Follow-up question - Maxdiff study attributes

Question

Hi,

I ran the analysis and noticed only 99 attributes in the Utilities file. But I had 100 items in my MaxDiff exercise. How do we report this? or is it something wrong?

Notes from help:

"Estimating the Model using CBC/HB: On the Attribute Information tab, modify all attributes to have "User-specified coding."  (Note that there will be k-1 total attributes in the study, representing your k total items in the MaxDiff design.)"

Thanks,
Lawrence

Answer 1

Everything is just fine.  Please see Appendix K in your CBC/HB manual.  The utility of the final (100th) item is constrained to be zero and the estimates of the remaining 99 items are with respect to the 100th item being zero.  Appendix K gives advice regarding how to place these raw utilities on a probability scale that is normalized for each respondent (so that the scores sum to 100 for each respondent.)

To verify that everything has gone well for your estimation, including a proper specification of the prior covariance matrix, make sure to compare the average utility scores for the 100 items from your study to the Counts scores for the 100 items.  (Counts are the number of times each item was selected best minus the number of times each item was selected worst, assuming a level-balanced design).  The results should have high correlation when X-Y scatter plotted.

Comments

Hi Bryan,

Thanks for your answers. Per your statement the 100th item will be zero for each respondent.

I have two questions regarding conversion of raw utilities to probability scale,

1. I have done the first step on rescaling procedure. On the second step, I need to take the average of all cases for an item and then subtract that average value from each individual value. Please confirm whether my understanding is correct here

2. Can you please explain about counts scores you have specified above?