topic Re: /white command for robust standard errors under heteroskedasticity in Statistical Procedures

/white command for robust standard errors under heteroskedasticity

SolarCat — Fri, 11 Mar 2016 09:27:30 GMT

I've been using proc reg with the /white tag and i find it drops data points from the analysis. What is the reason for this?

Re: /white command for robust standard errors under heteroskedasticity

RW9 — Fri, 11 Mar 2016 09:36:49 GMT

I am no statistician, so can't help with the logic, however Google throws up a fair few papers on the subject:

https://support.sas.com/rnd/app/examples/ets/hetero/

For example.

Re: /white command for robust standard errors under heteroskedasticity

SolarCat — Fri, 11 Mar 2016 09:45:26 GMT

Thanks but that paper doesn't help. I googled for a while before asking
this question. 😞

##- Please type your reply above this line. Simple formatting, no
attachments. -##

Re: /white command for robust standard errors under heteroskedasticity

Rick_SAS — Fri, 11 Mar 2016 10:25:54 GMT

PROC REG and other regression procedures drop observations for which any explanatory variable or response variable has a missing value. It only uses "complete cases": the set of observtions for which all variables in the model are nonmissing. See the doc section on "Missing Values" for more info.

Re: /white command for robust standard errors under heteroskedasticity

SolarCat — Thu, 17 Mar 2016 19:38:17 GMT

I apologize. I used a weight command, weighting by the square root of the variable, which removed all zero values. Thanks for your help!