Have an idea?

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

How should the Dual None response option be incorporated into analyses?

Using SSI Web Version 8, we are conducting a health service delivery study that includes 11 4-level service process attributes.   The study uses a partial profile experimental design with a Dual None response option.   We note that the Dual None response option generates a utility coefficient (18.01) and a standard deviation.   We typically analyze our data using HB, Latent Class, and Randomized First Choice Simulations.  How should the Dual None data be incorporated into these analyses?

Thank you!
asked Feb 25, 2014 by cunnic Bronze (1,440 points)

2 Answers

+1 vote
Asking respondents about their purchase intent for partial-profile defined products is somewhat tenuous.  I realize that ACA did this for decades, but I don't think it was ever argued that the purchase intent ratings for ACA's calibration concepts were accurate in an absolute sense for predicting real-world adoption rates.  They were more used as a relative indication of purchase interest, as a way to try to scale ratings-based utilities to be more appropriate within choice simulators, and as a way to try to identify inconsistent vs. consistent respondents.

Regarding partial-profile and CBC with the None, the Patterson/Chrzan paper presented at our 2003 Sawtooth Software Conference showed that the None usage (for standard None) was affected by the number of attributes shown in the partial-profile display.  I know that Dual-Response None is a little bit different animal, but I suspect the finding would generalize to the dual-response None.

So, in short, I'd proceed cautiously with trying to interpret the None when using Partial-Profile CBC.  If you  believe you are on reasonably firm ground, then you could turn the None "on" in our market simulator, so that a certain percent of respondents' "votes" in the simulator are assigned to the None category rather than to the other products you place in the simulator.

One does that by editing the Scenario, clicking the Advanced button, and setting the None weight to 1.0.
answered Feb 25, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (172,690 points)
Thank you, Bryan.  This is very helpful.
+1 vote
Hello, cunnic.

I'd agree with Bryan that one should be cautious interpreting "none" in the context of partial profile questions:  we know that the size of the none utility may depend on the number of attributes that appeared in each of your partial profile questions.   

Of course we know that we should be cautious interpreting "none" in full profile questions as well:  while it would be great if including "none" in our CBC experiments allowed us demand level scaling of utilities, we know it doesn't always work out that way.  

In sum, I would follow Bryan's advice, proceeding cautiously and include none if it seems to make sense, whether I was using full or partial profile CBC.
answered Feb 25, 2014 by Keith Chrzan Platinum Sawtooth Software, Inc. (90,475 points)
Good advice.  Thanks Keith!