In the context of health studies, I have to explain to non-statistician colleagues the difference between performing a one-tailed test at alpha=5% and a two-tailed test at alpha=10% when comparing two samples.
Indeed, being able to present a lower alpha for the apparent same statistical power can be falsely appealing for someone without the comprehension of the concept behind it.
In my comprehension, we should only use one-sample tests when the difference can "physically" only go in one direction (CMIIW). However, when comparing groups, I cannot figure out an instance where it is physically impossible for the value to be, say, higher in the control group.
Reading the numerous material here on SSE (the best one being this one to me) further supports my opinion that performing one-tailed tests is hardly ever the right thing to do in this case.
However, when explaining to non-statisticians (or maybe students), I feel that giving an example of what would be a good use case of one-tailed tests will help them grasp the concept better.
Therefore, using the simple example of two-sample tests such as t-tests or log-rank tests, what would be a legitimate comparison case where a one-tailed test is recommended?
If the answer can be health-related, this is a bonus.