Survey Weights

[Jia De Gedeon Lim]

Hey Everyone,

At some point in our research assignments we might need to use probability weights so I've taken the liberty to include some useful code here from Ajay's solowhh_factpred.do. (There might be other bits of useful code lying around that I didn't notice)

//////////////////////
///(1) weighting options

//Do we want to attempt to reweight or not?
local useweights 0

local weightfolder1 stylized
local weightfolder0 stylized_noweight

//////////////////////
//(2) If we don't want weights, equalize them
if !`useweights' {
    foreach var of varlist weight71 wt82 weight99 {
        replace `var' = 1
    }
}

I think that the first part "local useweights 0" basically states whether or not you want to use weights in your analysis. The second part sets all weights to 1 if you don't want to use weights for your analysis. In this way, you can include the command [pweight=___] in your entire code and just change the macro `useweights' to 0 or 1 (?) if and when you want to run weighted or unweighted regressions etc.

Also, this is a nice FAQ that explains how to construct variables containing weighted summary statistics since commands like egen doesn't support weights (although the package -egenmore- might be useful): http://www.stata.com/support/faqs/data-management/weighted-group-summary-statistics/

Thanks.

Best,
Gedeon

[Jia De Gedeon Lim]

Hi Everyone,

Sorry for the spam. The "summarize" command doesn't work with pweights but according to http://www.stata.com/support/faqs/statistics/weights-and-summary-statistics/, the use of aweights will give us the same values if we are interested in only point estimates. It also goes into detail about the use of weights; how the summarize command works; and how the svy: mean command works etc.

Best,
Gedeon