Combine two datasets to make a t-test

[Camilla Christensen]

Hello.

I have two different data sets. One contains 2448 women, and one contains 552 women. The 552 woman is a part of the 2448 women. What I want to test is to see if my 552 women are representative for the 2448 women. For that reason I want to to a t-test to see, if there is a statistical significant difference between the two groups in respect to 6 different variables (mean birthweight, distribution of sex etc). I have all my information about the 2448 in state file A, and my information about the 552 women in stata file B. I now want to combine the two data sets (without merging!!) to be able to do my t-test.

Can anybody help me?

Thanks a lot

[bamboo Zhu]

#1 I may suggest you use "append" to combine two data files and create a new group variable to define file 1 and file 2. Then you can do the t-test in one file.

[Camilla Christensen]

#2 when I append the two datasets, all the values in one of the data sets changes to 0 or 1. Why is that?

[Bruce Weaver]

Hello.

I have two different data sets. One contains 2448 women, and one contains 552 women. The 552 woman is a part of the 2448 women.

At the risk of asking a silly question, is there a dichotomous variable in the larger dataset that has one value for the 552 women and another value for the other 1896 women?

[Camilla Christensen]

At the risk of asking a silly question, is there a dichotomous variable in the larger dataset that has one value for the 552 women and another value for the other 1896 women?

No, there is not. Would that help me?

[Bruce Weaver]

Yes. You could use such a variable as the group variable for an unpaired t-test.

ttest y, by(group)

There are other issues you have not raised. Here are a few things that come to mind.

1. Do you have 6 univariate questions or one multivariate question? (I'm thinking of the Huberty & Morris 1989 article as I ask that.)
2. If you are going to use t-tests, I'd suggest using either the -unequal- or -welch- option, as the t-test will be sensitive to heterogeneity of variance, given such a large discrepancy in sample sizes.
3. The estimated mean differences with CIs would be more informative than the p-values from 6 t-tests. (These can be seen in the t-test output in the 'diff' row.)

HTH.