Stata - How to winsorize your data

Welcome to my classroom!
This video is part of my Stata series. A series where I help you learn how to use Stata. In this video, we look at how to use the winsor command to winsorize your data/variables. Often a better alternative to dropping certain observations.
Note: What I show here is my take on the topic. I would be happy to receive comments!
Useful links:
►Twitch: / steffens_classroom
►Twitter: / steff5001
►Workpage: www.rug.nl/sta...
►Subscribe: cutt.ly/Qfu9cmV

[@SteffensClassroom]

Happy my brother had some fun editing this one -.-

[@vartuhitonoyan4923]

Awesome job, brief and to the point. Thank you so much.

[@whyosrs9863]

steffen legend

[@vanisher49]

Thank you this was very helpful!

[@sitinabilahmohdshaari6771]

many thanks

[@failedomelet9978]

Pretty good video my dude
Had a bad time near the end though

[@SteffensClassroom]

Whatever makes his day xD

[@juhokim6149]

Thank you

[@bellisma77]

Thanx for your videos. Is there a specific criteria for choosing the cut off points here?

[@SteffensClassroom]

Hi!
Often, 5 or 10 percent is used. I need to look more into the literature, but always justify your choices.

[@bellisma77]

I wanted to apply winsor2 to my thesis data and afraid it won’t be suitable.

[@SteffensClassroom]

@@bellisma77 Well, it depends on how the data looks. Do you have a lot of outliers? Is it skewed? In any case, it is a good idea to conduct your analysis with and without applying winzorizing.

[@bellisma77]

@SteffensClassroom not that much outliers and i want to replace them by winsorizing with +/- 3 sigma

[@SteffensClassroom]

@@bellisma77 You just need to do it yourself, which fortunately is not very difficult in this case.
sum x
scalar low = r(mean) - 3 * r(sd)
scalar high = r(mean) + 3 * r(sd)
gen x2 = cond(x < low, low, cond(x > high & x < ., high, x))
This should do it.

[@FemkeHuisman]

If you should winsorize all continuously scaled variables, does that mean variables with 'real numbers' (stata: 'float' and 'double'), or also integer variables (in stata named 'byte', 'int', 'long')?

[@SteffensClassroom]

Be sure what you mean by continous here. You may be mixing things up here. I typically would only consider string vs numeric when it comes to winzoring, as winzoring works the same way for all numeric types in Stata (to the best of my knowledge).

[@FemkeHuisman]

@@SteffensClassroom Thanks!!

[@Kerrmunism]

Wait but why are one of their eyes blue tho?

[@darvidtorres]

How to winsorize for panel data?

[@SteffensClassroom]

Afaik, same startegy. You are winsorizing prior to estimation.

[@aminaahmedalibelal5676]

Thanx for your simple and clear explaination. Is it possible to replace the values with say the mean?
winsor2 x1 if z ==1,replace(mean)? i want to replace x1 which its z =1 with the mean value. Is this correct?

[@SteffensClassroom]

Honestly, never tried. However, what is your reasoning for wanting to replace your values with the mean? It doesn't sound like a good idea.

[@aminaahmedalibelal5676]

@@SteffensClassroom i have a panel data and i detected outliers using Cook's distance, and leverages, and i want to replace these points with mean or median value. I guess using replace command will be suitable right? thank you Dr. for you reply

[@SteffensClassroom]

@@aminaahmedalibelal5676 Hi again. The replace command would of course work. My question was more like why you would like to replace the values with the mean/median. There must be a reason I am missing :)
Good luck with your project!

[@aminaahmedalibelal5676]

@@SteffensClassroom Hi. I have a percentage of 2.30% of outliers and i do not want to trim them so my panel data will not be unbalanced. Thanks for your wishes. Hope you loaded more video's on STATA.

[@Josobro]

snans blundertale.

[@JetEriksen]

Funny Skeleton

[@xtra3678]

Sans is ness confirmation

[@crusaddy982]

Ketchup

[@kittentamer2164]

Funnybones

[@mitchellroberts9626]

doot

[@Suicutie]

undersnail

[@GrandleJams]

Haha, Megalovania

[@abujad4226]

Thank you