Living arrangements often reflect important quality-of-life indicators for elderly adults. In particular, increased income can prompt changes in household living arrangements for elderly adults. Using a differences-in-differences approach, we examine whether a supplemental income program in Mexico for adults aged 70 and older influenced household size and composition. We compare outcomes at baseline and at six-month follow-up for elderly adults in the treatment group with those in the control group that did not participate in the program. We find that household size increased by 3% in the treatment group relative to the control group. We also find a statistically significant increase in the number of girls aged 6–11 in the household, likely the granddaughters or great-granddaughters of program recipients. Increases in household size were greatest for adults aged 70–79, couples, households receiving two or more supplemental incomes, and households in the top income tercile. Household size did not increase for households of adults aged 80 and older, singles, households with only one supplemental income recipient, and households not in the top income tercile. These results suggest that when older adults have more income, they use part of this income to house their grandchildren.
Latin America is aging faster than any other region in the world. Between 2015 and 2030, the number of adults aged 60 years and older will increase 71% in Latin America compared with 66% in Asia, 64% in Africa, 41% in North America, and 23% in Europe (United Nations 2015). Mexico is among the fastest aging countries in Latin America. Mexico’s old-age dependency ratio,1 while remaining below that of the United States, is expected to nearly triple, from 11.4 to 32.2, between 2015 and 2050; that in the United States is expected to nearly double, from 24.6 to 40.3 (OECD 2017). High poverty rates among the elderly are complicating this demographic transition in Latin America (Paz et al. 2004). Mexico in particular has the sixth-highest poverty rate (24.7%) among OECD countries for persons above 65 years of age (OECD 2018).
To address growing elderly populations and their high poverty rates, more than 50 low- and middle-income countries have introduced noncontributory pension or supplemental income programs (Newson and Bourne 2011; Willmore 2007). Noncontributory pensions are also called social pensions or unconditional cash transfer programs for older adults. Recipients of such programs make no contributions for the government benefits they receive.2 Many countries in Latin America have introduced and allocated considerable resources for such programs (Levy and Schady 2013).
Previous research has indicated that programs providing additional income to older adults may affect household composition. Programs that increase income of older adults in developed countries, such as the United States, increase the proportion of older adults who live independently (e.g., Costa 1997, 1999), but studies of developing countries have shown more nuanced effects. For example, Chen (2015) found that noncontributory pensions in China decreased shared living with adult sons but not daughters and that assistance from children in general decreased. Hamoudi and Thomas (2014) found that in South Africa, family members with lower levels of education who would likely receive lower wages were more likely than others to move in with pensioners and, presumably, care for them.
Previous research has also examined the support pensioners may provide and how this may affect household composition. Edmonds et al. (2005) found a 7% increase in children aged 5 or younger, a 10% increase in women aged 18–23, and a 10% decrease in women aged 30–39 among South African households with a noncontributory pension recipient. This suggests that pensioners use their additional income to shelter young mothers and their children. In studying South African noncontributory pension recipients, Duflo (2003) found that young girls living with female pensioners experienced anthropometric improvements in weight (for height) and height (for age). Young boys did not experience such improvements, nor were such improvements associated with living with male pensioners.
Previous research also found nuanced results for Mexico. Juarez (2009), analyzing a food assistance program for older adults in Mexico City that provides additional income, found no changes in living arrangements. Gutierrez et al. (2015), using a regression discontinuity design to study the same program in Mexico City, found no changes on household composition at the cutoff age for program eligibility but did find an increase in school enrollment for coresiding children. This suggests that public resources for older adults might benefit other age groups. Previous studies in Mexico analyzing Oportunidades, a conditional cash transfer program for families with children, also found changes in living arrangements. Rubalcava and Teruel (2006) found that recipients are more likely to shelter new members of the extended family who had resided elsewhere, whereas young adults in the recipient household may start their own families and move elsewhere. Stecklov et al. (2005) found that Oportunidades reduces migration to the United States but not domestic migration. Behrman et al. (2005) found that Oportunidades makes males aged 9–15 less likely to leave their household but has no migration effects for girls.
We build on previous research by using a randomized experimental design with individual- and household-level panel data to track changes in living arrangements resulting from a noncontributory pension program in Mexico. Our work may be more generalizable than that on South Africa (Edmonds et al. 2005) given South Africa’s history of apartheid and its effects on housing and living arrangements. Our analysis may be more generalizable than that on China (Chen 2015) given the patriarchal structure of living arrangements in China. Our work may also be more generalizable for Mexico and Latin America because we focus on semi-rural areas, which are home to most Latin American persons, rather than Mexico City. Finally, our research may yield new insights for Mexico given effects on household composition may depend on program recipient. Oportunidades, the subject of much previous research, may, with its payments to mothers with children, have different effects on household composition than a program for older adults would have.
We address the following research questions. Do noncontributory pension programs for the elderly affect household living arrangements? If so, how? Do noncontributory pensions lead to independent living arrangements? Do other household members—including recipients’ children, grandchildren, and great-grandchildren—move after noncontributory pensions are introduced? Our results will improve understanding of the potential effects of noncontributory pensions on recipients’ living arrangements.
