Abstract
The economic impact of remittances on migrant-sending countries has been a subject of debate in the scholarly literature on migration. We consider the topic using a household-level approach. We use a new survey, “Georgia on the Move,” to examine migrant-level, household-level, and contextual variables associated with the probability that a household in the Republic of Georgia receives remittances. We then apply propensity score matching to estimate how remittances affect particular types of household expenditures, savings, labor supply, health, and other measures of well-being. Separate analysis of the subsample of households with a migrant currently abroad distinguishes the effects of remittances from the effects of migration as such. In Georgia, remittances improve household economic well-being without, for the most part, producing the negative consequences often suggested in the literature. We find evidence for an important aspect that has not been widely discussed in prior studies: remittances foster the formation of social capital by increasing the amount of money that households give as gifts to other households.
Introduction
The economic impact of migrant remittances on the communities that receive them has sparked considerable interest and debate among scholars of international migration. This literature, however, has not examined the economies and societies of the former Soviet Union, despite the fact that international migration and remittances have come to play vital roles in these societies after the Soviet Union collapsed in 1991 (Korobkov 2007). Moreover, the debate on remittances has largely overlooked their potentially important contribution to social capital formation in migrant-sending communities.
We examine the economic and social effect of remittances on households in the Republic of Georgia, including their role in fostering and strengthening social ties between and within households. Like other former Soviet republics in the South Caucasus and Central Asia, Georgia has experienced high levels of emigration since the Soviet collapse, but the economic and social consequences have remained unclear in the absence of empirical data. We analyze the fall 2008 Georgia on the Move (GOTM) survey. We first estimate probit models linking migrant-level, household-level, and contextual variables with the probability of remittance receipt among Georgian households. We then match “remittance” to “nonremittance” households with similar estimated propensities for remittance receipt to see how remittances affect particular types of household expenditures, savings, labor supply, health, and other measures. Propensity score matching accounts for differences between remittance and nonremittance households on observable variables that may confound estimates of the effects of remittances. Separate analyses for the subsample of households with a migrant currently abroad distinguish the effects of remittances from the effects of migration as such.
In Georgia, remittances improve household economic well-being without generating the negative consequences often suggested in the literature. They also produce and reinforce social capital by increasing the transfers that Georgia-based households make to other households, and in other ways. Thus, in contrast to the emphasis in other studies on how migrant social networks promote migration streams (Garip 2008; Massey et al. 2005), our findings suggest migrant remittances can establish and sustain reciprocal ties among households in the origin country.
Background: Migration in Post-Soviet Georgia
Migration has played a formative role in the economic and social life of Georgia since it gained independence with the Soviet Union’s demise in 1991.1 Georgia’s residential registration system broke down following the Soviet collapse, and vital records are suspect. Observers agree, though, that Georgia’s net out-migration rates are among the highest globally. The United Nations (2009:183) put net emigration from Georgia from 1995 to 2005 at 598,000. According to Badurashvili (2004), Georgia lost approximately 1 million of its nearly 5.5 million 1989 residents to emigration by 2002. Mansoor and Quillin (2007:33) cited data indicating that from 1990 to 2003, about 20 % of Georgia’s 1989 population migrated abroad. Among 25 ex-Communist nations covered in their report, only Albania and Armenia lost higher percentages to net out-migration from 2000 to 2003 (Mansoor and Quillin 2007:31).
In the early years of Georgia’s independence, political and military conflicts associated with the Soviet collapse produced large outflows of refugees and non-ethnic Georgians, primarily to Russia. Secessionist conflicts in several Georgian provinces created a sizable population of internally displaced persons. Since the mid-1990s, as Georgia experienced protracted economic crisis, economic motives were paramount for Georgian citizens going abroad. Outflows remained substantial and steady, if somewhat smaller in volume than in the initial phase. Western Europe (particularly Greece and Turkey) and North America became more common destinations. The 2003 Rose Revolution ushered in the Saakashvili administration, which initiated market reforms, improved the investment climate, and reoriented the country politically toward the West. These developments, as well as hostile Russian actions against Georgia starting in 2004 and culminating in the August 2008 invasion, probably furthered the trend toward European and North American destinations among Georgian labor migrants.
Bank transfer data from the National Bank of Georgia (NBG) show substantial growth in the volume of remittances received by Georgians during the Saakashvili presidency.2 Remittances through official channels totaled 63 million US dollars (2.2 % of gross domestic product, GDP) in 2000, growing to $197 million (4.9 %) in 2003. By 2008, they surged to $1.002 billion, outstripping Georgia’s robust GDP growth and accounting for 7.8 % of Georgia’s national product. These figures represent only official transfers, but data we report in this article suggest that they capture the lion’s share of remittances.3 Although migrant remittances play a significant role in Georgia’s economy, their precise effect depends on how they are used and whether their negative consequences outweigh their benefits.
The Economic Impact of Remittances
Studies of remittances contain both pessimistic and optimistic perspectives.4 According to the pessimistic view, remittances perpetuate dependency of migrant-sending communities (Lipton 1980; Reichert 1981; Rubenstein 1992). They are spent mainly on consumption, not invested in productive activity. They fuel inflation, a taste for imported goods, and a standard of living that requires more remittances and thus encourages additional migration, ultimately depriving communities of their most capable and productive workers. Remittances foster inequality because households that do not receive remittances cannot keep up with those who do (Adams 1989; Garip 2012). The associated envy can destroy the traditional social fabric and ties that keep poor communities together in the face of material hardships. Remittances can produce moral hazards, lowering incentives for members of households that receive them to work in the local labor market or in domestic businesses (Amuedo-Dorantes and Pozo 2006). Because remittances are an unreliable source of revenue in the long term, families who depend on them become vulnerable. Overall, the pessimistic outlook on remittances sees them as perpetuating a cycle of dependency and thwarting the positive development of communities that receive them in large volumes.
