Ken Blake, Ph.D.

Comparing the averages of two variables: The two-sample t-test.

A televised debate can be a make-or-break event for a political candidate. In this video, we look at examples of two different types of experiments that test whether watching a candidate participate in a televised debate tends to boost approval of the candidate. All data are made-up.

In the first example, you recruit 60 people, randomly divide them into two groups of 30. You let one group watch a candidate perform in a televised debate while, elsewhere, you let the other group watch the candidate sit for an in-depth television interview by a journalist. Afterward, you measure and average each group's approval of the candidate, then use an independent-samples t-test to determine whether the two group averages are nonrandomly different.

The second example takes a different approach. In this approach, you recruit 30 people, measure their approval of the candidate, then let them watch the candidate perform in a televised debate, then measure their approval of the candidate a second time. Once that's done, you average the pre-debate approval ratings, average the post-debate approval ratings, then use a paired-samples t-test to determine whether the two averages are nonrandomly different.


Note the key difference between the two types of experiments. In the first, there was no connection between the people in the two groups. That's when you want to use an independent-samples t-test. In the second, there was only one group of people, and each group member's approval was measured at each of two different points in time. In that situation, you need to use a paired-samples t-test.

Want to practice what you saw in the video? Here's the dataset.