Mexico’s Changing Demographics and Retirement Income Programs
In 1997, Mexico changed its social security system from a defined benefit (DB) system to a defined contribution (DC) system with personal retirement accounts (Aguila 2011). A DB plan is equivalent to a pay-as-you-go system, such as the U.S. Social Security system. Employees and employers make contributions, and employees receive retirement benefits based on their earnings history, tenure of service, and age. In a DC plan, the employer and employee contribute regularly, and individual benefits are based on the amount of money accumulated during the employee’s working life, including investment earnings.
The Mexican DC system includes some redistributive components. First, it includes a minimum pension guarantee. If workers at retirement have accumulated funds that would result in a benefit lower than the monthly minimum wage, then the government provides a subsidy to make up the difference. Second, the government contributes to DC individual retirement accounts: a small amount, called the social quota, equivalent to 5.5% of the minimum wage for Mexico City (Aguila 2011).
The transition to a DC plan likely worsened poverty rates among older Mexicans. One way it may have done so is by increasing eligibility requirements. The reform increased the minimum years of contribution for benefit eligibility from 10 years in the DB system to 25 years in the DC system. Lower-income workers generally have more frequent and longer-lasting spells of unemployment, making qualification for them difficult. They are also more likely to work in the informal labor market, which includes the self-employed and employees in noncompliant small firms not participating in social security programs (Perry et al. 2007). This puts low- and middle-income workers at a higher risk of poverty in old age because such workers are less likely to accumulate enough retirement funds to support themselves or even to meet eligibility requirements for a minimum pension guarantee (Aguila et al. 2011).3 Previous research has confirmed that pension coverage decreased after Mexico’s reforms (Mesa-Lago 2006), whereas the proportion of the population in the informal sector increased (Perry et al. 2007). Today, nearly 60% of workers are in the informal sector (INEGI 2018).
A noncontributory pension system may address these problems by providing a safety net that is independent of labor histories. For example, in the United States, the Supplemental Security Income (SSI) program is a last-resort noncontributory program designed to provide a basic standard of living for low-income older adults (Social Security Administration 2011). Without SSI, poverty among recipient families in 2010 would have been 65% rather than the actual rate of 43% (Center on Budget and Policy Priorities 2014). Most SSI recipients have no other source of income (Center on Budget and Policy Priorities 2019). From this, we infer that a noncontributory pension in Mexico may have a great impact, given that less than one-third of older adults receive social security benefits (Ramírez 2017).
Living Arrangements and Additional Retirement Income
Housing is but one challenge that Mexico confronts with its older population. Where older adults live and with whom they live will have significant implications for the well-being of Mexican households and the Mexican economy. Residential arrangements, for example, may reflect care arrangements. Litwak and Longino (1987) suggested that as the elderly age and their health declines, they are more likely to move closer to adult children. Bradley (2011) interpreted the older age distribution and greater rates of disability among older movers from Florida to the Northeast as suggesting that older adults move closer to family as their health begins to decline. Even in the absence of adult children, older adults may move in with extended family for companionship during later life. Very few older women without children in Latin America, for example, live alone; most live with extended family (De Vos 2014).
The relationship of household members to older adults can influence mental and physical health. One study of the U.S. elderly found that living alone is associated with a greater probability of experiencing depressive symptoms than those living with spouses or other family members experience (Wilmoth and Chen 2003). Older adults in China who are in their preferred living arrangements have better self-reported health than those who are not (Sereny and Gu 2011). Cultural traditions and social insurance provide resources for encountering such adversities as unemployment or illness (Tosi and Grundy 2018). In Latin America, the lack of unemployment insurance, long-term care services, and private pensions makes families play a fundamental role in providing for the well-being of their members (Bongaarts and Zimmer 2002).
At least three approaches can help us understand how additional income for older adults may lead to changes on household composition: the risk-sharing model, the altruism model, and the exchange model.
According to the risk-sharing model, additional income for older adults enables families to maximize their total production. Family members can ensure one another’s income against uncertain events or income shocks (Kotlikoff and Spivak 1981; Rosenzweig 1988). For example, grandparents taking care of grandchildren can help families smooth fluctuations in resources (Butcher 1993) or minimize living costs and maximize family well-being (Frankenberg et al. 2003). With scarce economic resources and credit constraints, the economies of scale associated with larger households may determine the optimal household size (Salcedo et al. 2012; Zimmer and Kwong 2003). Increased income can also change informal care and household living arrangements for families (Edmonds et al. 2005). For instance, additional income has increased labor force participation of prime working-age women in South Africa (Edmonds et al. 2005) and the United States (Hoerger et al. 1996). In Europe, families have used living arrangements to exchange physical, domestic, in-kind, and monetary assistance with the elderly (De Vos et al. 2004).
