Have an idea?

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

Counts in SMRT dont add to 100%


Ideally the counts generated using SMRt should add to 1.0 (100%). But it doesn't do so.
What could be the reason?

Attribute 6           
Total Respondents    316       
Yes    0.23       
No    0.22       
Within Att. Chi-Square    1.71
D.F.    1
Significance    not sig

Attribute 1           
Total Respondents    316       
American Express    0.21       
Citi Ultima    0.21       
HDFC Bank Infinia    0.24       
HDFC Bank Diners    0.25       
Within Att. Chi-Square    18.20       
D.F.    3       
Significance    p < .01
asked Mar 29, 2012 by mal_saw2001 Bronze (875 points)

1 Answer

+1 vote
There's a simple explanation for that.  Counts are the number of times a level is chosen divided by the number of times a level is available to be chosen.

If you happen to have a design with (say) 3 levels of brand, where each level appears exactly once in each choice task, and if no None choice is available, then indeed the counts percentages within brand will sum to exactly 100%.

But, if there is a None choice, then the counts within that attribute will sum to 100 minus the None percentage.

More commonly, each level isn't available exactly once per task.  When that happens, the counts percentages will not sum to 100%.  This happens when there are more or fewer levels within an attribute than concepts you are displaying per screen.  It can also happen if you use a method such as Balanced Overlap that often will choose to repeat a level within the same task.

For example, if there are 4 concepts to be displayed, and you only have 2 levels within the attribute, then each level will typically be displayed twice per attribute.  The counts percentages will sum to less than 100 in that case.  If there are more levels in an attribute than concepts per screen, then the counts percentages will sum to more than 100.

If this bothers you, you can always rescale the counts to sum to 100% within each attribute.  It's only appropriate to compare counts percentages within an attribute (not across attributes), so it doesn't matter if you multiply the counts percentages within an attribute by a scaling factor.  The interpretation of relative preferences for levels within that attribute are the same.
answered Mar 29, 2012 by Bryan Orme Platinum Sawtooth Software, Inc. (198,815 points)
Thanks Bryan