditions are included in the overall hospitalization rate for ACSC AHRQ 2015 b Dehydration was not analysed as a separate category K Dimitrovová et al Social Science M edicine 252 2020 112908 3 with a lower proportion of elderly and higher purchasing power Table A3 in Appendix B To deal with this concern and following the ap proach used in similar studies González and Viitanen 2009 Rocha and Soares 2010 we included demographic and socioeconomic control variables at the municipality level and used municipality fixedeffects in order to account for any other preexisting differences across muni cipalities Standard errors were clustered at the municipality level in all of our models to account for the possible serial correlation in the error terms and to avoid overestimation of the significance of estimated coefficients Bertrand et al 2004 The second major concern is the parallel trends assumption The main DiD assumption implies that the preexisting trends in the hospitaliza tion rate for ACSC in both groups of municipalities be parallel before the FHU implementation conditional on the set of municipality char acteristics that we control for Pischke 2005 In order to test for this assumption we performed an event study by including leads and lags in our model Furthermore this model allowed us to explore the dynamic effect of the FHUs implementation on the hospitalization rate for ACSC over time Autor 2003 Miraldo et al 2018 Pischke 2005 Rocha and Soares 2010 and is represented as follows y FHUs implementation for k periods Year Municipality Year x RHA X mrt k k mt t m t r mrt mrt 0 6 5 2 3 4 5 2 where FHUs implementation for k periods is a dummy variable that equals 1 if the municipality m in year t has at least one FHU implemented for k years These are dummy variables for the adopting municipalities for each preimplementation period up to 6 years and for each postim plementation period up to 5 years leaving as base category the year when the first FHUs was implemented year 0 Note that due to the progressive implementation of the FHUs not all the adopting municipalities are observed the same number of pre and post implementation years eg a municipality that implemented FHUs in 2006 will be observed on 6 preimplementation periods and 9 post implementation periods while a municipality that implemented the FHUs in 2015 will be observed 15 preimplementation periods and only one year of the postimplementation year 0 As a consequence in our event study we binned up the endpoints Schmidheiny and Siegloch 2019 ie we included up to 6 preimplementation period dummies where the last dummy contains all preimplementation per iods from 6 years backwards 6 and similarly we included 6 postimplementation period dummies where the last dummy contains all the observations from the fifth postimplementation period onwards 5 Table A4 in Appendix C If our data followed the parallel trends assumption the coefficients of the preimplementation periods leads 6 to 1 should not be significant suggesting that there were no differences in trends between adopting and nonadopting municipalities prior to the FHU implementation However as further reported in the results section our data do not follow the parallel trend assumption which means that the adopting and nonadopting municipalities already had a different trend of the hos pitalization rate for ACSC before the FHUs implementation To deal with this issue we added municipalityspecific linear time trends to Equation 2 In this way we allow for each municipality to follow its own overall linear trend and we can be assured that we are controlling for all timevarying factors at the municipality level that could bias our results Angrist and Pischke 2008 Furthermore the parallel trends assumption is complied after the inclusion of the municipalityspecific linear time trends and the model is represented as follows y FHU implementation for k periods Year Municipality Municipality trend X x mt k k mt t m m t mt mt 0 6 5 2 3 4 5 3 where trend is a linear trend ie equals 1 in 2000 2 in 2001 and so on Note that in Equation 3 regionalspecific time trends were not added because of multicollinearity 3 Results 31 Descriptive analysis Between 2006 and 2015 448 FHUs were created in 126 munici palities As a result in 2015 there was an average of 067 FHUs min 014 max 179 per 10000 inhabitants in the adopting municipalities and 152 municipalities without any FHUs Official records show that approximately half of the Portuguese population 5 million inhabitants was assigned to an FHUs in 2015 Ministério da Saúde 2019 311 Hospitalization rate for ACSC per 1000 inhabitants The average hospitalization rate for ACSC during the study period was 112 per 1000 inhabitants 99 in adopting municipalities vs 122 in nonadopting municipalities Table A3 in Appendix B Fig 1a shows the hospitalization rate for ACSC in adopting vs non adopting municipalities in the years prior and after the FHUs im plementation Note that while the nonadopting municipalities remain unchanged along the x axis ie the year 6 shows the hospitalization rate for ACSC in 2000 year 0 in 2006 and so on for the adopting municipalities due to the progressive implementation of FHUs we normalized the year of the implementation of the first FHU to year 0 eg for a municipality that implemented the first FHUs in 2010 the year 0 shows the hospitalization rate for ACSC in 2010 the year 1 in 2011 and so on likewise the year 1 shows the hospitalization rate for ACSC in 2009 the year 2 in 2008 and so on We can observe that the implementation of FHUs occurred in municipalities that already presented lower rates of hospitalizations for ACSC and that this in dicator was increasing at a lower rate previous to the FHUs im plementation in the adopting municipalities as compared to non adopting municipalities Note however that the differences between adopting and non adopting municipalities shown in Fig 1a might be affected by time effects due to the normalization to year 0 To further explore this we regressed the hospitalization rate for ACSC on year dummies for each municipalityyear observations The errors of these regressions which represent the variation in the hospitalization rate for ACSC that cannot be explained by the yearly variation that is common to all munici palities time effects are shown on Fig 1b We can observe that the differential trend between adopting and nonadopting municipalities before the FHUs implementation is even clearer after accounting for the time effects 312 Hospitalization rate for incentivized and nonincentivized ACSC The average hospitalization rate of the incentivized ACSC during the study period ie circulatoryrelated ACSC and diabetesrelated ACSC was 299 per 1000 inhabitants 258 in adopting municipalities vs 333 in nonadopting municipalities and 145 per 1000 inhabitants 127 in adopting municipalities vs 160 in nonadopting municipalities re spectively The average hospitalization rate of the nonincentivized ACSC during the study period ie respiratoryrelated ACSC and urinary tract infection ACSC was 519 per 1000 inhabitants 463 in adopting mu nicipalities vs 565 in nonadopting municipalities and 121 per 1000 inhabitants 118 in adopting municipalities and 124 in nonadopting municipalities respectively Fig A1 in Appendix D Similarly to Fig 1a we can observe that the implementation of FHUs occurred in municipalities that already presented lower rates of diseasespecific ACSC except for the urinary tract infection ACSC and that these indicators were increasing at different rates previous to the K Dimitrovová et al Social Science M edicine 252 2020 112908 4 FHUs implementation in the adopting municipalities as compared to nonadopting municipalities When accounting for the time effects the differential trend be tween adopting and nonadopting municipalities before the FHUs im plementation becomes even clearer Figs A1 and A2 in Appendix D An exception to this was the urinarytract infection ACSC where adopting and nonadopting municipalities follow more similar trends prior to the FHUs implementation 32 Differenceindifferences results 321 Hospitalization rate for ACSC per 1000 inhabitants Results from the DiD estimates Eq 1 would wrongly suggest that the FHUs implementation significantly decreased the yearly hospitali zation rate for ACSC by an average of 090 per 1000 inhabitants p 001 Table 2 Column 1 As mentioned above the main un derlying assumption of the DiD estimate is the parallel trend assumption However the results from Equation 2 show that the preim plementation periods of the event study are statistically significant suggesting the presence of preFHUs differential trends Eq 2 Fig 2a This implies that the effect of the FHUs implementation in the hospitalization rate for ACSC in Equation 1 is partly explained by the existence of nonparallel preFHUs trends The results from Equation 3 show us that the parallel trend assumption is met after the inclusion of the municipalityspecific linear time trends Eq 3 Fig 2b which then justify the inclusion of these trends in our first DiD specification In Table 2 Column 2 we can observe that after the inclusion of the municipalityspecific linear time trends to Equation 1 the DiD esti mate is no longer significant β 024 p 021 322 Hospitalization rate for incentivized and nonincentivized ACSC Results from the DiD estimates Eq 1 for the hospitalization rate of the incentivized ACSC rates of circulatoryrelated ACSC and dia betesrelated ACSC per 1000 inhabitants and nonincentivized ACSC rates of respiratoryrelated ACSC and urinary tract infection ACSC per 1000 inhabitants are also presented in Table 2 Similarly to the hospitalization rate for ACSC results for the hospitalization rate of circulatoryrelated ACSC initially suggest that the FHUs implementation had an effect in the reduction of this in dicator Table 2 Column 3 However after adjusting for the munici palityspecific linear time trends the DiD estimate is no longer sig nificant β 003 p 064 Table 2 Column 4 Regarding the hospitalization rate of diabetesrelated ACSC we did not find a statis tically significant effect of the FHUs implementation even in our first DiD estimate β 005 p 040 Table 2 Column 5 suggesting that the FHUs implementation did not have an impact in the reduction of diabetesrelated ACSC Regarding the nonincentivized ACSC on one hand the hospitali zation rate of respiratoryrelated ACSC is not significant after the ad justment for the municipalityspecific linear time trends β 013 p 026 also suggesting that the FHUs implementation did not have a significant impact in the reduction of respiratoryrelated ACSC Table 2 Column 8 On the other hand results show that the FHUs im plementation significantly decreased the yearly hospitalization rate of urinary tract infection ACSC by an average of 012 per 1000 inhabitants Table 2 Column 9 Since the preexisting trends of this indicator follow the parallel trends assumption Fig 3 adding the municipality specific linear time trends to our initial model would not be necessary Nevertheless even if we did the DiD estimate remains unchanged β 012 p 005 suggesting that the FHUs implementation de creased the hospitalization rate of urinary tract infection ACSC Table 2 Column 10 Results from the event studies Equation 2 and Equation 3 for the hospitalization rate for incentivized and nonincentivized ACSC that show the presence of preFHUs differential trends and justify the in clusion of the municipalityspecific linear time trends to the DiD esti mates for the circulatoryrelated ACSC and the respiratoryrelated ACSC are presented in Figs 3 and 4 respectively 33 Sensitivity analysis We performed several sensitivity analyses First since the creation of FHUs was not uniform across adopting municipalities the intensity of the FHUs implementation or exposure to the FHUs will depend on Fig 1 Hospitalization rate for ACSC in adopting vs nonadopting municipalities in the years prior and after the FHU implementation Notes The normalization of the implementation of the first FHUs to year 0 implies that the sample of the adopting municipalities gets smaller from year 1 onwards since the municipalities that implemented FHUs in 2015 will only be observed until year 0 those that implemented FHUs in 2014 will be observed until year 1 and so on At year 9 the sample of adopting municipalities is constituted by municipalities that implemented FHUs in 2006 See Table A4 in Appendix C for more detailed information Normalization in Fig 1a also implies that the hospitalization rate for ACSC for the adopting municipalities in the years prior to the implementation of the first FHUs consists of observations of adopting municipalities coming from later years as compared to the nonadopting municipalities As a consequence Fig 1a may be affected by time effects On Fig 1b we control for these time effects by plotting the errors of regressing the hospitalization rate of ACSC on year dummies for each municipalityyear observation K Dimitrovová et al Social Science M edicine 252 2020 112908 5 the number of FHUs created per municipality and the number of in habitants in that municipality who are the potential users In order to test if there was an effect of the intensity of the FHUs implementation as measured by the average number of FHUs functioning per 10000 inhabitants we divided the adopting municipalities into terciles of intensity of FHUs implementation Table A5 in Appendix E If there was an effect of the FHUs it should at least appear among the more exposed ie in the highest tercile of intensity of implementation Results from Equation 1 were similar to the original model and even though we can observe that the FHUs implementation was associated with a lower hospitalization rate for ACSC the estimate was not sta tistical significant even in the highest tercile as compared with the non adopting municipalities β 047 p 019 Table A5 Column 1 We repeated this analysis for all dependent variables and all results remained similar to those in the original model we did not find a statistically significant effect of the intensity of the FHUs implementa tion in the rate of circulatoryrelated ACSC diabetesrelated ACSC or respiratoryrelated ACSC We did find a negative and significant effect for the rate of urinary tract infection ACSC in the second tercile as compared to the nonadopting municipalities β 017 p 005 which supports our previous findings Table A5 Column 5 Second we converted our DiD estimate into two estimates ac cording to the type of FHUs FHUs model A and FHUs model B FHUs implementation x AfterA FHUs implementation x AfterB m mt m mt 1 2 Thus we reestimated the equations with AfterA 1 mt if the municipality had at least one FHUsA opened at year t and no FHUsB and AfterB 1 mt if the municipality had at least one FHUB opened at year t As mentioned previously it is necessary to be an FHUsA first before converting into an FHUsB FHUs model B are more demanding in terms of the goals established in the incentivized indicators but are also the ones that are entitled to individual financial incentives We obtained results consistent with our original model We found a negative and nonsignificant association for both FHUs model A and model B on the hospitalization rate for ACSC the rate of circulatoryrelated ACSC and the rate of respiratoryrelated ACSC and a positive and nonsignificant association for the rate of diabetesrelated ACSC For the rate of urinary tract infection ACSC we found a negative and statistically significant effect of the FHUs model A and FHUs model B of β 011 p 01 and β 025 p 001 respectively These results suggest that the negative effect found in this rate is greater in FHUs model B Third we know from literature that hospitalizations for ACSC are mostly prevalent at older ages especially after the age of 65 years and therefore we performed the same analysis by taking into consideration only the hospitalizations for ACSC in the elderly population The results support our previous findings first we observe that there were differ ential pretrends before the FHUs implementation and second the re sults from Equation 1 with municipalityspecific linear time trends were not significant for any of our dependent variables except for the rate of urinary tract infection ACSC β 035 p 01 Fourth we performed the same analysis by grouping the 12 hospi talizations for ACSC into two groups and reestimated the equations where the ymt is the rate of all incentivized hospitalizations for ACSC ie diabetes shortterm complications diabetes longterm complica tions uncontrolled diabetes lowerextremity amputation dia betes hypertension heart failure angina without proce dure and the rate of all nonincentivized hospitalizations for ACSC ie COPD or asthma in older adults asthma in younger adults bacterial pneumonia urinary tract infection dehydration We did not find a statistically significant effect of the FHUs implementation on the incentivized hospitalizations for ACSC β 002 p 083 nor in the nonincentivized hospitalizations for ACSC β 023 p 013 Finally we performed the same analysis for each of the 12 hospitalizations for ACSC separately as long as we had enough cases in each category ie except for hypertension dehydration asthma in Table 2 Differenceindifference results Equation 1 for the hospitalization rate for ACSC per 1000 inhabitants 20002015 Rate of ACSC per 1000 inhabitants Incentivized ACSC by P4P Nonincentivized ACSC by P4P circulatoryrelated ACSC diabetesrelated ACSC respiratoryrelated ACSC urinaryrelated ACSC 1 2 3 4 5 6 7 8 9 10 Eq 1 β SE Eq 1 trends β SE Eq 1 β SE Eq 1 trends β SE Eq 1 β SE Eq 1 trends β SE Eq 1 β SE Eq 1 trends β SE Eq 1 β SE Eq 1 trends β SE DiD FHUs implementation x After 08980 02441 03395 00318 00465 00132 04496 01322 01206 01172 02622 01932 00763 00674 00515 00494 01612 01181 00608 00579 Year fixedeffects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Municipality fixedeffects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Socioeconomic variables Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Regionalspecific time trends Yes Yes Yes Yes Yes Municipalityspecific linear time trends Yes Yes Yes Yes Yes Observations 4416 4416 4416 4416 4416 4416 4416 4416 4416 4416 Rsquared 0772 0854 0665 0744 0548 0614 0712 0798 0655 0656 Notes Socioeconomic variables include number of inhabitants proportion of elderly and purchasing power Standard errors SE clustered at municipality level in parentheses p 001 p 005 p 01 K Dimitrovová et al Social Science M edicine 252 2020 112908 6 Fig 2 Event study leads and lags coefficients for the hospitalization rate for ACSC without Equation 2 and with municipalityspecific linear time trends Equation 3 Fig 3 Event study leads and lags coefficients for the hospitalization rate for incentivized and nonincentivized ACSC without municipalityspecific linear time trends Equation 2 K Dimitrovová et al Social Science M edicine 252 2020 112908 7 younger adults diabetes shortterm complications uncontrolled diabetes and lowerextremity amputation diabetes Again we did not find a statistically significant effect of the FHUs implementation on any of the individualized hospitalizations for ACSC full results available from the authors upon request 4 Discussion 41 Key findings The FHUs in Portugal were implemented in municipalities with better health outcomes a larger and younger population and greater purchasing power We did not find a significant impact of the FHUs implementation in the reduction of the hospitalization rate for ACSC This result also held for hospitalizations for ACSC related to health conditions targeted in the P4P diabetesrelated ACSC and circulatory related ACSC 42 Interpretation The Portuguese primary care reform was an ambitious effort aimed at improving the quality of primary care primarily through the creation of small multidisciplinary FHUs with functional and technical autonomy and paid by capitation and according to performance and in which each citizen is assigned to a family physician Between 2006 and 2015448 FHUs were created in 126 municipalities However the opening of FHUs was left to the will of voluntary groups of healthcare professionals which led to the opening of FHUs in richer younger and healthier municipalities This was not a surprise since more health professionals are expected to live in urban and younger areas with better socioeconomic conditions However the voluntary and progressive nature of the FHUs implementation offered an interesting natural experiment to evaluate its impact To date no evaluation of the FHUs creation has been undertaken in terms of health outcomes and thus in this study we analysed the FHUs implementation on population health outcomes measured by the hos pitalization rate for ACSC over a long period of time 20002015 Hospitalizations for ACSC have been largely studied as an indirect measure of access to timely and effective primary care Agabiti et al 2009 Caminal et al 2004 Specifically studies show that an adequate supply of primary care physicians and a longterm relationship between the primary care physician and the patients reduces the hospitalization rate for ACSC van Loenen et al 2014 We did not find a statistically significant effect of the FHUs im plementation on the hospitalization rate for ACSC between 2000 and Fig 4 Event study leads and lags coefficients for the hospitalization rate for incentivized and nonincentivized ACSC with municipalityspecific linear time trends Equation 3 K Dimitrovová et al Social Science M edicine 252 2020 112908 8 2015 β 024 p 021 Additionally the national crude hospi talization rate for ACSC increased from 865 per 1000 inhabitants to 1306 per 1000 inhabitants between 2000 and 2015 These results suggest that other important characteristics other than the primary care supply and quality in primary care are determinant for the hos pitalization rate for ACSC We may hypothesize that population ageing Dimitrovová et al 2017 the increase in chronic conditions and the rise of multimorbidity Dantas et al 2016 are responsible for the increase in hospitalizations for ACSC We found a statistically significant effect of the FHUs implementa tion in only one nonincentivized area of the P4P the hospitalization rate of urinary tract infection ACSC Although the overall hospitaliza tion rate of urinary tract infection increased between 2000 and 2015 the slope was less pronounced in the adopting municipalities The hospitalization rate of urinary tract infection ACSC is one of the few acute conditions classified as ACSC and we may assume that patients assigned to FHUs had faster and easier access to primary care which in this case has proven to be essential to avoid hospitalization In this study we also assessed the impact of the FHUs implementa tion on the diseasespecific hospitalization rate for ACSC related to the incentivized areas in the P4P The rationale for this analysis is that if the conditions included in the P4P are ambulatory care sensitive and the scheme is effective in the desired quality improvement it should also lead to a reduction in these avoidable hospital admissions Harrison et al 2014 However we did not find a statistically significant effect of the FHUs implementation on the diseasespecific hospitalization rate of ACSC targeted in the P4P Our results differ from those of Harrison et al 2014 who found that the introduction of the Quality and Outcomes Framework in the United Kingdom was associated with a moderate and sustained de crease in incentivized hospitalizations for ACSC compared with those that were not directly incentivized by the P4P scheme Harrison et al 2014 Nevertheless as the authors conclude the improvements that occurred under the P4P scheme were greater for indicators related to measurements eg recording of blood pressure rather than for in dicators related with intermediate outcomes eg blood pressure con trol which would be more likely to have an immediate impact on the ACSC Therefore they argue that this modest improvement on inter mediate outcomes may have not been sufficient for them to attribute the observed effect on hospitalizations for ACSC solely to the Quality and Outcomes Framework scheme Harrison et al 2014 For example studies show that P4P schemes are usually implemented along with other interventions such as raising physicians awareness of clinical care standards public reporting Gupta and Ayles 2019 audit and feedback and electronic decisionsupport tools which can also have a considerable influence in quality improvement Mendelson et al 2017 Other studies also showed a positive and significant association with P4P and reduced avoidable hospitalizations Chen et al 2010 Fiorentini et al 2011 but these studies do not capture the overall trend in hospitalizations for ACSC before the introduction of the P4P In fact in a systematic review Houle et al 2012 show that results from uncontrolled studies suggest that P4P improves quality of care but that the majority of studies with more sophisticated methodologies such as randomized trials and interrupted time series failed to confirm these findings and did not find any improvements in clinical outcomes after P4P implementation Houle et al 2012 Regarding our findings on one hand we may hypothesize that the P4P scheme even though it was previously associated with improve ment in processes of care eg blood pressure checks and diabetes checks did not have an impact in the reduction of diseasespecific hospitalizations for ACSC because it was not directly designed to target the avoidable hospitalizations Still there were two performance in dicators related to intermediate outcomes and thus more likely to have had an impact on hospitalizations for ACSC such as have at least one hemoglobin A1C result record 85 performed during the analysis period and have at least one blood pressure record during the analysis period with SBP values 150 mmHg and DBP 90 mmHg Table A1 and A2 in Appendix A Furthermore some studies suggest that the link between improvements in processes of care and health outcomes is often not straightforward Vlaanderen et al 2019 On the other hand we may hypothesize that the P4P simply did not achieve the expected results There is a large body of literature in cluding several systematic reviews showing that P4P is not consistently effective in improving quality of care and that there is still an un certainty in the literature on the effect of P4P on patient outcomes Houle et al 2012 Mandavia et al 2017 Mendelson et al 2017 This uncertainty may be due in part to the differences in the P4P design and the context in which they are implemented Harrison et al 2014 Our results complement the previous literature and reinforce that P4P in primary care in Portugal has not been consistently effective in im proving patient health outcomes as measured by the rate of disease specific hospitalizations for ACSC 43 Strengths and limitations One of the main strengths of our study is that the progressive im plementation of the FHUs provides us with a natural experiment that allows us to compare treated adopting and nontreated non adopting municipalities within the same country Portugal a com parison that is not possible in the United Kingdom since the Quality and Outcomes Framework was implemented nationally Another strength of our study is the long data series 16 years which allowed us to esti mate the longterm effect on health outcomes and allowed us to esti mate trends more accurately Regarding the limitations of our study first part of the effect of the FHUs implementation might be captured by the municipalityspecific linear time trends that we added to our model since their inclusion reduces the degree of freedom and might potentially affect the sig nificance of our results Angrist and Pischke 2008 Wing et al 2018 Nevertheless results for the hospitalization rate of urinary tract infec tion ACSC show that even after adjusting for the municipalityspecific linear time trends the effect of the FHUs implementation did not change This gives us confidence that even in our most conservative specification with municipalityspecific linear time trends the effect of the FHUs implementation is still captured Second we can suspect that some changes of the primary care re form other than the creation of FHUs and the introduction of P4P may have affected the primary care practices in nonadopting municipalities ie the Personalized Healthcare Units For example during the pri mary care reform since 2010 the Personalized Healthcare Units were required to report virtually the same performance indicators even though they were not eligible for any teambased institutional in centives ACSS nd These changes in data recording may have led to an improvement in clinical practice and thus to an overall performance of these primary care practices Also some Personalized Healthcare Units may have been interested later on in obtaining the FHUs status for which a formal application to the Ministry of Health is needed and in which the health professionals must present expected future targets on a series of performance indicators This process may have also led indirectly to an overall improvement of these primary care practices Nevertheless what we aimed to capture in this study are the unique features of the FHUs the shift from the primary care provision in single handed practices into multidisciplinary teams with technical and functional autonomy and partly paid by P4P Third due data unavailability it is not possible to perform this analysis using individual data Our study uses aggregated data by mu nicipality and therefore assumes that at least one FHU within a mu nicipality will have an effect on the population of this municipality Even though we performed a sensitivity analysis by grouping the adopting municipalities into terciles of FHUs implementation intensity to overcome this issue to some extent results should be taken with K Dimitrovová et al Social Science M edicine 252 2020 112908 9 caution considering that data lacks the real individual exposure to FHUs Nonetheless we did not find a significant result of the FHUs implementation on the hospitalization rate of ACSC even among the highest exposed municipalities This result also held for conditions specifically incentivized by the P4P scheme Finally our study focused on only some of the conditions included in the P4P scheme those that are ambulatory care sensitive based on the set of guidelines for the adult population as defined by the US Agency for Healthcare Research and Quality AHRQ 2015 and that should lead to a reduction in hospital admissions and to an overall healthier status of the patient due to better disease management Other performance indicators incentivized under the P4P scheme namely related to cancer screening vaccination and child and maternal heath are not included in this analysis We therefore cannot comment on the effect of P4P on health outcomes related to those conditions 5 Conclusion No significant impact of the FHUs implementation on the reduction of the hospitalization rate for ACSC was found This result also held for conditions specifically incentivized by the P4P scheme This finding in line with the recent literature on P4P questions the capacity of this payment mechanism to achieve better health outcomes and invites a more careful and evidencebased action toward its wider diffusion CRediT authorship contribution statement Klára Dimitrovová Conceptualization Methodology Data cura tion Formal analysis Writing original draft Julian Perelman Conceptualization Writing review editing Manuel Serrano Alarcón Conceptualization Methodology Formal analysis Writing review editing Acknowledgments We thank to the Central Administration of the Health System Administração Central do Sistema de Saúde IP for providing data on all inpatient stays at all public nonspecialized Portuguese NHS hospitals for the 20002015 year period and for the data on the opening of all Family Health Units created in Portugal between 2006 and 2015 We thank the three anonymous reviewers for their insightful and constructive comments which helped us to improve the manuscript and we are grateful to Ana Costa Ana Rodríguez and Miquel Serra from Centre for Research in Health and Economics of Universitat Pompeu Fabra CRESUPF for their valuable suggestions We are also thankful to the participants of the 7th Workshop of the Associação Portuguesa de Economia da Saúde to the participants of the VI Taller EvaluAES and to the members of Nova SBE Health Economics Management Knowledge Center The first author was supported by a