MaxDiff anchoring

In a MaxDiff exercise, respondents evaluate small groups of items and select their best and worst options. These choices reflect relative preferences — but relative judgments alone can be misleading.

For example, a respondent who doesn't eat sugar may still select a best and worst ice cream flavor, even though they wouldn't purchase any of them. Their choices could resemble those of an enthusiastic ice cream fan despite very different real-world behavior.

MaxDiff anchoring addresses this limitation by introducing an absolute reference point. Anchoring helps distinguish whether items are genuinely important or unimportant, or whether respondents would actually buy or consider them, rather than only how items compare to one another.

Anchoring addresses this by introducing an absolute reference point, helping distinguish whether items are genuinely important or something a respondent would actually consider buying, rather than only how items compare to each other.

Anchoring appends a grid question to the end of the MaxDiff exercise. The grid includes a subset of items spanning from each respondent's most to least favored options based on their earlier choices. Respondents then indicate which items they would actually buy or consider important.

Results are displayed with a utility boundary line separating important from unimportant items. Items above the line have positive utility (e.g., buy), and items below have negative utility (e.g., not buy). Responses to the anchoring question and the MaxDiff results are used together to estimate this boundary.

When anchoring is enabled, the exercise is analyzed using anchored MaxDiff methods rather than standard MaxDiff analysis.

This implementation, known as the Direct Binary Approach Method, was proposed by Kevin Lattery and has been presented at several Sawtooth Software conferences.

1. Open the Advanced tab.
2. Toggle on MaxDiff anchoring.
   

1. Enter the anchor question text, including any rich text, images, or formatting.
2. Specify the “Important” and “Not important” labels.
3. Optionally, override the number of items shown per respondent. By default, 7 items are shown (or all items if the exercise contains only 6). A minimum of 5 is required. As the number of items in the exercise grows, consider showing more anchoring items to help the algorithm estimate a more accurate utility boundary.

 

Anchored scores are displayed on a positive probability (ratio) scale, with the anchor indexed at 100:

• 50: The item is half as preferred or important as the anchor.
• 100: The item has an equal likelihood of falling above or below the anchor.
• 200: The item is twice as preferred or important as the anchor.