T-tests: Comparing Change Scores

[Anthony Crimarco]

Hello everyone,

This will probably be a rudimentary question, but I need clarification none the less:

I am doing t-tests comparing nutrition habits between two groups: a Standard group and a Restaurant group. I am comparing the number of meals that each group had before and during the study. I did paired t-tests comparing the changes in meals consumed by each group from baseline and the study period. A summary of each is posted below:
Standard Group:

Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval]

Baseline 15 6 1.679002 6.502747 2.398899 9.601101
Follow-up 15 13.33333 3.026339 11.72096 6.842481 19.8241
diff 15 -7.333333 1.88646 7.306227 -11.37939 -3.28728

The Standard group decreased the number of meals consumed at a restaurant during the last 3 weeks by -7.3±7.3 meals during the study period (P=0.0016).

Restaurant Group:

Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval]

Baseline 15 17.66667 4.08326 15.8144 8.908944 26.42439
Follow-up 15 16.93333 5.041227 19.52459 6.120977 27.74569
diff 15 -0.7333333 3.919994 15.18207 -9.140883 7.674217

The Restaurant group decreased the number of meals consumed at a restaurant during the last 3 weeks by -0.7±15.2 meals, which was not statistically significant (P=085).


I then did the two-sample t-test to get group comparison means for the meals consumed during the study period. However, my adviser told me that she wants me to compare the change scores between the groups and not just the average scores at follow-up. I am not sure how to do this. My understanding is that the change scores are those values in tables I posted above (i.e. the "differences" section in the table). But I am not sure how to do the between-group comparison for those specific change scores.I know how to simply do the between-group comparisons at either baseline or follow up etc.

Does anyone have any suggestions or insights? Your help is greatly appreciated.

[Dave Airey]

I think your advisor is asking you to compute differences between meals before and after, and then compute an unpaired ttest on the differences between groups. It would be better to use all the data in one model like a repeated measures ANOVA or mixed model though.

[Clyde Schechter]

I'm confused by your outputs. The separate analyses within each group appear to be paired t-tests: the bottom line says diff, and there is no "combined" output row. And the numbers of observations are consistent. But with the paired t-test, the entries in the "Group" column of the output are the names of the variables, and Follow-up is not a legal Stata variable name, because of the hyphen. If you did an unpaired ttest of two different groups, then Group would contain the value labels of the group variable, and those could contain hyphens. But then the Obs for the diff row would be left blank and there would be a row called "combined" with obs equal to the total of the obs in the Baseline and Follow-up rows.

So please show the exact code that generated this output, as it looks to me to be impossible.

In any case, I'm going to assume that what you have done so far is actually correct (more or less) and that you have two variables, one called baseline and the other called followup. To test for the difference in change you would do this:

gen change = followup - baseline
ttest change, by(group)

where group is a dichotomous variable that distinguishes the Standard Group from the Restaurant Group.

All of that said, the combination of the outcome variable being a count, the rather small N's, and a small mean with large standard error in the baseline mean for the Standard group suggest to me that t-tests may not be appropriate for this data in any case. But that's a longer story, and one you should take up with your advisor.

Added: Crossed with #2.

Also added: Please read the Forum FAQ. And pay special attention to #12 to learn how to use code delimiters to improve the readability of Results that you post here.

[Anthony Crimarco]

Thank you both of you!

Clyde's code worked. That was exactly what I was looking for.

Sorry about the output issue. This is my first time on the forum. I saw that my tables didn't format correctly. And yes to clarify I had only posted the paired t-tests.