Temporary migration as a mechanism for lasting cultural change: evidence from Nepal

Achieving gender equality and empowerment of all women and girls is one of the 17 sustainable development goals of the United Nation. Beliefs about women's roles in society or the home often make achieving these goals challenging. Despite how entrenched beliefs about gender may be today, evidence suggests that beliefs may partially be a cultural by-product of idiosyncratic historical conditions. For example, the historical reliance on plow agriculture may have caused women to specialize in household production (Alesina et al., 2013), or reliance on pastoralism may have led some societies to develop practices to restrict women's sexuality (Becker, 2019). This raises the following question – can this process work in reverse? In other words, can external conditions or shocks also cause gender norms to erode? Can societies unlearn gender roles through even temporary experiences with more-equitable roles? Motivated by these questions, our paper studies whether temporary migration spells of husbands in rural Nepal leads couples to permanently adopt more-equitable norms in decision-making. Importantly, we also study the economic consequences of this shift in decision-making.

We use panel data on rural Nepali households to study how a husband's migration interacts with intrahousehold decision-making power and expenditure decisions within households over time. We show that during migration spells, decision-making power (both sole and joint) shifts from a husband to his wife. The shift is large and represents 10% of total decisions. In the Nepali context, this is not an a priori obvious result – Nepali married couples often live with the husband's parents, who typically command some authority in the household and could easily step in for their son upon his leave. Moreover, while husbands are migrating, they still report exerting some control over decisions, suggesting that this is not merely mechanical.

When decisions shift to the wife, we provide evidence that households decrease their spending on temptation goods (alcohol and tobacco products) by 105–120%. We provide some evidence that this reduction is offset by increased spending on children's education and clothing, although we interpret those results with some caution. Insofar as these reallocations are caused by the shift in decision-making power, this demonstrates how altering gender roles in decision-making could have meaningful economic impacts on developing countries.

Of course, such a change in spending habits could easily be attributed to an income effect. Higher wages motivate migration, and labor market or migration frictions likely mean that substantial wage arbitrage opportunities persist across locations. Naturally, we find that migration spells increase household income. To assess whether the spending changes are attributable to an income effect or changes in decision-making power, we exploit the fact that many households allow their sons to migrate. Although sending sons for migrant work increases household income, we do not observe changes in decision-making power or shifts in expenditures similar to the ones when husbands migrate. This suggests that the shifts in expenditures that occur when husbands migrate is specific to the migration of a husband (a primary household decision-maker) rather than income effects induced by migration spells in general. In other words, decision-making appears to be the channel that mediates the changes in expenditures that we observe.

Women's decision-making power is a key component of women's economic empowerment. Kabeer (1999) defines empowerment as the process by which people expand their ability to make strategic life choices, particularly in contexts in which this ability had been denied to them. Therefore, decision-making power is an important outcome in and of itself: if male migration triggers women's participation in household decision-making, then women's empowerment could be an important welfare implication of migration.

Moreover, decision-making may mediate other outcomes, such as how the household spends money. If women gain meaningful decision-making power, then assuming the husband and wife do not share identical preferences, we also expect changes in household economic decisions. In fact, many programs target female beneficiaries based on an expectation that women prefer to invest more in children (Duflo, 2012; Handa, 1996; Hoddinott and Haddad, 1995; Thomas, 1990), and specifically girls (Duflo, 2003; Qian, 2008; Thomas, 1990, 1994). Along these lines, several studies have demonstrated increased investment in children's education following migration (Antman, 2011, 2012; Edwards and Ureta, 2003; Yang, 2008),

This finding is not universal: e.g., McKenzie and Rapoport (2011) provide evidence of reduced educational attainment among migrant households in Mexico.

Our work lies at the intersection of literature on gender and migration. Much of the work at this intersection studies how female migration interacts with women's agency and gender inequality (e.g., Hughes, 2019; Parrado et al., 2005; Stecklov et al., 2010; Sultana and Fatima, 2017). Our results specifically contribute to a growing literature studying how migration affects household decision-making among female family members left behind. Clemens and Tiongson (2017) exploit a natural experiment and use a regression discontinuity design to show that when household members migrate from the Philippines to Korea, women take on greater responsibility for household decisions. Furthermore, they spend more on education, health, and quality-of-life goods (e.g., ceremonies); they also borrow less, save more, and are more likely to send children to private school and visit private clinics. They present suggestive evidence that these effects work through not only remittances but a change in household decision-making power. Antman (2015) finds a similar pattern in Mexico: decision-making power shifts in favor of the spouse left behind during a migration spell, and the household tendency to devote a larger share of resources to girls rather than boys during this period. In contrast to these findings, Göbel (2013) finds no evidence of differences in how female- and male-headed households spend remittances that their migrant spouses send home. Other work has highlighted that geographic separation could introduce asymmetric information about household decisions, therefore allowing the wife to make de facto independent decisions about the household budget (Chen, 2013; Ambler, 2015; Ashraf et al., 2015).