PhD research Grant FCT PhD Programmes PDBD1058292014 from Fundação para a Ciência e a Tecnologia The final content is of responsibility of the authors Appendix A Supplementary data Supplementary data to this article can be found online at https doiorg101016jsocscimed2020112908 References ACSS Metodologia de Contratualização para os Cuidados de Saúde Primários nd WWW Document httpwww2acssminsaudeptPublicações CuidadosdeSaúdePrimáriostabid118languageptPTDefaultaspx accessed 22720 Agabiti N Pirani M Schifano P Cesaroni G Davoli M Bisanti L Caranci N Costa G Forastiere F Marinacci C Russo A Spadea T Perucci C 2009 Income level and chronic ambulatory care sensitive conditions in adults a multicity populationbased study in Italy BMC Publ Health 9 457 httpsdoiorg101186 147124589457 AHRQ 2015 Prevention Quality Indicators PQI Log of ICD9CM and DRG Coding Updates and Revisions to PQI Documentation and Software through WWW Document httpswwwqualityindicatorsahrqgovDownloadsModulesPQI V50ChangeLogPQIv50pdf version 50 accessed 22720 Ammi M Fortier G 2017 The influence of welfare systems on payforperformance programs for general practitioners a critical review Soc Sci Med 178 157166 httpsdoiorg101016jsocscimed201702019 Angrist JD Pischke JS 2008 Mostly Harmless Econometrics an Empiricists Companion first ed Princeton University Press Oxfordshire Autor DH 2003 Outsourcing at will the contribution of unjust dismissal doctrine to the growth of employment outsourcing J Labor Econ 21 142 httpsdoiorg10 1086344122 Baier N Geissler A Bech M Bernstein D Cowling TE Jackson T van Manen J Rudkjøbing A Quentin W 2018 Emergency and Urgent Care Systems in Australia Denmark England France Germany and the Netherlands Analyzing Organization Payment and Reforms Health Policy New York httpsdoiorg101016j healthpol201811001 Basu S Phillips RS 2016 Reduced emergency department utilization after increased access to primary care PLoS Med 13 e1002114 httpsdoiorg101371journal pmed1002114 Bertrand M Duflo E Mullainathan S 2004 How much should we trust differencesin differences estimates Q J Econ 119 249275 httpsdoiorg101162 003355304772839588 Caminal J Starfield B Sánchez E Casanova C Morales M 2004 The role of pri mary care in preventing ambulatory care sensitive conditions Eur J Publ Health 14 246251 httpsdoiorg101093eurpub143246 Campbell SM Reeves D Kontopantelis E Sibbald B Roland M 2009 Effects of pay for performance on the quality of primary care in England N Engl J Med 361 368378 httpsdoiorg101056NEJMsa0807651 Chen JY Tian H Taira Juarez D Hodges KA Brand JC Chung RS Legorreta AP 2010 The effect of a PPO payforperformance program on patients with dia betes Am J Manag Care 16 e11e19 Dantas I Santana R Sarmento J Aguiar P 2016 The impact of multiple chronic diseases on hospitalizations for ambulatory care sensitive conditions BMC Health Serv Res 16 348 httpsdoiorg101186s1291301615842 Dimitrovová K Costa C Santana P Perelman J 2017 Evolution and financial cost of socioeconomic inequalities in ambulatory care sensitive conditions an ecological study for Portugal 20002014 Int J Equity Health 16 145 httpsdoiorg10 1186s1293901706427 Doran T Kontopantelis E Valderas JM Campbell S Roland M Salisbury C Reeves D 2011 Effect of financial incentives on incentivised and nonincentivised clinical activities longitudinal analysis of data from the UK Quality and Outcomes Framework BMJ 342 d3590 httpsdoiorg101136bmjd3590 Fiorentini G Iezzi E Lippi Bruni M Ugolini C 2011 Incentives in primary care and their impact on potentially avoidable hospital admissions Eur J Health Econ 12 297309 httpsdoiorg101007s101980100230x Gillam SJ Siriwardena AN Steel N 2012 Payforperformance in the United Kingdom impact of the quality and outcomes framework a systematic review Ann Fam Med 10 461468 httpsdoiorg101370afm1377 González L Viitanen TK 2009 The effect of divorce laws on divorce rates in Europe Eur Econ Rev 53 127138 httpsdoiorg101016jeuroecorev200805005 Gulliford MC Ashworth M Robotham D Mohiddin A 2007 Achievement of me tabolic targets for diabetes by English primary care practices under a new system of incentives Diabet Med 24 505511 httpsdoiorg101111j146454912007 02090x Gupta N Ayles HM 2019 Effects of payforperformance for primary care physicians on diabetes outcomes in singlepayer health systems a systematic review Eur J Health Econ httpsdoiorg101007s10198019010974 Harrison MJ Dusheiko M Sutton M Gravelle H Doran T Roland M 2014 Effect of a national primary care pay for performance scheme on emergency hospital ad missions for ambulatory care sensitive conditions controlled longitudinal study BMJ 349 httpsdoiorg101136bmjg6423 Houle SKD McAlister FA Jackevicius CA Chuck AW Tsuyuki RT 2012 Does performancebased remuneration for individual health care practitioners affect pa tient care A systematic review Ann Intern Med 157 889 httpsdoiorg10 7326000348191571220121218000009 INE Instituto Nacional de Estatística Statistics Portugal Dados estatísticos nd WWW Document httpsineptxportalxmainxpgidinemainxpidINE accessed 12419 Islam MK Kjerstad E 2018 Coordination of health care the case of hospital emer gency admissions Eur J Health Econ httpsdoiorg101007s002330149580 5 0 Kringos DS Boerma W Van Der Zee J Groenewegen P 2013 Europes strong primary care systems are linked to better population health but also to higher health spending Health Aff 32 686694 httpsdoiorg101377hlthaff20121242 Macinko J Starfield B Shi L 2003 The contribution of primary care systems to health outcomes within Organization for Economic Cooperation and Development OECD countries 19701998 Health Serv Res 38 831865 httpsdoiorg1011111475 677300149 Magán P Alberquilla Á Otero Á Ribera JM 2011 Hospitalizations for ambulatory care sensitive conditions and quality of primary care Med Care 49 1723 https doiorg101097MLR0b013e3181ef9d13 Mandavia R Mehta N Schilder A Mossialos E 2017 Effectiveness of UK provider financial incentives on quality of care a systematic review Br J Gen Pract 67 K Dimitrovová et al Social Science M edicine 252 2020 112908 10 e800e815 httpsdoiorg103399bjgp17X693149 Mendelson A Kondo K Damberg C Low A Motuapuaka M Freeman M ONeil M Relevo R Kansagara D 2017 The effects of payforperformance programs on health health care use and processes of care a systematic review Ann Intern Med 166 341353 httpsdoiorg107326M161881 Ministério da Saúde 2019 BICSP Bilhete de Identidade dos Cuidados de Saúde Primários WWW Document httpsbicspminsaudeptptbiufsPaginasdefault aspx accessed 21519 Ministérios das Finanças e da Administração Pública e da Saúde 2008 Diário da República 1a série No 77 18 de Abril de 2008 Portaria no 3012008 de 18 de Abril Miraldo M Propper C Williams RI 2018 The impact of publicly subsidised health insurance on access behavioural risk factors and disease management Soc Sci Med 217 135151 httpsdoiorg101016jsocscimed201809028 OECD 2015 OECD Reviews of Health Care Quality Portugal 2015 Raising Standards OECD Publishing Paris Perelman J Lourenço A Russo G Miraldo M 2016 Pagamento pelo desempenho nos Cuidados de Saúde Primários experiências cruzadas Edições Almedina 978972 4067810 Pischke JS 2005 Empirical Methods in Applied Economics Lecture Notes London School of Economics WWW Document httpeconlseacukstaffspischke ec524evaluation3pdf accessed 92618 Purdy S 2010 Avoiding Hospital Admissions what Does the Research Evidence Say The Kings Fund Rocha R Soares RR 2010 Evaluating the impact of communitybased health interventions evidence from Brazils Family Health Program Health Econ 19 126158 httpsdoiorg101002hec1607 Ryan AM Krinsky S Kontopantelis E Doran T 2016 Longterm evidence for the effect of payforperformance in primary care on mortality in the UK a population study Lancet 388 268274 httpsdoiorg101016S0140673616002762 Schmidheiny K Siegloch S 2019 On Event Study Designs and DistributedLag Models Equivalence Generalization and Practical Implications IZA Institute of Labour Economics Discussion Paper Series IZA DP No 12079 CEPR Discussion Paper No DP13477 Vaghela P Ashworth M Schofield P Gulliford MC 2009 Population intermediate outcomes of diabetes under payforperformance incentives in England from 2004 to 2008 Diabetes Care 32 427429 httpsdoiorg102337dc081999 van Loenen T van den Berg MJ Westert GP Faber MJ 2014 Organizational as pects of primary care related to avoidable hospitalization a systematic review Fam Pract 31 502516 httpsdoiorg101093fampracmu053 Vlaanderen FP Tanke MA Bloem BR Faber MJ Eijkenaar F Schut FT Jeurissen PPT 2019 Design and effects of outcomebased payment models in healthcare a systematic review Eur J Health Econ 20 217232 httpsdoiorg 101007s1019801809898 Wilson KJ 2013 Payforperformance in health care what can we learn from inter national experience Qual Manag Health Care 22 215 httpsdoiorg101097 QMH0b013e31827dea50 Wing C Simon K BelloGomez RA 2018 Designing difference in difference studies best practices for public health policy Research Annu Rev Publ Health httpsdoi org101146annurevpublhealth K Dimitrovová et al Social Science M edicine 252 2020 112908 11 18102022 APS 4 Políticas Sociais PSM e DiffinDiff Microeconomia IV 20222 Professores Adriano Dutra Teixeira André Luiz Pereira Mancha Cristine Campos de Xavier Pinto Monitores Frederico Marco Pereira Gomes Pedro Picchetti Antes de começar a atividade por favor leia as instruções abaixo 1 A tarefa é em Grupo e deve ser entregue de 28102022 a 31102022 por um único membro do Grupo com prazo máximo às 23h59 de 31102022 2 A entrega da atividade será pelo Blackboard e deve conter 1 pdf com as respostas 1 código comentado dofile ou script e 1 arquivo com os outputs logfile ou R Markdown com todas as saídas deste código Estes arquivos não podem ser entregues de forma compactada 3 Todas as entregas serão sujeitas ao filtro do SafeAssign no Blackboard para avaliar plágio 4 A ausência de algum desses arquivos sua entrega de forma incompleta a entrega de arquivos com respostas incompatíveis entre si ou a entrega com evidência de plágio pode gerar a anulação total da nota da atividade 18102022 1 Com base nas referências do Bloco de Políticas Sociais A 15 ponto Em linha com o modelo de matrícula escolar de Ferreira Filmer Schady 2017 visto em aula a teoria de Pais Silva e Teixeira 2017 assume que a decisão sobre se as crianças participam da força de trabalho atividades de lazer ou vão à escola é determinada por uma pessoa adulta da família Estruture formalmente com equações e suas palavras o modelo microeconômico de Pais Silva e Teixeira 2017 deixando claro quais as premissas o problema de maximização dos pais e qual o principal resultado teórico do modelo Finalize sua exposição elencando a hipótese econômica do artigo Dica Em sua resolução sugerimos que tenha em mente qual a pergunta de investigação deste artigo Não é necessário trazer o desenvolvimento integral do modelo A ideia aqui também não é fazer um resumo do modelo e sim estruturálo para ter um início meio e fim O desfecho será a hipótese econômica B 10 ponto O artigo de Dimitrovová Perelman SerranoAlarcón 2020 avalia o impacto da implantação das Family Health Units Unidades de Saúde da Família nos municípios de Portugal nos desfechos de saúde da população em especial nas internações hospitalares classificadas como evitáveis Explique com as suas palavras qual a estratégia de identificação empregada na análise empírica do artigo e quais os exercícios de robustez foram realizados Explique também o racional que baseou a exposição dos resultados na Tabela 2 e Figuras 1 2 3 e 4 Em seguida interprete ao menos 1 resultado aquele que seu Grupo julgar mais importante para a análise de cada Tabela e Figura Finalize identificando qual é o principal resultado deste artigo Justifique Dica antes de ler o artigo sugerimos a leitura do Bônus 2 da Aula teórica de Políticas Sociais A reforma de Atenção básica que ocorreu em Portugal tem características semelhantes à Estratégia Saúde da Família do Brasil e pode ajudar no entendimento do artigo C 05 ponto Com base na leitura do artigo de Gitter Manley Barham 2013 e nos exemplos de Teoria da Mudança que vimos em aula elabore um diagrama teórico que estruture a Teoria da Mudança do Red de Protección Social RPS considerado um Conditional Cash Transfer da Nicarágua explicando os mecanismos específicos de atuação do RPS de acordo com a análise realizada neste artigo 18102022 2 O Programa Bolsa Família PBF é um caso brasileiro de Conditional Cash Transfer CCT O programa possui estudos sobre seus efeitos na redução de pobreza acumulação de capital humano situação de saúde e oferta de trabalho Em 2021 o Bolsa Família contava com 146 milhões de famílias beneficiárias cujos benefícios totalizavam mais de R 15 bilhões O PBF teve uma mudança importante em seu critério de elegibilidade em 2007 de modo a cobrir adolescentes com idade de 16 e 17 anos a partir do Benefício Variável Jovem BVJ Este componente de benefício variável do Bolsa Família fornece transferências de renda e impõe frequência escolar sobre as famílias elegíveis com adolescentes de 16 e 17 anos Mais detalhes sobre esta mudança podem ser consultados no artigo de Chitolina Foguel MenezesFilho 2016 Nesta investigação sua Equipe foi contratada com o objetivo de determinar o efeito da expansão do Programa Bolsa Família oriunda do Benefício Variável Jovem do PBF na probabilidade de adolescentes de 16 e 17 anos estarem frequentando a escola e não estarem trabalhando A Pesquisa Nacional por Amostra de Domicílios PNAD contém microdados dos indivíduos e suas características de escolaridade bem como atributos sociais demográficos e econômicos Embora a pesquisa não contemple a informação sobre se a família é beneficiária do Bolsa Família a partir das regras de elegibilidade do programa podemos inferir as famílias que são potencialmente beneficiárias Colete os microdados das PNADs 2006 e 2009 de modo a obter informação dos indivíduos antes e depois da mudança no critério de elegibilidade De posse dos microdados defina os seguintes grupos Grupo de tratamento famílias que estão entre as 20 mais pobres e que possuem adolescentes de 16 ou 17 anos em sua composição Grupo de controle famílias que estão entre as 20 mais pobres e que possuem adolescentes de 15 anos em sua composição A análise será realizada em etapas Dica i recomendamos a leitura do artigo de Chitolina Foguel MenezesFilho 2016 para entendimento do Benefício Variável Jovem do Programa Bolsa Família Dica ii considere o plano amostral da PNAD em suas estimações Dica iii os microdados da PNAD podem ser baixados neste site do IBGE Recomendamos que consulte sempre o arquivo de dicionário da PNAD Aqui um vídeotutorial para abrir os microdados no Stata e aqui um vídeotutorial para abrir os microdados no R Dica iv o Datazoom é outra alternativa simples de abrir os microdados das PNADs no Stata Aqui tem um vídeotutorial que pode ajudar também Dica v ao fazer a classificação de grupos de tratamento e controle o ideal é que esta definição seja feita de acordo com a identificação das famílias Vamos pensar nos casos em que um mesmo domicílio pode ter várias famílias e isso é especialmente comum em domicílios brasileiros Nesse caso podemos ter várias famílias beneficiárias pelo Bolsa 18102022 Família em um domicílio Para identificar as famílias precisamos gerar uma chave que identifica unicamente as famílias da PNAD assim como fizemos na Aplicação 2 com a PNS Aqui vocês podem fazer isso usando as variáveis de i ano da PNAD year ii número de controle v0102 iii número de série v0103 iv número da família v0403 Dica vi Na execução de suas estimações sugerimos consultar o material da Aplicação deste Bloco Lá tem exemplos que podem ajudar os Grupos que estão trabalhando em Stata e R Etapa I Teoria Econômica A Estabeleça a Pergunta de pesquisa a ser investigada por sua Equipe com base nas informações disponíveis Formalize uma Teoria Microeconômica que fundamente os argumentos teóricos de como o Benefício Variável Jovem do PBF afetará sua variável de resultado Deixe claro quais as referências da literatura que serviram de base para a construção de seu argumento teórico Identifique a Hipótese econômica resultante de sua Teoria Micro Dica aqui existem várias possibilidades de desenvolver seu argumento microeconômico Uma possibilidade é adaptar um dos modelos vistos em aula outra possibilidade é buscar um outro modelo microeconômico no Google Scholar por exemplo para usar como referência teórica Esperamos que este modelo esteja bem alinhado à pergunta de pesquisa e que tenha como desfecho a Hipótese Econômica a ser testada nos itens seguintes 25 pontos Etapa II Análise descritiva prémudança B Usando a PNAD 2006 elabore uma tabela de estatísticas descritivas para caracterizar os grupos de tratamento e controle previamente à mudança Interprete seus resultados 07 ponto Etapa III Estimação do PSM C Explique por que o Propensity Score Matching é uma estratégia adequada para parear os grupos de tratamento e controle neste contexto Usando a PNAD 2009 identifique qual a variável de resultado qual a variável de tratamento e quais as variáveis explicativas observáveis que serão usadas na estimação Estime o PSM e o efeito de tratamento usando as técnicas de pareamento vistas em aula Interprete os seus resultados 10 ponto D Sempre que realizamos um PSM é importante verificar o balanceamento das variáveis explicativas do seu modelo e verificar se há evidências em favor da hipótese de suporte comum Verifique e interprete seus resultados 08 ponto 18102022 Etapa IV Estimação do DiferençasemDiferenças E No diffindiff vamos comparar mudanças ao longo do tempo na probabilidade de adolescentes estarem frequentando a escola e não trabalhando nos domicílios de tratamento e controle Faça uma tabela que mostre a probabilidade média de adolescentes estarem frequentando a escola e não trabalhando em 2006 pré e 2009 pósmudança nos grupos de controle e tratamento Obtenha primeiro a estimativa de diffindiff via diferenças de médias e discuta quais as hipóteses necessárias para a validade do método neste contexto Interprete seus resultados Obtenha agora a estimativa de diffindiff via regressão múltipla Qual resultado é preferível Discuta seus resultados 10 ponto F Colete agora os microdados da PNAD de 2003 Para investigar a robustez dos resultados estime o mesmo modelo de matrícula escolar usando apenas amostras prévias à mudança do Programa Bolsa Família Em outras palavras para este item utilize os anos de 2003 e 2006 que são períodos anteriores à criação do BVJ Este é um Teste de Placebo no qual 2006 será definido como o ano póstratamento e 2003 o ano de prétratamento Assim redefinimos a variável dummy de pósprograma na equação do diffindiff tornandoa agora igual a 0 quando o ano é 2003 e igual a 1 quando o ano é 2006 Estime e interprete seus resultados Bônus de 10 ponto A organização eficiência e qualidade dos comentários do código valem 10 ponto O código a ser enviado junto das respostas é nossa maneira de compreender e checar todos os passos que vocês efetuaram na análise empírica inclusive os passos relacionados à limpeza criação e ajustes das variáveis Por isso é imprescindível o envio do código e suas saídas outputs nos arquivos da entrega de seu Grupo Esperamos que estes arquivos estejam organizados bem comentados e funcionando perfeitamente Caso os resultados expostos nas respostas não sejam condizentes com o código ou este não esteja em estado averiguável e replicável a nota completa do exercício pode ficar comprometida 18102022 Referências Chitolina L Foguel M N MenezesFilho N A 2016 The impact of the expansion of the Bolsa Família Program on the time allocation of youths and their parents Revista Brasileira de Economia 70 183202 Gitter S R Manley J Barham B L 2013 Earlychildhood nutrition and educational conditional cash transfer programmes The Journal of Development Studies 4910 1397 1411 Dimitrovová K Perelman J SerranoAlarcón M 2020 Effect of a national primary care reform on avoidable hospital admissions A differenceindifference analysis Social Science Medicine 252 112908 Pais P S M de Figueiredo Silva F Teixeira E C 2017 The influence of Bolsa Familia conditional cash transfer program on child labor in Brazil International Journal of Social Economics Research on Economic Inequality Own and Sibling Effects of Conditional Cash Transfer Programs Theory and Evidence from Cambodia1 Francisco H G Ferreira Deon Filmer Norbert Schady Article information To cite this document Francisco H G Ferreira Deon Filmer Norbert Schady Own and Sibling Effects of Conditional Cash Transfer Programs Theory and Evidence from Cambodia1 In Research on Economic Inequality Published online 20 Nov 2017 259298 Permanent link to this document httpsdoiorg101108S1049258520170000025008 Downloaded on 10 January 2018 At 0406 PT References this document contains references to 0 other documents To copy this document permissionsemeraldinsightcom The fulltext of this document has been downloaded 19 times since 2017 Access to this document was granted through an Emerald subscription provided by emeraldsrm226864 For Authors If you would like to write for this or any other Emerald publication then please use our Emerald for Authors service information about how to choose which publication to write for and submission guidelines are available for all Please visit wwwemeraldinsightcomauthors for more information About Emerald wwwemeraldinsightcom Emerald is a global publisher linking research and practice to the benefit of society The company manages a portfolio of more than 290 journals and over 2350 books and book series volumes as well as providing an extensive range of online products and additional customer resources and services Emerald is both COUNTER 4 and TRANSFER compliant The organization is a partner of the Committee on Publication Ethics COPE and also works with Portico and the LOCKSS initiative for digital archive preservation Related content and download information correct at time of download Downloaded by James Cook University At 0406 10 January 2018 PT 259 Research on Economic Inequality Poverty Inequality and Welfare Volume 25 259298 Copyright 2018 by Emerald Publishing Limited All rights of reproduction in any form reserved ISSN 10492585doi101108S1049258520170000025008 OWN AND SIBLING EFFECTS OF CONDITIONAL CASH TRANSFER PROGRAMS THEORY AND EVIDENCE FROM CAMBODIA1 Francisco H G Ferreira Deon Filmer and Norbert Schady ABSTRACT Conditional cash transfers CCT have been adopted in many countries over the last two decades Although the impacts of these programs have been studied extensively understanding of the economic mechanisms through which cash and conditions affect household decisions remains incomplete In particular relatively little is known about the effects of these programs on intrahousehold allocation decisions This chapter uses evidence from a program in Cambodia where eligibility varied substantially among siblings in the same household to illustrate these effects A simple model of school ing decisions highlights three different effects of a childspecific CCT an income effect a substitution effect and a displacement effect The model predicts that such a CCT should unambiguously increase enrollment for eligible children but have an ambiguous effect on ineligible siblings Downloaded by James Cook University At 0406 10 January 2018 PT 260 FRANCISCO H G FERREIRA ET AL The ambiguity arises from the interaction of a positive income effect with a negative displacement effect These predictions are shown to be consist ent with evidence from Cambodia where the CESSP Scholarship Program CSP makes modest transfers conditional on school enrollment for chil dren of middleschool age Scholarship recipients were more than 20 per centage points more likely to be enrolled in school and 10 percentage points less likely to work for pay However the school enrollment and work of ineligible siblings was largely unaffected by the program A possible fourth effect operating through nonpecuniary spillovers of the intervention among siblings remains largely outside the scope of the analysis although there is some tentative evidence to suggest that it might also be at work Keywords Conditional cash transfers sibling effects Cambodia JEL classification I24 I32 1 INTRODUCTION Many programs in the developing world make cash transfers to poor house holds which are conditional on the school enrollment of schoolaged children The impacts of these conditional cash transfers on various individual and household outcomes including school enrollment learning achievements health and nutrition outcomes household consumption and savings have been studied extensively in a number of settings Yet there is relatively little evidence in the literature on the effect these programs have on intrahousehold allocation decisions and in particular on the educational and work trajecto ries of eligible and ineligible siblings Such intrahousehold spillovers are not merely an academic curiosity the effect of transfers on household members other than those to whom they are targeted may matter for the design of various program reforms currently under consideration Because program impacts on school attendance are frequently concentrated on children in grades where in the absence of the program school dropout is large see Schultz 2004 on the PROGRESA pro gram in Mexico and Schady and Araujo 2008 on the Bono de Desarrollo Humano program in Ecuador there have been calls for cash transfers to be directed exclusively toward children in these transition grades on the grounds that this would be a more costeffective way of increasing school attainment Attanasio Meghir Santiago Shephard 2009 de Janvry Sadoulet 2006 Such childspecific conditional transfers could potentially have impor tant implications for the school enrollment of ineligible siblings but these Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 261 effects are hard to sign ex ante Depending on the magnitude of the trans fer its income effect might lead to increased enrollment for all children in the household whether or not the transfer is conditioned on any action on their part Various models of schooling and child labor predict that greater family income during childhood increases school enrollment and reduces the time children spend working quite independently of any conditionality BenPorath 1967 Basu Van 1998 Baland Robinson 2000 On the other hand cash transfer programs conditional on the school enrollment of one specific child might lead parents to reallocate child work away from the recipient and to other children in the household Evidence of this effect has been found in some settings including Colombia BarreraOsorio Bertrand Linden PerezCalle 2011 More generally the transfer may provide an incentive for parents to specialize in the education of the recipient leading to a displacement of less schooling for his or her siblings Finally there is also evidence in the literature that the educational deci sions and experiences of older siblings may affect the educational outcomes of younger siblings directly Such nonpecuniary spillovers operate for exam ple through information sharing and emulation behavior Oettinger 2000 Qureshi 2011 Nicoletti Rabe 2014 To the extent that the cash transfer changes enrollment or attendance decisions for the targeted recipient these spillovers might potentially lead to knockon effects on sibling enrollment separately from any pecuniary effect from the transfer In this chapter we assess the impact of a childspecific conditional cash transfer program in Cambodia on the school enrollment and work of recipi ents and their ineligible siblings A very simple model of child occupational choice is used to provide a basic framework for the empirical analysis The main prediction of the model is that childspecific conditional cash trans fer programs will unambiguously increase school enrollment among eligible children but will have an ambiguous effect on ineligible siblings The effect on eligible children reflects mutual reinforcement between a positive income effect which affects the entire household and a positive but childspecific substitution effect which is brought about by the reduction in the opportu nity cost of schooling for the eligible child The displacement effect is also positive for these children as it refers to situations in which they displace their siblings from school The ambiguity of the effect on ineligible siblings arises from the opposing positive income and negative displacement effects of the transfer on these children2 This simple theoretical framework proves helpful for understanding our main results which come from an analysis of Cambodias CESSP Scholarship Program CSP This program makes very modest transfers equivalent to Downloaded by James Cook University At 0406 10 January 2018 PT 262 FRANCISCO H G FERREIRA ET AL between 2 and 3 percent of the total expenditures of the average recipient household conditional on school enrollment for children of middleschool age The results show that children who received scholarships were about 20 percentage points more likely to be enrolled in school and 10 percentage points less likely to work for pay However the school enrollment and work of ineligible siblings was largely unaffected by the program These results are robust to a variety of specification checks These findings have important implications for the way in which we think about household decisions regarding school enrollment and child labor and for the design of cash transfer programs We highlight three First the very large effect of the CSP program on the behavior of recipients con firms that scholarship and conditional cash transfer programs may be an effective way of increasing school enrollment in lowincome countries see Filmer Schady 2008 2014 on Cambodia Chaudhury Parajuli 2010 on Pakistan Baird McIntosh Özler 2011 on Malawi This is important because most of the evidence on these programs refers to Latin America where income levels are generally higher and institutions are stronger than in many African and Asian settings where such programs are now being implemented3 Second and as a cautionary counterpoint our model suggests that very narrow age ranges for benefit eligibility could potentially lead to the unin tended displacement of some poor children from school Although there was no net impact from the program on the enrollment of ineligible siblings in Cambodia this likely reflects offsetting effects with ineligible siblings in some households being enrolled thanks to a positive income effect while some ineligible siblings in other households were displaced In general it is clearly possible that either effect dominates Third the model cautions against interpreting a comparison of program effects for recipients and their siblings as definitive evidence about the relative size of the income and substitution effects of the transfer The much larger net impact on eligible children is certainly consistent with an important role for the substitution effect and this would confirm other findings in the litera ture4 But it cannot be interpreted as identifying these effects since there is a third effect namely the displacement effect which contributes negatively to sibling enrollment and positively to own enrollment In a context where ineligible children in transferreceiving households are relatively numerous more work is needed on quantifying the displacement effect The rest of the chapter proceeds as follows The next section describes our simple schooling decision model and introduces a childspecific CCT Section 3 briefly discusses the CSP and the data we use for the evaluation Section 4 Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 263 discusses our empirical specification The main results are presented in Section 5 Section 6 concludes the study 2 THE MODEL This section presents a simple model of schooling decisions which adapts the early insights of BenPorath 1967 and of a large subsequent litera ture to the specific context of a multiplechildren household facing a child specific CCT intervention The model is intended solely to provide a simple framework for interpreting our empirical results It abstracts from a number of important considerations that a more general model might consider such as endogenous parental labor supply child heterogeneity or nonpecuniary educational spillovers among siblings The model has two periods and is partial equilibrium in nature There is a continuum of households each consisting of a parent and two children Parents live only for the first period but care about the perfect foresight expectation