Strong Preferences or Simplifying Heuristics? Using Internal Validity Tests and Latent Class Analysis to Better...

Area of Science:

• Behavioral economics
• Econometrics
• Marketing science

Background:

• Internal-validity tests (IVTs) are crucial in discrete choice experiments (DCEs) for evaluating decision heuristics, choice logic, response consistency, and tradeoffs.
• Current standards for classifying unacceptable data quality based on IVT failures and incorporating these failures into choice models are lacking.

Purpose of the Study:

• To assess internal-validity test (IVT) failures in discrete choice experiments (DCEs).
• To utilize latent class analysis (LCA) for identifying choice patterns indicative of statistically informative DCE data.
• To evaluate preference heterogeneity while controlling for attribute dominance.

Main Methods:

• A DCE was conducted with 4 attributes (3 ordered) and 12 experimental choice tasks, alongside 2 constructed IVT choice tasks.
• Respondents with IVT failures were interviewed to understand their choices.
• Preference heterogeneity was evaluated using a 4-class LCA with attribute-specific alternative-specific constants, compared against a 1-class model without these constants.

Main Results:

• Out of 201 respondents, 34 exhibited IVT failures, with 38%-42% providing reasons beyond nonattendance or simplifying heuristics.
• A 4-class LCA model revealed significant differences in coefficients for two ordered attributes compared to a 1-class model, indicating potential bias from simplifying heuristics.
• The probability of attribute-specific dominance varied with the number of choice tasks exhibiting dominance, influencing class membership.

Conclusions:

• "Failures" in internal-validity tests (IVTs) should be viewed as unexpected responses requiring further investigation.
• While understanding questions can offer insights, they increase respondent burden and may not fully explain simplifying heuristics.
• Latent class models that control for attribute dominance provide a data-driven method for assessing simplifying heuristics and attribute dominance thresholds, surpassing subjective rules of thumb.