st: Repeated Cross Sectional Data and Conditional Logit (e.g., -asclogit-)

[Misha Spisok <misha.spisok@gmail.com>]

Hello, Statalist!

In brief, how can I carry out a conditional logit analysis with
repeated cross-sectional data?

The problems I have identified are, in brief:
1. Non-constant fweights or inability to use -expand- or -expandcl-
2. -asclogit- is not "xt" (hence the "repeated")
3. id_t != it_t+k for all (any) k (hence the "cross-sectional")

My data look like the following:

State	Year	Pop	Alternative	n	x1	x2
1	1990	25361609	1	20668	29	27
1	1990	25361609	2	4026	37	11
1	1990	25361609	3	22516	90	68
1	1990	25361609	4	21413	96	58
1	1990	25361609	5	9716	41	87
1	1990	25361609	6	4974	37	68
1	1990	25361609	7	25278296	53	41
…	…	…	…	…	…	…
1	2000	29165850	1	9068	96	67
1	2000	29165850	2	3239	80	92
1	2000	29165850	3	7124	30	83
1	2000	29165850	4	2025	38	41
1	2000	29165850	5	11261	23	22
1	2000	29165850	6	28353	48	93
1	2000	29165850	7	29104780	58	97
…	…	…	…	…	…	…
50	1990	2536160	1	795	38	4
50	1990	2536160	2	1870	82	8
50	1990	2536160	3	1999	55	33
50	1990	2536160	4	335	68	67
50	1990	2536160	5	1964	55	12
50	1990	2536160	6	2426	15	4
50	1990	2536160	7	2526771	96	25
…	…	…	…	…	…	…
50	2000	2916584	1	336	25	39
50	2000	2916584	2	2781	85	40
50	2000	2916584	3	1456	34	8
50	2000	2916584	4	2716	4	73
50	2000	2916584	5	1666	36	31
50	2000	2916584	6	947	51	75
50	2000	2916584	7	2906682	92	16

x1 and x2 are alternative specific regressors.

I could (in the sense that I am able to) change my data to be amenable
to -asclogit- except for the use of -expand- (or -expandcl-) on n.
But, assuming I could use -expand- or -expandcl-, I would create a
variable "choice" which indicated whether a given alternative was
chosen and "id" for each observation ("case" in the language of
-asclogit-).  I would then want to do something like the following:

asclogit choice x1 x2 [fw=n], case(id) alternatives(Alternative)

Here are some of the problems I'm encountering.

1. The data are grouped, but expanding using -expand- or -expandcl- is
too much data due to memory constraints.  I wonder what Stata would do
after, say, -asclogit-  even if I could successfully use -expand- or
-expandcl-.  Using the -fweight- option does not work because weights
must be constant.  Is there a way to deal with non-constant fweights
that I haven't seen in the documentation?  Am I looking at this the
wrong way?

2. Even if I could expand the data or be allowed to use non-constant
fweights, I can't find a "panel data" (i.e., -xt-) version of
-asclogit-.

3. The data are not--obviously--truly micro data, so, even if I could
expand the data, I couldn't claim to follow observation i in state j
in year t in the subsequent year, t+1.  However, I do have the entire
population, so it is not a sample, as I would understand a sample to
be a proper subset of the population.  For example, with -asclogit-
case(id) would not be accurate in any meaningful sense.

What I have in mind is Wooldridge (2002), in which he writes, "For
example, the identical distribution assumption is unlikely to hold for
a pooled cross section, where random samples are obtained from the
population at different points in time.  This case is covered by
independent, not identically distributed (inid) observations...we do
not cover the inid case explicitly in derivations because little is to
be gained from the additional complication."  (pp. 5-6)

I've coded something up in Matlab for a single year, but this seems
like something Stata could do, but I'm just not aware of the proper
trick or command.

Thank you for your time and attention.

Misha
Using Stata 10.1 but with access to Stata 11

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/