The optimistic perspective is often associated with the “new economics of labor migration” (NELM), which departed from the neo-classical perspective (Todaro 1969) by treating the household rather than the individual as the appropriate unit for analyzing labor-migration behavior (Lucas and Stark 1985; Stark 1991; Stark and Bloom 1985; Taylor 1999). According to NELM, households send members to work abroad in order to diversify income sources in the face of uncertain local labor market and agricultural conditions. Where credit and formal insurance are lacking, remittances provide income insurance or “smoothing” (Amuedo-Dorantes and Pozo 2011b). They alleviate poverty and improve living standards simply by providing income to families toward the bottom of the income distribution (Koc and Onan 2004). They can offset losses of labor and human capital resulting from migration. Moreover, some remittances are spent on productive activities, and remittance-driven consumption can have multiplier effects by increasing demand for goods and services (Massey and Parrado 1994): if these are locally produced, remittances benefit a circle of producers and service providers much wider than just the households that receive them (Adelman and Taylor 1992). Remittance-driven consumption geared toward imports may spur entrepreneurial locals to initiate new production to meet incipient demand. Remittances can also be invested in human capital that can have long-term developmental benefits: they can be used to improve education (Arguillas and Williams 2010; Borraz 2005; Lu and Treiman 2011; Zachariah et al. 2001) or health care for household members (Amuedo-Dorantes and Pozo 2011a; Kanaiaupuni and Donato 1999). As for inequality, the impact of remittances depends on their magnitude relative to other sources of income and on where families receiving them are in the income distribution (Acosta et al. 2008; Stark et al. 1986).
Empirical studies assessing the impact of remittances in specific national contexts arrive at mixed results (Garip 2012). As several recent reviews of the literature on remittances have indicated (Cohen 2005; de Haas 2010), the reasonable conclusion is that remittances play both positive and negative roles in the societies that receive them. Their net impact cannot be deduced theoretically but depends on who migrates, how the remittances are used, and whether remittances spur productive responses at the local level. These variables, in turn, depend on a host of local conditions, such as the development of local infrastructure and capital markets, the size and composition of the local workforce, the availability of investment opportunities, and local entrepreneurial culture. The empirical literature on Mexico is especially well developed (e.g., Durand et al. 1996a, b; Garip 2012; Massey and Parrado 1994), although Asian and African countries have also been examined. Despite the surge of labor migration in former Soviet countries, particularly in the Caucasus and Central Asia (Korobkov 2007; Mansoor and Quillin 2007), there are no systematic empirical analyses of the role of remittances in the region.
Remittances as Source of Social Capital Formation
We propose that remittances can strengthen social capital in the receiving communities in a way not previously considered in the literature: by fostering interhousehold transfers. We adopt the concept of social capital proposed by Portes and Sensenbrenner (1993:1323): “expectations for action within a collectivity that affect the economic goals and goal-seeking behavior of its members, even if these expectations are not oriented toward the economic sphere.” Interhousehold transfers may reflect and reinforce broader social norms emphasizing community or (extended) family solidarity, a form of social capital Portes and Sensenbrenner call “value introjection.” Interhousehold transfers also may be a type of “reciprocity transaction,” which creates mutual obligations between individuals or households or a basis for “enforceable trust,” providing expectations of future rewards accruing to the donors as a result of their position within the community. Although the latter two aspects of social capital reflect instrumental considerations, they are fundamentally social because social goods, such as prestige, respect, and authority, are typical currencies of exchange and trust in such transactions. By transferring funds to other households, Georgians cultivate and sustain broad norms of solidarity that prevail to varying degrees within different collectivities, and they also secure concrete reciprocal obligations (from the recipients) and expectations of future rewards (from the larger collectivity). In turn, the social ties and solidarities that are fostered by such exchanges may serve as resources for collective economic, political, and social actions.
Social capital formation may be a motive for sending remittances in the first place, one that (as we discuss later) overlaps somewhat but remains conceptually distinct from the altruism, insurance, and other motives discussed in the economics literature on remittances (e.g., Rapoport and Docquier 2005). We hypothesize, in addition, that remittances also build social capital by helping households that wish to invest in it overcome liquidity constraints. Other studies have discussed how remittances have provided “building blocks for community” (Cohen 2005:97) via investments in community organizations and collective goods projects and how migration networks facilitate the migration process (Garip 2008; Massey et al. 2005). The potential social capital formation role that remittances can play by increasing levels of interhousehold transfers has received scant attention, though. Some scholars have mentioned that remittances can be used for “maintenance of social networks” (Seddon 2004:415) or give origin households “flexibility in social roles and use of community- and kin-based networks” (Cohen 2005:96). However, they have not examined these aspects of remittances in depth, nor have they provided quantitative evidence to support these claims. Taylor and Dyer (2009) observed that remittances benefit migrant-sending communities indirectly as a result of the economic interactions between households that receive remittances and those that do not, but their analysis focused on market-based interactions, not on direct transfers of funds between households. In contrast, Cohen (2005:95) cited research on Mexico suggesting that remittances produce a “decline in kin-based support networks” by encouraging private celebrations of life cycle rituals.
