Analyze and report on Choice-Based Conjoint questions in modern reports
View and analyze part-worth utilities and attribute importance using the Zero-Centered Differences methodology.
Analysis of a Choice-Based Conjoint question is a time and resource intensive process, and it is not generated when you create your report. You must start the process manually.
The analysis uses a Hierarchical Bayes (HB) estimation method to calculate individual-level part-worth utilities for each respondent. These individual-level utilities power both the attribute importance outputs and the Conjoint Market Simulator. The key reporting outputs — the Zero-Centered Differences (ZCD) chart and table — are derived from these individual-level estimates.
Generate the choice-based conjoint analysis
To start working with a Choice-based Conjoint question in modern reports you first need to generate the analysis.
The analysis for Choice-based Conjoint questions is not run automatically when you create the report because the analysis is a resource intensive and time consuming process that may take up to one hour to complete.
You can generate the analysis for your conjoint question when the required number of responses have been received. If the minimum number of required responses is not collected, you cannot generate the analysis.
To meet the required number of responses:
- Only live data is included in the analysis. Test data is not processed.
- Only completed responses are included in the analysis. Incomplete responses are excluded.
To generate the analysis:
Choice-based Conjoint analysis results
A bar chart, table, and convergence plot display the analysis results.
You can switch between the chart or table which show the Zero-Centered Difference (ZCD), Importance, and Relative Importance for the attributes and levels in your Choice-Based Conjoint question. For more information, see Part-Worth Utilities and Zero-Centered Differences (ZCD). A convergence plot is also displayed, which you can use to check the statistical validity of the analysis.
Chart view
The chart view displays all attribute levels grouped and color-coded by attribute. Each level's Zero-Centered Difference (ZCD) value is displayed along the horizontal axis, representing how much that level contributes to consumer preference relative to the other levels of the same attribute.
Table view
The table view displays a crosstab with all levels grouped by attribute. For each level, the ZCD value is shown. For each attribute, the Importance and Relative Importance values are provided. Attributes are sorted from most important to least important.
Convergence plot
The convergence plot shows the stability and reliability of the statistical model over time. The plot tracks whether the repetitions of the algorithm have settled on a consistent answer.
You must check this chart for convergence before you proceed with your analysis. Otherwise you might be making business decisions based on "noise" rather than actual data.
If the plot shows massive swings even after thousands of iterations, it suggests your sample size might be too small, or your respondents were answering randomly (high error).
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 or part-worth for the population. Each colored line represents a different attribute level.
- Gray shaded area: This area on the left side of the chart is the "Burn-in" period. The algorithm is still searching for the right neighborhood for the parameters. These iterations are discarded because the model hasn't reached a steady state yet.
In a healthy model, you want to see the lines "settle down" into a horizontal, stable band after the burn-in period. A convergence plot that represents a successful estimation has a well-defined "fuzzy caterpillar" appearance. When the plot looks like a dense, horizontal "hairy caterpillar," it means the model has converged. The values are oscillating around a stable mean, suggesting the model has found the "true" estimate.
In general if the scale of the y-axis (Mu) is between -1.0 and 1.0 you can have a high degree of confidence that the model has converged.
The following table outlines how the convergence plot may appear, what it means, and action you can take to achieve statistically valid results for your 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 "bad" respondents. |
| Two lines far apart | Failed Convergence | Re-evaluate model constraints or data quality. This plot appearance represents failed convergence if the start of individual lines after the burn-in period differs drastically on the y-axis (Mu) from the end of the line. |
Part-Worth Utilities and Zero-Centered Differences (ZCD)
Learn about the values displayed in the choice-based conjoint chart and table views.
During the choice-based conjoint analysis, individual-level part-worth utilities are estimated for every respondent 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 studies, attributes, and respondents.
What is a Zero-Centered Difference?
A ZCD value represents how much a level contributes to consumer preference, expressed on a common scale where all attributes together sum to a fixed total. Because the scaling is applied to each respondent's utilities individually based on that respondent's own range of preferences ZCD values account for individual differences in how strongly or weakly people discriminate between options.
The ZCD for a level is calculated in two steps for each respondent:
Step 1 — Calculate a respondent-level multiplier:
Multiplier = (100 × number of attributes) ÷ (sum of all attribute ranges across all attributes)
Where each attribute's range is the highest part-worth level minus the lowest part-worth level for that attribute, for that respondent. This multiplier rescales each respondent's utilities so that all attributes together sum to 100 points per attribute on average.
