Analyze and report on MaxDiff questions in modern reports

Learn more about the different data views and analysis tools for MaxDiff questions in modern reporting.

The MaxDiff tile in a modern report can have up to three views or tabs:

1. The Summary view displays preference scores for each MaxDiff attribute, automatically sorted by rank and overall score.
2. The MaxDiff Analysis view features a convergence plot and Zero-Centered Differences (ZCD). These ZCD scores are derived from part-worth utilities, ensuring all attributes are evaluated on a standardized scale.
3. The TURF Analysis view displays total reach and frequency results for the MaxDiff question. The system uses the first-choice method to convert part-worth utilities into a TURF-compatible dataset before processing.

   The TURF Analysis view can only be turned on after the MaxDiff Analysis view is turned on, and MaxDiff analysis has finished running.

Parent topic: MaxDiff analysis

Summary view

The Summary view is available without turning on any toggles, and before the minimum number of completed live responses required for MaxDiff Analysis is reached. This means you can keep checking the summary as new responses come in.

Each row in the summary table represents an attribute. For each attribute, the following information is displayed.

Column	Description
Most	The number or percentage of times that the attribute was chosen as most important.
Did not choose	The number or percentage of times that the attribute was seen by participants, but not chosen at all.
Least	The number or percentage of times that the attribute was chosen as least important.
Rank	Rank is derived from Score. The attribute with the highest score (that is, the most important attribute) is ranked first, followed by the lower-ranking or less important attributes.
Score	The Score for an attribute is calculated as: Score = (number of times the attribute was selected Most − number of times the attribute was selected Least) ÷ (number of times the attribute appeared) Score values can range from −1 (always chosen as least preferred) to 1 (always chosen as most preferred).

Switch to the Stacked Bar view to see your MaxDiff data displayed in a different way. The stacked bar chart uses the first three colors of your modern report theme.

Generate the MaxDiff analysis in a modern report

The MaxDiff tile shows how many completed live responses are required before you can generate the MaxDiff Analysis views. Once you have enough responses, you can start.

Note: Only completed live responses count toward analysis. MaxDiff analysis excludes incomplete responses from live data, and does not work with test responses.

1. Open the modern report and go to the MaxDiff question tile.
   When enough responses are available, Generate Analysis appears on the tile. Otherwise, the tile explains how many completed responses are still needed.
2. Click Generate Analysis.
   Analysis can take noticeable time depending on sample size and design. You can continue working elsewhere and return later. When results are ready, the Summary view updates and you can enable MaxDiff Analysis.
3. Optional: To rerun analysis after new responses arrive, refresh report data if needed, then click Regenerate Analysis in the tile.

MaxDiff Analysis view

Once the MaxDiff analysis finishes running, you can view a Zero-Centered Difference (ZCD) chart and table, and a convergence plot.

MaxDiff analysis uses the same Hierarchical Bayes (HB) estimation method as Choice-Based Conjoint, applied to the single list of attributes you defined for the MaxDiff question. HB calculates individual-level part-worth utilities for each participant. The Zero-Centered Differences (ZCD) chart and table are derived from those individual-level estimates and give a more precise, statistically robust picture of consumer preference than the count-based score available in the Summary view and in standard reports.

For background on authoring MaxDiff and experimental design, see Create a MaxDiff question and How the MaxDiff experimental design works.

ZCD chart view

The chart view shows one bar per attribute. Each attribute's Zero-Centered Difference (ZCD) value is plotted along the horizontal axis. Attributes are listed in authored order.

ZCD table view

The table view lists each attribute in authored order alongside its ZCD value. You can apply a banner to the ZCD table to break the values out by segment — for example, by age range or country — so you can compare how preferences differ across groups.

Note: When you change the selected banner, the platform refilters the analysis results that have already been calculated. The MaxDiff analysis itself is run on the full qualifying response set; banners segment the output, they do not retrigger analysis.

Tip: Apply banners to the ZCD table to compare segments.

Convergence plot

The convergence plot shows the stability and reliability of the HB estimation over time. The plot tracks whether successive iterations of the algorithm have settled on a consistent answer.

Check this plot for convergence before you rely on the ZCD outputs. Otherwise you might be drawing conclusions from "noise" rather than a stable estimate. Large swings even after thousands of iterations can suggest that the sample size is too small, or that respondents answered randomly (high response error).

Filters and the report's custom theme do not impact the convergence plot. The plot reflects the underlying estimation, not the filtered view of results.

The components of the convergence plot are the:

• X-axis (Rep): The number of repetitions the algorithm has run.
• Y-axis (Mu): The mean utility (part-worth) for the population. Each colored line represents a different attribute.
• Gray shaded area: The "burn-in" period on the left side of the chart. During burn-in, the algorithm is still searching for the right neighborhood for the parameters. These iterations are discarded because the model has not reached a steady state yet.

In a healthy model, the lines settle into a horizontal, stable band after the burn-in period. A successfully converged plot has a "fuzzy caterpillar" appearance: a dense, horizontal band where the values oscillate around a stable mean.

The following table outlines how the convergence plot may appear, what it means, and what action to take to achieve statistically valid results for your MaxDiff analysis.

