Page 89 - Azerbaijan State University of Economics

THE JOURNAL OF ECONOMIC SCIENCES: THEORY AND PRACTICE, V.80, # 1, 2023, pp. 83-93

In the original regression, GDP per capita is used solely however, in our model we
preferred logarithmic transformation of GDP per capita. There are several reasons for this
change. First, as a number GDP per capita is huge compared to other variables so we
could not get reasonable coefficient in that way. The second as most importantly, GDP
per capita is highly skewed for Azerbaijan since we think high economic growth over the
past thirty years, so we need to normalize data. Moreover, change of GDP per capita is
more important rather than GDP per capita itself. Another reason for the logarithmic
transformation is to decrease heteroskedasticity (non-constant error term) problem.

Moreover, we tried to add unemployment rate ratio for male to female into our
regression but in that case, our explanatory variables became insignificant there
happened multicollinearity problem which might cause bias problem, so we had to
drop unemployment ratio variable.

During our analysis, we discovered that some of our variables such dependent variable
wage ratio of male to female have trending issue, which is common for timeseries
analysis (See Table 4). The main issue of trending is that sometimes variables show
trends in same or opposite directions where it misleads coefficient and cause to think
there might be relationship between them. The trending can be linear or growth level.
Yet even if trending is not observable, for the safety it is advisable to add time variable
into regression model to eliminate any trending related problems (Wooldridge, 2012,
p.363-364).

Table 4: Trend analysis

Source: The results are obtained through the author’s analysis

89