According to the altruism model (Becker 1993), parents care about the well-being of children and grandchildren, and they provide more resources to less well-off family members (McGarry and Schoeni 1995). Under this model, additional income of older adults would be negatively correlated with children’s income as parents provide more resources or allow those most in need to coreside with them. Disadvantaged family members are more likely to receive intergenerational transfers than those who are better off (Silverstein et al. 1995). Grandparents may also provide caregiving for grandchildren rather than receive caregiving from their children and assume caregiving roles in response to parents’ needs rather than their own needs (Gonzalez Bernal and de la Fuente Anuncibay 2007).
In the exchange model (Cox 1987), transfers represent an exchange for provision of services from the recipient. Extra income may increase transfers to family members who provide services, allowing such family members to coreside with the older adult. Hamoudi and Thomas (2014) found that pensioners were more likely to live with younger adults who had lower human capital and a lower opportunity cost for providing care to the older adult. Pensioners transfer resources to younger adults in exchange for care services.
Our study expands on Hamoudi and Thomas’s (2014) work by considering the possibility that pensioners might use extra income to reduce transaction costs of providing rather than receiving care. By coresiding with other family members, pensioners can minimize the costs of providing care. Coresiding with other family members, for example, eliminates transportation costs that pensioners may incur for caregiving. Because of their age, pensioners may not be able to provide continuous and consistent care to other family members unless they coreside with them. Hence, pensioners may be more likely to coreside with family members for whom they provide care. Data on composition of households for pensioners would allow us to discern whether pensions affect household living arrangements, particularly for persons who might receive care from older adults.
For this research, we draw on data collected during implementation of a noncontributory pension program in the Mexican state of Yucatan. In 2007, Yucatan began to implement a noncontributory pension program (Reconocer) for adults aged 70 and older. The program was introduced in three stages. In the first stage, which began in September 2007, noncontributory pensions were implemented in 10 localities with 2,500 to 6,500 inhabitants. In the second stage, which began in December 2007, these pensions were implemented in 16 localities with 6,500 to 20,000 inhabitants. In the third stage, which began in 2008, the pensions were implemented in localities with more than 20,000 inhabitants.
The third stage also featured a cluster randomized control trial design for evaluating the program. This trial provides the data we analyze. Among the 11 localities eligible for the third phase—Hunucma, Kanasin, Uman, Merida, Motul, Oxcutzcab, Progreso, Tekax, Ticul, Tizimin, and Valladolid—a pairwise matching procedure is used to ensure similarity in household and community indicators (as measured by the 2005 census) across pairs of towns. One of the matched pairs was chosen randomly for analysis. Within that pair, one of the localities, Valladolid, was randomly selected to receive the noncontributory pension (treatment), with the other, Motul, selected for the control (comparison) group.
A figure and two tables in the online appendix show how similar Valladolid and Motul are. Figure A1 shows no evidence of different trends in economic development (measured by the Human Development Index, 2000–2010) and income distribution (measured by the Gini coefficient; 1990–2010) for Valladolid and Motul. Table A1, based on data from the 2005 Mexican Census, shows similarities between Valladolid and Motul on a poverty index; in the percentage of adults who are illiterate; and in the percentage of households without a refrigerator, without electricity, or with an earthen floor. Table A2, based on census data from 1990 to 2010 and a regression-based test of differential pretreatment trends between Valladolid and Motul, shows no statistically significant coefficients for the interactions between treatment and year. This implies that the assumption of common trends cannot be rejected for any of the outcomes considered. The joint F test also indicates that the interaction terms before 2010 are not jointly statistically significant. In short, Valladolid and Motul experienced similar overall conditions in economic hardship before, during, and immediately after the implementation of the noncontributory pension. Hence, we can be more confident that evaluations of the noncontributory pension program that detect changes in household composition are a result of the pension and not of other economic changes.
The evaluation survey in both Valladolid and Motul collected data on community, household, and individual characteristics before the program was announced or implemented (baseline) and approximately six months after the program was implemented in Valladolid. Baseline (Wave 1 (W1)) data collection began in August 2008 for Valladolid and in November 2008 for Motul. The follow-up survey (Wave 2 (W2)) was conducted in both Valladolid and Motul from July to September in 2009. The surveys were collected as part of a project between the state of Yucatan and the RAND Corporation. The data are not based on a sample but rather capture virtually all residents 70 or older in Valladolid and Motul.
The noncontributory pension program gave adults 70 or older in Valladolid MXN $550 per month (US$70.20 at 2013 purchasing power parity (PPP)) starting in December 2008. This benefit was approximately one-third of the monthly minimum wage in Yucatan (MXN$1,865.95 in January 2013 or US$238.20 at 2013 PPP). The amount of the noncontributory pension was similar to a noncontributory pension provided by the Mexican federal government (MXN$500 in 2008; Aguila et al. 2016a) to older adults living in select areas with less than 20,000 inhabitants during the time of this study. The federal noncontributory pension amount was set as a flat rate to alleviate income insecurity in old age among those living below a food poverty line (Macias Sanchez 2016).
