Regional differentiation significantly influences individual decisions regarding labour market behaviour. Numerous studies worldwide emphasize the relevance of youth migration and identify recent graduates as key drivers of economic development (Busch and Weigert 2008; Di Cintio and Grassi 2011; Liu and Shen 2014; González-Leonardo et al. 2022). Research also shows that young people, who in some contexts constitute more than half of all migrants (Karachurina and Mkrtchyan 2010; Bernard et al. 2014), often migrate to obtain high-quality education or secure employment with higher wages (Knapp 2013).

Cross-regional labour migration after graduation has attracted particular scholarly attention because the motives for relocation and its outcomes are not directly related to educational or family considerations. In other words, researchers focus on the pure effects of migration (Di Cintio and Grassi 2011). Career advancement and wage increases are therefore among the primary outcomes of post-educational migration, and studies suggest that returns from post-educational relocation are realized more rapidly than those from educational migration (Nowok et al. 2013).

Nonetheless, migration may yield mixed results, as migrants often experience losses in social networks, educational opportunities, and job connections. Empirical evidence documents several successful cases with substantial returns from migration for graduates (Yankow 1999; Venhorst and Cörvers 2018; Ward 2022). At the same time, alumni frequently encounter challenges; for instance, they may struggle to adapt to new labour market conditions (Alexander 2021). Such difficulties can delay the accumulation of human capital and partially affect productivity and job satisfaction [ibid.].

This raises the question: what evidence exists regarding post-educational migration in Russia?

The issue of substantial regional differentiation in Russia is linked to several unique characteristics, making it a significant and interesting area of research. First, many Russian regions differ simultaneously in terms of economic structure, gross regional product (GRP) levels, social and healthcare indicators, cultural factors, the quality of higher education institutions, and, of course, labour market conditions (Oshchepkov and Kapelyushnikov 2015). Consequently, uncertainty regarding the most favourable region for relocation influences individual migration decisions.

Second, internal migration patterns in Russia differ from those observed in other countries. Unlike in many developed nations, the peak of migration activity occurs among the youngest working-age population, reflecting the early start of education and careers (Gabdrakhmanov 2019; Karachurina and Mkrtchyan 2021). Previous studies also indicate that Russia exhibits lower overall migration flows compared to some developed countries such as the USA, the United Kingdom, and Australia, while showing patterns similar to Germany and exceeding those of the Czech Republic, Slovakia, and Poland (Karachurina and Mkrtchyan 2021). Therefore, special attention should be given to migration occurring during the school-to-work transition.

Third, the effectiveness of regional labour markets is closely correlated with the overall economic development of a region (Oshchepkov and Kapelyushnikov 2015). A consistent pattern can be observed in Russia, with migrants moving from the East to the West and from the South to the North (Mkrtchyan and Florinskaya 2018b; Gabdrakhmanov 2019; Emelina et al. 2022). Researchers have identified leading regions that consistently attract migrants, as well as “outsider” regions from which outgoing flows originate.

The strongest leaders are Moscow and Saint Petersburg, which feature a concentration of high-quality universities, a wide range of employment opportunities across various sectors, and favourable living conditions. These agglomerations are characterized by both high standards of living and strong labour market indicators, particularly high wages (Mkrtchyan and Florinskaya 2018a, 2018b; Gabdrakhmanov 2019). Another group of leading regions includes areas primarily specializing in mining and manufacturing, such as the Ural, Siberian, and Far Eastern regions. These regions are associated with high wages but limited social infrastructure and underdeveloped healthcare systems. Employment in these regions also has specific requirements, often depending on specialized skills and constrained labour demand (usually this refers to oil-producing regions with rotational employment).

In contrast, the Southern and Caucasian regions typically experience high unemployment rates, low-quality jobs, and low wage levels, placing them in the second group [ibid.]. These territories are characterized by both weaker labour market conditions and lower living standards.

Several key studies focus specifically on youth and graduate migration in Russia. The most significant factors influencing graduates’ migration decisions are wages, as well as regional levels of poverty and unemployment (Antosik and Ivashina 2021). The authors also note that young people are often attracted to dynamic labour markets, where job search times are minimized, a wide selection of vacancies is available, and wages are flexible.

