Past research on the influence of conditional cash transfers—widespread antipoverty programs—on migration has tended to focus on beneficiaries as a homogenous unit. Drawing on feminist critiques of the contemporary international antipoverty agenda, this article views both conditional cash transfer programs and migration patterns from a gender-sensitive lens. Conditional cash transfers rely on a gendered division of labor in which the informal work of women is particularly called upon in order to fulfill program requirements. This work contends that conditional cash transfers emphasize gender responsibilities for women as mothers and caretakers, which mark their belonging in the domestic sphere and limit the likelihood of their migration while making no such demands on beneficiary men or nonbeneficiaries. Using logistic and multinomial logistic regression models and data from the Mexican Family Life Survey, the analysis finds evidence supporting the hypothesis that conditional cash transfer participation disproportionately limits migration for beneficiary women. This study broadly argues that the impact of such antipoverty programs is more gendered than previously thought and emphasizes the importance of examining previously studied outcomes in ways that consider the specific subject locations of recipients in order to better understand both the logics underlying development policy and the process of migration itself.
Conditional cash transfer (CCT) programs are one of the most widely implemented antipoverty strategies in the world, which had been adopted in more than 65 countries by the early 2000s (Fiszbein et al. 2009). Although their exact form varies slightly by context, CCTs are unified by their basic structure of conditionality, which allocates monetary assistance on the basis that recipients regularly fulfill requirements aimed at building their human capital. Such government-sponsored welfare programs have thus been widely examined, with many migration studies using various economic lenses to hypothesize that CCTs should influence migration for recipients because they sufficiently temper the incentive to leave by increasing beneficiary families’ incomes; facilitate moves by formerly credit-constrained recipients; or help recipients become more skilled and, hence, more marketable prospective migrants over time (Angelucci 2013; Azuara 2009; Stecklov et al. 2005).
Despite generally positive evaluations of CCT success, however, social investment programs such as CCTs are also driven by an underlying maternalist logic, meaning that they attempt to maximize their efficacy by directly relying on beneficiary mothers, rationalizing that mothers are more likely, in turn, to invest in the well-being of their families because of the traditional roles they are expected to play as caretakers within the home (Chant and Sweetman 2012; Rankin 2001; Roberts and Soederberg 2012; World Bank 2011). CCTs often systematically structure the conditionality of their terms such that beneficiary women are held primarily responsible for their household’s success or failure in any given disbursement cycle. Men, on the other hand, are viewed as having very little to no responsibilities under the mandates of CCT participation. As a result, I hypothesize that contemporary antipoverty policy’s reliance on a traditional gendered division of labor to ensure program success will have gender-specific implications for beneficiaries’ likelihood of migration in ways that have not been explicitly considered in previous analyses.
To examine this empirically, I study the likelihood of migration given participation in Mexico’s CCT program, PROGRESA/Oportunidades/PROSPERA (subsequently referred to as Oportunidades). Drawing on longitudinal data from the Mexican Family Life Survey (MxFLS), I use logistic and multinomial logistic regression models to find evidence that supports the hypothesis that Oportunidades constrains the likelihood of migration for beneficiary women but not beneficiary men. The results presented here have two important implications. First, they contribute to theorizations in the gender and migration literature, suggesting that some types of migration can be conceptualized as a unique form of the gendered division of labor made manifest on a spatial scale given the frequent coupling of migration for work and the normative relegation of men to the so-called public sphere. Second, these findings suggest that maternalist development programs are directed by policy logics that underwrite a traditional gendered division of labor that can result in disproportionality across domains of social life and that also hinge on a gendered division of labor. Overall, my findings suggest that previous work examining the influence of CCTs must give greater consideration to gender as a social structure that shapes social phenomena, including both migration and antipoverty interventions.
Beyond “Add Women and Stir”
Migration remains a pertinent global issue that impacts a growing number of sending and receiving countries as well as the people separated by the borders that demarcate them (Castles et al. 2005; Ratha et al. 2016). The types of people who migrate, the reasons for migration, and the absolute magnitude of migration flows suggest the need for research directions that are sensitive to the diversity of those experiences (Brettell and Hollifield 2014). Historically, however, mainstream migration research has often prioritized rational, market-orientated models of human behavior, ranging from the neoclassical theory of migration to the new economics approach (Harris and Todaro 1970; Stark and Bloom 1985; Todaro 1969). For a more comprehensive review, see Massey et al. (1993).
