0

When I want to specify my own robust standard errors in a regression command like plm, how do I do this directly without running several functions?

library("plm")
library("lmtest")
Data <- iris
Data$time <- as.numeric(rep(1951:2000,3))

here is what I would run in plm:

regression <- plm(Sepal.Length ~ Sepal.Width + Petal.Length, 
    data = Data, 
    index = c("Species", "time"), 
    model = "within", 
    effect = "twoways")

now say I'd like a covariance matrix which clusters at the individual (species level):

results <- coeftest(regression,vcov=vcovHC(regression,type="HC0",cluster="group"))

My question is how I can include these standard errors directly in plm without having to run it first. This is not a big deal as the covariance matrix is calculated in a separate step anyways but would feel more elegant.

Jakob
  • 1,325
  • 15
  • 31

2 Answers2

2

In the plm-world, the covariance matrix is always passed as an additional argument, e.g. to summary(). But you can try the following: In the plm object, overwrite the vcov that is already there:

regression$vcov <- vcovHC(regression, type="HC0", cluster="group")

But no guarantee this does not break anything else if you apply various functions to your plm object later.

You might also want to look at the summary.plm object created by summary.plm, e.g.

res <- summary(regression, vcov = vcovHC(regression, type="HC0", cluster="group"))

Also, you might want to look at ?summary.plm (doc for summary.plm) how to pass the vcov function to summary (the example above passes the matrix the function creates).

Helix123
  • 3,502
  • 2
  • 16
  • 36
1

If this is a question about elegancy, try packages dplyr, (or pipeR or magrittr) for a 'pipe' operator. When writing functionA(functionB((x) or 

y <- functionB(x)
z <- functionA(y)

you can use a great alternative as such:

z <- x %>% functionB %>% functionA

In your case, this would make:

results <- 
plm(Sepal.Length ~ Sepal.Width + Petal.Length, 
    data = Data, 
    index = c("Species", "time"), 
    model = "within", 
    effect = "twoways") %>% 
coeftest(., vcov=vcovHC(.,type="HC0",cluster="group"))

in which you use . to explicitly mark that you're inputting what is before the %>% pipe operator.

Hendrik D
  • 71
  • 7
  • Thats quite nice, thanks! (p.s.: typos in my question corrected, should also be Sepal.L(!)ength in your answer). The downside is that I only get the result of the second function in `results`, correct? – Jakob Apr 01 '16 at 09:42
  • The problem with your answer is that I can't use this version inside another command where plm is an argument. Do you know how I can include the covariance matrix inside plm? Your code does one after the other – Jakob Apr 01 '16 at 10:26