winsorize the data

[Rochelle Zhang]

Dear Statalisters,

I want to set the top and bottom 1% of the sample to r(p1) r(p99) respectively - winsorize the data.

I did

sum x, d

replace x = r(p1) if x< r(p1)

replace x = r(p99) if x>r(p99)

in stata13, the first replace works, but the second gives me zero observations,

I assume I can't use two replace consecutively.

please let me know if that is the case.

Best,
Rochelle

[Sarah Edgington]

Did you check whether there were any cases that satisfied your second condition? That is, are there any values of x greater than the 99th percentile? If your largest value is equal to the 99th percentile there would be no change made to x.

[Aspen Chen]

in stata13, the first replace works, but the second gives me zero observations,

Just one possibility: how many observations have values above the 99th percentile? You can check with

count if x>r(p99)

[Clyde Schechter]

Actually, -count if x > r(p99)- may not produce the intended result here, because it will count observations for which x is a missing value.

count if x > r(p99) & !missing(x)

[Aspen Chen]

Good point about the missing values, Clyde. Although I suspect Rochelle's data has no missing values, it is certainly good practice to add the extra condition in all cases. And this would also apply to Rochelle's second replace command.

[Rochelle Zhang]

Many thanks to Sarah, Aspen, and Clyde !!! You were all correct, I did not realize I do not have observation values greater than r(p99), I thought using two replace statements are incorrect.


Rochelle

[Rich Goldstein]

note that there are user-written programs for this; try -search winsorize-

[Victoria Rogers]

use
ssc install winsor
winsor Currentvariablename, gen(Newvariablename) p(0.01)

[Pallavi Chavan]

I have a panel dataset. I have imported it using xtset command. Now I would like to winsorize my dataset using one key variable in the dataset. What is the best way to do this? Can someone suggest the commands needed for this? Do I have download an additional module to do that? Thanks