

Proportional hazards assumption 

Message/Author 

Gareth posted on Friday, July 04, 2008  4:24 am



In Cox regression, can Mplus test the proportional hazards assumption? How is this requested? 


There is not an explicity test to request in Mplus, but you can take the common approach of including interaction terms with time as described in for example the Singer & Willet (2003) book "Applied longitudinal data analysis", pp. 562  


Hi Bengt. I have a question about making sure I create the time interaction term correctly in the Cox model. If I simply create the interaction term in the dataset and then use it in the Cox model, I get the same estimates as when I do the same thing in SAS, but my understanding is that that's the wrong way to do it in SAS. Per a biostatistician I work with and some reading I have done, the correct way to do it in SAS is to create the interaction term within the proc PHREG step (which produces different estimates), or to create a personperiod data set (which I haven't tried). (This information corresponds with what the Singer and Willet book says about there being two different types of statistical methods for handling these interaction terms (p. 564). Also, if you're wondering why I don't simply do the analysis in SAS, it's because I'm doing a path analysis, and this is just the direct path I'm talking about here.) So do I need to create a personperiod data set in order to do the analysis correctly in Mplus? I have continuous time (~2000 events occurring on individual days over almost 20 years), so I am reluctant to have to create a personperiod data set if I can avoid it (also not sure it would even work). And then if I did have a personperiod data set, I have no idea if the rest of the path analysis would work, either. Any help you can offer would be greatly appreciated. Thanks! 


Perhaps you refer to how to do nonproportional hazard modeling in continuoustime survival analysis. This is discussed in the paper below: Muthén, B., Asparouhov, T., Boye, M., Hackshaw, M. & Naegeli, A. (2009). Applications of continuoustime survival in latent variable models for the analysis of oncology randomized clinical trial data using Mplus. Technical Report. which you find together with Mplus scripts at http://www.statmodel.com/examples/penn.shtml#lily where we go through models that combine survival analysis with growth modeling. Here, nonproportional hazard modeling is done by splitting the time dimensions into different intervals. 

Back to top 