Our data permit empirical tests for a positive effect of remittances on interhousehold transfers, which would suggest that our theoretical argument is plausible and merits further empirical exploration in future studies. If remittances encourage interhousehold transfers, they enhance well-being beyond the households that directly receive them, and they establish and reinforce social ties, connections, and obligations that may have long-term economic and social benefits.
Remittances in Georgia: Research Questions and Approach
We investigate two empirical questions: (1) What variables are associated with the probability that a Georgian household receives remittances? (2) What effects do remittances have on consumption expenditures and investments in productive, human, and social capital in Georgia? First, we analyze all the households in the sample, including those without members living abroad, because some receive remittances from migrants outside the household. The correlates and impact of remittances among all households provide a comprehensive picture, but households with absent migrants are far more likely to receive remittances. Unobserved household characteristics associated with having an absent migrant, such as prior household wealth or migration-related debts, may jointly affect remittance receipt and expenditures or other remittance-related outcomes, thus potentially confounding estimates of the correlates and impact of remittances in the analysis of all households. Therefore, we conduct parallel analyses, using the subsample of absent-migrant households, which incorporate migrant characteristics in our model specifications and disentangle the economic impact of remittances as such from the effect of migration.
To assess their effect, we do not examine how remittances are spent. Instead, we measure the average effect of remittance receipt on household economic outcomes and behavior. It is irrelevant how the specific funds that come in the form of remittances are spent because money is fungible (Taylor 1999). If, for example, remittance money is used to purchase food, then the funds that in the absence of remittances would have been spent on food might be saved or spent on travel instead. The issue instead is how remittances affect overall household expenditures, investment, savings, and other related outcomes.
Selection into migration is nonrandom, and among households with absent migrants, the probability of receiving remittances is systematically related to other variables that might affect economic and social outcomes within households. We control for some variables that jointly affect the probability of having an absent migrant, the probability that he or she sends remittances, and the outcomes of interest. However, unobserved variables may still have joint effects and therefore bias our estimates of the effects of remittances. We mitigate the problem by replicating our analyses among absent-migrant households only and by matching remittance and nonremittance households on their estimated propensity to receive remittances. By eliminating one source of confounding unobserved variables—those that affect the probability of having an absent migrant and the outcome in question—we obtain a more precise measure of the specific impact of remittances. Other unobservable variables may still jointly affect remittance behavior of migrants and the expenditures of Georgian households and thus bias our estimates of the effect of remittances. Our data lack instruments that might account for such unobserved factors econometrically. Nonetheless, our data’s rich set of relevant variables and the performance of our models for remittance receipt lend confidence that our matching-based estimators accurately capture the effects of remittances.
Data: Georgia on the Move (GOTM) Survey
The GOTM survey was part of a six-country study of the relationship between migration and development funded by the Global Development Network (GDN). Parallel surveys were conducted in Colombia, Fiji, Ghana, Macedonia, and Vietnam. General findings from the comparative study are reported in Chappell et al. (2010). The survey was designed and implemented (using face-to-face interviews) by the Caucasus Research Resource Centers (CRRC) and International School of Economics at Tbilisi State University (ISET), with the help of external advisors and the GDN’s Project Management Team. Target sample volume was allocated equally across three strata: absent-migrant households (at least one member currently living abroad), return-migrant households (at least one member who previously lived abroad for at least three months), and nonmigrant households (with neither current nor return migrants).
Primary sampling units (PSUs) were voter precincts randomly sampled within rural villages, cities, and Tbilisi, with the number of PSUs in each proportionate to population size.5 The researchers conducted block enumerations of households by migration status within each selected PSU. Households with both absent and return migrants were randomly assigned to either stratum. The enumeration permitted a random sampling of households within each migration-status stratum. Because of some errors in the enumeration and variation in response rates (overall, 70 %) by strata, the final sample of 1,482 households included 464 absent-migrant households (31.3 %), 345 return-migrant households (23.8 %), and 673 nonmigrant households (45.4 %). The interviews were conducted in November–December 2008, after the August invasion by Russian troops.
Two quality-control measures were applied to check the validity of the GOTM data (Tchaidze and Torosyan 2009). First, sample distributions by age, gender, household size, marital status, and education were compared with two other large Georgian surveys: an annual survey of Georgian households conducted by the National Statistical Office of Georgia (www.geostat.ge), and the Caucasus Barometer survey of the Caucasus Research Resource Center (http://www.crrccenters.org/). Some statistically significant discrepancies emerged, but they were small in magnitude and could be explained by the different sampling frames and purposes of the studies. Second, the response deviation score technique (Murphy et al. 2004) was applied to four questions to assess all 68 interviewers. None exceeded the standard threshold (50 % deviations) indicating misconduct.
Methods, Variables, and Hypotheses
Modeling Remittance Receipt by Households
We estimate probit regressions for the probability that a household receives migrant remittances. The survey asked all households whether they received any money or goods from migrants abroad who are not household members, and asked absent-migrant households whether they have received money or goods from their absent migrant(s).6 Both questions explicitly refer to the previous 12 months. Although the survey asked about the amount of remittances received, how they were transferred, and how they were spent, we use a simple dummy variable as our sole measure of remittance receipt. There are many missing observations regarding the amount received: 33 % of the absent-migrant households receiving remittances refused to answer, and 23 % said they do not know.7 The data are even more sparse in regard to remittances from nonhousehold members. Thus, the data are unsuitable for modeling the amount of remittances received or using their volume to assess their impact, as others have done (Semyonov and Gorodzeisky 2008). The manner of transferring remittances is not our focus, but 87 % of households receiving remittances from members and 90 % receiving them from nonmembers received transfers via banks or licensed transfer agencies, implying that official bank transfer data capture most transfers.
