4.3 Estimation strategy and instrument
Our objective is to estimate the following model:
$$ CORRUPTION_{ij} = \alpha + \beta MIGRATION_{ij} + \gamma X_{ij}^{\prime } + \pi_{j} + Year\_2011 + u_{ij} , $$
(1)
where, for individual
i living in country
j, the dependent variable
CORRUPTION stands, in alternative specifications, for the variables capturing bribery experience and attitudes towards corruption, the explanatory variable
MIGRATION stands for migration-related variables (remittance-recipient and non-recipient migrant households). Vector
X includes sociodemographic controls, the
π
j
are country/region fixed effects,
Year_2011 is a dichotomous variable for year 2011,
13 and
u
ij
is the error term.
Given the binary nature of the corruption related variables paid a bribe, contacted public official and was asked for bribe, we estimate the models that use them as dependent variables with binary probit. The models explaining attitudes to corruption are estimated with OLS. We first report conditional correlations, i.e., the results of the models that do not address the potential endogeneity of migration and remittances. We then use the instrumental variable (IV) approach to deal with potential endogeneity and determine the effect of migration on corruption experiences and attitudes.
The endogeneity of migration could arise from reverse-causality, if corruption at home pushes people abroad or if people emigrate with an intent of accumulating resources toward engaging in corruption in the home country at a later stage (e.g., paying a doctor for major surgery or accelerating the licencing process when starting a new business). The very process of preparing to emigrate can make people more prone to corruption so as to secure speedy delivery of passports, visas, certificates of health and so on. If such sources of endogeneity are present, we would expect the effect of emigration on corruption, estimated in correlational analysis, to be biased upward. In addition, unobserved characteristics of people (and/or households) may exist that are correlated with both the willingness to migrate and the propensity to bribe public officials. This could produce an upward or downward bias in the correlational analysis, depending on the proclivity to corruption of emigrants and their household members.
Following Böhme et al. (
2015) and Höckel et al. (
2015), we use the interaction between historical municipality-level migration networks and the economic conditions at the main migrant destination countries (municipality-level variable) as an instrument for emigration at the household level. Specifically, for each municipality, we take the summation of the GDP growth rates of major migrant destinations weighted by the pre-existing migrant networks (the number of migrants relative to municipality population) in those destinations. Formally, the instrument can be expressed as follows:
$$ Migration{ - }Growth{ - }Interaction_{c} = \mathop \sum \limits_{j} \frac{{Migrants_{c, j, 1971} }}{{Population_{c, 1971} }} \times \left( {\frac{1}{5} \mathop \sum \limits_{\tau } GDP growth_{\tau ,j} } \right) , $$
(2)
where
c is municipality,
j = 1, 2, …,
J is the migration destination country, and
τ = 2001, …, 2005 are the five years over which the destination countries’ growth rates will be calculated. The choice of the 2001–2005 period for the destination countries’ growth rates, and of the year 1971 for pre-existing migrant networks, is explained in more detail below. Note that the same time horizon for growth rates (2001–2005) is used to construct the instrument for respondents interviewed in both 2010 and 2011, and, as such, the instrument is not time-varying.
We expect the municipality-level networks-destination-growth interaction to predict current emigration in two ways. First, pre-existing migrant networks are known to be powerful predictors of the emigration decision. Existing migrant networks reduce migration costs for subsequent migrants by conveying information about the destination country prior to the move, by providing financial assistance, facilitating employment and accommodation, and giving support in various forms after the move (see, e.g., Massey et al.
1998). It also has been shown that networks played a crucial role in explaining successive waves of Yugoslavian out-migration (Brunnbauer
2009). Second, high GDP growth rates in the destination countries, and favorable economic conditions more generally, imply better job opportunities, which act as a pull factor for prospective migrants (Antman
2011). The network component of the instrument reinforces the destinations’ growth component: the larger the network is, the more existing migrants will be able to convey information, particularly information about favorable economic conditions, about the destination and help migrants upon arrival.