Early and still much contemporary theorizing on migration assumes a young independent male subject whose primary concerns involve work and compensation. This has resulted in the development of migration theories tied to the market as a matter of primacy rather than conceptualizations of migration as a spatial manifestation of other social processes as well, including gender (Mahler and Pessar 2006). Despite a dearth of scholarship prior to the 1970s, reviews conducted by Pedraza (1991), Curran et al. (2006), Donato et al. (2006), Manalansan (2006), Silvey (2006), Mahler and Pessar (2006), Palmary et al. (2010), Herrera (2013), and Hondagneu-Sotelo (2017) charted a rich history of work throughout the past half-century that has moved the field of migration from women’s absolute invisibility, to an “add women and stir” approach, and finally to a promising theoretical place that investigates gender as fundamentally constitutive to the migration process.
Drawing on Connell’s (2014) relational theory of gender, Donato et al. (2006) discussed how gender is constitutive of migration. Beyond merely looking at men or women as individual actors who migrate, gender is defined as relational: constructions of maleness and femaleness must be understood in terms of how they operate as cultural foils to each other, in hierarchies of domination and subordination that are constantly being reinscribed and contested across individual, interactional, and institutional domains. In line with this notion of gender as constitutive of migration, I propose that migration can be reflective of not merely economic arrangements of labor across space but also gendered arrangements of labor across space—arrangements reliant on a gendered division of labor in which men are typically relegated to the public sphere, while women are consigned to the private. The gender division of labor perpetuates the idea that the feminine domain comprises care work and domestic work related to the home (even if it is not in one’s own home), while the masculine domain comprises manual, compensatory, and skilled work outside the home. In this way, the gendered division of labor informs the very meaning, practices, and institutional environments in which migration can take place, creating a political economy and normative context that pulls (or restricts) gendered actors to particular geographic locations to perform specific types of work.
This is not to say that women will be less likely to migrate compared with men because of this division of labor, but rather that it is important to understand how gender can configure labor across space in order to make sense of migration patterns. For instance, migration streams characterized by women’s relocation for domestic work presume women’s innate advantage to perform work associated with the private sphere; these streams thus rely on a gendered division of labor to actively encourage women’s labor migration instead of discouraging it. Conversely, a gendered division of labor also can (and often does) underwrite women’s obligations within their own homes so that they are less likely to migrate. Conceptualized in this way, this study aligns with contemporary attempts in the gender and migration literature to move migration away from being theorized as a singular act conducted by an autonomous individual to being a relational process that is often structured by existing forms of social organization.
Migration, Maternalism, and Conditional Cash Transfers
The Economic Argument for CCTs’ Influence on Migration
I contend that the case of conditional cash transfers and their relationship to migration can be illustrative of how the gender division of labor is constitutive of migration. CCT programs are one of the most widely implemented antipoverty strategies in the world and are unified by their basic structure of conditionality, which allocates monetary assistance on the basis that recipients regularly fulfill requirements aimed at building their human capital (Fiszbein et al. 2009; Lomelí 2008). Failure to comply with such requirements may result in expulsion from the program or, at the very least, in a temporary loss of the cash transfer (Adato and Hoddinott 2010; Handa and Davis 2006; Rawlings and Rubio 2005).
On their surface, CCTs are welfare programs that function as cash infusions into poor households, so it is unsurprising that the research on CCTs has largely taken an economic approach to understand their efficacy and repercussions. In regard to the influence of CCTs on migration in particular, findings have been mixed. Studies by Stecklov et al. (2005), Behrman et al. (2005), Rodríquez-Oreggia and Freije (2012), and Neto (2008) have provided some evidence that CCTs decrease the likelihood of migration. Other studies by Angelucci (2012, 2013), Azuara (2009), de la Rocha (2009), Rubalcava and Teruel (2006), and Winters et al. (2007) have demonstrated positive relationships between migration and program participation. Regardless of the direction of the relationship between CCT participation and migration, however, this overall body of research has tended to theorize the nature of that association only from an economic lens. For instance, Stecklov et al. (2005) argued that by increasing families’ incomes, the cash from the transfer sufficiently tempers the incentive for beneficiaries to leave. Angelucci (2013, 2012) contended that the transfer helps facilitate moves by formerly credit-constrained recipients. Azuara (2009) discussed the long-term effects of human capital investment in helping recipients become more skilled and, hence, more marketable prospective migrants over time. Economic-based explanations of migration behavior in relation to CCT receipt have thus dominated research in this area. Although economic frameworks and other theoretical emphases or considerations, such as gender, are not mutually exclusive, the latter have not been as widely investigated. In fact, economic factors themselves can be understood as gendered. Monetary compensation, for instance, is frequently associated with public sphere employment (i.e., the masculine domain), which often leads to disparities between men and women in terms of the labor they perform and the extent to which they are compensated, if at all. Given that CCTs are essentially contingent cash infusions that largely compensate beneficiary women for performing domestic and care tasks in the service of program goals, past theorizations that have employed economic frameworks without considering gender’s fundamental relationship to CCTs, migration, and beneficiary families have likely captured only a partial picture.