To model the probability of receiving remittances, we use the survey’s information on the sex, age, main activity, education, self-reported religiosity, and prior migration experience of each household member currently living in Georgia, variables similar to those used in analyses of remittance behavior in other contexts (e.g., Semyonov and Gorodzeisky 2005). See Table S2 in Online Resource 1 for full variable definitions. Our unit of analysis is the household, so we constructed household-level measures for these variables using the individual-level data. We also include contextual variables: the type of locality (Tbilisi vs. other urban vs. rural area) and the percentages of absent-migrant and return-migrant households in the PSU, derived from the sampling enumeration. Our models for absent-migrant households add variables characterizing the migrant groups (members of the same Georgian household living together in the same foreign destination or, in some cases, a single individual): sex, age, education at departure, and employment status of the migrant group head; the group’s size and age composition; main reason for migrating; whether a job abroad had been arranged prior to departure; frequency of contact with Georgian household; duration abroad; and country of destination.
Our main purpose for modeling who receives remittances is to obtain propensities in order to measure how remittances affect household spending and other outcomes. We base our specifications on common explanations of remittance behavior and the variables in our data. Rapoport and Docquier (2005) described typical motives for remitting: altruism, repayment of loans incurred for migration or education, insurance against income volatility and other uncertainties, investments in productive or human capital to enhance lifetime earnings, inheritance concerns (migrants use remittances to ensure that they are not disowned and that they have the opportunity to return to the household if they wish), a strategic incentive to deter the migration of others in the household (which will tend to drive down migrants’ starting wages), and payment for services (e.g., care for assets, children, and elderly relatives). Taylor (1999:75) grouped these motives under three headings: pure altruism, pure self-interest, and NELM motives (“migrants and their households of origin are bound together by mutually beneficial, informal contracts, including an agreement to provide insurance to one another”).
The NELM category partly overlaps with the notion of social capital investment that we propose: by remitting, the migrant maintains and strengthens his/her relationship to the household back home. Failure to remit jeopardizes the ties that bind migrants to their households, which can have adverse economic and non-economic consequences for the migrant in the short and long term. Social capital formation encompasses altruistic motives but can also be instrumentally motivated. It is broader than “NELM motives” because it includes psychic, normative, and status benefits to the remitter beyond the economic gains implied by “informal contracts,” and it proposes theoretical motives for sending remittances to households other than one’s own and for interhousehold transfers back in Georgia. Thus, it extends the logic NELM applies to households to interhousehold networks.
As detailed in Table 1, different theoretical considerations point to six underlying variables positively associated with the probability of remitting. Most of these relationships are implied by multiple theoretical mechanisms. We lack direct measures for them, so we rely on proxy measures, as follows.
Predictors of remittance provision, measures, and mechanisms
| Underlying Variable . | Measures . | Possible Mechanisms . |
|---|---|---|
| Migrant’s Disposable Income | Number of adults in the migrant group, economically motivated migration, education and age of migrant group head, full or almost full-time employment of migrant group head, prearranged job for migrant group head, destination in Europe/North America, duration abroad | Altruism and all other mechanisms |
| Costs of Services Provided by the Recipient Household | Number of retirees (+ or −), number of children, number of young adults (+ or −) in the Georgia-based household, residence in Tbilisi (where opportunity costs for household services are higher), number of working-age adults (−) | Exchange for services, insurance |
| Income Volatility in Home Location | Rural residence of Georgia-based household, Tbilisi residence (−) | Insurance |
| Georgia Household’s Bargaining Power | Average education level of adults in the Georgia household, number of (especially male) adults, proportion of absent and return migrants in the PSU (better information about migration incomes on the side of the household, enhanced normative expectation of remitting) | Inheritance protection, deterrence of additional migration, insurance, repayment of loans, social capital formation |
| Cost of Migration and Education | Education of the head of migrant group, destination in North America | Repayment of loans |
| Strength of Social Ties Between Migrants and Georgia-Based Households | Religiosity, Georgia residence in rural village or town other than Tbilisi, frequency of contact with the Georgia household, prevalence of migrant households in the community, duration abroad (−) | Altruism, social capital formation, insurance |
| Underlying Variable . | Measures . | Possible Mechanisms . |
|---|---|---|
| Migrant’s Disposable Income | Number of adults in the migrant group, economically motivated migration, education and age of migrant group head, full or almost full-time employment of migrant group head, prearranged job for migrant group head, destination in Europe/North America, duration abroad | Altruism and all other mechanisms |
| Costs of Services Provided by the Recipient Household | Number of retirees (+ or −), number of children, number of young adults (+ or −) in the Georgia-based household, residence in Tbilisi (where opportunity costs for household services are higher), number of working-age adults (−) | Exchange for services, insurance |
| Income Volatility in Home Location | Rural residence of Georgia-based household, Tbilisi residence (−) | Insurance |
| Georgia Household’s Bargaining Power | Average education level of adults in the Georgia household, number of (especially male) adults, proportion of absent and return migrants in the PSU (better information about migration incomes on the side of the household, enhanced normative expectation of remitting) | Inheritance protection, deterrence of additional migration, insurance, repayment of loans, social capital formation |
| Cost of Migration and Education | Education of the head of migrant group, destination in North America | Repayment of loans |
| Strength of Social Ties Between Migrants and Georgia-Based Households | Religiosity, Georgia residence in rural village or town other than Tbilisi, frequency of contact with the Georgia household, prevalence of migrant households in the community, duration abroad (−) | Altruism, social capital formation, insurance |
More migrant disposable income makes more funds available to transfer, whatever the motives. Migrant group’s earnings should be higher for older, male, better-educated, and currently employed (full-time or nearly full-time) heads of the migrant group, a prearranged job, location in West Europe or North America (where earnings are higher than in former Soviet republics and other destinations), economic motives for migration, and longer time living abroad (reflecting earnings growth over time or positive selection due to fewer failing migrants at higher durations). The larger the migrant group and the more children it has, the higher its costs and the lower its disposable income.
