winsorization: different num of obs are changed if win by group v.s. win the whole sample

[Darcy Pu]

Hello Forum,

We find that the numbers of obs changed are different if we win by group v.s. win the whole sample.
We attach a screenshot of this.

In theory, win by group or win by the whole sample should change the same num of obs.
Could anyone please help explain?

Thanks
Darcy

[Andrew Musau]

In theory, win by group or win by the whole sample should change the same num of obs.

How so? Isn't the point of winsorizing by group that the cutoffs are group-specific? So that with no groups you have 2 cutoffs and with \(N\) groups, you have \(N \times 2\) cutoffs.