Question: Limma without eBayes, is it indistunguishable from Ordinary Least Square?
gravatar for CantExitVIM
8 weeks ago by
CantExitVIM10 wrote:

One post has caught my attention (, which highlights how to "skip" the eBayes function. First, I understand that Bayes methodology is why limma is so popular and the benefits of utilizing such methods. But if one calculated the t-values and p-values as specified (code below), is there anything that distinguishes the results from ordinary least square if lmFit(method='ls") is used? In other words, if I wanted to compare empirical bayes results  to ordinary least square results, would this be an appropriate way to examine the differences?


How to calculate t-value and p-values manually:

fit2$t <- fit2$coef/fit2$stdev.unscaled/fit2$sigma
fit2$p.value <- 2 * pt(-abs(fit2$t), df = fit2$df.residual)
ADD COMMENTlink modified 8 weeks ago by Gordon Smyth35k • written 8 weeks ago by CantExitVIM10
gravatar for Gordon Smyth
8 weeks ago by
Gordon Smyth35k
Walter and Eliza Hall Institute of Medical Research, Melbourne, Australia
Gordon Smyth35k wrote:

Yes, you can do a regular t-test using the code you give instead of an empirical Bayes moderated t-test. The code you give is from page 61 of the limma User's Guide. Yes, the code will give you same results that you would get from any statistical package able to fit a similar linear model.

Just a note on terminology though. limma always does least squares (either weighted or unweighted) and empirical Bayes doesn't change that. It is at the test statistic stage that the empirical Bayes comes in.

ADD COMMENTlink written 8 weeks ago by Gordon Smyth35k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 356 users visited in the last hour