Online Program

Using the Proportional Odds Model to Evaluate State Legislators’ Use of Information Sources in the U.S. Policy Process (303468)

Ordinal data often results from survey items that use Likert type scale where response options are ordered to indicate an intensity, or degree of sentiment. The responses reflect an ordinal scale which presents applied statisticians with unique challenges. Although these data are often treated as interval level data (cf. Muthen & Kaplan, 1992) that may not be the best practice (Agresti, 2010) because ordinal scales lack many of the distributional properties of interval level data. Amplifying analysts’ difficulties is the issue of clustering. When subjects respond to multiple items, there is a clustering of correlated responses within subjects. Hence, analysts are faced with the decision of whether to fit a simple marginal model or a more complicated random effects model. We explore these analytic options using original survey data, census demographics on legislators and districts, and characteristics of 24 state legislatures across the U.S. The survey used Likert scales to assess how frequently legislators obtained information from various sources to guide decisions in their lawmaking duties. We examine the effects of these analytic decisions on the interpretation of model results.