İşgücü piyasası hareketliliği ve Türkiye'de işgücü piyasasına marjinal tutunamayanlar: Türkiye hanehalkı işgücü anketine bir uygulama, 2000-2002

The adequacy of conventional analyses of labor market mobility based on the standard three state model is subject to debate. In this thesis I engage in this debate and propose a five state model. This choice is motivated by evidence from the Turkish Household Labor Force Survey (HLFS) 2000-2002, namely the fact that transition rates from non-participation to employment are considerable, and the numbers involved are higher than those between unemployment and employment. I subdivide non-participants further first by separating students, then by focusing on individuals who are not classified as unemployed because they did not search for a job within the three month window that preceded the survey. I extend the window to a year and call individuals who searched sometime between the three to twelve month period marginally attached. I then subject the five state model (student, marginally attached, unemployed, employed and a residual category I refer to as not attached) to further scrutiny, to determine if the finer distinctions are warranted. In deciding whether particular pairs of states can be pooled, I follow the Likelihood Ratio test methodology of Jones and Riddell (1999). Since the short panel component of HLFS suffers from attrition (initial response followed by non-response) and substitution (initial non-response followed by response), the transition rates obtained from the balanced panel are biased. To adjust them, I use the Rescaled Additively Non-Ignorable (RAN) model of attrition and substitution developed by Tunalı, Ekinci, and Yavuzoğlu (2012). This model is attractive because attrition and substitution probabilities are functions of the labor markets states in both periods, and the adjustments to the balanced panel cells yield marginals which are consistent with the official statistics published by TURKSTAT. Results for the balanced panel for working age (15 and over) individuals reveal that, those who show up as marginally attached, employed, and unemployed in either or both periods are underrepresented, whereas students and not attached ones are overrepresented. I then calculate the adjusted forward transition rates and subject them to equivalency tests. The transition behavior of students is markedly different from the others. Notably the average quarterly transition rates to employment from marginal attachment and unemployment turn out to be statistically indistinguishable. As in Jones and Riddell, I conduct the pooling tests using both Multinomial (MNL) and Binary (BNL) Logit models. MNL results show that marginally attached individuals should be kept as a different labor market state. Additionally, tests based on BNL indicate that marginally attached and unemployed individuals present similar transition behavior to employment. I conclude that individuals who do not get classified as unemployed on the grounds that they ceased search warrant close attention, in that they remain attached to the labor market and transit into employment at rates close, if not equal to, those measured for the unemployed. Keywords: Labor Market Mobility, Marginal Attachment, Unemployment, Multinomial Logit, Binomial Logit, Likelihood Ratio Test, HLFS, Attrition/Substitution, Rescaled AN Model, Turkey