of their childrens wellbeing in Period 2 They take all decisions in Period 1 so as to maximize household welfare which is a function of the familys consumption level in Period 1 and of the expected utility of both children W c U U p 1 2 Following Becker 1991 and Baland and Robinson 2000 we assume that this function is additively separable as follows5 β W c U U U c U c U c p p 1 2 1 2 1 The subscripts p 1 and 2 denote the parents and each child respectively U is an individual utility function that is common across all individuals in the household which satisfies the usual properties U 0 U 0 U0 0 U c lim c 0 and U c lim 0 c Adults earn income A from labor and capital which is determined exogenously to the model A is distributed according to a cumulative distribution function FA with positive mass eve rywhere on a nondegenerate support 0A This exogenous adult income is the only source of exante household heterogeneity in the model Parents use all of their endowment of time in period 1 to supply labor inelastically They choose between two occupations for their children in Period 1 children can either work in which case they are paid a wage w or they can go to school6 The model assumes that there are no school fees but this is merely an innocuous simplifying assumption If each child had to pay a fixed fee f to attend school then the total cost of schooling would be Downloaded by James Cook University At 0406 10 January 2018 PT 264 FRANCISCO H G FERREIRA ET AL w f instead of just the opportunity cost w and adult disposable income when both children are enrolled would be B A2f The remainder of the model would be unchanged The choice of occupation for child i is denoted by σi which takes the value 1 if child i is sent to school and 0 if she is sent to work7 There is a posi tive return to schooling so that their Period 2 income is higher if they attend school in Period 1 h than if they do not θ There are no capital markets and parents cannot leave financial bequests to their children so that the only way to invest in their future is through education8 The households problem is then to maximize Eq 1 by choice of σ σ 1 2 subject to σ σ c A w 2 p 1 2 2 σ θ σ c h i 1 12 i i i 3 σ 01 i 12 i 4 where w 0 and h θ 0 The discrete nature of the control variable σ implies that the optimal decision for each household cannot be obtained from calculus Instead the utility levels arising from each possible decision must be compared against one another Since instantaneous utility is concave in income these levels will depend on the exogenous level of adult income A For example households will choose to enroll both their children if β β θ U A U h U A w U h U 2 and 5 β β θ U A U h U A w U 2 2 2 They will choose to enroll one child but not both if β θ β U A w U h U U A 2 U h and 6 β θ β θ U A w U h U U A w U 2 2 In fact it can be shown that Proposition 1 There exist positive income levels A and A A A such that Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 265 i for A A σ σ 0 1 2 ii for A A A σ σ 0 1 i j i j iii for A A σ σ 1 1 2 Proof See Appendix A Proposition 1 states that when households are identical in all dimensions other than adult income and school enrollment has an opportunity cost given by the forgone earnings of children in period 1 then enrollment deci sions vary monotonically with family income Above a certain adult income level A all children are enrolled in school and none work Below a lower threshold A no children are enrolled and all work Between the two thresh olds households enroll one child but not the other In that income range and under our simplifying assumption that siblings are identical in schooling ability in labor productivity and in how much their parents value them the decision of which child to enroll from each household is random with child i being sent to school with probability πi π π 1 1 2 9 Fig 1 which illustrates the proof of Proposition 1 shows that A marks the income level at which the discounted gain in expected child utility from enrollment β θ U h U equates the opportunity cost in firstperiod con sumption from forgoing the earnings of a first child U A w U A w 2 Point A marks the corresponding income level for the second child and it is clear that the existence of the intermediate range depends on utility being strictly concave in firstperiod consumption There are clear parallels to the previous literature This strong negative relationship between household income and child labor is reminiscent of Basu and Vans 1998 result that child labor arises only from households in poverty although our model has no need for a strong luxury axiom As in Baland and Robinson 2000 the result is driven by missing capital markets if families could borrow in Period 1 against the childs income in Period 2 then for sufficiently large returns to education ie for a sufficiently large value of hθ child labor could be eradicated Without that ability to borrow poor households for whom the marginal value of period 1 consump tion is very high use child labor as an inferior alternative consumption smoothing mechanism In this setting a conditional cash transfer is a monetary payment t in period 1 which is made if and only if a child is enrolled in school Since we Downloaded by James Cook University At 0406 10 January 2018 PT 266 FRANCISCO H G FERREIRA ET AL are interested in a situation where some children are eligible for the transfer but others are not even within the same household assume without loss of generality that only child 1 is eligible for the transfer This policy leaves the households problem unchanged except for Eq 2 which is replaced by σ σ τσ c A w 2 p 1 2 1 2 The effect of this policy on the households decisions is twofold First it changes the adult income thresholds at which first one and then both chil dren are enrolled Parents will now enroll both children if τ β τ β θ U A U h U A w U h U 2 7 The income level which satisfies Eq 7 as an equality is t A It is easy to see once again from the concavity of the utility function that At A Parents will enroll one child but not both if Eq 7 does not hold and τ β θ β θ U A w U h U U A w U 2 2 8 t A which solves Eq 8 as an equality is less than A The second change in household behavior is that under the maintained assumption that siblings are identical the choice of which child to enroll for those parents who enroll a single child is now no longer random They all choose to enroll child 1 who is eligible for the transfer which leads to the potential displacement effect If any ineligible siblings living in house holds with incomes in the interval A At were in school prior to the introduction of the CCT π 2 0 they are now displaced by the policy A A A U U A 2w U A w U A w U A β U h U θ B O Fig 1 Enrollment Decisions and Adult Income in the Basic Model Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 267 Fig 2 illustrates the changes brought about by the conditional cash trans fer The twochild enrollment threshold falls from A to At because of a pure income effect the value of the transfer is added to both sides of inequality 7 and the dashed curve t t U A w U A lies below the original curve U A w U A simply because that difference declines with income Ineligible siblings living in households with incomes in the interval At A see their enrollment rise unambiguously as a result of the CCT10 The reduction in the onechild enrollment threshold from A to At on the other hand arises from both an income and a substitution effect The value of the transfer is added only to the lefthandside of inequality 8 so that the dashed curve t U A w U A w 2 lies below the original curve U A w U A w 2 because of a full price effect comprising both an income and substitution effect In these households in the interval At A enrollment rises for eligible children child 1 and there is no effect on their ineligible siblings child 2 This allows us to state Proposition 2 The introduction of a childspecific conditional cash transfer that alters the household budget constraint from Eq 2 to Eq 2 will i unambiguously increase enrollment of the eligible children child 1 and ii have an ambiguous effect on the enrollment of the ineligible children child 2 Aτ Aτ A A U U A 2w U A w τ U A w τ U A τ β U h U θ Fig 2 Introducing a Conditional Cash Transfer Downloaded by James Cook University At 0406 10 January 2018 PT 268 FRANCISCO H G FERREIRA ET AL Proof Denote by πi the proportion of children of type i i 12 enrolled by families with a single child attending school prior to the introduction of the transfer eg if children of types 1 and 2 were enrolled with equal prob ability by families with income in the interval A A then π π 05 1 2 Then pretransfer enrollment for child i is given by π E F A F A F A i 1 1 2 i i 9 Posttransfer enrollment is different for the two types of children and given by t E F A 1 1 10 t E F A 1 2 10 Changes in enrollment are obtained from subtracting Eq 9 from Eq 10 π π τ E F A F A F A 1 0 1 1 1 11 where the inequality arises from the fact that t F A F A and t F A F A This proves i π π τ E F A F A F A 1 2 2 2 12 Since t A may be greater than A and indeed will be greater for t w the sign of ΔE2 may depend on the specific value of π 2 01 This proves ii In sum our simple model of schooling decisions for a multichild house hold predicts that a childspecific conditional cash transfer will lead to increased enrollment for eligible children This increase reflects a combination of three effects First there is a displacement effect among those households that only enroll one child they tend to replace their ineligible children with their eligible siblings in school Second some households that would not send any children to school in the absence of the program are now compelled to send an eligible child to school due both to a substitution effect the oppor tunity cost of that decision falls from w to w t and to an income effect an increase in unearned Period 1 income reduces the utility loss from forgoing that opportunity cost The effect on the enrollment of ineligible children is ambiguous The dis placement effect works against them with families that send a single child to Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 269 school shifting away from them toward their eligible children Furthermore they do not benefit from a substitution effect since the opportunity cost of their going to school remains equal to w However some ineligible children can benefit from an income effect Those are children in households whose income levels in the absence of the transfer were just insufficient to enroll both children but who given the extra income from the transfer are now willing to forgo the child earnings from their second child as well These are households with exogenous incomes between t A and A in Fig 2 This is a pure income effect since they receive no additional transfer for this added enrollment11 The model also allows an assessment of the likely relative size of the impacts on recipients and their siblings Subtracting Eq 12 from Eq 11 gives a comparison of the changes in enrollment for eligible and noneligible children π π τ τ E E F A F A F A F A if and only if 1 2 1 2 A sufficient but not necessary condition for the enrollment impact of the transfer to be greater for recipients than for their ineligible children therefore is that preprogram enrollment rates for eligible children be no higher than those of their ineligible siblings π 1 π 2 In other words if eligible children were ini tially either equally or less likely to be enrolled than ineligible children then their enrollment would increase by more as a result of the transfer The nec essary condition is evidently much weaker if the measure of the population between A and A is not very different from that between t A and t A then one would need a pretransfer situation in which almost all eligible children were already enrolled π1 1 while almost all ineligible children were not π2 0 for the enrollment effect to be greater for ineligible siblings It is hard to conceive that a CCT would be targeted to type1 children if this were the case This model suggests in essence that the effects of a childspecific CCT on the enrollment of recipients and ineligible children will be different Increases in enrollment are likely to be larger for eligible than for ineligible children Indeed the effect on ineligible children is impossible to sign exante given offsetting income and displacement effects With these predictions in mind we now turn to the empirical analysis of a childspecific CCT in Cambodia 3 PROGRAM AND DATA Cambodia has had programs that offer scholarships to poor children mak ing the transition from primary to lower secondary school for a number of Downloaded by James Cook University At 0406 10 January 2018 PT 270 FRANCISCO H G FERREIRA ET AL years These programs have operated in some regions of the country and not others and have been funded from a variety of sources including government budgets loans from multilateral and bilateral donor agencies and NGOs One of the programs that predated the CSP known as the Japan Fund for Poverty Reduction JFPR scholarship program was targeted at girls and children from ethnic minorities making the transition from primary school to lower secondary school Filmer and Schady 2008 evaluate the program and conclude that despite the small amount of the transfer which like the CSP program we study in this chapter accounted for only 23 percent of the total consumption of the median recipient household the JFPR increased enrollment rates by almost 30 percentage points Program effects were par ticularly large among girls in the poorest households In the time period we study the CSP operated in 100 of the approximately 800 middle schools in Cambodia These schools were selected on the basis of administrative data which indicated that poverty rates in the areas served by these schools were high and by implication secondary school enrollment rates low In addition there was a requirement that none of the selected schools participate in other scholarship programs including the JFPR The selection of CSP recipients within eligible schools was done in three stages First using administrative data from the 100 CSP schools program officials identified all of the primary feeder schools for every CSP school A primary school was designated a feeder school if it had sent graduating students to a given CSP school in recent years Second within feeder schools all sixth graders were asked to complete a CSP application form regardless of whether these students or their par ents had previously expressed an interest in attending secondary school The application form consisted of 26 questions about characteristics that were highly correlated with the probability of school dropout as indicated by analysis of a recent nationwide household survey the questions were also reasonably easy for students of this age to answer and for peers and teachers to validate In practice the form elicited information on household size and composition parental education the characteristics of the home the mate rial of roof and floors availability of a toilet running water and electricity and ownership of a number of household durables Forms were filled out in school on a single day Students and parents were not told beforehand of the content of the forms nor were they ever told the scoring formula both deci sions designed to minimize the possibility of strategic responses for example by a student seeking to maximize her chances of receiving the award Once completed forms were collected by headteachers and sent to the capital Phnom Penh There a firm contracted for this purpose scored them using Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 271 the responses and the set of weights that reflected how well each character istic predicted the likelihood of school dropout in the nationwide household survey The formula used was the same for every school and once calculated the scores could not be revised12 Finally within every CSP school all applicants were ranked by the score regardless of which feeder school they came from In large CSP schools with total enrollment above 200 50 students with the lowest values of the score were then offered a scholarship for seventh eighth and ninth grades in small CSP schools with total enrollment below 200 students 30 stu dents with the lowest values of the score were offered the scholarship13 In total just over 3800 scholarships were offered in the year of the program we study14 The list of students offered scholarships was then posted in each CSP school as well as in the corresponding feeder schools Once children had been selected to receive a CSP scholarship their families received the cash award three times a year Payments were made at widely attended school ceremo nies with the school principal publicly handing over the cash to parents We analyze the impact of the program among the first cohort of eligible children These children filled out the application forms in May 2005 and the list of scholarship recipients was posted in November 2005 We use data on children at two points in time First we have access to the composite dropout score as well as the individual characteristics that make up the score for all 26537 scholarship applicants Second we fielded a household survey of 3453 randomly selected applicants and their families in five provinces Battambang Kampong Thom Kratie Prey Veng and Takeo15 The household survey was collected between October and December of 2006 approximately 18 months after children filled out the application forms Since application forms were filled out at the end of the sixth grade estimates of program effects on the school enrollment of applicants based on the household survey refer to the beginning of the eighth grade16 Table 1 summarizes the characteristics of CSP recipients and nonrecip ients as reported on their application forms separately for all applicants lefthand panel and applicants within 10 ranks of the cutoff of the score righthand panel The first four columns of each panel show that as expected recipients are generally poorer than nonrecipients For example in the full sample CSP recipients are less likely to own a bicycle 55 percent of recipients own one vs 75 percent of nonrecipients less likely to own a radio 26 vs 39 percent and less likely to live in a dwelling whose roof is made of solid materials such as tiles cement concrete or iron 45 vs 66 percent The differences between recipients and nonrecipients are smaller when we limit the sample to children whose value of the score is closer to the cutoff Downloaded by James Cook University At 0406 10 January 2018 PT Table 1 Characteristics of CSP Recipients and Nonrecipient Applicants Overall Within 10 ranks of cutoff Non recipients Recipients Diff Pvalue Dummy RD Coef Dummy RD Pvalue Non recipients Recipients Diff Pvalue Dummy RD Coef Dummy RD Pvalue Male 0392 0234 0157 0000 0057 0025 0389 0305 0084 0003 0062 0156 Live with mother 0861 0774 0086 0000 0023 0248 0851 0844 0007 0752 0013 0648 Mother attended school 0501 0363 0138 0000 0050 0073 0481 0477 0004 0898 0057 0201 Live with father 0683 0526 0157 0000 0002 0944 0627 0624 0003 0935 0039 0398 Father attended school 0578 0413 0165 0000 0015 0554 0569 0494 0075 0014 0049 0258 Parent is civil servant 0051 0023 0029 0000 0007 0487 0046 0026 0020 0181 0015 0411 Number of other children in household 1302 1334 0033 0445 0097 0146 1378 1398 0020 0790 0040 0749 Number of adults in household 2963 2726 0236 0000 0116 0125 2878 2902 0024 0811 0107 0373 Disabled household member 0164 0197 0033 0030 0025 0291 0193 0157 0036 0140 0052 0112 Own bicycle 0754 0545 0209 0000 0002 0922 0720 0672 0049 0071 0003 0938 Own oxhorses cart 0371 0255 0116 0000 0018 0474 0354 0307 0047 0107 0063 0073 Own motorbike 0114 0034 0079 0000 0005 0677 0080 0063 0018 0240 0007 0757 Own car or truck 0018 0001 0016 0001 0003 0500 0016 0003 0013 0059 0000 0964 Own radio 0392 0255 0137 0000 0048 0081 0373 0305 0068 0018 0024 0621 Own TV 0386 0144 0242 0000 0007 0789 0346 0247 0099 0001 0068 0110 Roof made of solid materials 0656 0445 0211 0000 0034 0205 0629 0575 0054 0089 0048 0282 Downloaded by James Cook University At 0406 10 January 2018 PT Floors polished wood tiles 0014 0010 0003 0370 0007 0200 0005 0016 0011 0040 0015 0118 Floors wood planks or bamboo 0943 0900 0043 0000 0046 0001 0952 0924 0028 0060 0013 0603 Drinking water piped into house 0003 0001 0002 0136 0002 0220 0000 0000 0000 0000 Drinking water well pump 0793 0766 0028 0085 0033 0074 0780 0753 0027 0309 0019 0554 Drinking water vendor purchased 0015 0014 0001 0800 0007 0250 0014 0023 0010 0189 0012 0221 Toilet flush 0031 0005 0027 0010 0008 0098 0027 0009 0019 0202 0006 0618 Toilet pit latrine 0077 0051 0026 0004 0014 0203 0055 0071 0016 0248 0033 0108 Lighting electricity from a generator 0012 0002 0010 0197 0002 0547 0005 0001 0003 0515 0002 0712 Lighting electricity from a battery 0518 0317 0201 0000 0005 0850 0483 0419 0064 0041 0032 0461 Cooking fuel electricity gas kerosene 0003 0000 0003 0102 0000 0810 0000 0001 0001 0314 0004 0318 Note Information based on application forms for sample of applicants covered by household survey significant at the 1 percent level at the 5 percent level Downloaded by James Cook University At 0406 10 January 2018 PT 274 FRANCISCO H G FERREIRA ET AL The final two columns in each panel of Table 1 report the coefficient and pvalue in a regression of each characteristic from the application form on a quartic in the composite score school fixed effects and dummies for the age of the child and her birth order This corresponds to our basic estima tion specification discussed in more detail below and is a standard check on the validity of the regression discontinuity RD specification Imbens Lemieux 2008 This specification check suggests that differences between recipients and nonrecipients are unlikely to be an important source of bias to our estimates of program impact In the full sample the coefficients on the dummy variable for scholarship recipients are significant for only two charac teristics the probability that a child is a boy and the fraction of households with floors made of wood planks or bamboo We estimate the impact of the CSP program separately for boys and girls throughout which removes any possible bias associated with differences in the gender composition of recipi ents and nonrecipients The difference in the proportion of households with floors made up of wood planks or bamboo is not significant when the sample is limited to children whose score places them within 10 points of the cutoff As a robustness check on our estimates of CSP program effects we therefore also present results for this smaller sample An additional specification check recommended by Imbens and Lemieux 2008 is to ascertain whether there is bunching in the density of scores at the cutoff Bunching might arise for example from manipulation of the scoring through strategic answers by students to the questions in the applica tion forms Such behavior would invalidate the assumption of continuity in the distribution of observed or unobserved characteristics at the cutoff which is essential for RD identification Fig 2 in Filmer and Schady 2014 a companion paper that uses the same dropout risk score data confirms that there is no unusual density mass to either side of the distribution of normal ized dropout risk score around the cutoff In order to place our results in context Table 2 summarizes enrollment and work outcomes for children in the control group separately for appli cants and their siblings and by gender We consider six different outcome variables The first three are the probability of enrollment working for pay and working without pay These are binary indicator variables for exam ple enrollment takes on the value of one if a child is enrolled in school and zero otherwise The remaining variables correspond to the number of hours an applicant or their sibling attended school worked for pay and worked without pay conditional on the relevant binary variable taking on a value of one For example hours in school refer only to children who are Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 275 enrolled in school The measures of hours of school attendance and work refer to the last seven days17 The table shows that enrollment of boys is higher than that of girls in this sample of applicants who were not offered a scholarship 63 percent of boys and 54 percent of girls are enrolled Among their siblings who are on aver age younger overall school enrollment is higher 86 percent for boys and 80 percent for girls Applicants who are enrolled in school attend on average for about 26 hours per week and their siblings attend for approximately 21 hours About 31 percent of applicant boys in the control group worked for pay compared to 37 percent of girls among those who work for pay average hours are 24 for boys and 28 for girls Siblings are much less likely to work for pay 9 percent of boys and 17 percent of girls Work for pay among children in this age group is concentrated in the farm sector and construction for boys and in the farm sector and garment industry for girls18 Work without pay is much more widespread among applicants and their siblings 64 percent of applicant boys and 51 percent of applicant girls work without pay On average these children work for about 19 hours Among sib lings the incidence of work for pay is once again lower 52 percent among boys and 47 percent among girls Table 2 Summary Statistics on Enrollment and Work of Nonrecipient Applicants and Their Siblings Males 718 Females 718 Parents Non recipient applicants Siblings of non recipient applicants Non recipient applicants Siblings of non recipient applicants Fathers of non recipient applicants Mothers of non recipient applicants Enrolled 0628 0860 0544 0804 0484 0347 0498 0397 Enrolled hours 2596 2134 2612 2150 936 719 927 670 Work for pay 0309 0093 0366 0167 0285 0218 0463 0290 0482 0373 0452 0413 Work for pay hours 2403 2196 2776 2461 3019 2403 2304 2386 2652 2490 2307 2304 Work for no pay 0637 0517 0513 0466 0778 0742 0481 0500 0500 0499 0416 0437 Work for no pay hours 1913 1818 1956 1683 3562 3203 1453 1319 1476 1277 1890 1847 Note Standard deviations in parentheses Downloaded by James Cook University At 0406 10 January 2018 PT 276 FRANCISCO H G FERREIRA ET AL The last two columns of the table focus on patterns of work among appli cants parents Many more adults work in the nopay sector than in the for pay sector a pattern that is apparent for both men and women Work hours are approximately 2430 hours per week in the forpay sector and 3236 hours in the nopay sector 4 IDENTIFICATION STRATEGY The basic identification strategy we use in this chapter is based on RD The regressions we estimate take the following form α β δ δ δ δ ε Y f S R R R S R S XX Male Female Male Female ihs s h i ihs 1 2 3 4 13 where Yihs is an outcome variable for example the probability that child i in household h who applied for a scholarship in CSP school s is enrolled in school αs is a set of CSP school fixed effects fSh is the control function a flexible parametrization of the dropout score In our main results we use a quartic in the score and allow the function to differ for males and females we also test for the robustness of the results to this choice of functional form Xi includes a set of single year age dummies a set of birth order dummies a dummy for males a dummy for siblings and the interaction between siblings and males The variables RMale RFemale take on the value of one if the observation is a male female applicant who was offered a scholarship the variables RSMale RSFemale take on the value of one for male female siblings of applicants who were offered a scholarship and εihs is the regression error term All regressions are limited to schoolaged children ages 71819 Standard errors account for clustering at the level of the primary feeder school In this setup the parameters δ1 and δ2 are estimates of the program impact on male and female recipients respectively while the parameters δ3 and δ4 are estimates of the program impact on the male and female siblings of recipi ents respectively Note that because we include the main effects for boys and siblings in the vector Xi as well as the interaction terms between them we are comparing treated applicants to control applicants and not to their siblings and treated siblings to control siblings and not to applicants A number of things are worth noting about this specification First because the score perfectly predicts whether or not an applicant is offered a scholar ship this is a case of sharp as opposed to fuzzy RD Second because we Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 277 focus on the impact of being offered a scholarship rather than that of actually taking up a scholarship these are IntenttoTreat ITT estimates of program impact20 Third because we include school fixed effects in all specifications the identification of the scholarship effect is entirely driven by withinschool differences between applicants who were offered scholarships and those who were not and withinschool differences between their siblings Finally as with every approach based on RD the estimated effect is local Specifically it is an estimate of the impact of the scholarship program around the cutoff However the cutoff falls in terms of the dropout score varies from school to school This is because the number of students offered the scholarship was the same in every large and small CSP school respectively but both the number of sixth graders and the distribution of the underlying characteristics that make up the dropout score varied21 In practice the value of the cutoff varies from a score of 21 to 40 in the schools attended by the study sample with the median at 28 The estimates of δ are therefore weighted averages of the impacts for these different cutoff values For the three indicator variables the probability of enrollment of work ing for pay and working without pay the models are estimated by OLS For the other variables the hours spent in each of these activities in the past 7 days we present results both from OLS regressions and the marginal effects from Tobit specifications the latter take account of the fact that the variables are censored with a substantial fraction of the sample reporting zero22 5 RESULTS 51 Main Results Before turning to the estimates of Eq 13 we motivate our results by showing outcomes as a function of the ranking based on the dropout score relative to the cutoff We do this by plotting average outcomes at each value of the rela tive ranking and overlaying a quartic in the score2324 Fig 3 has six panels corresponding to enrollment work for pay and work without pay for appli cants and their siblings In each case distinct jumps at the cutoff would suggest that the program affected behavior For applicants Panel A suggests that the program had large effects on enrollment approximately 20 percentage points Panel B suggests that the probability of work for pay dropped and Panel C suggests that the program led to a small increase in the likelihood that children engaged in unpaid work For siblings Panels A and B suggest little relationship between scholarships Downloaded by James Cook University At 0406 10 January 2018 PT 278 FRANCISCO H G FERREIRA ET AL and enrollment or work for pay Panel C suggests a small increase in work without pay Fig 3 is thus consistent with the CSP program having had a large effect on the schooling of children who were offered scholarships but little or no effect on their siblings The results of parametric estimates of program impact described in Eq 13 are reported in Table 3 For each outcome we show estimates of program impact on male and female recipients as well as on male and female siblings We also test for differences in the recipient effects by gender δ1 δ2 in the sibling effects by gender δ3 δ4 and whether the genderspecific recipi ent and sibling effects are the same δ1 δ3 and δ2 δ4 The first two rows of the table confirm that the program had large effects on recipients School enrollment increased dramatically by 22 percentage points for boys and 20 percentage points for girls This increase came hand in hand A Enrollment B Work for pay C Work for no pay Applicants 0 2 4 6 8 1 Probability 0 2 4 6 8 1 Probability 0 2 4 6 8 1 Probability 0 2 4 6 8 1 Probability 0 2 4 6 8 1 Probability 0 2 4 6 8 1 Probability 25 15 5 5 15 25 Relative ranking Siblings Mean Quartic Mean