<p>A couple of useful links</p>
- Fixed effects assumes a single true value imperfectly measured due to sampling error by different studies
- Random effects assumes no single true value but a range of different effect sizes so that the variation in effect size observed between studies is due to both sampling error and due to the underlying variation. I currently think of this along with heterogeneity of treatment effects, and imagine that the treatment must interact with different population characteristics such that its efficacy varies.
These approaches are used to assign weights to the studies. With FE, the bigger the study, the greater the weight. With RE, the number of studies available is crucial since we can only understand the variation in the effect distribution if we sample that distribution more often.
Also see notes on treatment heterogeneity by control event rate.
Schmid, CH 1998 An empirical study of the effect of the control rate as a predictor of treatment efficacy in meta-analysis of clinical trials. Stat Med 10 Mar 2019 10:22 824 74363 0 3635058121 ↩