CBC or ACBC for optimal messaging

Question

Hello and thank you in advance for your thoughts!

We are looking to optimize the components of an advertising message.
We have 5-6 attributes with 3-4 level each.  Each level is simply a nuance in wording.

We are thinking CBC would be best as there may be interaction effects that would be missing in ACBC.

Please let me know your recommended approach.

Thanks again!

Answer 1

I'm not sure about which would be the best approach for you...but I don't understand your comment that interaction effects would be missing in ACBC.  ACBC has proven to be just as useful or perhaps even more so for capturing interaction effects between attributes taken 2 at a time as CBC.

Comments

Thank you.  We are trying to identify the words and phrases that would be oncluded in a communication to optimize an advertising message.  Typically we use MaxDiff and TURF but the some team members prefer conjoint.  Of all methods, which would you recommend?  We will have a robust sample size.

---

So much depends on whether you think the interaction effects (synergies between pairs of phrases) are due to correlations in respondent preference (e.g., the same people that like phrase 1 also are motivated by phrase 7) vs. within-person interaction effects (each person doesn't think phrase 1 goes well with phrase 7).

If you think the interactions are of the first type, then MaxDiff would seem to work fine.  If you think they are more of the second type, then a conjoint methodology (where respondents evaluate the overall "like" for a combination of phrases) would work better.

However, sample sizes to detect the interaction effects need to be larger than if you were only after main effects.   For example, think about two 3-level attributes.  If thinking about main effects, you are dividing your concepts shown by 3.  If thinking about pairs of attributes (2-way interaction effects), you are dividing your concepts shown by 9.   To get equal precision for interaction effects would involve collecting 3x as much data.