How can I use MaxDiff and Van Westendorp to create a Value Matrix for software feature selection?

“Value Matrix” is a technique sometimes used for feature selection for software bundles and plans. We do not express a view on validity of this technique and do not recommend using it for your projects.

The technique centres on building a 2x2 matrix, where each dot is a feature. The vertical axis is willingness to pay (measured through Van Westendorp) and the horizontal axis is “value” (measured through MaxDiff). The details of how it is created appears to be inconsistent from one analyst to another.

That said, it is possible to create this matrix using tools. For this, you might need two sets of data.

First, to identify respondents who valued one attribute in MaxDiff the most, there are a couple of ways:

  1. Through the simulator:

    1. In the preference share simulator, you can put all the items into one scenario and see shares of preference for each item.

    2. Under Advanced settings of that scenario, you can click Segregate. This will create a variable that you can use in segmentation.

    3. Under the Segmentation tab, you can add a new segment with a new condition “Segregation based on simulation” and pick the relevant scenario and item.

    4. Press Save and apply to take segmentation into effect.

  2. Through Excel export:

    1. Under the individual preferences tab of the Excel export, you see everyone’s preferences for separate items.

    2. Using Excel formulas, you can find which item was most preferred by each respondent (looking for the highest value).

Second, to see the “normal price point” in Van Westendorp, you can look at the standard outputs. The table on the right will have it listed:

Four intersections in Van Westendorp PSM

However, it will not provide confidence intervals. Van Westendorp results are not usually viewed in the context of statistical significance testing. There is no theory that backs up the application of the normal price point in practice. It is, of course, possible to calculate a confidence interval through computational statistics (bootstrap / jackknife).