W O R L D B A N K E D U C AT I O N G L O B A L P R A C T I C E Ru s s i a n Fe d e r a t i o n : A n a l y t i c a l S e r v i c e s a n d A d v i s o r y Ac t i v i t y : P 1 7 0 9 7 8 Returns to Education in the Russian Federation: Variation across regions and implications for policy development in priority regions Ekaterina Melianova* |Suhas Parandekar* |Artëm Volgin* * Education Global Practice, Europe and Central Asia June 9, 2020 Correspondence This working paper is the third in a series of working papers Email: sparandekar@worldbank.org investigating the returns to education in the Russian Feder- Data and Code ation. This paper uses regionally representative household Thanks are due to Rosstat for making the survey data to determine the rates of return to education in anonymized Statistical Survey of Income and Participation in Social Programs different regions. Returns show a wide dispersion together micro-data readily available for researchers with the labor market context. The paper’s policy recom- around the world. The code used for this paper is made freely available for all mendations would be particularly helpful to support human researchers at https://bitbucket.org/ capital development of federally targeted economically and zagamog/edreru/src/master/ socially depressed regions. KEYWORDS Returns to Education, Russian Federation, Regional Analysis JEL Codes: I26, I28, J240, R110 June 9, 2020 1 | ESTIMATING REGIONAL RETURNS TO EDUCATION 1.1 | Motivation for this study The diversity of economic conditions across Russian regions suggests fruitful policy analytical use of regional level returns to education. Regional economic development in the Russian Federation is a heavily studied topic, with nu- Acknowledgments: Country Director: Renaud Seligman; Regional Director: Fadia Saadah; Practice Manager: Harry Patrinos; Program Leader: Dorota Nowak; Peer Reviewers: Cristian Aedo; Ruslan Yemtsov; Husein Abdul-Hamid; Team members: Polina Zavalina; Zhanna Terlyga. Thanks to seminar participants at the World Bank Moscow office on Jan. 29, 2020 for useful feedback. Any errors are a responsibility of the authors. 1 2 P170978: WP03 - Variation across Regions merous studies focused on macroeconomic issues and investigations regarding convergence of growth trajectories, decomposition of inequality and efficiency of public spending. Examples of these studies are: Lugovoy et al. 2007, Hauner 2008, Gluschenko 2011 and Kufenko 2014. A recent World Bank report described the three main factors that explain the wide scale of diversity in Russia’s regions, so that some regions have income levels that match Singapore or New Zealand, and others match Bolivia or Honduras: (i) the persistent Soviet legacy; (ii) diverse physical geogra- phy; and (iii) dominance of oil and gas in some regions (World Bank 2018). The report analyzed the determinants of the Economic Potential Index (EPI) of Russian regions: urbanization; the presence of high-tech industries; advanced human capital; and connectivity (access to markets). These four factors explain 60% of the variation in EPI. In this study we create a typology of regions using various measures for the quantity and quality of labor demand and supply, including a measure related to the EPI. For the EPI analysis, the measure of advanced human capital was the regional percentage of population with a higher education degree. While that report examined regional development with an overview of all sectors, and recommended that regional development can be spurred through investment in human capital, this paper seeks to derive deeper insights regarding human capital. It seeks to answer three questions: What is the variation of the returns to education across regions in Russia? What are the regional variables that may be causing the regional variation (as determined through a random effects regression model)? and What are the policy implications of this regional variation? After concluding this introductory section with a review of available regional estimates of the returns to education in the Russian Federation, we present our own estimates of the regional returns to education. We compute regional returns to education as a combination of a fixed coefficient and random coefficients, using the levels of education. The returns can also be termed as the wage premium to the respective levels of education. The final section of the paper presents the returns to education in context of regional conditions related to the labor market supply and demand. In light of the government strategy to target depressed regions, we suggest that human capital development may benefit from an examination of the differential returns to education by region. 