A concern in interpreting our results is that migration is not a randomly assigned event: households select into migration. Moreover, this decision may be correlated with other factors that also affect household decision-making or finances. For example, one concern may be that households selecting into migration are more likely to see changes in decision-making power, i.e., migrant households are more open to adopting more-equitable norms. Past work suggests that, if anything, the bias goes the opposite way. Nobles and McKelvey (2015) find that when women are more in control of household resources, husbands are less likely to migrate. This suggests that households in which the husband migrates are not likely to be the type of households in which women's empowerment grows over time; in fact, they may be the opposite type of household, thus working against our results.

We address these issues in several ways. Similar to Antman (2015), our main empirical approach uses household fixed effects (FE) and thus compares households to themselves over time as the husband's migration status changes. This approach controls for any unobserved time-invariant differences among households. Selection into migration – likely on the basis of youth, earnings potential, risk tolerance, or other unobservables that could easily affect our outcomes – will be controlled for by these FE. Moreover, we include a specification whereby we limit our sample to households in which the husband migrates at some point, therefore conditioning on selection into current/eventual migration (so only selection into the timing of migration spells remains as a confounder). Lastly, we include household-specific time trends to control for unobserved time-varying factors, such as differential trends in consumption. While our estimates rely on several assumptions, we argue that our empirical approach controls for the factors that most plausibly threaten our interpretation. Finally, we argue that the narrative as a whole provides a compelling case that our results are not simply spurious.

Our novel contribution to this literature is to highlight the potential for persistence of these effects (which past work has not studied) and to offer an explanation for this. We explore persistence by identifying asymmetries in the magnitude of behavioral changes when the husband leaves vs. when he returns. This decomposition allows us to see whether decisions return to their original, premigration-spell trend on the husband's return. Following a migrant's return, we find suggestive evidence that husbands resume their role in economic decisions, but decisions are more likely to be made jointly. Despite this, expenditures appear to return to premigration levels, which suggests that women may lack bargaining power when making joint decisions. Nevertheless, a lasting change in who participates in decisions is an important first step toward more equality in decision-making. This finding underscores the potential role of temporary migration spells in triggering lasting cultural change around gender norms.

We show how these persistent effects can be explained by a habit-formation model (e.g., Becker and Murphy, 1998). Migration disrupts the household's usual decision-making process, which necessitates that the household form new habits. During this time, the household “learns by doing” – e.g., couples may realize that women are indeed capable of making financial decisions. Preferences slowly change over who makes decisions, and the new habit of more-equitable decision-making starts to stick. The longer a migrant stays away, the more likely this habit forms. Consistent with this narrative, we find that the contemporaneous changes in decision-making power are concentrated in households that experience longer (at least 6 months) migration spells. This result helps rule out an alternative explanation for persistence – that migrants “bring home” norms or ideas from the country they visit (Batista and Vicente, 2011; Beine and Sekkat, 2013; Bertoli and Marchetta, 2015; Böhme, 2015; Lodigiani and Salomone, 2020).

In addition, as we discuss, Nepali migrant destinations are not likely to be places with less extreme gender norms.

Identifying the habit-formation mechanism broadens the policy impact of our paper by illustrating how exactly could policies be designed to improve women's empowerment. This contributes to a growing literature studying how policies can be effectively designed to change pervasive gender norms. For example, correcting individual beliefs about others’ privately held beliefs on gender may lead to less distortion of female labor supply in Saudi Arabia (Bursztyn et al., 2020). Similarly, giving women individual bank accounts, therefore incentivizing them to work, may indirectly change gender norms (Field et al., 2021). Our work suggests a new mechanism that policies aimed at similar goals could apply: promoting learning and habit formation through a temporary disruption in the decision-making process may facilitate long-term cultural change.

The remainder of the paper is organized as follows. We begin, in Section 2, with a description of the unique Nepali context and our data, which underscores the policy relevance of our analysis. In Section 3, we present our main empirical strategy and discuss our identifying assumptions. Results are presented in Section 4. In Section 5, we discuss how our results can be explained by a model of habit formation. We close in Section 6 with some concluding remarks.

