Statalist


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

st: predictions after clogit


From   Ginger Smith Carls <smithga@umich.edu>
To   statalist@hsphsun2.harvard.edu
Subject   st: predictions after clogit
Date   Thu, 13 Nov 2008 11:01:43 -0500

Hello statlisters,

I would like to generate predictions after clogit. For example, my dependent variable is a hospitalization. I want to predict the number of fills of a specific drug in my sample under different scenarios. The diffent scenarios are defined by different values of an indepdent variable. One scenario is at the sample mean. The other is to add some amount to the variable. What is the best way to do this? Some of the options are:

a. predict outcome for diferent scenarios and calculate the difference
b. use mfx to calculate the marginal effect of the change

When I run mfx, it says there is no effect. When I run predict, the predicted value is identical for each scenario. I tried other predict options and the results did not make sense either. One thing to note about my data is that more than one "yes" or 1 is allowed to occur within each group. Most groups have only one success, but some have more than 1.

The reason I want to do this, rather than simply report the odds ratio for the variable of interest, is that the odds ratio may not approximate relative risk in my case because the outcome is relatively common. I would like to report . when x var increases by 20 points, yvar increases by some percent.

It is my understanding that the odds ratio does not approximate relative risk unless the outcome is rare or in special designs that are not the case here. I don't think my outcome is sufficiently rare to assume that odds ratio~= relative risk, so I am looking for an alternative way to report results.

Thanks for your help,
Ginger

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index