Have an idea?

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

Is there a limit to the number of compound attributes that should be used?


I'm wondering if there is a rule of thumb limit to the number of compounded attributes that can be used in a study. I refer to the Sawtooth paper "Formulating Attributes and Levels in Conjoint" where you take 3 attributes and combine them to create 8 compound combinations which account for both main and interaction effects. For ease of understanding I will just code using numbers where 1 is the attribute level and 0 is absence of an attribute level:
000 (none of the attribute levels)
111 (all attribute levels)
Where ist is possible to disaggregate the levels to get the individual values for each attribute (taking the average of each combination that contains an attribute level). Is there a rule of thumb which suggest that a number of these types of attributes become too difficult for respondents to handle or could I construct a full experiment with this type of compound attribute?

Any advice would be great!!

asked Nov 12, 2017 by Jasha Bowe Bronze (1,680 points)

1 Answer

+1 vote
Best answer

Respondents could easily handle 10 attributes, but you'll sooner run into a problem as an analyst if you try to code this as a 1,024 level compound attribute.  So I think what respondents can handle isn't the right question to ask - rather, it's how many levels can you code for a compound attribute before the standard errors around them get too large - and I think you'll see that probably three binary attributes (which become an 8-level compound attribute) may be the limit.  The next step up is 16 levels and you can run a design test to see that you'll start increasing your standard errors pretty badly.
answered Nov 14, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (53,875 points)
selected Nov 14, 2017 by Jasha Bowe