SPSS command: SPSS CONJOINT
Conjoint analysis: part-worths, importance and simulations
Conjoint analysis estimates how much each attribute level adds to a product's appeal from respondents' ratings or rankings of whole profiles.
Free · No account · Files stay on your device · Updated 18 October 2026
What it tells you
Crosstabs runs the SPSS CONJOINT procedure in your browser on a plan from ORTHOPLAN or a plan file, giving part-worth utilities with standard errors, importance values, fit statistics and preference simulations per respondent and pooled.
- The part-worth utility of every attribute level, with a standard error, for the group and for each respondent.
- How important each attribute is: its utility range as a share of all ranges, averaged over respondents.
- How well the model reproduces preferences (Pearson's R, Kendall's tau) and predicts holdout cards.
- Predicted shares for products you define, under maximum-utility, Bradley-Terry-Luce and logit rules.
When to use it
Use it when
- Respondents rated or ranked full product profiles built from an orthogonal plan.
- Pricing and packaging decisions: how much a feature is worth relative to price, and which configuration wins share.
- Segmenting customers on their individual utilities (export them and cluster).
Use something else when
- Choice-based conjoint (respondents choose one option from each choice set): that needs a multinomial logit or hierarchical Bayes model, not OLS on ratings.
- You need interactions between attributes; the model is additive main effects only.
- The design cards are not orthogonal or are too few for the parameters: the panel refuses a design with no residual degrees of freedom.
How to run it in Crosstabs
- 01
Open your data
Open a .sav, CSV or Excel file in the workspace and switch to Statistics mode. The file is read in your browser.
- 02
Choose Conjoint analysis and the plan
In the Classify group, pick Conjoint analysis (CONJOINT). Under Plan source choose Plan file (CSV with CARD_ and STATUS_) and upload it, or Plan generated with ORTHOPLAN in this session.
- 03
Pick a model per factor
For each factor set Model to Discrete, Linear (numeric levels), Ideal or Anti-ideal (numeric, at least three levels). For linear factors set Expectation to None, More is better or Less is better to flag reversals.
- 04
Describe the responses
Under Data are choose Scores (higher = more preferred), Ranks (1 = most preferred) or Sequence (card numbers in preference order); pick a Subject identifier and tick one response column per rated card in card order — Select card1…cardN columns does this for you.
- 05
Add simulated products
Plan cards with STATUS_ = 2 are simulated automatically; use Add product under Simulated products to define more, then click Run conjoint analysis.
- 06
Review individuals and export
Choose a respondent under Individual subject to see their utilities and fit, and use Download utilities for every subject (CSV) for segmentation. Weights are not used: each subject counts once.
Worked example
Computed by the Crosstabs engine from the synthetic data below when this page was built — not typed in.
Data
The 9-card meal-kit plan from the ORTHOPLAN example plus its 4 holdout cards, rated 1–10 by 24 synthetic respondents. Each respondent's ratings come from individual part-worths scattered around a common pattern (cheaper is better, weekly delivery and a free first box help) plus rating noise, rounded to whole scores; seed fixed. Price is modelled as Linear with “Less is better”; the other factors are Discrete. Three products are simulated.
| Factor | Level | Utility Estimate | Std. Error |
|---|---|---|---|
| Price per box | 49 | −6.057 | 0.320 |
| Price per box | 59 | −7.293 | 0.386 |
| Price per box | 69 | −8.529 | 0.451 |
| Menu style | Classic | 0.255 | 0.075 |
| Menu style | Vegetarian | −0.231 | 0.075 |
| Menu style | Family | −0.023 | 0.075 |
| Delivery | Weekly | 0.351 | 0.057 |
| Delivery | Fortnightly | −0.351 | 0.057 |
| Trial offer | None | −0.472 | 0.057 |
| Trial offer | First box free | 0.472 | 0.057 |
| (Constant) | 13.273 | 0.390 |
Scores: higher values mean more preferred. Group results pool all 24 subjects; standard errors use the pooled residual variance. Price is linear, so its utility is B × level.
| Factor | Importance |
|---|---|
| Price per box | 48.274 |
| Menu style | 18.549 |
| Delivery | 15.282 |
| Trial offer | 17.895 |
Averaged importance: each subject's utility ranges as a percentage of their total, then averaged.
| Item | Value | Sig. |
|---|---|---|
| Pearson's R | 0.998 | <.001 |
| Kendall's tau | 0.930 | <.001 |
| Kendall's tau for Holdouts | 0.667 | .087 |
Correlations between observed and estimated preferences; significance is one-tailed.
| Product | Maximum Utility (%) | Bradley-Terry-Luce (%) | Logit (%) |
|---|---|---|---|
| Budget weekly | 79.2 | 39.3 | 66.3 |
| Premium trial | 12.5 | 29.5 | 16.3 |
| Mid fortnightly | 8.3 | 31.2 | 17.4 |
24 of 24 subjects have all-positive predicted totals and are used for the BTL and logit models.
How to read it
Across 24 respondents, price per box matters most (48.3% of importance), followed by menu style (18.5%). The pooled linear price coefficient is B = −0.124 per currency unit, and 0 of 24 respondents show a price reversal. The group model reproduces the mean ratings with Pearson's R = 0.998 and ranks the holdout cards with Kendall's τ = 0.667. In the simulation, “Budget weekly” wins the largest first-choice share (79.2% under maximum utility).
Output tables explained
- Utilities
- Group part-worth utility and standard error for every factor level plus the constant, estimated from the mean responses with the pooled residual variance.
- Importance Values
- Averaged importance per factor: each respondent's utility ranges as a percentage of their total, then averaged; also drawn as a bar chart.
- Coefficients
- For linear factors the slope B, and for ideal or anti-ideal factors γ1 and γ2, with standard errors.
- Correlations
- Pearson's R and Kendall's tau between observed and estimated preferences, and Kendall's tau for holdouts when there are at least three holdout cards; significance is one-tailed.
- Number of Reversals
- How many respondents have a coefficient whose sign contradicts the declared expectation, or an ideal point that is actually a minimum.
- Preference Probabilities of Simulations
- Predicted share for each simulated product under maximum utility, Bradley-Terry-Luce and logit; BTL and logit use respondents whose predicted totals are all positive.
Assumptions and notes
- Preferences are additive in the attribute levels (main effects only) and are measured on an interval scale for scores; ranks are reversed so that higher means preferred.
- There are more design cards than parameters, so each respondent's model has residual degrees of freedom.
- Respondents with a missing or invalid response for any rated card are excluded, and the panel says how many.
How the results are validated
Reference values come from statsmodels OLS on SPSS-style effect-coded designs and SciPy correlations; for all-discrete factors, the utilities also agree with R's conjoint::caUtilities to about 4e-14. For score, rank and sequence data, tests require every respondent's and the group's utilities, standard errors, constant, linear and ideal-point coefficients, Pearson and Kendall correlations (including holdouts) and simulated shares to agree within 1e-9, and importance values within 1e-8. Validation evidence
Questions
How should my data be laid out?
One row per respondent and one column per rated card (design plus holdout cards) in card order, holding the score or rank; for sequence data the columns hold card numbers from most to least preferred. The plan itself comes from ORTHOPLAN or a CSV with CARD_ and STATUS_.
Why are the price utilities all negative?
With a linear model a factor's utility is B × level, so for a negative slope every level is negative and the constant absorbs the offset. Compare differences between levels, not their signs; discrete factors are effect-coded and sum to zero.
How is importance calculated?
For each respondent, a factor's importance is its utility range divided by the sum of all ranges; the reported value is the average over respondents. It depends on the levels you chose — a wider price range makes price look more important.
Which simulation rule should I report?
Maximum utility counts each respondent's first choice and suits big-ticket decisions. BTL and logit spread each respondent's preference across products and suit frequently bought items; logit is the most sensitive to utility differences.
What is a reversal?
A respondent whose linear coefficient has the wrong sign for the expectation you declared (for example preferring higher prices), or whose ideal point is a minimum. Many reversals suggest careless responding or a misspecified expectation.
Related procedures
- Orthogonal Design (ORTHOPLAN)ORTHOPLAN
- Categorical Regression (CATREG)CATREG
- Linear RegressionREGRESSION
- K-Means ClusterQUICK CLUSTER
- GLM UnivariateUNIANOVA
- Free SPSS alternative: full coverage list
Run Conjoint Analysis on your own data
Open a .sav, CSV or Excel file. Everything runs in your browser; the file never leaves your device.
SPSS is a trademark of IBM Corp. Crosstabs is not affiliated with or endorsed by IBM.