The relationship between migration and gender norms is of considerable interest in Nepal and throughout much of South Asia. In 2015, remittances made up >30% of Nepali gross domestic product (GDP) – a share considerably higher than in most countries (World Bank Group, 2018). As a South Asian country, Nepal is home to gender norms that often constrain women's opportunities in the labor market, education, or autonomy over decisions (Asian Development Bank [ADB], 2010).

See Jayachandran (2015) for a review of how gender norms, especially those present in South Asia, might explain gender inequality.

Our main data source is the “Heifer Nepal Smallholders in Livestock Value Chain” impact evaluation panel survey data collected in 2014, 2016, 2017, and 2018 from rural Nepali households residing in communities across seven districts (see Figure A1 in the Appendix for a map). In each round of the survey, female respondents were asked about household demographics and finances, economic activity including migration status of household members, and economic decision-making. We include only those households in which a married, female respondent was interviewed in the first round and at least one other time. Our full sample is an unbalanced panel of 2,508 households.⁴

The survey data were collected for the evaluation of a livestock transfer program. Details of the randomized program evaluation process are provided by Janzen, et al. (2018). One concern may be that this program affected migration, leading us to conflate migration spells with program treatment effects. However, the treatment is not correlated with the decision to migrate, mitigating this concern.

The correlation coefficient between a randomized treatment dummy and a dummy variable indicating that the husband migrated in the last year of our data is 0.01.

2.2

Decision-making

Our decision-making data are based on a survey question (posed to the wife) that asks who in the household makes the decision regarding a particular expenditure. We first create variables for each good that describe who makes the decision – the husband alone, wife alone, or husband and wife jointly.

While the survey allowed for additional compositions of decision-making power between household members, we focus on cases where either the husband has sole control, the wife has sole control, or the husband and wife jointly make the decision. This implies that for a given category, the three indicators need not sum to one. Note that the compositions allowed changes between survey rounds, which required us to harmonize survey responses over time; see Section C2 of the Appendix for more details.

Then, we take the average of the indicators across goods within each category; the final variable represents the fraction of goods within a category over which the husband, the wife, or both have control. Expenditure categories include temptation goods (alcohol and tobacco), formal health care, ceremonies and celebrations, children's education, children's clothing, and adult women's clothing.

See Section C1 of the Appendix for details about the items that compose each category, which changes slightly between survey rounds. When only one item makes up a category, the outcome variable is binary. It is missing only when the entire set of items in its category are missing, which happens when the respondent replies “I don’t know” or refuses to answer to the question. Moreover, when respondents answer “not applicable”, meaning no decision was made, we code all decision-making outcomes for that item as zero. Furthermore, we report summary statistics for each expenditure category broken down by migrant status in Section C1 of the Appendix in Table C1.

Across all households in our sample, 34% of expenditure decisions are made jointly, while men alone decide 14% and their wives alone decide 11%. But, averaging across households hides substantial heterogeneity by the husband's migrant status. Figure 1 reveals a striking pattern that illustrates the main narrative of our empirical results. Before husbands migrate, 27% of decisions are made jointly, while 18% are made by the husband alone and 11% are made by the wife alone.

Figure 1

Data are pooled 2014–2018 values. The figure depicts the mean fraction of all expenditure decisions (considered in this paper) that are made either by the individual or jointly as indicated; 95% confidence intervals are also displayed. See Section C1 of the Appendix for more details on how each expenditure category is constructed. “Future” indicates that the husband will migrate in the next period, “current” indicates migration in the current period, “past” indicates that the husband migrated in the previous period, and “never” indicates that the husband did not migrate during any of the survey rounds. These groups are mutually exclusive and, therefore, we omit the 102 observations that both migrated in the previous period and will migrate in the future period.

During migration spells, decision-making power shifts in favor of the wife as she becomes more than twice as likely as before to make sole decisions. Although she is unlikely to retain newly acquired sole decision-making power following her husband's return, a much higher share of subsequent decisions (47%) are made jointly. This anticipates our main finding: a husband's migration spell seems to increase his wife's involvement in household decision-making, which she partially retains upon his return. Moreover, these summary statistics highlight why decision-making is not simply mechanically related to migration status – clearly, husbands remain involved to some extent upon migrating.

