[Limdep Nlogit List] weighting and proportions data

Durham, Cathy cathy.durham at oregonstate.edu
Thu Aug 10 03:54:33 EST 2006


Use of the sample size by region sounds even better, and I think I've
seen those numbers provided in
NAS publications before.  Good luck.

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au]On Behalf Of George
Frisvold
Sent: Wednesday, August 09, 2006 10:25 AM
To: Limdep and Nlogit Mailing List
Subject: RE: [Limdep Nlogit List] weighting and proportions data


Cathy,
Yes, I agree that weighting by farms rather than land area units makes
more
sense.
Better still, I think I'll try and find out from the National Ag
Statistical
Service the number of farms they
actually sampled (n), rather than the universe of farms (N).  Using n is
more logically consistent.
GF


-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au]On Behalf Of Durham,
Cathy
Sent: Tuesday, August 08, 2006 3:41 PM
To: Limdep and Nlogit Mailing List
Subject: RE: [Limdep Nlogit List] weighting and proportions data


George, I wouldn't interepret that as an appropriate use of weighting
since you just have the one you don't have N choices made, and of course
they will all be huge. If I was reveiwing it I might think it was
plausible to weight by the number of farms in the region, since that is
the number decision makers.
Cathy
-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au]On Behalf Of George
Frisvold
Sent: Wednesday, August 02, 2006 10:26 PM
To: Limdep and Nlogit Mailing List
Subject: [Limdep Nlogit List] weighting and proportions data



Hello All,

I have a data set with proportions data for land allocated to different
activities.
So the data is proportion of acreage in a region applying a particular
technique.
Looking at related literature, a number of studies run multinomial logit
regressions (implicitly) treating land units as N.
There seem to me to be 3 problems here:
(1) This has, as the LIMDEP manual says, "the surprising side effect of
producing implausibly
small standard errors"  because each unit of land is treated as an
observation.
(2) Further, one can get even smaller standard errors by changing units
(e.g. from square miles to hectares to acres)
(3) Because individuals are making decisions about multiple units of
land
doesn't seem to fit the notion of a binomial trial
(unless you think of a farmer carring out 500 "separate" trials on 500
acres.

I'd welcome any ideas about a defensible way to weight the data.

Thanks,

George F

_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au

_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au



More information about the Limdep mailing list