Have an idea?

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

Non-generic concept labels in CBC studies

Hi everyone!

I am designing a Discrete Choice Experiment using the CBC software for SSI Web (with access to the ADM).
The main characteristics of my design are:
- traditional full-profile
- balanced overlap task generation method (because I am equally interested in main effects and interactions)
- three alternatives + the none option
- six attributes (no alternative-specific attributes)
- no prohibitions

For the purpose of my DCE, it makes sense to label the alternatives in a way that is not generic -as recommended by the software- but provides respondents with additional information instead. If I ignored the software advice,  I would then be unable to disentangle with the current design the effect of extra information brought by the labels (alternative-specific constant) from order effects. Since I would like to avoid including an extra attribute in my design and then adopting generic labels, I was wondering whether anyone knows how to overcome this impasse. There must be something I am overlooking...

Many thanks in advance for any hint!
asked Dec 11, 2013 by lotika (455 points)

1 Answer

+1 vote
If you want to use the ability to use labels on top of each column, but you want to use those labels to signal something meaningful (in terms of utility) to the respondents, then there is a workaround.

Of course, I would suggest just adding another attribute to the experimental design, but you have stated that you don't want to add another factor to the design for some reason...you just want to use the current design.

To use the "column labeling" available within our latest CBC software but also get utility estimates for those labels (because they are meaningful to choice likelihood), do this:

1.  Specify those column labels that convey utility meaning to respondents within our software (ignoring our warning).

2.  Collect the data (do NOT use the concept sorting option in the CBC software, as it will make it impossible for you to figure out which products were shown in which columns).

3.  Export the data to .CSV format (single .CSV format).

4.  Modify the .CSV file to add another attribute to the experimental design (your column label attribute).

5.  Take the new .CSV file and submit it for utility estimation with our standalone CBC/HB or Latent Class software (NOT the built-in software within SSI Web for utility estimation).

6.  Import the resulting .HBU file into our SMRT market simulator or our online market simulator for analysis.
answered Dec 12, 2013 by Bryan Orme Platinum Sawtooth Software, Inc. (156,655 points)
Dear Bryan, many thanks indeed for your prompt reply.
However, it seems to me that by doing in this way I am still not able to disentangle order effects (e.g. deriving by always choosing the first alternative) from the extra utility brought by the labels. In fact, the concept labels do not change their order across respondents. Moreover by not choosing a specific concept sorting option as you were suggesting,  I get random allocation of concepts across columns. Thus, I do not see how I can then figure out which product is shown in which column. I am scared I did not fully understand your point here.