1.2 | Previous estimates of regional returns for Russia Until quite recently, the only tried and tested set of available survey data that contained adequate information to calculate the rate of returns to education was the Russian Longitudinal Monitoring Survey (RLMS), implemented by the Higher School of Economics (HSE). The RLMS is a nationally representative household survey, but the survey size and design is too small to include regionally representative samples. Cheidvasser and Benítez-Silva 2007 had used the RLMS to derive rates of return at a level that roughly corresponded to Russia’s eight federal districts. The authors had examine data from the 1995 to 1998 rounds of the RLMS. In this period of time, of substantive economic and social upheaval following the collapse of the Soviet Union in 1991, the returns of the education were low overall, and they were relatively even lower for metropolitan Moscow and St. Petersburg. Baeva 2013 examined returns to education for regions in the Siberian Federal district. Using data from the en- terprise based Survey of Wages by Occupation by Rosstat for the years 2007, 2009 and 2011, she found that the premium to Higher education was 61% for the Russian Federation and 56% for the Siberian Federal District. At the regional level, the premium ranged from 40% for Krasnoyarsk to 72% for Novosibirsk. The author also presents de- tails about considerable variation in the returns to vocational education and a closer examination of returns for the Irkutsk region. Oshchepkov 2018 also utilized data from the Survey of Wages by Occupation by Rosstat, for the years 2005, 2007, 2009, 2011, 2013 and 2015. Only returns to Higher education are computed in this paper, and a typical specifications results in estimates of a wage premium for Higher education for all of the Russian Federation as 81%. P170978: WP03 - Variation across Regions 3 The dispersion indicates a range from 54% return for the Republic of Mordovia to 127% for the Tuva Republic. A very useful practice in this paper is the correct interpretation of coefficients on dummy variables in semi-logarithmic regressions that was recommended by Halvorsen, Palmquist, et al. 1980. The author presents the regional estimates of returns to education using ordinary least squares (OLS) regression, with a modified Mincerian specification that includes gender, public or private sector and broad classification of industry. An interesting aspect of Oshchepkov 2018 is the use of data from all five rounds of the occupational wage survey for 79 of the Russian regions, that results in (79 x 6) or 474 coefficient estimates from which wage premium style returns (i.e., not dividing by the years of higher education) can be computed. The author reports a second stage re- gression, using the computed coefficient estimates as dependent variables and regressing them on a set of region level variables, with a specification that includes fixed effects for each region and each year. If there are unobserved re- gional or temporal fixed effects that are correlated with the error term in this second stage regression, the specification is said to result in valid estimates of effects of regional characteristics. Treating regression coefficients as dependent variables could be perilous if there is a systematic time-varying relationship between regional returns to education and the regional characteristics. From a policy analytic perspective, it is of particular interest to trace the time- and region- varying effects as policy makers can use such effects to proactively influence the returns to education. In spite of the possible methodological issues, the paper provides an interesting perspective to the topic of returns to education in the Russian Federation. The literature in this field is likely to grow as more regionally representative household or enterprise data sets become available for the Russian Federation. 1.3 | Data To estimate returns to education in Russian regions, we use the most recent (2018) round of the Statistical Survey of Income and Participation in Social Programs, collected by Rosstat. The primary purpose of the Rosstat survey was to obtain statistical information, reflecting the role of wages, income from self-employment, property income, pensions, and social benefits in ensuring the material well-being of families. The survey contains data on trends in income and poverty variation among households with different socio-economic status. There are also variables on people’s participation in social programs, their pension and health insurance, material and social security of low-income families, and the impact of social policy measures on people’s well-being. The sample selected for the empirical modeling consists of individuals aged 25-64 who are out of school and have positive labor market experience and income. 1.4 | Methods The Mincerian equation with an added gender dummy is the main focus in the regional investigation of returns to education in Russia: in this section we look at how these returns vary across regions. Additionally, we explore the determinants of the established variation through a random effects regression analysis. The equations of interest are as follows: First level: Log (Wage)ij = b0j + b1j ·Educ+b2j ·Exp+b3j ·Exp2 + b4j ·Gender+ ij (1) Second Level: b0j = γ00 + γ0n · Z + u00 ; b1j = γ10 + γ1n · Z + u10 ; bij = γi0 f or i 0 (2) 4 P170978: WP03 - Variation across Regions where an individual i is nested within a region j , Log (Wage) is the logarithm of monthly wage, Educ stands for highest attained level of education, Exp and Exp2 reflect the years of working experience and its quadratic term respectively, Gender is a dummy variable for gender, Z is an n × i matrix of regional characteristics, and u00 , u10 are the first- and second-level errors respectively. The random effects models were estimated using restricted maximum likelihood (REML). Individual Wald tests and likelihood ratio tests were exploited to evaluate the significance of fixed and random effects, respectively. Weights were used in the modeling to ensure the representativeness of the sample across Russian regions (the weighting variable was divided by 1000 to allow the convergence of the multilevel models). 