Conjoint

Conjoint helps you learn which parts of a product or offer influence people's choices. Instead of asking someone to rate each feature in isolation, you show complete options with different combinations of features and ask which one they would choose.

Use it to compare product packages, pricing, service levels, or other offers that people evaluate as a whole. You can inspect the choices people actually made, run a modeled analysis to estimate the effect of each feature, and compare new combinations in the simulator.

A Conjoint question takes one place in your survey, but includes several rounds. In each round, a person sees two to four complete options and chooses one. Each option combines one level from every attribute in the question.

For example, a delivery service study might use Price ($10, $20, or $30), Delivery time (same day or two days), and Support (standard or priority). One round could ask someone to choose between a $10 service with two-day delivery and standard support, and a $20 service with same-day delivery and priority support. Later rounds show different combinations.

You can also offer a “None of these” choice when people could realistically decline every option. If the question is Required, the person must choose a product or None in every round. If it is optional, they can leave rounds unanswered. They can go back and change a choice before submitting.

Conjoint measures preferences relative to the levels and products included in your study. A result for $20, for example, only makes sense alongside the other prices and features you tested.

You need 2 to 10 attributes, with 2 to 7 levels for each and no more than 30 levels in total. We recommend starting with no more than 7 attributes. Keep level labels short, distinct, and easy to understand. Attribute names must be unique, and level names must be unique within each attribute.

Iterate generates the product combinations when you save. It varies the levels shown and where products appear, so respondents compare different combinations instead of seeing the same choices repeatedly.

After saving, the attributes, levels, number of options per round, number of rounds, and whether None is offered are locked. You can still edit the question text, instructions, option label, None label, and Required setting. To change the study design, duplicate the question and edit the duplicate before saving it. Responses to the original stay with the original question.

Open View responses to see an observed summary for every level. It updates as responses arrive and shows how often a level appeared in an offered product, how often a product containing it was chosen, and the percentage chosen when shown.

For example, if a $20 level appeared in 40 offered products and products containing it were chosen 12 times, its Chosen when shown rate is 30%. A level can appear in more than one product in a round. If someone chooses None, the offered product levels still count as shown. Unanswered rounds do not count.

These are observed rates, not the effect of changing only that level. A $20 product might also have faster delivery, so its choices reflect the whole product. The modeled Scores tab helps separate the relative effects of the levels.

You can open an individual response to see the products that person chose in each round.

Modeled preferences is part of Research Assistant. If you do not see it, contact our support team to learn more about adding it to your plan.

The analysis learns from each person's choices between complete products. It estimates how much each level changes preference while allowing different people to value the same level differently. Skipped rounds add no choice information. The model checks that its estimates have settled before results become available. If those checks fail, the analysis can fail rather than show unreliable estimates. Contact us if you need help with a failed run.

A run uses a stable sample of up to 2,000 people who answered at least one round. If more people answered, the run uses 2,000 of them. Dashboard filters do not change the modeled results. The modeled results are a snapshot, while observed counts continue updating as new responses arrive.

To update the model after more responses arrive, use Run again. Another run becomes available 24 hours after the previous one completes and after at least 10 new responses or 10% of the previous response count, whichever is larger. The button's tooltip shows what you are waiting for.

The Scores tab shows Attribute importance and Level values. Both come from the completed model, not directly from the observed counts.

Importance measures how much the levels of an attribute can change preference within the range you tested. For each person, the model compares that attribute's highest and lowest level values, then turns the attribute ranges into percentages. The chart averages those percentages across the modeled respondents, and mean importance adds to 100%.

Importance depends on the levels you include. If you test only a narrow price range, Price may appear less important than it would in a study with a much wider range. A high importance percentage does not mean that percentage of people named the attribute as their top priority.

Level values show how each level affects preference relative to the average level of the same attribute. Positive values increase preference and negative values decrease it. The levels of each attribute are centered around zero, and a product's values add together to describe that product.

For example, if same-day delivery has a positive value and two-day delivery has a negative value, the model estimates that people tend to prefer same-day delivery when the other product features are held in the comparison. A value of +2 does not mean “twice as preferred” as +1. Focus on differences between levels of the same attribute.

If the survey offered None, the Scores tab also shows a None choice value. It compares declining the products with a product made from the average level of each attribute.

The whiskers show uncertainty in each modeled estimate. Given the responses and the model's assumptions, the displayed range contains 95% of the probability for that estimate. Wider ranges mean less precision, so treat small differences cautiously. These ranges describe uncertainty in the overall estimate, not how varied individual preferences are.

The Counts tab gives a fuller table of the observed choices, including Times shown, Chosen, and Chosen when shown for each level. None has its own row when the question offered it. Counts may include responses received after the model ran, so they can differ from the snapshot used for Scores and the Simulator.

Open the Simulator tab after a completed analysis. It starts with two example products. Choose one level for each attribute in each product, then add more products if you like, up to five. Results update as you change the products. None is included as another choice when the question offered it.

The simulator offers two ways to compare the products:

For example, Product A might have 45% preference share against Product B's 35% and None's 20%. That describes this comparison among these modeled respondents. It is not a sales forecast, a market-share estimate, or a promise that 45% of future customers will buy Product A. Adding or changing a product can change every share.

Your survey CSV has one column for each Conjoint round. The value names the selected option and lists its attribute levels, or shows your None label when None was selected. Unanswered rounds and skipped questions are blank. The export keeps one row per survey submission and contains people's recorded choices, not modeled level values, importance, or simulator results.

If you have questions about setting up Conjoint or interpreting the results, reach out to our support team.