The key empirical challenge for this analysis is that the decision to migrate is not randomly assigned. It therefore may be correlated with factors that also affect the outcomes, resulting in an issue of omitted variables. Moreover, many variables that factor into one's decision to migrate (such as earnings potential) are fundamentally unobserved. To overcome this challenge, similar to Antman (2015), we use household FE to compare changes in the outcomes within households over time as the husband's migration status changes. Put differently, this approach estimates the average change in the outcomes that households experience when the husband migrates, while holding constant all fixed household characteristics that may correlate with both migration and the outcomes. In this section, we describe the approach, outline potential weaknesses to this empirical strategy, and explain how we overcome them.

Our basic regression approach is presented in the equation below: (1) Yit=β0+β1CurrentMigrantit+δi+νt+eit {Y_{it}} = {\beta _0} + {\beta _1}{CurrentMigrant_{it}} + {\delta _i} + {\nu _t} + {e_{it}} where Y_it is the outcome of interest for household i in period t. CurrentMigrant_it is a dummy for the situation that the respondent's spouse in household i is migrating in time t. We consider two broad categories of main outcomes: women's decision-making power in a household; and expenditures.

See Section 2 and Appendix (Section C) for how these variables are defined.

The coefficient of interest in Eq. (1), β₁, captures the average difference in household Y_it when a husband is migrating and when he is not, compared to the difference over the same time period for other households in which a husband's migrant status is not changing – this includes both households that never see migration over any period and those who see it in some period though not the present one. Put another way, β₁ is the average change in Y_it when a husband migrates, compared to households in which the husband is not migrating, above and beyond differences attributable to fixed characteristics.

The advantage of using household FE is that any time-invariant household-level characteristics will be held constant. For example, differences in underlying risk preferences or ability across households do not affect our estimates. In addition, we can control for any time-varying factors that might also be related to the migration decision and the outcome variables. One particularly important time-varying control variable is whether a household has past migration experience. Following Antman (2015), our second estimation includes Experience_it, a dummy equal to one if the respondent's spouse in household i migrated prior to time t. This approach allows us to more precisely estimate the impact of a husband's migration spell by comparing households of similar experience levels. Furthermore, we can include a vector of other time-varying controls, X_it. Incorporating migration experience and other time-varying controls, the estimation becomes (2) Yit=β0+β1CurrentMigrantit+β2Experienceit+Xitγ+δi+νt+eit {Y_{it}} = {\beta _0} + {\beta _1}{CurrentMigrant_{it}} + {\beta _2}{Experience_{it}} + {X_{it}}\gamma + {\delta _i} + {\nu _t} + {e_{it}}

Our vector of time-varying controls, X_it, includes the presence of a mother-in-law or father in-law and the following time-varying economic shocks: whether or not the household experienced a natural disaster, serious illness, death of a household member, falling agricultural prices, decrease in income, loss of employment, a Mercalli scale measure of earthquake intensity at the VDC level, and a dummy corresponding to each control indicating whether it is missing (in which case the control was recoded to zero). We do not include additional controls that may introduce endogeneity through dynamic effects of migration. To ensure that our main results are not driven by this type of endogeneity working through the controls that we do include, we present our main results both with [Eq. (2)] and without controls [Eq. (1)].

For β₁ to represent a causal effect, we must assume that changes in decision-making and expenditures would have looked similar in the absence of migration spells across all households, conditional on household FE, year FE, and observed time-varying characteristics of households. While this is arguably a weak assumption, one concern may be that households experience unobserved shocks that cause the husband to migrate and the outcomes to change. While this is a valid concern, we argue that our results taken together form a narrative that does not seem consistent with such a story.

As discussed before, the comparison group in Eqs. (1) and (2) includes both households for whom we do not observe migration over the observed time periods and those whose husbands do migrate at some point, albeit not during the present period. Even after controlling for past migration experience in Eq. (2), including nonmigrants as part of the comparison is not ideal – their lack of migration could indicate systematic differences from those whom have previously migrated (which we control for), or importantly – will migrate in the future (which we do not control for). For this reason, we also estimate Eq. (2) using a limited sample of households in which the husband is a migrant during at least one of the time periods we observe. This approach, in effect, holds constant the decision to migrate at some point, even in the future. Using the limited-sample approach, we assume that changes in our outcomes would have looked similar across households regardless of when their decision to migrate takes place (again conditional on various controls). In other words, we assume the timing of migration is as good as random.

