# Weighting / Calibration vs. Constraint

Specific attribute levels in study is not in line as per the market. Below level Importance for example:

Level A: 63%
Level B: 11%
Level C: 8%
Level D: 5%
Level E: 13%

These are calculated using the exp of level then dividing it with sum of exponential of all levels of that attribute.

Actual market:
Level A: 28%
Level B: 40%
Level C: 22%
Level D: 6%
Level E: 4%

Should I use constraints to get the levels in the new order or just use weighting on the attributes levels only.

Main questions are as below:
What is the implication if raw utility of particular attributes does not add upto 0. (This is because weights are either negative or positive to get to the desired market standard)
Is it statistically correct? Will it have any impact on other results?

What is suggested and the rationale behind that.

+1 vote
Raj,

I don't know enough about your study. For example, is there some reason to believe that preferences for Levels A - E of an attribute should align with their shares of occurrence in-market?   For example, folks might like Level A but it may only be available on higher priced products or on less respected brands.  In those cases it would not be unusual for the relative preferences of the levels to be different from their market shares (I'm assuming here that your study has more than a single attribute).  Without knowing more it's hard to say whether any correction is even necessary.

Your other question is about scaling of the utilities.  They sum to zero when you use effects coding of your design matrix.  But other types of coding (e.g. dummy coding) are also used and do not usually result in utilities centered on zero.  In fact, any additive transformation of your utilities will still allow you to use them in simulations.
answered Apr 10, 2018 by Platinum (90,475 points)

This is a multi attribute study and brand levels are not matching to market share and client is bit concerned on the direction of levels preference.

With weighting / calibration some of the weights have to be negative to match the market share which is also one of my concerns.

Any help would be highly appreciated!!

Thanks & regards,
Raj
Raj,

If you have strong (and correct) beliefs about the order of preference for an attribute's levels, then yes, I'd advise constraints.  But do be careful - sometimes we assume we know the order in advance and then constraints  over-write what could otherwise be important findings.  For example, we are sometimes wrong to assume that consumers prefer lower prices to higher ones - particularly in safety related categories, low price may be taken to mean lower quality = less safety.  So do feel free to add constraints, but after careful thought.
Thanks Keith, finally can resolve this.

Yes, order of levels is something which we have observed in actual sales data , the importance of levels is something I assume is biased because of awareness and could be sampling as well.

Agreed with you other points as well.