If remittances pay for services to the migrant provided by household members at home, then factors raising the costs of those services should increase the probability of remitting. The more dependents (children, young adults in education, and retirees) in the Georgia household, the greater the associated costs. In contrast, working-age adults and employed young adults provide alternative sources to cover these costs. Retirees may care for young children at discounted rates, so their effect (like that of young adults, depending on whether they work) can be positive or negative. The opportunity costs of caring for household members should be greater in Tbilisi, which offers more employment opportunities than villages and other cities.
Potential and actual income volatility, emphasized by the insurance explanation of remittance behavior, is typically greater in rural areas. It should also be lower in Tbilisi, where the country’s resources and opportunities are concentrated, than in other cities.
Remittances are more likely to the extent that the Georgia-based household has economic and/or normative power to motivate reluctant or self-interested migrants. For example, the Georgia household can threaten to cut off the migrant from a future inheritance or a potential return to the household (de la Brière et al. 2002) or to send more migrants abroad, which tends to diminish migrant wages (Stark 1995). Measures associated with greater resources, more potential migrants, and more information about migrant earnings in the Georgia-based household proxy its bargaining power. All mechanisms except for altruism imply that these measures correlate positively with the probability of remitting.
If the repayment of loans to cover the costs of migration or pay for migrants’ prior education drives remittances, remittance probabilities should be higher for highly educated migrant group heads and migrants in North American destinations (which cost more to reach).
The altruism, insurance, and social capital mechanisms all imply that the stronger the social and/or economic connection between a migrant and a Georgian household, the greater the probability of remitting. We propose that more religious households should have stronger ties to migrants because of the traditional values associated with religion and because shared religious practice often reinforces community-based identities. Both within-household and also broader community ties are weaker in Tbilisi than in other cities and, especially, rural villages, given the more “modern” lifestyles associated with the capital. Households reporting more frequent interactions with their migrant members abroad presumably have, on average, closer ties with them.8 The longer a migrant is abroad, the weaker ties may become. Finally, larger percentages of absent and return migrants in the PSU reflect a more entrenched process of migration in the area that may increase social pressure on migrants to remit.
We cannot adjudicate between the different theoretical explanations for remitting behavior because multiple theories point to similar patterns of effects. For some variables, though, expectations differ regarding the signs of effects implied by different mechanisms (length of time abroad, residence in Tbilisi, and imputed income of the Georgia-based household).
Measuring the Impact of Remittances
To assess the economic impact of remittances, we first apply t tests to the differences in unconditional means on our outcome measures for nonremittance and remittance households. This provides an initial sense of the magnitude of the raw group differences between the two types of households. However, unconditional comparisons do not indicate whether remittances have a causal impact because observed differences in the group means may reflect group differences in variables that jointly influence the receipt of remittances and the outcomes in question. This is especially true for our analyses of all households: having an absent migrant is strongly associated with receiving remittances and is likely associated with expenditures, business activity, labor supply, and so on. Remittance selectivity could also affect our analyses for absent-migrant households: those that receive remittances may be wealthier, have more human capital, or have more working-age members than those that do not; and these variables, in turn, probably affect expenditures and the other outcomes of interest. Therefore, we use propensity score matching to account for nonrandomness in remittance receipt.
We first estimate propensity scores from the results of our probit models for remittance receipt. We then use these estimated propensities to calculate counterfactual expected outcomes for each household that received remittances under the hypothetical condition that they had not received remittances. In effect, we match each of these “treated” households to households that did not receive remittances but have the same propensity to receive them. Under the assumption that residual factors affecting treatment assignment net of treatment propensity are ignorable, we can interpret the average differences across treated observations in their observed outcomes and their counterfactual outcomes as average treatment effects for the treated (ATTs).9 ATTs based on propensity score matching offer intuitive measures of the effects of a dichotomous variable that require fewer assumptions than standard parametric techniques. We use kernel matching, which computes the ATT for each treated observation as a weighted average of its differences in the outcome from all the untreated observations in the common support region, each weighted by an inverse function of the difference in propensities. Compared with other matching procedures, kernel matching makes maximum use of the information in the data. Analytical standard errors cannot be computed with kernel matching, so we calculate bootstrapped standard errors.10
The outcomes we analyze include measures of expenditures in the prior 12 months on consumption and investments in human and social capital, labor supply, school enrollment by 17- to 25-year-olds, household earnings excluding remittances, business ownership and landownership, number of rooms in dwelling, and household Internet access. All expenditure and income data are denominated in Georgian lari, equivalent to about $0.60 (USD) at the current exchange rate. To reduce the number of expenditure analyses, we combine 14 of the GOTM’s original 22 expenditure items into four broad categories: school expenses (school fees, school supplies, and “other school expenses”); leisure (holiday-related expenses, leisure items, and leisure activities); housing needs (water supply, cooking fuel, heating fuel, and electricity); and household goods (clothes, kitchen appliances, electrical appliances, and furniture). The eight other specific categories remain distinct: religious activities, personal services, medical care, rent, motor vehicle, savings, debt payments, and gifts to others. We also sum expenditures across all categories (“total household budget”) and across all items except savings, gifts, and debt payments (“total expenditures”).11
We hypothesize that migrant remittances have either a positive effect or no effect on these forms of household expenditure. Increased consumption may have salutary multiplier effects if the items consumed are domestically produced or the increased demand spurs domestic producers to provide the items or services. Investments in productive capital (business activities), human capital (education and health), and savings can have long-term positive effects on development. A positive effect of remittances on the volume of transfers to other households implies the social capital formation benefits that we proposed earlier.