1.4.1 | Left Hand Side (LHS) variable The outcome to be investigated is the logarithm of monthly monetary remuneration before income tax payment at the main place of work. 1.4.2 | Right Hand Side (RHS) variables Education, experience, and gender are the first-level variables as in an OLS equation. We then computed the intra- class correlation coefficient (ICC) on a base model of the logarithm of earnings to examine the percentage variance of earnings explained due to variation across regions. In the base model with covariates, we find an ICC value of 0.20, which is high enough to justify modeling regional random effects. We then compare the base model with a model including Education as a random regional effect, and used Wald tests, likelihood ratio tests and other information tests (AIC, BIC) to determine which model provides a better fit. These criteria point to the inclusion of Education as a random regional effect in addition to the fixed effect of Education. Next we tested a set of fixed regional effects. We checked for the influence of regional level educational quantity and educational quality measures to explain the variation in education payoffs across Russian regions, and also included a set of variables to represent labor market conditions. To measure educational quantity or access, we used the number of students enrolled in vocational education per 10,000 residents (voc_edc) and the number of students enrolled in higher education per 10,000 residents (high_edc). As a measure of educational quality, standard deviations from the national mean of the Russian school-leaving and university entrance examination, the EGE, were incorporated. We also added variables regarding economic development and the labor market - these are the gross regional product, the level of urbanization, the regional unemployment level, the share of employment in jobs related to natural resources exploitation and the ratio of recent graduates who migrated to other states compared to the graduates who stayed in the same region. Appendix Figure A1 shows descriptive statistics of the variables used - the univariate distribution of each variable, and their respective bivariate correlations. For improved context, the matrix represented in A1 also includes regional aggregates for the main variables of interest - education (in years) and logarithm of monthly wage. The figure indicates a rich and varied pattern of correlations - some of these are straightforward - such as the relationship between wages and regional product (grp). The sparklines and bi-variate scatter plots in A1 also indicate the presence of a number of outliers for almost every variable. In a regional context, random effects regression deals effectively with such a data structure. All region-level variables were normalized with Z-standardization before being plugged into the analysis to obtain meaningfully interpretable moderation effects in cross-level interaction models. For the statistically significant interactions, marginal returns to schooling, conditioned on thresholds of region-level characteristics (-1, 0, 1 standard deviations), were evaluated: P170978: WP03 - Variation across Regions 5 {b1j |Z = 1} = γ10 + 1 × γ1n {b1j |Z = 0} = γ10 {b1j |Z = −1} = γ10 − 1 × γ1n (3) Appendix Table A1 demonstrates descriptive statistics of the key variables of interest by regions. Appendix Figures A2 to A6 present maps of the basic Mincerian specification for each region, using the same color code so as to depict the transition over the years. The figures show the declining returns over the years. 1.5 | Estimation Results of Regional Analysis Of the eight variables tested for regional effects, it turned out that six of the eight variables passed the test - the only variables that did not meet the criteria was the migration ratio and the standardized EGE score variable. After adding these six regional fixed effects to the specification, the next step was to check for interactions of the second level variables with education levels. The investigation revealed that with one exception, none of the second-level characteristics have a statistically significant interaction with education as a random effect. The only variable that had an independent random effect at the regional level as well as a statistically significant regional interaction with education was voc_edc, the regional coverage of vocational education. Substantively, it was found that growth in the number of students covered by vocational programs leads to higher