What Do Conditional and Unconditional Quantile Regression Models Tell Us Something Different About Wage Inequality in Turkey?
Ebru Çağlayan Akay, Fulden KömüryakanThis study comparatively analyzes wage inequality in the Turkish labor force by estimating the generalized Mincer wage equation with quantile regression methods using the 2018 Turkish Household Budget Survey data. This is the first study on wage inequality in Turkey that includes a comparative analysis of conditional and unconditional quantile regression methods. The results indicate that conditional quantile regression estimates wage inequality to be lower than it actually is. In contrast, in unconditional quantile regression, wage inequality is higher. The results further provide evidence of wage inequality in the Turkish labor market and suggest that wage inequality is higher in low-wage segments.
Koşullu ve Koşulsuz Kantil Regresyon Modelleri Türkiye’de Ücret Eşitsizliği Hakkında Farklı Ne Söylüyor?
Ebru Çağlayan Akay, Fulden KömüryakanBu çalışmanın amacı, Türk işgücündeki ücret eşitsizliğini genelleştirilmiş Mincer ücret denklemi çerçevesinde, koşullu ve koşulsuz kantil regresyon yöntemleri ile tahmin ederek karşılaştırmalı olarak analiz etmektir. Bu amaç için 2018 Türkiye Hanehalkı Bütçe Anketi verileri analiz edilmiştir. Bu çalışma, Türkiye’de Mincer ücret denklemini koşulsuz kantil regresyon yöntemi ile inceleyen ve koşullu ve koşulsuz kantil regresyon tahmin sonuçlarını karşılaştırmalı olarak analiz eden ilk çalışmadır. Analiz sonuçlarına göre, koşullu kantil regresyon yönteminin ücret eşitsizliğini olduğundan daha az belirlediği, koşulsuz kantil regresyon yöntemine göre ise ücret eşitsizliğinin daha fazla olduğu belirlenmiştir. Sonuçlar, Türk işgücü piyasasında ücret eşitsizliğinin olduğunu ve ücret eşitsizliğinin ücret seviyesinin düşük olduğu kesimlerde daha fazla olduğu hakkında kanıtlar sunmaktadır.
Wage inequality arises when the welfare level of some individuals in society is higher than others. Although wage inequality is relatively lower in developed economies, it remains a much more serious socioeconomic challenge for emerging market economies, along with labor force competition. Therefore, wage inequality is one of the most examined topics in literature.
This study examines the wage inequality by analyzing the Turkish Household Budget Survey (HBS) conducted by the Turkish Statistical Institute (TURKSTAT) in 2018. The HBS was administered to 15,551 households in Turkey. We analyze 5,455 employees who reported positive income in the survey year.
This study further comparatively analyzes wage inequality in the Turkish labor force by estimating the generalized Mincer wage equation with quantile regression methods. The generalized Mincer wage equation contains three sets of variables as human capital, demographic features, occupation and work types
Although ordinary least squares (OLS) is the best unbiased linear estimator to analyze the relationship between variables, it is based on the conditional mean of the distribution and breaks in cases of long-tailed distribution. Thereby OLS could take any arbitrary value in the case of long-tailed distributions. Since outliers and long-tailed distributions are quite common in the analysis of wage equations, Koenker and Bassett (1978) proposed a new method, termed as conditional quantile regression (CQR) to overcome the limitations of OLS and address this challenge. CQR is based on the estimation of conditional quantiles of distributions. In this way, CQR allows the estimation of the conditional quantile functions and determines the statistical differences of parameters in different quantiles of distribution. CQR is robust to heteroskedasticity and outliers, is more efficient than OLS in the case of the nongaussian error term, and elicits detailed information regarding the differentials in a distribution by computing several regression curves corresponding to different conditional quantiles in the distribution. Although the CQR method has several advantages, it can become limited in some situations. If the variables for different conditional quantiles alter, the interpretation of these variables may be affected. Since the distribution of the dependent variable at the θth quantile depends on the conditional distribution of the given explanatory variable in the CQR method, the distribution of inequality in the dependent variable may be affected by the inequality in the given explanatory variable. Hence, the inequality can be estimated as more or less than its actual value. Therefore, CQR may estimate coefficients that are sometimes not generalizable or interpretable for a population.
Firpo, Fortin, and Lemieux (2009) proposed a regression method termed as unconditional quantile regression (UQR) to overcome these limitations of CQR. UQR is used to estimate the effect of explanatory variables on the unconditional quantiles of a dependent variable. UQR makes it possible to analyze different relationships between the variables and elicits a more detailed result among different groups in the sample. UQR estimates more interpretable and generalizable coefficients than CQR as it marginalizes the effect over the distributions of other explanatory variables in the model. Basically, CQR estimates the impact on the dependent variable conditional on the explanatory variable for an observation, whereas UQR estimates the impact on the dependent variable at the θth quantile for all observations in a sample unconditional on the explanatory variable.
This is the first study on wage inequality in Turkey to apply a comparative analysis of conditional and unconditional quantile regression methods. The results suggest that CQR determines wage inequality as lower than it actually is. In contrast, in UQR, wage inequality is higher. The results further provide evidence on wage inequality in the Turkish labor market and suggest that wage inequality is higher in low-wage segments.