In most cases, only one or two eligible adults per household received the Yucatan noncontributory pension. The age restriction of 70 was the same as that for the Mexican federal government noncontributory pension during the period of this study (Aguila et al. 2016a) as well as that for older adults receiving noncontributory pensions in other Latin American countries, including Argentina, Chile, and Uruguay (Barrientos and Lloyd-Sherlock 2002). The age threshold provides a disincentive for workers in Mexico, where the retirement age is 65, to move to the informal sector and avoid paying social security contributions (Bosch and Guajardo 2012).
The survey included items taken or adapted from other longitudinal studies, such as the Mexican Health and Aging Study (MHAS), the U.S. Health and Retirement Study (HRS), the New Immigrant Study in the United States, a survey for evaluating the Mexican Oportunidades program, and other family life surveys. Informed consent was required from participants at each survey stage. Specifically, while assuring individual confidentiality, we asked respondents for their consent to take part in the general survey (oral), to allow the research team to review administrative and health records (written), and to collect anthropometric and biomarker measurements (written). These documents were available in Spanish and Mayan and provided to all project participants to keep for future reference. The institutional review boards of both the RAND Corporation and the State of Yucatan approved the study procedures (protocol approval number 2008-0513-CR07). Complete descriptions of the study protocols, including a detailed description of its design, sampling frame, and follow-up procedures, are available elsewhere (Aguila et al. 2014, 2016b).
Both the baseline and follow-up surveys collected data on self-reported measures, anthropometrics and biomarkers on health and disability, health care usage and out-of-pocket expenditures for the elderly respondent, financial and in-kind transfers among beneficiaries’ children and neighbors, economic activity of older workers, and household-level data on food availability. The surveys also included a household roster with detailed information about family members, including age, gender, education, and relationship to the elderly respondent.
For this study, we built a separate data set with individual lines for each family member, allowing us to track changes in living arrangements across time. We carefully merged family member data by name, age, gender, and education across waves. This data set includes family members who live in the household and children who live outside the household, monetary and in-kind transfers from family members, and hours of informal care provided to the elderly respondent. Our methods were similar to those used to construct the U.S. HRS family data file. In contrast to similar previous studies (e.g., Edmonds et al. 2005), our data enable a panel design.
Table 1 shows descriptive statistics for our samples in Valladolid and Motul at W1. As shown by t tests for the differences between the two towns, respondents were similar in age, marital status, education, paid employment, and monthly household income. Motul did have a larger share of males 70 and older.
As in the U.S. HRS, the older adult was the only person interviewed for the vast majority of households. The respondent provided information for other family members, who were interviewed only when they were identified as the most knowledgeable about household finances and then asked only about this topic. Our data include other family members only if they were living in the household with an adult aged 70 or older in W1 or W2. Because multiple older adults were interviewed in 35% of households, standard errors were clustered at the household level to adjust for within-household correlation.
We identified individuals who moved by computing the distance between housing units in W1 and W2 using global positioning system (GPS) devices and ellipsoidal distances based on Vincenty’s (1975) formulae. Between W1 and W2, 56 older adult respondents, or 0.05% of the sample of older adults, moved to a different housing unit. We were able to locate and interview 16 (29%) of these 56 respondents to include them in our analysis. We compared baseline characteristics of those whom we could/could not locate and found no statistically significant differences between them (results available upon request). We checked the consistency of the variables used across time. We imputed missing information in the independent variables by using the Software Imputation and Variance Estimation Software (IVEware 0.2) designed by University of Michigan.
Response rates at W1 were 91.5% in Valladolid and 95.3% in Motul; response rates for W2 were 87.9% in Valladolid and 81.9% in Motul (response rates calculated per methods of American Association for Public Opinion Research 2011).
We examined a series of dependent variables to capture changes in household living arrangements. These variables are (1) household size; (2) the percentage of respondents living alone or with a spouse only; (3) the percentage of respondents who are single or couples and live with other family members; and (4) changes in the number of household members in elementary school (6–11 years old), middle and high school (12–19 years old), and in 10-year age groups from age 20 to 69 (i.e., 20–29, 30–39, 40–49, 50–59 and 60–69). Most previous studies regarding program effects on household composition analyzed household size (e.g., Edmonds et al. 2005). Previous research has also found differences among older adults with partners, without partners, and coresiding with other family members (De Jon Gielverd and van Tilburg 1999).
Table 2 shows descriptive statistics of the outcome variables. Average household size was 3.4 individuals per household in treatment and control groups. The proportion of older adults living alone was 13% in treatment and control groups. The average number of household members by age groups were similar in both Valladolid and Motul, with the exception of the 40–49 age group.
The independent variables in our analysis are age, age squared, gender (male = 1, female = 0), marital status (married or in union = 1, single = 0), and years of education—all variables not affected by the introduction of the noncontributory pension program. We also conducted our analysis by household income terciles, using the same definition as the HRS and MHAS. We followed the same imputation method as the MHAS for nonresponse or responses using unfolding brackets in economic variables. Household income includes family transfers (both monetary and in-kind), salary income (even from sporadic jobs), income from businesses or farms, pensions, income from properties, capital income (checking and savings accounts, fixed investments, stocks, bonds and shares), and transfers from governmental institutions.
