CBC utilities

Once responses to your CBC exercise are collected, preference scores (utilities) are estimated using Hierarchical Bayesian (HB) utility estimation. These utility scores represent the relative desirability or worth of each attribute level. Higher utility scores indicate a greater preference. You can view these scores in the Summary tab of the Exercise analysis.

To find the CBC utilities:

1. Go to Analysis.
2. Click on the Utilities tab for the desired CBC exercise.

Important note:
Utilities should only be compared within the same attribute, not between different attributes. For example, the utility of a level for one attribute should NOT be compared to the utility of a level from a different attribute. Since utilities are additive across attributes, the total utility of a product alternative is the sum of the utilities for its attribute levels, with one level chosen from each attribute.

Each attribute is displayed on its own chart (or table, if you switch to that view), showing the utilities for each level. In Discover, utilities are zero-centered, which has been the standard for decades.

Within each attribute the scores sum to 0. Due to the zero-centering of utilities within each attribute, even if all levels within the same attribute are considered excellent by respondents, some levels are made to be positive and others to be negative (so they sum to 0). A negative score does not mean that respondents dislike a level, it just means they like it relatively less than the other levels within the same attribute. The inverse logic also applies. A positive utility does not necessarily mean that respondents like that level, they just relatively prefer it to other levels in the same attribute that have lesser scores).

Example:

Consider these three levels and their zero-centered utilities:

Everybody likes to win money, so winning any of these three large amounts of money would be a great thing. But, winning $300,000 is better than $200,000, which is better than $100,000. Since the utilities are constrained to sum to 0 within each attribute, the least desirable of the levels has to be made negative. However, just because it has a negative score doesn’t mean that respondents wouldn’t like to win $100,000!

Zero-centered diffs scaling:

The raw utilities are multiplied by a constant so that the range of utilities for each attribute averages 100 across all respondents. This rescaling ensures that each respondent has nearly the same weight when computing average utilities across the sample.

Advanced note:

Even though utilities are rescaled to zero-centered diffs for reporting in Discover, if you want to export the data and create your own market simulator in Excel, you should use the raw utilities (instead of the zero-centered diffs) to ensure the market simulator math is accurate.

Below the attribute level utilities report, you'll find the Relative attribute importance report. These importances sum to 100 and show how much each attribute impacts choice. The importances are calculated at the individual level, with the displayed values representing the average for the entire sample.

Be cautious when interpreting the importance scores. They are directly related by the range of levels you include in your conjoint analysis. For example, if you use a price attribute ranging from 100 to 200, the importance of price will appear lower than if the price range were 100 to 300. Similarly, if you conduct one study with 3 attributes and another with 6 attributes, the importance scores for the second study will be, on average, half the size of those in the first study, since the importances are made to sum to 100 within each study.

Should you desire, you can check the option to display 95% confidence intervals.

Assuming your respondents are representative of the population from which they were drawn randomly, you are 95% confident that the true population's utility scores fall within this interval.

You can also use the 95% confidence intervals to assess whether one level is preferred to another within the same attribute. If the confidence intervals do not overlap between two items, you are at least 95% confident that one level is preferred to the other (again within the same attribute).

As a reminder, do not try to compare single attribute levels between different attributes. This is not proper or supported in conjoint analysis.

In the upper right-hand corner of the settings panel is a download menu (). Within the menu there are five files that are available for download:

• Summary (.xlsx)
• Charts (.png)
• CBC Design & Choices (.xlsx)
• Respondent Utilities (.xlsx)

In analysis it’s common to compare groups of people using their exercise scores. Segmenting allows you to take the results from a conjoint exercise and split them into groups. These groups come from different responses/values for one of your questions or defined variables.

For example, you could use a demographic question or variable in your survey (like the respondent’s location) to segment the results to see how people in North America responded versus people in Europe.

To apply segmenting, click the Segmenting dropdown and select a question or variable from the menu. When a variable is selected, the charts will automatically update and the Segmentation icon turns green. Attribute utilities and importance are now split into groups according to the selected variable. Respondents that don’t have any response (or value for defined variables) will be missing from the results.

To clear the segmenting or apply a different one, click the dropdown and select No segmenting or choose another option from the menu.