英语论文网

cases to collect a set of 2737 random observations (of individuals, not households) with 18 explanatory variables and weekly wage as the dependent variable.

In this paper, self-employed individuals were not omitted. Even though their weekly earnings are difficult to observe, this group constitutes only a small fraction of the sample. The main criterion when scrubbing the data was reported earnings. The missing values had to be passed as this would have important distortions for the estimation. Retired people were excluded for the dataset unless they were younger than 65 (60 for women) and had returned to the workforce. Hence, a sample of working age individuals, those having part-time or full-time jobs, and those recorded sick or disabled, will be examined in my study.

The issue of data quality and credibility is a very important one. The British Household Panel Survey is carried out at the Institute for Social and Economic Research of the University of Essex and is very popular among researchers in social sciences for both quantitative and qualitative analyses. Thus, it is a well-recognised and reliable source of microeconomic data. Regrettably, as any other survey, it has its limitations. As long as a human factor is included, a researcher cannot be sure to what extent the answers to the survey questions are unbiased. The issue of bias is not critical in this study as I deal with employment and other generic figures where little bias is expected. However, a more relevant question is whether the numeric and categorical observations are reported accurately. It is important to take such concerns into account when drawing conclusions, yet this should not obscure the weight and significance of a study. As well, a number of limitations exist in the dataset designed for this particular paper. One of them is the underlying uncertainty in sample representativeness of the whole population. Second, my paper only looks at the period of one year which means that proxy variables for some characteristics have to be used (e.g. work history/experience). This automatically introduces uncertainty and problem of auto-correlation. Therefore, the findings obtained in this paper will have to be treated with great caution given the data and methodology limitations.

Firstly, a dummy variable defining individual’s gender was produced and will be used as a group indicator in the Blinder-Oaxaca decomposition. Secondly, a natural logarithm of a reported weekly wage was generated as this is necessary for estimating log-linear models (see section “method”). One of the major discrepancies between human capital theory and practice is the length of schooling. BHPS does not provide a measure for years of education, therefore, typical studies use the ‘highest educational attainment’ variable. Since it is multi-categorical, there is a possible solution. A scale of educational levels can be constructed to obtain a measure for education. Olsen and Walby (2004) use a point system where individuals are ‘awarded’ points for their educational attainment, that is, a respondent is given a number of points equal to presumable number of years of schooling depending on the highest educational attainment. For instance, 13 points for A-levels, 16 points for a university degree, and 17 points for a higher degree (see: Appendix 1). However, I have not included this constructed variable directly in the regression. For potential experience, I