We use dummy variables indicating at least one unemployed household member, at least one member in poor or very poor health, and at least one member aged 17–25 studying full-time (among households with members in that age range).12 The skeptical view implies that the probability of having an unemployed household member will be higher in households that receive remittances because they provide a disincentive to work locally. The optimistic perspective anticipates that remittances are invested in human capital and thus are associated with lower probability that a household member has poor health and a higher probability that young adults are enrolled in education.
We also look for a negative effect of remittances on the adjusted logged earnings (excluding remittances) of household members left behind, another prediction of the skeptical view. If remittances act as a disincentive to work, then other household earnings would be lower, on average, in remittance-receiving Georgian households.13 We omit household earnings from our models predicting receipt of remittances because earnings at the time of the survey are more likely endogenous to receipt of remittances during the preceding year than vice versa. In any case, we ascertained that income does not significantly affect remittance receipt.
If they increase the probability that a member of the Georgia-based household operated a business in the previous year, remittances foster productive investment. If they help households acquire land, the developmental impact depends on how the land is used. Improved housing quality might represent conspicuous consumption, but such consumption may benefit the local construction industry. Internet access might provide entertainment and leisure but can also be used to solidify social ties, obtain information about economic opportunities, and advance business purposes.
Descriptive Statistics
Table 2 displays the descriptive statistics on remittance status and the covariates in our remittance models for all households and by household migration status.14 Overall, 28.3 % of GOTM households and 71.9 % of absent-migrant households received remittances in the prior year. Having an absent migrant abroad strongly predicts remittance receipt. Still, 8.5 % of households without an absent migrant received remittances. In 60 % of such cases, remittances were sent by extended family members; in the rest, by “others.” These extrahousehold remittances are difficult to account for when using NELM reasoning because NELM focuses on their role in the household economy. They are one piece of evidence that remittances can contribute to social capital formation by establishing and solidifying ties across households. Compared with households without absent migrants, absent-migrant households are more likely to live in cities other than Tbilisi and, as we would expect, in PSUs with higher concentrations of absent-migrant households. Otherwise, the differences in mean characteristics are modest, and generally not statistically significant.
There is some tendency for absent-migrant groups to be male-headed (61.5 %). Only 14.4 % have been abroad for less than one year, with duration missing for 3.7 %. Most migrated for economic reasons (83.6 %), and most have a head working full-time or almost full-time. They tend to have fewer children and more adults than the households left behind in Georgia. The vast majority (94.6 %) have completed at least secondary schooling. Although Russia remains the most common destination country (33.7 %), many migrant households live in North America or Western Europe (39.9 %).
Table 3 shows the descriptive statistics on the outcome variables we analyze for all households (overall and by remittance status), providing a sense of the magnitude of possible remittance effects on outcomes. Corresponding statistics for the absent-migrant households are in Table S3 in Online Resource 1. The largest expenditure categories for households in the GOTM data are household goods, housing necessities, and health care. Schooling expenses, savings, and gifts to others also figure prominently. Spending on religious activities, leisure, and personal services is lower. Unemployment and poor health are widespread in contemporary Georgia: 38 % of the GOTM households have at least one member unemployed, and 45 % have at least one member in poor or very poor health.15 Roughly one-half of households with 17- to 25-year-olds have at least one enrolled in school. Fourteen percent of households engaged in small business activity in the preceding year; 12 % have Internet access.
On average, households that receive remittances spend more than those that do not in all categories, and they have substantially higher total budgets and total expenditures. The differences regarding business activity, landownership, number of rooms, Internet access, and school enrollment of young adults are quite small. Remittance households have substantially lower adjusted earnings than nonremittance households. Finally, a smaller proportion of them have a member with bad health.
Results
Predicting Receipt of Remittances
The dependent variable in our probit model for remittance receipt among all households is a dummy variable denoting receipt of any remittances, whether from absent migrants or from nonhousehold members (Table 4).16 The probability of remittance receipt is positively affected by rural residence, the household’s level of religiosity, and the number of young and school-age children. The difference between residence in Tbilisi versus other cities is not significant, perhaps because Tbilisi’s lower income volatility and weaker social ties are offset by higher costs of providing services. Household’s mean education level also has no significant effect. Remittance receipt is negatively linked to the number of male (but not female or young) adults and the number of retirees in the household. Male adults may offer alternative income sources to pay for services provided to absent migrants, and elderly household members may offer these services for free.17 As expected, having an absent migrant dramatically increases the probability of receiving remittances.18 The percentages of absent and return migrants in one’s voting precinct also increases the likelihood of remittance receipt, consistent with the social capital perspective. The model predicts remittance receipt well. However, its pseudo-R2 of .380 falls to .066 if we remove the absent-migrant dummy variable.