Considering Gender in the Relationship Between CCTs and Migration
This study seeks to emphasize the gendered aspects of CCTs and the probable implications these may have on migration for individuals within beneficiary families. Social investment programs such as CCTs embody two fundamental characteristics of contemporary antipoverty interventions: (1) the focus on rational, market-driven actors who are held at least partially accountable for their own state of poverty; and (2) the intersection of this idea with gendered ideologies that emphasize a traditional division of labor by holding women primarily responsible for the care of beneficiary families (Molyneux 2006; Rankin 2001). Many of these programs attempt to maximize health, education, and other human capital outcomes by directly investing in women, rationalizing that they are more likely, in turn, to invest in the well-being of their families because of the roles they are expected to play as mothers and caretakers (Chant and Sweetman 2012; Roberts and Soederberg 2012; World Bank 2011).
In the case of CCTs, the structure of conditionality formally and informally calls on women to be the ones to physically receive the transfers, perform community chores related to CCT program administration, and ensure the compliance of other household members so that the state continues to subsidize their children’s education and other household members’ health care (Bradshaw 2008; Molyneux 2006). Men, in contrast, are viewed as having very few to no responsibilities under the mandates of CCT participation. Qualitative research in Nicaragua, Mexico, Chile, and Argentina has supported this claim, revealing how responsibilities from CCTs fall disproportionately on women as a signifier of “good mothering” and largely leaving beneficiary men alone, subsidizing a clear division of labor between women in the home and men outside it (Bradshaw and Víquez 2008; Molyneux 2006; Tabbush 2010).
As a result, CCTs heavily rely on a traditional gendered division of labor to implement their antipoverty agenda, which I argue is likely to shape the context in which gendered labor across space will occur. By compensating women for remaining in the private sphere to ensure program success, CCTs likely shift the calculus of migration decision-making to discourage women’s migration out of the home. With the possibility of losing welfare money or expulsion from the program entirely, beneficiary women may feel far less inclined to move, whether because they no longer have the need to do so (given that women are often considered “migrants of last resort”) or because they feel obligated to stay. Comparatively, beneficiary men are likely to be significantly less constrained by CCT requirements merely because very few are placed on them.
Hypothesis 1 (H1): Within beneficiary families participating in a CCT program, women will be significantly less likely to migrate compared with men.
Beyond their potential influence on this division of labor within beneficiary households, CCT programs themselves, ceteris paribus, may also have an influence on the likelihood of beneficiary women’s migration compared with nonbeneficiary women. Tying program obligations to maternalist gender expectations is hypothesized to intensify these obligations for beneficiary women above and beyond what would be the case for their nonbeneficiary counterparts.1 Hence, I additionally hypothesize the following:
Hypothesis 2 (H2): Women who participate in CCT programs will be significantly less likely to migrate compared with nonbeneficiary women.
Why CCT participation is expected to limit the likelihood of women’s migration is argued to occur via the three following pathways:
Physical presence constraints: Participation requires predominantly women to remain at the place of origin. Women are generally expected—and often required—to be the physical recipients of the cash transfers themselves in the home communities in which they originally enrolled. For the Mexican case investigated here, recipients are often required to provide an electronic fingerprint every time a cash transfer is disbursed, thus preventing other family members from assuming that responsibility. In some cases, they must also attend maternal and child health lectures as well as complete community service requirements related to CCT program event setup and cleanup. The requirement that beneficiary women remain at the place of origin on a regular basis to fulfill program conditions may thus limit the likelihood of their migration.
Hypothesis 3 (H3): Beneficiary women who directly receive the transfer from the CCT program will be significantly less likely to migrate than others who are not the primary recipients.
Increases in domestic and care work: Participation also requires additional informal work from beneficiaries—particularly women—which are not formal requirements of the program itself but are associated with its inherent obligations. For example, ensuring that children regularly attend school as a formal requirement may mean that women must walk their children to school, help with homework, attend school meetings, and take over more of the chores normally assigned to those children now attending school. Beyond just physical presence, the additional household and care work should also limit the likelihood of women’s migration by increasing certain responsibilities that only they, to some extent, are expected to meet.
Traditional gender ideology reinforcement: CCTs moralize traditional gender expectations, equating being a “good” woman and being a “good” mother to beneficiary women’s successful management of program requirements. Bradshaw and Víquez (2008) discussed this tie between the maternalist expectations aimed at women in general (i.e., caring for children and other household dependents, cleaning the home, and so on) and the maternalist logic of CCTs specifically, which builds on, reproduces, and perhaps intensifies these preexisting gender expectations. This aspect of gender ideology, which emphasizes women’s place within the home and men’s place outside it, should additionally limit beneficiary women’s migration.
