The Role of Migration Experience in Migrants’ Destination Choice

Abstract In this article, we employ a panel household survey from Tajikistan to study labor migrants’ location choices in Russia. We find that labor migrants from Tajikistan consider a wide variety of economic, demographic, and geographical characteristics of Russian regions when making location choices. We also find that experienced migrants are less responsive to current regional characteristics that might suggest path dependence in destination choices by experienced migrants.


Introduction
It has long been noticed that migrants' distribution over the receiving country does not replicate the one of natives (Chiswick and Miller, 2004). Migrants often concentrate in particular locations. Understanding the reasons of observed settlement patterns is important for many reasons. The first reason is because of the impact that migrants might have on the natives' labor market outcomes (Longhi et al., 2008;Kerr and Kerr, 2011), settlement decisions (Borjas, 2006;Mocetti and Porello, 2010), and other outcomes. The second reason is that migrant success in the receiving country considerably depends on a settlement decision (Damm, 2009;Edin et al., 2004;Chiswick and Miller, 2005).
The economic literature on migrants' location choices points to several important factors determining migration destinations. Among them, the labor market characteristics of the receiving regions and co-ethnics concentration have been extensively explored. Regional economic characteristics represent opportunities in the labor market, while concentration of co-ethnics represents available migrant networks. Migrants rely on networks that provide information and direct assistance with job search, accommodation, legal issues, and so on.
Networks also provide ethnic goods such as food, clothing, social organizations, religious services, and so on. Thus, current migrants' choices depend on the migration decisions of their predecessors.
Research results about the importance of the two aforementioned factors vary considerably depending on migrant characteristics and data source. For example, in the case of the USA, Bartel (1989) and Kaushal and Kaestner (2010) showed, using the census data, that the level of immigrants' concentration is important while economic factors have little or no effect.
Immigrant concentration was a significant factor in many other studies and survives even control for location fixed effects (Jaeger, 2007). Other studies (Zavodny, 1999;Jaeger, 2000;Jaeger, 2007), employing admission data from the Immigration and Naturalization Service, show the considerable role of regional economic factors in attracting migrants. The relative importance of the factors depends on various migrant's personal characteristics: education, age, gender, marital status, previous occupation, country of origin, visa type, and legal status (Scott et al., 2005;Pena, 2009;Jaeger, 2000;Jaeger, 2007;Bartel, 1989;Kaushal and Kaestner, 2010). In general, economic factors turn out to be more important for migrants who are employment-based, better educated, more qualified and legal, while the effect of ethnic concentration is higher for the low-skilled.
Most of the discussed papers study the location choices of the long-term or permanent and predominantly legal migrants, while location choices of the short-term and circular migrants seem to be understudied (a notable exception is a series of papers by Bauer et al. (2005Bauer et al. ( , 2007Bauer et al. ( , 2009). In the case of repeated migration episodes, personal migration history plays an important role. Path dependence in location choice arises because migrants have higher expected returns and lower costs at familiar destinations. First, there are unobserved characteristics of destinations that do not vary with time (e.g., migrant's preexisting connections to people residing in the region). Second, migrants build new connections upon arrival both with the existing ethnic network, local population, and employers. They make other investments in a receiving region-specific human capital (e.g., knowledge of local laws, rules, and local labor market).
With accumulation of experience in the receiving country's labor market, migrants can rely on their own knowledge and connections to find employment. Thus, experienced migrants become attached to their previous destinations and may not be sensitive to the current situation in the regions.
In this article, the author explores the location choices of temporary labor migrants in Russia, both legal and illegal, using individual data from a sending country -Tajikistan. The article has two main aims. First, to find out what regional characteristics are important for migrants' location choices. Second, to find out how migration experience affects the determinants of location decision. To the author's knowledge, this is the first article that empirically investigates the destinations of temporary migrants in Russia using micro-level data and one of a few that accounts for migration experience while exploring the factors of location choice.
The data employed in this study are the 2007 and 2009 1 waves of the Tajikistan Living Standard Survey (TLSS) and its extension, the Tajikistan Household Panel Survey (THPS) in 2011. 2 The author finds that both regional economic characteristics and the presence of co-ethnics determine labor migrants' choices. Apart from traditional factors, the author accounts for the differences in migration policy across Russian regions. In the studied period of 2007-2011, there were migration quotas defined at the regional level. Introduced to fight illegal migration, these quotas had mixed success in shaping migration streams. For the majority of regional characteristics, the author finds that the effects on the probability of migrants choosing a region declines with migration experience. So, current regional characteristics are much less relevant for the experienced migrants.
The rest of the article is organized as follows: after describing the patterns of Tajik migration to Russia in Section 2, the author discusses the theoretical foundations in Section 3. The author presents the data description and descriptive statistics in Section 4. Finally, the author presents the empirical results and their robustness in Section 5.