However, not all graduates relocate to Moscow or Saint Petersburg. Many choose smaller economic centres near their home region or the region where they studied (Antosik and Ivashina 2021). Furthermore, the decision to relocate among young professionals is strongly influenced by the opportunity to find employment that matches their skill level. This tendency is particularly evident in migration toward regions engaged in innovation and research (Moskvina 2019). For example, some graduates move to the Far Eastern territories, which are rich in natural resources and provide opportunities for research activities [ibid.].

The quality of higher education institutions also plays a vital role in migration decisions, both for educational and post-educational migration (Kashnitsky et al. 2016; Moskvina 2019; Borzykh 2022). For recent graduates, regions with high-quality educational institutions can serve as a positive signal of a more developed socio-economic environment, thereby increasing the attractiveness of migration to those areas (Bernard et al. 2014).

In summary, post-educational migration in Russia is shaped by specific characteristics of both regional educational systems and labour markets. These, in turn, influence individual migration decisions and migration outcomes. At the same time, university graduates retain valuable human capital that lagging regions need to bridge existing gaps (Antosik and Ivashina 2021).

Most existing studies do not analyse individual-level migration effects, focusing instead on geographical trends, the scale of migration outflows, and general migration drivers. Furthermore, research on internal migration often relies on macro-level data, whereas our study emphasises individual-level observations. Data collected at the individual level, rather than in aggregate form, allows researchers to assess in greater detail both the factors influencing migration decisions and migration outcomes.

By doing so, this study aims to fill a gap in the literature by estimating wage gains from migration for recent graduates. The article contributes by highlighting the value of wage differences between migrants and non-migrants, which can inform policy decisions and regional development strategies.

The empirical part of the research is based on Monitoring of Graduates’ Employment (MGE) provided by the Ministry of Labour and Social Protection and the Federal Service for Labour and Employment of the Russian Federation. It represents various datasets that combine nationwide administrative data about recent graduates. Educational data transmission is compulsorily implemented by almost all state and non-state educational organisations across the country (excluding security force organisations). It is merged with a dataset containing individual information about employment characteristics, and then supplemented with data on gender, age and the presence of various social transfers. The main advantage of MGE, in comparison with previous existing data, is the availability to analyse and evaluate the individual career and educational trajectories of graduates at the country-wide level.

In this paper, we focus exclusively on individuals who graduated in 2021 and emphasize the labour migration of university graduates. We use total monthly wages at the individual level for the period from July 2021 to July 2022. The initial sample, after removing duplicates, comprises approximately 669.000 graduates.

First, we restrict the sample to full-time bachelor’s and specialist degree graduates. The resulting sample includes about 250.000 bachelors and 75.000 specialists¹. Second, additional datasets on continuing education, employment and its characteristics, as well as average admission scores to universities (i.e., average Unified State Exam grades), are merged with the educational database. Accounting for technical issues during the initial data collection, the overall sample ultimately consists of 235.400 bachelors and 70.600 specialists.

Finally, data on school education are incorporated, which are directly used in the probit regression model. Matching school-level data to the overall sample posed certain challenges; consequently, the regression sample consists of 183.266 bachelors and 26.890 specialists. To address the reduction in observations, we tested the final dataset for sample bias and confirmed its absence.

Since the MGE dataset provides only individual-level data, regional characteristics were added through the employment region variable, for which we calculated aggregate indicators. A more detailed description of all selected variables is provided in Table A1 in Appendix 1.

To identify the most significant determinants of migration and to estimate the returns from migration as one of the main outcomes, we employ two distinct models. First, we implement a probit model with Heckman correction, controlling for graduates’ specializations. In the context of labour migration, non-random selection may arise in the probability of employment due to unobserved values of the dependent variable. Consequently, all estimates based solely on the employed subgroup may be subject to sample selection bias. The concept of sample selection correction was introduced by J. Heckman and later extended to the evaluation of probit models using the Heckman correction (Heckman 1976; Van de Ven and Van Praag 1981).

