Gender vs salary gap

#1
Hi all,

Gender vs salary gap is a hot topic for a few years now. Being a HR professional I'd like to introduce my company to use statistics (hypothesis testing) to interpret information instead of just drilling down data line by line which can take lots of effort and yield results that we may not be confident in concluding.

I had a little basic in statistics, 10 years ago...I do need your kind help to find out if the test I choose is correct to begin with.

Current raw data observation, 10% gap between male and female employees
Research question: to find out if gender plays a part in explaining the 10% difference in salary gap

I choose Regression to test out my data. It returns Rsquare=0.1%, significance F = 0.03, Gender P-value = 0.03

Finding is that even though Gender can statistically explain the difference in salary gap, it only explains 0.1% of it.

Question: Is the test I used correct to begin with?
Question: Did I correctly state the question and finding?
 

Miner

TS Contributor
#2
You may be missing other variables such as length of service, education, skills, number of job hops. Pay is very complex. Two people (same gender) with the same credentials start at the same company. One stays with that company their entire career. The other leaves, works for three other companies then returns. The second employee will almost always be at a higher managerial level and pay level than the one that stayed.
 
#3
yeap, I was keeping it simple in my example. In my case, I will be having different sample accounting for the business unit, pay level (aka type of roles). My current hypothesis is testing if Gender play a role in explaning pay difference. If yes, by how much? If not, then other factors may be in play.
 

hlsmith

Not a robit
#4
If you are trying to get back into stats, I would also make sure you are meeting the assumptions of linear regression as well. Some times I would imagine salary values can be skewed and capped.

Running a linear regression is a good first step. Though I will reinforce @Miner 's comment that you will need to control for other confounding variables and there could be interactions between variables. Also, you need to keep in mind your sample size when adding more variables.

Lastly, I think the legal field has articulated methods for examining discrimination. Some times they are crude, but they may also provide some direction.

Welcome to the forum!!
 

Miner

TS Contributor
#5
I caution you against overly simplistic analyses that do not consider all the potential covariates and interactions. Otherwise, you can run into Simpson's Paradox and misleading results. Your R^2 of 0.1 is telling you that there are a lot of missing variables.

Do you have higher/lower paying positions that are skewed toward one gender or another? If so, you may need to weight your analysis to account for this. Rather than a pay gap, there may be fundamental reasons why each gender selects or is selected for particular roles. This might be outright bias, or it could be self-selection where one gender prefers or avoids certain roles.
 
Last edited: