🧭 Two Uses, One Method: Why Purpose Matters

Forced-choice conjoint experiments are a standard tool in political science and sociology. The literature has largely overlooked that these experiments serve two distinct purposes: uncovering respondents' multidimensional preferences, and estimating causal effects of specific attributes on a profile's selection probability in a multidimensional choice setting. This distinction is both analytically and practically important because the appropriate quantity of interest depends on the study's purpose.

📊 What Most Researchers Do—and Where That Falls Short

The vast majority of social scientists using conjoint analyses, including many scholars interested in preferences, adopt the average marginal component effect (AMCE) as the main quantity of interest. That practice is problematic for studying preferences for three reasons:

• AMCE is essentially a causal quantity, which is conceptually at odds with the goal of describing patterns of preferences.
• AMCE generally does not identify preferences; instead it conflates preferences with compositional effects that are unrelated to underlying preference structure.
• Because of this conceptual and empirical mismatch, AMCE is neither conceptually nor practically suited to explore respondents' preferences.

🛠️ A New Target: The Average Component Preference

To address these limitations, a new estimand is proposed: the average component preference. This estimand is designed specifically to capture patterns of preferences rather than causal effects on selection probability. A method for estimating the average component preference is presented, offering researchers a practical way to recover preference patterns from forced-choice conjoint data.

✨ Why It Matters for Research Design and Inference

The distinction between describing preferences and estimating causal attribute effects matters for identification, interpretation, and empirical practice. Researchers interested in respondents' preferences should align their quantity of interest and estimation strategy accordingly and avoid treating AMCE as a general-purpose descriptor of preference patterns.