schooling premiums concerning both vocational and university education. However, the independent second level effect is negative and four time larger in magnitude, so the finding about the interaction effect does not seem to be significant from a policy analytical viewpoint. The results from the random effects regression and the mean values of the random effects are presented in Appendix Table A2. The addition of the fixed effects for education together with the random effects described in Appendix Table A2 leads to an estimation of the marginal effect for education for each region. We utilize the correction for dummy variables as recommended by Halvorsen, Palmquist, et al. 1980. The results are presented in Figure 1.1. The error bars represent 95% confidence intervals around the estimates. 1.6 | Limitations of the Analysis It is important to recognize a number of limitations of this study that can be tackled by researchers in the future. First, the estimates presented here do not take account of the considerable migration of workers that takes places within the Russian Federation. The study examines wages of individuals currently living in specific regions and attributes the education of these individuals to the same region as a simplifying assumption. Second, geographically contiguous regions and regions connected by transport pathways experience the phenomenon that people may live in one region and work in another region; the successive rings around Moscow city is a classic example of this phenomenon. In spite of these limitations, the next sections of the paper attempt to show that the analysis of returns to education at a regional level do provide useful policy insights. 2 | CATEGORIZATION OF PRIORITY REGIONS The Presidential Executive Order on National Goals and Strategic Objectives of the Russian Federation (2018-2024) defined in December 2018 a set of 13 National Projects and 9 National Development Goals with a budget of nearly 6 P170978: WP03 - Variation across Regions (a) Higher Education Figure 1.1 (b) Vocational Education P170978: WP03 - Variation across Regions 7 26 trillion rubles for a six-year period. This substantive amount is the equivalent of 17% of GDP every year. The national goals include cutting poverty by half by 2024, to improve housing conditions for 5 million people annually and to improve life expectancy. Given Russia’s size and uneven geographic and economic conditions, the success of the strategic goal depends on the implementation performance at the regional and municipal levels. A sub-national focus will enhance the probability of success of the three pillars of the country’s development strategy: growth, the environment and human capital. The Federal Government identified ten poor regions as strategic priorities in Russia. These are the lowest rank- ing regions according to indicators of regional income, poverty levels, unemployment rates and investment climate: Adygea, Republic (Maykop), Pskov Oblast, Altai Krai (Barnaul), Kurgan Oblast, Kalmykia, Republic, Chuvashia, Repub- lic, Altai, Republic, Karelia, Republic, Tyva, Republic, and Mary El, Republic. The Federal Government is working on a strategy for inclusive growth and job creation in these regions. As Human Capital is expected to be an important element of the development strategy for these regions, it will be useful to examine the variation in the rates of return to education in these ten regions. Accordingly, in Figure 1.1, the names of nine of the ten regions for which data was available are highlighted in red color. It should be recalled that these returns are not simply the OLS returns, but are calculated after aggregating the fixed and random effects taking account of regional characteristics and hence are expected to be more accurate than OLS results. The 95% confidence intervals are also presented in the figure. The priority regions are dispersed across the distribution of the rates of return to education both for vocational and higher education. Premiums to education range from 10.1 % (Karelia Republic) to 38.2% (Altai Republic) for university level and from 10.4% to 20.6% for vocational level for the same two regions. The returns for vocational and higher education are roughly moving in step, with the exception of higher returns for higher education for Kurgansk Oblast and the Tuva Republic. 2.1 | Quantity and Quality of Skilled Labor Supply In order to better place policy recommendations for regions in context of their particular situation, we devised an algorithm or heuristic to classify regions according to certain variables of interest. We identified a set of variables that capture the quantity and quality of skilled labor supply and the quantity and quality of skilled labor demand. For skilled labor supply quantity, we utilized the proportion of the labor force with a higher education degree and for skilled labor supply quality we utilized the mean university entrance exam (EGE) score for the region. Both of these are proxy variables for underlying constructs. In order to have a reasonable comparison across dimensions, the variables were standardized. In the case of the EGE score, we standardized the score to 500 for the mean for all of the Russian Federation and 100 standard deviation. For all other variables we use a mean 0 and standard deviation of 1. The plot of regions according to the two dimensions of labor supply quantity and quality is presented in Panel (a) of Figure 2.1. Four regions are outliers and are not seen in the graph - St. Petersburg and Moscow in Quadrant I and Ingushetiya Republic and Karachayevo-Cherkessiya in Quadrant IV. The graph also presents the numbers of regions in each of the quadrants. Quadrant membership, or tags from quadrants is the central piece of our classification of regions. 