s-news
[Top] [All Lists]

logistic regression, unbalanced design

To: s-news@lists.biostat.wustl.edu
Subject: logistic regression, unbalanced design
From: Manzanita2001@aol.com
Date: Fri, 21 Oct 2005 16:38:27 EDT
 
Dear all,
 
I am using SPlus 6.1 and running a logistic regression analysis on the prevalence of fish abnormalities.  I have several predictor variables that are factors and each factor variable has several levels.  My data set is unbalanced. 
 
I do not have a baseline group and for most variables I have many levels.  Therefore, odds ratios are less useful to me than a standardized prevalence for each level of a factor.  I believe that because my data set is unbalanced the intercept in the output is not the grand mean and the regression coefficients not directly interpretable (is this correct?).  I would like to be able to use something similar to "model.tables" to obtain an intercept that I could interpret as the grand mean (mean prevalence) and regression coefficients and standard errors that can be used with the intercept to calculate a standardized prevalence (after back-transforming the logit transformation).
 
Any advice would be most welcome.
 
Thank you!
 
Sincerely,
Mindy Nelson
<Prev in Thread] Current Thread [Next in Thread>
  • logistic regression, unbalanced design, Manzanita2001 <=