Essays on labor market dynamics using the survey of income and living conditions

This dissertation is composed of two chapters on labor market dynamics based on the Survey of Income and Living Conditions (SILC) conducted by the Turkish Statistical Institute (TURKSTAT), the only publicly available household level panel data in Türkiye. SILC is collected across Europe under the supervision of European Statistical Institute (EUROSTAT). It offers a wide range of information on households and individual members, such as demographic characteristics, labor market status, income, asset ownership and living conditions by following households and household members for up to four years. EUROSTAT also supervises and coordinates a second effort that started earlier, known as Labor Force Surveys. Since the short panel component of the Household Labor Force Survey (HLFS), the main data source for labor market analysis in Türkiye, is not publicly available, T-SILC stands as a good candidate for studying labor market dynamics. Increased availability of panel data has exposed the vulnerability of surveys to non-response. Attrition, non-response after an initial response, and reverse attrition, response after an earlier non-response, are examples of missing data problems that loom large. The potential selectivity in exits and reentry can compromise inferences on dynamics that can be drawn from panel data. In Chapter 1, we examine household attrition patterns in T-SILC. Using data from the 2006-15 period, we document that 15.5 percent of households in the sample frame of T-SILC attritted, and an additional 4.15 percent reverse attritted before completing the rotation plan. TURKSTAT keeps track of non-response and uses weights that depend on age, gender, a narrow region/location designation and household size, to render the balanced panel representative. This approach would work if attrition is random conditional on the observables used in weighting. Yet we find that the likelihood of attrition conditional on observables is influenced by the labor market status of the household head. Households with heads who were unemployed (employed) in the period before attrition are more (less) likely to attrit compared to non-participants. The 19.6 percent combined attrition rate plus our finding that non-response has a statistically significant link with the labor market status of the household head exacerbates concerns that attrition in T-SILC can result in biased inferences while studying labor market dynamics. Other outcomes that can be affected are income inequality, poverty or deprivation status, and even living conditions of households that are intimately connected with the employment status of the household head. In Chapter 2, we investigate the suitability of T-SILC for studying labor market dynamics. Since Chapter 1 documents that the probability of attrition is non-ignorable with respect to the labor market status of the household head, attrition stands as a major threat. We side-track the attrition problem by using fresh subsamples which rely on first and second round information only. Although T-SILC has other shortcomings such as the gap between the reference periods of earnings and detailed labor market information, lack of public/private distinction and tenure information at the current job, it has the potential to shed light on labor market dynamics. Our assessment of the suitability of T-SILC for studying labor market dynamics is based on three connected investigations, which can be seen as the initial steps of separate substantive analyses. Using data from the 2006-17 period, firstly we study the annual transition patterns obtained from T-SILC over a broad set of labor market states: full-time (FT) and part-time (PT) formal and informal wage and salary employment (WS), non-WS employment, unemployment, and non-participation. Use of a rich state space enhances our ability to evaluate the adequacy of T-SILC in reflecting what goes on in the labor market. In light of our finding that non-response is a significant concern, we include attrition (reverse attrition) as a terminal (origin) state in our forward (backward) transition analyses. We show that state dependence is the dominant pattern, except for unemployed individuals. Nearly half of the FT informal WS workers change their status one year later. Annual transition rates to attrition are sizeable, particularly from unemployment. Secondly, we test whether the transition rates show the footprints of economic shocks that Türkiye experienced during the period under investigation: the 2008-09 global crisis and the extraordinary minimum wage hike that took place at the beginning of 2016. We compare our findings with those obtained from HLFS to see how well T-SILC fares. Test results reveal that transition rates estimated on T-SILC are sensitive to both shocks. Our rich state space designation allows us to evaluate the patterns and mechanisms that lead to changes in the transition rates. Lastly, we assess the suitability of T-SILC for studying labor market turnover via commonly used measures, including hiring, job separation, and job-to-job transition rates. Our results are broadly inline with the findings of studies that use HLFS. We conclude that T-SILC is a promising data source for deepening the investigations of the dynamics we explored.