Today I ran a fixed effects model in plm using unbalanced panel data with N>>T (N=5970 and T=10). Unfortunately the data is from a database that does not allow me to share it. However, my equation is the following:
pl <- plm(LS~FIAS+EXPT+EMPL+AV+TURN+FIPL+EBIT+FIEX, index=c("IDNR", "CLOSDATE_year"), effect = "twoways", model="within", data=data)
where LS is the wage share on micro level. The model runs smoothely and prints my results. I read in another paper that because of the structure of the data I have to adjust the standard errors by replacing them with Driscoll-Kraay robust standard errors. As far as I understand,
coeftest(pl, vocov.=vcovSCC)
should do the trick. The results are exactly the same however. So my question is: Am I performing the correction wrong or is plm automatically doing the adjustment?
Appendix:
> summary(pl)
Twoways effects Within Model
Call:
plm(formula = LS ~ FIAS + EXPT + EMPL + AV + TURN + FIPL + EBIT +
FIEX, data = data, effect = "twoways", model = "within",
index = c("IDNR", "CLOSDATE_year"))
Unbalanced Panel: n = 2718, T = 1-10, N = 12942
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-1.1183188 -0.0191726 -0.0003545 0.0160751 0.5363292
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
FIAS -4.8960e-02 1.0766e-02 -4.5475 5.490e-06 ***
EXPT -1.4202e-02 2.6751e-03 -5.3093 1.124e-07 ***
EMPL 5.3522e+03 3.0152e+02 17.7510 < 2.2e-16 ***
AV -2.6384e-01 1.0277e-02 -25.6739 < 2.2e-16 ***
TURN 5.9762e-02 2.8426e-03 21.0235 < 2.2e-16 ***
FIPL -6.6372e-01 5.0019e-02 -13.2692 < 2.2e-16 ***
EBIT -8.5062e-01 1.2782e-02 -66.5489 < 2.2e-16 ***
FIEX -6.8813e-01 8.4500e-02 -8.1436 4.285e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Total Sum of Squares: 81.846
Residual Sum of Squares: 41.643
R-Squared: 0.4912
Adj. R-Squared: 0.35479
F-statistic: 1231.5 on 8 and 10205 DF, p-value: < 2.22e-16
> coeftest(pl, vocov.=vcovSCC)
t test of coefficients:
Estimate Std. Error t value Pr(>|t|)
FIAS -4.8960e-02 1.0766e-02 -4.5475 5.490e-06 ***
EXPT -1.4202e-02 2.6751e-03 -5.3093 1.124e-07 ***
EMPL 5.3522e+03 3.0152e+02 17.7510 < 2.2e-16 ***
AV -2.6384e-01 1.0277e-02 -25.6739 < 2.2e-16 ***
TURN 5.9762e-02 2.8426e-03 21.0235 < 2.2e-16 ***
FIPL -6.6372e-01 5.0019e-02 -13.2692 < 2.2e-16 ***
EBIT -8.5062e-01 1.2782e-02 -66.5489 < 2.2e-16 ***
FIEX -6.8813e-01 8.4500e-02 -8.1436 4.285e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
UPDATE: I ran the same model with lm() and received the following output:
> l <- lm(LS~FIAS+EXPT+EMPL+AV+TURN+FIPL+EBIT+FIEX+factor(CLOSDATE_year)+factor(IDNR), data)
> summary(l)
Call:
lm(formula = LS ~ FIAS + EXPT + EMPL + AV + TURN + FIPL + EBIT +
FIEX + factor(CLOSDATE_year) + factor(IDNR), data = data)
Residuals:
Min 1Q Median 3Q Max
-1.11832 -0.01917 -0.00035 0.01608 0.53633
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 1.011e+00 3.316e-02 30.493 < 2e-16 ***
FIAS -4.896e-02 1.077e-02 -4.548 5.49e-06 ***
EXPT -1.420e-02 2.675e-03 -5.309 1.12e-07 ***
EMPL 5.352e+03 3.015e+02 17.751 < 2e-16 ***
AV -2.638e-01 1.028e-02 -25.674 < 2e-16 ***
TURN 5.976e-02 2.843e-03 21.023 < 2e-16 ***
FIPL -6.637e-01 5.002e-02 -13.269 < 2e-16 ***
EBIT -8.506e-01 1.278e-02 -66.549 < 2e-16 ***
FIEX -6.881e-01 8.450e-02 -8.144 4.29e-16 ***
factor(CLOSDATE_year)2009 6.487e-04 1.573e-02 0.041 0.967102
> coeftest(l, vocov.=vcovSCC)
t test of coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 1.0112e+00 3.3160e-02 30.4930 < 2.2e-16 ***
FIAS -4.8960e-02 1.0766e-02 -4.5475 5.490e-06 ***
EXPT -1.4203e-02 2.6750e-03 -5.3093 1.124e-07 ***
EMPL 5.3522e+03 3.0152e+02 17.7510 < 2.2e-16 ***
AV -2.6384e-01 1.0277e-02 -25.6739 < 2.2e-16 ***
TURN 5.9762e-02 2.8426e-03 21.0235 < 2.2e-16 ***
FIPL -6.6372e-01 5.0019e-02 -13.2692 < 2.2e-16 ***
EBIT -8.5062e-01 1.2782e-02 -66.5489 < 2.2e-16 ***
FIEX -6.8813e-01 8.4500e-02 -8.1436 4.285e-16 ***
factor(CLOSDATE_year)2009 6.4867e-04 1.5728e-02 0.0412 0.9671020
As the coefficients are the same, but the SEs get updated that this suggests that indeed plm is directly estimating robust Standard Errors, right?