Explanatory variables for the main equation are described in Table A1 (Appendix 1). The selection equation for employment probability includes the presence of a disability and participation in targeted training. Targeted training programs provide financial support to students during their education and require them to enter a contractual agreement with an employer, sometimes even in the early stages of their studies. Additionally, we include another variable in the selection equation, which is discussed in the Limitations section.

Second, we estimate a modified Mincer equation with Heckman correction. Given differences in both migration decisions and wages across specializations, we estimate multiple wage models. Considering disparities in labour demand, the effect of migration may vary by specialization. Accordingly, we focus on STEM fields, medical, agricultural, social sciences, humanities, and education.

The dependent variable for the probit model is a migration dummy, while for the Mincer equation we use the natural logarithm of average monthly wages.

The general equation for both models is as follows:

$y_i^{\ast }=x_i^{\prime }\beta +{\epsilon }_i$, (1)

where $y_i^{\ast }$ is the latent variable in the probit model and the main equation for the Mincer model, $x_i^{\prime }$ is the vector of explanatory variables, β is the vector of coefficients, and ε_i represents the random error term.

For the probit model, the observed migration dummy is defined as:

$y_i^{\text{probit}}=(y_i^{\ast }>0)$. (2)

Selection equation for both models:

$z_i^{\ast }=w_i^{\prime }\gamma +u_i$, (3)

where $z_i^{\ast }$ is the latent variable in the selection equation, $w_i^{\prime }$ is the vector of explanatory variables, γ is the vector of coefficients and u_i represents the random error term.

Then the binary variable representing employment status, taking the value 1 if the individual is employed and 0 otherwise, is defined as:

$z_i=\{\begin{array}{l}1,\text{if}{z}_i^{\ast }\ge 0,\text{i.e.}{u}_i\ge -w_i^{\prime }\gamma ,\\ 0,\text{if}{z}_i^{\ast }<0,\text{i.e.}{u}_i<-w_i^{\prime }\gamma .\end{array}$ (4)

Result for the probit model:

$y_i=\{\begin{array}{c}{y}_i^{\text{probit}},\text{if}{z}_i=1\\ \text{unobserved, if}{z}_i=0\end{array}i=(1,\dots ,n),$ (5)

where y_i represents the probability of migration, adjusted for the probability of employment, as estimated by the probit model with Heckman correction.

Result for the Mincer equation:

$y_i=\{\begin{array}{c}{y}_i^{\text{Mincer}},\text{if}{z}_i=1\\ \text{unobserved, if}{z}_i=0\end{array}i=(1,\dots ,n),$ (6)

where y_i denotes the natural logarithm of the average monthly wage, adjusted for the probability of employment, as estimated in the Mincer model with Heckman correction.

The error random terms are defined as:

$({\epsilon }_i,u_i)\sim \left(\left[\begin{array}{l}0\\ 0\end{array}\right]\left[\begin{array}{cc}{\sigma }^2& \rho \sigma \\ \rho \sigma & 1\end{array}\right]\right).$

Although the MGE database is comprehensive, it has several limitations. First, the sample includes only officially employed graduates. If a graduate has not secured official employment after graduation, no information is available regarding their subsequent labour market outcomes. Second, the MGE contains limited information on individual characteristics and abilities and lacks data on socio-economic factors entirely. Yet, individual and family characteristics can explain a substantial portion of migration decisions as well as the returns from migration.

To address this, we use proxies for abilities, including the attainment of an honours degree and high university admission scores. We hypothesize that more capable graduates are less likely to engage in labour migration, either because they have already relocated to regions with higher-quality universities during educational migration or because they possess a competitive advantage in the local labour market. Additionally, we consider selection prior to university, reflected in the possession of a school certificate with honours.

First, school achievements serve as early indicators of ability, while a university diploma with honours reflects post-graduation competence. Second, we hypothesize that graduates with honours certificates who did not migrate during educational migration are less likely to migrate during the labour market stage. A school certificate with honours thus acts as a proxy for abilities in the transition from secondary to higher education. High-performing individuals are more likely to obtain quality higher education and acquire specific skills demanded in the labour market. Such graduates enjoy competitive advantages locally and, ceteris paribus, are less likely to relocate, particularly if they already have employment and established social networks. Consequently, this variable is included in the selection equation of the probit model as an instrumental variable to correct for selection bias before the school-to-work transition. So, in the selection equation of the Heckman probit model, we include three variables: disability, participation in targeted training, and school performance.