Our choice of independent variables was guided by previous literature. We included age to account for varying motives to move during later life (Litwak and Longino 1987). We included gender because females are more likely to use pension income to assist children (Duflo 2000, 2003). We included marital status because marriage increases the likelihood of there being two pensioners in the household, loosening household budget constraints (Becker 1993). We included education as an indicator of human capital that plays an important role in determining living arrangements (Hamoudi and Thomas 2014). We included household income as an indicator of socioeconomic status that has been shown to be an important determinant of living arrangements (Costa 1999).
We conducted the analysis of the effects of a noncontribution pension for older adults on household composition in several ways. In addition to presenting unadjusted difference-in-difference (DID) intention-to-treat (ITT) estimates, we estimated the causal effect of the noncontributory pension program on living arrangements using standard ordinary least squares (OLS) DID regression, logit, negative binomial regression, and propensity score matching.
We estimated logit regressions for binary outcome variables; results presented are odds ratios. Negative binomial regression is a type of generalized linear model in which the dependent variable is a count of the number of times that an event occurs. Because most of the outcomes examined here are counts of the number of household members in different age groups, the distributional assumptions of negative binomial regression may be more appropriate than OLS regression (Cameron and Trivedi 2013). The coefficients of the negative binomial DID are reported in incidence rate ratios.
Propensity score matching is a method whereby treated observations are matched to similar nontreated observations. Like OLS regression, propensity score matching accounts for differences in confounding variables but has the advantage of not assuming linearity. Nonparametric propensity score matching corrects for differences in baseline characteristics between treatment and control groups (Heckman et al. 1998). We imposed common support across groups and used kernel matching. We obtained standard errors through the bootstrap method with 1,000 replications. We compared parametric estimates using OLS, logit, and negative binomial regressions with nonparametric propensity scores matching DID estimates.
In contrast to previous research on older adults in developed countries (Costa 1997, 1999), we did not find that increased income leads to independent living among older adults in Mexico. Similar to Edmonds et al. (2005), we found no changes in the proportion of elderly adults living alone. This suggests that independent living is not a normal good for older adults in Mexico.
Results connote the preferences and intrahousehold sharing resources mechanisms that additional income confer on older adults. The noncontributory pension program increased average household size by 3.0% (0.101 / 3.42 = 0.030) in the treatment group compared with the control group (see Table 3 DID with propensity score matching and Table 2 average household size for the treatment group). Our results suggest in particular that additional income prompts grandchildren to live with older adults. The average number of children aged 6–11 in the household increased by 23.2% (0.044 / 0.19 = 0.232) in the treatment group compared with the control group (see Table 3 DID with propensity score matching and Table 2 average number of household members for that group). These estimates based on DID with propensity score matching are statistically significant when the Holm-Bonferroni correction for multiple hypothesis testing is used. These results fall in line with those of Edmonds et al. (2005), who also found an increase in the proportion of young children in the household following receipt of pension. These findings suggest that living with small children is a normal good for older adults in Mexico.
It is worth exploring this relationship further. On the one hand, older adults may enjoy spending time with their grandchildren and will use their extra income to lower the transaction costs of “consuming” their company. Time with grandchildren can be thought of as a service that the pensioner purchases by spending the time and energy to travel to their location. Older adults may lower these transaction costs by using their extra pension income to house their grandchildren.
On the other hand, older adults may use their extra pension income to lower the transaction costs of providing services instead of receiving services. Becker (1993) argued that parents draw utility from helping their children, particularly those who are in greater need. Our results suggest that, as Becker argued, older adults draw benefit from helping their adult children. Because our results do not show a corresponding increase in the number of adults moving in with pensioners, we hypothesize that the grandchildren who move in with older adults are the children of parents who cannot support them. Qualitative interviews with some W2 households support this hypothesis (results available upon request).
In other words, older adults use their extra income to act as surrogate parents. Riley and Riley (1993) described family as a latent resource that can be triggered in times of need. Building on this concept, we hypothesize that an infusion of income might also activate family assistance if family members experience a continuous need for assistance, which pensioners with additional income might address. We may also consider that housing grandchildren may be a result of a family insurance mechanism to help family members after income shocks and smooth fluctuations in income as predicted by risk-sharing models (e.g., Butcher 1993; Rosenzweig 1988).
Overall, the results for the DID with propensity score matching and DID of the means are similar in sign, magnitude, and statistical significance. The logit and negative binomial DID regressions also confirm our findings.
Heterogenous Treatment Effects
We also analyzed heterogeneous effects by the gender of family members living in the household, groups of age of the older adult (70–79 years vs. 80 or more years), singles and couples, number of pension recipients in the household (one pension recipient vs. two or more recipients), and household income tercile. Figures 1, 2, 3, 4 and 5 show the DID with propensity score matching estimates with corresponding 95% confidence intervals. We could not include the age group 0–5 years old in our analysis because of the small number available for analysis.
