Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Interactions of cbc logit output for NGENE priors

Hey there

I created my cdc design with ngene software and included random priors for variables and interactions for the pretest with sawtooth. Now, after the pretest I want to emend the model with informations from the pretest.

As I can use the output-effects and standard errors on the one side for the variable priors, I am wondering how to get information for the interaction priors. Is it right to use the information from the interaction-search (2LL P-Value for Interaction Effect)?

Thanks for your help!
kind regards,
asked Oct 24, 2018 by bs77 Bronze (815 points)

1 Answer

0 votes

If I recall correctly, I believe Ngene will want the coefficients and standard errors, right, not the p-values.   So you should run a utility model from your pretest data that includes the main and interaction effects you want your Ngene design to optimize, and then use those coefficients and standard errors.
answered Oct 24, 2018 by Keith Chrzan Platinum Sawtooth Software, Inc. (105,750 points)
Hi Keith,

Yes, I do think you recall right. They want the coefficients and the standard errors vor the variables and a distribution (e.g. n)., but its not very clear what they want for the priors of the interactions.

However, during the design development in NGENE I preasumed some interaction effects and implemented them within the ngene-code. When I analyse the pretest of the cbc experiment in sawtooth analysis manager with Logit analysis type, do i have to tell sawtooth, that interactions were already used within the model  (Analysis settings -> Attribute coding), or does sawtooth consider this automatically? I am asking because without applying them in the settings as mentioned before, they are not plotted in the LOGIT summary.

If it would, I assume that I could take this information for the improvement of the ngene design model. Is that right?

Last but not least (sorry for this long post) I am wondering if it is possible in sawtooth analysis manager to sum up part-worth utilities of the attribute levels (e.g. min, med, max) to one  value (like it does with the variable importance)?

Again many thanks for your help!
You do need to specify interaction effects - they're not computed in our software unless you specify them.  Please contact support@sawtoothsoftware.com and they can walk you through how to do this.  

I don't understand your third paragraph.  

And no, if you want to do any summary of the utilities other than what the software has as default, you would need to do those summaries manually.