Have an idea?

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

Analysis manager and two data sets

Dear All,

I need to combine data sets for 2 cbc studies.

Both studies are almost the same: 7 atttibutes, same levels, same number of tasks, concepts per task etc.

There is only one difference for attribute no 6.

Study 1 - contains 5 levels (old one)
Study 2 - contains 3 levels (new one)

I have estimated utilities for both studies but I would need combined outputs.
I have calculated and then transformed utilities for att.6 to 3 levels and theoretically can combine both raw utilites in an excel simulator (for preference share etc) but I like capabilities of analysis manager and simulator implemented into Lighthouse.

I tried simly copy and paste raw utilities from  study 1 to analysis manager of study 2 and then go to lighthouse simulator but this trick doesn't work.

I would be thankful for any idea how I could combine outputs for both studies into Lighthouse analysis and simulator manager.

Thank you!
asked May 17 by robson Bronze (905 points)

1 Answer

+1 vote
This could be a tricky thing to do properly...to combine the two studies.  Within each study, the utilities are zero-centered (and thus affected by the other levels included in the attribute).  So, combining the 3-level attribute and the 5-level attribute is more than just taking the two sets of utilities and combining them.

If the attribute in question was categorical such as brand or color, it's not as straightforward as if the attribute was continuous such as price or speed and that the full range of levels included in the study was the same whether using the 3-level version or the 5-level version.

The most proper way to do it I think would be to format both raw choice files in a way that the 3-level and 5-level attribute are coded for utility estimation such that the 3 levels common to the two line up and that HB imputes the utilities for the missing two levels from the upper-level alphas and covariances.

Then, I would take the resulting utility file and use Sawtooth Software's standalone desktop simulator or our web-based simulator to read in the full set of utilities.
answered May 17 by Bryan Orme Platinum Sawtooth Software, Inc. (198,715 points)
Thank you Bryan,

the attribute which differs is diameter size so we can treat it as continuous. It was just amended from 5 levels to 3 levels, so previously estimated ranges are more detailed than current ones and formating into 3 levels is possible. Btw. it's constant sum study.

I like your idea to format raw choices from 5 to 3 levels but the issue is that designs are slightly different  (for other attributes than attribute no 6 although all other attributes and levels are the same)  thus I wanted simply to use calculated raw utilities from Study1  and then with respective amend for attribute 6  import them into (or copy) to the Lighthouse simulator (for Study 2). But  such operation doesn't seem to be possible or I don't see a respective option to import additional utilities (to already present ones).
Even though the experimental design (combinations of attribute levels) shown to the new respondents was different, you could still export the two sets of data to .CSV file...stack them, recode attribute 6 so that first respondents now show that they saw levels 1, 3, and 5 only of attribute 6 and re-estimate.  Imposing a utility constraint on that attribute in the new estimation might also help.

Another approach is a faster hack: modify the utilities for the respondents who got levels 1, 3, and 5 of the continuous attribute and use linear interpolation to impute the values for each respondent for the missing levels 2 and 4.
Thank you Bryan!