Step 1 has the simplest possible aggregation with a single grouping column (RACE), a single aggregating column (BASE_SALARY), and a single aggregating function (mean). This result is easy to consume and doesn't require any more processing to evaluate. Step 2 slightly increases the complexity by grouping by both race and gender together. The resulting MultiIndex Series contains all the values in a single dimension, which makes comparisons more difficult. To make the information easier to consume, we use the unstack method to convert the values in one (or more) of the levels to columns.

By default, unstack uses the innermost index level as the new columns. You can specify the exact level you would like to unstack with the level parameter, which accepts either the level name as a string or the level integer location. It is preferable to use the level name over the integer location to avoid ambiguity. Steps 3 and 4 unstack each level, which results in a DataFrame with a single-level index. It is now much easier to compare salaries from each race by gender.