Have an idea?

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

exclude variables from choice simulator

Is there any possibility to exclude an attribute from the choice simulator?
I already excluded the attribute from HB-Analysis but it seems like it is still an option in the simulator.
Thanks a lot!
asked May 21, 2019 by bs77 Bronze (790 points)

1 Answer

0 votes
The way we've built the Choice Simulator into Lighthouse Studio, the simulator can support (allow you to toggle among) multiple utility runs that you might have estimated.  For example, you might have estimated HB utilities using just main effects.  And, you may have estimated a second HB utility run using main effects plus an additional interaction effect.

In your case, you have an HB utility run where you decided to exclude an attribute from the HB estimation.  (I'm assuming you did this because it truly wasn't a real attribute that respondents would have had any preference for; but maybe it was just a "dummy" attribute used for display purposes--such as painting a line on the screen?)

No matter how many utility runs you've conducted, the simulator refers to the original attribute table when building the attribute grid for scenario specifications in the simulator.  That's because it wants to be prepared for any possible HB utility run with any possible utility estimation settings.

When you want to exclude the effect of an attribute in the market simulator, you specify "NA" in the attribute grid for the scenario specification for that attribute.  The NA level specification tells the simulator to add zero utility when encountering this attribute.

I hope you are not making the mistake of omitting an attribute from HB estimation that in fact really does explain some signal in the data.  If you do that, the precision of your estimates for the other attribute levels will be diminished.  Rather, if you want to ignore (hold constant) the effect of a meaningful attribute in the simulator, you should go ahead and estimate utilities for it in the HB run, but then use the "NA" level specification for all the products in the simulation scenario for that attribute to be held constant (removed).

If you truly want to strip all mention of an attribute from the Sawtooth Software Choice Simulator, there is a power trick for doing that; but it would involve exporting a .HBU file, editing the header of that file with a text editor, then importing that .HBU file into a new project involving the standalone installation of the Choice Simulator (rather than the version of it integrated within Lighthouse Studio).
answered May 21, 2019 by Bryan Orme Platinum Sawtooth Software, Inc. (175,190 points)
thanks a lot for your answer.

my intention was to test the effect of that attribute on the results as that the "planned to omit" attribute shows a very homogeneous behavior  (all participants think it has a negative effect). and also i am more interested in the effects of the other (three) attributes on the preference.

So am I right: the NA level specification in the LHS simulation gives the specific level zero utility (not negative, not positive) but it considers the existence of the attribute.

But maybe that was a wrong approach.  Can you recommend how to find (calculate?) the importance of an attribute related to the choice made? I don't think that the "attribute importance" reflects this? Does it?

My situation: CBC study. 1 nominal attribute (7 levels), 2 ordinal attributes (4 levels), 1 ordinal level (2 levels). None option available. 15 randomized choice tasks (2+1 choice cards per choice set).
The choice simulator helps to get an overview about the scenarios, as we have 32 scenarios in 7 landscapes (=224 scenarios).
I want to know which attribute shows the most impact on the choice made (and whats the size of it).

Thanks a lot!