Winsorizing in conjunction with logarithm or not?

[Daan Peters]

Hi all!

I am conducting a research with panel data, and I am using a variable with some very extreme outliers. My supervisor has told me to winsorize the variable and making a logarithm of the variable, to account for the outliers. I don't know whether it is correct to do both or to do one of the two. Not sure whether it is relevant or not, but after this step I will use the variable in a regression (regress). Hopefully you can give me some advice on what to do!

Kind regards,

Daan

[Nick Cox]

I am not sure that we can or should arbitrate when your supervisor is giving you instructions. Why not ask her?

Who says what is correct? My personal position is odd in that I wrote winsor (SSC) but never use it. But evidently winsorizing is regarded as routine in some fields.

Evidently your variable of concern is all positive, else why consider logarithms? Taking logarithms will often tame high outliers, thus reducing or removing much of any incentive to winsorize. Occasionally it will produce outliers with very low positive values, so watch out.

[Carlo Lazzaro]

Daan:
welcome to this forum.
As an aside to Nick's helpful reply and following his suggested caution in commenting on supervisor's instructions, I would ask her/him notwithstanding why -regress- (and not, say, -xtreg-), is the first choice for panel data analysis.