2

I am performing a meta-regression of nine studies that look at cancer risk vs level of substance in the drinking water. Each study has a reference strata and one to five higher strata. The studies have quite different ranges of exposure. I have run the glst program in STATA which is a random slope model, and I have run the Meta-dose program in SAS which is a random intercept model. My results for fixed models are identical; my results for random model are quite different. One shows a highly significant positive coefficient and the other shows a non-significant coefficient. Do I want the random slope model or the random intercept model, and Why? What are the differences in interpretation? Is there a model in either program that is both random slope and random intercept? Why would i want to do that, and how would I do it?

Thanks, Steve L

amoeba
  • 93,463
  • 28
  • 275
  • 317
Steven Lamm
  • 51
  • 2
  • Using a random intercept is a way to model static correlations between measures within group. A random slope is used if you think a covariate's linear affect randomly differs between groups – gammer Mar 24 '17 at 03:40
  • Maybe my answer to this question will give some rough idea: http://stats.stackexchange.com/a/235382/109647 – T.E.G. Mar 24 '17 at 05:53
  • See also https://stats.stackexchange.com/questions/120964/fixed-effect-vs-random-effect-when-all-possibilities-are-included-in-a-mixed-eff for discussion – Tim Jan 25 '18 at 10:03
  • Possible duplicate of [Fixed effect vs random effect when all possibilities are included in a mixed effects model](https://stats.stackexchange.com/questions/120964/fixed-effect-vs-random-effect-when-all-possibilities-are-included-in-a-mixed-eff) – mdewey Jul 31 '18 at 16:59

0 Answers0