The effects of non-contributory pensions on material and subjective well being

Bando Grana, Rosangela; Galiani, Sebastián; Gertler, Paul J. Working Paper The effects of non-contributory pensions on material and subjective well being IDB Working Paper Series, No. IDB-WP-840 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Bando Grana, Rosangela; Galiani, Sebastián; Gertler, Paul J. (2017) : The effects of non-contributory pensions on material and subjective well being, IDB Working Paper Series, No. IDB-WP-840, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0000850 This Version is available at: https://hdl.handle.net/10419/173894 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/3.0/igo/legalcode The Effects of Non-contributory Pensions on Material and Subjective Well Being Rosangela Bando Sebastián Galiani Paul Gertler IDB WORKING PAPER SERIES Nº IDB-WP-840 September 2017 Office of Strategic Planning and Development Effectiveness Inter-American Development Bank September 2017 The Effects of Non-contributory Pensions on Material and Subjective Well Being Rosangela Bando Sebastián Galiani Paul Gertler Inter-American Development Bank University of Maryland and NBER University of California at Berkeley and NBER Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Bando, Rosangela. The effects of non-contributory pensions on material and subjective well being / Rosangela Bando, Sebastián Galiani, Paul Gertler. p. cm. — (IDB Working Paper Series ; 840) Includes bibliographic references. 1. Income maintenance programs-Peru. 2.Poverty-Peru. 3. Well-being-Peru. I. Galiani, Sebastián. II. Gertler, Paul. III. Inter-American Development Bank. Office of Strategic Planning & Development Effectiveness. IV. Title. V. Series. IDB-WP-840 1300 New York Ave, NW Washington, D.C. 20577 Rosangela Bando | [email protected] Copyright © Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 AttributionNonCommercial-NoDerivatives (CC-IGO BY-NC-ND 3.0 IGO) license (http://creativecommons.org/licenses/by-nc-nd/3.0/igo/ legalcode) and may be reproduced with attribution to the IDB and for any non-commercial purpose, as provided below. No derivative work is allowed. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the UNCITRAL rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this CC-IGO license. Following a peer review process, and with previous written consent by the Inter-American Development Bank (IDB), a revised version of this work may also be reproduced in any academic journal, including those indexed by the American Economic Association's EconLit, provided that the IDB is credited and that the author(s) receive no income from the publication. Therefore, the restriction to receive income from such publication shall only extend to the publication's author(s). With regard to such restriction, in case of any inconsistency between the Creative Commons IGO 3.0 Attribution-NonCommercial-NoDerivatives license and these statements, the latter shall prevail. Note that link provided above includes additional terms and conditions of the license. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. http://www.iadb.org 2017 1 The Effects of Non-Contributory Pensions on Material and Subjective Well Being Rosangela Bando Inter-American Development Bank Sebastian Galiani University of Maryland and NBER Paul Gertler University of California at Berkeley and NBER September, 2017 Abstract: Public expenditures on non-contributory pensions are equivalent to at least 1 percent of GDP in several countries in Latin America and is expected to increase. We explore the effect of non-contributory pensions on the well-being of the beneficiary population by studying the Pension 65 program in Peru, which uses a poverty eligibility threshold. We find that the program reduced the average score of beneficiaries on the Geriatric Depression Scale by nine percent and reduced the proportion of older adults doing paid work by four percentage points. Moreover, households with a beneficiary increased their level of consumption by 40 percent. All these effects are consistent with the findings of Galiani, Gertler and Bando (2016) in their study on a non-contributory pension scheme in Mexico. Thus, we conclude that the effects of non-contributory pensions on well-being in rural Mexico can be largely generalized to Peru. Acknowledgements: The authors thank Ada Kwan, Juan Manuel Hernandez and Dylan Ramshaw for providing key inputs and advice for this work, and acknowledge a grant from GTZ for financial support. The authors declare that they have no financial or material interests in the results of this study. Keywords: Non-contributory pensions, poverty, mental health and well being. JEL Codes: H4, H3, I1 and I3. 2 I. Introduction While pensions are believed to be critical for protecting material well-being after retirement, only 20 percent of seniors worldwide receive pension benefits (Pallares-Miralles, Romero and Whitehouse, 2012). For those who have coverage, the benefits are often inadequate (ILO, 2014; Gasparini et al., 2007). Additionally, poverty rates among the elderly are substantially higher in countries where social security coverage is limited; the number of people who are 60 years of age or older is estimated to double by 2050 (United Nations, 2013); and the life expectancy of the elderly is also estimated to substantially increase by 2050 (Bosch, Melguizo and Pagés 2013). For these reasons, improving the effectiveness of pensions and expanding pension programs compel immediate attention. A number of governments have responded to high poverty rates among the elderly with noncontributory pensions. In OECD countries, 59 percent of the income of individuals over age 65 comes from public pension transfers (OECD, 2015). In Latin America, at least 15 countries have implemented non-contributory pension programs covering about 20 percent of the region’s population (Bosch, Melguizo and Pagés, 2013; Pallares-Miralles, Romero and Whitehouse, 2012). In Latin America, these programs constitute a large part of social safety nets. For example, in Mexico, the Adultos Mayores program is the second largest social program behind the conditional cash transfer program Progresa (formerly Oportunidades), and in Peru, Pension 65, a noncontributory pension program for the elderly, is second only to the conditional cash transfer program Juntos (Rubio and Garfias, 2010; Aguila et al., 2013, MIDIS, 2012). In this paper, we explore the effects of Pension 65 in Peru. The program’s main goal is to provide