Conjoint: Personalized Trade-offs Make Conjoint Better
Adaptive Choice-Based Conjoint (ACBC) is one of the newer conjoint methodologies. Think of it as a combination of Choice-Based Conjoint (CBC) and Adaptive Conjoint Analysis (ACA), yielding the familiar choice task of choosing from a group of profiles (as in CBC), and a customized, interactive survey experience that “learns” from previous responses to make choice options as relevant as possible (as in ACA). The result is a more relevant respondent experience.
Research And Experience With ACBC Has Shown The Following:
The interview is perceived as more engaging and relevant to respondents
The interview takes longer than traditional CBC
Hit rates (the ability to correctly predict an outcome) are higher in ACBC than CBC
When Should You Use ACBC?
Your study has 5 or more attributes. For very simple studies, traditional CBC will work fine, however, once you reach 5 attributes (e.g., brand, feature A, feature B, feature C, price), moving to ACBC will provide a better experience for respondents and better results for you.
Price is an attribute. Remember that traditional CBC is great for pricing research. ACBC takes the best parts of CBC’s ability to research price and makes them even better (see discussion on summed pricing, below)
You’re not trying to do too much else in your survey. In other words, you’d got the time for the ACBC exercise, which could take about 10 minutes. We have found that screening, profiling, and a few other questions can be placed in the survey with a relatively large number of attributes, and we can stay under 15 minutes.
Sample size: Usually, it’s “the more the merrier.” Having a larger sample size gives the ability to finesse some of the modeling and, of course, slice the data in more ways. However, ACBC also works with very small sample sizes. Because Hierarchical Bayes (HB) statistics underlie the model, each and every respondent has their own personal model, thus enabling a valid look at even the smallest groups.
More on individual-level modeling via HB. Individual level statistics can be carried forward into outside analyses such as segmentation analysis. The power of carrying such valid choice data into segmentation modeling is great – think about how much better preference results from conjoint are than simple rating scales.
How It Works
Build Your Own (BYO) Configurator
The respondent chooses from all the levels of all the attributes to “build” their ideal product or service
If “summed pricing” is used (see description below), the respondent sees incremental price differences as they choose levels of each attribute.
Screeners
A number of profiles are created that are similar to the profile built in the configurator stage
The respondent indicates whether or not the profile is a possibility
Must-Haves And Unacceptables
Based on choice patterns, respondents are asked whether certain levels must be included in their preferred solution, or if they should not be included (they are unacceptable).
Identifying these “non-compensatory” decision-making criteria (see description below) helps create an even more relevant respondent experience.
Choice-Tournament
Like traditional CBC, the respondent is shown the surviving possibilities from the screening task in sets of 3 profiles at a time and chooses the winner.
Calibration Section (Optional)
Displays BYO concept, choice task winner and about 4 other previous concepts and asks for likelihood to purchase on a 5-point scale.
Used if you have secondary data sources regarding likelihood to purchase on a Likert scale – helps establish the “none” threshold.
New Benefits
One of the great benefits of ACBC is its ability to handle non-compensatory decision making.
A non-compensatory screening approach occurs when a respondent sets a minimum (or maximum) acceptable value (cutoff, standard, or threshold) for a certain attribute. The respondent is specifying that they “must-have” something, or that it is “unacceptable”.
The result is a more relevant set of choice tasks, unique to the individual respondent.
Summed pricing provides the most realistic price-feature matching possible. When a profile is “loaded” with high-end features, price is high; low-end products have lower prices.
This precision is made possible by assigning incremental prices for up to all the levels of all the attributes in the study.
Even more precision is possible by using “price adjustments” that vary the incremental price added to, for example, a specific brand. This technique allows real-market brand premiums to be better displayed in the exercise.
Price is then varied by a random amount (usually +/- 30%), which helps test the boundaries of possible prices.
Thousands of prices are shown, allowing modeling of price as a continuous (numeric) variable (as opposed to the constrained categories shown in traditional CBC)
The result improves our ability to independently assess other attributes (decoupling them from price).
For more information about how we might help you with an ACBC project, contact us at contactus@elucidatenow.com.