Figure 1 shows changes in the number of household members by gender and age group. The figure shows an increase in the number of girls 6–11 years old, specifically granddaughters and great-granddaughters, in pensioners’ households, but no effects for boys.
Figure 2 shows the effects on living arrangements on households with adults 70–79 years old and households with adults 80 years or older. Among households with adults 70–79 years old, there are two statistically significant effects: increased household size, and an increase in particular in the number of persons 6–11 years old in the household. Among households with adults aged 80 or older, there are no significant changes in the household size or composition variables.
In Fig. 3, we show the changes on living arrangements for households headed by a single older adult and those headed by an older adult with a spouse. The only statistically significant changes here are for households headed by a couple. Such households saw an increase in household size generally and the number of members 6–11 years specifically. They also saw an increase in the proportion of households with other family members and a corresponding decrease in the proportion of households without other family members.
Figure 4 shows changes in households by the number of pensioners in the household (one or more than one). The only statistically significant effects here are on households that had at least two adults receiving the noncontributory pensions. Such households saw statistically significant increases in household size as well as in the number of persons 6–11 years of age. They also saw an increase in the number of households comprising a couple living with other family and a corresponding decrease in the proportion of households comprising only a couple.
Figure 5 shows changes on households by income tercile. The only statistically significant effects are for households in the highest-income tercile. Such households saw statistically significant increases in household size and in the number of residents 6–11 years of age. They also saw a statistically significant increase in the proportion of households headed by a couple with family and a corresponding decrease in the proportion of households headed by a couple but without family living with them. Households in the other two income terciles show no statistically significant changes in household composition variables, although point estimates are similar for all three terciles.
Poverty data add further context to our results. In our baseline data, 48.3% of adults 70 years old or older in Valladolid and 51.3% in Motul were below the Mexican national poverty line, which in late 2008 was MXN$1,954.55 per month or US$301.43 PPP (CONEVAL 2015). Across all of Mexico, 45.7% of the adults aged 65 and older lived in poverty, and 11.0% lived in extreme poverty, with poverty across Yucatan being higher still (CONEVAL 2016). In short, older adults both in our study and across Yucatan and Mexico typically have low incomes. The noncontributory pension, accounting for an average of 26.0% of household income, represents a significant increase in income for these households. Higher-income households, as well as those headed by couples and those receiving two or more noncontributory pensions, may be better positioned to share this additional income with their granddaughters and great-granddaughters.
We checked the robustness of our results in several ways. One of the main identifying assumptions of the DID methods is common trends between treatment and control groups. As discussed in the Data section, we found similar trends between Valladolid and Motul when comparing aggregated indicators (GINI index and human development index; see Fig. A1 in the online appendix). We also conducted OLS regressions to test differences in pretreatment trends using data from the Mexican census from 1990 to 2010 (see Table A2 in the online appendix). We found no significant differences between Valladolid and Motul in the proportion of households with earthen floor, number of rooms, access to potable water, access to sewage, and access to electricity—all variables commonly used to analyze poverty and household characteristics in Mexico. Finally, we found that Valladolid and Motul do not differ in the proportion of older adults living alone. We therefore conclude that these two localities have had similar socioeconomic trends over time.
Potential sources of attrition in our analysis are nonresponse and death. We compared the sociodemographic characteristics of attriters and panel respondents for Valladolid and Motul and differences across both towns. We found no statistically significant differences within Valladolid and Motul. We conducted a similar analysis comparing deceased and panel respondents. We found statistically significant differences within Valladolid and within Motul but few statistically significant differences across towns (see Table A3 in the online appendix). The deceased were older, less educated, and less likely to have worked for pay. The deceased in Valladolid were also slightly older than the deceased in Motul. Overall, these results do not point to attrition bias.
One possible concern for our analysis is that access to other public programs may have changed as a result of introducing the pension program. We found, however, a reduction of only 3.3 percentage points between W1 and W2 in the proportion of older persons in Valladolid receiving Oportunidades (now called Prospera) for elderly adults living in eligible households (see Table A4, online appendix). We did not observe effects on other government programs. Despite a notable difference between the towns in the proportion of households receiving Oportunidades at baseline (11.1% in Valladolid vs. 18.1% in Motul), the validity of the DID estimator rests on the similarity in the change in characteristics across groups over time rather than on their baseline levels (Angrist and Pischke 2009).
Another possible concern for our analysis is the overlap of data collection periods with summer breaks from school. Baseline data collection for Valladolid overlapped local school breaks by one week (August 10–17, 2008) during which 174 data collection interviews were conducted. Baseline data collection did not overlap school breaks in Motul. We conducted the same analysis shown in Table 3 excluding the interviews collected during the summer break week to assess possible bias. The results of this analysis (available on request) were qualitatively and quantitatively unchanged. We also compared our outcome variables at baseline for those interviewed in Valladolid in the one-week overlap period with those interviewed in the month afterward (August 18 to September 19, 2008). Again, we found no statistically significant differences between both groups (results available upon request).
