## RE: st: Multinomial logit in stata

Maarten - I was thinking of the proportions as arising from binomially distributed counts, one each from given combinations of categorical independent variables. The distribution of these estimates depends on the sample sizes. Without knowing these sample sizes or imposing some constraints you could not even write a likelihood for such data. In your example where there is a continuous independent variable you could do beta (or dirichlet) regression, but it would be a stretch to assume that there is one set of Beta or Dirichlet parameters that describes the distribution of all the sample proportions. This would only be plausible if the budgets were all the same. Even so, the loss in power would be enormous unless you had many such cities.

Al

-----Original Message----- From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Maarten buis Sent: Friday, March 03, 2006 3:37 PM To: statalist@hsphsun2.harvard.edu Subject: RE: st: Multinomial logit in stata

Al: Say you wanted to know how the proportions of city budget spent on programs 1 till 5 dependent on the size of the city, strength of the republican party, and the number of dogs eating hotdogs of brand 3, than I don't see why you would need the size of the budget in order to make inference. (the absolute amount spent on the various programs might also be of substantive interest but that is another question) Maarten

--- "Feiveson, Alan H. (JSC-SK311)" <alan.h.feiveson@nasa.gov> wrote:

> Maarten - But suppose I told you I had two types of coins - one that > came up 45% heads, the other 53% heads. Does the probability of a head

> depend on the type of coin? If you knew the number of trials that > produced those proportions you could do the inference. If all you were

> given were the two numbers 45 and 53, you could do nothing useful. On > the other hand if you were given N1 sample proportions from Coin 1 and > N2 proportions from Coin 2, you could do inference even if you did not

> know the number of flips per proportion, but you would also have to > assume the numbers of flips were the same for each proportion > otherwise the sample proportions would not be identically distributed.

> Even in this case, unless you had a large number of proportions per > coin, you would sacrifice a lot in power compared to knowing the > actual number of flips. > > Al > > -----Original Message----- > From: owner-statalist@hsphsun2.harvard.edu > [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Maarten > buis > Sent: Friday, March 03, 2006 3:07 PM > To: statalist@hsphsun2.harvard.edu > Subject: RE: st: Multinomial logit in stata > > Alan and Albert > Inference could be done if the unit of analysis is cities (in Alberts > example), e.g. proportion of city budget spent on categories 1, 2, 3, > 4, and 5. If you wanted to do the inference on lower level units > (individuals/companies/dogs who choose to buy/eat hotdog brand 1, 2, > 3, 4, or 5 within > cities) than the problem is even more severe than Alan's remark suggest: > than you also have the ecological inference problem. See for example: > http://www.stat.columbia.edu/~cook/movabletype/archives/2006/03/the_ec > ol ogical.html . Note that the Papke and Wooldridge paper I > reccomended falls in the ecological inference trap. > HTH, > Maarten > > --- "Feiveson, Alan H. (JSC-SK311)" <alan.h.feiveson@nasa.gov> wrote: > > Unless you had repeated observations for the same covariate pattern,

> > I > > > don't see how proportions alone would be enough information to do > > proper inference. Those same proportions could have arisen from 100,

> > 1000 or 100000 observations. Clearly the inference would be > > different for those situations. > > ----------------------------------------- > between 1/2/2006 and 31/3/2006 I will be visiting the UCLA, during > this time the best way to reach me is by email > > Maarten L. Buis > Department of Social Research Methodology Vrije Universiteit Amsterdam

> Boelelaan 1081 > 1081 HV Amsterdam > The Netherlands > > visiting adress: > Buitenveldertselaan 3 (Metropolitan), room Z214 > > +31 20 5986715 > > http://home.fsw.vu.nl/m.buis/ > ----------------------------------------- > > > > ___________________________________________________________ > To help you stay safe and secure online, we've developed the all new > Yahoo! Security Centre. http://uk.security.yahoo.com > * > * 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/ > > * > * 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/ >

----------------------------------------- between 1/2/2006 and 31/3/2006 I will be visiting the UCLA, during this time the best way to reach me is by email

Maarten L. Buis Department of Social Research Methodology Vrije Universiteit Amsterdam Boelelaan 1081 1081 HV Amsterdam The Netherlands

visiting adress: Buitenveldertselaan 3 (Metropolitan), room Z214

+31 20 5986715

http://home.fsw.vu.nl/m.buis/ -----------------------------------------

___________________________________________________________ To help you stay safe and secure online, we've developed the all new Yahoo! Security Centre. http://uk.security.yahoo.com * * 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/

* * 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/

Tag: