Hello,
I'm writing a meta-analysis of prevalence in R. Using the influence() function I determined that two of the included studies are outliers. I would like to check if the outcome of the rma() function after removing the two outliers is significantly different from the outcome of the same function when all studies are included. Is there a simple way to do this in R?
Thanks in advance for any help.
dear Krzysztof,
I think your question is mainly related to the general use of R rather than meta-analysis specifically. The influence.rma() function already provides various diagnostic outputs for each individual studies quantifying the effect that the exclusion of the study would produces. Irrespective of whether this an accepted thing to do in meta-analysis, if you want to fit a model without some specific studies you probably have to subset your main dataset to filter out undesired observations (you can use the dplyr package which is quick and intuitive to use: https://dplyr.tidyverse.org/reference/filter.html).
Dear KB,
Be careful here! Please distinguish between two situations: (a) the dubious study gives a deviant result (i.e., gives a deviant estimate of your target estimand); (b) the dubious study has deviant collateral characteristics (covariates, care of execution, sponsors, ...).
In situation (a) the verdict 'this study is an outlier' is logically the same as 'in/exclusion of this study makes a change in the final estimate', as both verdicts portray rejection of the same null hypothesis. (No second P-value needs quoting.)
Anyhow, with two supect studies (your situation), please handle the 2X2 accept-reject possibilities carefully.
Always consider carefully the reason of the different results especially in the case where you have to outcomes with more studies pointing to both outcomes. Even if only a few studies point to a certain outcome these studies might still be the correct ones.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis