Originally Posted by
jfsully
OK, I am not a statistician by a long shot. As a physician, I think I am reasonably good at critically reading studies and basic stats, but I may not have the jargon down, and I may misunderstand even some basic stats. When I refer to a trend, I am saying that a study calculated, say, an odds ratio that was not 1, but the confidence interval included 1. Sometimes it is close enough to 1 that it seems clear there's nothing there, and we should embrace the null hypothesis. Other times, it's clear the results are diluted by study limitations. When I was on an IRB, we had a resident statistician who would frequently point out that a proposed study was underpowered to test the hypothesis they wanted to test. So they were never going to disprove the null hypothesis, whatever they found. Let's say their hypothesis was correct, but they were unable to clearly demonstrate it due to the study design. In that case they may find a "trend" toward their hypothesis being true, but their CI may still include the null hypothesis. If study finds this and was well-powered and properly designed, you're probably done. If not, the "trend" may be a signal that there may be something useful there to retool and study again.