Finally, if inherent ability or some other unobserved characteristic is causing certain households to trend differently (e.g., high-ability households’ incomes may grow faster), then household-specific time trends, α_i, hold this constant. As noted previously, the survey data were collected for an evaluation of a livestock transfer program. If the program sends people on a different growth path, then household-specific time trends will control for effects of the treatment. To estimate α_i, we include H_it, the interaction of a time period dummy variable and a household dummy variable. This approach partials out of β₁ any changes due to household-specifictrends.

The estimation thus becomes (3) Yit=β0+β1CurrentMigrantit+β2Experienceit+Xitγ+δi+νt+αiHit+eit {Y_{it}} = {\beta _0} + {\beta _1}{CurrentMigrant_{it}} + {\beta _2}{Experience_{it}} + {X_{it}}\gamma + {\delta _i} + {\nu _t} + {\alpha _i}{H_{it}} + {e_{it}}

One concern with Eq. (3) is that the household-specific time trends pick up not only households’ pretrends in the outcomes but also the dynamic effects of migration should they exist. Using a past study on state-level changes in unilateral divorce laws, Wolfers (2006) shows that policy-level time trends can lead to misleading results, such as overestimating the effect of a policy or even estimating it in the wrong direction.

One way to circumvent this issue is to estimate the effect relative only to pretrends. In our case, we cannot apply this solution because our “policy” (a migrant spell) happens for some households at the beginning of our data set, so we lack pretrends for part of our sample. Moreover, we see migrant spells end in our data, which makes it unclear whether or not to treat future periods as pretrend periods.

Our approach considers two broad categories of main outcomes: women's decision-making power in a household; and expenditures. Our hypothesis is that changes in women's decision-making power drive changes in expenditures. To examine this pathway, we need to rule out other mediating factors, particularly income. Income effects are a main reason for migration, so we expect to observe income effects associated with the migration of any family member. The main concern is that any changes in spending may merely reflect the fact that households earn more during migration spells, which has an income effect. Toward this aim, we again estimate Eqs. (1)–(3) using income and other financial outcomes, in order to test for the existence of other such financial factors that might be driving changes in expenditures.

To rule out these mediating factors, we exploit the migration of sons. The logic is that when sons migrate, we expect to observe a similar income effect but no related shift in decision-making power. To the extent that these households are similar to households that send husbands, this approach provides a useful test for whether observed differences in spending patterns are simply reflecting income effects. Implementing the approach is simple; we again estimate Eqs. (1)–(3), this time defining the CurrentMigrant_it dummy equal to one if a son in the household migrates.

There may be multiple sons who migrate or different sons who migrate over different time periods. We simply include a dummy variable for the situation that there is at least one son who is currently migrating in order to most closely mirror the main analysis and avoid making arbitrary specification decisions.

A unique contribution of this paper is to consider the persistence of effects after a husband returns from his migration outpost. To do this, we turn to a first-difference model, which allows us to decompose the change in outcomes when the husband leaves separately from the change when the husband returns: (4) ΔYit=α0+α1Returnedit+α2Leftit+α3Experienceit−1+X→it−1'γ+δi+νt+uit \Delta {Y_{it}} = {\alpha _0} + {\alpha _1} {Returned_{it}} + {\alpha _2}{Left_{it}} + {\alpha _3}{Experience_{it - 1}} + \vec X_{it - 1}^\prime \gamma + {\delta _i} + {\nu _t} + {u_{it}}

This first-difference approach differs from the earlier specification in a few ways. First, the dependent variable, ΔY_it, is now the change in the outcome of interest (financial outcomes, women's decision-making power in a household, or expenditures) for household i between periods t − 1 and t. Similarly, Returned_it and Left_it are dummy variables for whether or not the husband returned from or left for a migrant spell between t − 1 and t. For conceptual clarity, it may be useful to think of these variables as decomposing the change in current migrant status, ΔCurrentMigrant_it. Similar to Eq. (2), this approach also includes household FE (δ_i) to control for any time-invariant differences across households and year FE (ν_t) to control for unobserved shocks that affect everyone in a given year. We also control for the household characteristics in the base period (t − 1) using the same set of time-varying controls (X_it−1) and prior experience with migration (Experience_it−1). These variables hold constant the initial conditions of a household before any change in migration status potentially occurs.

Analogous to how β₁ represents the within-household change

A “within-household change” refers to the change within a household over time. This is the relevant interpretation due to the inclusion of household fixed effects.

In this section, we report our results. To summarize, we show that when husbands migrate, their wives make more decisions and households spend their money differently. Observed changes in expenditures appear to be mediated by changes in decision-making. Upon a migrants’ return, decision-making power is more equitably shared.

