Here are a couple of readable articles about some of the finer points of statistics. We don't need to know how to do the mathematics but we need the take-home messages.
The first article, on meta-analysis, explains when to use a fixed effects model and when a random effects model. The short answer is - use fixed effects if all the studies are the same (e.g. same drug, same dose, same sort of patients etc.) If they are different (e.g. 15 different ways of trying to get people to exercise more) use random effects. With a fixed effect model the meta-analysis gives you the (single) common effect of the various studies; if random effects you get the average of the (many) effects.
The other one has a brilliant graph which demonstrates why it's important not to slice your data up. The focus is correlation coefficients, but as the authors point out right at the end, why use a correlation coefficient when you could calculate a regression coefficient instead? The correlation coefficient was invented 100 years ago. Time to move on!
Comments