2.2 | Quantity and Quality of Labor Demand To match the classification of regions by quantity and quality of labor supply, we also carry out a similar classification for labor demand. For the quantity dimension of labor demand, we use the total share of set of specific industries in the regional GRP from Rosstat (latest available figures). We include the industries that are likely to contribute most in terms of labor force demand, excluding the oil and gas industry and excluding the mostly public sector education 8 P170978: WP03 - Variation across Regions and health sectors. The objective is to arrive at a qualitative grouping of regions, but future research can also test sensitivity of the classification to alternative choices of sectors. The sectors chosen for this purpose were: agriculture, hunting, forestry, fishery and fish breeding, manufacturing, wholesale, retail trade and repair services, hotels and restaurants, transport and communications. The percentage contribution to GDP for these sectors by region ranged from 35% (Tuva Republic) to 81% (Khanty-Mansisk). As a measure of quality of labor demand we utilize an indicator of product complexity computed by Lyubimov, Gvozdeva, and Lysyuk 2018. This paper is based on a methodology that was initially proposed and implemented by the economists Ricardo Hausmann and Céesar Hidalgo to capture the productive potential of an economy on the basis of the diversity of its products and exports (Hausmann and Hidalgo 2011; Hausmann et al. 2014). Lyubimov, Gvozdeva, and Lysyuk 2018 develop an “Economic Complexity Index” (ECI) utilizing production as well as export data. It is possible to explain intuitively the conceptualization of the complexity index on the basis of product diversity and the export basket. When we compare less developed economies with more developed ones, we see that more developed economies are able to manufacture a more diverse range of products because they have stronger production networks. Also, given the competitive international marketplace, the quality of products can be gauged by the prevalence of that product in the mix of traded goods. This method takes care of two problems - if a country has high exports of commodities, example from natural resource extraction, it does not score high on diversity; and if a country does manufacture a diverse range of goods, but these are not internationally competitive, it would also get a low score. Lyubimov, Gvozdeva, and Lysyuk 2018 extend the logic to regional measures of complexity. As human capital quality is closely linked to the complexity of products, the ECI is a very useful variable for purposes of classification of regions. The position of regions along the two standardized dimensions is shown in Panel (b) of Figure 2.1. (a) Labor Supply (b) Labor Demand FIGURE 2.1 Ranking of Regions on Quantity and Quality dimensions P170978: WP03 - Variation across Regions 9 2.3 | Bringing Demand and Supply classification together The purpose of classifying regions according to proximate measures of labor demand and labor supply is to situate the variation in regional returns to education in context. We seek to combine the quadrant classification displayed in Figure 2.1 with the pattern regarding returns to education. In order to do so, we compare a region’s position in the demand panel on the left hand side and the supply panel on the right hand side. If a region is better placed on the demand dimension than it is with regard to the supply dimension, we term it as demand dominated; and vice versa. With four quadrants for each of the classifications, there are 4 times 4 or 16 categories that need to be simplified into 2 groups (supply or demand dominated). The decision is straightforward when a region is high on both quality and quantity of demand parameters (Quadrant I in Panel (a)) or low on both quality and quantity of supply parameters (Quadrant IV in Panel (b)). In case of ties, for 28 of the 80 regions with available data, we use the quality dimension to break ties. We also generate a two-fold classification of the returns to education, using the classification of regions above and below the median for both returns to higher education and returns to vocational education for each region, presented in Figure A1. When reducing from four dimensions to two, we use the returns to higher education to break ties. The result of this heuristic is a combined table that examines the returns to education in the context of labor supply or demand dominance. The classification is presented in 2.2 for the 80 regions for which data was available, with the priority regions highlighted using red color for the region names. Even though the priority regions are economically disadvantaged, it is very useful to note how they are spread across the four cell of Figure 2.2. Policy analysis to aid development of strategies for the regions will benefit from the kind of analysis presented in this paper and even more fine-tuned analysis in the future for devising policies for specific regions. 