Plot appearance	Status	Action needed
Stable, horizontal band	Converged	Proceed with analysis.
Strong trend (up or down)	Failed convergence	Re-evaluate model constraints or data quality.
Occasional spikes	Unstable	Check for outliers or low-quality respondents.
Two lines far apart	Failed convergence	Re-evaluate model constraints or data quality. This appearance signals failed convergence when the start of individual lines after the burn-in period differs drastically on the y-axis (Mu) from the end of the line.

What is a Zero-Centered Difference?

During MaxDiff analysis, individual-level part-worth utilities are estimated for every participant using Hierarchical Bayes (HB) estimation. Rather than reporting raw part-worth values directly, the report transforms these utilities into Zero-Centered Differences (ZCD). ZCD is a scaled version of the part-worth that makes values easier to compare across attributes and participants.

A ZCD value represents how much an attribute contributes to consumer preference, expressed on a scale that is centered on zero. Because the scaling is applied to each participant's utilities individually based on that participant's own range of preferences, ZCD values account for individual differences in how strongly or weakly people discriminate between attributes.

The ZCD for an attribute is calculated for each participant and then averaged across the participants in the selected segment.

Interpreting ZCD values

• A positive ZCD value indicates that consumers prefer this attribute more than the average attribute in the list.
• A negative ZCD value indicates that consumers prefer this attribute less than the average attribute in the list.
• A ZCD value near zero indicates that consumers feel relatively neutral about this attribute compared to the others.

Tip: ZCD and the count-based Score in the Summary view answer related but different questions. Score reflects how often each attribute won or lost in the sets it appeared in. ZCD reflects each participant's underlying preference structure as estimated by HB and is generally more stable, less sensitive to which attributes happened to appear together, and better suited to segment comparisons.

Run TURF analysis on MaxDiff question data

After the MaxDiff analysis completes, navigate to the TURF Analysis tab to run your analysis. The system automatically formats your part-worth utilities into a structured TURF dataset using the first-choice method.

What is the first-choice method?

For each participant, the single attribute with the highest part-worth utility is counted as selected. Every other attribute (even ones with high positive part-worth values) is counted as unselected.

The first-choice method produces a more conclusive result by ignoring borderline values. If we were to count all attributes with positive part-worth values as selected, the ones hovering between 0 and 1 could be problematic because they are right at the threshold between being counted as selected and unselected. The part-worth utilities are so low on these borderline values that they could easily flip if a participant retakes a survey. By only focusing on the top attribute, the first-choice method eliminates the wobble problem and reduces ambiguity.

The first-choice method is used to convert MaxDiff utilities into a TURF dataset. However, this does not change the underlying TURF algorithm.

Note:

• Use options on the TURF Analysis tab to omit attributes from combinations rather than adding the MaxDiff question to report filter criteria in ways that conflict with your analysis goals.
• If you change report filters after running TURF analysis, refresh report data and use Regenerate Analysis on the TURF Analysis tab when the UI presents it.

1. In the MaxDiff question tile, open the options menu and select Tile Settings.
2. Click the TURF Analysis tab.
3. Turn on Show TURF Analysis.
   The TURF analysis settings appear.

4. Configure your TURF analysis settings.
   1. Under Exclude These From Analysis, select the options you want to omit entirely from TURF analysis.
      Important: To ensure the most accurate TURF analysis results you must exclude Did not answer responses.

      You can exclude any other options you want. For example, you may want to exclude Other - specify answer options.

      When you use this setting to exclude options, the TURF analysis treats the excluded options as though they do not exist. They are not included in the overall tabulation of selected responses. They are not considered in frequency calculations. Use this setting to exclude any answer options that may skew TURF analysis if they are included. That way, they aren't considered alongside the options you want to evaluate.
   2. Set the Combination Size.
      This setting determines the maximum number of options included in a solution. For example, if you set it to 4, the resulting TURF analysis will show you possible combinations in four different sizes: the best combination featuring one option, the best combination featuring two options, the best combination featuring three options, and the best combination featuring four options.
      Combination size can range from 0 to 20.
   3. Optional: Drag and drop Available Options for Analysis into the Always Include All Of These column.

      This setting determines the options that will always be included in a combination.

      Note that if you have options that will always be included, you may have empty rows in your results table. For example, if you have three options that are always included, there will be no best combination featuring one option or best combination featuring two options because, at minimum, the combination must have the three options you specified.

   4. Optional: Drag and drop Available Options for Analysis into the Always Exclude All Of These column.
      This setting determines the options that will be included when TURF analysis runs, but excluded from the resulting combinations. For example, let's say you have issues with a vendor who supplies two of the options you're evaluating. You can still include both options in the TURF analysis and consider them relative to other options, but you don't want the combinations to feature either of them.
   5. Select or clear Allow Dropping Options.
      Selecting this option allows the TURF analysis to swap options in and out to find the most optimal combinations at every combination size. We recommend selecting this option.
   6. Select or clear Include Alternatives.
      Selecting this option allows the TURF analysis table view to display alternative combinations at every combination size that come close to the best of combination in terms of reach and frequency. If you select this option, type the number of alternatives to include.
5. Click Apply.
6. When the analysis finishes, you can interpret the results in the table and graph.

Related information

• TURF analysis