Hypothesis 4 (H4): The direct effect of CCT participation on women’s migration will be mediated by the measure of domestic and care work as well as the measure of traditional gender ideology (i.e., the number of hours spent within the last week on cooking, cleaning, and childcare).
Because the data used for the analysis do not include direct measures of the proposed mechanisms, proxies were used. The rationale for choosing direct receipt and hours spent on chores as indirect measures is provided in the Methods section.
Setting: Mexican Case Study
In 1997, Mexico piloted its CCT program, Oportunidades, which has been heralded as one of the most successful CCT programs in terms of both population coverage and observed outcomes (Barber and Gertler 2008; Behrman et al. 2005; Fernald et al. 2008; Skoufias et al. 2001). Oportunidades is an ideal case to study. Having served as a model for global emulation, its design is representative of most other CCT programs around the world.
Oportunidades began by first offering two kinds of assistance to beneficiaries—a food grant and an education scholarship—both dispensed bimonthly. The food grant for the amount of US$16 per month (in 2001 dollars) would be distributed to families if all members attended an annual clinic check-up and some members attended regular public health information sessions (Stecklov et al. 2005). The education scholarship functioned slightly differently: the amount the family received varied by the number of children present. All children aged 7–18 qualified to participate and were required to attend school up to 85 % of the time each month. Failure to meet the attendance requirement would lead to a loss of the cash transfer for that month for that particular child. By 2001, the program expanded considerably to offer additional subsidies and grants, including financial incentives for adolescents to finish high school, cash transfers to purchase school supplies and uniforms, nutritional supplements for infants, and cash transfers for senior citizens (Secretería de Desarrollo Social 2014). Oportunidades covered more than 4 million families in 2002, the year of the first survey wave utilized in this analysis (Stecklov et al. 2005). The amount of money directly received from the program is substantial, estimated to amount to an average of 22 % of a family’s original pre-participation income (Angelucci 2013).
The program also relies heavily on the informal work of women. Rivero (2002) reported that some beneficiary women felt as though the program “treated [them] badly” or that they “were asked to do things in ways that offended their dignity,” including seemingly unnecessary chores or tasks. Fieldwork that I conducted involving interviews with 35 beneficiary women supports this as a widespread sentiment shared by 65 % of the sample. One woman expressed that the program “felt like a second job” and that she and her beneficiary neighbors were asked to “work an extra ten hours per week” to meet program requirements.2
Data and Methods
Data for this analysis are from the Mexican Family Life Survey (MxFLS), which is the first nationally representative survey of the Mexican population. The baseline sample design is probabilistic, stratified, and multistaged in which every phase is an independent sample of households in Mexico during 2002. The analysis draws on data from Wave 1 (2002) and Wave 2 (2005–2006), taking independent variables from Wave 1 and dependent variables from Wave 2 to account for time ordering issues. I limit the sample to those who could plausibly be treated as adults (ages 15+) and thus could conceivably be making their own migration decisions. The final sample includes a total of 9,880 unique men, 11,923 unique women, and 8,406 unique households. Only individuals present in Mexico during Wave 1 and who were present for the survey collection are included in the analysis.
MxFLS is characterized by a moderate percentage of missing observations along some variables, particularly that for migration. Wave 2 managed to relocate and reinterview about 90 % of the original sampled households, the remainder of which can partially account for the missing observations. I partly address the issue of missing cases by running the analysis separately on a listwise deleted data set and on a multiply imputed data set created via chained equations using the mice package in R (van Buuren and Groothuis-Oudshoorn 2011). The imputed data contain imputed missing information for both the dependent and independent variables. Only results from using the imputed data set are reported because the results of the two data sets were substantively similar. Additionally, given the complex survey design on migration behavior that is characteristic of MxFLS, some explanation of the migration variables constructed for the purposes of the analysis is provided in section A1 of the online appendix.
The analysis uses logistic and multinomial logistic regression models focusing on the first observed migration event, divided into separate analyses for any international migration, any domestic migration, and any migration. Although many theories focus on international migration or treat any migration as equal, the motivations for and dynamics informing international versus domestic migration decisions may vary considerably, so the following analysis considers each separately as well as together.
