When to use random effects

We know that the studies in our systematic reviews represent different people in different circumstances, so you should almost always use the random effects model to account for those differences.

That means you’ll need to weight your studies by two types of precision:

1. Precision of the difference between the “true effect” in one study and the actual results in that study

2. Precision of the difference between the “true effect” in one study and the “true effect” in another study.

We already know that the first type of precision is either reported by each study or easily calculated using a study’s standard deviation.

As for the difference between each study’s “true effect” and the average true effect, computer programs will help you with these calculations.

A word of warning- although the random effects model takes into account that each study has a unique truth, it’s not robust enough to combine wildly different studies. It still assumes that combining studies into an overall average effect is meaningful.

So, when you’re conducting a meta-analysis, don’t assume that the Random Effects model can account for all differences between studies. You still have to use common sense and ask yourself if your studies should be combined.