Condfstat and CGM2 interaction

[EdJeeOnGitHub]

Hi - big fan of the package!

There appears to be a bug where even with only one clustering variable the conditional fstat changes depending on which cmethod argument is used in felm. For example:

library(lfe)
#> Loading required package: Matrix
library(tidyverse)

set.seed(6)
fake_data <- matrix(rnorm(1000*5, 0, 1),
                    nrow = 1000,
                    ncol = 5) %>% 
  as_tibble() %>% 
  mutate(group = factor(rep(1:5, 200)))
#> Warning: `as_tibble.matrix()` requires a matrix with column names or a `.name_repair` argument. Using compatibility `.name_repair`.
#> This warning is displayed once per session.


lfe_fit_cgm <- felm(formula = V1 ~  1| group | (V2|V3 ~ V4 + V5) | group,
                data = fake_data,
                cmethod = "cgm")
lfe_fit_cgm2 <- felm(formula = V1 ~ 1 | group | (V2|V3 ~ V4 + V5) | group,
                     data = fake_data,
                     cmethod = "cgm2")



condfstat(lfe_fit_cgm)
#>                 V2       V3
#> cluster F 1.945907 2.995397
#> attr(,"df1")
#> [1] 1
condfstat(lfe_fit_cgm2)
#>                 V2        V3
#> cluster F 1.945907 0.0120056
#> attr(,"df1")
#> [1] 1

My understanding is that with only one clustering dimension cgm and cgm2 should return identical results?