For the household-level analysis, a multinomial logistic regression model specification is presented for three definitions of migration (i.e., domestic, international, and any migration), including baseline controls constructed at the household and community levels. The four possible multinomial outcomes include whether in the follow-up survey wave men and women within the same household migrated, only men within the same household migrated, only women within the same household migrated, or no household member migrated. Only men from the household having migrated is the reference category for ease of comparison. The aim of the household-level analysis is to account for migration risk within beneficiary and nonbeneficiary families. Amount received from CCT participation is controlled so that the primary covariate (i.e., household CCT participation) can capture the direct effect of program participation beyond the financial amount received. This analysis is intended to test H1, which states that women within beneficiary households should be significantly less likely to migrate than men in those households. In the results, support for this hypothesis would be an effect of CCT participation significantly restricting only women’s migration from the household relative to only men’s migration, with agnostic predictions for no household member migrating or both men and women from the same household migrating.
For the individual-level analyses, five model specifications are presented for both men and women across three outcomes: international, domestic, or any migration.
The first model is a simple baseline including all controls.
The second model includes the baseline specification alongside a measure for household CCT participation, in which an individual is coded as participating if anyone in their household claims to be a direct beneficiary. This measure for household participation is used to illustrate the total direct effect of the program on migration for anyone living in a household that participates.
The fourth model includes the baseline specification and the measure for time spent on chores and childcare, which serves as a proxy for additional domestic labor required by CCT participation and gender ideology. The estimated coefficient for chores in this model merely serves to illustrate the direct effect of chores on migration.
The fifth and final model includes the baseline specification, the measure for individual CCT participation, and the measure for chores. As a test of H4, which hypothesizes that the direct effect of Oportunidades participation on women’s migration will be mediated by the measure for domestic and care work as well as the measure for traditional gender ideology, I again use a Clogg test; this test determines whether the estimated coefficient for individual CCT participation in Model 3 is significantly different and greater than the coefficient for individual CCT participation in Model 5.
It could be argued that women who are more gender-role conforming to begin with are also more likely to participate in the program and therefore less likely to migrate overall. Given that no direct measure for gender ideology is included in the survey, the analysis uses the number of hours spent on household chores and care because this measure is often used to control for gender ideology in the family demography literature (Shelton and John 1996). Time spent on chores and care is also included as a test of the remaining mechanisms because chores in one sense capture the actual performance of domestic work and in another sense capture the performance of gender. As a result, the actual direction of the coefficient estimate is expected to be negative, and the inclusion of the covariate itself is expected to reduce the magnitude of the estimate for individual CCT participation among women.
Separate analyses for men and women are conducted in correspondence to other studies focusing on gender and migration, which operate under the premise that the estimated coefficients for every covariate should systematically differ across gender because of the divergent processes these two groups experience (Curran et al. 2006). Because of the disproportionate bias conventionally produced by clustering standard errors for nonlinear parametric models, I do not use clustering methods to adjust the standard errors (Greene 2011). Given the separation of the analyses by gender anyway, accounting for clustering across couples is less of a concern. Although clustering within families or households might still be an issue, I attempt to address that issue by both including and removing repeat members within households for gender-separated subsamples to look for any substantive difference in the results. There are no such differences, so only results from the analysis using the sample including all members in the household are reported.3
To illustrate the findings graphically, I calculate counterfactual simulations of predicted probabilities for each of the groups of interest (i.e., beneficiary women and nonbeneficiary women; beneficiary men and nonbeneficiary men) using the Zelig package (Choirat et al. 2015; Imai et al. 2008); these are plotted in Figs. 1 and 2 for ease of comparison. Given the inability to compare predicted probabilities across models and given that the analysis relies on gender-separated samples, I compare women only with other women and compare men only with other men. To specify the counterfactual scenarios for the predicted probabilities themselves, I vary respondents’ reported sex and direct program participation status for each group’s predicted probability while keeping all other values equal to their full sample median values (in the case of numeric variables) or to their full sample modal values (in the case of categorical variables). Confidence intervals are estimated for each predicted probability via simulations produced by the Zelig package. Only predicted probabilities are plotted for the household-level analyses. Confidence intervals are not plotted because of limitations in the ability of the statistical packages to simulate confidence intervals for multinomial logistic regression results from imputed data from which some variables are constructed post hoc.
In terms of the covariates themselves, most controls used in the analysis are similar to conventional measures of sociodemographic traits in other migration studies, including measures for age, marital status, position within the household, education, income, assets, employment over the past year, school enrollment, access to credit, economic shocks, amount of remittances and other transfers received, intentions to move, having relatives already living in the United States, self-reported health, whether their home has a flush toilet, migration experience and networks, and rurality (Massey and Espinosa 1997; Stark and Taylor 1991; VanWey 2004, 2005). Further explanations for some variables’ inclusion on theoretical terms are discussed in section A2 of the online appendix. Table 1 includes descriptive statistics for all final variables used, and Table 2 includes the definitions of each variable.