economic security to persons who are 65 years of age or older and living in poverty (Presidencia del Consejo de Ministros, 2011). At the time this study was conducted, the program provided beneficiaries with US$ 78 every two months. This study makes use of a strong identification strategy by exploiting an exogenous poverty cutoff to determine eligibility. As a result, we are able to analyze household survey data using a sharp regression discontinuity approach. We estimated effects for a sample of households that is within 0.3 standard deviations of the threshold. As a result, program participation was statistically ignorable in the neighborhood that we studied. We find that households with a beneficiary increased their level of consumption by 40 percent and that the program reduced the proportion of older adults doing paid work by 4 percentage points. These effects contributed to their subjective welfare as indicated by a 9-percentage-point reduction in the Geriatric Depression Scale. However, we do not find impacts on the use of health services, physical health outcomes, enrollment of minors in school or household composition. However, we find that transfers to persons residing outside the household increased as the 3 proportion of households that reported expenditures on transfers rose from 46 percent to 61 percent. Several studies have focused on the effects of non-contributory pension schemes on the health and material welfare of beneficiaries. Some examine the effects of such schemes on consumption (Fan, 2010; Blau, 2008; Case and Deaton, 1998), physical health (Kadir and Barret, 2014), and labor supply (de Carvalho, 2008; Bosch, Melguizo and Pagés, 2013; Grueber and Wise, 1998). Other studies have analyzed the effects of pensions on other family members. For example, Case and Deaton (1998), Duflo (2003), Hamoudi and Thomas (2014) and Fan (2010) explore program effects on children’s school enrollment, household composition and private transfers. Our work is also related to the work of Finkelstein et al. (2012) and Baicker et al. (2013) who find access to Medicaid health insurance lowered self-reported depression in low-income adults. Indeed, the literature shows unemployment results in more depression because of the lack of work, but also in less depression as people can spend more time in pleasant activities (Knabe et al., 2010; Krueger and Muller, 2012; and Ruhm, 2001). In contrast, in previous work, we took a comprehensive approach in examining the influence of Mexico’s non-contributory pension schemes of Adultos Mayores on both material and subjective well-being (Galiani, Gertler and Bando, 2016). Indeed, pensions may allow older adults to reduce their time working and increase their time enjoying life. We found that beneficiaries used part of the transfer to finance an increase in household consumption and used the rest to offset reduction in labor earnings from beneficiaries reducing paid work. These changes resulted in an improvement in mental health as measured by the Geriatric Depression Scale. 1 When we compare the results in this paper with the effects of the Adultos Mayores program in Mexico, we find that we can broadly generalize the estimates for Mexico to Peru. We find that the effects of the programs are not that different across the two countries. The depression score in Peru decreased by 8.68 percent, while it decreased in Mexico by 9.11 percent. Paid work decreased by 4 percentage points in both countries. In addition, consumption rose by 40 percent in Peru and by 14 percent in Mexico. For food consumption, households in Peru allocated 67 percent of the increase, while in Mexico, they allocated 54 percent. This study is important in that it constructs external validity of the effects of non-contributory pensions, since in principle, the effects of any program are contingent on the context of the study (Angrist, J., 2004; Campbell, 1969; Fisher, 1935). Understanding program effects in multiple 1 Mental health is a widely accepted indicator of quality of life among the elderly (Campbell et al., 1976; Walker, 2005). 4 economic and cultural contexts is necessary in order to construct external validity and inform policy. A number of studies use similar multi-country strategies to generalize cause-and-effect constructs. For example, Cruces and Galiani (2007) examine the effects of fertility on labor outcomes in three counties, Banerjee et al. (2015) study microcredit in six countries, Gertler et al. (2015) study health promotion in four countries, Dupas et al. (2016) examine the effects of opening savings accounts in 3 countries, and Galiani et al. (2016) investigate slum upgrading in three countries,. This paper is organized as follows. Section II describes the Pension 65 program. Section III describes the data, and section IV describes the identification strategy. Section V presents the empirical results. Section VI compares our findings with the results obtained in Mexico. Section VII concludes. II. The Pension 65 Non-Contributory Pension Program The program provides beneficiaries with a pension of US$ 39 per month, which is paid out in bimonthly transfers (Presidencia del Consejo de Ministros, 2011). In addition, beneficiaries receive care in public health facilities at no cost and are eligible for the Integral Health Insurance Program (Seguro Integral de Salud (SIS)) (MIDIS, 2016). The program significantly increased the number of pension beneficiaries in 2013 as coverage expanded from 40,700 to 247,700 beneficiaries between January and November of that year. To be eligible, a person has to be at least 65 years old, possess a government-issued identification document that attests to his or her age and be certified as living in a household that is below the poverty line. Persons who receive benefits from other pension programs are not eligible. The government defines poverty based on its Household Targeting System (Sistema de Focalización de Hogares (SISFOH)) index. A person’s SISFOH index score is a weighted average of a number of household characteristics. 2 A household is classified as poor if its score falls below a set threshold value. Government-defined poverty thresholds are set for geographic areas known as “conglomerates” (conglomerados). The SISFOH index is used universally for targeting all government programs, including the Pension 65 program, and the data used to construct the SISFOH index were collected long before the Pension 65 program established the 2 These characteristics include the type of fuel used for cooking; electricity; water and sewerage access; the materials that the floor, walls and roof are made of; health insurance and assets. Assets include refrigerators, washing machines, laptops, and cable and Internet connections. They also include the level of education of the head of household and the extent of overcrowding. 