www.cmm.bristol.ac.uk/forum

Hello,

Hope this finds you all well! I've found this forum really helpful and useful and thank you for all great supports!

Just got a quick question re a prediction in MCMC model (and in general non-MCMC model too).

For instance, I've built a four-level linear model with MCMC (non-specified prior method), predicting depressive symptoms score by a function of age (continuous), education (<middle school graduate, high school graduate, and >=college graduate as reference), income (quartiles), and other continuous and categorical independent variables.

I would like to present predicted values across educational categories, accounting for all other variables (assuming at means or other reasonable ways). After building the model, when I go to the "prediction" window, I have got options to specify my prediction model.

In this case, I've chosen only fixed part constant (not random intercepts) and fixed part educational categories (one for <middle school graduate, and the other for high school graduate), and performed a calculation. Here, I am wondering whether the predicted values that I've got from this procedure would just ignore the unspecified independent variables, or do some underlying procedure to handle those unspecified (adjusted at mean or else)?

Would appreciate your comments!

Best,
Yongjoo

The predictions window just takes each parameter that you specify, multiplies it by the corresponding data and then adds them all up. You can confirm this by performing the calculating yourself and comparing the results. If you want to perform a prediction for particular values, or include other variables at their means then I would suggest looking at the customised prediction window (details available in the manual supplement: http://www.bristol.ac.uk/cmm/media/soft ... nt-web.pdf).

Thank you for your help, Chris! So helpful!