How to get Cronbach's alpha in survey data?

[Man Yang]

Hello, I need to calculate the alpha using six survey items in a dataset but as far as I know, svy is not compatible with alpha. Is there a way for me to calculate alpha while taking survey weights into consideration? Thanks!

[Marcos Almeida]

You may wish to take a look at this thread.

[Steve Samuels]

Ben Jann's alphawgt program "Cronbach's alpha for weighted data" (at SCC) takes aweights. For the weight, specify [aw = your_sampling_weight].

I've tried a number of commands which take both aweights and pweights and compared the results; the estimates (though not standard errors) were always the same. This is documented someplace on the Stata web site, but I don't have the reference.

[Man Yang]

Thanks for your response Marcos and Steve. Since the sampling weight I am dealing with is jackknife survey weights, I don't think aw or pw will work in my case. I tried to get the covariance matrix first and then calculate alpha by hand, but the corr command cannot work with svy as well.

[Steve Samuels]

Pseudo-values for a jackknife require a full-sample weighted parameter estimate: \(\widehat{\theta}\). (See the Manual entry for svy jackknife). To compute this, survey datasets with jackknife replicates also include a "final" weight. Round that to an integer (perhaps after dividing by 10 or 100 or 1000) and use it as a frequency weight (fw).

[Steve Samuels]

Pseudo-values for a jackknife require a full-sample weighted parameter estimate: \(\widehat{\theta}\). (See the Manual entry for svy jackknife). To compute this, survey datasets with jackknife replicates also include a "final" weight. Round that to an integer (perhaps after dividing by 10 or 100) and use it as a frequency weight (fw).