If you work in or around consumer insights, you have most likely used or at the very least heard of “Conjoint” techniques. The literature on the topic is intensive and can be overwhelming for those who are not that familiar with it.
While we can automate your curiosity and research projects with the SightX platform, we still need to write, learn, and study about these topics as researchers. Hopefully, this write-up will bring some additional clarity for you and add another experiment to your research tool kit.
What follows are some of the most pressing questions we often hear with answers that will keep you from sinking into a technical academic paper.
When it comes to decision making, consumers often choose from a variety of alternatives. Those alternatives are often referred to as product attributes, and each attribute can have several levels.
Do you want to know how consumers value different attributes that make up an individual product or service?
To illustrate, here is an example: You are considering purchasing a car (product), one of the attributes can be the color of the car, and the choices of color are referred to as attribute levels (e.g. grey, white, black). A second potential attribute can be the price point of the car. The manufacturer may have their stock price and the fully loaded option ($24K vs. $39K). A third attribute could be the energy type (e.g. electric, gas, or hybrid).
What is the best combination of the three to offer consumers that will result in the biggest product adoption or sales?
The ultimate goal of the conjoint analysis is to measure consumer preferences for these different product features, or to learn how changes to price affect demand for products or service, and even to forecast the likely acceptance of a product if brought to market.
The most popular conjoint analysis is Choice-based Conjoint Analysis (CBC) which is also known as Discrete Choice Modeling (DCM). The purpose of conjoint analysis is to assess how consumers evaluate a particular product, and the tradeoffs they conduct across various attributes and their respective levels.
For example, would consumers be more likely to express higher likelihood of purchasing a medium-priced electric car with the color grey rather than higher priced gas car with the color white? Or is it some other set of alternatives?
Conjoint analysis derives estimates on the utilities of the product’s attribute levels (aka part-worth).
While there are several estimation models such as the Ideal Point model and Vector model, the most widely used model is the part-worth model, which unlike other models, it does not make any a priori assumptions regarding the utility caused by a specific level of any attribution. In other words, the outcomes will be a more accurate depiction of the consumer preference.
The part-worth is an estimate of the overall preference (or utility) associated with each level of each attribute used to define the product or service you are researching.
There are several statistical techniques to estimate part-worth, and the most widely used techniques with the most accuracy are Latent Class Analysis, and Regression modeling based on Hierarchical Bayesian (HB).
The values of the part-worth utilities provide information on how attractive attribute levels are. Remember the example of car types across various price points and colors?
If you want to know the relative importance of each attribute, then the researcher needs to calculate Attribute Importance via determining how much difference each attribute can make in the total utility of a specific product. That difference is the range in the attributes’ utility value.
You calculate percentages from relative ranges, obtaining a set of attribute importance values that add up to a 100. The higher the percentages are, the more important it is.
In the end, you should be able to discern which features go together for the most impact. (Ideally, you are working with a software platform that automates this process for you.)
In addition to those insights, sometimes insights leaders are interested in taking these utility scores and using them to simulate market share, where a market simulation provides information on the relative share of respondents who prefer predefined products in a certain context.
This enables researchers with the ability to test various market scenarios and assess things such as price demand curves, or the impact of product adjustments, and competitiveness between various products.
The first step in conducting a market simulation begins with specifying relevant products. Then the total utility of these products is computed at the individual or target level. The total product utility is the sum of its part worth utilities (Green and Krieger, 1988). From there, consumer insights leaders can, for example, compare the total utility value of products to that of a “none of the above” option.
The higher the difference between the total utility score of an alternative and the utility of “none of the above” option, the more likely it is for users to accept the alternative. Therefore, a product’s total utility below the “none of the above” option is the value indicating that the users are not likely to accept the offer.
Researchers can easily apply the logit model to estimate market share. Market share is predicted by simply exponentiating the total utility of a product and then dividing this value by the sum of all products’ exponentiated values and that “none of the above” option. Here is the formula:
If you are a bit scared of formulas, like most people, all the formula asks you to do is what is mentioned in the last paragraph “market share is calculated by exponentiating the total utility of a product and then dividing this value by the sum of all products’ exponentiated values and the “none of the above” option”.
The calculation can be done using a basic calculator with the “EXP” (or ex) button on it.
Below is an example with total utility scores, exponential scores, and the market shares associated with it.
|Total Utility||Exp. (total)||Market Share|
If any of the above needs further explanation, or you’d like to see conjoint analysis in action and discuss how it can help your research, please reach out at any time!
To read more technical papers, and/or if you want dive into this or any other research topic in more depth, please contact SightX at firstname.lastname@example.org.