AMCE and ATE Explained for Conjoint Surveys
A plain-language explanation of AMCEs, ATEs, reference levels, uncertainty, and the data behind conjoint estimates.
In Brief
An AMCE asks how changing one randomized attribute level changes the average chance that a profile is chosen, relative to a reference level. An ATE asks how assignment to a treatment changes an outcome on average. Both are average effects, but they answer different questions and should be labeled clearly.
Reading the Estimate
A conjoint coefficient is usually read as a change in choice probability. An AMCE of 0.08 means that profiles with that level were about eight percentage points more likely to be selected than profiles at the reference level, averaging over the other randomized attributes in the design.
Why the Data Are Long
Every profile shown contributes information, including the profiles that were not selected. A proper long-format file therefore contains one row per displayed profile and keeps respondent IDs so the analysis can account for multiple choices made by the same person.
Communicating Uncertainty
Report the estimate with its confidence interval and name the reference level in the figure or table. These are average effects across respondents and tasks; they should not be described as though every individual reacted in the same way.
FAQ
Is an AMCE a percentage-point effect?
With a binary choice outcome estimated by a linear probability model, an AMCE can usually be read as a percentage-point change in the probability of selection. Always confirm the model and outcome coding before using that interpretation.
Why cluster standard errors by respondent?
One person typically completes several tasks, so their profile rows are related rather than independent. Clustering by respondent allows the uncertainty estimate to reflect that repeated-measures structure.