Best-Worst CBC with paper and pencil

Question

I would like to conduct a Best-Worst CBC Conjoint where I will
generate the design with LightHouse and export/send the design
to an external programming vendor for survey programming/hosting.
The conjoint response data will need to be imported into LightHouse
via paper and pencil template. I have never conducted a
Best-Worst CBC Conjoint before (but have done Discrete Choice CBC) - so
have a few questions to make sure I understand correct steps:

(1) For design testing - do you proceed exactly the same as Discrete Choice CBC to test the design and use the same 0.05 standard error acceptable cutoffs for main effects?

(2) For the paper and pencil template - for each choice task are there 2 separate responses/columns? For example, if showing 3 concepts per task and (for task #1) left-most concept is selected as best and middle concept is selected as worst - you would enter for task 1 a "1" for chosen as
best and a "2" for chosen as worst?

(3) Once the conjoint response data is read into LightHouse - do you proceed with default HB utility estimation as you would with Discrete Choice CBC (OR do you need to do anything different/change any
software settings)?

(4) Will HB produce exactly the same outputs as with Discrete Choice CBC (attribute importance, level preferences, raw utilities) that can ALL be interpreted and used EXACTLY the same way for analysis and simulation
as with Discrete Choice CBC?

Thank you.

Answer 1

Yes, you'd use the same design testing procedure for best-worst CBC as for CBC.  The 0.05 standard error threshold is just a practical recommendation we've made based on our experience that leads to reasonably precise estimates of utilities for CBC.  If you're seeing 0.05 or lower standard errors for best-worst CBC estimates, that's a good target.  And, you should be able to accomplish that with a bit lower sample size than if using best-only CBC (best-worst contains more info).

Best-worst CBC paper-and-pencil templates indeed expect two response columns per task to receive the two answers.

Once you read the best-worst CBC data into Lighthouse Studio, you proceed as you normally would with HB estimation for best-only CBC.  

Same outputs as you see with standard best-only CBC.  

Of course, the utilities from best-worst CBC represent a compromise solution between the best tasks and the worst tasks.  It's possible that the preference strategies respondents seem to be following to make "best" choices could be different from the preference strategies respondents are following to make "worst" choices.  And, the response error could be different when making best choices vs. making worst choices.  Academics have published many papers on models that exploit those potential differences.  But, for practical simplicity, we just combine the two types of choice tasks and estimate a single model.  This seems to work well enough for practical consulting purposes.

Comments

To add a comment...when you conduct HB analysis within Lighthouse Studio for best-worst CBC, by default the software assumes you want to use both best and worst choices for the analysis.  You can change it to "best only" or "worst only" if you want to isolate the utilities from just half of the choices.  You do that under the CBC/HB Analysis settings, under the "Estimation Settings" link.  The setting says, "Tasks to include for best/worst data".

---

Thank you for the very helpful response.