CCT Participation and Gender
For the household-level analysis, the results of interest are the estimated coefficients for whether a household participated in Oportunidades. If CCT participation lowers the likelihood that only women from the same beneficiary household migrate relative to only men migrating from the same beneficiary household, this would support H1. Compared with only men from the same household migrating, only women migrating have a lower log odds, at –0.29 (p < .001, Table 3) for international migration, = –0.77 (p < .001, Table 4) for domestic migration, and –0.42 (p < .001, Table A1 in the online appendix) for any migration. Predicted probabilities calculated from varying program participation while all else is held equal are shown in Fig. 1 to illustrate these findings visually. Ultimately, the results provide support for H1 and the theoretical premise that CCTs reduce the likelihood of migration for women in beneficiary families but not for men. In fact, the predicted probabilities for beneficiary men shown in Fig. 1 for all migration types appear to suggest that being a CCT beneficiary may increase the likelihood of migration for men, corresponding to similar findings by Angelucci (2012, 2013), Azuara (2009), de la Rocha (2009), Rubalcava and Teruel (2006), and Winters et al. (2007).
For the individual-level analysis, the estimated coefficients for the covariates are relatively consistent across all five models for each of the gender- and destination-disaggregated analyses. The estimates also generally align with similar findings and theory in the literature on Mexican migration. Looking to Model 3 across the individual-level analyses for women, significant and negative estimates for individual CCT participation would support H2, which predicts that beneficiary women will be less likely to migrate compared with nonbeneficiary women, ceteris paribus. Compared with nonbeneficiary women, women who are direct recipients of the program are estimated to have lower log odds of migrating: –0.06 for international migration (Table 5), –0.48 for domestic migration (Table 7), and –0.36 for any migration (Table A2, online appendix). Although the estimate in the model for international migration is not statistically significant, both the estimates for domestic and either migration are significant at the p < .05 level. In contrast, men who are direct recipients of the program do not significantly differ from other men across any of the migration outcomes (Tables 6 and 8, and Table A3 in the online appendix), most likely because fewer of them are tasked with being direct recipients to begin with or, if they are given that title formally, they are not impacted by receipt in the same ways that beneficiary women might be.
The results generally support the hypothesis that beneficiary women who directly receive the transfers are less likely to migrate compared with their nonbeneficiary counterparts, whereas beneficiary men are not significantly different from their nonbeneficiary counterparts. Based on the final model specification in all the individual-level analyses, Fig. 2 illustrates the test of H2 by plotting predicted probabilities of migration for beneficiary and nonbeneficiary women, disaggregated by migration destination type. It also plots the predicted probabilities of migration for beneficiary and nonbeneficiary men, disaggregated by migration destination type. The center points represent the predicted probabilities themselves, and the whiskers show the bootstrapped 95 % confidence intervals.
Physical Presence Constraints, Housework and Care Work, and Gender Ideology
As for the tests of the proposed mechanisms, the first comparison of interest lies in the estimated coefficient for household CCT participation in Model 2 and individual CCT participation in Model 3 for all the individual-level analyses. The expectation is that individual CCT participation (which measures whether the respondent directly receives the cash transfer and therefore must be subject to more of the locational constraints mandated by the program) should have a significantly larger and negative effect on migration compared with household CCT participation (which counts respondents as being a CCT recipient regardless of whether they specifically are the ones who directly receive the money). A significant difference in this regard would show that being the one held directly responsible for the cash transfer and other program obligations especially reduces one’s likelihood of migration. For women, individual CCT participation decreases the log odds of international migration by –0.06, compared with log odds of 0.26 for household CCT participation. Individual CCT participation decreases the log odds of domestic migration by –0.48, compared with –0.38 for household CCT participation. Finally, individual CCT participation decreases the log odds of any migration by –0.36, compared with –0.17 for household CCT participation. The calculations of z scores for the Clogg test are –0.46 for the international analysis, –0.16 for the domestic analysis, and –0.34 for the any migration analysis. Although the actual estimated coefficients and direction of the Clogg test z scores correspond to the hypothesis stated in H3 (in which direct participants are less likely to migrate compared with all those who participate in any given household), the results from the Clogg test do not show that the difference between them is statistically significant. In the case of men, because none of the estimated coefficients for men’s migration are significant, as expected, direct receipt is not expected to matter anyway.
