This is an econometrics exercise. Basically, we have various variables we can use to explain the pay gap, including the worker's education, the worker's job, and whether the worker works long hours or short hours. All those variables are included in the econometric regression, but there are other variables we cannot include. It's much harder to model attitudes towards women or their career progression. It is much harder to model social institutions. There is the tax benefit system. Is that really neutral towards both partners working in couple families? There are various factors that the econometric exercise doesn't capture.
You can see from this chart that in some countries the unexplained variable is much bigger than in other countries. For instance, in Korea a lot of workers are on annual contracts that are renewed, but their earnings progression isn't as strong as for workers who hold a full-time or a permanent contract, so to speak. Then the question becomes why we have this pay gap, since on the whole, young women in Korea are now much better educated than are young men. There's a generation gap, in that women 20 to 30 years ago were not in that strong an educational position, but there is also an element of disadvantage, an unwillingness among employers to invest in women, and there are still very traditional expectations. Korean employers still expect their women workers to either leave employment when they become pregnant or take leave and maybe not come back, so their level of investment is much lower, but it's very difficult to capture that in econometric regression.
If you ask me what that unexplained bit means, it means two things: there are a certain number of things the econometrics don't pick up, but there may also be an element of discrimination in the labour market. Unfortunately, I cannot tell you how large or small that particular element is.