4.4

Upon migrants’ return, decision-making power is shared more equitably

In a final empirical contribution, we explore whether the changes in spending habits and decision-making persist after husbands return. Figure 2 reports the coefficients α₁ and α₂ from estimating Eq. (4) for decision-making aggregated across all categories.

Table F1 (Section F of the Appendix) reports the same results and also the results disaggregated by expenditure type in table format.

The wide confidence intervals likely reflect a loss of power in Eq. (4) relative to the FE approach used in Eqs. (1)–(3). Separating the “left” and “return” changes in migration status is statistically demanding, and the sample size is reduced after differencing.

Figure 2

Coefficient plots of α₂ (left panel) and α₁ (right panel) from Eq. (4) are depicted (see Table F1 for coefficients listed in a table); 95% confidence intervals displayed, with robust standard errors clustered on VDC. The different colors/shapes each indicate a separate regression. Blue dots are from regressing husband's decision-making power on Eq. (4) and therefore represent changes in the fraction of decisions over which the husband has the only say. Hollow circles represent the change in the fraction of decisions made jointly. Hollow triangles represent the change in the wife's sole decision-making power. F-tests comparing the absolute value of α₁ and α₂ for the fraction of decisions made by the husband alone, the husband and wife jointly, and the wife alone yields p-values of 0.919, 0.036, and 0.557, respectively.

We find that when a husband leaves, the growth in decisions made by the wife alone is 9 percentage points differentially higher (or more positive) compared to changes over time periods when the husband's migration status does not change. At the same time, overall involvement (joint or alone) by the husband grows differentially lower (more negative or less positive), although these estimates are not statistically significant. However, if and when the husband returns, the wife forfeits all of the sole decision-making power she gained during the migration spell (11% lower growth rate). Yet, instead of being redistributed back to both the husband-only and joint decisions (as we might expect based on the changes when the husband leaves), decisions mostly shift to being made jointly. Specifically, there is a 14 percentage point greater growth in decisions made jointly upon the husband's return. An F-statistic comparing the absolute value of α₁ to α₂ (as reported in Table F1) for joint decisions has a p-value of 0.036, providing evidence that the growth in joint decisions upon the husband's return is larger than the low (or possibly negative) growth rate when he leaves. In other words, while the wife ultimately loses any sole decision-making power she had gained, there is suggestive evidence that, on net, she is left with more shared decision-making power than before her husband's migration spell. This is because the growth in joint decisions is differentially larger than the smaller (or perhaps negative) change in joint decisions estimated between the periods when the migrant leaves (coefficient is −0.040 but is statistically indistinguishable from zero).

An alternative specification where we include an indicator variable for whether the husband had gone away or was at home in both t−1 and t is shown in Table F2 (Section F of the Appendix). This coefficient is not distinguishable from zero in all but one case, indicating that growth rates between periods where the husband is away are indistinguishable from when he is home. However, it is not a priori obvious whether it makes sense to compare the return/leave changes to changes among households in which the husband has gone away or is at home in both periods. With this in mind, we bundle the households that experience no change in the husband's migration status in the omitted group. The interpretation is similar either way.

Even if women ultimately gain shared decision-making power after a husband returns, it may not ultimately affect the decisions the household makes. This could happen if, for instance, husbands have disproportionately more bargaining power and therefore household decisions are still more likely to reflect their preferences. Table 5 reports the results from the regression in Eq. (4) with changes in expenditures as the dependent variable. Each row represents a different regression with a different dependent variable.

Table 5

Dynamic shifts in expenditures: first-difference approach (asinh)

	α1: returned	α2: left	P-value: |α1| = |α2|	N
Δ Temptation goods	1.084*** (0.361)	−0.952** (0.386)	0.819	3,491
Δ Health care	0.328 (0.369)	0.672** (0.329)	0.467	3,368
Δ Ceremonies or celebrations	−0.403 (0.286)	−0.909*** (0.299)	0.163	3,473
Δ Child education	−0.310 (0.236)	−0.450* (0.258)	0.615	3,389
Δ Child clothing	−0.107 (0.213)	−0.119 (0.243)	0.955	3,407
Δ Adult women clothing	−0.027 (0.153)	−0.120 (0.165)	0.693	3,563

Notes: Robust standard errors clustered at the VDC level are in parentheses. Each row represents a separate regression where Returned and Left represent dummies that describe the change in the husband's migrant status between the previous and current periods. The P-value is from an F-test testing for equality between the absolute values of the Returned coefficient and the Left coefficient. Controls for the base period are included. The control set across all regressions includes a separate dummy each for various shocks (the household experienced a natural disaster aside from the 2015 earthquake, serious illness, death of a household member, falling agricultural prices, decrease in income, or loss of employment), a Mercalli scale measure of earthquake intensity at the VDC level, a dummy indicating whether the husband has migration experience, a dummy indicating a resident mother-in-law, a dummy indicating a resident father-in-law, and a separate dummy each indicating when the control is missing. Zeros are imputed for missing values. All amounts are in rupees and transformed by the inverse hyperbolic sine.

