[Limdep Nlogit List] Panel bootstrap with fixed effects models

[Eaton, Derek]

Dear all,

I would appreciate any guidance in understanding how LIMDEP's panel bootstrap works in a fixed effects model. Using the panel bootstrap, each bootstrapped sample will typically have a number of individuals (strata) appearing more than once (over all of their time periods of course). But the fixed effects commands use a specification ;Pds=  to identify each individual. These will no longer be unique for any individual appearing more than once in the sample. The procedure does not seem to break down, but I am not sure what it is exactly doing.

In the case of the unconditional 2-way Poisson FE model, I get coefficient standard errors from the bootstrap that are about 100 times larger than those of the unconditional estimator (admittedly with only 100 replications). Could this be related to how the estimator is handling replicated values for ; Pds ?

I have used the following procedure, based on reading section R15.5 of the LIMDEP Version 9.0 manual:

MATRIX		; bb = Init(vars,100,0.0) $ vars is the number of coefficients estimated
PROC
	 DRAW		; N=699 ; Pds=ni ; Rep $ 
	 POISSON	; LHS = y  ; RHS = x
			; Fem ; Time=PD ; Pds=ni $
	 MATRIX	; bb(*,j) = b $
	 DRAW		; N = 0 $ 
	ENDPROC
EXEC		; j = 1, 100 $

Thanks!
Derek Eaton
-+-+-+-+-+-+-+-+-+ 
Derek Eaton 
Researcher, International Trade & Development 
Agricultural Economics Research Institute (LEI) 
part of Wageningen University & Research Centre 
  
t: +31-(0)70-335 82 43 
f: +31-(0)70-335 81 96 
p: P.O. Box 29703, The Hague, 2502LS, The Netherlands 
www.lei.nl/uk 
derek.eaton AT wur.nl