3 | POLICY RECOMMENDATIONS FOR PRIORITY REGIONS Returns to education tend to fall with level of economic development when comparing across countries (Psacharopou- los and Patrinos 2018). When examining the case of differential returns within the Russian Federation, we do find that St. Petersburg and Moscow city figure in the ranks of low returns. However, as studied by Lyubimov, Gvozdeva, and Lysyuk 2018, the more well-off regions in the Russian Federation as well as the no so well-off regions are diverse in the make-up of their productive networks. We attempt to exploit this diversity to come up with tailored policy rec- ommendations for regions. These are preliminary and demonstrative recommendations for groups of regions. Further analysis would need to be carried out for a specific region as the grouping used here is quite wide. For sake of brevity the analysis presented here combines the findings regarding returns to higher education and returns to vocational education, but it would be beneficial to separate them for a more granular view. The Table 3.1 provides an indicative list of policies that would be useful on the basis of an examination of the returns to education and the context of a region. Higher returns in general indicate the scope for greater investment in the supply or quantity of education as more people would be attracted to obtain higher levels of education. Lower returns to education indicate a scope for increased investment in the quality of education provision, and making better industry-education connections in terms of skills provided. When labor supply conditions are relatively good and labor demand conditions are lagging from other regions, it is an indication towards job creation policies, through innovation and entrepreneurship. When labor demand conditions are dominant, it would be an indication for better matching between jobs and skills, innovation to enhance labor productivity and diversify educational offerings. Other things constant, one would expect returns to be high when labor demand conditions are dominant and competition between employers drive up wages. However, as other things change with regional diversity, we find cases where labor supply 10 P170978: WP03 - Variation across Regions FIGURE 2.2 Variation of Education Returns and Regional Labor Market Context P170978: WP03 - Variation across Regions 11 is dominant at the same time as returns are high. With better availability of data at the regional level in the future, it would be feasible to come up with better targeted policy decision making. TA B L E 3 . 1 Policies fitting Regional Context High Returns to Education Low Returns to Education Labor Demand • Improved career guidance for high school • Policies to improve quality of professional col- dominates graduates leges, higher investment in World Skills Labor Supply • Policies to encourage deeper teacher profes- • Deepen supply of extra-curricular activities sional development in general and university for better soft-skills education • Investments in general education and poli- • Investments and policies on the industrial cies to improve quality of provision of general side private sector firm formation; diversifi- education so students come out with skills cation or cluster specialization etc. needed by the market Labor Supply • Policies to develop entrepreneurship and en- • Policies to integrate industries to become dominates courage job creation, including innovation part of global value chains, support specific Labor Demand policies industry clusters • Policies to develop problem solving skills • Policies for dissemination and connectivity and financial literacy, including strengthening of educational systems like university consor- extra-curricular education tiums • Investments in university quality, e.g. interna- • Investments in industrial development, iden- tionalization of universities tification of economic activities for which re- gion may have comparative advantage 12 P170978: WP03 - Variation across Regions References Baeva, Olga Nikolaevna. 2013. “Assessment of the return on education at the regional level (Otsenka otdachi ot obrazovaniya na urovne regiona)”. Manager (Upravlenets) 2 (42): 64–69. Cheidvasser, Sofia, and Hugo Benítez-Silva. 2007. “006. The Educated Russian’s Curse: Returns to Education in the Russian Federation during the 1990s”. Labour 21 (1): 1–41. Gluschenko, Konstantin P. 2011. “Studies on income inequality among Russian regions”. Regional Research of Russia 1 (4): 319–330. Halvorsen, Robert, Raymond Palmquist, et al. 1980. “The interpretation of dummy variables in semilogarithmic equa- tions”. American Economic Review 70 (3): 474–475. Hauner, David. 