5 eligibility threshold. The Ministry of Economic Affairs and Finance (MEF, 2010) provides details on the estimation of the SISFOH scores and poverty thresholds. III. Data Sources The data used in this study come from two surveys carried out by the National Institute of Statistics and Informatics (Instituto Nacional de Estadística e Informática or INEI). The sampling frame was restricted to the 12 out of 24 departments in Peru in which 70 percent of program beneficiaries resided as based on administrative records. 3 Households were then randomly sampled based on the following eligibility criteria: having at least one adult between the ages of 65 and 80, whose available SISFOH information could determine household poverty status and whose SISFOH score(s) were 0.3 standard deviations above and below the SISFOH eligibility threshold. 4 There were two rounds of data collection. The first round was conducted in November and December of 2012, and the second round, in the period from July to October of 2015. In the first round, data were collected on 4,031 individuals in 3,031 households. INEI excluded 58 households that had errors in their eligibility score in the SISFOH system from the second round. Of the 2,973 remaining households, 234 were not found and therefore lost to attrition. We further excluded another 155 households from the analysis whose SISFOH scores at baseline were more than .3 standard deviations from the eligibility cutoff. Excluding these observations allows us to reduce the average distance of the SISFOH score from the eligibility threshold by 52 percent. 5 All in all, excluding all of these households did not likely affect our results as treatment status is uncorrelated with exclusion status (p-value = 0.559), and the baseline characteristics of the excluded households are not statistically different from those included in the sample (Table A1 in Appendix A). In summary, the analysis sample used in this study consists of 3,342 individuals living in 2,584 households. The survey questions were designed to collect detailed information on the older adults and their households, as well as basic information on all other household members. More specifically, the survey collected labor information for persons 14 years of age or older. This information included labor market participation, hours worked and monetary compensation. Anthropometric 3 Amazonas, Ancash, Cajamarca, Cusco, Hunuco, Junin, La Libertad, the provinces of the Lima Region (Cajatambo, Canta, Huarochiri, Oyón and Yauyos), Loreto, Pasco, Piura and Puno. 4 For a detailed description of the selection of the sample, see the Ministry of Development and Social Inclusion (MIDIS y MEF, 2013). The INEI monitored actual transfers from January 2012 to June 2015, and the data can therefore be used to check for actual transfer reception. 5 The score distance from the eligibility threshold in the final sample is between -0.32 and 0.31. If we were to include the 155 observations that were located in the tail of the distribution, the score would take on values of between -0.46 and 0.86. 12 b. Health and Well-Being Table 4 shows the results for health and well-being. The values of program estimates given in Column (2) for Panel A show that physical health was not affected. More specifically, hypertension, waist circumference, BMI and memory scores were not altered by the program. Consistent with this, older adults did not feel that their health had improved or that they were having less difficulty than before in performing daily activities. The physical health scores confirm this. Table 4 Panel B, which focuses on subjective well being, shows a different story. The program reduced the older adults’ score on the Geriatric Depression Scale by 8.68 percent (from 0.43 to 0.39). In addition, the contribution-to-household expenditures score increased by 12.92 percent (from 0.83 to 0.94), and the self-worth score rose by 6.54 percent (from 0.57 to 0.61). However, the program did not affect the satisfaction score, which remained at 0.74, or the empowerment score, which remained at 0.88. The overall well-being score, shown in the last row of Panel B, indicates that the program led to an increase in well-being equivalent to 0.17 standard deviations. As the program made beneficiaries eligible for the public Integral Health Insurance Program (Seguro Integral de Salud (SIS)), we find that the share of older adults affiliated with this insurance program increased by 12 percent (from 79 percent to 89 percent). However, we find no effects on the use of health services. Table 5 reports estimates of program effects on health perception, insurance and health services. c. Household Income and Consumption Table 6 reports impact estimates for household labor income and consumption expenditures, with income and expenditures being presented in US dollars (US$) and in terms of adult equivalents. Column (2) shows that the program did not affect total household labor income. Indeed, total labor income remained at US$ 38.46. The program did not affect the labor income of older adults either, which remained constant at US$ 25.94. However, the program increased household expenditure by 39.73 percent (from US$ 45.16 to US$ 63.11). Older adults allocated 67 percent of their expenditure to food consumption and 33 percent to non-food consumption. 13 Table 4. Impact on health and well-being Mean in control group RD with conglomerate fixed effects RD with conglomerate fixed effects and controls Adjusted p-values (1) (2) (3) (4) Panel A. Physical health Hypertension 0.44 -0.07 -0.07 0.124 (0.03)* (0.03)* Waist circumference 89.01 -0.52 -0.79 0.654 (1.29) (1.38) BMI 23.31 -0.10 -0.06 0.527 (0.14) (0.13) Memory 11.25 -0.07 -0.11 0.661 (0.25) (0.24) Physical Health Index 0.00 -0.03 -0.03 (0.05) (0.05) Panel B. Subjective Well-being Depression symptoms index 0.43 -0.04 -0.04 0.124 (0.02)* (0.02)* Satisfaction with quality of life 0.74 0.00 0.00 0.767 (0.02) (0.02) Empowerment 0.88 0.03 0.03 0.196 (0.02) (0.02) Contribution 0.83 0.11 0.11 0.003 (0.02)*** (0.02)*** Self-worth 0.57 0.04 0.04 0.101 (0.02)** (0.01)*** Subjective well-being index 0.00 0.17 0.17 (0.04)*** (0.03)*** Source: Authors’ calculations. Note: Based on 3,342 observations. Standard errors, clustered at the conglomerate level, are shown in parentheses. Coefficients as percentages of the mean in the control group are shown in brackets. Controls include each individual's age, sex, marital status and years of schooling. Pvalues adjusted according to Anderson (2008) for the family of outcomes listed in the table. 