Give three ways to deal with possible Type I error. Give an example for each and explain how we might address the problem.

Question

What will be an ideal response?

Answer 1

Ans: A Type I error occurs when we claim that we have a significant result, and we do not. That is, we say there is a difference between our groups when there really is not. There are a number of reasons this might happen. First, we might reduce the alpha level to make it harder to make a mistake in that direction, but doing so increases the likelihood of a Type II error. Second, we might switch from a one-tailed to a two-tailed test. A two-tailed test makes it harder to find a significant difference because you are considering an alpha level half of that of a one-tailed test. Third, if we conduct multiple analyses, we are increasing the likelihood of finding something significant—it is sort of like throwing the bean bag in corn hole . . . you would guess that the more times you try, the more likely you are to hit the target. To address this problem, we might make the alpha level more stringent—that is, we might make it more difficult to find a significant result. We might also use a Bonferonni adjustment and divide the alpha level by the number of statistical tests we are using. In conclusion, most researchers conducting behavioral science research are not as concerned with this as they might be, but for others in fields like medicine where it is crucial that we do not say a medicine works when it does not—a Type I error could be critical.