Wave 2 data collection had greater overlaps with local school breaks. In both towns, summer break for primary and high school students was July 3 to August 24, 2009. Wave 2 data collection was July 7 to August 26, 2009, in Valladolid, and July 7 to August 27 in Motul. During these overlap periods, we conducted 1,096 (97.8% of the total) interviews in Valladolid and 741 (93.7%) in Motul. After August 24, we collected 25 surveys in Valladolid and 50 surveys in Motul. If older adults in both towns experienced a similar inflow of children during the summer vacations, then the effect of this temporary inflow would offset any increase we found among households interviewed after August 24. We conducted the same analysis in Table 3 but excluding interviews conducted after August 24, finding similar effects in magnitude and statistical significance (results available upon request). This implies that inflow of children during the summer does not pose any threat to our findings.
We also explored the effects of the noncontributory pension on living arrangements using an alternative data set and methods. We conducted a sharp regression discontinuity design (RDD) among individuals above and below 70 years old in Valladolid, specifically for older adults aged 70–79 and 60–69. Exploiting the age eligibility of the program, we estimated the local impact of the noncontributory pension for adults aged 70–79. We obtained the RDD estimates using data from the 2010 census. We used as our control group households with adults aged 60–69 but without adults aged 70 or older (i.e., older households not eligible for the benefit). Table A5 of the online appendix shows the results. Consistent with our previous findings, we found an increase in average household size but virtually no change for the proportion of persons living alone.
We designed the rollout schedule of the program and the timing of public information campaigns in close cooperation with the State government of Yucatan. The public information campaigns for the pension program started one month before the disbursement of the first pension payment and one month after W1 data collection was finished. Furthermore, the information about the program was disseminated only in Valladolid. We therefore surmise that bias effects resulting from announcement of the program are negligible. In addition, because Valladolid is 95 miles away from Motul, spillover bias effects on survey responses are unlikely. We also obtained information about differential shocks between W1 and W2 for treatment and control groups. Both towns, which are on the eastern side of the Yucatan peninsula, reported being affected by droughts (April and May 2009), suggesting similar aggregate effects.
Discussion and Conclusion
This research examines the effects of a noncontributory pension program on changes in household living arrangements in Mexico. We found that the noncontributory pension program affected household size, particularly increasing the number of granddaughters and great-granddaughters aged 6–11 living in the household, and specifically for pension recipients aged 70–79, living with a spouse, in a household receiving two or more noncontributory pensions, and in the top income tercile.
It is worth exploring possible mechanisms underpinning these results. In their research on income transfers to elderly Black South African women, Edmonds et al. (2005) hypothesized that increased income may enable pensioners to house their young grandchildren while their daughters work away from home. They found that increased pension income was associated with a 7% increase in children aged 0–5 and a 10% decrease in prime working-age women aged 30–39 living in pension recipients’ homes. We did not find a change in the number of women in their 30s who moved out of pensioners’ households. It is possible that women in this age group do not move out of pensioners’ households but rather send their young daughters to live with pensioners, enabling these women to work more hours. This is consistent with previous research showing that Latin American grandmothers take on primary caregiving roles while their adult children work away from home (Da 2003). As Edmonds et al. (2005) also argued, this result may be unrelated to family preferences but rather is the result of changes in the labor market. The infusion of pension income into the local economy may ignite a demand for labor, which decreases the comparative advantage of caring for children among working-age adults. These adults may send their children to live with their parents as a result.
Contrary to previous studies, our research does not show that older adults use increased pension income to draw support from their families. Rather, our study suggests they use their increased income to provide support for their families. Hamoudi and Thomas (2014) found that pensioners in South Africa were more likely to live with men and women of lower educational attainment. They noted that although pensioners may use their extra income to help support these family members, less-educated men and women were more likely to reciprocate by providing older adults with personal care. In contrast, pensioners in our study were more likely to live with a dependent group—young girls—unlikely to provide personal care. This difference may reflect cultural differences, with older Mexicans who are able to do so providing care. Research on a perhaps more culturally similar group in Brazil did not find an effect of increased pension income on household composition (Kassouf and Rodrigues de Oliveira 2012). However, the authors of that study did not examine changes in the composition of children in the household but rather only different groups of adults. Although De Vos (2018) did find an increased probability of living in an extended family among low-income older women receiving a noncontributory pension, the author was not specific about the age composition within these households.
The gender-specific nature of our results is consistent with that in other studies. As in our study, Chen et al. (2017) found differential effects for male and female family members of older pensioners. Specifically, they found that increased pension income in China decreased shared living with sons but not daughters. Our finding of an increase in the number of granddaughters and great-granddaughters aged 6–11 living in the household housed by their grandparents may stem from pensioners providing care to their grandchildren most in need. Young girls are more likely than young boys to represent a financial burden for Mexican households; in Mexico, young girls are less likely to attend school (Parker and Pederzini 2000) and receive a lower share of resources when household heads return from a migration spell (Antman 2011). It is therefore reasonable to expect that pensioners would more readily house young girls because young girls are more likely in need of support.