14 Table 5. Impact on individuals’ health perceptions, health insurance and use of health services Mean for control group RD with conglomerate fixed effects RD with conglomerate fixed effects and controls Adjusted p-values (1) (2) (3) (4) Panel A. Health perception Perception of good or very good health (1 if yes, 0 otherwise) 0.58 0.01 0.00 0.571 (0.04) (0.04) [1.35%] [-0.06%] Perception of difficulty performing daily activities (1 if yes, 0 otherwise) 0.44 -0.04 -0.04 0.440 (0.04) (0.04) [-9.59%] [-9.41%] Panel B. Health insurance Health insurance (1 if insured, 0 otherwise) 0.79 0.10 0.09 0.191 (0.04)** (0.04)** [12.31%] [11.95%] Panel C. Use of health services In the previous month had primary care visit 0.32 0.05 0.05 0.355 (0.03) (0.03)* [15.83%] [16.26%] In the previous month had visit, medication or exam 0.52 0.08 0.08 0.381 (0.05) (0.05) [14.55%] [14.66%] In the previous 3 months had dental, ophthalmological or optometric care or vaccination 0.23 0.06 0.06 0.355 (0.04) (0.04) [27.45%] [23.77%] In the previous 12 months was hospitalized or had surgery 0.06 0.01 0.01 0.475 (0.02) (0.02) [21.42%] [21.26%] Source: Authors' calculations. Note: Based on 3,342 observations. Standard errors, clustered at the conglomerate level, are shown in parentheses. Coefficients as percentages of the mean in the control group are shown in brackets. Controls include an individual's age, sex, marital status and years of schooling. Pvalues adjusted for type I error in multiple hypothesis testing by Anderson (2008). 15 To get a sense of how these changes relate to the pension transfers, consider the following. The program transferred US$ 39 (125 Peruvian Soles (S$)) per month per person. Considering that the average household size is 2.84, and additionally that, on average, the sample includes 1.29 older adults per household. Therefore, the average transfer per adult equivalent to each household was US$ 39*1.29/2.84 = US$ 17.71. This amount is not statistically different from the increase in consumption (p=0.948). Consistent with this, we find household consumption changes in line with the total transfer. In other words, households with two older adults increase consumption twice as much as households with one older adult. Appendix C shows estimates by the number of older adults in the household. Table 6. Impact on household income and expenditures Mean in control group RD with conglomerat e fixed effects RD with conglomerate fixed effects and controls Adjusted p-values (1) (2) (3) (4) Labor income per adult equivalent (AE) 38.46 4.24 4.99 0.262 (6.37) (6.73) [11.02%] [12.97%] Labor income per AE excluding older adult 25.94 4.87 6.16 0.262 (6.62) (6.46) [18.77%] [23.75%] Household expenditure per AE 45.16 17.94 18.05 0.012 (4.63)*** (3.94)*** [39.73%] [39.97%] Household food expenditure per AE 31.68 12.03 12.16 0.012 (3.68)*** (3.21)*** [37.99%] [38.38%] Household non-food expenditure per AE 13.49 5.91 5.89 0.012 (1.77)*** (1.97)** [43.81%] [43.71%] Source: Authors’ calculations. Note: Based on 2,584 observations. Standard errors, clustered at the conglomerate level, are shown in parentheses. Coefficients as percentages of the mean in the control group are shown in brackets. Controls include age, marital status, sex and education of the head of household. Pvalues adjusted according to Anderson (2008) for the family of outcomes listed in the table. 16 d. Benefits to Other Family Members and Transfers Increases in household consumption may benefit other household members, in addition to the older adults. Thus, we seek to determine if pension transfers affected school enrollment, where we define enrollment as the percentage of household members who are 3 to 15 years old and enrolled in an educational institution. Table 7 in Panel A shows the results of this analysis. No effects were found. We then look at whether pensions influence living arrangements. As may be seen from the same panel, we do not find any effects on household size. Next, we try to determine if transfers at the older-adult and/or household-level change. Panel B shows impact estimates for current transfers at the household level. The share of households with individuals who reported having received a transfer in the previous six months decreases from 51 percent to 43 percent. However, column (4) shows this effect is not statistically significant when adjusting for multiple testing. We find no impact when transfers to older adults are excluded. We also find the share of private transfers sent increased from 46 to 62 percent. We therefore conclude that the receipt of non-contributory pensions did not affect children’s school enrollment or household composition. These results differ from those of Duflo (2003) and Hamoudi and Thomas (2014), who find that the receipt of pensions did influence these two variables. We do not find evidence that the receipt of these pensions leads to a decrease in transfers either. Our results for this variable therefore differ from those of Fan (2010), who finds that pension transfers translate into decreases in private transfers to the elderly equivalent to 39 cents for every pension dollar. In contrast, the receipt of a pension is likely to benefit family members who reside elsewhere. d. Robustness Tests In this section, we discuss the sensitivity of our results to alternative specifications. In summary, our findings are robust. First, we compare the results just discussed with those obtained with the inclusion of controls. In the empirical section, we also report estimates while also conditioning on a set of observable control variables. Nevertheless, we expect local estimation to replicate the conditions of a local experiment. If so, the introduction of controls should not affect our point estimates previously reported. However, their introduction may increase the efficiency of the estimator of the parameter of interest. 