If I have a drug purported to lower heart rate and I give it to a group of volunteers, I would conduct a paired t-test to determine if their HR before is lower than their HR after taking the drug.

I find significant results, so now I accept the premise that the drug lowers HR.

Now, what if I conduct the same experiment, except I have two groups to whom I give the drug. One of these groups is different in one way (say, overweight) and the other group is not. Otherwise, they are completely controlled.

If I want to test that the decrease in heart rate is more significant in group A compared to the decrease in heart rate in group B (i.e. group A's HR dropped more than group B), I would use an unpaired t-test. Is that correct the correct usage of a paired vs. unpaired t-test?

  • $\begingroup$ You do the same experiment, measuring HR pre- and post-treatment, yes? To compare the effect in both groups, you can measure and compare effect sizes $\endgroup$ Commented Nov 5, 2018 at 18:47

1 Answer 1


Let's write down explicitly what the groups are.

V1a: Before group

V1b: After group

V2a: Regular weight group, before treatment

V2b: Regular weight group, after treatment

V3a: Overweight group, before treatment

V3b: Overweight group, after treatment

In situation #1, you are comparing V1a and V1b. These groups contain the same people, so a paired test, such as the paired t-test, would be appropriate.

In situation #2, you are comparing V2b-V2a and V3b-V3a. Let's call these V4 and V5, respectively. Then you have a basket of data, V4, on subjects of regular weight, and a basket of data, V5, on subjects who are overweight. This is a situation to do two-sample testing, such as the unpaired t-test.

If you're doing situation #2, a technique worth reading about is called difference in differences.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.