Seminar ‘Occupational Propensity of Qualification Improvement in a Late-industrial Economy: Evidence from Russia"
On May 19 a joint research seminar of the Laboratory for Labour Market Studies and the Centre for Labour Market Studies was held at HSE. Vasiliy Anikin, Assistant Professor at the Department of Applied Economics delivered the report ‘Occupational Propensity of Qualification Improvement in a Late-industrial Economy: Evidence from Russia’.
On May 19 a joint research seminar of the Laboratory for Labour Market Studies and the Centre for Labour Market Studies was held at HSE. Vasiliy Anikin, Assistant Professor at the Department of Applied Economics delivered the report ‘Occupational Propensity of Qualification Improvement in a Late-industrial Economy: Evidence from Russia’.
Abstract:
It is taught that in a late industrial economy qualification improvement should be considered an occupational process. Drawing on the representative sample of the Russia Longitudinal Monitoring Survey (RLMS-HSE, 2012), I postulate hierarchical nature of training probability studied as a binomial distribution with logit link function. It is shown that consideration for the structural context of occupations and wage-diversity within them is crucial to the explanation of the variation in training, which is normally ignored by researchers. Applying two-level modelling with cross-class interactions gives more specific information about occupational propensity of training in Russia, as compared to a standard random-intercept model: there is a highly scattered variation in probability of training within skilled non-manual workers (managers, professionals, and semi-professionals) and low-spread and almost zero variation – within so-called generic labour (clerks, sales workers, and manual labour). Multiple interactions between occupational classes, gender, area of living, even skills used at working place result in increasing the explanatory power of the model. Neither gender, nor habitual area individually has significant effect on training but a combination of the two with occupational class has a considerable effect. This property of Russian labour market is understudied in literature (not only in Russia). In order to estimate the two-level model with multiple cross-class interactions, I applied Markov chain Monte Carlo (MCMC) estimation using Metropolis Hastings sampler (univariate – for both fixed and random effects) and hierarchical centring at level 2, which led to the more accurate and precision weighted estimates, rather than I got from Maximum Likelihood. Moreover, using MCMC resulted in publishing one-peak conditional posterior distributions of the intercept and main predictors close to Normal. Neither iterative generalized least squares (IGLS), nor restricted IGLS (RIGLS) methods were useful to obtain any sufficient estimation from the full model including a sequence of predictors, as well as to estimate fully random model on the small samples.