17 Table 7. Impact on benefits to other household members and transfers Mean in control group RD with conglomerate fixed effects RD with conglomerat e fixed effects and controls Adjuste d pvalues (1) (2) (3) (4) Panel A. Benefits for other household members % HH members age 3 to 15 enrolled in school† 0.81 -0.05 -0.02 0.951 (0.06) (0.06) Household size per adult equivalent 2.84 0.04 0.74 1.000 (0.24) (0.2) Panel B. Transfer to and from household Received private transfer in last 6 months 0.51 -0.08 -0.09 0.249 (0.04)* (0.03)** Received private transfer excluding older adult 0.39 -0.04 -0.06 0.951 (0.07) (0.07) Sent private transfer in last 3 months 0.46 0.15 0.16 0.010 (0.05)*** (0.05)*** Panel C. Transfer to and from older adult Transfers received (US$) 15.81 -0.25 -2.18 1.000 (5.4) (4.53) Transfers sent (US$) 2.98 -2.00 -1.76 0.924 (2.07) (2.1) Received transfer 0.44 -0.07 -0.08 0.735 (0.06) (0.05) Sent transfer 0.06 0.03 0.03 0.596 (0.02) (0.02) Source: Authors’ calculations. Note: Panels A and B based on 2,584 observations. Panel C based on 3,342 observations. Standard errors, clustered at the conglomerate level, are shown in parentheses. Coefficients as percentages of the mean in the control group are shown in brackets. Controls include each individual's age, sex, marital status and years of schooling. P-values adjusted according to Anderson (2008) for the family of outcomes listed in the table. † The proportion of households with no minors between the ages of 3 and 15 is 42 percent. This share is the same for beneficiary and non-beneficiary households (p=0.248). 18 For individual outcomes, controls include each individual's age, sex, marital status and years of schooling. For household outcomes, controls include age, marital status, sex and education of the head of household. We compare the results shown in Column (2) with those given in Column (3) in Tables 3 to 7. We find that, for all variables in Tables 3 through 7, the estimates are both similar in magnitude and statistical significance. This evidence is consistent with the assumption that eligibility thresholds successfully provide local exogenous variation in treatment assignment. Next, we use monitoring information to incorporate differences between planned and actual treatment. We estimate program effects excluding the 260 non-eligible households that were identified ex-post. We also estimate local average treatment effects using eligibility as an instrument for the receipt of transfers. We find that these alternative specifications yield estimates that do not differ from our intent-to-treat estimates in our preferred specification. However, instrumental variable estimates are less efficient than ordinary least squares. We conclude that our average local treatment effects are within the margin of error of the intent-to-treat estimates. Tables that compare these estimates with our intent-to-treat estimates may be found in Appendix D. We conclude our results are robust to alternative specifications. VI. Generalizing the Results In this section, we compare our findings with those of Galiani, Gertler and Bando (2016). The Pension 65 program in Peru and the Adultos Mayores program in Mexico have three main features in common. First, both are federal programs intended to provide social security coverage to the elderly in poor areas. Second, both programs provide bi-monthly transfers of similar amounts (at the time these studies were conducted, the bi-monthly transfer in Mexico was equivalent to US$ 95, while it was equivalent to US$ 78 in Peru). Third, both programs have minimum eligibility requirements, since they both target persons above a set age threshold who are living in poverty. However, the two programs differ in two important ways as well. First, the Mexican government originally implemented the Adultos Mayores program only in rural areas (see Galiani, Gertler and Bando (2016) for a rigorous evaluation of the program’s implementation in rural localities with fewer than 2,500 habitants). Over time, however, Adultos Mayores was expanded to urban areas. The Peruvian government, on the other hand, did not introduce any geographic restrictions based on population size. Second, until the 2013 fiscal year, persons in Mexico did not become eligible for the Adultos Mayores program until they reached 70 years of age; whereas, in Peru, people have been eligible at age 65 for the Pension 65 program ever since its inception. In summary, we find that the results in the two countries are similar: the Geriatric Depression Scale scores in Peru decreased by 8.68 percent, while in Mexico they decreased by 9.11 percent; 19 paid work decreased by 4 percentage points in both countries; and consumption rose by 40 percent in Peru and by 14 percent in Mexico. In Peru, 67 percent of the increase in consumption was allocated to food, while in Mexico the corresponding figure was 54 percent. The magnitude of program effects thus does not differ to a statistically significant extent across the two countries. Figure 1 illustrates the comparison of the consumption, depression and labor variables in Mexico and Peru. Figure 1. The effects of non-contributory pensions on mental health, labor performed by older adults and household consumption Source: Authors’ calculations. Note: The results for Mexico correspond to the effects of the Adultos Mayores program in that country. These effects are reported in Galiani, Gertler and Bando (2016). The results for Peru correspond to the effects of the Pension 65 program. The two populations have many similarities. The average age of the beneficiaries is around 71.5 years in both countries, and approximately half of the population is male. Household consumption per adult is equivalent to US$ 45 for Peru and US$ 40 for Mexico. There were some significant differences between these sample populations, however. The program in Mexico targeted rural populations, while the program in Peru did not. As a result, the households in the sample for the -20 020 40 60 Non-contributory pension effect (percent) Depression Labor (pct. points) Household consumption Mexico Peru 20 Mexican study were larger, and the education level of the older adults was lower than in the Peruvian sample population. Another difference was that 59 percent of older adults work in Peru, while the corresponding figure was 36 percent in Mexico. Because of these differences, the labor impact of non-contributory pension systems is similar in magnitude in the two countries but is smaller as a percentage of initial outcomes in Peru than it is in Mexico. The two surveyed populations are similar in terms of the age and gender of older adults, as well as household consumption levels. However, there are some significant differences between the two populations that need