The much higher probability that absent-migrant households receive remittances calls for a separate analysis incorporating characteristics of absent-migrant groups (Table 5). Twenty households had multiple absent-migrant groups. We initially handled this by treating the unit of analysis as the migrant group rather than the Georgia-resident household (the first set of results in Table 5), but this introduces nonindependent observations: households with multiple migrant groups appear in the sample multiple times. Lacking enough of these households to adjust econometrically for the correlations of their residuals, we instead treat the Georgia-resident households as the unit of analysis and select only one migrant group from households with multiple groups: the remitting group if only one remits, or a randomly chosen group if either all do or all do not remit. This approach facilitates using our probit model results for the propensity score analysis of the impact of remittances, wherein Georgia-resident households are the logical unit of analysis. The results differ only slightly.
As predicted, migrant groups whose heads work full-time or almost full-time, have been abroad for more than one year, migrated for economic motives, live in North America or Western Europe, and have more frequent contact with the Georgia-resident household are significantly more likely to remit. There is weak evidence of a positive effect of the absent-migrant group head’s age (the log transformation fits best). Contrary to expectations, the number of children in the migrant group and its head’s gender and education have no effects. Its number of adults has a negative effect, presumably because it is associated with less, rather than more, disposable income. The effects of variables characterizing the Georgia-resident household and its context are similar to those that we found in our analyses of all households, although for the absent-migrant households, more aggregated specifications of some variables (e.g., total number of children rather than separate effects for young and school-age children) were optimal. The model’s pseudo-R2 is an acceptable .307.
Measuring the Impact of Remittances
The specifications of the probit models reported in Tables 4 and 5 satisfy the balancing condition necessary to apply propensity score matching: within each block of the propensity score distributions, covariate means across treated and nontreated groups are equal. (See Table S3 in Online Resource 1 for details regarding the propensity score distributions.) Overall, one-half of the GOTM households have propensities less than .103, but the median propensity for absent-migrant households is .811, and only 10 % have propensities less than .310. These disparities in the relative proportions of treated and untreated observations in the two samples justify separate analyses on both of them: similar results increase confidence that findings are not artifacts of disproportionate numbers of treated and untreated units.
The impact of remittances on the economic and social well-being of Georgian households is evident from the unconditional differences in mean outcomes across remittance and nonremittance households and the estimated average treatment effects for the treated (ATTs) based on kernel matching (Table 6). The evidence is consistent across samples and methods: remittances increase total household budgets, total expenditures, spending on household goods, savings, and interhousehold transfers. These effects are large, ranging from about one-quarter to one-third of a standard deviation of the variable in question (see the standard deviations for the whole sample reported in Table 3). Remittances also increase spending on personal services, medical care, school expenses, and debt payments: the ATTs are significant using a one-tailed test for one sample and using a two-tailed test for the other. Our inferences for spending on religious activity, rent, and vehicles depend on which sample we consider. Remittances do not affect spending on leisure pursuits or housing needs (water, fuel, and electricity).
Remittances have no association with unemployment or nonremittance earnings (despite the significant difference of unconditional means with respect to earnings). They do not increase school enrollment rates of 17- to 25-year-olds, but they are associated with diminished health risks. The results are mixed for business ownership (a weak positive effect among absent-migrant households only) and number of rooms (strong positive effect among all households only). Remittances are not associated with landownership or Internet access. In sum, remittances in Georgia do not provide disincentives to work, but they do improve health.
Discussion
In Georgia, migrant remittances from abroad have a range of positive economic effects for households that receive them without producing the negative effects often feared in the literature. They directly improve living standards by increasing expenditures on household goods. Some of the items purchased may be imported, but increased spending on personal services undoubtedly has positive spillover effects in the local service sector, and some spending on household items goes to local retailers and producers. Remittances provide insurance and (implicitly) access to loans by enhancing savings and debt payments.19 They improve human capital through more spending on education and medical care, and diminished probability of poor health among household members. Migrant remittances do not create disincentives to work (moral hazard), nor are remittances typically “wasted” on leisure. It is unclear whether remittances promote small business activity, and they do not encourage the acquisition of land or sending young adults to school. However, their largely positive role in Georgia’s longer-term economic development transcends the benefits for the households that receive them.
Several empirical results indicate remittances contribute to social capital formation. First, 8.5 % of households without absent migrants receive remittances from members of other households, often but not always extended family members. The standard NELM framework cannot explain this phenomenon, but social capital formation—establishing and solidifying social ties among communities larger than the household in the expectation of normative and/or economic rewards in the future—is a plausible motive. Second, a household’s religiosity and frequency of contact with absent migrants—both measures of the tightness of its connection to absent migrants and others—are significant predictors of remittance receipt. Third and most important: remittances have a strong positive effect on the amount of money that Georgian households give to other households. We do not know who receives these transfers or whether the transfers reinforce solidarity, establish future obligations, or bestow future economic and non-economic benefits on the donors; the answers to these questions require different data and qualitative studies of the interhousehold transfers facilitated by remittances. However, our data offer indirect evidence that the social capital mechanism we posit is at work: remittances have no effect on interhousehold transfers (ATT = 2.39 among absent-migrant households) for Georgia households in Tbilisi, where social ties between households tend to be weaker. Their effect is large (ATT = 89.38) and highly significant for households elsewhere, where traditional patterns of interhousehold connections are more prevalent. The impact of remittances on interhousehold transfers obtains precisely for those households most likely to have strong social ties to other households.
Altogether, our theoretical reasoning and our empirical findings suggest that remittances can be invested in social capital as well as in human and productive capital. Further exploration of this potential role of remittances should be a goal of future research on remittances. Contextual characteristics that increase or decrease the likelihood that remittances play such a role should be elucidated theoretically and empirically. For example, perhaps the social capital formation aspect of remittances will be greater where public social security provision is especially weak, as in Georgia.
Although our study provides new empirical insights into the economic and social roles of remittances in contemporary Georgia and a promising propensity score–based approach to measuring their effects that can be used in other contexts, it nonetheless suffers from some important limitations. First, some outcome measures have relatively large numbers of missing cases, which we handle using listwise deletion. They may not be missing at random, and more complete data might produce different conclusions. Second, the effects of remittances may vary for different types of households or at different points of the propensity distribution: given space limitations, we leave analysis of such potential variations for future research. An instrumental variable approach would be preferable to deal with potential unobserved variables that may jointly affect remitting and outcomes, but the GOTM lacks variables to serve as valid instruments for remittance sending. Finally, the GOTM data do not support analysis of the mechanisms producing some of the effects we have found and their larger socioeconomic impact at the community level. Our evidence that they contribute to social capital formation would be stronger if we could show how those transfers were used, to whom they were given, and what expectations and social relationships resulted from the exchange. Nevertheless, by precisely measuring a range of the effects of remittances at the household level, we have provided empirical grounds for other researchers to explore mechanisms and consequences more extensively in future studies—both of Georgia and of other contexts where remittances play a major role in local economies.
Acknowledgments
The authors acknowledge the Global Development Network for funding the survey analyzed herein; the University of Delaware’s Title VIII Program for a grant that supported data analysis and writing; Robert Tchaidze, Eric Livny, Randy Filer, Vladimir Popov, Scott Radnitz, and the anonymous reviewers for helpful suggestions and input; and Sophie Shkirtladze and Maka Chitanava for research assistance. Earlier versions were presented at the University of Delaware Title VIII Conference (Sofia, Bulgaria, June 2010), the Population Association of America (Washington, DC, April 2011), the Center for Demography and Ecology at the University of Wisconsin–Madison (July 2011), and the Program on New Approaches to Research and Security in Eurasia (Bishkek, Kyrgyzstan, July 2011).
Notes
For more detailed descriptions of Georgia’s migration patterns since 1991, see International Organization for Migration (IOM) (2008) and Tchaidze and Torosyan (2009).
The transfer data can be found online (http://www.nbg.gov.ge/). See also Table S1 in Online Resource 1 for more details.
NBG specialists told us most remittances are transferred via the NBG because these transfers are reliable, flexible, quick, and cheap, and remittances are not taxed or questioned.
For more-detailed summaries of the debate over the economic effects of remittances, see Taylor (1999), Cohen (2005), Massey et al. (2005), Rapoport and Docquier (2005), de Haas (2010), and Garip (2012).
Precincts in heavily militarized zones, remote mountainous areas, territories occupied by Russian forces, those with fewer than 50 voters, and those with more than 50 % non-Georgian speakers were excluded at the PSU stage for practical reasons. Overall, 7.7 % of voting precincts containing 3.7 % of voters were excluded.
Because the GOTM captures remittances paid by households that left Georgia entirely, the absence of such households in the sample is not a problem for the analysis of the effects of remittances among Georgia-resident households.
Data on any remittance receipt are missing for only 5 % of households.
Frequency of contact may be endogenous if transferring remittances provides occasions for contact. However, most remittances are sent via bank transfer. Thus, our eight-category measure of frequency of contact is a reasonable proxy for the strength of the ties between the Georgia-resident household and migrant group.
This “ignorability” assumption stipulates that no unobservable variables affecting whether a household receives remittances influence the outcomes that we examine. Such unobservable variables would bias our estimates of remittance effects in one direction or the other, depending on how they affect remittance receipt and the outcomes. However, this concern is mitigated by the rich set of covariates and good fit of our remittance receipt models.
We estimate all models using Stata 11.0. For detailed introductions to propensity score matching, see Smith (1997) and Morgan (2001). For a technical treatment of kernel matching and details about estimation, see Becker and Ichino (2002).
Households may have spent money on categories not covered in the survey, so our measures of total budget and total expenditures contain errors and underestimate the true totals. The expenditure data may be error prone because of difficulties remembering or reluctance to reveal high levels of expenditures. However, we do not think that these measurement errors vary systematically by household remittance status, so we doubt that they bias our findings.
Other age ranges for school participation that we tested yielded the same results.
We adjust household earnings for household size in the standard fashion by dividing total household earnings by the square root of household size. Earnings data were missing for at least one member in 18.0 % of households. We assigned a value of zero to those individuals. Additional analyses on households with complete data revealed no differences in results.
We mean-substitute missing values on religiosity and (for absent migrants) duration abroad. We exclude six households with missing data on other variables. The GOTM sample means on some variables can be compared with official data found online (www.geostat.ge).
We define unemployment as not working but looking for work. Parallel analyses that included those not working and not looking among the unemployed yielded the same results.
In optimizing our specifications for this model and the model estimated on the sample of absent-migrant households, we omit most nonsignificant covariates, although we retain some to satisfy the balancing condition necessary for the estimation of propensity scores.
We tested for variation in the “elderly” effect by the presence of children and found no significant interaction.
Households with a return migrant are also significantly more likely to get remittances, but we had to remove this variable to satisfy the balancing property necessary for the propensity score analysis.
Because debt payments are a separate expenditure category, it is unlikely that transfers to other households are capturing repayment of money borrowed to pay the costs of migration.
