Public opinion researchers frequently work with semantically balanced ordinal scales, such as five-point Likert scales. These items are used to measure candidate evaluations, public spending preferences, positions on social issues, and candidate and party placement, for example. Because the items are semantically balanced about a midpoint or implied midpoint, researchers may be interested in both the tendency to respond above versus below a midpoint (the response direction) and the tendency to choose extremal categories above or below a midpoint (the response intensity). While political analysts are commonly interested in both response intensity and direction, traditional methods used for modeling semantically balanced ordinal scales do not address these issues well. In this paper, we discuss adjacent category logit models that allow researchers to simultaneously model direction and intensity in ordinal scales with midpoints or implied midpoints. We apply the models to data on approval ratings of House incumbents, finding that the tendency to evaluate incumbents favorably increases through the 1980s (peaking in the late eighties) and declines thereafter until 1 996 (the last year for which we have data). We also find that respondents, both favorable and unfavorable, express stronger opinions during most presidential.election cycles than in midterm election years. odeling trends, cycles, or temporal change in public opinion is commonplace in political analysis. Researchers have devoted a great deal of attention to understanding how the public's ideological predilections, policy preferences, or approval functions vary over time. At the core of this literature is the notion that public opinion shifts or changes over time are politically meaningful, translatable into public policy outcomes, and bear on issues of democratic governance. To address the issues above, researchers often use ordinal scales semantically balanced about a midpoint or implied midpoint (for example, a scale with categories strongly = 1, = 2, neither approve nor = 3, disapprove = 4, strongly = 5 or a scale with categories strongly = 1, 2, disapprove 3, strongly = 4). In analyzing such items, researchers sometimes assign equal interval scores to the scale categories and inappropriately use ordinary least squares regression to model the scale as a function of covariates. Analysts also sometimes combine scale categories above and below the midpoint or implied midpoint. This approach is problematic because in combining, for example, strong approvers with approvers, and strong disapprovers with disapprovers, the intensity of the response is lost. In contrast, researchers occasionally fold semantically balanced ordinal scales about the midpoint or implied midpoint, combining strong approvers with strong disapprovers and approvers with disapprovers. In this approach, respondents on different sides of an issue (direction) are grouped together, and it is therefore not possible to assess whether or not the approvers or disapprovers are more (less, equally) intense in their opinions. Sometimes, analysts will model the item using the proportional odds
Read full abstract