to be identified, as they allow us to learn how the effects of noncontributory pensions vary in different contexts. We identify two main differences. First, the percentage of older adults who are working is higher in Peru. (A full 51 percent of the older adults reported having worked in the previous week for pay in Peru, while in Mexico the corresponding figure was 23 percent.) Accordingly, older adults’ labor earnings amount to US$ 23 in Peru but to only US$ 16 in Mexico. Both programs triggered a decrease of four percentage points in paid work. This change represents a 20 percent decrease (from 23 percent to 18 percent) in Mexico, but a decrease of only nine percent in Peru (from 51 percent to 46 percent). In addition, the household size in terms of adult equivalents is larger in Mexico, where an average household has 5.6 adult equivalents, while a household in Peru has 3.2. In addition, the average older adult in Peru has almost eight years of education, while the average older adult in Mexico has only two. These differences may, in part, be a result of the difference in targeting criteria, since the Adultos Mayores program in Mexico targets rural populations, while Pension 65 in Peru does not. We conclude that the results for Peru contribute to our knowledge about the effects of noncontributory pensions and allow us to apply that knowledge to a different context. The evidence suggests that the findings of Galiani, Gertler and Bando (2016) in rural Mexico can be reasonably well generalized to Peru in qualitative terms and, in many cases, in quantitative terms as well. VII. Conclusions. In order to study the effects of non-contributory pensions in Peru, we exploit a regression discontinuity design around the poverty score threshold for eligibility. Since we focus on a sample of households within 0.3 standard deviations from the threshold, this study provides a stronger identification strategy than that of previous studies. We find that the receipt of non-contributory pensions in Peru benefited older adults in several ways. For instance, it led to improvements in mental health, as evidenced by a reduction of nine percentage points in the overall Geriatric Depression Scale score. We do not find impacts on the 21 use of health services or health, but the receipt of those pensions did decrease the amount of paid work performed by older adults by 4 percentage points. The bulk of the cash transfer was used to finance an increase in consumption of 40 percent. In addition, recipient households are more likely to support members who reside elsewhere, as the share of households that made transfers to other individuals or households increased from 46 percent to 61 percent. More importantly, we find that our results are qualitatively similar to those of Galiani, Gertler and Bando (2016) in Mexico and hence both sets of results help us to construct external validity. Our findings should be viewed in the light of a number of caveats that point to directions for future research. First, we have observed these program effects after only one year, at most, since beneficiaries started receiving these program transfer payments, and it is possible that households may adjust their behavior in the long run. For example, Zhu and Xiaobo (2015) find that retirement leads to an immediate increase in life satisfaction, but they also find that the level of satisfaction decreases with time (see also Galiani, Gertler and Undurraga, 2016). A second caveat is that the data do not allow us to study how the receipt of non-contributory pensions may affect persons of working age near retirement age. Galiani, Gertler and Bando (2016), however, do not find anticipation effects in Mexico. The number of people in need of non-contributory pensions is likely to increase significantly in the coming years, and government expenditure on non-contributory pension schemes will probably climb. The findings of this study suggest that public expenditure on such pension systems results in welfare improvements among beneficiaries. Moreover, these pensions benefit not only older adults but also other household members. Therefore, non-contributory pensions appear to be an effective means of enhancing welfare among the older population and of reducing poverty. 28 Table B1. Definition of variables used in the tables (continued) Variable Definition Panel D. Well-being Satisfaction To construct this variable we used the following questions: "How content are you… With your health status? With yourself? With your ability to carry out daily activities? With your interpersonal relations (neighbors, friends)? With the place where you live? With your relationship with your children? With your relationship with other family members? With your life in general?" The points for each question for the possible response options were as follows: Very content=1; Content=1; Not very content=0 ; Not content=0. The score is the sum of the points for each question, divided by eight. Empowerment To construct this variable we used the following questions: "Do you think… That your family takes you into account when making decisions on household expenditures? That your family takes you into account when making important decisions for the household? That you support household expenditure? That you decide freely about what to spend your money on? That your family treats you with respect? That your family respects your wishes, opinions and other interests?" The points for each question for the possible response options were as follows: Always=1; Yes, most of the time=1; Sometimes=0; Rarely=0; Never=0 The score is the sum of the points for each question, divided by six. Continued 29 Table B1. Definition of variables used in the tables (continued) Variable Definition Contribution To construct this variable, we used the following question: "How much of your income do you contribute to household expenditure in the household where you live?" The values for this variable for the possible response options were as follows: All=1; Almost everything=1; More than half=1; Half=1; Less than half=1; Not very much=1; No contribution=0; Has no income=0. Self-worth To construct this variable, we used the following questions: "Do you consider that you: Provide economic support for the household? Provide support by doing household chores (cleaning, cooking, etc.)