Replace part of the output of lm() function in Base R-CodePudding

Object res1 is a regression function from the metafor package. Object res2 is a regression function from the stats R base.

I was wondering if it might be possible to get the vcov() of res1 and substitute it for vcov() of res2?

Below is my unsuccessful solution.

# An Example:
library(metafor)

dat2 <- escalc(measure="OR", ai=waward, n1i=wtotal, ci=maward,
n2i=mtotal, data=dat.bornmann2007)

res1 <- rma.mv(yi ~ 0 type, vi, random = ~ 1 | study/obs, data=dat2)

res2 <- lm(yi ~ 0 type, data = dat2)

vcov(res2) <- vcov(res1) ## apparently this won't work!

CodePudding user response：

You don't say how you are planning to use the result of the vcov() call, and that's crucial. If all you are doing is calling vcov() and you want a different result, simply put the metafor object as an attribute of the lm object, assign it a new class (say "lmPlusMetafor"), and define a vcov.lmPlusMetafor method to extract the result from the metafor part.

Here is the code to do this:

library(metafor)

dat2 <- escalc(measure="OR", ai=waward, n1i=wtotal, ci=maward,
n2i=mtotal, data=dat.bornmann2007)

res1 <- rma.mv(yi ~ 0 type, vi, random = ~ 1 | study/obs, data=dat2)

res2 <- lm(yi ~ 0 type, data = dat2)

attr(res2, "metafor") <- res1

class(res2) <- c("lmWithMetafor", class(res2))

vcov.lmWithMetafor <- function(object, ...) vcov(attr(object, "metafor"))

Now when you run vcov(res1) or vcov(res2) you'll get the same result.

However, this is probably not good enough for anything useful. For example, if you now run summary(res2), I think you'll see standard errors etc. based on the original values, not the metafor ones.

If you want everything to work, you'll need a lot more changes: you'll need to see how the things you want are calculated, and work out how to do them with this new "lmWithMetafor" object. It won't be easy.

CodePudding user response：

I am not sure if this will satsify what you need, but assigning the function vcov to an object may help?

vcov_1 <- vcov(res1)
vcov_2 <- vcov(res2)   

vcov_1 <- vcov_2

> vcov_1 <- vcov(res1)
> vcov_1
               typeFellowship    typeGrant
typeFellowship   0.0018436986 0.0000618297
typeGrant        0.0000618297 0.0014597749

> vcov_2 <- vcov(res2) 
> vcov_2
               typeFellowship  typeGrant
typeFellowship    0.008567784 0.00000000
typeGrant         0.000000000 0.00556906


> vcov_1 <- vcov_2
> vcov_1
               typeFellowship  typeGrant
typeFellowship    0.008567784 0.00000000
typeGrant         0.000000000 0.00556906