*

p ≤ 0.10,

**

p ≤ 0.05,

***

p ≤ 0.01.

Among the categories for which we observed a change in the main result – temptation goods, children's education, and children's clothing – there are no asymmetric changes. In fact, spending on temptation goods appears to rebound by almost the exact same amount. This underscores that even if migration may change who participates in decision-making, the relative bargaining position of the decision-makers may play a large role in what decisions are made. Nevertheless, any lasting changes in decision-making power are an important first step toward more-equitable decision-making.

For children's education, evidence from the fixed effects estimate (presented in Table 4) showed that households spend more on children's education during a husband's migration spell. Table 5 now suggests a reduction in the change in spending when a migrant leaves relative to when his status does not change. Specifically, the estimate suggests that growth in child education expenditures is differentially lower between periods in which the husband leaves versus between periods in which his migration status does not change (significant at the 10% level). As discussed in the empirical strategy section, unlike the estimates presented in Table 4, the first-difference approach estimates the within-household change in growth in the outcome, not the within-household change in levels of the outcome. In other words, it is comparing growth rates, which further requires that we compare subsequent periods to one another rather than treating all periods in which the husband is home the same. Therefore, the possibility that growth is differentially lower between periods when the husband leaves does not preclude the possibility that while the husband is away (compared to all periods when he is home), spending on education is differentially larger than in other periods (as shown in Table 4). Even if this is possible, it is still an odd result. We emphasize that the main purpose of the first-difference specification is to test for asymmetries between the return and leave effects.

We have shown that when husbands migrate, decision-making power tends to shift toward the wife, and there is suggestive evidence that she retains some decision-making power after he returns. A natural next question is: what is driving this pattern?

When a household member migrates for work, the composition of the resident household members also changes, which could affect the household economic decision-making power through several channels. First, shifting control over household decisions to his wife could be purely practical, as the migrant is less able to manage the household from a distance. Second, a husband's migration could increase his wife's bargaining power: migration affects labor allocation not only by the migrant, but also by those at home (Abdulloev et al., 2014; Lokshin and Glinskaya, 2009), thus affecting each individual's relative contribution to household income.

In addition, geographic separation could introduce asymmetric information about household decisions, therefore allowing the wife to make de facto independent decisions about the household budget. Ashraf et al, (2015) present results from a field experiment in El Salvador, which show that migrants increase savings when they could easily monitor and control the savings account back home. Ambler (2015) uses another experiment in El Salvador to show that recipients send home more remittances when their income is revealed to their spouse. A third study by Chen (2013) in China shows that the decisions that change during migration spells were those not easily monitored from afar or after the husband's return.

If husbands and wives have heterogeneous preferences, any of these three channels would ultimately lead to different household decisions. We are unable to distinguish between these three channels in order to identify what exactly drives contemporaneous changes in decision-making. However, none of these explanations on their own are likely to explain the possibility of persistence over time. One potential explanation consistent with persistent effects is that the male migrant is “bringing home” norms and customs from their destinations. This seems unlikely since Nepali migrants mostly migrate to the Middle East, Far East, or to other cities within Nepal. These destinations are not known for their liberal gender norms, and if anything, we might expect this mechanism to result in fewer decisions being made by women upon the migrant's return.

In order to explain persistence, we suggest that a different mechanism must be at work: habit formation. According to Wood and Neal (2007), “Habits are learned dispositions to repeat past responses. They are triggered by features of the context that have covaried frequently with past performance, including performance locations, preceding actions in a sequence, and particular people.” In economics, the original model of habit formation by Becker and Murphy (1998) assumes that past consumption affects current utility (what Becker and Murphy describe as “rational addiction”). This assumption describes the inherent “stickiness” of habits. Once formed, habits are difficult to override.