Tajikistan -Russia migration facts and 2008-2009 financial crisis
Tajikistan is a mountainous country with a predominantly rural and fast-growing population.
Over the past quarter of a century, Tajikistan has had the lowest per capita GDP among the The collapse of the Soviet Union and the civil war caused a surge in migration. Internal migration was mainly caused by the civil war and external migration, in the first years of independence, occurred mainly for ethnic reasons, while in the following years economic reasons prevailed. Migration was a response to a difficult economic situation in the country and was a widespread survival strategy for Tajik households. Tajikistan's economy is highly dependent on remittances: the ratio of remittances to GDP reached almost 50% and was the world's highest in 2008 (World Bank, 2007Bank, , 2008Bank, , 2009a. Estimates of the number of workers sent abroad vary. According to the household surveys, the stock of Tajik citizens who worked abroad was 300,000 in and 560,000 in 2009(Kumo, 2012. Choudinovskikh and Denissenko (2013) give an estimate of 430,000 that is based on a Labor force survey 2008. About 95% of Tajik migrants go to Russia (according to the TLSS data). The number of registered labor migrants from Tajikistan in Russia was 250,000 in 2007Russia was 250,000 in , 360,000 in 2009Russia was 250,000 in , and 330,000 in 2011 15-16% of the registered labor migration, making Tajikistan the second largest among all sending countries.
Russia started attracting labor migrants from the post-Soviet space in the period of rapid economic growth of the 2000s. Officially registered foreign workers made up 2.5-3.5% of the Russian labor force in (Russian State Statistical Committee, 2011. In some industries, this share was much higher. For example, in construction where almost 40% of registered foreign workers were employed in 2009-2010; their share was about 16%. Trade was the second most important area of employment, attracting 18% and 17% of foreign workers in 2009 and 2010, respectively. Migrants were predominantly male and low-skilled. As we will see later, migrants from Tajikistan share this typical profile. Since 2007, a policy of regional migration quotas was introduced. The quotas were the regional limits for the number of work permits that could be issued for foreign low-skilled workers. When first introduced by the law in 2007, the quotas were not binding. The number of permits was 6 million for the whole country. The regional distribution of quotas was first defined by central authorities, taking into account the socioeconomic situation in the regions. by the Presidential decree, by 50% from the planned nearly 4 million. Complications were also introduced to the application procedures for the firms that wanted to hire migrant workers. As a result, the number of issued work permits fell in 2009 by 30% compared to 2008. Migrants also faced more abuse by employers and a higher risk of deportation (Marat, 2009).
The crisis heavily hit the Tajik economy through a slump in the prices of its exports (cotton and aluminum) and the drying of the migrant remittances stream. These two factors resulted in a reduction of GDP growth rates from 21% in 2008 to 4% in 2009. Still, the crisis only spurred the participation of Tajik households in labor migration. Kröger and Meier (2011) show that members of households affected by the slump in export prices were more likely to send household members to work abroad. Additional household members were sent to support the stream of remittances (Marat, 2009;Danzer and Ivaschenko, 2010). Danzer and Dietz (2018) show that, on average, the duration of migration and net income from migration increased during the financial crisis in spite of the shrinking average wages.