Quartic Mean Quartic Mean Quartic Mean Quartic Mean Quartic 25 15 5 5 15 25 25 15 5 5 15 25 25 15 5 5 15 25 25 15 5 5 15 25 Relative ranking 25 15 5 5 15 25 Relative ranking Relative ranking Relative ranking Relative ranking Fig 3 Program Effects on Applicants and their Siblings Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 279 with a sharp reduction in the probability that CSP recipients work for pay of 12 percentage points in the case of boys and 9 percentage points in the case of girls Finally recipients were also more likely to work without pay a result that is significant for girls Paid work may be more difficult to combine with school ing than unpaid work because paid work generally involves less flexible hours and a greater intensity of work as suggested by Edmonds 2007 Edmonds and Schady 2012 The results in Table 3 are consistent with this pattern Before discussing the effects on siblings we consider how the CSP pro gram affected the hours spent on each of these activities These results are presented in Table 4 The lefthand panel of the table presents the marginal effects from Tobit regressions and the righthand panel presents correspond ing results estimated by OLS The coefficients on hours of schooling sug gest that recipients spent 68 more hours in school than nonrecipients The reduction in hours worked for pay is smaller between 1 and 3 hours Note that the estimated effects on hours worked without pay are negative ranging from a reduction in 30 minutes to a reduction of almost an hour and three quarters So while the program effect on the incidence of work without pay is positive Table 3 recipients worked fewer hours Table 425 Table 3 Program Effects on Recipients and Siblings by Gender School Enrollment Work for Pay Work Without Pay Own effectmale 0215 0120 0043 0031 0032 0037 Own effectfemale 0200 0088 0074 0026 0025 0029 Sibling effectmale 0011 0007 0046 0019 0020 0030 Sibling effectfemale 0000 0034 0028 0019 0021 0028 R2 031 021 011 Pvalue OwnM OwnF 070 042 048 Pvalue SibM SibF 060 028 063 Pvalue OwnM SibM 000 000 095 Pvalue OwnF SibF 000 002 010 Note Sample size is 8182 in all regressions Sample includes all children ages 7 to 18 All specifications include a set of school dummies a set of single year age dummies a set of birth order dummies a dummy for the gender of the child dummy variables for siblinggender and a quartic in the score Standard errors adjust for clustering at the applicant primaryschool level significant at the 1 percent level at the 5 percent level Pvalues are from an Ftest of equality of parameter estimates Downloaded by James Cook University At 0406 10 January 2018 PT 280 FRANCISCO H G FERREIRA ET AL Table 4 Program Effects on Recipients and Siblings by Gender Tobit Marginal Effects OLS Hours of schooling Hours worked for pay Hours worked without pay Hours of schooling Hours worked for pay Hours worked without pay Own effectmale 7667 1213 0480 6130 3225 1552 1193 0370 0845 0945 1091 1052 Own effectfemale 8275 1162 0339 6626 3297 1690 0993 0325 0693 0783 1035 0787 Sibling effectmale 1104 0010 0674 1022 0848 0276 0686 0603 0790 0595 0671 0825 Sibling effectfemale 0158 0546 0174 0209 0825 0125 0683 0424 0702 0596 0712 0731 R2 021 012 010 Pvalue OwnMOwnF 061 079 089 066 096 091 Pvalue SibMSibF 024 039 061 024 098 071 Pvalue OwnMSibM 000 001 016 000 002 006 Pvalue OwnFSibF 000 008 045 000 001 004 Notes Sample size is 8182 in all regressions Sample includes all children ages 718 All specifications include a set of school dummies a set of single year age dummies a set of birth order dummies a dummy for the gender of the child dummy variables for siblinggender and a quartic in the score Standard errors adjust for clustering at the applicant primaryschool level Significant at the 1 percent level at the 5 percent level Pvalues are from an Ftest of equality of parameter estimates Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 281 We next turn to a discussion of CSP program effects on the siblings of applicants focusing on both changes in participation in Table 3 and hours Table 4 Table 3 suggests that siblings of CSP recipients increased the likeli hood of work without pay by between 3 and 5 percentage points although these effects are not statistically significant Table 4 shows that the implied change in hours is very small and also not significant Both tables also make it clear that the school enrollment choices of siblings were unaffected by the program The values of the Ftests reported in the table show that we can eas ily reject the null of equal applicant and sibling coefficients on schooling and hours worked for pay In sum Tables 3 and 4 make clear that consistent with the predictions of our simple model the effect of the CSP program on the schooling and work for pay of applicants is significantly larger than its effect on their siblings The null effect on sibling enrollment suggests that on aver age any positive income effect from the transfer on ineligible brothers and sisters was roughly outweighed by the negative displacement effects 52 Robustness Checks We conducted a large number of robustness checks to our main results These include specifications that limit the sample to children with a score that places them within 10 ranks of the schoolspecific cutoff specifica tions that allow for schoolspecific control functions in addition to the schoolspecific intercepts and specifications in which the control func tion is defined in terms of an applicants ranking relative to the schoolspe cific cutoff as in Fig 3 rather than in terms of the score To investigate whether effects are heterogeneous across different household types we also estimate specifications that separately consider program effects on older and younger siblings and on richer and poorer households None of these changes has a qualitatively important effect on our basic results and we review each of them in turn 521 Sample Restricted to Households Within 10 Ranks of Cutoff A standard check on the RD specification involves testing whether the esti mated coefficients are robust to limiting the sample to observations that are close to the cutoff We do this by restricting the sample to children in households with a score that places them no further than 10 ranks from the schoolspecific cutoff This comes at a cost our sample is reduced by almost twothirds from 8182 observations to 2920 Downloaded by James Cook University At 0406 10 January 2018 PT 282 FRANCISCO H G FERREIRA ET AL The results from this robustness check are reported in the lefthand panel of Table 5 In terms of work the coefficients in this smaller sample tend to be somewhat larger for boys For example among applicants we estimate a reduction in work for pay of 18 rather than 12 percentage points Among girls the only notable change is that the coefficient on work for pay for appli cants is reduced substantially from 9 to 5 percentage points and is no longer significant In terms of schooling the results for the smaller sample are extremely close to those estimated for the full sample of children Because the results for the smaller sample are very similar to those that use the full sam ple of children we conclude that our main set of results is not driven by pos sible biases introduced by using observations that are far from the cutoff 522 SchoolSpecific Control Function Although our basic specification allows for schoolspecific intercepts it imposes a common control function across schools This assumes that a given change in household socioeconomic status as measured by the com posite score is associated with an increase in the probability of enrollment or work of the same magnitude across all schools Conceivably such an assumption of equal control functions may not do justice to the data For example there may be differences in school quality which affect not only whether school enrollment is higher in some schools than in others at all levels of socioeconomic status a difference in intercepts across schools but also the gradients between socioeconomic status and enrollment a differ ence in slopes across schools The righthand side of Table 5 reports results from specifications that allow for schoolspecific quartic trends and intercepts This places large demands on the data for each school there are two intercepts for boys and girls and eight polynomials in the score quartics for boys and girls for a total of 570 terms The righthand panel of Table 5 shows however that the results from this more flexible formulation are very close to those that impose a com mon control function For example in this specification the CSP program effect on the probability that applicant boys are enrolled in school implies an increase of 19 percentage points compared to 22 points in the specifica tion that imposes a common control function while that for girls implies an increase of 21 percentage points compared to 20 points In terms of work for pay the coefficients in Table 5 imply a reduction of 12 percentage points for boys and 7 points for girls compared to 12 percentage for boys and 9 points for girls in Table 3 Sibling effects remain small and insignificant It does not appear that the assumption of a common control function across Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 283 Table 5 Program Effects on Recipients and Siblings by Gender Alternative Estimation Approaches Restricted to Within 10 Ranks of Cutoff Control Function Is SchoolSpecific Function Enrolled Worked for pay Worked without pay Enrolled Worked for pay Worked without pay Own effectmale 0215 0175 0098 0188 0120 0036 0056 0052 0055 0040 0040 0047 Own effectfemale 0192 0046 0074 0209 0070 0123 0039 0039 0047 0034 0035 0040 Sibling effectmale 0001 0014 0078 0017 0006 0045 0029 0029 0044 0028 0029 0041 Sibling effectfemale 0018 0049 0052 0000 0010 0070 0028 0031 0046 0028 0031 0040 R2 031 023 014 035 027 019 Pvalue OwnMOwnF 071 002 071 068 033 015 Pvalue SibMSibF 054 030 060 064 093 066 Pvalue OwnMSibM 000 000 073 000 000 079 Pvalue OwnFSibF 000 092 065 000 001 007 Note Sample size is 2920 for the sample of children in households within 10 ranks of cutoff and 8182 for the full sample All specifications include a set of school dummies a set of single year age dummies a set of birth order dummies a dummy for the gender of the child dummy variables for siblinggender and a quartic in the control function Standard errors adjust for clustering at the applicant primaryschool level Significant at the 1 percent level at the 5 percent level Pvalues are from an Ftest of equality of parameter estimates Downloaded by James Cook University At 0406 10 January 2018 PT 284 FRANCISCO H G FERREIRA ET AL schools introduces substantial biases into our estimates of the impact of the CSP program 523 Defining Control Function in Terms of Ranking Rather than the Score In our main set of results the control function is defined in terms of the score on the application form rather than in terms of an applicants ranking rela tive to the schoolspecific cutoff An alternative is to define the control func tion in terms of an applicants ranking relative to the schoolspecific cutoff Table 6 reports the results from specifications that are based on an applicants rank with a common control function lefthand panel and schoolspecific control functions righthand panel In both cases the estimated program effects on applicants are very similar to those from our basic specification As before the effects on siblings are quantitatively small and in all but one instance work without pay for males are not significant at conventional lev els We conclude that our main results are robust to this alternative approach to defining the control function 524 Sibling Effects Differentiated by Relative Age Firstborn or earlierborn siblings have typically been found to be less likely to attend school26 We investigate the extent to which our results could mask heterogeneity by the relative age of siblings In order to isolate the issue of relative age we reestimate our basic model but now allow the impacts to dif fer by whether a sibling is younger or older than the applicant We do not differentiate by gender to keep sample sizes reasonable however results that disaggregate by the gender of both the applicant and her sibling are similar to those we report but substantially less precise Table 7 shows that our results are not an artifact of aggregation sibling effects are not significantly or sub stantively different depending on the relative age of the sibling In all cases the program effect on siblings remains statistically insignificant We also explored whether restricting the analysis to siblings who are close in age to the applicant alters our findings For this purpose we estimated our basic model but restricted the sample to children both applicants and siblings ages 1418 Results from these estimates are very similar to those we report in the study that is strong own effects on schooling and work for pay but small and insignificant impacts on siblings Finally we restricted the sample to male applicants only and analyzed the effect on brothers and to female applicants only and analyzed the effects on sisters on the grounds that samesex siblings might be closer substitutes for each other As with Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 285 Table 6 Program Effects on Recipients and Siblings by Gender and Defining Control Function as WithinSchool Ranking Control Function is Function of WithinSchool Ranking Control Function is SchoolSpecific Function of WithinSchool Ranking Enrolled Worked for pay Worked without pay Enrolled Worked for pay Worked without pay Own effectmale 0187 0142 0114 0181 0120 0075 0040 0045 0053 0045 0047 0054 Own effectfemale 0175 0047 0112 0188 0058 0135 0037 0036 0047 0039 0039 0048 Sibling effectmale 0018 0030 0120 0021 0007 0093 0032 0036 0049 0035 0037 0050 Sibling effectfemale 0027 0008 0064 0020 0003 0083 0032 0032 0046 0033 0035 0048 R2 031 021 012 035 027 019 Pvalue OwnMOwnF 083 008 098 090 029 040 Pvalue SibMSibF 084 037 038 098 083 089 Pvalue OwnMSibM 000 000 085 000 000 063 Pvalue OwnFSibF 000 002 009 000 001 008 Note Sample size is 8182 in all regressions Sample includes all children ages 718 All specifications include a set of school dummies a set of single year age dummies a set of birth order dummies a dummy for the gender of the child dummy variables for siblinggender and a quartic in the control function Standard errors adjust for clustering at the applicant primaryschool level Significant at the 1 percent level at the 5 percent level Pvalues are from an Ftest of equality of parameter estimates Downloaded by James Cook University At 0406 10 January 2018 PT 286 FRANCISCO H G FERREIRA ET AL the age restriction we do not find that restricting the sample in these ways changes our findings27 525 Richer and Poorer Households The simple model in Section 2 predicts that school enrollment should rise with household income A with the poorest households enrolling no chil dren the richest households enrolling both siblings and intermediate house holds enrolling one child but not the other It further predicts that the income effect of the transfer should lead to the enrollment of siblings among the betteroff intermediate households namely those between At and A Displacement effects offsetting this positive income effect on average should occur primarily lower down the income distribution Because our RD strategy identifies local effects around the dropout score eligibility cutoff we cannot investigate effect heterogeneity with respect to socioeconomic status merely by looking at how the effect varies along the observed distribution of household income or wealth However exploit ing the fact that the cutoff score itself varies across schools we can assess whether our results vary across the distribution of schoolspecific cutoffs Table 8 reports results for the dichotomous school enrollment decision corresponding to Column 1 in Table 3 for samples split at different Table 7 Program Effects on Recipients and Siblings by Age Relative to Applicants Age School Enrollment Work for Pay Work Without Pay Own effect 0198 0092 0059 0020 0020 0024 Sibling effectyounger 0016 0028 0029 0016 0016 0025 Sibling effectolder 0025 0018 0069 0032 0038 0037 R2 032 021 011 Pvalue SibYSibO 021 021 030 Notes Sample size is 8182 in all regressions All specifications include a set of school dummies a set of single year age dummies a set of birth order dummies a dummy for the gender of the child dummy variables for siblinggender and a quartic in the score Standard errors adjust for clustering at the applicant primaryschool level Significant at the 1 percent level significant at the 5 percent level Pvalues are from an Ftest of equality of parameter estimates Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 287 percentiles 10th 25th median 75th and 90th of the distribution of school specific cutoffs Effects on transfer recipients remain strong and significant in both subsamples for all thresholds except for boys attending schools with cutoffs above the 90th percentile As the model predicts effects on eligible applicants are stronger for poorer households or those attending schools with lower cutoffs reflecting lower preprogram enrollment rates The effects on siblings remain statistically insignificant across all speci fications up to the 75th percentile of the distribution of school cutoffs in line with our earlier results For siblings of recipients attending schools above the 90th percentile of the cutoff distribution however there is some evidence of a negative effect on sibling school enrollment significant at 5 percent for boys and 10 percent for girls This runs counter to our model predictions and suggests that our simplifying assumptions that children are identical in ability and parental preference for example or that there are no nonpecuniary externalities at work within the household may be too strong28 A richer model that incorporates child heterogeneity and nonpecuniary spillovers may be needed to account for the result in the last columns of Table 8 526 School Visits A final possible concern is the possibility of systematic reporting biases in our measure of school enrollment based on household survey data Conceivably parents of scholarship recipients could be more likely to lie to enumerators about school enrollment than those of nonrecipients although it is less clear why they would lie about the school enrollment and work status of ineli gible siblings As Filmer and Schady 2011 report however results from an analysis of data on directly observed school attendance from four unan nounced school visits are very similar to those that use the household survey The impact of being offered a scholarship on physically verified attendance is equal to 25 percentage points when pooling across all four visits February March 2006 AprilMay 2006 June 2006 June 2007 and equal to 20 per centage points when restricting the analysis to June 2007 when the applicants would have been in eighth grade if they did not repeat school grades 6 DISCUSSION AND CONCLUSION Cash transfer programs both conditional and unconditional have become very popular in the developing world In many countries they have become Downloaded by James Cook University At 0406 10 January 2018 PT 288 FRANCISCO H G FERREIRA ET AL Table 8 Program Effects on Recipients and Siblings for Richer and Poorer Schools Threshold 10th percentile Threshold 25th percentile Threshold median Threshold 75th percentile Threshold 90th percentile Higher Lower Higher Lower Higher Lower Higher Lower Higher Lower Ownmale 0210 0328 0198 0315 0173 0252 0121 0244 0010 0239 0032 0125 0033 0098 0042 0049 0065 0037 0095 0031 Ownfemale 0190 0267 0180 0276 0197 0203 0170 0218 0217 0207 0026 0143 0027 0087 0037 0044 0050 0033 0072 0028 Siblingmale 0009 0008 0004 0072 0017 0035 0070 0028 0142 0026 0020 0099 0020 0057 0030 0030 0051 0024 0054 0020 Siblingfemale 0006 0012 0011 0002 0016 0015 0040 0022 0121 0023 0018 0085 0019 0069 0028 0033 0042 0024 0059 0020 Observations 7547 635 6870 1312 4200 3982 2227 5955 841 7341 R2 031 029 032 027 035 028 035 030 041 030 Pvalues Own M OwnF 062 075 067 070 066 041 057 057 012 041 SibMSibF 051 097 079 028 098 056 061 083 079 089 OwnMSibM 000 000 000 000 000 000 000 000 015 000 OwnFSibF 000 003 000 000 000 000 000 000 000 000 Note Higher Lower columns report results for the sample attending schools whose schoolspecific eligibility cutoff is above below the indicated percentile in the distribution of schoolspecific thresholds Sample size is 8182 in all regressions All specifications include a set of school dummies a set of single year age dummies a set of birth order dummies a dummy for the gender of the child dummy variables for siblinggender and a quartic in the score Standard errors adjust for clustering at the applicant primaryschool level significant at the 1 percent level at the 5 percent level Pvalues are from an Ftest of equality of parameter estimates Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 289 the largest social assistance program covering millions of households as is the case in Brazil Mexico Ecuador and South Africa Many of these pro grams also seek to increase the educational attainment of children However because enrollment rates are already very high at some school grades it has been suggested that cash transfers in particular those which are conditioned on school enrollment could be made more efficient if they were narrowly tar geted at child ages and grades where in the absence of the program dropout rates are high Attanasio et al 2009 de Janvry Sadoulet 2006 In this chapter we construct a simple model of schooling decisions and how they respond to a childspecific CCT The model predicts that such a program would unambiguously increase school enrollment among eligible children but would have a smaller and ambiguous effect on the school enrollment of their ineligible siblings We take the predictions of this model to data from a childspecific CCT in Cambodia This analysis shows that the program significantly increased the school enrollment of eligible children but left average schooling outcomes for their siblings unaffected Such an outcome is consistent with a combination of income densities and prefer ence parameters such that the positive income effect on sibling enrollment is exactly offset by the negative displacement effect This canceling out of the two effects is obviously not a requirement of the model though it appears to have been the case in Cambodia In general the displacement effect might be smaller or larger than the income effect An example of the latter situation seems consistent with the findings of Barrera Osorio et al 2011 from their analysis of a program in Bogotá Colombia That program makes reasonably large transfers equivalent to about 8 per cent of expenditures for the median recipient household conditional on the school enrollment of specific children selected for the program29 Barrera Osorio et al compare families in which two one or no children were selected into the program They conclude that the program positively affected the school enrollment of recipients but that this came in part at the expense of their siblings who were more likely to drop out of school and enter the labor market Similar intrahousehold spillovers from childspecific policy inter ventions have also been documented in other contexts Manacorda 2006 for instance uses historical data from the United States to show that mini mum working age laws that enabled children of a particular age to legally join the labor market led to a reduction in their siblings labor force participation and an increase in their siblings school participation Our simple model of twochild households can account for the broad pattern of results for Cambodia Colombia and the early 20th century United States It is also consistent with the evidence that enrollment effects Downloaded by James Cook University At 0406 10 January 2018 PT 290 FRANCISCO H G FERREIRA ET AL on recipients are heterogeneous and tend to be stronger among poorer households who were unlikely to have enrolled any children in school to begin with The model does less well however in accounting for heterogene ity in the effect on siblings the data provide no support for the prediction that the enrollment effect on siblings should be higher for betteroff households among the targeted population This suggests that other factors that the model abstracts from such as heterogeneity in ability tastes or parental preferences among children within the household or nonpecuniary educa tional spillovers among siblings may play an important role in intrahouse hold allocation decisions and should be the subject of future research More generally our results suggest that the characteristics of ineligible siblings not just those of children at whom the program is targeted should also be taken into account in the design of conditional cash transfers Households are complex decisionmaking entities and there are potentially important spillovers across household members including those excluded from the program by policy design These spillovers are likely to depend on the details of the program the age grade and genderspecific patterns of school enrollment and the opportunities available to children outside school Understanding these differences across settings and programs should be a priority for future research NOTES 1 A previous version of this paper was published as Policy Research work ing Paper 5001 Own and Sibling Effects of Conditional Cash Transfer Programs Theory and Evidence from Cambodia Impact Evaluation Series No 36 The World Bank httpdatatopicsworldbankorghnpfilesedstatsKHMimp09apdf 2 Nonpecuniary educational spillovers are not modeled explicitly but they are consistent with the ambiguity result If spillovers are positive they reinforce the income effect If they are negative eg if younger siblings seek to differentiate their choices so as to avoid competition with older siblings they would add to the pecuni ary displacement effect 3 The literature from Latin America is extensive see among others Schultz 2004 and Behrman Sengupta Todd 2005 on Mexico Schady and Araujo 2008 on Ecuador Attanasio Fitzsimons Gutierrez Meghir and Mesnard 2010 on Colombia Glewwe and Olinto 2004 on Honduras and Maluccio 2010 on Nicaragua Fiszbein and Schady 2009 review these and other studies 4 See Attanasio Meghir and Santiago 2012 Baird McIntosh and Ozler 2011 Bourguignon Ferreira and Leite 2003 de Brauw and Hoddinott 2011 Todd and Wolpin 2006 and Schady and Araujo 2008 5 Our parental welfare function is a simple transformation of that in Baland and Robinson 2000 adjusted for the fact that parents do not consume in period 2 Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 291 Our discount parameter β plays the same role as their altruism parameter δ For reasons which will become obvious we focus on two rather than n children but nothing of substance hinges on this other than considerable presentational simplicity 6 As in Basu and Van 1998 and Baland and Robinson 2000 we abstract from the actual decisionmaking process within the household As argued by Basu and Van 1998 the model does not conflict with recent evidence and theories which ask for the rejection of the unitary model of the household This is because we assume that a childs labor supply decision is taken by a parent this decision could be different if the decisionmaking were shifted to another member of household p 415 7 Although the binary nature of this decision problem simplifies the presentation the qualitative results extend to a version of the model in which σ is continuous in 01 so that children may divide their time between school and work This extension which also allows for child leisure is not presented here to economize on space but it is available from the authors on request 8 The assumptions that parents do not consume in Period 2 and cannot bequeath to their children allow us to abstract from financial savings which although essential for the efficiency argument in Baland and Robinson 2000 are not important for our purposes 9 In a richer model children might be allowed to differ in school ability labor pro ductivity or parental preference Such differences would alter the model in two ways First for households with A A A child heterogeneity would alter the pre transfer decision of which child to enroll This change can easily be accommodated by our set up by allowing π i π j More importantly however child heterogeneity would lead to a finite elasticity for the displacement of ineligible by eligible children once the transfer is introduced This extension is left for future work and we note only that the magnitude of the displacement effect we consider here is likely to be an upper bound since it assumes perfect substitutability across children 10 Adult labor continues to be supplied inelastically Appendix B shows that this is a reasonable approximation for Cambodia 11 If there were positive nonpecuniary spillovers from the enrollment say of an older sibling on the younger siblings educational attachment this effect would have the same sign as the positive income effect However it need not necessarily affect the same children We return to this point below 12 Scholarship recipients and their scores were posted at feeder schools and at CSP schools There was a complaint mechanism whereby community members could appeal the decisions made on the basis of the score either because they believed that an applicant had misrepresented their characteristics on the form or because they believed an applicant was poorer or less poor than indicated by the score In practice however less than 1 percent of applicants appealed the decisions and the recipient status of even fewer was revised as a result of a complaint 13 In practice within every large school the 25 students with the lowest dropout score were offered a scholarship of 60 and the 25 students with the next lowest scores were offered a scholarship of 45 in small schools the comparable numbers were 15 students with scholarships of 60 and 15 with scholarships of 45 We do not focus on this distinction in this chapter but do so elsewhere see Filmer and Schady 2011 Rather we compare applicants who were offered a scholarship regardless of the amount with others that were not Because the identification strategy is regres siondiscontinuity we are implicitly comparing applicants who were offered a 45 Downloaded by James Cook University At 0406 10 January 2018 PT 292 FRANCISCO H G FERREIRA ET AL scholarship with those who were offered no scholarship at all Students who were offered a 60 scholarship help estimate the control function that relates enrollment to the dropout score 14 Occasionally there were tied scores at the cutoff In these cases all applicants with the tied score at the cutoff were offered the scholarships 15 The sample was based on randomly selected schools in these five provinces The survey was limited to applicants ranked no more than 35 places above the cutoff in these schools This restriction was imposed to maximize the number of schools while maintaining the density of observations around the cutoff an important consideration when estimating program effects based on regressiondiscontinuity as discussed below 16 We also have access to a third data set These come from four unannounced visits to the 100 CSP schools in February April and June 2006 and in June 2007 in which the physical attendance of applicants was verified These allow us to validate our schooling impacts on recipients as we discuss below but do not allow an analysis of labor impacts nor sibling effects 17 The definitions of these variables follow the questionnaire on which the data are based Work for pay is defined as work for pay on a farm public or private sector or in a business belonging to someone else Work for no pay is defined as work for no pay on a farm private or public sector own account or in a business belonging to yourself or someone else in your household 18 Our survey did not collect information on what work for pay children are engaged in so we make use of a recent nationwide household survey the 2004 Cambodia SocioEconomic Survey CSES We limit the sample in the CSES to rural areas which most closely corresponds to the catchment areas of the CSP schools In rural areas 35 percent of boys age 1018 who work for pay are farm workers and another 26 percent work in construction among girls 35 percent of those who work for pay