Question: How to Winsorize using Standard Deviations instead of percentiles?

[Adam Aberra]

Greetings,

Is there a way of Winsorizing past a certain threshold of standard deviations instead of percentiles?

In particular I would like to Winsorize + / - 3 sd.

The dataset I'm working with is constantly being updated so such a code (if it exists) would be most helpful.

I have installed winsor and winsors2. I have read the help documentation for each of them but to no avail.

Thanks for your help.

[Nick Cox]

You just need to do it yourself.

su y 
scalar low = r(mean) - 3 * r(sd) 
scalar high = r(mean) + 3 * r(sd) 
gen y2 = cond(y < low, low, cond(y > high & y < ., high, y))

That seems an especially poor to winsorize, however. (It's not even winsorizing, but a disreputable relation.) Any outliers pull out the mean and SD any way.

[Adam Aberra]

Thank you, Nick. That answers my question.