Theoretical model and empirical estimation
To explore the determinants of location decisions of Tajik migrants in Russia, the author employed a standard discrete choice model, which has been used in many previous studies (Bartel, 1989;Bauer et al., 2009;Jaeger, 2007). This approach, in the case of micro-level data, allows for the choice between a limited number of destinations to be modeled.
Let us assume that a potential migrant makes a location decision each period. When making this migration decision, a migrant compares costs and benefits of migration to each of the possible destinations and decides on the location with the highest net benefits. The benefits from migration are mostly related to the expected earnings in that particular destination. The costs of migration depend on many factors that include the direct costs of obtaining tickets, travel documents and passports, costs of settling up at the destination, and housing and registration costs.
Formally, let individual i in period t choose between j = 1, …, N destinations that bring utility U ijt . The probability of choosing destination j is then: Utility of the individual i from destination j in period t, U ijt = f(G ijt ), is a function of the factors that contribute to the gain from migration G ijt : where p ijt -probability of employment, w ijt -migrant's wage, and C ijt -migration costs.
In turn, each of the factors p ijt , w ijt , and C ijt depends on four components. These are destination characteristics Z jt-1 (such as labor market characteristics and co-ethnic concentration), migrant's personal characteristics X it (age, gender, education, and migration experience), their interactions X it *Z jt-1 and a random component e ijt .
The author assumes a linear relation between the utility and the factors X it , Z jt-1 , and e ijt : Substituting Eq. (3) into Eq. (1) and assuming that returns to individual characteristics, α, are the same at all destinations, we get: Assuming that e ijt is an independent and identically distributed random variable with Weibull distribution Eq. (4) could be written as: Obtained expression is the conditional logistic regression model or McFadden choice model.
Parameters β and γ are estimated by maximization of the likelihood function that is a sum of logs of Eq. (5) for all observations.
An important property of this model is the independence of irrelevant alternatives (IIA).
It can be easily seen from Eq. (5). The property means that the relative probability of choosing two alternatives depends only on these alternatives' characteristics and does not depend on the choice set. This property allows for destinations that did not enter the sample to be ignored.
Although, in the case of spatial choice, we might expect this assumption not to be fulfilled, as less distant alternatives can be expected to be closer substitutes for each other. As shown in (Cushing, 2007) and (Dahlberg and Eklöf, 2003), results of a conditional logit model in a migration context are sensitive to the IIA assumption with the problem of omitted variables.
If a model is well-specified, results of the conditional logit model are comparable to mixed logit and multinomial probit models that relax the IIA assumption. An important way, at least partly, to account for a region's unobserved heterogeneity, is control for groups of regions that are similar (Schneider and Kubis, 2009). In the case of Russia, a common way to aggregate regions are geographic macro regions -federal districts. To partly eliminate the problem of the IIA assumption violation, controls for the federal districts will be included.

Data and descriptive statistics
The samples of TLSS and THPS are representative on the national, regional, and rural/urban As can be seen in Table 1, Tajik migrants are predominantly young males with a secondary general education from large rural households. In Russia, Tajik migrants occupy low- The average time spent abroad is 2.9 years with a range of 1-17 years. Table 2 shows how migrant characteristics vary with migration experience. We see a clear trend of more females in the later migration cohorts. Also, migrants with less experience have a lower education level. Indeed, the trends we see are a combination of changes in selection into migration and selection out of migration. However, different socio-demographic profiles should be taken into account in the analysis.
The survey asked each migrant which town or city they had been living in, and also for those who were still officially a migrant, which city were they currently living in. The author assumes that in the case of return migrants, they report their last destination as they theoretically could change it during the trip (the model assumes that migrants may choose the location for each period). Although the survey question asked about the migrant destination city, the  analysis will be carried out on a regional level. First, we cannot be sure whether a migrant actually stays in the city as opposed to just naming the largest city of the region. Second, regional-level statistics are more reliable in Russia. According to the aforementioned model, migrant destination choice is driven by the probability of employment, wages, and migration costs. These can be predicted with various regional characteristics (Z jt-1 in our model's notation). As discussed earlier, several groups of regional factors contribute to a migrant's expected utility from destinations, with labor market characteristics and networks being of particular importance. Regional unemployment, population size, population density, labor force sectoral structure, the size of the construction sector 5 and, partly, migration quotas 6 all contribute to predicting migrant employment probability (see sample averages in Table A2 in Appendix). The average regional wage 7 is a proxy for differentials in migrant wages. As a robustness check, the author constructed an indicator which is a weighted average of mean wages in ISCO occupational groups, in which migrants work.
This measure could be a better proxy for a migrant's expected wages. Indeed, local wages are not perfect proxies because regions differ in terms of the migrant position in the distribution 4 For 9% of migrants, exact destination is unknown. 5 Regional statistics is taken from sources of the Russian State Statistical Committee (Russian State Statistical Committee 2018c; Russian State Statistical Committee, 2018d). 6 Information on migration quotas was taken from the orders issued by the Russian Ministry of Health and Social Development (2007Development ( , 2008aDevelopment ( , 2008bDevelopment ( , 2009Development ( , 2010. 7 Nominal wages are used as labor migrants send considerable part of their earnings as remittances, so they are expected to consider nominal earnings.  (Table 3).
Economic dynamics measures might affect migrant choices both through wage, employment, and migration costs. For example, adverse economic dynamics not only means an increase in current unemployment and a reduction in current wages and economic activity, but may also affect the migrant position through changes in attitudes toward migrants, and the willingness to employ them. On the one hand, during times of crisis, local populations become more hostile toward migrants, and there might be a higher risk of deportation. On the other hand, businesses might be more willing to replace their more expensive local workers with cheap migrant labor. Another consideration concerns the change in the construction sector output. The considerable change in the industry may create a deficit of local labor resources and make employers more likely to employ migrant labor. We do not see a considerable decline in the data (Table A2 in Appendix) because regional characteristics of the year preceding migration are used in the regressions.
Networks are another important factor that drives location decisions. Networks reduce costs of migration and improve employment prospects and wages (e.g., ethnic discrimination is absent when migrants are hired by more prosperous co-ethnics). There is no doubt that migrants rely on personal connections while making location decisions. For example, about 56% of return migrants in the 2009 TLSS report that connection with relatives, friends, or acquaintances is considered to be the main reason for choosing the destination country. About 16% report that the reason was a pre-arranged job, and nearly 20% report the reason as, they   (Bauer et al., 2007;Beine and Salomone, 2013;Beine et al., 2015).
Yearly data on the permanent Tajik  It is difficult to argue why this change happened. One explanation could be the extremely high cost of living in Moscow, especially housing prices that one incurs when settling with a family.