are farm workers and 27 percent work in the garment industry 19 Eightyeight percent of the applicants lived with at least one schoolage sibling at the time of the household survey 20 That said a question in the household survey asked if the applicant had enrolled in school at the beginning of the 200506 school year just as the scholar ships were offered and approximately one year before the survey was fielded 974 of applicants responded affirmatively which suggests a takeup rate of nearly 100 and thus that ATE estimates would be almost identical to the ITT estimates we report 21 All else being equal in CSP schools that received more applications and in those in which children have characteristics that make it more likely they will drop out a child with a high dropout score is more likely to be turned down for a scholarship than a similar child applying to a school that receives fewer applications or serves a population with a lower average dropout score 22 See Black Galdo and Smith 2007 for an application of the Tobit model in an RD framework 23 Because the cutoff falls at different values of the underlying score in different schools depending on the number of applications the distribution of applicant char acteristics and whether a school was defined as large or small it is not informa tive to graph outcomes as a function of the score Rather for these figures we redefine an applicants score in terms of the distance to the schoolspecific cutoff so that for example a value of 1 represents the nexttolast applicant to be offered a Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 293 scholarship within a school 0 the last and a value of 1 represents the first applicant within a school who was turned down The figures then graph outcomes as a function of this relative rank 24 These parametric regressions include a quartic in the relative rank but not the vector of school fixed effects or child characteristics Note that these differ slightly from the models estimated below which control for the composite score CSP school fixed effects and age and birth order dummies We note that using locally weighted least squares regressions as in Fan Gijbels 1996 instead of a quartic produces almost identical results 25 We also carried out this analysis for an additional activity household chores We found small and statistically insignificant program effects on time spent in house hold chores These results are available from the authors on request 26 This has been documented in settings as diverse as Brazil Emerson and Souza 2008 Nepal Edmonds 2006 and Taiwan Parish Willis 1993 Edmonds 2007 provides a thoughtful review 27 These results which we do not include in the chapter for the sake of brevity are available from the authors upon request 28 Nonpecuniary spillovers from older to younger siblings operating through information about schools and teachers or through incentives for emulation offer a potential explanation Qureshi 2011 finds that increases in schooling of older sisters in rural Pakistan lead to measurable increases in school enrollment for younger broth ers In a study of secondary examination GCSE performance in the UK Nicoletti and Rabe 2014 find similar spillovers from older to younger siblings and interest ingly find that these effects are more pronounced for poorer households Heterogeneity of this kind where effects weaken with family socioeconomic status would provide an offsetting channel for the income effect heterogeneity that we modeled in Section 2 and might account of the results we find with respect to the 90th percentile of the school cutoff distribution 29 The amount of the transfer and its value as a share of expenditures for the average recipient household are not reported in the chapter We are grateful to Felipe Barrera for providing us with this detail ACKNOWLEDGMENTS We thank the editor Sanghamitra Bandyopadhyay an anonymous referee as well as Felipe Barrera Luis Benveniste Eric Edmonds Ariel Fiszbein Karen Macours and seminar participants at LACEA Buenos Aires and Tulane University for helpful comments Thanks are also due to the World Bank Education Team for Cambodia and the members of Scholarship Team of the Royal Government of Cambodias Ministry of Education for valu able assistance in carrying out this work This work benefited from fund ing from the World Banks Research Support Budget P094396 as well as the BankNetherlands Partnership Program Trust Fund TF055023 Ryan Booth provided excellent research assistance The findings interpretations Downloaded by James Cook University At 0406 10 January 2018 PT 294 FRANCISCO H G FERREIRA ET AL and conclusions expressed in this chapter are those of the authors and do not necessarily represent the views of the World Bank its Executive Directors or the governments they represent REFERENCES Attanasio O Meghir C Santiago A 2012 Education choices in Mexico Using a structural model and a randomized experiment to evaluate PROGRESA Review of Economic Studies 791 3766 Attanasio O Meghir C Santiago A Shephard A 2009 Improving the education component of conditional cash transfers in urban settings IDB Education Division Working Paper No 1 Attanasio O Fitzsimons E Gutierrez M I Meghir C Mesnard A 2010 Child educa tion and work choices in the presence of a conditional cash transfer program in rural Colombia Economic Development and Cultural Change 582 181210 Baird S McIntosh C Özler B 2011 Cash or condition Evidence from a cash transfer experiment Quarterly Journal of Economics 1264 17091753 Baland JM Robinson J A 2000 Is child labor inefficient Journal of Political Economy 1084 663679 BarreraOsorio F Bertrand M Linden L L PerezCalle F 2011 Improving the design of conditional cash transfer programs Evidence from a randomized education experi ment in Colombia American Economic Journal Applied Economics 32 167195 Basu K Van P H 1998 The economics of child labor The American Economic Review 883 412427 Becker G 1991 A treatise on the family Cambridge Mass Harvard University Press Behrman J Sengupta P Todd P 2005 Progressing thorough PROGRESA An impact assessment of a school subsidy in Mexico Economic Development and Cultural Change 541 237275 BenPorath Y 1967 The production of human capital and the life cycle of earnings Journal of Political Economy 754 352365 Black D A Galdo J Smith J 2007 Evaluating the worker profiling and employment services system using a regression discontinuity approach American Economic Review 972 104107 Bourguignon F Ferreira F Leite P 2003 Conditional cash transfers schooling and child labor Microsimulating Brazils Bolsa Escola program The World Bank Economic Review 172 229254 Chaudhury N Parajuli D 2010 Conditional cash transfers and female schooling The impact of the female school stipend programme on public school enrolments in Punjab Pakistan Applied Economics 4228 35653583 de Brauw A Hoddinott J 2011 Must conditional cash transfer programs be conditioned to be effective The impact of conditioning transfers on school enrollment in Mexico Journal of Development Economics 962 359370 de Janvry A Sadoulet E 2006 Making conditional cash transfer programs more efficient Designing for maximum effect of the conditionality The World Bank Economic Review 201 129 Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 295 Edmonds E 2006 Understanding sibling differences in child labor Journal of Population Economics 194 795821 Edmonds E 2007 Child labor In T P Schultz J Strauss Eds Handbook of development economics pp 36073710 Amsterdam Elsevier Science NorthHolland Edmonds E Schady N 2012 Poverty alleviation and child labor American Economic Journal Economic Policy 44 100124 Emerson P Souza A P 2008 Birth order child labor and school attendance in Brazil World Development 369 16471664 Fan J Gijbels I 1996 Local polynomial modelling and its applications New York NY Chapman Hall Filmer D Schady N 2008 Getting girls into school Evidence from a scholarship program in Cambodia Economic Development and Cultural Change 563 581617 Filmer D Schady N 2011 Does more cash in conditional cash transfer programs always lead to larger impacts on school attendance Journal of Development Economics 961 150157 Filmer D Schady N 2014 The mediumterm effects of scholarships in a lowincome country Journal of Human Resources 493 663694 Fiszbein A Schady N 2009 Conditional cash transfers Reducing present and future poverty Washington DC The World Bank Glewwe P Olinto P 2004 Evaluating the impact of conditional cash transfers on schooling An experimental analysis of Honduras PRAF program Unpublished manuscript University of Minnesota Imbens G Lemieux T 2008 Regression discontinuity A guide to practice Journal of Econometrics 1422 61535 Maluccio J 2010 The impact of conditional cash transfers on consumption and investment in Nicaragua Journal of Development Studies 46 1438 Manacorda M 2006 Child labor and the labor supply of other household members Evidence from 1920 America American Economic Review 965 17881800 Nicoletti C Rabe B 2014 Sibling spillover effects in school achievement IZA Discussion Paper 8615 November Oettinger G 2000 Sibling similarity in high school graduation outcomes Causal interdependency or unobserved heterogeneity Southern Economic Journal 663 631648 Parish W Willis R 1993 Daughters education and family budgets Taiwan experiences Journal of Human Resources 284 862898 Qureshi J A 2011 Additional returns to investing in girls education impact on younger sibling human capital PhD thesis Harris School of Public Policy Studies University of Chicago Schady N Araujo M C 2008 Cash transfers conditions school enrollment and child work Evidence from a randomized experiment in Ecuador Economía 82 4370 Schultz T P 2004 School subsidies for the poor Evaluating the Mexican progresa poverty program Journal of Development Economics 741 199250 Todd P Wolpin K 2006 Assessing the impact of a school subsidy program in Mexico Using a social experiment to validate a dynamic behavioral model of child schooling and fertility American Economic Review 96513841417 Downloaded by James Cook University At 0406 10 January 2018 PT 296 FRANCISCO H G FERREIRA ET AL APPENDIX A PROOFS Proof of Proposition 1 Consider first the conditions under which both children would be enrolled σ σ 1 1 2 if and only if β β θ U A U h U A w U h U 2 A1 β β θ U A U h U A w U and 2 2 2 A2 β θ U A w U A U h U A1 implies A3 β θ U A w U A U h U A2 implies 2 2 A4 Concavity of U implies that U A w U A U A w U A 2 2 so A4 is always implied by A3 A3 is the necessary and sufficient condi tion for σ σ 1 1 2 Given the Inada conditions on U A such that β θ U A w U A U h U Concavity again implies that β θ U A w U A U h U A A and β θ U A w U A U h U β θ U A w U A U h U A A At equality A3 corresponds to point B in Fig 1 Now consider the conditions if any under which a single child would be enrolled σ σ i j 1 0 i j if and only if β θ β U A w U h U U A 2 U h A5 β θ β θ U A w U h U U A w U and 2 2 A6 A5 is just the converse of A1 and implies the converse of A3 A6 implies β θ U A w U A w U h U 2 A7 The Inada conditions imply that A such that β θ U A w U A w U h U 2 β θ U A w U A w U h U 2 Concavity of U implies that β θ U A w U A w U h U 2 β θ U A w U A w U h U 2 A A and β θ U A w U A w U h U 2 A A At equality A7 corresponds to point O in Fig 1 Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 297 It follows that i σ σ 0 1 2 A A ii σ σ i j 1 0 i j A A A iii σ σ 1 1 2 A A Table A1 Program Effects on Parents Did Activity Hours of Activity Tobit marginal effect OLS Worked for pay Worked without pay Hours worked for pay Hours worked without pay Hours worked for pay Hours worked without pay Male applicant Father 0047 0040 2083 1700 5628 1711 0078 0078 2158 3769 3428 3868 R2 020 020 018 022 Mother 0030 0021 0687 0189 1409 0334 0048 0056 0956 2544 2023 2600 R2 020 017 019 017 Female applicant Father 0032 0023 0032 2596 1664 2763 0038 0036 1201 1949 1466 1906 R2 013 008 010 011 Mother 0011 0073 0017 2411 0486 1803 0031 0032 0816 1428 1076 1461 R2 017 007 012 011 Notes Significant at the 1 percent level at the 5 percent level Sample sizes are 489 for the sample of male applicantsfathers 758 for the sample of male applicantsmothers1425 for the sample of female applicantsfathers and 1889 for the sample of female applicantsmothers All specifications include a set of school dummies a set of single year age dummies and a quartic in the score Standard errors adjust for clustering at the applicant primaryschool level Downloaded by James Cook University At 0406 10 January 2018 PT 298 FRANCISCO H G FERREIRA ET AL APPENDIX B PROGRAM EFFECTS ON PARENTS The model proposed in the chapter assumes that parents continue to sup ply labor inelastically in response to a transfer that is conditioned on child schooling This is consistent with the literature on CCTs see Fiszbein and Schady 2009 especially Chapter 4 for a review of the evidence from a num ber of countries In addition we find no evidence of significant changes in parental labor supply in Cambodia Table A1 summarizes the results of esti mating a simplified version of equation 1 focusing both on the incidence of different kinds of work lefthand panel and hours rightside panel with the latter estimated by Tobit and OLS as before We estimate this model for fathers and mothers and for male and female applicants separately The only significant finding in the Appendix Table is that mothers are less likely to work without pay when their daughters receive a scholarship a differ ence of 7 percentage points although the change in hours about 2 hours on average is quite small There is little evidence of large reallocations of parent labor in these data Downloaded by James Cook University At 0406 10 January 2018 PT Own and Sibling Effects of Conditional Cash Transfer Programs 299 Downloaded by James Cook University At 0406 10 January 2018 PT Full Terms Conditions of access and use can be found at httpswwwtandfonlinecomactionjournalInformationjournalCodefjds20 The Journal of Development Studies ISSN 00220388 Print 17439140 Online Journal homepage httpswwwtandfonlinecomloifjds20 EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes Seth R Gitter James Manley Bradford L Barham To cite this article Seth R Gitter James Manley Bradford L Barham 2013 EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes The Journal of Development Studies 4910 13971411 DOI 101080002203882013812200 To link to this article httpsdoiorg101080002203882013812200 Published online 15 Aug 2013 Submit your article to this journal Article views 464 View related articles Citing articles 2 View citing articles EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes SETH R GITTER JAMES MANLEY BRADFORD L BARHAM Department of Economics Towson University Towson MD USA Department of Agricultural and Applied Economics University of Wisconsin Madison WI USA Final version received July 2013 ABSTRACT Conditional cash transfer CCT programmes have been linked to improvements in education but effects on nutritional status are unclear We develop a theoretical household model demonstrating how CCTs educational requirements may constrain households to shift resources from younger to older children to sustain school attendance This could limit households capacity to invest in young childrens nutritional status particularly given a negative income shock In a Nicaraguan pilot CCT recipients consumption and nutritional status increased on average but less in households with schoolage children Effects are stronger in communities dealt an exogenous income shock 1 Introduction Conditional cash transfer CCT programmes provide payments to mothers based on their households utilisation of health services and childrens school attendance CCTs have multiple goals that appear quite complementary such as increasing schoolage childrens school attainment decreasing poverty enhancing womens autonomy and decisionmaking capacity and improving earlychildhood devel opment While these programmes have been shown to be successful in achieving some of these goals their effect on younger childrens under four years old nutritional status is less clear Fiszbein Schady 2009 Manley Gitter Slavchevska 2013 In particular CCTs have had limited success at improving heightforage z HAZ scores of young children a common measure of nutritional status Behrman Hoddinott 2005 Fernald Gertler Neufeld 2008 Maluccio Flores 2005 Macours Schady Vakis 2012 In Nicaragua two different pilot programmes failed to generate statistically significant increases in HAZ Maluccio Flores 2005 Macours et al 2012 though the earlier programme decreased stunting and the later programme improved cognitive development One contending explanation for weak child nutrition outcomes is that poor families face difficult tradeoffs in human capital investment for younger and older schoolaged children Sending school aged children off to work is an unfortunate but sometimes necessary event in poor households especially in response to negative income shocks If CCTs provide income conditional on school attendance then participating households may be constrained in their use of child labour and expenditure patterns in ways that trade off human capital investments in younger and older children Red de Protección Social RPS the Nicaraguan CCT studied in this article linked payment size to the average cost of school attendance including opportunity cost for a typical household but did not adjust payments for family size beyond a small payment for direct schooling costs This design Correspondence Address Dr Seth R Gitter Department of Economics Towson University 8000 York Road Towson MD 21252 USA Email srgittergmailcom The Journal of Development Studies 2013 Vol 49 No 10 13971411 httpdxdoiorg101080002203882013812200 2013 Taylor Francis feature was implemented in part to discourage fertility choices linked to payment streams Maluccio Flores 2005 Although the opportunity cost coverage was intended to supplement household income to the same degree as having the child work outside of the home the payment cap may have limited RPSs ability to fully cover those costs particularly in households with more schoolage eligible children Opportunity costs of CCT participation extend beyond sending schoolage children to work In addition to working outside the home older childrens attendance at school may be at the cost of time spent caring for younger siblings In Oportunidades Mexicos national CCT mothers increased their time spent caring for younger children when older children increased their school attendance Dubois RubioCodina 2012 In Colombia sisters of CCT recipients worked more and attended less school BarreraOsorio Bertrand Linden PerezCalle 2008 and in Cambodia a CCT increased school enrolment for recipients but not their siblings Ferreira Filmer Schady 2009 Thus households face opportunity costs of income and time associated with a childs school attendance and increased schooling investments induced by CCTs could hinder younger childrens development This tradeoff is more likely among CCT recipients under duress because of chronic low incomes andor serious negative income shocks Nicaraguas RPS provides a useful example because the panel data include a major shock to income that many recipients experienced Specifically the price of coffee dropped substantially in the early 2000s with large repercussions for the coffeeproducing areas that comprise about half of the sample When wages dropped households needed income more than ever and they were forced to consider sending schoolage children out to work and or keeping such children home from school to help care for younger siblings while adults in the household went off to work If the estimated opportunity costs of child labour and time as evaluated by RPS organisers fully cover the income the child might have generated as well as any additional income the household might need in order to make sure that the child was adequately prepared for school we will see no adverse effects the household will have a clear gain in utility from programme participation However if the transfers do not fully cover that cost the potential for households to benefit is more limited Poorer households choosing to participate may have to cut consumption by members other than the schoolbound child As the marginal utility of income rises the tradeoffs made by households become starker and it is in this context that we find evidence to support our theory Maluccio 2005 takes advantage of the coffee price shock in his analysis of the programme and we reconsider the data in light of our theory to explain some of his results including some that he finds puzzling Overall he finds that RPS seems to work children get more education they work less and household consumption expenditures increase Further RPS functioned as a safety net when recipient households in coffeeproducing areas faced the income shock However Maluccio 2005 glosses over a result that seems counterintuitive he finds that programme effects on child heightforage are lower in coffeeproducing areas Since marginal effects on nutritional status should be higher among the poorer households as found by Manley et al 2013 in their metaanalysis this is a curious result Maluccio struggles to explain it ultimately ascribing it to the smaller sample size 2005 p 36 While smaller sample size might explain a larger standard error such an issue is unlikely to change the sign the coefficient on RPS effects on HAZ overall is 036 and significant while the interaction term the modifier on RPS effects in coffee regions is 024 Thus in recipient areas producing coffee the overall effect of the programme is considerably reduced Next we offer a theory to explain this discrepancy and a broader view of how the number of older children in CCT programmes might affect the HAZ of younger siblings and then we find empirical support for the theory via econometric analysis of the RPS data In the theory section we develop a household model which shows that school and earlychildhood nutritional invest ment tradeoffs are most likely to occur when households face the following conditions 1 they have older children who would not have gone to school without the cash payment 2 the cash payment is sufficient to get the children to attend and 3 the transfers are smaller than the total cost of schooling including direct and opportunity costs In other words if households gain utility from sending children to school they may decrease their current consumption to do so The 1398 SR Gitter et al conditional cash transfer reduces the cost making schooling more accessible Our empirical analysis finds that RPS participants experienced substantially less benefit when faced with the tradeoff between older childrens schooling and younger childrens health particularly during an exogenous income shock 2 Theoretical Model Conditional cash transfer programmes are designed to foster the human capital development of both schoolaged children and younger children ages six and under1 If consumption goods improving younger childrens health are normal goods then it follows that unconditional cash transfers should have a positive impact However with finite resources substitution between investment in younger and older children can occur especially if transfers are conditioned in a manner that requires investment in older children Below we propose a simple household model to show how transfers conditional on schooling can potentially have diminished or even negative effects on younger childrens development because of this required investment We build on previous work on stochastic shocks to production and school attendance Gitter Barham 2009 Jacoby Skoufias 1997 Kruger 2007 by considering their influence on younger siblings health The household utility function Equation 1 has three parts represented by adults a the younger children yc and schoolaged children o The utility function for adults u reflects only their consumption ca The young childs human capital h is a function of the young childs consumption cyc and care from elders z The third part is the school age childrens human capital H which has two inputs consumption co and time spent on education e We assume that all three functions have diminishing marginal returns to any input The household faces four constraints The first is that the older children can divide their total time represented by the N the number of older children between three activities education e childcare zo and wage labour Lo Adults in the household divide their time between child rearing za and wage labour La The total production of childcare z and income wL is the sum of the production of the adults and the older children The older childrens productivity in childcare is a fraction αz of adults produc tivity where 0 αz 1 This is based on findings in Dubois and RubioCodina 2012 that mothers are better caretakers than a childs siblings Similarly in the wage market older children earn a wage that is some fraction of adults childcare productivity αw where 0 αw 1 The relationship between αz and αw could depend on local conditions and child characteristics such as gender and age It is worth noting that the core results do not change if we assume that adult and child labour are perfect substitutes in terms of childcare or wage labour As to the budget constraint households spend all of their money with no borrowing or saving allowed in this model U ¼ uðcaÞ þ hðcyc zÞ þ XN i¼1 Hiðco eÞ st N ¼ e þ zo þ Lo 1 ¼ za þ La z ¼ za þ azzo L ¼ La þ awLo wL ¼ ca þ cyc þ co 1 We next consider two potential forces that could affect human capital outcomes of the younger child h cyc z a cash transfer conditional on schooling and a shock combined with a cash transfer conditional on school attendance EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes 1399 development However an opposite effect is evident for households with two or more school agechildren in those communities There in coffee communities control households show some improvement in HAZ scores over time while treatment households do not This is consistent with the proposition that treatment households are sending their older children to school rather than having them take care of their younger siblings or work outside of the home to improve household consumption The role of family size in shaping programme impacts on HAZ deserves deeper attention 4 Econometric Specification and DifferenceinDifference Estimation We use the gold standard differenceindifference measures of programme impacts Hoddinott Skoufias 2004 Maluccio Flores 2005 Schultz 2004 to see whether outcomes in treatment households depend on the number of schoolage children Our regressions allow us to control for possible confounding factors to explore more cleanly the implications of the theoretical model presented in Section 2 for per capita consumption and heightforage z scores of younger children under five years Maluccio 2005 tests the interaction between RPS and being in a coffeeproducing area using a specification similar to Equation 3 This is called a triple difference approach and we redeploy that approach here The two main coefficients of interest are the programme effect α4 and how an additional schoolage child changes that effect In our model we explore whether the number of schoolaged children influences the effect of RPS by interacting those terms α7 Specifically we interact the number of older children aged 813 with all combinations of year and treatment dummies In order to test whether the income shock exposes the tradeoff associated with the CCT we would ideally add a fourth interaction distinguishing children in households with older siblings from children without them However since the number of regressors doubles with each additional set of interactions such a specification is too unwieldy and we settle for running two triple difference regressions of the form specified below in Equation 3 one on households in coffee producing regions and another in noncoffee producing areas Here the unit of observation depends on the outcome in question We use per capita consumption PCC for our first dependent variable and in this case each household is an observation For height for age z scores HAZ each observation is a child between 2 and 5 years old OlderChildren refers to the number of children in the household aged 813 years5 Note that one of the household characteristics included in the Xi vector is total household size so the OlderChildren variable is capturing more than the effect of adding one person of any age to a household This is particularly important with PCC Using per capita consumption expenditures as a dependent variable helps to identify how households responded to both the shock and potential transfers 5 Results 51 Per Capita Consumption The results of triple difference regressions on per capita consumption are shown in Table 4 Households randomised into CCT communities were on average better off through the time period and as one might expect such households were much better off after the programme actually began that is in 2002 as shown by the CCTYear coefficients Consumption expenditures rose substantially total expenditures increased roughly 25 per cent while food consumption climbed by about a third The results on the triple interaction term CCTYearMembers813 are consistent with our second prediction that each additional schoolaged child would reduce the transfers impact on per Flores 2005 show that treatment and control groups are not substantially different across most major indicators The average payment was 27 a month about the same as for the Mexican CCT programme Oportunidades though the share of household income accounted for by the transfer was considerably larger in Nicaragua To avoid incentivising fertility household payments and eligibility were deter mined exante of programme implementation and additional children were not funded For the most part payment size was not linked to household size Only a yearly 20 school supplies transfer was awarded per child which was not enough to compensate households for the full opportunity cost of sending additional children to school it merely offset school fees and transportation for each addi tional child Thus households with any schoolage children received approximately the same transfer regardless of the number of children However to receive the education transfer all children had to attend school Maluccio 2005 Furthermore the transfer was enforced as 10 per cent of households lost benefits due to failing to meet conditions Maluccio Flores 20053 To the extent that the additional transfer failed to cover the twofold opportunity costs sending each child to school imposed an additional cost on the household RPS also included conditions aimed at improving younger childrens physical development Children under 24 months got monthly care and growth monitoring from clinics while older children received bimonthly care and less attention Adequate weight gain was also a condition of payment Furthermore RPS increased the provision of iron supplements though as in PROGRESA these failed to decrease anaemia rates Maluccio Flores 2005 Previous analyses of RPS show positive but limited effects Maluccio and Flores 2005 find that RPS decreased stunting by 55 per cent a statistically significant outcome Another study by Maluccio 2005 found similar effects but also noted that RPS effects on HAZ were diminished in coffee communities p 35 In a different study Hoddinott and Wiesmann 2008 show that households in RPS increased calories from fruits and vegetables as well as the number of unique foods consumed Thus households improved their nutritional intake but this appears to have had only a limited impact on childrens height particularly in coffeeproducing communities 32 Describing the Shock The shock of interest is the steep decline in coffee prices that occurred during the implementation of RPS Figure 1 shows the change in coffee prices over the time period in question along with the years of data collection The first round of data collection occurred as coffee prices were falling The final Figure 1 Data collection timeline EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes 1401 round of data was collected during the low point of coffee prices when prices were only slightly above 60 cents per pound less than onethird of their high point in 1997 A previous analysis of the sample populations in Nicaragua Maluccio 2005 shows that the shock was associated with a 16 per cent decline in per capita consumption between the baseline year 2000 and the second year of treatment 2002 The shock affected about half of the communities As Maluccio 2005 notes coffee is only grown at certain altitudes and therefore is only feasible in certain communities Coffee communities were identified using surveys given to community leaders which ask questions about the presence or importance of coffee crops Of the 42 surveyed communities onehalf were coffeegrowing commu nities with 11 treatment and 10 control communities in both coffee and noncoffee communities The community measure may not control for heterogeneous effects associated with a households partici pation in the coffee sector however it is a more inclusive measure because it encompasses households that may have experienced indirect impacts from the coffee shock through falling local wages 33 Sample Statistics We have two outcome variables of interest First is per capita consumption the average amount of resources available to household members see Maluccio and Flores 2005 for more details on this measure Second we use heightforage z scores HAZ to measure younger childrens development Growth patterns of children under age five are similar for all ethnic groups WHO 2006 and growth charts allow the conversion of child heights into HAZ based on observed means and standard deviations for children of a given age and sex as compared to a reference population Heightfor age is often described as an indicator of longterm nutritional status among children and associated with longterm physical development Hoddinott Kinsey 2001 Strauss Thomas 1998 Victora et al 2008 Walker et al 2011 Waterlow et al 1977 It is also an indicator of a childs underlying health status and children showing lower levels of physical development for their age are often delayed in their mental development as well Many studies have evaluated childrens growth with reference to such a standardised population in order to estimate the health effects of natural disasters economic crisis and various policy interventions see for example Balk et al 2005 Ferreira Schady 2009 Hoddinott Kinsey 2001 Paxson Schady 2005 RPS collected data on anthropometrics for all children in treatment and control groups under the age of five Maluccio 2005 analyses a sample of all children aged 6 to 48 months at the time of measurement We use a slightly different sample focusing on children aged 24 to 60 months We do this for two main reasons First in this data children under the age of 24 months measured in 2002 would have been in utero after the start of the programme Such children are beneficiaries of prenatal care provided by the programme and thereafter received extra attention as part of the programme Our age restrictions still ensure that children evaluated after the programme began were no more than age three at the time of the intervention a level seen as an upper limit of the time during which child growth is highly susceptible to damage Victora et al 2008 Second as noted by GranthamMcGregor et al 2007 a pattern of growth faltering common in developing countries involves a pronounced dropoff in heightforage continuing through the first 12 18 months as difficult conditions combine with high susceptibility of nutritional status to risk factors In accordance with this global estimate we note in this dataset a large drop in heightforage through the first 18 months or so as shown in Figure 2 Since variation in HAZ is largely attributable to a secular trend during this period comparatively subtle effects such as those associated with a pro gramme like this are likely to be more noticeable if they happen after that window To investigate the theory described in Section 2 we examine the determinants of two outcomes per capita consumption and HAZ by household type For the descriptive statistics we break down households into those with at least two children aged 813 44 per cent of the sample versus those with one or no children of that age In the baseline there are an identical number of treatment and control communities Note that the results are similar whether we use the number of school aged children or a binary dummy for those households with two or more children 1402 SR Gitter et al Table 1 summarises the data on per capita consumption and younger childrens z scores with a differenceindifference comparison of baseline vs treatment year across households with zero or one schoolage child 813 years old as opposed to those with two or more Households with more schoolaged children see half the increase in total per capita consumption 501 vs 1012 this may be due to the smaller per capita transfer for larger households Finally in terms of HAZ scores small treatment households saw an increase of 03 over controls while treatment households with more children dropped by 013 compared to similar controls This result is partly attributable to the increase Figure 2 Height for age lowess regression Table 1 Dependent variable outcomes Per capita total consumption household Schoolaged children Zero or one Two or more Control Treatment Control Treatment Baseline 3746 3893 2918 3058 Treatment year 2002 3242 4401 2881 3521 Differenceindifference 1012a 501a 352 209 Height for age children 2460 months Schoolaged children Zero or one Two or more Control Treatment Control Treatment Baseline 174 189 202 191 Treatment year 2002 187 173 188 19 Differenceindifference 029b 013 192 062 Notes C is September 2000 Nicaraguan córdobas US1 is equivalent to roughly C1285 a indicates differenceindifference significant at 5 per cent b at 10 per cent tstats in brackets below EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes 1403 in the control groups HAZ scores for larger households Even ignoring this change smaller families saw HAZ scores increase 016 compared to 001 in larger households Before we can evaluate the effects of the shock we must establish that the groups are similar at baseline As shown in Table 2 52 per cent of the sample is in coffee communities which are evenly split among treatment and controls Adults in coffee communities do appear to have lower education levels but are otherwise similar to their counterparts Accordingly we control for education in the regression analysis This difference could bias the results if maternal education influences the pro gramme effects or the triple interaction of programme effects community type and the number of school age children However including those interactions does not change the main results Nor are those terms statistically significant predictors of heightforage Age and household size differences will be controlled for directly when we use regression analysis but first we consider some simple averages In Table 3 we review the effects of the coffee price shock verifying that it amplifies contrasts between households and particularly the difference between large and small families First with respect to consumption control households with zero or one schoolage child in coffee communities saw consumption fall roughly 25 per cent from the baseline period while larger households declined less Households in the control group with two or more children in coffee communities only lose about 10 per cent of per capita consumption 2940 to 2617 This supports the theory in Section 2 which anticipates offsetting consumption losses by increasing child labour4 The other per capita consump tion effect of note is that in both types of communities coffee and noncoffee the treatment house holds were able to maintain or increase per capita consumption levels during the shock while control households were not In three of the four group comparisons the differenceindifference DID for per capita consumption was positive and statistically significant between treatment and control households There are several observations to draw from the lower half of Table 3 regarding our other outcome of interest HAZ First HAZ scores are generally lower in coffee communities than noncoffee commu nities which could reflect the inherent tradeoffs associated with having a local economic activity that is relatively intensive in the use of older childrens labour Second treatment households in coffee areas with zero or one school age children saw HAZ increase by 019 across the study period while control households in the same categories saw HAZ decline by the same amount A simple DID suggests that in coffee areas treatment is associated with an increase of 038 z scores in households with no or one older child This is the type of salutary effect that programme designers might seek for earlychildhood Table 2 Descriptive statistics by group Noncoffee communities Coffee communities Schoolaged children Zero or one Two or more Zero or one Two or more Control Treatment Control Treatment Control Treatment Control Treatment Number of children 813 baseline 049 052 254 246 045 052 240 241 Total household size 588 570 882 824 565 553 831 829 Child age in months children 2460 months only 4157 4131 4143 4179 4046 3957 4179 4166 Age of childs mother years 2882 2821 3612 3649 2802 2738 3667 3587 of mothers who can reada 067 071 055 051 056 036 044 043 Note aThe share of mothers who can read is the only variable evincing a statistically significant difference across sub groups or coffee and noncoffee communities and that is only at the 10 per cent level 1404 SR Gitter et al Table 3 Outcomes by village coffee vs noncoffee status Per capita total consumption household Noncoffee communities Coffee communities Schoolaged children Zero or one Two or more Zero or one Two or more Control Treatment Control Treatment Control Treatment Control Treatment Baseline 3613 3891 2901 2980 3877 3895 2940 3127 Treatment year 2002 3383 4539 3064 3612 3084 4235 2617 3430 Differenceindifference 877a 240 469 135 1133a 255 626b 190 Height for age z scores Noncoffee communities Coffee communities Zero or one Two or more Zero or one Two or more Schoolaged children Control Treatment Control Treatment Control Treatment Control Treatment Baseline 154 156 175 16 191 217 225 21 Treatment year 2002 164 153 171 158 21 198 206 21 Differenceindifference 013 002 038b 019 064 05 179 071 Notes a indicates differenceindifference significant at 5 per cent b at 10 per cent tstats in brackets below EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes 1405 21 The Impact of a Conditional Cash Transfer on Young Child Development Suppose that the household receives the transfer conditional on the older child attending school full time so that e 1 The effects of the conditional transfer CT on the younger child h are the sum of the impacts on his or her consumption cyc and changes in childcare z We would expect that increasing household income potential by providing a conditional cash transfer would improve the younger childs human development through increased spending on the younger childs consumption and increasing quality of childcare However the size of the effect will depend on several eventualities including whether the transfer fully compensates for the total loss of older childrens income and costs associated with the older childrens schooling As older children move from work to school they go from contributing income and childcare to consuming more household resources Adults may increase childcare when older siblings cannot take care of the younger but adults also might increase their labour time to compensate for the loss of the labour of older children now in school In this case a conditional transfer could also reduce childcare and the welfare of younger children 22 An Income Shock Combined with a Cash Transfer Conditional on Schooling Previous research shows that income shocks can make CCTs conditionality of schooling more likely to bind de Janvry Finan Sadoulet Vakis 2006 Gitter and Barham 2009 Maluccio 2005 as households typically use child labour as a buffer against shocks All three papers showed CCTs had greater impacts on schooling and child labour during a shock and the last two examined RPS and the coffee shock When a household faces a consumption deficit the conditional transfer suddenly represents a dual opportunity cost Households choosing to take the transfer lose the potential income older children might bring home and instead face the added cost required to support a child participating in education If the transfer is not enough to cover these costs the household has committed to supporting one member to a certain degree without ensuring that consumption in the rest of the household will increase With the transfer household consumption may stay the same or even increase but depending on intrahousehold distribution the younger child may or may not see an increase in consumption andor may not continue to receive the same level of childcare limiting improvements in his or her nutritional and development status Without enough income to cushion the cost of sending a child to school the underlying tradeoff involved in sending children to school becomes apparent 3 Data Summary and Descriptive Statistics 31 Programme and Data Description Our data come from Nicaraguas Red de Protección Social see Hoddinott Bassett 2008 for detailed descriptions and comparison to other CCT programmes in Mexico and Honduras and Maluccio and Flores 2005 for a detailed description of the programme2 We use two survey rounds one collected in 2000 before the programme had begun and another in 2002 in the second year of payments It is worth noting that RPS like other CCTs randomised treatment at the community level Thus we have data on 42 communities with half each in control and treatment groups Maluccio and capita consumption The final prediction of the theoretical model is that a negative income shock should amplify the triple interaction term CCTYearMembers813 so we would expect this term to be larger in magnitude in coffee communities Indeed the triple interaction term is larger in absolute terms and statistically significant at the 10 per cent level in the coffee growing region implying that each additional schoolage child cut into the effect of the programme in all CCT areas but more significantly in those that also experienced the negative shock The effect of transfers on per capita consumption for households with more schoolaged children is smaller although this marginal effect is only statistically significant at the 10 per cent level for per capita consumption of food The magnitude of the term suggests that each additional schoolaged child reduces the effect of the CCT by roughly onesixth6 Note that in nonCCT coffee areas schoolage children in 2002 helped buffer household consumption from the effects of the shock as shown by the YearMembers813 coefficient which is significant at the 5 per cent level This is consistent with those children attending school when it is affordable and going to work or providing childcare to enable adults to work as necessary when it is not Finally note that household size was included and it has the expected strong negative coefficient This ensures that the estimated effects of schoolage children describe their effects not simply as an additional mouth to feed but reflect their particular status as children of this age 52 Child HeightforAge This articles central finding about the role of family size and programme structure on HAZ scores is shown in Table 5 The dependent variable is the heightforage of children under age five and the first two columns present results for all households with and without control variables for maternal education and literacy The second two columns separate results by coffee and noncoffee communities with the aforementioned control variables The regression estimates confirm the predictions of the theory elaborated above The first two rows of the pooled results show weak declines in height for age from 2000 to 2002 and in CCT vs nonCCT Table 4 Per capita consumption outcomes of CCT and household structure All areas All areas Noncoffee Coffee Total cons Food cons Food cons Food cons Year 4750 4978 3003 7031 1638c 1120c 1520 1534c CCT 2168 1375 1918 835 2999 2064 3092 2838 CCTyear 9105 8837 8258 9223 2598c 1888c 2799c 2349c CCTyearmembers813 1408 1445 1017 1711 984 834a 1257 1020a CCTmembers813 1155 789 1078 513 1040 804 1163 1127 Yearmembers813 756 508 73 875 723 552 843 412b Members813 1519 820 772 911 760 578 817 820 Household size 3624 2224 2127 2342 221c 154c 220c 215c Constant 57167 38164 37241 39167 2817c 1922c 3129c 2263c Observations 2383 2383 1244 1139 Rsquared 017 016 017 015 Notes Robust standard errors in parentheses errors clustered at community level a significant at 10 per cent level b at 5 per cent level c at 1 per cent level EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes 1407 communities but those effects are not statistically significant In the next row the pooled model shows that for households with no schoolaged children RPS increased HAZ by almost 04 z scores CCTYEAR Given that the estimated effect of most CCTs on HAZ is typically less than 02 this 04 increase represents a substantial gain However as seen in the triple interaction term CCTYearMembers813 each additional schoolage child in a CCTrecipient household reduces that gain by more than half and that outcome is driven by the coffee community experience as reflected in the largest estimate shown in that column The two rows below the triple interaction CCTMembers813 and YearMembers813 show that older children in general help the nutritional status of younger children all coefficients in pooled and separate models are positive and again they are occasionally significant such as in CCT and nonCCT areas combined in coffee areas in 2002 It is only the combination of CCT treatment older children and an income shock that make for negative effects on HAZ scores of younger siblings This result is consistent with older children not being available to help to compensate for the negative shock through labour efforts or childcare andor decreased consumption of the younger child In the pooled regression the statistical significance is muted when additional controls for maternal age and literacy are added but the magnitudes and signs are the same The significance holds up most robustly in the fuller specification for coffee commu nities and this outcome is consistent with the arguments raised above regarding negative CCT impacts on HAZ scores during times of shock Finally household controls for baseline consumption and maternal literacy show the expected positive relationship with childrens heightforage and male children MALE 1 but do not show statistically significant differences with female children Table 5 Height for age associations with CCT year and older children in the home Coffeenoncoffee community All All Noncoffee Coffee Year 016 009 004 017 010 012 018 014 CCT 021 009 010 017 015 016 024 020 CCTyear 039 026 014 043 015b 014 018 021b CCTyearmembers813 021 017 007 029 009b 010 015 013b CCTmembers813 010 007 008 014 008 011 017 010 Yearmembers813 013 009 005 015 006b 008 010 008a Members813 002 001 001 005 005 009 014 007 Age in months 001 002 001 002 000c 000c 000b 000c Male 004 002 008 003 006 007 012 006 Per capita spending baseline 014 011 013 011 002c 002c 003c 003c Mothers age 001 000 001 001 001 001 Mother can read 1 043 035 037 009c 013b 013c Constant 167 201 193 195 016c 026c 036c 034c Observations 1420 1202 598 604 Rsquared 008 01 01 011 Notes Robust standard errors in parentheses errors clustered at community level a significant at 10 per cent level b significant at 5 per cent level c significant at 1 per cent level 1408 SR Gitter et al Consider carefully the latter two columns of Table 5 to compare coffee producing communities with noncoffee communities Heightforage declined more in coffee communities in the second year of the study reflecting the shock though the difference is not statistically significant CCT areas were marginally and insignificantly worse off the first year but the second year that was strongly turned around as the payments began The statistically significant improvement of about half a standard deviation in the height of younger children is larger than has been linked to most previous pro grammes Again those gains are strongly undercut by the presence of older schoolage siblings in a household Each sibling cut the programme effect by about 03 of a standard deviation an amount that is statistically significant This loss is biting as we see that in coffee communities in general including CCT and nonCCT areas older siblings tended to be a help on average adding about 015 standard deviations to the heightforage of their younger siblings This effect was significant at the 10 per cent level An effect of comparable size is associated with having schoolage siblings in a CCT area prior to the shock though it is not statistically distinguishable from zero Finally we see negligible effects of having schoolage siblings when averaged across all situations that is across the years and whether the community was randomised into the CCT or not Age is strongly negatively correlated with height for age and per capita spending at the baseline is also significant and with the expected positive sign Maternal education is always highly significant and salutary for the childs nutritional status as well To sum up all signs are as expected At a baseline level younger childrens nutritional status is not much affected by having older children around In coffee communities they are overall helpful perhaps as they help cushion the income shock by providing childcare or leaving school to work However in households committed to the CCT they become strongly negative during the shock as older children are unable to contribute to their siblings HAZ and instead become an added constraint on the households potential investment in earlychildhood development As a result younger children apparently lose out as parents dedicate more scarce resources to schoolage children 6 Discussion We find that participation in RPS Nicaraguas CCT programme helped households cope with negative income shocks on younger childrens development However the effectiveness of the programme in improving consumption expenditures and the nutritional status of young children is limited in households with more schoolage children This unfortunate limitation is the product of three major facets of the situation First RPS transfers to households were a fixed amount indepen dent of the number of household members or the number of children This was done to avoid promotion of fertility but it may have had negative side effects by constraining household choices Second an income shock made the potential tradeoffs more urgent These two issues combined to render painful the conditionality constraint that households send older children to school and may have forced those childrens consumption and schooling outcomes to be pursued at the expense of the earlychildhood development of their siblings Omitted variable bias is also a possibility in any regression analysis Our threeway differencein difference specification differencing across time treatments and number of schoolage children in the household plus running separate regressions for coffeeproducing and noncoffeeproducing areas reduces the potential for competing explanations but it is possible that some unobserved phenomenon correlates with each of the dimensions across which we are taking differences We would expect such an omitted variable to also impact consumption however an examination of the consumption data does not support the presence of an omitted variable in RPS coffee growing communities 7 Conclusion and Policy Implications Maluccio 2005 describes how RPS apparently increased child enrolment in education decreased child labour and improved the level of consumption expenditures in poor areas However our reexamination of the data shows that improvements in consumption for households with more schoolaged children were lower This drop in effectiveness may be due to the insensitivity of the EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes 1409 transfer to the number of schoolage children in the household Programme designers should reconsi der the relationship between payments and household size particularly since there is little evidence that CCTs influence fertility Stecklov Winters Todd Regalia 2007 We show how CCTs could fail to help young children in recipient households experiencing a consumption shock We link this possibility to the conditionality of the transfers which requires schoolage children to attend school When not attending school older children generate labour earnings or participate in childcare contributions which take on added significance when income is exogenously reduced Programme impact could be improved if funds were increased to compensate for lost child labour including both wage labour and help in child rearing It is worth noting that CCTs have not been designed with the intention of adapting to shocks Bourguignon 2000 Grosh Del Ninno Tesliuc Ouerghi 2008 Grosh et al 2008 rightly point out that targeting after a crisis may be particularly difficult given the need for realtime data in lowcapacity countries and the fact that monitoring earnings in informal markets makes monitoring and targeting difficult This work suggests that there are tradeoffs between investment in education and early childhood nutrition and households are most likely to face those tradeoffs during a shock These types of shocks are common in areas that are often the focus of CCTs and they may interact with programme stipulations to have unintended negative effects on earlychild development One strategy that many countries used during the global economic downturn was to expand cash transfer programmes by increasing funding and starting new programmes Fiszbein Ringold Srinivasan 2011 Likewise higher payments to more households during shocks may help overcome the issues raised in this article Acknowledgements The authors are grateful for financial support provided by the InterAmerican Development Bank The authors thank helpful comments of reviewers as well as participants and organisers of both IDB sessions for helpful input on ways to improve this article and invite further comments All errors are solely our responsibility Notes 1 The RPS programme targeted school children aged 713 Most impacts on human capital development are seen before the age of three Bhutta et al 2008 Walker et al 2011 but for simplicity we chose to have two groups of children 2 Data are available at httpwwwifpriorgdatasetnicaragua 3 Household level data on enforcement are not available so we cannot control for household size or differences in communities in enforcement 4 Maluccio 2005 and Gitter and Barham 2009 using the same data find increases in child labour commensurate with this observation 5 This age range was chosen because RPS provided a school transfer for those aged 713 However we found that essentially none of the sevenyearolds worked so we dropped them from the older cohort 6 Only three households have more than five schoolaged children so the negative effect of having older children is almost never predicted to supersede the positive effects of receiving the transfer References Balk D Storeygard A Levy M Gaskell J Sharma M Flor R 2005 Child hunger in the developing world An analysis of environmental and social correlates Food Policy 30 584611 BarreraOsorio F Bertrand M Linden L L PerezCalle F 2008 Conditional cash transfers in education Design features peer and sibling effects Evidence from a randomized experiment in Colombia Working Paper 13890 Cambridge MA NBER Behrman J Hoddinott J 2005 Program evaluation with unobserved heterogeneity and selective implementation The Mexican PROGRESA impact on child nutrition Oxford Bulletin of Economics and Statistics 67 547569 Bhutta Z A Ahmed T Black R E Cousens S Dewey K Giugliani E Shekar M 2008 What works Interventions for maternal and child undernutrition and survival The Lancet 3719610 417440 1410 SR Gitter et al Bourguignon F 2000 Comments on Crises and the poor Socially responsible macroeconomics by Nora Lustig Economia 1 1 2326 de Janvry A Finan F Sadoulet E Vakis R 2006 Can conditional cash transfers serve as safety nets to keep children at school and out of the labor market Journal of Development Economics 79 349373 Dubois P RubioCodina M 2012 Child Care Provision Semiparametric Evidence from a Randomized Experiment in Mexico Annals of Economics and Statistics 105106 155184 Fernald L C H Gertler P J Neufeld L M 2008 Role of cash in conditional cash transfer programmes for child health growth and development An analysis of Mexicos Oportunidades Lancet 371 828837 Ferreira F Schady N 2009 Aggregate economic shocks child schooling and child health World Bank Research Observer 242 147181 Ferreira F Filmer D Schady N 2009 Own and sibling effects of conditional cash transfer programs Theory and evidence from Cambodia Policy Research Working Paper no 5001 Washington DC World Bank Fiszbein A Ringold D Srinivasan S 2011 Cash transfers children and the crisis Protecting current and future investments Development Policy Review 29 585601 Fiszbein A Schady N 2009 Conditional cash transfers Reducing present and future poverty Washington DC The World Bank Gitter S R Barham B L 2009 Conditional cash transfers shocks and education An impact evaluation of Nicaraguas RPS Journal of Development Studies 45 17471767 GranthamMcGregor S Cheung Y B Cueto S Glewwe P Richter L Strupp B and the International Child Development Steering Group 2007 Developmental potential in the first 5 years for children in developing countries Lancet 369 6070 Grosh M Del Ninno C Tesliuc E Ouerghi A 2008 For protection and promotion The design and implementation of effective safety nets Washington DC The World Bank Hoddinott J Bassett L 2008 Conditional cash transfer programs and nutrition in Latin America Assessment of impacts and strategies for improvement etrieved from httpssrncomabstract1305326 Hoddinott J Kinsey B 2001 Child growth in time of drought Oxford Bulletin of Economics and Statistics 63 409436 Hoddinott J Skoufias E 2004 The impact of Progresa on food consumption Economic Development and Cultural Change 53 3761 Hoddinott J Wiesmann D 2008 The impact of conditional cash transfer programs on food consumption in Honduras Mexico and Nicaragua Retrieved from httpssrncomabstract1269417 Jacoby H Skoufias E 1997 Risk financial markets and human capital in a developing country Review of Economic Studies 64 311335 Kruger D 2007 Coffee production effects on child labour and schooling in rural Brazil Journal of Development Economics 82 448463 Macours K Schady N Vakis R 2012 Cash transfers behavioral changes and cognitive development in early childhood Evidence from a randomized experiment American Economic Journal Applied Economics 4 247273 Maluccio J 2005 Coping with the coffee crisis in Central America The role of the Nicaraguan Red de Protección Social FCND Discussion Paper No 188 Washington DC International Food Policy Research Institute Maluccio J Flores R 2005 Impact evaluation of a conditional cash transfer program The Nicaraguan Red de Protección Social Research Report 141Washington DC International Food Policy Research Institute Manley J Gitter S Slavchevska V 2013 How effective are cash transfers at improving nutritional status World Development 48 133155 Paxson C Schady N 2005 Child health and economic crisis in Peru World Bank Economic Review 19 203223 Schultz T 2004 School subsidies for the poor Evaluating the Mexican Progresa poverty program Journal of Development Economics 74 199250 Stecklov G Winters P Todd J Regalia F 2007 Unintended effects of poverty programmes on childbearing in less developed countries Experimental evidence from Latin America Population Studies 612 125140 Strauss J Thomas D 1998 Health nutrition and economic development Journal of Economic Literature 36 766817 Victora C G Adair L Fall C Hallal P C Martorell R Richter L Sachdev H S for the Maternal and Child Undernutrition Study Group 2008 Maternal and child undernutrition Consequences for adult health and health capital Lancet 371 340357 Walker S P Wachs T D GranthamMcGregor S Black M M Nelson C A Huffman S L and Richter L 2011 Inequality in early childhood Risk and protective factors for early development Lancet 378 13251338 Waterlow J C Buzina R Keller W Lane J M Nichaman M Z Tanner J M 1977 The presentation and use of height and weight data for comparing the nutritional status of groups of children under the age of 10 years Bulletin of the World Health Organization 55 489498 WHO 