Hi everyone,   I want to compare the goodness of fit of a MNL model, that I conducted through proc phreg and ties=breslow and a HB model that I conducted through proc bchoice.   Which is the easiest way to compare the goodness of fit of those two models and how to I implement them in the codes? I head that the Bayesian Information Criterion is useful but I couldn't find how you include it in the above named procedures.   Thanks for your help. Best, Alex ... View more

Hi everyone,   I want to compare the overall if of an MNL model, a fixed effects Bayes model and a random-effects hierarchical Bayes model. The log likelihood measure seems to be most appropirate for that, however how do I calculate the overall log likelihood for the HB model in the PROC BCHOICE procedure?   Thank you for your help. Best, Alex ... View more

Hi everyone,   I am running a BCHOICE procedure and was wondering how I can get the covarance and correlation matrix as an actual data output (and not via running stats=corr or stats=cov). Furthermore I wondered why the stats=cov table, for example, shows covariances and variances of the covariances itself (e.g. REmean"Attrlvl1"). What sense does it make to show the covariance matrix of the covariance? Possibly, my thinking is wrong so I am glad for help. Lastly, if I want to draw conclusions to how much my parameters are associated what is the best way? The standard BCHOICE output gives me the mean random estimates (e.g. REmeanAttrlvl1) and its covariaces (e.g. RECovAttrlvl1, Attrlvl 2). So I could interpret them just as I always would with covariates (e.g. Attrlvl1 and Attrlvl2 are positiely associated). However, doesn't the correlation between those two give better interpretable results? How would you proceed?   Thank you very much for your help.   Best, Alex ... View more

Hi everyone,   If I run e.g. a conditional logit model with proc phreg I get the standard error, the p value of the coefficients and the log likehihood as a general goodness of fit measure for the conditional logit model. However, if I want to run a hierarchical Bayes model with only random effects with BCHOICE, I get the standard deviation and the 95% HPD Interval but no p-values and no standard errors in the summary output.   So my question is what are the best goodness of fit measures for the hierarchical Bayes model and how can I calculate them in an easy way in SAS?   Thank you very much for your help.   Best, Alex ... View more

Hi everyone,   I am trying to run a hierarchical bayes model on a choice based conjoint analysis with the following attributes and -levels (this is just an example to make the explanation easier):   Att1: Color Att1Lv1: Red Att1Lv2: Blue   Att2: Size Att2Lv1: small Att2Lv2: large   The raw data is dummy coded: 1 for a present attribute and 0 for not-present. 1 for a yes choice and 0 for a no choice.   I am using the following code: ods graphics on; proc bchoice    data=CSDesignEffect outpost=EffectPost                 nmc=500 thin=2 seed=123                 nthreads=4                 Hitprob;                 class     red blue small large                         None                          N Set;                 model choice=                         red blue small large                 /     choiceset=(Set);                 preddist covariates=CSDesignEffect nalter=5 outpred=Predout;                 / subject=N;                  run;   Now I am trying to specify the reference attribute levels which are supposed to be "blue" and "large". I tried the command ref=last but it wouldn't work - for some reason SAS shows parameter estimates for blue which should actually be "0" since they are the reference. For "large" it seems to work. My second question: SAS shows either zeros, ones or nothing behind the parameters in the summary. E.g. "Red 0", "Blue 0", "small 0", "large 0". What does this mean?   Thank you very much for your help.   Best, Alex ... View more