Linear models were the only things that were used for a very long period. But as it’s true of all statistical analysis, if you make assumptions that are not true, your conclusions aren’t going to be true. When you use a linear model in a moderate or mediator analysis, what it means is that for every value of the target, whether it’s a treatment or a risk factor, whatever, that the relationship between the outcome and the moderator, mediator is linear. Similarly, for every value of the moderator and mediator, if there are more than two targets, that’s also linear.
The moderators try to tear apart the population into subgroups where the response to treatment for example, or the effects are much the same. So it’s a process of dissecting the population. The mediators are dissecting the treatment. What components of the treatments seem to be the ones that account for the differences here? If we could do that, we could improve the treatment. So there are two different ways of dissection. But the proving of the causality is a new study in which you are randomly manipulating whatever causal factor you’re hypothesizing.
To learn more from Dr. Helena Kraemer listen to the podcast episode below.