Two bug fixes for T=3 panels.

[mronkko]

Four very minor code changes that only apply to T=3 panels. Fixes #73

[tappek]

Much appreciated, thank you for this! Will look at this more closely. I think a reproducible example for the fixes in pgmm is given by what you put in #73 while setting transformation = "ld":

library(multilevel)
library(plm)
data(univbct)

bliese_panel_data <- pdata.frame(univbct, index = c("SUBNUM", "TIME"))

est <- pgmm(
     JSAT ~
         plm::lag(JSAT) + plm::lag(COMMIT) + plm::lag(READY) | plm::lag(JSAT, 2),
     data = bliese_panel_data,
     effect = "individual", # Only individual fixed effects, no time effects
     model = "twostep",
     transformation = "ld"
)
summary(est)

[mronkko]

Yes, that is a reproducible example. I was planning on writing a bug report, but then realized that this was easy to fix (just two lines of code) and submitted a pull request instead.

[mronkko]

The code that you posted actually does not run with my patch because the summary functions fails for the ld model. The patch that I wrote addresses 1) Getting the ld model to estimate and 2) Getting the summary of d models to work when T=3. The logic for when to run mtests for ld models is not immediately obvious to me. The mtest is currently run as a part of summary for T=3 ld models even if calculating these tests is impossible.

[tappek]

It seems like this changes some results, also for coefficients, in the test suite. Need to investigate more closely.