Variables that are selected arbitrarily for the model construction are often used in many publications devoted to wages and labour market in various parts of the world. These variables describe respondents’ characteristics: monthly net earnings, education, occupation, nature and size of employee’s firm, sector of employment (agriculture, industry, services or other), marital status, age, size class of the place of residence, gender and relationship with the head of the household. For these features (except for age and earnings, which were quantitative or qualitative, depending on the model we built) we create 33 binary variables identifying actual attributes of observations.
Data used in our investigation came from Labour Force Survey (LFS) which is being performed by Central Statistical Office in Warsaw every quarter of a year since 1992. This survey is the main source of labour market information in Poland and is consistent with Eurostat’s and International Labour Organization’s guidelines. We used LFS databases from the fourth quarter of 2005 and the first quarter of 2009.
We built nine models - six of them were ordered multinomial logit models, which were estimated by Maximum Likelihood method, and three of them were exponential models, which were estimated by Ordinary Least Squares method (after linearizing them). Among these nine models, some of them are general models (estimated on the basis of all observations and variables) and partial models i.e. estimated separately for men and women that simplifies gender wage gap analysis.
The main difference between situation observed in the years 2005 and 2009 is that private institutions employees rated their wages lower than public institutions employees in 2005 however this state of affairs was completely reversed in 2009. Respondents, who lived in towns up to 10 thousand inhabitants, declared higher earnings than employees from villages in 2005. In 2009 this tendency changed and the parameter standing by this variable is statistically significant. Variables describing sector of employment are insignificant in both multinomial models, but they are significant in exponential model, that is the main difference between these two types of models.
We observe gender wage gap in 2005 and 2009 in Poland (that seems to became even deeper in the year 2009). Irrespective of the year we analyzed, women working in private institutions were better paid than women in public institutions. Men’s situation was converse – public workers earned more than private. The interesting observation is that the character of employer’s institution influenced men earnings in 2005, but was no more significant in 2009, however this factor became meaningful for women earnings then.