# How do I determine the utility of level combinations from different attributes?

Do I understand it right, that in order to determine the overall utility of the combination of A1L1 & A2L1 I have to calculate A1L1 + A2L1 + A1L1*A2L1?

Lets say I have one attribute (A1) with different car brands (Brand A , Brand B, Brand C) and an attribute (A2) with different colors (red, blue, white).

In a multinominal logit model I received for Brand A an effect of 1.80, for color red I received an effect of 0.41 and for the interaction of Brand A * Color red I received an effect of 0.46. Does it mean that the utility of a "red Brand A car " would be 1.80 + 0.41 + 0.46  = 2.67? And not just the effect of the interaction term Brand A * Color red= 0.46?

And if so, do I have to calculate this with the effects I receive in the model output or rather with the utilities (zero-centered differences) which I receive when I click on utility report?

So in my case it would be 113.84 + 26.28 + 29.03  then and not 1.80 + 0.41 + 0.46 ?
related to an answer for: How to interpret interaction effects?
asked Aug 21, 2017

## 1 Answer

0 votes
Yes, if you have estimated an interaction effect, then indeed the total utility of two levels from two attributes is given as you outlined.

If you decide to use raw utility scaling, then you use the raw utility scaling.  If you decide to use zero-centered diffs scaling, then you use zero-centered diffs scaling.  Just be consistent.

Raw utilities are used when you build market simulators and apply the logit rule to produce shares of preference.

Zero-centered diffs scaling is used when you are tabulating the results and want to place respondents on a similar scale.
answered Aug 21, 2017 by Platinum (159,785 points)