Compare a subset to entire population that includes the subset

[J. Sandhu]

Hello Statalist,

I have been asked to compare a subset to the entire population to see if the subset is significantly different from the overall population. So for example, I want to know if the proportion of Californians getting their vaccinations is statistically different from those getting vaccinations across the country (including California).

I know that this is incorrect: tab vaccine California , chi2

as this will compare Californians to non-Californians which is not what they want.

I don't believe I can use chi-squared test at all because the populations are not independent. Can anyone guide me to the correct test?

Thank you for any help you can give me,

J. Sandhu

[Clyde Schechter]

No, you can't use chi square to contrast non-independent populations. Actually, you can't use any test that I know of for this purpose. The thing is, the null hypothesis that Californias are different from Americans as a whole is logically the same as Californians are different from non-Californian Americans. So just go ahead and use whatever tests (chi square, or more complicated models like logistic regression, etc.) to contrast Californians with non-Californians. Whatever conclusion you reach about that contrast will be valid for the Californians vs All Americans contrast.

If your audience expects you to show descriptive statistics such as the proportions of Californians who were vaccinated and the proportions across the country, you can calculate them separately if need be. (But if you are using -tab, chi2-, the output will already show you California results, non-California results, and the totals.)

[Steve Samuels]

See also: http://www.statalist.org/forums/foru...tandard-errors Note the very strong request in the FAQ and when you registered to use full real names on Statalist.