Statalist The Stata Listserver

[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: winsorize the effect of outliers (?)

From   "Nick Cox" <[email protected]>
To   <[email protected]>
Subject   st: RE: winsorize the effect of outliers (?)
Date   Wed, 8 Nov 2006 16:13:59 -0000

Maarten Buis has already drawn attention to relevant 
stuff. But other comments are possible: 

1. Only those who recognise the reference "Cleary (1999)"
can say precisely what he or she did. I guess that -2 
and 2 means winsorize 2 at the bottom and 2 at the top 
but I don't then know what 1 and 0 means. I suspect 
a typo there. 

2. On the evidence here different amounts of winsorizing 
were applied to different variables. If that is 
based on inspection of the data any P-values that follow
are suspect, from at least one point of view. 

3. Winsorizing is a kind of ad hoc method some 50 years
old. Aren't there now better ways of getting model fits
robust to outliers? 

[email protected] 

P.S. I did write -winsorize-, but I never use it. Someone
asked how to do it.... 

[email protected]
> In an working paper I read: "To avoid the effect of outliers, 
> we winsorized the observations [time series panel data] 
> following Cleary (1999). The cutoff value are -2 and 2 for 
> "var1[_t-1]", -5 and 5 per "var 2[_t-1]", 1 and 0 for "var3". 
> Note: [_t-1] means a value taken in (t-1).
> Given these three variables (var1, var2 and var3), could 
> somebody tell me how to "winsorize" this way? Thank you. 

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index