2008. “Explaining differences in public sector efficiency: evidence from Russia’s regions”. World Devel- opment 36 (10): 1745–1765. Hausmann, Ricardo, and César A Hidalgo. 2011. “The network structure of economic output”. Journal of Economic Growth 16 (4): 309–342. Hausmann, Ricardo, et al. 2014. The atlas of economic complexity: Mapping paths to prosperity. Mit Press. Kufenko, Vadim. 2014. Economic growth and inequality: Empirical analysis for the Russian regions. Springer. Lugovoy, O, et al. 2007. “Economic, geographical, and institutional aspects of regional economic growth in Russia”. Tech-nical Report, Institute for the Economy in Transition, Moscow (in Russian). Lyubimov, Vanya, Margarita Gvozdeva, and Maria Lysyuk. 2018. “Towards increased complexity in Russian regions: networks, diversification and growth”. Oshchepkov, Aleksey. 2018. “What Drives Returns to Higher Education: Evidence from Russian Regions”. Center for Labour Market Studies, Higher School of Economics, Moscow. Psacharopoulos, George, and Harry Anthony Patrinos. 2018. “Returns to Investment in Education: A Decennial Review of the Global Literature”. Education Economics 26 (5): 445–458. World Bank. 2018. Rolling Back Russia’s Spatial Disparities. World Bank. doi:10 . 1596 / 29866. https : / / elibrary . worldbank.org/doi/abs/10.1596/29866. P170978: WP03 - Variation across Regions 13 Appendix TA B L E A 1 Descriptive Statistics for Regions in Russia, Rosstat 2018 Wage Experience Education, % Gender, % Regions N mean sd mean sd SE VE HE Males Females Altayskiy Kray 4646 22127.6 11952.2 23.6 11.0 17.456 54.50 28.05 48.90 51.10 Amurskaya Oblast 2557 33441.2 17409.0 23.2 11.2 16.347 50.65 33.01 49.59 50.41 Arkhangelskaya Oblast 3183 33438.1 16884.2 22.6 10.6 12.692 54.95 32.36 44.17 55.83 Astrakhanskaya Oblast 2836 26474.1 13737.6 23.0 11.3 13.646 55.08 31.28 50.99 49.01 Belgorodskaya Oblast 3692 26281.0 10811.9 23.8 11.1 12.351 54.47 33.18 49.76 50.24 Bryanskaya Oblast 3087 22482.3 9634.1 23.5 10.9 19.631 50.66 29.71 48.66 51.34 Chechenskaya Respublika 2010 27718.4 11793.2 18.7 10.6 25.721 26.37 47.91 65.37 34.63 Chelyabinskaya Oblast 6717 27990.8 14280.9 23.9 11.2 12.104 54.53 33.36 47.39 52.61 Chukotskiy Aok 1535 65574.1 32370.8 23.6 10.6 13.941 46.06 40.00 43.97 56.03 Chuvashskaya Respublika 3248 21453.7 12602.2 24.3 11.0 19.119 50.80 30.08 50.18 49.82 Evreyskaya AOb 1536 28532.1 17385.1 23.8 11.2 22.005 50.33 27.67 50.00 50.00 Irkutskaya Oblast 4686 29967.6 17443.1 22.3 11.2 17.520 47.06 35.42 47.57 52.43 Ivanovskaya Oblast 2876 24881.8 12496.8 23.3 10.9 20.341 49.90 29.76 47.77 52.23 Kabardino-Balkarskaya Res. 2006 23592.3 10766.2 21.7 11.6 21.137 40.53 38.33 52.04 47.96 Kaliningradskaya Oblast 2838 29749.2 15489.1 23.5 11.4 13.495 52.40 34.11 50.07 49.93 Kaluzhskaya Oblast 3155 29662.1 12879.5 24.1 11.2 13.312 52.11 34.58 47.92 52.08 Kamchatskaya Kray 2203 51160.5 29997.7 23.1 11.2 13.118 42.99 43.89 47.89 52.11 Karachayevo-Cherkessiya 1510 22900.6 12540.8 22.0 11.8 17.152 40.07 42.78 48.01 51.99 Kemerovskaya Oblast 5056 26287.0 13774.4 23.6 11.3 18.137 52.99 28.88 48.04 51.96 Khabarovskiy Kray 3731 42008.8 21837.8 22.3 11.2 11.900 44.33 43.77 46.15 53.85 Khanty-Mansiyskiy Aok 4335 50837.9 22261.7 22.8 10.5 13.564 46.78 39.65 49.60 50.40 Kirovskaya Oblast 3284 22941.0 13674.6 25.1 11.2 20.128 55.33 24.54 47.69 52.31 Kostromskaya Oblast 2518 23993.1 12090.9 23.6 11.1 12.669 61.28 26.05 47.82 52.18 Krasnodarskiy Kray 8730 32563.7 17499.8 23.0 10.9 15.888 48.57 35.54 50.02 49.98 Krasnoyarskiy Kray 5540 33954.6 21199.2 23.0 11.0 21.588 48.05 30.36 49.64 50.36 Kurganskaya Oblast 2468 20896.9 11539.5 24.4 10.7 21.394 52.47 26.13 48.38 51.62 Kurskaya Oblast 2956 23622.6 11475.0 23.9 11.0 14.783 52.17 33.05 50.30 49.70 Leningradskaya Oblast 4506 32124.3 17227.4 24.2 11.5 7.723 54.77 37.51 46.03 53.97 Lipetskaya Oblast 2869 25037.8 10813.5 24.1 11.0 13.106 53.82 33.08 49.60 50.40 Magadanskaya Oblast 1841 51000.8 23729.4 24.1 11.4 18.523 43.02 38.46 43.24 56.76 Moscow 29921 66263.5 26437.9 20.8 10.8 4.953 32.18 62.86 47.06 52.94 Moskovskaya Oblast 13431 46725.1 20563.7 22.6 11.4 10.975 39.13 49.89 47.51 52.49 Murmanskaya Oblast 3078 43992.5 28841.9 23.4 11.2 12.801 50.45 36.74 49.84 50.16 Nenetskiy Aok 1118 54467.3 23147.1 22.6 10.8 17.263 49.73 33.01 39.98 60.02 Nizhegorodskaya Oblast 6139 30912.9 13291.8 23.4 11.2 16.941 49.31 33.75 47.42 52.58 Novgorodskaya Oblast 2673 26856.0 12683.0 24.6 11.2 15.638 55.74 28.62 45.16 54.84 Novosibirskaya Oblast 5374 29229.9 14687.7 23.9 11.6 16.561 49.33 34.11 47.06 52.94 Omskaya Oblast 3978 25337.5 14613.1 23.6 10.9 22.197 51.31 26.50 51.11 48.89 Orenburgskaya Oblast 4190 24207.0 12519.9 23.3 11.0 15.131 53.68 31.19 51.29 48.71 Orlovskaya Oblast 2424 21901.2 10561.0 24.7 11.1 15.017 50.66 34.32 46.99 53.01 Penzenskaya Oblast 3103 23478.4 10982.9 24.2 11.0 20.722 51.40 27.88 51.02 48.98 Permskiy Krai 5290 29176.6 14449.4 23.4 11.0 13.894 58.32 27.79 48.17 51.83 Primorskiy Kray 4104 37839.9 18420.2 23.8 11.3 14.985 52.97 32.04 49.98 50.02 Pskovskaya Oblast 2382 23838.4 12015.3 25.0 11.0 17.632 55.33 27.04 48.11 51.89 Respublika Adygeya 2013 21350.3 