Federica Maiorano <[email protected]> seems to be having some trouble
with constraints:
> I am estimating a translog cost function + 1 input
> share cost equation using mysureg.
>
> When I attempt to estimate a constrained model, then I
> get an error message from mysureg. As an example, I
> copy below what happens with one constraint only.
>
> Please note that there do not seem to be any problems
> when the same commands are performed using sureg.
>
> I am obviously doing something wrong, but I cannot
> work out what it is. Any suggestions?
>
> -----------------------------------------------
> constraint define 1 [sl]_cons = [lnvc]lnpl
>
> mysureg (lnvc lnk lnpl lnpm lnaccess lncalls time k2
> pl2 pm2 acc2 call2 t2 kpl kpm kacc kcalls kt plpm
> placc plcalls plt pmacc pmcalls pmt acccalls acct
> callt) (sl lnpl lnpm lnaccess lncalls lnk time),
> constraint(1) cluster(idcomp)
>
> Fitting constant-only model:
> Constraints invalid:
> [lnpl] not found
>
> -----------------------------------------------
Aside from being an example estimation command in the Stata Press book
"Maximum Likelihood Estimation with Stata, 2nd Edition", there does not seem
to be a publicly available user-written estimation command called -mysureg-.
So I'll assume that Federica is using -mysureg- introduced on pg. 238 (the
ado-file for -mysureg- is given on pg. 286).
Federica discovered a bug in -ml model- that was fixed for Stata 9 in
the ado-file update on 15sep2005.
If Federica is using Stata 9, she should use -update- to get Stata up-to-date,
and the problem will go away.
For Stata 8, Federica can change -mysureg- to parse the -constraints()- option
as a "passthru", and add `constraints' back to the `mlopts' macro after
-mysureg- fits the constant-only model. Here are the lines that will be
affected (look for !!):
***** BEGIN: mysureg.ado
program mysureg, sortpreserve
...
syntax anything(id="equations" equalok) ///
[if] [in] [fweight pweight] [, ///
noLOg /// -ml model- options
Robust CLuster(varname) ///
noLRTEST svy noSVYadjust ///
Level(integer `c(level)') /// -Replay- options
corr ///
CONSTraints(passthru) /// !! NEW LINE
* /// -mlopts/svyopts- options
]
...
mlopts mlopts, `options'
local cns `constraints' // !! LINE MODIFIED
...
`qui' di as txt _n "Fitting constant-only model:"
ml model d2 mysuregc_d2 `eqns0' ///
...
nocnsnotes missing maximize
local initopt continue search(off)
if "`lrtest'" == "" {
local lf0 lf0(`k_eq' `e(ll)')
}
local mlopts `"`mlopts' `constraints'"' // !! NEW LINE
...
end
...
***** END: mysureg.ado
So, you ask: "What's going on? I want details."
As originally developed, -mysureg- passes constraints to each call of
-ml model-, even when fitting the constant-only model. Under version 8.1 (and
8.2 for Stata 8), -ml model- does not use -makecns- to tolerantly construct the
constraint matrix; it stops with an error message instead of just noting
when there was a problem with a constraint then move on. Thus -ml model-
exits with an error if any of the constraints involve predictors that are not
present in the constant-only model fit. As mentioned above, this bug was fixed
in Stata 9.
--Jeff
[email protected]
*
* 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/