Although the intergenerational dependence suggested in our study matches that found in other studies for China and South Africa (e.g., Chen et al. 2017; Edmonds et al. 2005), our results might not hold in developed countries. For example, studies of the United States and Canada have found that increased income from Social Security and other programs for older adults leads to less coresidence for pensioners and family members (Boyd 1991; Costa 1997, 1999; McGarry and Schoeni 2000; Orsini 2010). Thus, noncontributory programs in the United States and Canada would not necessarily lead to transfers from older adults to granddaughters. Nevertheless, older U.S. and Canadian immigrants are more likely to live with other family members because of economic necessity, familial values, and a desire to follow the cultural norms of their home countries—particularly for immigrants from Mexico, Latin America, and China (e.g., Angel et al. 1996; Diwan et al. 2011; Gonzales 2007; Gubernskaya and Tang 2017; Kaida and Boyd 2011). Therefore, we may expect results similar to our work for older adult immigrants in the U.S. and Canadian population.
The largest increase in the number of granddaughters and great-granddaughters is in households of adults aged 70–79, couples receiving two or more income supplements, and households in the highest income tercile. Consistent with the South African case (Edmonds et al. 2005), we did not find that increased income is associated with a higher probability of independent living. Rather, our results suggest that increased income enables grandparents with higher socioeconomic status to actualize their desire to help their more needy children, a finding that falls in line with altruistic and risk-sharing models. This interpretation falls in line with similar work done in Brazil, in which more than 80% of older adults who received a noncontributory pension reported sharing their increased pension income with household members (Barrientos et al. 2003). Our work highlights family behavioral changes as a mechanism through which other generations benefit from public policies directed at the elderly.
A limitation of our work is that six months is a short time frame in which to observe changes in living arrangements given the effort that moving may require. The effects of the noncontributory pension in our study may be limited to children because they can move more easily than adults can. Another limitation of the study is that we do not have information on the prior residence of granddaughters and great-granddaughters nor any detailed information on their parents’ household.
The Yucatan program may have several other implications. In South Africa, increased pension income led to improved anthropomorphic measures for young girls but not boys (Duflo 2003). Sustained pension income may lead to similar positive effects in Mexico. Increased pension income may also have economic benefits enabling, for example, pensioners’ working-age daughters (and granddaughters) to work away from home and helping their family guard against income shocks. Finally, the living arrangements that result from increased pension income may help reinforce the intergenerational ties that provide a protective effect in the face of hardship. In one U.S.-based study, children raised by their grandparents fared better on several measures than those raised by a single parent (Solomon and Marx 1995). This benefit, however, must also be weighed against the physical and emotional toll that grandparents experience in serving as surrogate parents (Hayslip and Kaminski 2005).
We would like to thank Leandro Carvalho, Richard K. Green, Dowell Myers, and participants at 2016 Lusk Research Seminar at the USC Sol Price School of Public Policy, the 2017 DITE Winter meeting at Duke University, and the 2018 PAA annual meeting for their valuable comments. The expert assistance of Stephany Maqueda and Jorge Peniche is gratefully acknowledged.
EA designed the intervention, conducted data collection for the study, framed the scope of the paper, conducted the statistical analysis, and was the primary text author. JP framed the scope of the paper, conducted literature review, participated in the statistical analysis, and helped write the manuscript. AV framed the scope of the paper, conducted literature review, and helped write and revise the manuscript. All authors reviewed and approved the final draft.
The design and data collection of this research was supported by funding from the State of Yucatan and by Grants R01AG035008, P01AG022481, and R21AG033312 from the National Institute on Aging (NIA). The construction of the data set at the family level was supported by a grant from the USC Lusk Center for Real State affiliated with the USC Sol Price School of Public Policy and the USC Marshall School of Business.
Data, surveys, and supporting material of the experiment conducted in the State of Yucatan, Mexico is available at https://cesr.usc.edu/yucatan_aging_data.
Compliance With Ethical Standards
Ethics and Consent
Informed consent was required from participants at each survey stage. The institutional review boards of both the RAND Corporation and the State of Yucatan approved the study procedures (protocol approval number 2008-0513-CR07).
Conflict of Interest
The authors declare that they have no conflict of interest.
The old-age dependency ratio is the ratio of the number of individuals at least 65 years of age per 100 individuals of working age (i.e., 15–64 years). It is a common indicator of the extent of challenges that a society may face in providing for its older population.
These programs, which are equivalent to the Supplemental Security Income (SSI) program in the United States, seek to decrease income insecurity in old age. In contrast, contributory pensions are equivalent to Social Security benefits in which the employee, employer, and government finance retirement benefits.
Workers who do not meet the eligibility criteria can withdraw the amount accumulated in the retirement account but cannot claim a minimum pension guarantee (Aguila et al. 2011).
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.