search

richarddmorey / BayesFactor

BayesFactor R package for Bayesian data analysis with common statistical models.
https://richarddmorey.github.io/BayesFactor/
131 stars 48 forks source link

Meta-analytic t tests - mix one- and two-sample #122

Closed DominikVogel closed 5 years ago

DominikVogel commented 5 years ago

meta.ttestBF() can be used to get a Bayes Factor for a series of "one- and two-sample designs." I assume this refers to paired and independent t-tests.

My question is if there is a way to mix t values from one-sample and two-sample designs. Let's assume I have four studies. Three of them use a two-sample design, and one uses a one-sample design. Is there a way to combine those four into a meta-analytic t test?

P.S. I also posted the question on StackOverflow. Sorry for cross-posting.

richarddmorey commented 5 years ago

There's nothing that says you can't do this, but the assumption is that the (standardised) effect size is the same across all the studies. Given the differences between paired and independent designs --- the paired design is going to have less variability because the participant variability is effectively subtracted out --- this assumption is very likely to be wrong. It will lead to a misleading quantification of the evidence.

DominikVogel commented 5 years ago

Hi Richard,

thank you very much. I did not think about it from that perspective. Totally makes sense.

Best, Dominik