Current habits may be triggered to form (or dissolve) when some sort of shock causes individuals to reevaluate their behavioral choices (Verplanken et al., 2008; Wood and Neal, 2009). For example, a move to a new city (Verplanken et al., 2008) or transferring to a new university (Wood et al., 2005) provides a sufficiently different context that encourages individuals to form new habits. Likewise, the migration of a prominent household decision-maker represents a similarly discontinuous change in context that disrupts the household's usual decision-making process.

Of course, new habits are not guaranteed to stick. When a male migrant returns, what is stopping the household from resorting to their previous decision-making habits? Becker and Murphy's original habit-formation model offers an explanation: current preferences evolve over time through a process of “learning by doing.”

The original “learning by doing” economics model, which builds on an even longer literature from education and psychology, is often attributed to Arrow (1962).

This learning process takes time. Therefore, we expect the habit to form more strongly in households with longer migration spells. We can test for this by estimating Eqs. (1)–(3) again, this time including an interaction for the situation that the trip was declared to last for >6 months. Table 6 reports the results of this approach using decisions aggregated across all categories as the outcome.

These data come from the same survey question used to identify migrants, which asks about resident status. Enumerators were asked to specify whether the residency (if not at home) was greater than or less than 6 months. We cannot rule out the possibility that enumerators answered based on expected time away rather than time away up until the date of the survey. But, even if this is the case, this simply introduces measurement error – households declaring longer trips are more likely to be observed at a point in time when the husband has been away for a longer time than in other households.

During a husband's migration spell, his wife may exert more control over household resources and therefore change how the household spends income. Given the prevalence of seasonal migration in developing countries, even these temporary changes could have consequences for economic development. The extent to which these changes persist after migration spells will magnify these consequences.

Using a panel data of rural Nepali households, we present evidence that a husband's migration leads to contemporaneous shifts in intrahousehold decision-making: a husband's absence increases the expenditure decisions over which the wife has full control by 10 percentage points. These households also shift away from expenditures on alcohol and tobacco in favor of children's education and clothing. We exploit the migration of sons – which increases household income but does not change decision-making – to show that this does not seem to be driven by an income effect.

To address the challenge of self-selection into migration and omitted variables that could threaten our interpretation of these findings, we rely on household FE, conditioning on eventual selection into migration, and including household-specific time trends. These approaches address the concern that time-invariant and time-variant unobservables among households, such as household risk aversion or differential income trajectories across households, drive our results. Our results provide a narrative largely consistent across these empirical strategies. We argue that this approach, in addition with the narrative that the results form as a whole, provides a compelling case for our interpretation.

One remaining challenge relates to the location of the consumer of the good. For each kind of expenditure, we do not directly observe the consumer. One might hypothesize that temptation goods are consumed more often by men than women. If this is the case, then when men migrate, expenditures on temptation goods will decline simply because the consumer is no longer present. This presents a challenge to the interpretation of our main results. Disentangling this channel presents an important area of future research. However, we believe that our results tell another story. First, shifts in decision-making power are strongest for temptation goods, but there is evidence for changes in other categories as well. These other categories include expenditures that are definitely not consumed by the male migrant (children's education and clothing; and adult women clothing). Our analysis of persistence is also relevant; women retain some decision-making power when male migrants return, which seems to reflect changes in decision-making that go beyond the absence of the primary consumer. This is true even in the temptation goods category.

This is evident in Table F1 of the Appendix.

One important contribution of our paper is to show that the economic decisions are more likely to be made jointly even after a male migrant returns. This could be driven by either the importation of norms from abroad or a habit-formation model. We find that migration spells that last longer have larger changes in decision-making, which we argue is relatively more consistent with a habit-formation model. This suggests that even temporary migration spells may help achieve gender equality in economic decision-making, with downstream effects on how households spend their income. When women play a role in decisions, they are more likely to invest in the next generation, increasing the likelihood that a pernicious cycle of intergenerational poverty can be broken.

The evidence presented in this paper suggests that a new habit of more-equitable household decision-making can be formed over a period of temporary migration. If the new habit “sticks” and women continue to take part in economic decisions, then it seems plausible that temporary male migration could provide a mechanism for lasting cultural change. Our analysis can only speak to short-term effects, but the results are promising. Women's labor force participation in Nepal, a key indicator of gender equality, offers suggestive evidence supporting this narrative. When compared to its South Asian neighbors, Nepali female labor force participation rate (LFPR) stands out: while nearly 80% of women participate in the labor force in Nepal, only 26% do in India.

Afghanistan has the next highest rate of female LFPR at 44%. Estimates are from the most recent year for which data are available for Nepal, i.e., 2010.