2

I'm using the sumlog function in the metap package of R, and have a few questions about the nuts-and-bolts of this function:

1)Some of my p-values are reported as '<0.001'. How do I include these? use 0.001?

2)A few studies in my review don't give a p-value and only state 'significant' or 'non-significant'. Is there any way to incorporate these?

3)I'm interested to know what this function is actually doing - as a opposed to simply averaging the p-values.

Many thanks for all the help!

Keno
  • 43
  • 3
  • Per what it's doing: https://en.wikipedia.org/wiki/Fisher%27s_method https://en.wikipedia.org/wiki/Extensions_of_Fisher%27s_method – alan ocallaghan Feb 23 '20 at 13:13
  • I think for p-values that are reported as <0.001, using 0.001 is valid and conservative. For "sig/non-sig" I don't think it's valid to include these. – alan ocallaghan Feb 23 '20 at 13:14

1 Answers1

2

The conservative thing to do is to use the largest value which would have been counted as significant so 0.05 is you think they used that as their cut-off. If they state <0.001 then that would be the 0.0005 as anything like 0.0006 would have rounded to 0.001.

Non-significant values represent more of a challenge as the rule above corresponds to using 1 which is very conservative. However you might want to explore truncated Fisher's method which is available in the package you are using as a wrapper around the package TFisher. This uses a subset of the $p$-values defined by a cut-off.

The vignette gives more details about the properties of the method and compares it with the others in the package.

Disclaimer: I am the author of the package you are using.

mdewey
  • 16,541
  • 22
  • 30
  • 57