Evaluating the public perception of a feebate policy in California through the estimation and cross-validation of an ordinal regression model

• Survey in California found roughly ¾ would have supported a feebates policy.
• Ordinal regression used to evaluate the variables that influence policy opinion.
• The cross-validated prediction accuracy of ordinal regression model is nearly 90%.

Understanding the key factors influencing policy perception can be critical for informing the design of public policies. Feebates is a unique public policy that is meant to influence vehicle purchases. It presents buyers with a rebate for purchasing low-emission vehicles and a fee for purchasing high-emission vehicles. Because feebates directly impacts the consumer, understanding the dynamics of public perception, support, and opposition is important. This study explores the public perception of a potential feebate policy within California, and evaluates the robustness of an ordinal regression model to predict policy sentiment. The authors conducted a series of 12 focus groups throughout the State, which were followed by a computer-assisted telephone interview (CATI) survey of 3072 California residents in Fall 2009. The survey results were used to gain insights into consumer response to the policy, while focus groups gauged participant understanding of the feebate concept and overall response in preparation for the statewide survey. The survey data was weighted to match key demographics of the population and probed respondents on sentiments towards climate change, foreign oil dependence, policy fairness as well as overall perceptions of the policy. The results suggested that roughly three quarters (~76%) of the population would have supported a feebate policy, while one-in-five (~22%) would have opposed it. To evaluate how key factors simultaneously influence policy support/opposition, the authors estimated an ordinal regression model on policy support, which could correctly re-predict 89.4% of the sample׳s policy support or opposition. To assess the model׳s robustness, it was validated through re-estimation with 10,000 randomly drawn subsamples. Models estimated using these subsamples were then applied to predict policy perception for the remaining hold-out sample. The model performed very well, as hold-out sample opinions were predicted at an average accuracy of 89.1%, with little variance in performance. The authors conclude with a discussion of the implications of these results on public support for feebates and comment on the use of ordinal regression to predict policy opinion.