Re: st: Re: Problems with -hetgrot-

[Richard Williams <Richard.A.Williams.5@nd.edu>]

At 06:14 PM 3/14/2004 -0600, Scott Merryman wrote:
> Richard,
>
> I believe there is an error in Greene's text.  As you reported, Greene 
> gives the
> LR statistics as 120.915.  However, if you compute this by hand, given the
> individual sigma^2, the result is:
>
> . disp 100*ln(15708.84) - 20*(ln(9410.91) + ln(755.85) + ln(34288.49) +
> ln(633.42) + ln(33455.51))
> 104.41512
>
> which is what you and I got.
I did the same thing!  But if he has an error, I don't think that is where 
it is.  In Greene, 4th edition, p. 597, he says that iterated fgls (the 
Model 3 I presented) can be used to compute the LR statistic.  But, he also 
says that "If only least squares results are available...[they may be 
used], possibly with some loss of power in small samples."  That is what 
your above calculation does; it uses the least squares estimates rather 
than the ML estimates.

What Greene apparently does is use the estimate of sigma from the OLS 
estimates (15708.84) with the ML estimates for the group sigmas.  The 
calculation (using the #s reported on p. 598_ is

. disp 100*ln(15708.84) - 20*(ln(8657.72) + ln(175.80) + ln(40210.96) + 
ln(1240.03) + ln(29825.21))
120.92892

What I don't understand is why he is using the 15708.84 from the least 
squares estimates, rather than using a ML estimate of sigma.  This would 
seem to be inconsistent with what he says on the middle of p. 597.

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/