I am running a glm, with a binomial distribution and a small sample size (in R using the glm function). Through forums, I have noticed that it is stated that, the p-value is a good approximate if the sample sizes are large, and that if we are dealing with small sample size, it is better to use resampling to do so. Some suggest using the function manyglm, as it calculates the p-value based on resampling by default. Is there any reference for this? Is it still ok to use manyglm for univariate analysis? Is it the same?
If not using manyglm, what techniques exist for obtaining p value by resampling?
Thanks so much for your help in advance.
Do you mean the function in the package mvabund? If so, there is information here:
https://rdrr.io/cran/mvabund/man/manyglm.html
and in
Wang Y., Neuman U., Wright S. and Warton D. I. (2012). mvabund: an R package for model-based analysis of multivariate abundance data. Methods in Ecology and Evolution. online 21 Feb 2012.
ps. You might describe your problem more. Are you just trying to run a model with a more robust standard error estimation (and thus p value)? Just wanting bootstrap estimates for a regression model? etc. I think manyglm is designed for different situations than what you want, but I'm not sure exactly what you are after.
Hi Daniel, thank you for your response. I did just want to run a model to get a more robust p-value. I am familiar with the mvabund package, and manyglm function (and have used it previously with multivariate data). I just wanted to know if the results (p-value) fo the manyglm model ran on only 1 response variable (so not multivariate) is equally or better valid than running a glm.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size