2006 World Health Organization child growth standards backgrounder 1 Retrieved from httpwwwwhoint childgrowth1whatpdf EarlyChildhood Nutrition and Educational Conditional Cash Transfer Programmes 1411 The influence of Bolsa Familia conditional cash transfer program on child labor in Brazil Paloma Santana Moreira Pais Felipe de Figueiredo Silva and Evandro Camargos Teixeira Rural Economics Department Federal University of Viçosa Viçosa Brazil Abstract Purpose The Brazilian Government created the Bolsa Familia program to combat poverty and the insertion of so many children into the labor market This program is an income transfer program subject to certain conditions such as a minimum school attendance for children under 17 years of age In 2006 almost half of the people with an income per capita of R30000 US13953 per month declared that they received this benefit Accordingly the purpose of this paper is to analyze the impact of Bolsa Familia on child labor in Brazil in 2006 Designmethodologyapproach The authors used a propensity score matching model with data from the National Household Sample Survey PESQUISA NACIONAL POR AMOSTRA DE DOMICÍLIOS PNAD for 2006 Findings Results indicate that the program increased the number of hours of child labor in Brazil However this outcome might be explained by the fact that those families who received Bolsa Familia were also those with higher socioeconomic vulnerability Thus they need to guarantee their survival with the income generated via child labor Social implications The Brazilian Government needs to invest not only in monetary transfer policies but also in the improvement of the job market to create opportunities for the social development of children Originalityvalue The contribution of the paper is the investigation into the effect of the Bolsa Familia program on the average time allocated to child labor the authors find that this time allocation could be reduced by requiring a compulsory school attendance Keywords Brazil Education Child labour Bolsa Familia program Paper type Research paper 1 Introduction Given child labors widespread occurrence the situations causes consequences and eradication policies are a recurrent theme in the economic literature According to the International Labour Office ILO 2013 168 million children worldwide are in child labor accounting for almost 11 percent of the child population as a whole of this group 5833 percent are involved in the farm sector 3214 percent are in the service sector and 953 percent are in industrial sector The Asia and Pacific regions have the most child labor 4628 percent of children followed by 3515 percent in SubSaharan Africa 745 percent in Latin America and the Caribbean and 1112 percent in other regions ILO 2013 Several causes of child labor have been discussed in the literature but poverty is the most cited given its prevalence in places where this type of labor prevails Ferro and Kassouf 2005 pointed out the importance of childrens earnings for families who have very low per capita income and already participate in child labor Cacciamali et al 2010 also stated that rather than attend school the lowincome population feels obliged to enter earlier into the job market to pursue their own survival or to complement the family income Thus the insertion of children into the job market early perpetuates the cycle of poverty and low schooling within the society Since researchers seek to reduce poverty in the long run and to increase the human capital of children in underprivileged families they have been studying the effect of several policies such as conditional cash transfers CCT on child labor Ravallion and Wodon 2000 evaluated the impact of Food for Education Program FEE on child labor for International Journal of Social Economics Vol 44 No 2 2017 pp 206221 Emerald Publishing Limited 03068293 DOI 101108IJSE0220150038 Received 23 February 2015 Revised 3 August 2015 12 October 2015 Accepted 13 October 2015 The current issue and full text archive of this journal is available on Emerald Insight at wwwemeraldinsightcom03068293htm 206 IJSE 442 Bangladesh and observed an increase in school attendance and a weak decrease in the occurrence of child labor FEE requires children of underprivileged rural families to have a school attendance of more than 85 percent to obtain the monetary transfer Ravallion and Wodon argued that parents were sacrificing other uses of their childrens time such as time spent on leisure rather than work to receive the monetary transfer Thus they concluded that FEE only had a moderate influence on child labor earnings However their study did not look into the impact of the program on the childrens hours of work and the authors admitted that the data do not allow them to capture any effects of the program on small amounts of child labor In Mexico Skoufias and Parker 2001 evaluated the consequences of Programa de Educacion Salud y Alimentacion PROGRESA on child labor and school attendance Although the goal of the program was not to influence child labor directly but to enhance the human capital of underprivileged families the researchers recognized that the program could have a positive effect on this type of labor by requiring childrens school attendance These authors used a differenceindifference model instead of the twostage model Ravallion and Wodon 2000 used They concluded that the program simultaneously increased school attendance and decreased child labor although its influence on female children was less than on male children given the complementarity between domestic work and school attendance However the estimates obtained by the authors reveal that the treatment and control samples have significant preprogram differences which can affect the results This problem could be solved by adopting a propensity score matching model There have also been numerous studies of CCTs for South American countries such as Colombia Uruguay Ecuador and Brazil Attanasio et al 2010 investigated the influence of Familias en Accion FA on school attendance and child labor in Colombia and observed an effective program FA required a school attendance of more than 80 percent and included other priorities such as health and nutrition The authors estimated the effects of the program within a differenceindifference framework and they found an increase in school attendance and a decrease in time spent on labor Amarante et al 2011 evaluated the influence of Plan Nacional de Atencion a La Emergencia Social PANES on child labor and school attendance in Uruguay PANES was a temporary program that lasted approximately three years from April 2005 to December 2007 and aimed to fight poverty among families under the national poverty line which amounted to 8 percent of the population The program featured requirements for school attendance and health checkups Amarante et al 2011 also used a differenceindifference model in addition to others and found that PANES did not affect school attendance child labor or household income Thus they argued that the monetary transfer was not generous enough to promote school attendance or that the determinants of child school attendance are more complex and require complementary interventions Edmonds and Schady 2012 used another variable time allocation to evaluate the influence of a CCT program Bono Desarollo Humano BDH on child labor in Ecuador Unlike the previous programs BDH did not require any school attendance They found a decrease in child labor for both paid and unpaid child labor using a twostage least square model 2SLS In Brazil Araújo 2009 2010 Cacciamali et al 2010 and Araújo et al 2010 discussed the influence of Bolsa Familia1 on child labor This CCT program requires that children attend school Cacciamali et al 2010 used a bivariate probit model while the others used a propensity score matching model Cacciamali et al 2010 found an increase in school attendance as well as an increase in the probability of child labor occurrence Additionally they pointed out a worse effect on rural children and as a potential solution they suggested a more precise policy toward them The studies proposed by Araújo 2009 and Araújo et al 2010 agreed with Cacciamali et al 2010 and confirmed that the program encouraged an increase in school attendance without benefiting child labor eradication Additionally 207 Cash transfer program on child labor Araújo 2009 highlighted the programs limitations due to the increased work associated with school which can affect students performance in school and their time allocation to other basic activities that are important to their development On the other hand Araújo 2010 suggested that the program can be an important tool for achieving child labor eradication however the paper pointed out that there are other determinants of child labor rather than poverty such as regional and gender factors and advocated that these other determinants should be considered in the same policies fighting child labor Despite the different methods and data sources used it is important to point out the possibility that the data collected on child labor cannot capture the true extent of the problem Given the illegal nature of such work those who answer the questionnaires tend to not give true statements especially when children are involved in work such as drug trafficking and child prostitution thus skewing the data results These Brazilfocused papers evaluated the influence of Bolsa Familia on child labor by analyzing the probability of participation in this type of activity and the proportion of children involved in it However Attanasio et al 2010 indicated that the precondition of school attendance by CCT programs eg FEE PROGRESA and PANES requires a time reallocation between school work and leisure It is worth noting that since schooltime allocation is comparatively short the influence of these CCT programs does not necessarily affect childlabor time allocation Therefore we contribute to the literature by evaluating the influence of the Bolsa Familia conditional cash transfer program on the time allocation of child labor which is represented by the their time spent on work To accomplish this objective we first delineated a set of variables that determine the participation in the Bolsa Familia program next we built a control group based on these key variables to create a test sample This procedure is explained in Section 4 We found that a higher school attendance increases the probability of a family participating in this program On the other hand a higher income per capita level decreases the probability of a family participating in it Additionally when a mothers education level is higher or the family lives in a household in a metropolitan urban area the probability of participating in the program decreases alternatively when a household has a greater number of children or is located in a rural location the probability of participation increases Contrasting the treatment group with a control group of similarcharacteristic nonparticipants we found that similar to previous studies participation in the Bolsa Familia program increases childlabor time allocation This result is a direct consequence of the high exposure of these families to economic vulnerabilities such as poverty and points out the importance of generating specific policies to eradicate child labor The rest of the paper is organized in the following way First we describe the Bolsa Familia conditional cash program and the state of child labor in Brazil Second a theoretical framework is proposed based on the literature followed by the empirical specification in Section 4 In Section 5 we present the empirical results and compare the results with those already found in the literature Finally we present the main remarks strengths and limitations of this paper 2 Bolsa Familia conditional cash transfer program and child labor in Brazil In Brazil conditional cash transfer programs were created mainly with the purpose of breaking the perpetuation of poverty and to generate conditions that allow children not to enter into the job market early The Bolsa Familia conditional cash transfer program was created in January of 2004 law number 10836 and is characterized as a monetary transfer program subject to two main preconditions First it requires compulsory school attendance for children between 6 and 17 years of age Second it requires a social assistance be provided to children and teenagers who are up to 15 years old and who have been removed from child labor by the Child Labor Eradication Program Programa de 208 IJSE 442 Erradicação do Trabalho Infantil Peti Children who participate in any labor activities should participate in the Service of Living and Strengthening of Links Serviço de Convivência e Fortalecimento de Vínculos SCFV of Peti and achieve a minimal monthly attendance of 85 percent Although these conditions guarantee access to basic social benefits such as education and social assistance the main consequence of an income transfer program is the immediate relief of poverty conditions According to the data from the Household Sample Survey PESQUISA NACIONAL POR AMOSTRA DE DOMICÍLIOS PNAD 2006 59 percent of the people that earned onefourth of the minimum wage R87502 US4070 participated in this program while 41 percent of those who earned between R8750 and less than half of the minimum wage also participated Figure 1 The proportion of people that participated in the program decreases as the income level increases achieving less than 1 percent when the income level reaches twice the minimum wage of 2006 These results are illustrated in Figure 1 According to the National Secretary of Income and Citizenship3 Secretaria Nacional de Renda de Cidadania SENARC 2014 R2399 million approximately US1039 million were spent on this program in the municipalities and Federal District between April 2006 and October 2013 In 2013 R417 million US181 million were spent on municipalities that participated in the Bolsa Familia program of which 1016 percent were in the northern region of Brazil 4640 percent were to the northeast 2777 percent were in the southeast 557 percent were in the midwest and 99 percent were in the southern region SENARC 2014 also pointed out that 14 million families received monetary transfers from this program We found a similar distribution when we compared both the groups of children and teenagers who attended school and the children who belonged to families that participated in the program which might suggest that a higher proportion of the program beneficiaries brings about a higher school attendance Figure 2 presents such trends and shows that the percentage of children between 14 and 17 years of age who attended school decreased at the same rate as the proportion of students belonging to families that participated in the program Furthermore SENARC 2014 pointed out that between October and November of 2013 959 percent of the 16085 million children between 6 and 17 years old who were in a family that participated in Bolsa Familia had an attendance record of equal or higher than 85 percent while only 41 percent did not obey the programs conditionality According to Cacciamali et al 2010 programs such as the Bolsa Familia combat two problems that perpetuate poverty across generations because they assure a minimum level of subsistence income for underprivileged families and they allow for the acquisition of human capital Though previous studies have indicated that Bolsa Familia does not impact 0 10 20 30 40 50 60 70 Up to 14 SM From 14 to 12 SM From 12 to 1 SM From 1 to 2 SM From 2 to 3 SM From 3 to 5 SM Above 5 SM Income brackets in minimum wages SM Source Research results based on PNAD 2006 database Figure 1 Proportion of people who participated of the Bolsa Familia program per income per capita level in 2006 209 Cash transfer program on child labor child labor Cacciamali et al 2010 also argued that the cash transfer raises the families earnings which allows their members to allocate more time to leisure or education without incurring losses with respect to the minimum income necessary to survive As such these programs could reduce the occurrence of child labor in poor families over time 3 Theoretical framework This work rests upon the theory about the time allocation proposed by Becker 1965 with adaptations of Ersado 2002 In this theoretical model a familys time is distributed across employment leisure and education and childrens timeallocation decisions take into account the private returns of each activity We adopted the neoclassical model of unit home labor supply in which the family makes joint decisions about domestic consumption and labor supply for its members Ersado 2002 assumed that the decision about whether children participate in workforce leisure activities or school is determined by an adult person Additionally Ersado suggested that a family composed of an adult and a child maximizes its function of utility into two periods t and t1 according to the following equation where U is a concave utility function based on the consumption set C of the childs schooling Sc of the leisure time of the adult and the child Lp Lc and of a vector of individual and family characteristics X In the first period the parent decides whether to send his child to school or work If the decision to send the child to school is made Ersado 2002 assumes that the child will earn a wage Wc in the first period and a wage Wu wage of an unskilled adult in the second period On the other hand if the adult decides to send the child to school the child will not earn any wage in the first period but will earn a wage Ws wage of a skilled adult in the second period One can suppose therefore that The familys resources will depend on the decision of the adult about sending the child to work or to school over the period t In the period t1 the consumption and the leisure of the child will be subject to his wage Wu or Ws which will depend on the childs time spent on education Thus the head of the family should maximize the utility function presented in Equation 1 subject to the availability of time and to the restrictions of resources in each period Source Research results based on PNAD 2006 database where Wt is a vector of wage rates for the adult and the child T the total time available for the family ie T Tp Tc Ct the value of the total consumption and Ωt the nonwage income The nonwage income Ωt includes the profits of the selfemployment in farming and nonfarming activities Π income coming from the familys asset interests At government transfers and other sources of revenues that are not derived from labor Yt as presented in the following equation Ωt Πt δAt Yt where δ is the interest rate The family head will also consider the childs time restriction since it must be divided between school work and leisure in addition to the unpaid domestic work as shown in the following equation Tc Lct Sct Ect where Ect is the time spent on paid or unpaid labors Furthermore the temporal trajectory of the family assets can be defined by the following equation At1 1 δAt Πt Yt WtEct WtTp Lpt Ct where At is the total of assets in the previous period and Πt Yt WtEct WtTp Lpt Ct is the saving or dissaving if negative in the period t after that periods consumption Using Equations 3 and 5 and solving for Ωt one has Ωt At 1 At Ct WtEct WtTp Lpt ΔAt where Equation 6 consists of an intertemporal measure of the nonwage income that allows the agents either to save or not It follows that the family head maximizes the domestic welfare Equation 1 subject to the restrictions of the childs time Equation 4 and nonwage income Equation 6 The solution to the maximization problem is a function of prices wagerate for the child unskilled and skilled labor household asset holding nonwage income and other factors such as family characteristics included in the vector Xt as presented in the following equation ΓWt Πt At Yt Xt Ψ where Ψ denotes all the observed and unobserved characteristics that affect the parents decisions The indirect utility function is obtained by replacing the choice vector Equation 6 in the utility function Equation 1 to define the maximum utility of the family V UΓWt Πt At Yt Xt Ψ Given that schooling can be seen as a form of valuing human capital a utility function evaluating the decision of whether or not to school a child is defined by Vs UΓWs Πt At Yt Xt Ψ The parent will decide to send the child to school instead of sending him or her to work in the period t if the parent understands that the future profits will be higher with the valuing of human capital as seen in the following equation Vs Vu 0 Cash transfer program on child labor 211 where Vu is the indirect utility function for the decision of not sending the child to school Vu UΓWu Πt At Yt Xt Ψ Therefore this model can be used to guide our empirical results From a CCTprogram perspective we can expect that an increase in the familys income can reduce the childs time allocation on work paid or unpaid Additionally given the preconditions of this program an increase in school attendance as well as an enhanced future wage in t 1 is expected 4 Empirical specification 41 Propensity score matching model As mentioned before several papers have evaluated the effect of such programs as the Bolsa Familia on child labor Araújo 2009 stated that these studies should be able to compare the results achieved by children of families under the CCT program and children of families that are not participating in the program However Becker and Ichino 2002 identified difficulties in existing microdata studies where the selection of control and treatment groups is not realized randomly To counter this problem they argued that the use of a propensity score matching model could minimize such bias since it compares the results of treated and control groups that are similar under other characteristics Thus we decided to use a propensity score matching model to determine the influence of the Bolsa Familia program on child labor in Brazil The propensity score corresponds to the probability that families will receive the programs monetary transfer given its characteristics To measure the programs effect on child labor we can use the average treatment effect ATE defined by Rosenbaum and Rubin 1983 as ATE Eyl Ey0 where E denotes the expected value of the treatments effect y1 is the result of the treated variable people who participate and y0 the result of the control group people who do not participate On the other hand we can estimate the effect of the treatment by the ATE upon the Treaty ATT ATT Ey1d 1 Ey0d 1 where ATT refers to the mean of the effect for persons who participate in the program Though this procedure will yield results about the effects of Bolsa Familia on child labor there is a problem in the estimation Equation 13 since it is not possible to find an individual who receives a monetary transfer and at the same time does not receive a monetary transfer ie someone belonging to both treatment and control groups Araújo et al 2010 suggested the replacement of Ey0d 1 by a group of people who effectively do not receive a monetary transfer Ey0d 0 Thus Equation 13 becomes ETM1 Ey1d 1 Ey0d 0 However estimation of a treatments effect by Equation 14 can generate biased results since the participation in the treatment group is not always random and might depend on a set of characteristics that qualify the participants Araújo 2009 Thus people with similar characteristics may or may not participate in the program and the results therefore do not depend on the characteristics of the participants even though the selection was based on observable characteristics Araújo 2010 We can solve the bias of selfselection by imposing some conditions First we should consider the conditional independence hypothesis Rosenbaum and Rubin 1983 which suggests that the results are independent of the treatment group and the average effect of Cash transfer program on child labor 212 the treatment is obtained on the basis of the differences among the results of the treated and nontreated groups To achieve this analysis we implemented a propensity score matching model which consists of finding the counterfactuals that represent what would have happened if the program had not been implemented Rosenbaum and Rubin 1983 defined the propensity score as the conditional probability of receiving the treatment or the probability of participating in the program under certain characteristics pX Prd 1X EdX where d 0 1 indicates exposure to the treatment and X is the multidimensional vector of pretreatment characteristics Becker and Ichino 2002 It is also necessary to adopt a statistical model with a limited restricted dependent variable to estimate the propensity score which we used in the Logit model Thus based on the estimate of the propensity score defined by Equation 15 the Average Treatment Effect upon the Treaty ATT can be estimated as follows ATT EEy1d 1 pXi Ey0d 0 pXid 1 where the expected value refers to the distribution of the probability of Xi given the treatment and y1 and y0 are the potential results for the two counterfactual situations ie with or without the treatment Becker and Ichino 2002 However Becker and Ichino 2002 identified a weakness in estimating the Treatment Effect Equation 16 by the propensity score the necessity of adopting a matching algorithm The most used matching algorithms are nearest neighbor matching by stratification Kernel or radial We used the nearest neighbor matching as in Araújo 2009 2010 where all units of the treatment group find a match in the control group Becker and Ichino 2002 We tested the matching quality by checking if the matching procedure is able to balance the distribution of the relevant variables in both control and treatment groups Caliendo and Kopeinig 2008 state that the basic idea of this test is to compare the situation before and after using the matching algorithm to verify whether differences remain between the groups after using the propensity score If after conditioning on the propensity score there is still a dependence on X this dependence suggests either misspecification in the model used to estimate or that there is a fundamental lack of comparability between the two groups Caliendo and Kopeinig 2008 Rosenbaum and Rubin 1983 proposed a procedure that consists of calculating standard bias SB and comparing this bias before and after matching the results to verify whether the standardized bias decreased or not Caliendo and Kopeinig 2008 state that in most empirical studies an SB below 3 or 5 percent after matching is seen as sufficient To further perform a robustness check we used the test pseudoR2 proposed by Sianesi 2004 This test indicates how well the regressors X explain the participation probability Sianesi 2004 suggests reestimating the propensity score on the matched sample ie only on participants and matched nonparticipants and comparing the pseudoR2 s before and after determining the matches The value of pseudoR2 postmatching should be significantly smaller which indicates no systematic differences between the treated and nontreated groups Additionally we used the limits of Rosenbaum 2002 to test for the existence of a bias caused by the omission of a relevant variable hidden bias For this test we assume that the probability of participating in the program πi is not only determined by observable factors Xi but also by an unobservable component ui πi Prd 1Xi FβXi γui and γ is the effect of ui If the estimation is free of hidden bias the participation probability will solely be determined by Xi and γ will be zero however if there is hidden bias two individuals with the same observed covariates have differing chances of receiving treatment Caliendo and Kopeinig 2008 Since this paper sought to determine the impact of Bolsa Familia on child labor a missing variable could undermine the implications of the matching analysis Cash transfer program on child labor 213 42 Data We used the data set available from the National Household Sample Survey PNAD 2006 provided by the Brazilian Geography and Statistics Institute and evaluated the data in keeping with reweighted results from 2009 which took into consideration the new population projections designed by the IBGE in 2008 The data set we used was from 2006 since the 2006 data included IBGEs supplementary survey about participants access to monetary transfers and the social programs developed by the government and other institutions We limited the sample to families who had individuals aged between 0 and 17 years and who had a net income per capita excluding income from government transfers of up to R30000 US13953 the sample totaled 21886 individuals The delimitation of the sample with relation to income was also adopted by Araújo 2009 2010 and Araújo et al 2010 and corresponds to the families who are suitable to receive the Bolsa Familia monetary transfer and thus are families that could be characterized as having a similar level of economic vulnerability The variables qualifying participation in the Bolsa Familia program and the demographic characteristics used in the Logit model were those related to the characteristics of the children and their families These variables characterize specific social and economic conditions that could contribute to the increased or decreased selection probability for the program In particular we considered characteristics about the children age in years and a dummy for the color 1 if the child is black and 0 for the others We expected to find a positive relationship between the age and the probability of participating in the program and a negative relationship between the color of the child and hisher participation Additionally we used the parents characteristics and family characteristics For instance we used fathers schooling and mothers schooling in years we expected a negative relationship between these variables and participation in the program Furthermore we used the net income per capita from the government transfers to capture the CCT programs effect Additionally we used a few dummies such as for families headed by women 1 if yes and 0 if not families living in rural areas 1 for rural and 0 for urban and for families living in metropolitan regions 1 if yes and 0 otherwise We used two more variables to capture the household population number of children aged 0 to 5 years in the family and the number of children of 6 to 17 years Our labor response variable was the number of hours of child labor hoursinf We expected to obtain a positive relationship between the probability of participating in the program and families headed by women families living in rural areas and families with a larger number of children aged 0 to 5 years or aged 6 to 17 years On the other hand we expected a negative relationship between the probability of participating in the CCT program and the net income per capita from government transfers as well as families living in metropolitan areas We expected these effects based on the theoretical framework described previously and in the literature such as Araújo 2009 2010 and Araújo et al 2010 5 Results 51 Characteristics of the families in the sample First we analyze the characteristics of the persons who constitute the sample Among the group of people who earned income per capita of up to R30000 US13953 a month in 2006 479 percent of the sample stated they received the benefit or were part of a family that had a member who received the benefit The northeast region had the highest proportion of people who were beneficiaries at 613 percent followed by the northern region at 39 percent The average income per capita of these people excluding the income coming from government transfers was of R17830 The southern region presented the highest income per capita with R19670 while the northeast region presented the lowest income per capita of R16673 The difference in the income averages indicates that the greatest 214 IJSE 442 vulnerability manifested in the northeast region which concurs with the result that this region featured the highest destiny of government transfers The number of schoolage children proves to be another determining factor for those receiving the benefits since having schoolaged children is a precondition for participating in the program For the selected income sample the mean of children per family aged between 6 and 17 years was 13 children whereas in the northeast region the average was 14 children On the other hand the average number of children per family in the southeast region