? Provide support in the form of childcare? Support others with your advice and experience? Represent a burden for the household?” (coding order reversed) The points for each question for the possible response options were as follows: Always=1, Sometimes=1, Rarely=0, Never=0 The score is the sum of the points for each question, divided by five. Well-being The average of standardized scores for satisfaction, empowerment, contribution and self-worth. We standardized each indicator according to the distribution in the control group for the corresponding year. All indicators had equal weights. Continued 30 Table B1. Definition of variables used in the tables (continued) Variable Definition Panel E. Household characteristics Income per adult equivalent Sum of labor income in the previous 4 weeks of all household members per adult equivalent in US dollars.1 See household size for the definition of adult equivalent. Income per adult equivalent excluding older adults Sum of labor income in the previous 4 weeks of all household members, excluding those aged 65 years or over, per adult equivalent in US dollars.1 See household size for the definition of adult equivalent. Household expenditure per adult equivalent Expenditure in the previous 4 weeks on food and on non-food items in the household in US dollars. 1 Household food expenditure per adult equivalent Expenditure in the previous 4 weeks on food and drink in or out of the household in US dollars. 1 Household non-food expenditure per adult equivalent Expenditure in the previous 4 weeks in US dollars for household maintenance, transportation and communications, domestic services, entertainment and cultural activities, personal care, clothes and shoes, health, transfers, furniture and electronics, and other goods and services (funeral services, marriage services, etc.). 1 Household size per adult equivalent Weighted sum of the number of household members. A weighting of 1 is given for persons older than 12 years and of 0.5 for persons 12 years old or younger. Age of head of household Age of the head of household in years. Married head of household Equals 1 if the head of household is married or living with a partner. Equals 0 if the head of household is widowed, divorced, separated or single. Male head of household Equals 1 if the sex of the head of household head is male. Equals 0 if the sex of the head of household is female. Education of head of household in years Education of the head of household. Assigns the following values to the last year completed: initial education: 2 years, elementary education: 8 years, secondary or advanced non-university education: 13 years, university education: 17 years, graduate studies: 18 years. The years of education are calculated on the basis of the last education level successfully completed. Note: The exchange rate used to convert Nuevos soles (S$) to US dollars (US$) was S$ 3.21 per US$ 1 in 2015 and S$ 2.58 per US$ 1 for 2012. Continued 31 Table B1. Definition of variables used in the tables (continued) Variable Definition Panel F. Enrollment Percentage of household members from 3 to 15 years old enrolled in an educational institution Number of household members from 3 to 15 years old enrolled in an educational institution, divided by the total number of household members between the ages of 3 and 15. This value is missing for households without members in that age group. Panel G. Current transfers to and from the household Receipt of current transfers in the previous six months (1 if yes, 0 otherwise) Transfers received in the previous six months in the form of alimony, pension transfers for food, remittances, survivor’s pensions, JUNTOS program transfers and other transfers from public or private institutions. Pension 65 transfers are listed separately and are not included in the calculation of this variable. Only transfers to older adults are considered. Receipt of current transfers in the previous six months excluding those to older adults (1 if yes, 0 otherwise) Transfers received in the previous six months in the form of alimony, pension transfers for food, remittances, survivor’s pensions, JUNTOS program transfers and other transfers from public or private institutions. Pension 65 transfers are listed separately and are not included in the calculation of this variable. Only transfers to household members other than older adults are included. Transfer expenditure in the previous 3 months (1 if any, 0 if none) Expenditures in the previous three months on tips to household members aged 14 or under, tips to non-household members, transfers, donations or gifts to family members not currently living in the household, periodic remittances to household members who live elsewhere, other expenditures, such as donations to institutions, church, charities, etc. Panel H. Social network transfers to and from older adults Social network transfer receipt (US$) Receipt of economic assistance in the previous six months by members of the social network of the older adult. Social network transfer provision (US$) Transfer of economic assistance in the previous six months to members of the social network of the older adult. Transfer receipt (1 if yes, 0 if no) Equals 1 if social network transfer receipt is non-negative. Transfer provision (1 if yes, 0 if no) Equals 1 if social network transfer provision is non-negative. Source: Authors’ calculations. 32 Appendix C. Impact on household income and expenditure by number of older adults in the household Table C1. Impact on household income and expenditure, by number of older adults in the household Full sample Households with one older adult Households with two older adults Mean in control group Effect Mean in control group Effect Mean in control group Effect (1) (2) (3) (4) (5) (6) Labor income per adult equivalent 38.46 4.24 41.31 5.27 31.62 1.52 (6.37) (6.4) (8.09) [11.02%] [12.76%] [4.81%] Labor income per adult equivalent excluding older adults 25.94 4.87 29.16 5.17 18.13 4.57 (6.62) (6.57) (7.11) [18.77%] [17.72%] [25.2%] Household expenditure per adult equivalent 45.16 17.94 48.36 13.40 37.82 28.86 (4.63)*** (5.15)** (3.84)*** [39.73%] [27.7%] [76.32%] Household food expenditure per adult equivalent 31.68 12.03 33.66 8.65 27.12 20.32 (3.68)*** (3.91)** (3.31)*** [37.99%] [25.69%] [74.91%] Household non-food expenditure per adult equivalent 13.49 5.91 14.71 4.75 10.70 8.54 (1.77)*** (1.99)** (1.86)*** [43.81%] [32.31%] [79.87%] Observations 2,584 1,829 752 Source: Authors’ calculations. Note: Standard errors, clustered at the conglomerate level, are shown in parentheses. Coefficients as percentages of the mean in the control group are shown in brackets. All estimates correspond to the RD with conglomerate fixed effects specification. 