Chi2 test with if command

[Christian Huber]

Dear statalister,

in my current project i would like to compare a sample attribute with the population's attribute in order to verify if the sample is representative for the population. For example, my sample includes the age of firms which I´ve also collected through a survey. I sent the survey to all firms in a databse (n=100) and received answers from about 50 (just as example). Now, I want to compare the age distribution of sample to the age distribution of the population database. My stata datatable looks like the following:

	Survey	AGE_Survey (of Survey respondents (n=50))	AGE_DB (in Database (n=100))
Firm 1	1	5	5
Firm 2	1	2	2
Firm 3	0		3
Firm 4	0		2

If i want to see the age distribution of the sample, i would for example use the following command (in order to omitt empty cells): tabulate AGE_Survey if Survey==1

How would I be able to compare the age distribution of AGE_Survey (Survey=1) with the age distribution of AGE_DB using Chi2 or similar tests? I want that stata omitts the empty cells of the column AGE_Survey when comparing.

I tried somehting like that: tab AGE_Survey if Survey==1 AGE_DB, chi2 but was not successful.

Maybe some has a recommendation?

Regards,
Alex

[Marcos Almeida]

Hello Christian,

Welcome to the Stata Forum.

You may try something like this:

. svy, subpop (if Survey ==1): tab Age_Survey Age_DB, row

Hopefully that helps!

Best,

Marcos

[Christian Huber]

Dear Marcos,

thank you for the fast reply. I think this goes into the right direction. However, Stata uses the answers from AGE_DB which fit the condition Survey=1. I wanted that Stata only considers Survey=1 for AGESurvey and not for AGE_DB.

Is there a way I could tell stata?

Regards,
Christian (Christian-Alexander)

[Marcos Almeida]

Well, if I understood correctly, I fear there is no way (neither point) in getting a 2-by-2 measure of assocation with a chi-square test, having missing values in one of the four cells.

Hopefully you will get further advice on this.

[Oded Mcdossi]

Here is a trick to present the two distributions side by side, however to compare the distribution you should check for tab_chi (SSC) by Nick Cox, I have no experience with this program but I think it does exactly what you are looking for. Read also the helpful notes by Richard Williams.

clear*
input Survey    AGE_Survey    AGE_DB    Firm
1    5    5    1
1    2    2    2
0    .    3    3
0    .    2    4
end
expand Survey*2, g(to_expand)
lab de to_expand 1 "Survey respondents (n=50)" 0 "Database (n=100)"
la val to_expand to_expand
tab AGE_DB to_expand, col