Upcoming talk: Jonathan Wand: Improving the Measurement of Interpersonally Incomparable Data
November 27th
Jonathan Wand is an Assistant Professor in the Department of Political Science at Stanford University and a Robert Wood Johnson Health Policy Scholar at the University of Michigan. His applied and computational statistical research interests include models of dynamic and strategic individual choice behavior, non-parametric and semi-parametric scaling methods, and shape constrained inference for testing formal models.Substantively, he works on elections, campaign finance, public opinion and health care policy. Jonathan is the recipient of both the Harold Gosnell Award from the APSA and the Robert H. Durr Award from the MPSA for his research on political methodology.
Abstract for the talk:
Attitudes and attributes of individuals are often measured by means of survey questions with ordered response categories, and these measures are commonly employed to make interpersonal comparisons. These types of comparisons, however, rely on the assumption that individuals agree on the meaning of the scale categories. The interpersonal incomparability of responses due to differences in standards is a central challenge in the study of surveys and public opinion. My talk will focus on the use of anchoring objects, such as anchoring vignettes, to improve our ability to draw reliable comparisons across individual’s. Relevant applications range from measuring patient pain to racial discrimination, and from customer satisfaction to the influence of political corruption.
I investigate how to compare survey responses across individuals by asking all individuals to evaluate a common set of anchoring vignettes, or other common survey items. I offer an axiomatic derivation of building scales that illuminates previously unrecognized assumptions implicit in an earlier non-parametric methods using anchoring vignettes, and also leads to a new non-parametric scaling method. I also propose a new semi-parametric method for accomodating measurement error that overcomes earlier uses of strong assumptions concerning within-group homogeneity of the use of scales and the underlying attributes that are compared across groups.