MaxDiff, short for Maximum Difference Scaling, is a research method designed to uncover what people genuinely care about. Instead of asking respondents to rate items one by one, MaxDiff asks them to choose what matters most and least from a small set of options. This forces clear trade-offs and results in more reliable prioritisation.

MaxDiff works best when you want to understand relative importance rather than absolute opinions. Common use cases include:

Because respondents only choose the best and worst options, MaxDiff often feels easier and more engaging than long rating grids.

​

You can add a MaxDiff question to any survey by selecting Add question and choosing MaxDiff from the question list. MaxDiff questions can sit alongside other question types, allowing you to combine prioritisation with broader context or qualifying questions.

Each survey supports one MaxDiff question per audience. You can still include other questions before or after the MaxDiff task.

Options are the items you want respondents to prioritise. These might be:

You can add up to 25 options, though for best results we recommend between 8 and 15. Larger lists increase respondent effort and can reduce data quality. Image-based options are not currently supported.

If you are unsure how to frame your MaxDiff question or which options to include, Compass can help. By describing your research goal in plain English, Compass can suggest whether MaxDiff is the right approach and support you with drafting clear, comparable options before you launch.

The platform automatically calculates how many question sets are needed based on the number of options you add. This is designed so that each option is shown to respondents roughly three times. Each question set always shows four options.

When you preview your survey, you will see a short introduction explaining the MaxDiff task to respondents, followed by the best and worst selection screens. Previewing is a good way to sense-check clarity before launch.

Once your survey is complete, MaxDiff results appear in the results dashboard alongside your other questions.

For each option, Attest shows three metrics:

You can view these metrics in charts or crosstabs and switch between views depending on how you want to analyse or present the data.

Attest uses a respondent-based approach to calculate MaxDiff scores. For each respondent, we count how many times they selected an option as best and as worst. We subtract worst from best and divide by the number of times the respondent saw that option.

This produces a normalised score per respondent. We then average these scores across all respondents to produce the final MaxDiff score. This approach ensures that each respondent contributes equally, even if options were not shown the same number of times.

You can filter MaxDiff results by demographics or other questions in your survey to understand how preferences differ across segments.

You can also compare groups using crosstabs. Statistical significance testing is supported for % best and % worst values, using Fisher’s Exact Test. Significance testing is not currently available for the MaxDiff score itself, as it is a normalised average rather than a simple count.

MaxDiff charts and tables can be saved to a board to help you build a story and share insights with stakeholders. You can add your own commentary or use AI to help summarise key findings.

If you need to export results, you can copy MaxDiff data directly from charts or crosstabs and paste it into a spreadsheet. Raw MaxDiff exports and PPT exports are not currently supported.

For best results, we recommend a minimum of 400 respondents when running MaxDiff, especially if you plan to analyse sub-groups. Directional studies may work with smaller samples, but larger sample sizes provide more confidence when making decisions. If you plan to analyse specific segments, ensure each segment has at least 100–150 respondents.

When embedding MaxDiff as part of a larger research survey, tailor your sample size based on the intended use of the insights:

The larger the stakes, the more respondents are recommended for statistical precision and stakeholder confidence.