Unexpected Behavior with vcov

[mgoplerud]

Hello,

I noticed some unexpected behavior with vcov. If you adjust any of the arguments, the other settings specified during the initial estimation are not inherited correctly. It seems to not be a problem with summary in the example I show below.

library(fixest)

res_pois <- fepois(round(Petal.Length) ~ Petal.Width + Sepal.Length | Species, iris, se = 'iid', ssc = ssc(adj = FALSE))
res_glm <- glm(round(Petal.Length) ~ Petal.Width + Sepal.Length + Species, iris, family = poisson())

# Target covariance
vcov(res_glm)[2:3,2:3,drop=F]
#>               Petal.Width Sepal.Length
#> Petal.Width   0.038988314 -0.005434063
#> Sepal.Length -0.005434063  0.006418182
# Uses correct pre-specified settings (agrees with GLM)
vcov(res_pois)
#>               Petal.Width Sepal.Length
#> Petal.Width   0.038988314 -0.005434063
#> Sepal.Length -0.005434063  0.006418182
# Overrides pre-specified arguments for ssc
vcov(res_pois, se = 'iid')
#>               Petal.Width Sepal.Length
#> Petal.Width   0.040063854 -0.005583968
#> Sepal.Length -0.005583968  0.006595235
# Overrides default setting for se/vcov: "se" is set to "cluster"
vcov(res_pois, ssc = res_pois$summary_flags$ssc) 
#>               Petal.Width Sepal.Length
#> Petal.Width  0.0037281535 3.720606e-04
#> Sepal.Length 0.0003720606 6.963743e-05
# If both are manually specified, agrees with provided settings
vcov(res_pois, se = 'iid', ssc = res_pois$summary_flags$ssc) 
#>               Petal.Width Sepal.Length
#> Petal.Width   0.038988314 -0.005434063
#> Sepal.Length -0.005434063  0.006418182

# Point estimates agree as expected
coef(res_pois)
#>  Petal.Width Sepal.Length 
#>   0.08475446   0.14744109
coef(res_glm)[2:3]
#>  Petal.Width Sepal.Length 
#>   0.08475446   0.14744109

# Not a problem with summary?
summary(res_pois)
#> Poisson estimation, Dep. Var.: round(Petal.Length)
#> Observations: 150 
#> Fixed-effects: Species: 3
#> Standard-errors: IID 
#>              Estimate Std. Error  t value Pr(>|t|)    
#> Petal.Width  0.084754   0.197455 0.429235 0.667752    
#> Sepal.Length 0.147441   0.080114 1.840401 0.065709 .  
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> Log-Likelihood: -235.1   Adj. Pseudo R2: 0.20737 
#>            BIC:  496.1     Squared Cor.: 0.941817
summary(res_pois, se = 'iid')
#> Poisson estimation, Dep. Var.: round(Petal.Length)
#> Observations: 150 
#> Fixed-effects: Species: 3
#> Standard-errors: IID 
#>              Estimate Std. Error  t value Pr(>|t|)    
#> Petal.Width  0.084754   0.197455 0.429235 0.667752    
#> Sepal.Length 0.147441   0.080114 1.840401 0.065709 .  
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> Log-Likelihood: -235.1   Adj. Pseudo R2: 0.20737 
#>            BIC:  496.1     Squared Cor.: 0.941817
summary(res_pois, ssc = ssc(adj = F))
#> Poisson estimation, Dep. Var.: round(Petal.Length)
#> Observations: 150 
#> Fixed-effects: Species: 3
#> Standard-errors: IID 
#>              Estimate Std. Error  t value Pr(>|t|)    
#> Petal.Width  0.084754   0.197455 0.429235 0.667752    
#> Sepal.Length 0.147441   0.080114 1.840401 0.065709 .  
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> Log-Likelihood: -235.1   Adj. Pseudo R2: 0.20737 
#>            BIC:  496.1     Squared Cor.: 0.941817

^{Created on 2022-09-18 by the reprex package (v2.0.1)}

[lrberge]

Hi, sorry for the delay. I agree the behavior was against what one expected. Now fixed.