Have an idea?

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

Different currencies in ACBC

We are planning to launch a conjoint survey in different countries and I'm not sure how to run the analysis with different currencies. I searched previous post on the forum and found Brian said this could pose a problem for ACBC because "With HB estimation, each respondent "borrows" some information from the other respondents to stabilize estimates.  If one person uses a currency 1/3 the magnitude of another respondent, then the price slope (the utility) will likely be about 3x the utility of the other respondent (assuming these respondents have fundamentally the same price sensitivity).  This can cause bias in the utility estimates, especially for respondents whose scaling of their currency is quite different from the mean scaling of currency."

My question is that is there a way to avoid estimating utilities and run simulation in different currencies? I'm afraid I don't have the luxury of sample size to do that.

OR can I set price as discrete levels "price 1, price 2, price 3..."? What is the drawback of setting price as discrete instead of continuous variable?

asked Jan 23 by xliu (220 points)

1 Answer

0 votes
This problem is specific to using a summed pricing attribute, because the data stores the attribute as the actual value, as opposed to a level index (i.e. brand is stored as 1/2/3 etc., while price is stored as $1,450, $989, $2,381 etc.).

By using scripting to dynamically change the price for different respondents, the data will be recorded with the different price values (i.e. 989 for a respondent from the United States,  but 55,750 for someone from Russia).

If you can get away with not having a summed pricing setup, but just doing a normal price attribute with set levels, you're in a much better position.  I wouldn't say there is a drawback of treating price as a normal variable.  Most people that end up using a summed pricing approach do so to try to make the pricing a bit more realistic (products full of good levels should, on average, cost a lot more than products made up of not so good levels).

Conjoint studies with price a normal attribute with discrete level is probably the vast majority of exercises done, so definitely nothing wrong there.  If changing the actual value shown based on different currencies,  it would make a lot of sense to make sure you are showing and modelling the same thing (it would be weird to show a large range of price in currency 1 and then show a small range in price of currency 2 and then combine everything for estimation).  It would also seem like using the currency group as a covariate in the HB estimation would be a good idea.
answered Jan 24 by Brian McEwan Gold Sawtooth Software, Inc. (37,610 points)