10505.9 23.4 11.3 20.666 43.67 35.67 49.53 50.47 14 P170978: WP03 - Variation across Regions TA B L E A 1 Descriptive Statistics for Regions in Russia, Rosstat 2018 Wage Experience Education, % Gender, % Regions N mean sd mean sd SE VE HE Males Females Respublika Altay 1381 20285.3 12029.5 23.0 10.6 23.027 45.26 31.72 43.08 56.92 Respublika Bashkortostan 7126 31100.8 15175.2 23.4 11.0 12.167 56.67 31.17 51.98 48.02 Respublika Buryatia 2469 29536.3 17237.4 22.1 10.6 17.173 45.61 37.22 48.12 51.88 Respublika Dagestan 3388 26377.3 11971.9 23.0 10.7 30.519 30.79 38.70 55.99 44.01 Respublika Ingushetiya 1207 23740.2 10168.5 18.2 9.6 10.025 18.89 71.09 61.14 38.86 Respublika Kalmykiya 1751 18568.8 11749.1 23.6 11.4 15.762 40.89 43.35 46.43 53.57 Respublika Karelia 2164 28510.2 16639.5 23.7 10.8 17.144 55.45 27.40 47.00 53.00 Respublika Khakasiya 2064 27288.1 16613.3 23.3 11.1 22.045 51.11 26.84 50.97 49.03 Respublika Komi 2972 35891.6 21554.4 23.8 11.0 16.689 53.47 29.85 46.67 53.33 Respublika Mariy El 2486 21133.1 11941.6 24.1 11.2 18.785 52.98 28.24 47.87 52.13 Respublika Mordovia 2236 21221.0 10837.3 23.1 11.2 15.519 49.11 35.38 48.35 51.65 Respublika Saha (Yakutia) 3243 45763.1 25001.6 23.2 11.3 18.440 45.76 35.80 46.69 53.31 Respublika Severnaya Osetiya 2114 22993.1 12762.5 21.8 11.3 12.677 40.92 46.40 48.91 51.09 Respublika Tatarstan 7212 30327.9 12928.8 23.5 11.1 18.691 48.64 32.67 51.48 48.52 Respublika Tyva 1704 23421.9 16851.3 21.4 10.0 19.777 44.78 35.45 40.43 59.57 Rostovskaya Oblast 6985 28287.2 12779.9 23.1 11.0 15.476 48.03 36.49 50.68 49.32 Ryazanskaya Oblast 2609 25889.2 11760.9 24.7 11.1 12.457 59.37 28.17 49.18 50.82 Saint-Petersburg 11352 48520.8 23771.0 22.8 11.4 5.259 38.15 56.59 46.04 53.96 Sakhalinskaya Oblast 2258 50325.1 25563.0 23.6 11.2 17.493 48.23 34.28 46.94 53.06 Samarskaya Oblast 6275 32584.4 15015.6 23.8 11.1 11.331 47.87 40.80 47.71 52.29 Saratovskaya Oblast 4572 23698.6 12322.4 23.7 10.8 14.961 50.22 34.82 50.42 49.58 Smolenskaya Oblast 2726 25517.8 12104.9 24.6 11.3 14.380 52.31 33.31 46.04 53.96 Stavropolskiy Kray 4945 25263.6 12696.7 22.6 11.3 16.946 43.80 39.25 47.48 52.52 Sverdlovskaya Oblast 7712 35983.2 15242.7 23.6 11.3 16.779 54.94 28.28 48.59 51.41 Tambovskaya Oblast 2781 22698.6 10440.1 24.1 11.0 16.397 53.54 30.06 50.67 49.33 Tomskaya Oblast 3074 29580.6 16745.7 22.1 11.1 13.500 47.56 38.94 46.78 53.22 Tulskaya Oblast 3516 27687.4 11814.7 24.3 11.3 17.491 54.69 27.82 48.98 51.02 Tverskaya Oblast 3157 26310.0 15025.1 25.5 11.1 14.824 56.57 28.60 44.73 55.27 Tyumenskaya Oblast 3095 31441.2 17278.6 22.7 11.2 16.123 52.89 30.99 50.05 49.95 Udmurtskaya Respublika 4073 24044.6 11540.9 23.9 11.3 20.108 51.04 28.85 46.99 53.01 Ul’yanovskaya Oblast 3109 23215.3 10596.4 24.8 10.9 19.170 53.84 26.99 50.37 49.63 Vladimirskaya Oblast 3502 25001.4 12605.8 24.5 11.4 19.503 50.77 29.73 46.49 53.51 Volgogradskaya Oblast 4836 24459.0 12915.8 23.2 11.0 15.881 50.91 33.21 49.69 50.31 Vologodskaya Oblast 2965 28248.9 16693.8 23.9 11.2 17.302 57.47 25.23 49.61 50.39 Voronezhskaya Oblast 4348 26261.9 11813.9 23.6 11.5 22.700 43.38 33.92 48.37 51.63 Yamalo-Nenetskiy Aok 3164 69356.7 28075.6 21.0 10.4 10.683 40.27 49.05 48.74 51.26 Yaroslavskaya Oblast 3361 30261.4 14682.8 24.1 11.4 16.215 53.73 30.05 47.01 52.99 Zabaykalskiy Kray 3017 28336.6 16350.4 23.0 10.6 24.561 47.40 28.04 47.07 52.93 P170978: WP03 - Variation across Regions 15 TA B L E A 2 Null model Mincerian Random Slope Cross-Level Interaction (1) (2) (3) (4) Constant 10.178∗∗∗ 10.032∗∗∗ 10.056∗∗∗ 10.065∗∗∗ (0.034) (0.034) (0.036) (0.036) Vocational 0.283∗∗∗ 0.279∗∗∗ 0.267∗∗∗ (0.009) (0.021) (0.021) Higher 0.638∗∗∗ 0.641∗∗∗ 0.622∗∗∗ (0.009) (0.025) (0.025) Coverage VE X Vocational 0.050∗∗ (0.025) Coverage VE X Higher 0.083∗∗∗ (0.030) Experience −0.026∗∗∗ −0.027∗∗∗ −0.027∗∗∗ (0.002) (0.002) (0.002) Experience squared −0.065∗∗∗ −0.065∗∗∗ −0.065∗∗∗ (0.002) (0.002) (0.002) Females −0.403∗∗∗ −0.404∗∗∗ −0.404∗∗∗ (0.005) (0.005) (0.005) Coverage VE −0.101∗∗∗ −0.142∗∗∗ (0.039) (0.043) Variance of Intecept 0.09 0.08 0.09 0.09 Variance of Vocational 0.02 0.02 Variance of Higher 0.04 0.04 Residual Deviance 0.45 0.35 0.34 0.34 sigma 0.67 0.587 0.584 0.584 deviance 119505.212 106528.235 106137.315 106129.127 df.residual 49184 49179 49173 49171 Observations 49,187 49,187 49,187 49,187 Log Likelihood −59,755.060 −53,289.500 −53,094.620 −53,096.640 Akaike Inf. Crit. 119,516.100 106,595.000 106,217.200 106,225.300 Bayesian Inf. Crit. 119,542.500 106,665.400 106,340.500 106,366.100 Note: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01 16 FIGURE A1 Correlations of Regional Level Variables with Wages and Education P170978: WP03 - Variation across Regions P170978: WP03 - Variation across Regions FIGURE A2 Mincerian Returns Basic Specification 2014 17 18 FIGURE A3 Mincerian Returns Basic Specification 2014 P170978: WP03 - Variation across Regions P170978: WP03 - Variation across Regions FIGURE A4 Mincerian Returns Basic Specification 2014 19 20 FIGURE A5 Mincerian Returns Basic Specification 2014 P170978: WP03 - Variation across Regions P170978: WP03 - Variation across Regions FIGURE A6 Mincerian Returns Basic Specification 2014 21