33 Appendix D. Comparison of intent-to-treat estimates with local average treatment effects Table E1 shows the number of households according to their eligibility status and receipt of transfers. Monitoring data are available for all the households except for 176 out of the 2,584 in the sample. Missing data do not differ across households above or below the eligibility threshold (p=0.612). Of those 2,408 households for which information is available, 260 received at least one pension transfer but transfers were later discontinued. Of these 260 households, 247 were households that had been deemed eligible when the program started (treatment). Thus, we estimate the program effects excluding these 260 households on the assumption that their exclusion improves data quality. Among the 2,148 households that were deemed eligible, 1,302 were in the treatment group. However, 177 never received a transfer. Out of the 846 eligible households in the control group, 20 received at least one transfer. Thus, we instrument actual treatment with treatment status before the program started. Tables D2, D3 and D4 show estimates for three specifications. The first column shows estimates with the RD model with conglomerate fixed effects and controls based on all 2,584 households and 3,342 individuals. This column shows the same estimates that are listed in Column (5) in Tables 3, 4 and 5. The second column shows estimates for the same model as in Column (1) but focuses on the 2,148 eligible households and 2,772 eligible individuals. The third column shows estimates for the same sample as the second column, but uses SISFOH score eligibility as an instrument for actual treatment. In summary, Columns (1) and (2) show intention-to-treat estimates and Column (3) shows local average treatment effects. Column (1) is based on the full sample, and Columns (2) and (3) are based on households whose treatment status was verified with monitoring data. Table D2 shows estimates of pension transfer effects on individual labor supply. Table D3 shows estimates of pension effects on health and well-being. Table D4 shows effects on household income and expenditure. In all three tables, the results do not differ to a statistically significant extent across models. Differences are larger for labor income in Table D2 between the RD model with controls (Column 1) and the local average treatment effect (Column 3). The average labor income in the control group for the full sample is US$ 22.93. Thus, the effect of these pensions varies from a reduction of 25 percent to a decrease of 56 percent in labor income. However, these two results do not differ to a statistically significant extent at the 10 percent level. As expected, local average treatment effects are larger than intention-to-treat estimates but are estimated less efficiently. We conclude that any errors related to eligibility classification are unlikely to explain differences between treatment and control groups. In addition, average local effects are larger and are consistent with intent-to-treat effects. Table D1. Number of households, by eligibility and transfer receipt Control Treatment Total Eligible Never received a transfer 826 177 1,003 Received at least one transfer 20 1,125 1,145 Non-eligible 13 247 260 With no monitoring information 66 110 176 Total 925 1659 2584 Source: Authors’ calculations. 34 Table D2. Impact on individual labor supply RD with conglomerate fixed effects and controls RD with conglomerate fixed effects and controls excluding noneligible households Local average treatment effect (1) (2) (3) Panel A. Work Worked during the previous week -0.03 -0.05 -0.06 (0.03) (0.04) (0.06) Hours worked during the previous week -1.39 -1.65 -1.97 (0.51)** (1.01) (2.06) Panel B. Paid work Worked during the previous week for pay -0.06 -0.11 -0.13 (0.02)*** (0.03)*** (0.05)** Hours worked during the previous week for pay -1.08 -2.44 -2.89 (0.76) (1.19)* (1.99) Labor income -5.73 -10.72 -12.79 (1.76)*** (2.62)*** (4.46)*** Source: Authors’ calculations. Note: Estimates for Column (1) are based on 3,342 observations. Estimates for Columns (2) and (3) are based on 2,772 observations. Standard errors, clustered at the conglomerate level, are shown in parentheses. Coefficients as percentages of the mean in the control group are shown in brackets. Controls include each individual's age, sex, marital status and years of schooling. 35 Table D3. Impact on health and well-being RD with conglomerate fixed effects and controls RD with conglomerate fixed effects and controls excluding noneligible households Local average treatment effect (1) (2) (3) Panel A. Physical health Hypertension -0.07 -0.09 -0.11 (0.03)* (0.03)** (0.06)* Waist circumference -0.79 -1.69 -2.02 (1.38) (1.48) (1.34) BMI -0.06 -0.35 -0.42 (0.13) (0.3) (0.52) Memory -0.11 -0.07 -0.08 (0.24) (0.25) (0.23) Physical health -0.03 -0.09 -0.11 (0.06) (0.06) (0.07) Panel B. Well-being Depression -0.04 -0.04 -0.05 (0.02)* (0.03) (0.03)* Satisfaction 0.00 0.01 0.02 (0.02) (0.02) (0.03) Empowerment 0.03 0.04 0.05 (0.02) (0.03) (0.02)** Contribution 0.11 0.12 0.14 (0.02)*** (0.02)*** (0.03)*** Self-worth 0.04 0.05 0.06 (0.01)*** (0.01)*** (0.02)*** Well-being 0.17 0.20 0.24 (0.03)*** (0.05)*** (0.07)*** Source: Authors’ calculations. Note: Estimates for Column (1) are based on 3,342 observations. Estimates for Columns (2) and (3) are based on 2,772 observations. Standard errors, clustered at the conglomerate level, are shown in parentheses. Controls include each individual's age, sex, marital status and years of schooling. 36 Table D4. Impact on household income and expenditure RD with conglomerate fixed effects and controls RD with conglomerate fixed effects and controls excluding noneligible households Local average treatment effect (1) (2) (3) Labor income per AE 4.99 1.21 1.45 (6.73) (8.06) (5.86) Labor income per AE excluding older adults 6.16 6.49 7.78 (6.46) (7.42) (5.32) Household expenditure per AE 18.05 14.01 16.76 (3.94)*** (4.3)*** (4.72)*** Household food expenditure per AE 12.16 9.38 11.22 (3.21)*** (3.83)** (3.92)*** Household non-food expenditure per AE 5.89 4.63 5.54 (1.97)** (1.73)** (1.61)*** Source: Authors’ calculations. Note: Estimates for Column (1) are based on 2,584 observations. Estimates for Columns (2) and (3) are based on 2,148 observations. Standard errors, clustered at the conglomerate level, are shown in parentheses. Controls include age, marital status, sex and education of the head of household.