In my post on Gartner’s EPIC work on April 5, 2021, I talked about using regression to assess pay equity. That is the ideal statistical analyses. But it has two important limitations. You can only use multi-variate regression when:
- the total number of employees in your comparison group is at least 30 individuals.
- there are at least five individuals in your underrepresented group. Thus, for gender pay equity you need at least five women in your comparison group.
When clients follow our advice in our EPIC methodology (for example, analyzing the company overall controlling for country differences … see this blog for more on that) they will tend to have these larger comparison groups. But there will be times when you might only have 10 employees doing the same role, including only one or two women. How do you assess pay equity in this situation?
Follow a Two Step Process to Evaluate Small Groups
For those (hopefully) limited number of roles where you have only a few employees to compare, then you should follow this two-step process.
- Use a two-sample t test to compare the salaries. Your two samples are the men and the women. In this test you aren’t putting in any of the reasons why the pay could be different. If you find that there is a statistically significant difference, then there is reason to move to step 2. But if you do not find a difference, then you can stop here. We will talk about how to report results in a future blog.
- If you do find a statistical difference, then you will want to do a visual analysis of the comparison group. I am not going to lie; this is awkward and never feels definitive. You put the employees in the comparison group in an Excel spreadsheet as the rows with all the variables that you would put in for regression as the columns. And then you do a visual analysis. And try to figure out why the two-sample t test identified the group as being paid less.