The groupby object has four methods that accept a function (or functions) to perform a calculation on each group. These four methods are agg, filter, transform, and apply. Each of the first three of these methods has a very specific output that the function must return. agg must return a scalar value, filter must return a boolean, and transform must return a Series with the same length as the passed group. The apply method, however, may return a scalar value, a Series, or even a DataFrame of any shape, therefore making it very flexible. It is also called only once per group, which contrasts with transform and agg that get called once for each non-grouping column. The apply method's ability to return a single object when operating on multiple columns at the same time makes the calculation in this recipe possible.