Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: interpreting negative and positive AIC- OLS VS. GLM

From   Arina Viseth <arina@UDel.Edu>
Subject   Re: st: interpreting negative and positive AIC- OLS VS. GLM
Date   Thu, 19 Aug 2010 15:01:35 -0400 (EDT)

Thank you very much again Maarten. Your help is very much appreciated.


---- Original message ----
>Date: Thu, 19 Aug 2010 17:12:00 +0000 (GMT)
>From: (on behalf of Maarten buis <>)
>Subject: Re: st: interpreting negative and positive AIC- OLS VS. GLM  
>--- On Thu, 19/8/10, Arina Viseth wrote:
>> From your experience do you have a recommendation for
>> assessing  model fit for this kind of model?
>The thing that realy bites when it comes to modeling 
>proportions with a linear model are the boundaries: 
>You cannot have a linear line that will respect these 
>boundaries forever (unless you have a horizontal line). 
>So at some point a linear effect will have to become
>nonlinear. The question is does that happen within
>the range of your data, or can a linear line reasonably
>represent your data. 
>The first thing I would do is just plot the distribution
>of your proportion and see if it gets close to one or
>both of the boundaries. If that is the case I would not
>use a linear model, and instead move towards one of the
>alternatives like a fractional logit model or -betafit-
>(which you can download by typing in Stata -ssc install
>betafit-). A nice tool for that is Nick Cox's -stripplot-
>(to install type in Stata -ssc install stripplot-), like
>in the example below:
>*------------------------- begin example -------------------
>use, clear
>stripplot governing, stack width(.01)
>*------------------------- end example ---------------------
>If most of your observations are somewhere in the middle
>and you thus think that linear regression is ok for your 
>data, I would still check the residuals to see if the 
>linear effects are appropriate and whether the boundaries 
>haven't introduced more heteroskedasticity then you feel 
>comfortable with.
>Hope this helps,
>Maarten L. Buis
>Institut fuer Soziologie
>Universitaet Tuebingen
>Wilhelmstrasse 36
>72074 Tuebingen
>*   For searches and help try:
*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index