• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site

Working paper

Проблемы измерения специального стажа на данных РМЭЗ-ВШЭ

In this study, we analyze some pitfalls of measuring employment tenure in the RLMS – HSE data, focusing on the inconsistency of responses to the “when you started your current job” question in two consecutive rounds of the survey. If job start dates differ between two interviews, this signals either that the job has been changed or that at least one of those two dates is incorrect. We determine conditions when the inconsistency between dates means measurement errors in tenure and analyze the incidence and magnitude of these errors as well as their correlation with individual characteristics and impact on regression analysis. Our fi ndings may be summarized as follows. Firstly, the prevalence of such errors in the RLMS – HSE data is high: in the period 1995–2014 about a quarter of all tenure observations were contaminated. Secondly, the magnitude of these errors is substantial. The mean absolute error (a difference between job start dates in two consecutive surveys) was about 3 years, while about a quarter of all errors exceeded 5 years. The mean relative error exceeded 50% of individual’s tenure length. Thirdly, both the probability and magnitude of measurement errors in tenure are not random and correlated with some individuals’ characteristics including wages. This suggests that the standard mincer-type wage equation underestimates returns to tenure in the RLMS – HSE data. Finally, we provide examples of how measurement errors in tenure may affect results of descriptive and econometric analysis. We show that in 1995-2014 due to these errors the average tenure varied in fairy wide limits, which challenges any conclusions about the rise or fall in the average tenure in Russia built on RLMS – HSE data. We also show that panel data estimates of returns to tenure are very sensitive to measurement errors as these errors affect variation in tenure length within individuals more than that variation between individuals. All in all, our results suggest that researchers using RLMS-HSE data should work with tenure data with great care and try to take into account measurement errors in their analyses.