Next Upcoming Google+ Hangout: Tuesday, August 27 @ 7PM (CST) - To Participate CLICK HERE

Search For Topics/Content
Top
Discussion Forum Topics
Find a Blog Entry

SPSS Topics Discussion > The value of random intecerpts model in SPSS Mixed

Hi Jeremy,

If you had the time I’d appreciate your advice on an issue I am having with multi – level regression model using SPSS Mixed. The data I have represents 250 nurses located within 36 units. The null model indicates significant clustering around my DV. I have specified the model to include the random grouping of unit and with fixed effects (level one and level 2)of my main predictors. However the model will not run(non – positive hessian matrix warning) when I add random slopes. I have had expert advice to suggest that this may be that the covariance of slopes is close to non-existence and that simplifying the model by removing the random slope may help.

Certainly the model does run when I do this. The simpler model does take into account the clustering in units (Random factor). The results of the simplified model indicate some interesting and significant fixed effects however now the model does not contain random effects(and thus not a full random co-efficents model). I think my model is thus best described as a two level random intercepts model. I suppose ultimately my question is how valuable is a model with fixed effects only?

Thanks in advance

Jennifer

May 28, 2012 | Unregistered CommenterJennifer

Hey Jennifer,

Thanks for the question! I'm afraid I'm not fully sure how to answer the question. I would agree with how you described your model, and the conclusions you took (based on the information I have), and I do think it is valuable.

To me knowing the slopes DO NOT vary is just as valuable as knowing they do. Try not to think of it as a failure to be able to predict variability in slopes, but instead think of it as "what does it mean that the slopes don't vary, and why might that be the case".

This is similar to my contention that a non-significant effect (particularly when you expected there to be an effect) is just as informative as a significant effect. It all has meaning. I hope that helps.

May 30, 2012 | Registered CommenterJeremy Taylor

It certainly does Jeremy. Many thanks

June 19, 2012 | Unregistered CommenterJennifer

Glad I could help, Jennifer!

June 19, 2012 | Registered CommenterJeremy Taylor