We work in phylogenetic project to create a new graphical toll to investigate temporal signal of various evolutionary scales.
In our work, we have a question need to your help :
Do we need to perform any kind of multiple testing correction test (e.g. Bonferroni adjustment) when doing
multiple linear regression tests on points being phylogenetically linked ?
For instance, in the below phylogenetic tree (root2tip.png), we would perform 4 different linear regressions between genetic and temporal distances at nodes 1, 2, 3 and 4 with points highlighted in blue, orange, green and red, respectively. Some datatests would be « nested » between each others (1 vs 2,3&4, 3 vs 4) while the others are not (2 vs 3&4).
I had a similar issue with a paper we just published. The short answer is yes because of the likelihood of chance significance with multiple testing. I had analyzed our data using hurdle regression with multiple predictors. The reviewers insisted on multiple correction for the each of the predictor significance values. We used a modified Bonferroni correction (False Discovery Rate - Benjamini and Hochberg, 1995) and the reviewers were satisfied.
Please see
https://stat.ethz.ch/R-manual/R-devel/library/stats/html/p.adjust.html
https://rcompanion.org/rcompanion/f_01.html
http://www.biostathandbook.com/multiplecomparisons.html
The above uses the free r software to do the correction, an example of the r script is given.
to download r
https://www.r-project.org/
General info
https://www.youtube.com/watch?v=K8LQSvtjcEo
Multiple testing correction adjusts the individual p-value for each variable to keep the overall error rate to less than or equal to the user-specified p-cutoff value. Now, in my opinion, you can only perform it once you are testing the statistically significant difference between means of the tested variable. In doing multiple linear regression I am doubtful if there is a need to perform multiple testing corrections. Or shortly, I would say there is no need to perform. I am proud to accept your reactions to this comment of mine.
Yes, I know. I guess the issue is whether, in your model, you are making specific predictions for your variables. If you are, then you probably don't need to control for multiple testing.
- Badges
- Science topic
- Similar topics
- Human-Computer Interaction
- Collaboration