Chi2 test with iweights

[Emilie Venborg]

Hi,

I am trying to do a chi2 test with weighted data (iweights), but I have not found a useful command.
I do not know which variable to use for psu - is it possible to do weighted chi2 without a psu variable?

Thank you in advance!

[Nick Cox]

It's hard to see that chi-square tests -- presumably you're referring to analysis of counted frequencies in a contingency table -- make any sense unless you are dealing with such counted frequencies, directly or through frequency weights.

There's, I guess, a more appropriate analysis for what you're doing -- but if it's survey data without survey details even the svy-savvy (not me) might struggle to advise.

[Rich Goldstein]

see

help svy_estimation

which shows you that tab can be used with survey data

[George Ford]

what command are you using?

[Emilie Venborg]

Hi George

I have tried some different commands including:
Code: tabulate occupation belief [iweight= weights2], chi2
Error: option chi2 not allowed

Code: crosstab occupation belief [iweight=weights2], chi2
Error: command crosstab is unrecognized

AI recommended the following command, but as I wrote in my question I am not sure which psu variable to use, so if possible I would like to avoid this command:
Code: svyset [weights2], psu([variable x])
strata ([belief])
svy: tabulate occupation, row chi2

[Leonardo Guizzetti]

Two things.

1) From -help weight-, you should take serious caution in using -iweights- unless you know that this is the correct approach. To quote from the documentation (emphasis mine):

4. iweights, or importance weights, are weights that indicate the "importance" of the observation in some vague sense. iweights have no formal statistical definition; any command that supports iweights will define
exactly how they are treated. Usually, they are intended for use by programmers who want to produce a certain computation.

I've not seen iweights used with survey data, so I can't advise further.

AI recommended the following command, but as I wrote in my question I am not sure which psu variable to use, so if possible I would like to avoid this command:

2) It's my firm belief that using "AI" when you are not already an experienced user is asking for folly.

If you have survey data, then you should be able to go to the producer of that survey for documentation on precisely how you need to account for its design. Again, we can't advise further since we don't know your survey.

[Nick Cox]

The mention of crosstab is a case in point about the limitations of AI. Unless someone knows otherwise, it appears to be an hallucination. Naturally, many people refer to cross-tabulation and more informally to a crosstab, and I found mention of a procedure with that name in SUDAAN, but as far as I know there is no such Stata command.