Step 2 — Apply the multiplier to each raw part-worth:
ZCD for a level = Multiplier × raw part-worth utility for that level
The ZCD value reported in the chart and table is then the average of each respondent's scaled value across all respondents in the selected segment.
Interpreting ZCD Values
- A positive ZCD value indicates that consumers prefer this level more than the average level within that attribute.
- A negative ZCD value indicates that consumers prefer this level less than the average level within that attribute.
- A ZCD value near zero indicates that consumers feel relatively neutral about this level.
- To identify the most compelling product configuration, select the level with the highest ZCD value from each attribute and bundle them together.
- You can compare ZCD values for levels within the same attribute. Do not compare ZCD values for levels from different attributes.
Importance
Importance measures how influential an attribute is to consumers' overall product decisions. When evaluating a product, does brand matter more than scent, or price more than format? The importance value answers these questions.
Importance is calculated at the individual respondent level and then averaged across the sample. For each respondent, their importance score for an attribute equals their attribute range multiplied by their personal multiplier.
Respondent's attribute importance = (attribute range for that respondent) × multiplier
Where attribute range is the highest part-worth level minus the lowest part-worth level for that attribute for that respondent, and multiplier is the respondent's personal scaling factor described above. The importance value shown in the report is the average of these respondent-level scores across all respondents in the selected segment.
The higher the importance value for a given attribute, the more consumers prioritize that attribute in their decision-making. High importance values indicate the product dimensions to which you should devote more time, energy, and resources.
Relative importance
Relative importance expresses each attribute's importance as a percentage share of the total importance across all attributes, making it easy to compare which attributes matter most at a glance. Relative importance values always sum to 100%.
Relative importance = (Attribute importance) ÷ (Sum of importance across all attributes) × 100%
Relative importance aligns with absolute importance and reaffirms which attributes should be prioritized. Because the individual-level scaling ensures that all attributes together sum to a fixed total for every respondent, relative importance in the new version is a particularly stable and interpretable metric.
Example: Interpreting the Choice-Based Conjoint results
You survey laundry detergent shoppers and ask them to choose between products that vary by Brand, Scent, Format, and Price. Based on their individual-level choices, the HB estimation produces part-worth utilities for each respondent, which are then transformed into ZCD values for reporting.
The table below shows example ZCD and importance outputs:
| Attribute | Level | ZCD | Importance | Relative Importance |
| Brand | EverClean | 22.27 | 124.24 | 31.06% |
| Brand | PureBloom | 3.19 | ||
| Brand | Store Brand | -3.23 | ||
| Brand | SwiftWash | -22.23 | ||
| Price | $8.99 | 13.68 | 95.88 | 23.97% |
| Price | $12.49 | 10.28 | ||
| Price | $15.99 | -4.02 | ||
| Price | $19.49 | -19.94 | ||
| Format | Concentrated | 5.27 | 91.62 | 22.91% |
| Format | Liquid | -1.92 | ||
| Format | Single-Use Pods | -3.35 | ||
| Scent | Fresh Linen | 13.16 | 88.26 | 22.07% |
| Scent | Lavender Mist | 4.32 | ||
| Scent | Fragrance-Free | -17.48 |
Interpreting these results:
- Looking at the ZCD values by attribute, the ideal product configuration for the broadest consumer appeal would be: EverClean brand, $8.99 price, Fresh Linen scent, and Concentrated format which combines the highest ZCD level from each attribute.
- Brand is the most important driver of choice (importance: 124.24, relative importance: 31.06%), meaning consumers' brand preferences vary more widely than their preferences across other attributes. The gap between EverClean (22.27) and SwiftWash (−22.23) is the largest spread of any attribute, confirming that brand is where positioning and marketing investment are most likely to shift consumer decisions.
- Price and Format are the second and third most important attributes, and notably close in relative importance (23.97% and 22.91% respectively). Price shows strong negative sensitivity at the high end. The ZCD drops sharply from 10.28 at $12.49 to −19.94 at $19.49, signaling that premium pricing is a significant deterrent for most shoppers.
- Scent is the least important attribute (88.26 / 22.07%), but the spread is still meaningful. Fresh Linen (13.16) and Fragrance-Free (−17.48) sit at opposite ends, suggesting that while scent is not the primary decision driver, a Fragrance-Free positioning carries real risk of alienating the majority of shoppers.
- The four attributes are notably balanced in importance — ranging from 22.07% to 31.06% — meaning no single attribute dominates consumer choice. This suggests that product optimization requires attention to the full bundle of attributes rather than focusing exclusively on any one dimension such as price or brand.