When conjoint analysis fails, psychological mismatch is often the first explanation offered.

Commentary commonly argues that failures occur because consumers lack stable or precise preferences, construct preferences during the task, or respond hypothetically rather than as they would in real market contexts shaped by brand, marketing, and availability. The conclusion is that conjoint rests on false assumptions about preference formation and therefore produces results that cannot generalize to the market.

These explanations sound plausible.
They are also premature unless statistical power has been established.

The Missing Question: Could the Study Recover Stable Structure?

Each of these psychological explanations implicitly assumes that the experiment was sufficiently powered to detect stable structure had it existed.

In practice, indicators of power, such as task count relative to parameter count, are often left unexamined altogether.

With too few choice tasks, weak experimental leverage, or poorly identified price effects, the data simply do not contain enough information to distinguish structure from noise. Under those conditions:

• “unstable preferences” are indistinguishable from imprecise estimation,
• “constructed preferences” are indistinguishable from random response variation,
• “artificial deliberation” is indistinguishable from noise inflating minor attributes,
• and “hypothetical bias” is indistinguishable from weakly identified willingness to pay.

In all cases, the empirical signature is the same: variance without recoverable structure.

Why Contextual Explanations Don’t Resolve the Ambiguity

Appeals to real-world complexity, brand effects, marketing influence, or supply-side constraints, do not resolve this problem.

Those factors may matter. But without sufficient power, there is no way to determine whether the model failed because it omitted context or because the study never had enough information to support individual-level inference in the first place. This sequencing error explains why even best-practice conjoint studies can fail without obvious methodological mistakes.

Psychology and context become catch-all explanations precisely when the data cannot adjudicate among alternatives.

Why Better Estimation Doesn’t Fix This

Advanced estimation techniques can stabilize results and improve apparent coherence. But they do not add information that was never collected.

Hierarchical Bayes is often treated as a remedy in these situations. It can stabilize estimates by pooling information across respondents, but it does not create signal that was never collected. HB can improve coherence; it cannot establish whether instability reflects psychology or insufficient power.

Advanced estimation alone does not diagnose.

The Correct Order of Explanation

Psychological mismatch may be a real contributor to failure, but it cannot be the first explanation.

Unless a conjoint study demonstrates sufficient power to recover stable structure, psychological explanations are not diagnoses. They are convenient explanations applied after the fact—ones that do not require interrogation of the study’s statistical power.