Location choice determinants
The author starts with identifying regional characteristics that can explain migrants' location choices. The results presented in Table 4  Most of the variables included in the model are statistically significant and have predicted signs. Migrants go to regions with larger population size and density and higher wages. The co-ethnics concentration variable-share of ethnic Tajiks in the regional population-is also significant and is a strong predictor of location choice. Regional economy structure also affects migrants' decisions. Migrants are more attracted to regions with more people employed in industrial production, which indicates a stronger economy, and less attracted to regions with 8 Permanent population is defined as individuals who spent 12 months or more in the country's territory.  Unemployment rate, % 1.20*** (0.08) 1.28*** (0.10) 1.20***(0.08) 1.14** (0.07) Change in unemployment rate (in pp.) 0.89* (0.06) Log of average wage (in 1,000 RuR) 6.72*** (3.14) 6.98*** (3.21) 6.73*** Notes: Dependent variable -migration of individual i to region j in period t; controls for federal districts included; the values presented are the odds ratios; standard errors clustered in primary sampling units in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. more people employed in utilities and personal and social services. Probably, in the former regions, the local population is ready to take predominantly low-skilled jobs in the sector and therefore provides less job opportunities for migrants. Characteristics of economic dynamics have predicted signs, but change in the unemployment rate is significant only at the 10% level, and change in construction sector is not statistically significant.
Some of the factors do not have intuitive effects. Migrants go more probably to regions with higher unemployment. We cannot expect that a generally low unemployment across Russian regions will be an important factor for migrants choosing their destination. Also, higher unemployment might mark regions unattractive for internal migrants. Higher relative migration quotas associate with less probability to choose a region. Thus, the quota mechanism is not able to route migration streams. We are going to see whether this result is stable across the years, as the mechanism of quotas distribution was not the same.
Multicollinearity can be an issue in our data. In the pooled sample regressions where the number of potential destinations is larger, variance inflation factor (VIF) statistics do not exceed 4, which can be considered acceptable. Although in regressions for separate years, there is a strong multicollinearity because the number of alternatives becomes smaller. Therefore, results for separate years are not reported.
To show how the choice model evolves over time, the author adds interactions of the destination characteristics with the year of survey. Figure 1 depicts the odds ratios of the regional characteristics for the three survey years. Results are quite stable across the years. Still, we can see some dynamics in the effects of some variables. We see increasing importance of wages, population size, and density. These results reflect slowly rising concentration of migrants in Moscow. Interestingly, during the crisis year of 2009, migrants more consider the networks.