For the second proposed mechanism of interest, the focus is on the difference between the estimated coefficient for the influence of individual CCT participation in Model 3 and the influence of individual CCT participation in Model 5 when hours of chores and care work are controlled for. The actual magnitude of the coefficient for hours of chores and care work is relatively small—as demonstrated by the log odds of –0.02 for women’s domestic and any migration and insignificant log odds for international migration—so it is not surprising to see that including it in the full model does not dramatically change the estimated effect of individual CCT participation. For women, the change from having no measure of chores in the model to including it goes from –0.06 to –0.05 for international migration, from –0.48 to –0.45 for domestic migration, and from –0.36 to –0.33 for any migration. Although the attenuation of the direct effect of individual CCT participation once chores are included corresponds to the expectations in H4, the z scores from the Clogg test show that this reduction is not statistically significant: the z score is 0.05 for domestic migration, 0.05 for any migration, and 0.01 for international migration. Again, in the case of men, because none of the estimated coefficients for men’s migration are significant, as expected, chores and care work are not expected to matter anyway.
Although both of these comparisons appear to correspond to H3 and H4, none of the Clogg tests are statistically significant. I discuss possible explanations in the upcoming Limitations section.
The estimated coefficients for the sociodemographic controls are also relatively consistent across all analyses and tend to align with similar findings and theory in the Mexican migration literature. For the household-level analyses (Tables 3 and 4, and Table A1 in the online appendix), access to credit, assets, having more adult women and more young children in the household, and living in a rural area all significantly and positively influence migration for both men and women and for only women relative to only men. Having more men in the household and more children ages 6–14 decreased the odds of migration for these groups compared with only men. The influence of the remainder of the covariates varies by migration destination type and comparison group.
For the individual-level analyses (Tables 5, 6, 7, and 8, and Tables A2 and A3 in the online appendix), the odds of migrating slightly decrease with age and are significant. Having a paid job is associated with lower odds of domestic migration for women, which is significant. Being currently enrolled in school has a significant and negative effect on migration for both groups, with its effects varying in significance. Effects of educational attainment vary, although women generally appear to be more likely to migrate if they have at least some high school education, and men with at least some high school appear less likely to migrate than their counterparts who have less education.
Having relatives living in the United States positively and significantly increases the odds of migration for both men and women. Having thought about moving in the past positively and significantly influences both men’s and women’s odds of migrating to either type of destination. Assets tend to negatively and significantly reduce the odds of migrating, whereas the effects of income tend to vary in direction and significance. Having a flush toilet generally has a large and positive influence on the odds of both men’s and women’s migration. The presence of younger and older children has varying effects for men and women for the different types of migration.
Migration network variables tend to have large, positive, and significant estimated effects for both women and men. Rurality tends to negatively impact the odds of domestic migration but positively impact the odds of international migration for both men and women. The influences of the remainder of the covariates vary by migration destination type and group but are largely not significant.
The operationalization of the mediating variables proved to be a primary limitation to this study. Gender ideology could have been better measured by directly asking respondents about their values and beliefs regarding gender-role expectations, but no such measure exists in MxFLS. Although the household decision-making measures in the survey could have been used, they were asked for only those people who were presently married or cohabitating, so their use would have resulted in systematic missing information for one-half the sample. Nonetheless, chore hours are an imperfect measure of gender ideology because hours spent on chores can vary based on other factors, such as time availability and access to outside help. However, the hope is that its use alongside controls such as household composition, employment status, and marital status allows for what remains of its estimated effect to partially isolate the influence of gender ideology. Additionally, chore survey items asked respondents only about how much time they spent on domestic tasks and caring for children and seniors, not necessarily any time potentially spent on Oportunidades-specific tasks, such as walking to clinics, attending public health lectures, or traveling to cash pickup locations. As a result, this measure imperfectly operationalizes what would have been additional time demands imposed on beneficiaries by the program. Finally, the analysis encountered a rare event issue in which both migration—particularly international migration—and program participation resulted in relatively small cell counts. The corresponding standard errors derived from the analysis using these variables, therefore, are relatively large and may have affected calculations that include them, such as the Clogg test. Because better measures than the ones available and higher event rates are not possible in the MxFLS, imperfect operationalization and small cell problems may partially account for the lack of significance in the analysis.
The results from the analysis support the contention that participation in Oportunidades—and plausibly other CCT programs—disproportionately reduces the likelihood of migration for women compared with men as well as nonbeneficiary women. The theoretical framework presented earlier is also partially supported by these findings, although the proposed mechanisms were not found to be statistically significant. Grounded on the claim that it is important to take a gendered perspective when studying the relationship between CCTs and migration, the framework presented notes the lack of a gendered approach in past research and emphasizes the integral role of gender in shaping both migration decision-making and the division of responsibilities imposed by maternalist antipoverty programs. Program expectations draw on preexisting gender ideologies that rely on a gendered division of labor to tie being a “good” woman and a “good” mother to being a responsible CCT beneficiary. By drawing on these gendered expectations, maternalist programs can in turn play a part in the process of social reproduction in which the ideologies themselves and the division of labor they underwrite work to reproduce, reinforce, and perhaps intensify traditional organizations of gendered labor.
