I'm running 2sls and making sure that my results replicate in both stata and r. In r, using felm, my code is:
a1 <- felm(y ~ 0|year + id | (x ~ z1 + z2 + z3)|cluster, data = db)
summary(a1$stage1)
In STATA, using ivreghdfe, my code is:
ivreghdfe y (x = z1 z2 z3), absorb(id year) cluster(cluster) first
The coefficients and standard errors generated by the code look the same in both approaches, but the two commands give me different F statistics, even while assuming the same degrees of freedom. ivreghdfe provides much more detail on which F statistics are being calculated and provides a number of ways to test instrument strength, but none of them correspond to the F statistic reported by R (neither the "projected model" or "full model"). My question is twofold: What test does the FELM F Statistic correspond to, and why does it differ from the statistics reported by STATA?
cmethod = 'reghdfe'
option to thefelm
. $\endgroup$ – Dimitriy V. Masterov Aug 28 '20 at 20:16