Lingyu Kong Lingyu Kong - 17 days ago 4x
R Question

Waldtest in R to get adjust F statistic with plm and result shown with stargazer?

I am working with an unbalanced short panel.
Raw data: bankFull.xlsx

What I actually want is only get the regression results with two side fixed effects and robust S.E reported, which is very easy in Stata. I followed online tutorial but ran into some problem always with

# Adjust F statistic
wald_results <- waldtest(FE1, vcov = cov1)
Error in model.matrix.pFormula(formula, data, rhs = 1, model = model, :
NA in the individual index variable

no matter how I adjusted the data! It almost drives me crazy.

here is my code:

bankFull <- openxlsx::read.xlsx("bankFull.xlsx",1)


FE1 = plm( RoA ~
data = bankFull, index = c("name","year"),
effect="twoways",na.action = na.omit,
model= "within")

# robust S.E.-----------
library(lmtest) # waldtest; see also coeftest.

# Adjust standard errors
cov1 <- vcovHC(FE1, type = "HC1")
robust_se <- sqrt(diag(cov1))

# Adjust F statistic
wald_results <- waldtest(FE1, vcov = cov1)

# show results. how can I get the F value?
stargazer(FE1, FE1, type = "text",
se = list(NULL, robust_se),
omit.stat = "f")

Secondly, as the code shown, I use stargazer to demonstrate the results. I also need the adjusted F value to be shown in the table. Is there any option in the package that I can use?


Use the development version of plm to get a robust F-test for your model (from Example:

data("Grunfeld", package = "plm")
mod_fe <- plm(inv ~ value + capital, data = Grunfeld, model = "within")
plm::Ftest(mod_fe, test = "F")

# with robust vcov
plm::Ftest(mod_fe, test = "F", .vcov = vcovHC(mod_fe))
plm::Ftest(mod_fe, test = "F", .vcov = function(x) vcovHC(x, type = "HC3"))

To feed the robust values (robust standard errors, t- and p-values, F-value associated p-value) use arguments se, t, p and for the F test simply add.lines of the stargazer command (and omit the F statistic generated by stargazer by default). Here is a full example for what you want: (section "Robust standard errors (replicating Stata’s robust option)").