Robustness of results
Results do change when a different combination of controls is included. In particular, when population density is dropped from the regressors' list, unemployment is no more significant and the coefficient for average wages is rising considerably ( Table A3 in Appendix). Still, we keep the specification with population density as a baseline model as the VIF statistics do not suggest that there is a multicollinearity problem.
A great concentration of migrants in Moscow creates a concern in our results. Dummy variable for Moscow that captures its time-invariant characteristics (e.g., historical network) is not statistically significant (  Another robustness check is using an alternative indicator of earnings-weighted sum of occupation-specific wages. 11 It gives a close result to a basic specification (Table A3 in Appendix).

Experience effects
Now let us turn to analysis of the effect of previous experience. To do this, the author adds to regressions an interaction between regional characteristics and migration experience measured by the number of years a migrant was abroad as suggested by equation (5). The sample in this case includes only the years 2009 and 2011. Figure 2 presents the predicted odds ratios of the regional variables depending on the migrant's experience abroad (regression coefficients can be found in Table A4 in Appendix).
Results show that recent migrants are more sensitive to regional characteristics. An exception is the change in the construction sector, as experienced migrants might have better connections with employers and access to information about demand in case of new construction projects.
The observed changes in regional characteristics' effects with migration experience may reflect the importance of experience itself, the difference in decision-making between the cohorts or the change in migrant characteristics over time. To rule out the latter explanation, 11 The indicator is calculated using migrants' occupations reported in the TLSS-THPS data and occupation-specific wages reported by Russian State Statistical Committee (Russian State Statistical Committee, 2018c).

Figure 2
Odds ratios of regional variables with number of years abroad. Note: 95% confidence intervals depicted.
the author adds to the regressions the interactions with education, age, and gender (Table A5 in Appendix). The coefficients for the interactions between experience and regional characteristics do not change their significance or become more significant and only slightly change in size. Thus, the change in composition of migrant streams does not explain the observed patterns of experience effects. We also see that education and gender of migrants affect the importance of regional characteristics for the destination choice. Male migrants and migrants with professional education as opposed to those having secondary general and below are more sensitive to economic and demographic characteristics such as unemployment, wages, size of construction sector, and population size and density.
The other two hypotheses are both valid. Indeed, there exists path dependence in destination choice, so experienced migrants choose their destinations with less regard for current regional characteristics. On the other hand, earlier migrant cohorts could be less oriented toward the regional labor market characteristics as well as the organized migration network.
Instead, they could rely more on their personal connections which are usually more randomly distributed throughout the country. Later cohorts rely more heavily on the organized network that emerges in regions with a strong economic position.

Conclusion
In this article, the author investigates the determinants of labor migrants' location choices within the receiving country by exploring the case of Tajik labor migrants in Russia. An emphasis is made on how previous migration experience affects the determinants of migrant choices.
The author finds that both the concentration of permanently settled co-ethnics and regional economic characteristics determine labor migrants' choices. The author did not find that the quota policy had a stable result on migrants' placements. Results show that economic dynamics had a weak effect on defining migrant locations, but the measures that were used could not capture the world economic crisis well enough.
Migration experience is an important factor affecting factors of location decisions. Such an experience expands migrant networks, both among other migrants and the local population, provides better knowledge of local labor market and formal and informal rules. Thus, migration experience is expected to improve employment probability and reduce costs of migration. At the same time, migration experience makes migrants less sensitive to current economic situation in the regions. The author finds that effects of the majority of factors decline with migration experience; therefore, current regional characteristics are much less relevant for the experienced migrants. These results suggest existing inertia in the geography of migrant concentration: once a location has been chosen, a migrant keeps on going there even if the situation has been changing. On the other hand, new streams of migrants are able to quickly decide to change the geographical location of settlement.
Temporary migrants are often considered to be the most mobile and ready to relocate part of the workforce. Our results show that previous migration experience affects current choices and reduces the responsiveness of migration streams to the changing economic environment in the receiving regions. In the absence of direct estimates of the sensitivity of migrant flows, these results may help predict how flows of migrants to Russia will react to macroeconomic shocks. The dataset of Russian regional statistics compiled by the author from the Rosstat open sources, generated during the current study, is available from the author on reasonable request.

Competing interests
The author declares that he has observed these principles.

Funding
The article was prepared within the framework of the HSE University Basic Research Program and funded by the Russian Academic Excellence Project '5-100'.

Author contributions
All the work in this paper was performed by EC. She read and approved the final manuscript.