The heterogeneous thresholds ordered response model: Identification and inference

Although surveys routinely ask respondents to evaluate various aspects of their life on an ordered scale, there is concern about interpersonal comparability of these self-assessments. Statistically, the problem is one of identification in ordered response models with heterogeneous thresholds. As a solution to the identification problem, King et al. (2004) proposed using anchoring vignettes, namely brief descriptions of hypothetical people or situations that survey respondents are asked to evaluate on the same scale they use to rate their own situation. While vignettes have been introduced in several social surveys and are increasingly employed in a variety of fields, reliability of this approach hinges crucially on the validity of the assumptions of response consistency and vignette equivalence. This paper proposes a joint test of these key assumptions based on the fact that the underlying statistical model is overidentified if the two assumptions hold. Monte Carlo results show that the proposed test has good size and power properties in finite samples. We apply our test to self-assessment of pain using data from the first wave of the Survey of Health, Ageing and Retirement in Europe. We find that, when using only one of the three available vignettes, or when the test is carried out separately by subgroups of respondents, the overidentifying restrictions are less likely to be rejected