Recall: The use of pweights with regress

[Aaron Kratzer]

Hello Statalisters,

I have been searching for an answer to the following question: "Can anyone point me to any documentation as to exactly what/how pweights are applied [when using with -regress-]?"

This is actually pulled directly from a seemingly unanswered statalist archived post from 2004 (unanswered because the posted replies do not seem to have any references or links to documentation, which seemed to be one of the main questions of the original post). After searching the internet at large, statalist updated forum, the statalist archives, the stata journal, the general user manual, and of course the -search- and -help- commands, I have been unable to find any specific information about how pweights are applied when used with -regress-. For other commands, I have been able to find links to articles at the bottom of the help page explaining the math behind the command, or occasionally links from statalist; but I have not found any such information for pweights. In particular, I am looking for how the use of pweights affects the estimation of coefficients and robust standard errors.

I understand what pweights should be, and have calculated them for my data, and I am just looking for a theoretical understanding of what stata does when I use the following:

regress mydepvar myindepvars [pw=mypweightvar]

Could someone help me find this documentation?

Many thanks for any help on this!

[Richard Williams]

If you type

help weight

you get a description of the different types of weights. Section 11.1.6 of the Users Guide says more, What exactly is it you want to know that is not in these sections?

[Aaron Kratzer]

Hello Richard!

Thank you for the quick reply. I am familiar with the definition of what a pweight is (namely, an inverse probability-of-sampling weight). Section [U] 11.1.6 only seems to define what the pweight should be, and not how it is used computationally or theoretically. I have also read through section 20.24.3, and it seems to start to discuss the statistical properties of pweighting the data, but it does not seem to clearly state how it affects calculation of coefficient estimates or the standard errors of those estimates. It also does not seem to reference any relevant published articles in reference to the method, as many other stata commands/options do.

In particular, I am looking for how the use of [pweight=...] in the -regress- command affects the estimation of coefficients and robust standard errors. Do you know of any articles/peer-reviewed content that this implementation of pweighting is based on?

Thanks for your help!

[FernandoRios]

Hi Aaron
The information you are looking for its there.
Look into the regress helpfile and the manual reference. Then go to Method and formulas. Finally, check the section of weights and robust estimations.
It details the estimations under aw iw pw and fw. It also details how to do robust hc2 and hc3 variances.

Fernando

[Aaron Kratzer]

Hello Fernando,

Thank you very much! I think that is exactly what I was looking for. I guess I overlooked the most basic location for the information!

For anyone looking for this in the future, this is on page 2254 of the Stata Base Reference Manual [R] of the version 15 release.

Again, thanks for the help!

[Marcos Almeida]

If you’re hungry for more on pweights , I recommend the reading of “Survey Weights: a step-by-step guide to calculation” (StataPress, 2018) written by Richard Valliant and Jill Dever.

The focus is on Survey probability weights, and you’ll come across lots of references worth reading as well.

[Aaron Kratzer]

Marcos, thank you very much! This is part of my master's project, and this looks like some great summer reading to dig in deep to survey weights. I am requesting that my university library get a few different Stata manuals.