Improvement of Choice-Based Conjoint Analysis through Likelihood to Purchase
Choice-based conjoint analysis is a widely used market research strategy to decipher the trade-offs people are willing to make in regards to a product or service.
There are a lot of strengths to this kind of conjoint model including imitating the scenarios that consumers face in the marketplace. The data is modeled by the attributes in the profiles shown as the explanatory variables and their selected choice as the response variable.
In this post elaborating on the specifics of CBC is not the purpose but rather how it can be improved through asking the respondent their likelihood to purchase the package they selected is.
The theory behind asking this follow-up question is to understand their level of preference for the package they selected. It gives us another dimension in preference.
Another way to think about it might be that just because the respondent prefers the profile to the others that it is stacked up against, it doesn’t mean that they would definitely purchase the package if it were in front of them.
It might just be the one they dislike the least.
The way we typically ask this question is to use display logic to present the question after they have selected one of the profiles in the CBC question. The reason to not show the likelihood to purchase question from the start is so the respondent can solely focus on the conjoint profiles and which one they most prefer.
Then after they have selected a conjoint profile, the likelihood to purchase question will be displayed directly under the conjoint choice packages.
The language for that question can be something along the lines of: “Based on the package you selected above, what is your likelihood to actually purchase the package?” The options for this question is usually some kind of Likert scale.
The interesting part of using this dual-data collection is to compare people’s utilities for the packages using their binary choice and their likelihood to purchase. In regular CBC analysis, the response variable is a ‘1’ if they selected the profile and a ‘0’ if they did not.
The modeling of this data will paint us a picture of the attributes that enhance or detract their preference. With their likelihood data, the response variable would be their scale point they selected.
This analysis will tell us the magnitude of their preference and we can model the revenue projections based on market size.