Number of CBC Questions
With Discover-CBC, the goal is to obtain enough information for accurate individual-level preference score estimation.
Based on our experience and statistics involving standard error computation, Discover-CBC examines your attribute list and recommends an appropriate Number of CBC Questions and Number of Concepts per Question.
You can change the recommended Number of CBC Questions. Clicking Use custom value shows the recommended range (a low and high value) and you can specify a different number. Decreasing the number of questions reduces the information available to estimate accurate preference scores.
If the recommended range for Number of CBC Questions exceeds what you think respondents can answer without getting tired and giving you bad data, then you should consider the following:
•Reducing the number of attributes or number of levels per attribute will reduce the recommended Number of CBC Questions.
•If you have a relatively large sample size (often in the 100s of respondents), then you can count on individual errors tending to cancel out due to the benefit of large numbers of respondents. You may decide you can afford to reduce the accuracy at the individual-level due to the abundance of respondent records.
•For the purposes of your research, you may decide that you are willing to use fewer questions than the recommended range and are willing to accept lower precision for preference score estimates.
Number of Concepts per Question
Based on the largest number of levels in any one of your attributes, Discover-CBC recommends the Number of Concepts per Question (aside from the "None" concept).
From a statistical viewpoint, showing more concepts per question is better for obtaining precise preference scores. However, showing more concepts per question makes it harder for humans to answer the questionnaire.
Generally, we suggest asking at least three Concepts per Question (in addition to the "None" concept). However, in certain cases such as emotionally-laden decisions involving different cancer treatments, it may be extremely difficult for respondents to consider more than two Concepts per Question.
If you change the Number of Concepts per Question, this may change the recommended Number of CBC Questions. As long as you didn't override the default Number of CBC Questions with a custom value you'll see the new recommendation.
Minimize Level Repeating (Level Overlap)
It is usually helpful to have a modest degree of Level Overlap within each CBC question, which Discover-CBC does by default. To illustrate the issue over Level Overlap, consider the following choice task:
Notice the brand attribute has no repeating levels across the three concepts. The weight attribute repeats the level "10 Kilos" (shown in Concepts 1 & 2). The shape attribute also repeats the level "Square" (shown in Concepts 2 & 3).
By default, Discover-CBC selects levels to display within product concepts leading to a modest degree of Level Repeating, as with weight and shape above. Much research has shown this to be an effective way to probe respondents' preferences.
In this example, Brand has minimal (no degree of) Level Repeating (no brand level is repeated within the same question). You may decide that sometimes it just looks more realistic within your choice tasks to minimize the Level Repeating for an attribute. If that is the case, check the box for any attributes that you want to minimize the level repeating.
Even if you click the box to Minimize Level Repeating for an attribute, if you are showing more Concepts per Question than levels for that attribute, Discover-CBC obviously has to repeat at least one of the levels to display the requested Number of Concepts on the screen.
Remove Ratings Questions
By default a CBC ratings question will be asked for each attribute that has "Level Preference Order" set to "Not Ordered" (attributes such as brand or color). You can optionally remove the ratings questions. However removing the ratings questions may lead to lower quality utility estimates! The rating questions:
1.Allow Discover-CBC to avoid designing choice tasks wherein one concept is logically preferred to another (dominated concepts).
2.Allow Discover-CBC to constrain the utilities during utility estimation so that they conform to each individual respondent's stated rank order of preference for levels within non-ordered attributes like brand, color, or style.
If you decide to remove the ratings question, you can counteract the loss of information via some combination of the following:
1.Increasing the number of choice tasks per respondent.
2.Increasing your sample size.
3.Exporting the data to a .CSV file for estimation under hierarchical Bayesian estimation (CBC/HB software).