Oaxaca decomposition - gender wage gap

[Thea Fredriksen]

Hi, I am currently analysing the gender wage gap in the UK for my dissertation. I am using data from the Labour Force Survey and have included the years 1997, 2010, and 2019, and will be looking at the wage gap at these specific points in time. I will use the Blinder-Oaxaca decomposition to see how much of the wage gap can be explained due to differences in some control variables. I am, however, having some problems with interpreting my results, as I was expecting the coefficient on the "explained" part to be positive when controlling for these variables. I am concerned that there is something wrong with my data etc. because of this, but I would be grateful if someone could help me understand my results and interpret whether this seems odd.

[FernandoRios]

Hi Thea
I think a good exercise will be for you to do the OB decomposition by hand.
Right now, you assume explained component SHOULD be positive, but that is not necessarily the case.
Thus, start by constructing all the components for the OB decomposition (Means and coefficients), and try to replicate the decomposition by hand.
This may help you understand why the effects are as the command suggest
HTH

[Thea Fredriksen]

Hi FernandoRios, thank you for your respond! I will definitely do that, however, I am just wondering if the results makes sense with the fact that I am analysing the wage gap? I feel like I have a good understanding of the OB, however I am struggling a bit with the fact that endowments are negative.

[FernandoRios]

Right, but that is a matter of comparing means. If you do this by gender, what do you see?
Do women have better characteristics than men?

[Thea Fredriksen]

More women have a degree which makes sense with regards to the numbers, I guess I am just uncertain about my data as I would assume historically more men have degrees.

[ericmelse]

Dear Thea,

Please see the FAQ tab at the top of this web page, with this paragraph - 12.5 Posting attachments: please don't.. - on screenshots.
These are mostly hard to read and possibly not as useful explaining what you are looking for (paraphrasing Nick Cox).
Instead, make good use of the code tags (use the # button in the new post bar of options) in between you can post your own code as well as the result tables etc. from the Stata Results window.

[FernandoRios]

well, that is different
its not a matter of the data was collected, or if indeed patterns have changed . But that is a question on the Data collection and official statistics.