was only 11 Second we highlight some important characteristics between the beneficiary and non beneficiary groups using descriptive statistics The mean of the mothers education was 493 years for the beneficiary families while it was 666 years for the nonbeneficiary families Similar results were obtained for the fathers schooling the mean of the treatment group was 283 years while the mean for the control group was 330 years These results seem to provide common evidence families whose parents have a higher education level are less likely to be the target of income transfer programs because usually these families have higher incomes and are less vulnerable The results obtained for the treatment and non treatment groups with respect to other key variables are reproduced in Table I Families headed by females had a similar result for both groups which indicates that the choice to participate in the program does not seem to be influenced by the gender of the person of reference in the family The same conclusion appears in the variable concerning the individuals color which did not differ markedly On the other hand the proportion of individuals residing in rural areas while receiving benefits was higher than the proportion of rural families who did not receive the benefit inversely the percentage of people residing in metropolitan regions and belonging to the treated group was lower than for the control group This survey of the characteristics of the sample seems to indicate a positive relationship between receiving the benefit and living in a rural area as compared to a negative relationship with respect to living in an urban area Bolsa Familia program beneficiaries Bolsa Familia program nonbeneficiaries Proportion Linearized SE Proportion Linearized SE School attendance No 0278094 0005789 0413155 0007132 Yes 0721906 0005789 0586845 0007132 Female as family head No 0996171 0000652 0991181 0001067 Yes 0003829 0000652 0008819 0001067 Child labor No 0874664 0005168 0929466 0003992 Yes 0125336 0005168 0070534 0003992 Rural areas No 0541331 0015216 0753555 0011834 Yes 0458669 0015216 0246445 0011834 Metropolitan area No 0900318 0004281 0738282 0006692 Yes 0099682 0004281 0261718 0006692 Black No 0938962 0004859 0935138 0004386 Yes 0061038 0004859 0064863 0004386 Source Results of the work Table I Characteristics of the children and teenagers in Brazil with PR capital family income of up to R30000 in 2006 215 Cash transfer program on child labor In Table I we see that for children of up to 17 years old 7219 percent of the group that received monetary transfers attended school whereas only 5868 percent of the group of nonbeneficiaries attended school Additionally it is worthwhile to examine the comparison between beneficiaries and nonbeneficiaries with respect to the proportion of children of up to 17 years performing some sort of labor Table I shows that 1253 percent of the beneficiaries were involved in some kind of work whereas among the nonbeneficiaries only 705 percent worked 52 Empirical results To estimate the treatment effect we must first estimate the propensity scores and resurface the balancing tests to see if the samples of treated and untreated individuals can be correctly compared Thus we divide the empirical results into two parts The first subsection presents the estimation results of the propensity scores from the logit model and the balancing test The second subsection presents the results and discussions of the estimated treatment effect 521 Estimation of propensity score and balancing test As discussed in Section 41 a nonexperimental method matching process was utilized to find a control group with characteristics as close as possible to the treatment group to empower us to evaluate the impact of the Bolsa Familia CCT program on child labor First we compared the group based on the means of a Logit model since we did not know the propensity scores previously Araújo 2010 This model estimates the probability of a family taking part in the program or not and the results are present in Table II Table II allows us to identify that the fathers schooling the age of the child age the fact the family is headed by a female headfemale and the color of the family black are not relevant variables in determining the probability of whether the families would be beneficiaries Though these variables were not significant in determining participation in Bolsa Familia they are included in the analysis because they constitute important factors in the prediction of child labor Araújo 2009 On the other hand we found that school attendance increases the probability of a family participating in the program which was expected since school attendance of children between 6 and 17 years of age was one of the conditions for the family to receive the benefit The net income per capita excluding income obtained from the government transfer reduces the probability of participation in the program Araújo 2009 2010 and Araújo et al 2010 also obtained this result We expected this result since the income level is a precondition of the Bolsa Familia program and according to Araújo 2010 it indicates that the government program has Coefficient SE Z pvalor Confidence interval School Attendance 0419767 0037709 1113 0000000 0345858 0493675 Mothers schooling 0051696 0003991 129 0000000 005952 004388 Fathers schooling 0003636 0004091 089 0374000 001165 0004383 Age 0004092 0004002 102 0307000 001194 0003751 Number of children 1 0067546 0018778 360 0000000 0030742 0104351 Number of children 2 0308758 0012028 2567 0000000 0285183 0332333 Headfemale 0200266 0195703 102 0306000 058384 0183305 Rural 0483943 0034456 1405 0000000 0416410 0551477 Metropolis 0605106 0036018 168 0000000 067570 053451 Family income 0002654 0000185 143 0000000 000302 000229 Black 0037256 0059414 063 0531000 007919 0153704 Constant 0150244 0062205 242 0016000 0028323 0272164 Source Results of the work Table II Estimate of the Logit model for the group of families of Brazil with net income per capita of the transfers up to R30000 in 2006 216 IJSE 442 a wellspecified focus ie people with lower income levels receive government monetary transfers The mothers level of schooling and the fact that the family lives in a metropolitan area both reduce the probability of participation in the program Some of those results were also obtained by Araújo 2009 2010 and Araújo et al 2010 but such papers considered the family heads schooling while the present study opted for the distinction in schooling level between men and women We observed that only the females schooling had a significant impact on the probability of participation in the program A larger number of children from 0 to 5 years and from 6 to 17 years increased the probability of participation as did a familys rural location this latter fact was also found by Araújo 2009 and Araújo et al 2010 As discussed in 41 we analyzed the quality of the matching pairing We tested the validity of the conditional independence hypothesis by performing the standardized bias reduction analysis and the pseudoR2 test and the results are presented in Table III Table III shows that all variables provided a reduction in bias except for the number of children between 0 to 5 years in the family number of children 1 Additionally the comparison among the results of both the pseudoR2 test before and after the matching process indicate a significant reduction when matched which corroborates the fact that there are no systematic differences in the distributions of the variables between the control and treatment groups These results suggest that both groups are compatible We also used the verification of the existence of bias from an omitted variable to test the quality of the matching procedure This idea takes into consideration the reality that the probability of a family participating in the program is based on characteristics observable Mean Variable Sample Treated Untreated Reduction of the bias School attendance Unmatched 072306 058451 249 Matched 068853 058451 Mothers schooling Unmatched 501470 661740 275 Matched 545480 661740 Fathers schooling Unmatched 273150 323230 259 Matched 286100 323230 Age Unmatched 828990 710490 282 Matched 795580 710490 Number of children 1 Unmatched 088980 091385 378 Matched 088070 091385 Number of children 2 Unmatched 204320 120430 474 Matched 164540 120430 Headfemale Unmatched 000371 000888 132 Matched 000439 000888 Rural Unmatched 042447 022901 423 Matched 034172 022901 Metropolis Unmatched 016632 033243 183 Matched 019667 033243 Family income Unmatched 16987000 18510000 392 Matched 17584000 18510000 Black Unmatched 006378 006599 593 Matched 006509 006599 Summary PseudoR2 LR pvalue Unmatched 0104 3145660 000 Matched 0054 1507940 000 Source Results of the work Table III Standardized bias analysis for the sample of matched and unmatched individuals 217 Cash transfer program on child labor and not observable by the researcher if a nonobservable characteristic has an impact on the results there is a smaller chance that the Bolsa Familia CCT program has influence on child labor Araújo 2009 Thus we analyze this bias by implementing the method described above that uses the Rosenbaum limits The results are presented in Table IV Table IV shows for the six values of the Rosenbaum limits Γ the estimated coefficients were highly significant indicating that there is no bias resulting from an omitted variable in the model Thus on the basis of the results presented in Tables III and IV the model was welladjusted with a balanced distribution of independent variables in the two groups and validated by the absence of bias from an omitted relevant variable 522 Estimation of treatment effect Since the estimated propensity scores were validated by the previously applied test we could then evaluate the impact of the Bolsa Familia CCT program on child and teenager labor in Brazil We carried out this procedure as Araújo 2009 2010 by examining the effect of the treatment on the participating group using the nearest neighbor matching method The results of this algorithm are in Table V Table V allows us to conclude that contrary to what we expected the Bolsa Familia CTT program increased the number of hours of child labor in Brazil The incidence of the program increased the childrens and teenagers time allotment for labor by 525 percent We also found a higher incidence of child labor among families benefiting from the program compared to nonbeneficiary families Our finding that children whose families benefit from Bolsa Familia are more likely to allot time to labor was also observed in Cacciamali et al 2010 Araújo 2009 and Araújo et al 2010 Cacciamali et al 2010 found that the Bolsa Familia program increased the incidence of child labor in rural and urban areas Araújo 2009 noted that the program increased the proportion of children and teenagers who study and work and Araújo et al 2010 observed that the program is an important instrument for raising childrens school attendance but does not restrict childrens labor activities enough Our results are probably due to the characteristics of the sample as well as the objectives of the program A reasonable conclusion about these results is that due to the complementarity of descriptive analysis and this methodology the families that are more likely to receive the benefits are those who are also more likely to insert their children into labor market as pointed out by Ferro and Kassouf 2005 These authors also stated that Variable Γ Critical pvalue Child labor 1 00000 11 00000 12 00000 13 00000 14 00000 15 00000 Source Results of the work Table IV Sensitivity test of the effect of treatment by means of the Rosembaum limits Response variable Treatment Control Estimator of the ATT nearest neighbor Hoursinf 11869 5250 0525 Note Significant at 1 percent Source Results of the work Table V Effect of the average treatment upon the treated one for the variable hours of child labor hoursinf for Brazil in 2006 218 IJSE 442 since the public managers seek to provide for the most needy people children with a higher probability of working because of poverty will also be those who will be targeted by this program Furthermore many families who participate do not have any other alternatives than to insert their children into the job market since their household income is so low Cardoso and Souza 2004 argued that the transfer is too small a monetary motivation for a family to forego child labor earnings Ramalho and Mesquita 2013 corroborate this argument and complement it by affirming that most of the time child labor earnings are higher than the benefit Therefore although we found an unexpected result the Bolsa Familia program structure may not be formulated well enough to fight several determinants of child labor Attanasio et al 2010 pointed out that requiring school attendance does not inherently prevent child labor since education and work may not be perfect substitutes They also suggested that the parents may be reallocating the time from other activities such as leisure to education increasing school attendance and thus are not substituting the child labor earnings with the monetary transfer In response Ramalho and Mesquita 2013 suggest that child labor eradication policies should not be restricted to monetary transfer but rather should support investments into access to education and improving its quality and should generate economic opportunities Similarly Kassouf 2015 recognizes the importance of such policies CCT to supplement the income of underprivileged children and teenagers for their survival or the survival of the childrens family However she argues that child labor eradication policies should not be restricted to enforcing school attendance at elementary levels rather programs should assure that children complete their educational cycle including high school and professional courses In other words Kassouf 2015 advocates for policies that increase the quality of schooling that would actually achieve a reduction in child insertions into the job market In short it is necessary to modify the Bolsa Familia program andor to generate strategies to fight other determinants of child labor beyond poverty in order to effectively reduce child labor in Brazil In particular families should have better socioeconomic conditions which may favor the development of children and contributes to the eradication of this sort of labor 6 Summary and conclusion We sought to analyze the influence of the Bolsa Familia conditional cash transfer program on child labor in Brazil using a propensity score matching model and the 2006 data available from PNAD We limited the sample to families who have a net monthly income per capita excluding income from the government transfers of up to R30000 US13953 and who have children and teenagers under 18 years of age as pointed out in the literature Our sample revealed that the school attendance of children and teenagers the number of children in the household and the familys residence in rural areas increased the probability of a family being a beneficiary of this CCT program On the other hand mothers schooling the familys residence in a metropolitan region and a higher family net income per capita excluding income from the government transfer decreased this probability By comparing analogous participant and nonparticipant families across these characteristics we found that the Bolsa Familia program increased the number of hours a child spent at work Even though this program aims to reduce families poverty by focusing on those who present higher socioeconomic vulnerability it is expected that these families will seek to complement their income and assure a subsistence consumption level by inserting their children early into the job market Therefore to reduce the incidence of child labor we conclude that it is necessary to invest in policies that not only enact income transfers but 219 Cash transfer program on child labor that also generate job opportunities for adults while providing socioeconomic opportunities for child development Thus policies that combat determinants of child labor other than poverty will also be indispensable to reaching the longwishedfor objectives of reducing and even eradicating child labor in Brazil We faced two limitations with respect to the data set First IBGE only made available a survey about CCT programs on PNAD from 2006 and so it is not possible to evaluate the impacts of the program on child labor in other years though the conditions of the Bolsa Familia program have not changed Second since these data are part of a secondary data set the information about child labor could be too low since respondents selfdeclare and the law forbids this type of work We suggest for future research that these limitations should be taken in account However the empirical results found in this paper are robust and a first step toward a better understanding of this research topic Notes 1 This program is better explained on the next section 2 Value based on the minimum wage as of the survey date in 2006 which was R35000 US16279 3 Which is included in the Ministry of Social Development and Combat to Hunger Ministério do Desenvolvimento Social e Combate à Fome References Amarante V Ferrando M and Vigorito A 2011 School attendance child labor and cash transfers an impact evaluation of PANES Working Papers No 222 Poverty Economic Policy Research PIERI p 36 Araújo AA 2010 O programa bolsafamília e o trabalho infantil no Brasil doctoral thesis Federal University of Viçosa 129pp available at httpalexandriacpdufvbr8000teseseconomia20 aplicada2010231836fpdf accessed February 10 2014 Araújo GS 2009 Programa Bolsa Família e o trabalho de crianças e adolescentes Limites e alcances master dissertation Federal University of Uberlândia 163pp available at wwwppgeieufubrsites ppgeieufubrfilesAnexosBookpageIEdissertacaoGuilhermeSilvaAraujo0pdf accessed March 17 2014 Araújo GS Ribeiro R and Neder HD 2010 Impactos do programa bolsa família sobre o trabalho de crianças e adolescentes residentes na área urbana 38th Encontro Nacional de Economia ANPEC Salvador Attanasio O Fitzsimons E Gomez A Gutierrez MI Meghir C and Mesnard A 2010 Childrens schooling and work in the presence of a conditional cash transfer program in rural Colombia Economic Development and Cultural Change Vol 58 No 2 pp 181210 Becker GS 1965 A theory of allocation of time The Economic Journal Vol 75 No 299 pp 493517 Becker SO and Ichino A 2002 A estimation of average treatment effects based on propensity score The Stata Journal Vol 2 No 4 pp 358377 Cacciamali MC Tatei F and Batista NF 2010 Impactos do programa bolsa família federal sobre o trabalho infantil e a frequência escolar Revista de Economia Contemporânea Vol 14 No 2 pp 269301 Caliendo M and Kopeinig S 2008 Some practical guidance for the implementation of propensity score matching Journal of Economic Surveys Vol 22 No 1 pp 3172 Cardoso E and Souza AP 2004 The impact of cash transfers on child labor and school attendance in Brazil working papers Vanderbilt University Nashville TN April Edmonds EV and Schady N 2012 Poverty alleviation and child labor American Economic Journal Economic Policy Vol 4 No 4 pp 100124 220 IJSE 442 Ersado L 2002 Child labor and school decisions in urban and rural areas crosscountry evidence Discussion Paper No 145 International Food Policy Research Institute Washington DC December Ferro AR and Kassouf AL 2005 Avaliação do impacto do programa bolsaescola sobre o trabalho infantil no Brasil Pesquisa e Planejamento Econômico Vol 35 No 3 pp 417444 International Labour Office ILO 2013 Marking Progress Against Child Labour Global Estimates and Trends 20002012 International Programme on the Elimination of Child Labour IPECILO Geneva Kassouf AL 2015 Evolução do trabalho infantil no Brasil Sinais Sociais Vol 9 No 27 pp 945 PESQUISA NACIONAL POR AMOSTRA DE DOMICÍLIOS PNAD 2006 Microdados available at wwwibgegovbrhomeestatisticapopulacaotrabalhoerendimentopnad2006 accessed June 1 2014 Ramalho HMB and Mesquita SP 2013 Determinantes do trabalho infantil no Brasil Urbano Uma análise por dados em painel 20012009 Economia Aplicada Vol 17 No 2 pp 193225 Ravallion M and Wodon Q 2000 Does child labour displace schooling Evidence on behavioural responses to an enrollment subsidy The Economic Journal Vol 110 No 462 pp C158C175 Rosenbaum PR 2002 Covariance adjustment in randomized experiments and observational studies Statistical Science Vol 17 No 3 pp 286304 Rosenbaum PR and Rubin DB 1983 The central role of propensity score in observational studies for causal effects Biometrika Vol 70 No 1 pp 4155 Secretaria Nacional De Renda De Cidadania Senarc 2014 Prestação de contas ordinárias anual Relatório de gestão 2013 Brasília March available at wwwmdsgovbrbolsafamiliaarquivos RELATPC3P93RIO P20DEP20GESTPC3P83OP202013pdfpagespeedce5cmS05YHYNpdf accessed June 5 2014 Sianesi IC 2004 An evaluation of the active labour market programmes in Sweden The Review of Economics and Statistics Vol 1 No 86 pp 133155 Skoufias E and Parker SW 2001 Conditional cash transfers and their impact on child work and schooling evidence from the PROGRESA program in Mexico Economía Vol 2 No 1 pp 4596 About the authors Paloma Santana Moreira Pais is a Professor at the Administration and Economics Department Federal University of Lavras and DSc in Applied Economics at the Federal University of Viçosa and MSc in Applied Economics at same department Paloma Santana Moreira Pais is the corresponding author and can be contacted at palomapaishotmailcom Felipe de Figueiredo Silva is a PhD Student in Agricultural Economics at the University of NebraskaLincoln and a Doctoral Student and in Applied Economics at the Federal University of Viçosa MSc in Applied Economics at same department Evandro Camargos Teixeira is a Professor at the Economics Department Federal University of Viçosa DSc in Applied Economics at the University of São Paulo MSc in Economics from the Federal University of Paraná For instructions on how to order reprints of this article please visit our website wwwemeraldgrouppublishingcomlicensingreprintshtm Or contact us for further details permissionsemeraldinsightcom 221 Cash transfer program on child labor O trabalho infantil é um assunto elefante para pautas de políticas públicas e sociais entre os diversos motivos abordados pelos pesquisadores a pobreza é o mais citado ou seja crianças em situações de vulnerabilidade social tende a começar a trabalhar mais cedo para poder ajudar a família por esse motivo há programas sociais com a finalidade de diminuir o trabalho infantil e tentar garantir que a criança tenha direitos básicos como educação assistência social e lazer Sendo assim Pais Silva e Teixeira 2017 baseiase na teoria sobre alocação de tempo proposta por Becker 1965 com adaptações de Ersado 2002 Considerando que o tempo de uma família é distribuído entre emprego lazer e educação e as decisões de alocação de tempo das crianças levam em consideração os retornos privados de cada atividade Para desenvolver o modelo proposto será levado como base no modelo neoclássico onde a divisão de tempo de uma família para trabalho educação e lazer é feita de maneira conjunta Segundo Ersado 2002 a decisão sobre a participação das crianças na força de trabalho atividades de lazer ou a escola é determinada por uma pessoa adulta Além disso Ersado sugeriu que uma família composta por um adulto e uma criança maximiza sua função de utilidade em dois períodos t e t1 de acordo com a seguinte Equação 1 V tU CtLp t Lc tSc t Xt Onde U é uma função de utilidade conchavada que depende das variáveis CtConsumo do tempo t LptLazer do adulto do tempo t Lc tLazer da criança no tempo t Sc tEscolaridade da criança n tempo t Xtcaracterísticas individuais e familiares Suponhamos que o adulto tenho duas opções Primeira mandar a criança trabalhar assim num primeiro momento t essa criança ganhará um salario W c e posteriormente t1 seu salário será de W u que representa o salário de um adulto não qualificado Segunda mandar a criança para a escola então em t a criança não ganhará nenhum salário e depois em t1 ela ganhará o salário W s que representa o salário de um adulto qualificado Por definição podemos supor que W cW uW s Assim na vida adulta o investimento em lazer vai depender do salário W u e W s Assim o chefe da família deve maximizar a função de utilidade apresentada acima sujeita à disponibilidade de tempo e às restrições de recursos em cada período de acordo com a Equação 2 CtW t Lp tLc tSc t ΩtW tT Onde W té um vetor de salários de adultos e crianças T é o tempo disponível para família Ct é o consumo Ωt é remunerações que não são proveniente do trabalho Podese especificar Ωt segundo a Equação 3 ΩtΠ tδ AtY t Onde Π t lucro proveniente de trabalho autônomo extras Atrenda proveniente dos interesses patrimoniais da família Y toutras fontes de receitas que não são derivadas do trabalho δ taxa de juros O adulto também considerará a restrição de tempo da criança uma vez que deve ser dividida entre escola trabalho e lazer além do trabalho doméstico não remunerado seguinte Equação 4 T cLc tSc tEc t Ec té o tempo gasto em trabalhos pagos ou não pagos A trajetória temporal de os bens da família pode ser definida pela seguinte Equação 5 At11δ AtΠ tY tW t Ec tW t T pLptCt Onde Até o total de ativos no período anterior Π tY tW t Ec tW t T pLp tCt é a poupança ou despoupança se negativa no período t após o consumo desse período Resolvendo Ωt de acordo com as Equações 3 e 5 temse a Equação 6 Ωt At1AtCtW t Ec tW t T pLp tΔ At onde a Equação 6 consiste em uma medida intertemporal da renda não salarial que permite os agentes para salvar ou não Seguese que o chefe de família maximiza o bemestar doméstico Equação 1 sujeito as restrições de tempo da criança Equação 4 e renda não salarial Equação 6 A solução para o problema de maximização é uma função que depende das variáveis preços taxa de salário para a criança não qualificado e mão de obra qualificada posse de bens domésticos renda não salarial e outros fatores como características incluídas no vetor Xt conforme apresentado na seguinte equação 7 Γ W t Π t AtY t Xt Ψ Ψ é referente a todas características observadas e não observadas que afetam a decisões A função de utilidade indireta é obtida substituindo o vetor de escolha Equação 6 na função de utilidade Equação 1 para definir a utilidade máxima da família apresentado na Equação 8 VU Γ W tΠ t AtY t XtΨ Assim podemos considerar que os ganhos futuros de um adulto vão depender se quando criança o adulto responsável tomou a decisão de mandar a criança para o trabalho ou para estudar Considerando há hipótese econômica do modelo onde a criança pode estudar no período t ao chegar na vida adulta período t1 terá um salário maior do que o adulto que quando criança período t foi mandado a trabalhar Sabese que no Brasil existe um alto índice de desigualdade social com a finalidade de diminuir esse índice de desigualdade foi criado o Bolsa Família que é um programa de CCT ou seja um programa de transferência condicional de renda que tem como premissas ajudar financeiramente famílias de baixa renda com criança em idade escolar uma das condições estabelecidas no bolsa família é que as crianças em idade escolar devem frequentar as escolas O objetivo o programa é garantir ajuda financeira para que as crianças não precisem deixar de estudar para poder trabalhar E futuramente quando for adulto a expectativa de salário seja maior Concluindo que o presente artigo analisado tem como hipótese econômica que os filhos de famílias assistidas pelo programa de Bolsa Família terá um desempenho econômico maior salários mais altos do que os filhos de famílias com vulnerabilidade econômica que não são assistindo pelo programa Bolsa Família O artigo traz como estratégia uma abordagem que tem como objetivo deixa o sistema de saúde mais eficiente para isso o foco da implementação do novo sistema é estabelecer melhorias aos cuidados primários e facilitar o acesso esperase que com cuidados primários mais adequados será possível reduzir as principais causas de mortes e distúrbios implicando em uma prevenção a saúde e diagnóstico precoce As mudanças realizadas em 2006 levou a criação de modelos de prestação de cuidados primários Para fortalecer os cuidados primários foi criado a Unidades de Saúde Familiar USF que consistia em equipes multidisciplinares voluntariamente constituídas de em média 20 profissionais de saúde GPs enfermeiros e técnicos gozando de autonomia funcional e técnica Também foi estabelecido que todo paciente coberto pela USFs tem direito a um médico família que estabelece cuidados a longo prazo e acompanhamento periódico ao paciente A tabela 2 apresenta as estimativas do parâmetro da Equação 1 que estipula uma regressão Onde a taxa de internação por 1000 habitantes Y mrt depende das variáveis de implementação das USF Ano do paciente município horário da internação e por fim as características de cada munícipio Pela tabela temos que o R2 do modelo considerando a amostra observada foi de 0772 ou seja 772 da variabilidade da taxa de internação por 1000 habitantes é explicada pelo modelo ao considerar o efeito tendência no modelo o R2 aumenta para 0854 ou seja podemos considerar que o passado influência nos dados Na Figura 1 pode se perceber que a implementação nas USFs fez com que a série histórica da taxa de intenção diminuiu no decorrer do tempo Podese perceber na figura 1b que ao tirar o efeito de tendência e sazonalidade a taxa teve um declínio ainda maior I Na Figura 2 mostram que os períodos de préimplementação do estudo do evento são estatisticamente significativos sugerindo a presença de tendências diferenciais pré USFs Percebase que na Figura 2 b que ao se considerar o efeito de sazonalidade na regressão tem uma diminuição média do erro das estimativas de número de internações Isso implica que o efeito da implantação das USFs implicou em uma diminuição na taxa de internação por CSAP na Equação 1 é parcialmente explicada pela existência de tendências préUSF não paralelas Os resultados da Equação 3 nos mostram que a suposição de tendência paralela é atendida após a inclusão de tendências temporais lineares específicas do município A Figura 3 traz os níveis de internação para pacientes que possuem comorbidades específica e paciente que são e não são incentivados a ter um tratamento com acompanhamento médico Percebe pelo gráfico de sazonalidade que quando não há incentivo prevê uma sazonalização maior dos dados em relação aos pacientes que são incentivados a ter acompanhamento periódico Na Figura 4 tem um resultado semelhando da Figura 3 e mostra que quando há incentivo no tratamento primário com a instalação das USFs diminui o efeito de sazonalidade diminuindo o valor do DiD no decorrer do tempo em relação as internações realizadas O programa Red de Proteccion Social RPS do Governo da Nicarágua representa uma abordagem a onde são oferecidos reder de segurança para pessoas com vulnerabilidade econômica ou seja é realizado uma Transferência Condicional de Renda CCT Que estabelece que crianças em idade escolar possa estudar ao invés de trabalhar assim quando chegar na vida adulta terá ganhos econômicos maiores O programa de RPS tem a finalidade de diminuir a desigualdade econômica e social do país assim obtendo mais mão de obra qualificada e diminuindo o nível de pobreza do país essasa medidas resulta em um aumento a segurança dos municípios e o desenvolvimento do lugar Como a Nicarágua estabeleceu suas diretrizes do programa de CCT Primeiro foi adotado uma abordagem sistemática para garantir uma rede se segurança de renda básica para os pobres em vez de depender de processos mais difusos e indiretos de redução da pobreza por exemplo por meio de programas de infraestrutura orientados pela demanda e segundo eles pretendem fazer mais do que apenas colocar dinheiro de curto prazo nas mãos das pessoas por mais que a primeiro momento isso seja necessário para as medidas de CCT mas também tem o objetivo de investir no capital humano de longo prazo Essas medidas são alcançadas ao condicionar as transferências de renda à participação das famílias nos serviço de saúde e educação com base nas premissa de que a atenção à saúde nutrição e educação de primeira infância aumenta significativamente as chances dessa criança sair da pobreza mais tarde na vida Ajudar financeiramente família de vulnerabilidade Melhoria na educação saúde primária e na nutrição das crianças mão de obra capacitada diminuição da pobreza e desigualdade social desenvolvimento econômico do país