It is also important to note that we do not focus on the duration of an individual’s stay in the destination region; therefore, migration is not classified as short- or long-term. Additionally, this paper places less emphasis on the effects of regional characteristics.

The main factors significantly influencing migration decisions in our study are related to educational and job characteristics (Table 1). First, we hypothesize that graduates enrolled in paid-basis programs are more likely to migrate, as they may seek to recoup their educational investments in more economically developed regions. Second, we expect that holding a diploma with honours and participation in continuing education will reduce the probability of labour migration. Third, working while studying may slow migration and bind graduates to their region of study due to established labour market networks.

Fourth, higher university admission scores are likely associated with a lower probability of migration. A higher score indicates better quality education and a more economically developed region. In such regions, high labour demand can compress wages from below. For graduates with lower abilities, this may translate into higher potential returns from migration, thereby encouraging relocation. Finally, we anticipate that gender will have either a weak or negligible effect on migration probability since both female and male graduates are in the early stages of their careers.

According to the MGE data, the total share of employed graduates in 2021 who migrated to work in another region amounted to 31.8%. Graduates relocate both to economically developed regions near their place of education and to major urban agglomerations, with Moscow emerging as the primary destination. The by-region matrix (Table A2, Appendix 1) shows that, on average, almost 13% of graduates from all federal districts leave their educational regions for Moscow, with the largest number relocating from nearby areas, including the Moscow Oblast. Beyond the central part of Russia, graduates also tend to migrate to regions close to their region of study.

High salaries in the Far Eastern and Ural regions are associated with a lower rate of graduate labour migration. Economic activity in these regions is largely based on natural resource extraction, providing employees with a competitive wage advantage. In contrast, economic specialization in the Caucasian and Southern regions, primarily in agriculture, encourages graduates to migrate to more technically developed regions. Consequently, wages in “outsider” regions are almost six times lower than in leading regions: North Ossetia – 23.100 rubles; the average across all regions – 50.200 rubles (Table A1, Appendix 1); Chukotka – 126.500 rubles).

Wages, along with certain educational and labour characteristics, are correlated with the probability of migration, particularly when considering educational specializations (Table 2). The largest share of migrants is observed among graduates in agricultural fields, who also earn the lowest wages. At the same time, they are the least likely to enrol in paid-basis programs, have the lowest average admission exam scores and perform near the average on other indicators.

Graduates in social sciences, by contrast, are the most likely to pursue education on a paid basis but also receive below-average wages. Notably, they are the only group that experiences a wage loss upon migration (-1.8%).

Migration patterns among STEM, medical, and humanities graduates are generally comparable to the average for all specializations. Engineers, mathematicians, and specialists in natural sciences tend to earn relatively high salaries and often relocate to regions with technological hubs that provide specialized educational and employment infrastructure (Emelina et al. 2022). Medical graduates are less likely to obtain honours degrees and, in nearly 40% of cases, pay for their education. They benefit from both relatively high salaries and some of the largest gains from migration.

Humanities graduates lead in many educational indicators but perform poorly in most employment-related measures. Nearly 30% hold diplomas with honours, and they have the highest university admission scores. Nevertheless, they are employed less frequently than graduates in other fields, and their average wages reach only 81% of the overall average for all other specializations, both in their region of study and after relocation.

Graduates in education represent a distinct group of specialists, characterized by a consistently high employment rate (78.7%) but relatively low wages. This “stability” is largely associated with the historically low funding for the education sector in Russia (Emelina et al. 2022). Education graduates also include a high proportion of top-performing individuals, as well as graduates who were employed during their studies. Many of them pursue continuing education (34%), which further contributes to a lower likelihood of migration.

Thus, we observe potential differences both in migration decisions and, in particular, in the outcomes of migration among graduates of different specializations. Consequently, we anticipate disparities in both the probability of migration and the returns from migration across the six groups of specializations. Specialization is included in the Heckman-corrected probit model as a categorical variable, while Mincer equations are estimated for each specialty separately.