I am very much in agreement with the first response posted by Nick.
This is one of the most common ways OLS is abused. Hence the comment
about examining what your DV distribution looks like, test all of your
assumptions, and verify that your results make sense.
I am not so sure that my second recommendation should be overlooked so
quickly. Data exploration is a must and I think that what I have
suggested would be a good starting point. However, if more powerful
analyses are available by all means analyze accordingly.
*******************************************
Carter Rees, M.A.
Associate Research Scientist
Wyoming Survey & Analysis Center
710 Garfield St.
Suite 320
Laramie, WY 82072
307-399-0496
*******************************************
-----Original Message-----
From: Nick Cox [mailto:[email protected]]
Sent: Friday, October 08, 2004 3:52 AM
To: [email protected]
Subject: st: RE: RE: A continuous DV between 0 and 1
I disagree here on two standard grounds.
* OLS is all too likely to produce predictions
outside [0,1].
* To degrade an approximately
continuous variable by categorising it in this
way is just throwing away information in the
data.
Other postings pointed to possible analyses.
Nick
[email protected]
Carter Rees, M.A.
> I would think that if it is a ratio variable an OLS
> regression would suffice. OLS is pretty robust when it comes
> to its assumptions so the S shape may not be a problem.
> However, your definition of "somewhat" S shaped has to be
> considered. If it is truly an S-shaped (binomial)
> distribution you should check to see if you can identify a
> logical cut point and turn your DV into a categorical 0/1
> variable and run a logistic regression. If not a simple 0/1,
> maybe multinomial or ordinal regression would be the way to
> go. In any event, if you do decide on a categorical DV check
> out Scott Long's Regression Models for Categorical Dependent
> Variables Using Stata.
________________________________________
Eugene Kang
> I am using STATA 8 to analyze a model with a ratio DV ranging
> from 0 to 1. The distribution of this DV is somewhat
> S-shaped. I was wondering if someone could point me to the
> correct statistical analysis to use (I can read the STATA
> manual and other books from there on).
*
* 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/