The information on the historical migrant networks comes from the 1971 Population Census of Yugoslavia (Baucic
1973b). During this census, information on the number of migrants, as well as migrants’ gender, age, education and, crucially for this study, destination countries,
14 was supplied by migrants’ family members and, when the whole household had emigrated, by neighbors. Only the records on “Yugoslav workers temporarily employed abroad” were collected by the census: the data thus capture only guest worker migration flows, which started in the mid-1960s and hit their peak in 1971, and underestimate the total stock of Yugoslav emigrants at that time. Importantly for our study, all information on migrant stocks is available at municipality/commune level.
We represent economic conditions at the migrant destinations by the average of the destination countries’ GDP growth rates in the 2001-to-2005 time-span. That period precedes the advent of the global recession and is characterized by strong economic growth across most of the world, sending a strong signal to prospective migrants. Our IV analysis will, thus, capture the effects of migration and remittances that are driven by the relatively recent economic developments in historical migrant destinations.
The exogeneity of the instrument has to be considered from the point of view of both its network and its GDP growth component. One can convincingly rule out any direct effect of the destination countries’ GDP growth rates on corruption outcomes in the migrants’ countries of origin: it is difficult to think of channels through which economic conditions in foreign countries would lead to differential corruption outcomes within a migrant’s source country—apart from the effect running through migration. The possibility that historical migration networks, the other component of the instrument, might be directly related to home-country corruption invites some discussion.
15 For example, it is not inconceivable that local guest worker emigration in the 1960s and 1970s might have been driven by the local-level corruption at that time. If such local corruption persisted, the historical migration networks may be directly related to differences in corruption incidence today. An ideal solution would be to control for the local-level prevalence of corruption in the Yugoslavia of the 1960s and 1970s. Such data, however, are not available. Instead, we use the 1971 municipality-level illiteracy rate, sourced from the 1971 Yugoslav Population Census, as a proxy for local-level development and corruption in both stages of the IV estimation.
16
To validate the network component of our instrument further, it is useful to outline the actual reasons behind the local-level variation in the early 1970s’ migration rates. First, the variation can be traced back to the different regional rates of migration that the countries of ex-Yugoslavia experienced at the turn of the 20th century and the interwar period (Baucic
1973a; Brunnbauer
2009). This, for example, explains why certain regions of Croatia and the Dalmatian coast of Bosnia and Herzegovina were the first to embrace guest-worker emigration opportunities, exhibiting the highest municipal rates of emigration. Prior to WWI, those regions were part of the Austro-Hungarian Empire (which had a particularly favorable emigration regime), had close access to seaports and, in addition, were subject to agricultural shocks (e.g., the
Phylloxerra epidemics that destroyed much of the profitable Dalmatian wine industry (Mlinaric
2009; Brunnbauer
2009)). Those factors contributed to early out-migrations from the region and to the development of strong cultures of migration that have persisted.
The demand-driven nature of bilateral recruiting and guest worker migration programs is another reason why emigration rates exhibited regional variation. Representatives of host countries’ manufacturing companies recruited workers directly, in some cases through the Yugoslav state employment agencies. However, the distribution of foreign recruiters within Yugoslavia was uneven. For example, manufacturers from West Germany, the most important destination of Yugoslav migrants, recruited workers from specific regions in Yugoslavia, because they had good experience with workers from these regions (Novinscak
2009).
It should also be stressed that the early guest-worker out-migration from Yugoslavia clearly was considered, by both migrants and the ruling elites, to be temporary. Migrants—often young, low-skilled men with agricultural backgrounds—went abroad to earn money that they planned to invest back home in building/extending houses or buying land and agricultural machinery (Pichler
2009; Novinscak
2009). Property sales at home were rare, and the immediate family (spouse, children) typically were left behind, confirming return intentions of the first guest-worker migrants (Brunnbauer
2009). This, again, does not support a possible conjecture that the guest worker migration was driven by the extent of local corruption.
Overall, the context and demand-side-managed nature of the guest worker out-migration in former Yugoslavia, coupled with the use of within-region variation in migration rates and a control for local-level development at the time of migration, render it very unlikely that the variations in historical local-level migration rates are related directly to today’s local-level corruption. This makes us confident that the network component of the network-growth interaction instrument is exogenous to present-day corruption.