OBJECTIVES: To compare measures of internal validity across similar stated-choice surveys, evaluate consequences of inconsistencies, and determine whether there is a basis for establishing a minimum standard for validity. METHODS: Axioms of utility theory require that valid preferences be stable, monotonic, and transitive. Counts of validity failures were obtained for each survey respondent in 8 stated-choice or discrete-choice experiment surveys. Each survey was administered to adults who were diagnosed with a specific disease. Each survey required respondents to choose between 8 to 12 pairs of treatment alternatives with varying treatment features. We also identified respondents with non-compensatory or lexicographic preferences, where subjects choose alternatives based on a single attribute. Using negative-binomial regression we estimated the effect of personal characteristics on internal-validity test results. RESULTS: Choices from 3929 respondents were evaluated for internal validity. Across all surveys, approximately 20% failed at least one stability test, 10% failed at least one monotonicity test and 15% exhibited lexicographic preferences. Non-white respondents (p<0.05) and respondents with less education (p<0.05) were more likely to fail one of these tests. The effect of income is concave. As income increases the likelihood of failure of one or more tests decreases; however, above an annual income of $100,000, the likelihood of failure increases. Gender, age and employment had no significant effect on internal validity. With the exception of respondents with lexicographic preferences, removing respondents who fail internal validity tests generally did not materially change point estimates in preference models. However, removing inconsistent respondents improved estimate precision by 1-5%. CONCLUSION: Our results suggest that a failure of non-lexicographic internal validity tests at a rate less than 25% does not bias preference estimates derived from stated-choice surveys. Identifying which respondent groups find stated-choice surveys challenging can help in developing and pretesting surveys that minimize cognitive errors in those groups.