Although CCT maternalism can structure inequitable outcomes patterned by gender, this particular study does not test for that possibility and thus makes no claims in that regard. Despite evidence that the maternalist structure of CCTs reduces the likelihood that beneficiary women migrate, migration is a complicated process that runs the gamut from agency to coercion. In some circumstances, lowering the likelihood of migration may benefit women in their everyday lives given that at least some migrations are disruptive events that serve as last resorts in light of no other options. In other cases, incentivizing recipients with cash transfers and citing maternal duty in their service may remove women’s options to move or work outside the home, which in their own right have been shown to increase women’s autonomy. It is also entirely and simultaneously possible that CCTs grant women greater autonomy and power in other ways (Franzoni and Voorend 2012). Having more money and perhaps greater discretion over how CCT funds are spent can serve as a form of power for beneficiary women. One could argue that CCT disbursements at least partially compensate women for work that they would be expected to accomplish without pay anyway. As a result, no value claims are being made in regard to whether migration or CCTs themselves are good or bad, although maternalism itself should be interrogated more critically.
Ultimately, what the results do support is the claim that gender must play a more prominent role in theorizations of both migration and development policy, specifically CCTs. For the former, the present analysis provides support for how gender can be constitutive of migration, seeking to conceptualize migration as a process that often reflects gendered arrangements of labor across space. In contrast to former approaches that have typically framed migration as a behavior primarily reflective of market forces, this revision aims to broaden the general understanding of migration to more fully incorporate the dimension of the social. This expansion calls not merely for the consideration of women within existing migration theory or the framing of migration as a process that differentially affects men and women. Instead, it attempts to demonstrate that gender itself constitutes how we understand who should perform certain types of labor and where they are expected to exist in public and private life.
Likewise, for theorizations on the consequences of maternalist development policy, this study also aims to continue the critical work of incorporating gender into conceptualizations of CCT program design and the implications that follow. Although many previous studies on CCTs have evaluated its successes by focusing on specific outcomes (e.g., educational attainment, health care seeking, consumption smoothing), they have not necessarily engaged with the critical literature on CCTs that focuses more exclusively on program maternalism and gender bias in their own analyses. This study is one of the first to both acknowledge the valuable insights of the critical literature on CCTs and to apply them to disaggregate and study a specific outcome.
More research on how particular programs affect various outcomes for categorically disaggregated groups is needed to advance research in these areas. Although the present study focuses on gender and migration, scholarship that wishes to continue this work would ideally consider how not only gender but also other social categories work to shape the experiences of their embodied subjects across a number of outcomes, including (but not limited to) migration, education, health, nutrition, marriage, and reproduction.
The author thanks the anonymous reviewers for their invaluable comments as well as Nathalie Williams, Kyle Crowder, Aimée Dechter, and Hedy Lee for their guidance throughout this process. Lindsey Beach, Frank Edwards, Hannah Curtis, Thiago Marques, and the University of Washington’s Gender and Sexuality Research Cluster are also due a hearty thank you for providing the intellectual and social community so desperately needed to see a project like this from start to finish. A special thank you must be given to María Vignau Loría, who assisted with field work in Mexico and is generally a lovely person as well as to the Department of Sociology at the University of Washington for providing funding toward that effort. A final and much deserved thanks must also be given to Justin Goodman, who was not here for the start of this project but has certainly ushered it to its finish.
Perhaps more gender role–conforming women opt into these programs compared with less–role-conforming women, creating a selection mechanism that may bias the results. To account for this possibility, a measure for gender ideology is included in the model specification.
During the summer of 2014, I conducted fieldwork along with a research assistant who aided with translating and conducting interviews. We interviewed 35 beneficiary women of Oportunidades and 25 nonbeneficiary women across six randomly selected neighborhoods within the Mexican state of Oaxaca as well as Distrito Federal. Interviews lasted for an average of about 30 min and covered topics ranging from gender attitudes, family background, everyday routines, employment history, and CCT program participation (or nonparticipation). The median age of interviewees was 26; 30 resided in rural or semirural neighborhoods, and 30 resided in the urban outskirts of Mexico City. Interview transcripts were first open-coded and then thematically coded to identify larger themes in attitudes and experiences.
Multilevel models are also used in other analyses in which men and women are combined in one sample. However, once the sample is divided by reported sex, there are not enough clusters by family or household to allow for a multilevel analysis because of insufficient degrees of freedom. Given the importance of estimating each coefficient separately by reported sex, I choose simple logistic regression models over a multilevel specification.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.