Have an idea?

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

Effects coding in MBC Excel simulator - can it be "patched in"

Hi folks,

I'm aware the MBC software uses dummy coding but I'm in need of using effects coding for an MBC project (so that the intercepts represent the grand mean). I am wondering whether it's theoretically possible to export a regular dummy-coded Excel simulator from MBC, but then estimate utilities in the CBC/HB system (which uses effects coding), and overwrite the design and data inside the dummy-coded Excel simulator for each dependent variable with the effects-coded design+data.

This seems theoretically possible to me, but before I go down that rabbit hole I'd thought I'd check and see whether anyone had tried it? If so is there any time saving vs. just building an effects-coded simulator from scratch?
asked Jan 30, 2022 by mgard (120 points)

1 Answer

0 votes
I don't see a reason why this shouldn't work, though I don't understand the reason for doing so for the MBC market simulator.  Whether using dummy-coding or effects-coding, the predictions of the market simulator should essentially be identical (within the expected degree of noise from using one HB estimation versus another from a different random starting point).  So, it would seem like doing a lot of extra work for no change in the prediction results.

It's true that if you're in a situation in which you need to interpret the utilities, effects coding zero-centers the utilities within each attribute, making it easier to interpret the effects.  But, most of us use the market simulator and variations from a base case prediction to explain the effects of the attributes to audiences (sensitivity analysis).  This approach is much more intuitive than trying to display utility scores and explain them.
answered Jan 31, 2022 by Bryan Orme Platinum Sawtooth Software, Inc. (201,765 points)
Thanks Bryan! Actually, the reason I want to use effects coding is that I have an independent variable where each level (including the off state) was shown a specified % of the time. I want to be able to simulate the effect of each level of this variable on choice, but I also want to disable this variable in simulation so that I get the (weighted) average impact of the independent variable across all levels.