How does a choice experiment (CE) model derived under standard preference axioms perform for respondents with incomplete preferences? Using simulated data, we show how such miss-specification results in unnecessary noise and bias in welfare estimates, and can be avoided.