Dimensionality Redux

Abstract

Professor Jacoby raises a number of important points in his comment on my study of public opinion on government spending. This rejoinder underscores why I believe the evidence warrants the conclusion that attitudes toward welfare spending are partially distinct from attitudes toward other social spending. Before proceeding, I would like to highlight some grounds on which we agree. Although our models differ in terms of the number of dimensions needed to represent spending attitudes, we concur that American citizens hold real attitudes on what is arguably one of the most important issues in the domestic policy arena of the United States. In light of the enduring attitudes-nonattitudes controversy, the question of whether spending attitudes are better described in unidimensional or multidimensional terms should not obscure the fact that they are coherent and meaningful. We differ, however, in the particulars. To appreciate why this difference matters, it is helpful to situate our debate in a broader context Converse (1964) argued that citizens' issue preferences are weakly held and largely unstructured; however, he identified an important exception. Based on responses to the open-ended National Election Study (NES) like-dislike questions, Converse found that citizens evaluate domestic politics in spend-save terms. Given the impressionistic nature of the evidence and the lack of suitable measures to explore dimensionality, researchers suspended judgment on the matter. In a subsequent pair of articles, Professor Jacoby (2000, 1994) found that responses to a series of social spending items fit an underlying cumulative scale. Items referencing programs in other policy domains (e.g., crime, the environment, etc.) did not fit the scale. Around the time these studies appeared, Gilens (1999, 1996) demonstrated that attitudes toward welfare and food stamps depend on stereotypical beliefs about blacks. Such results are likely an outgrowth, in part, of public discourse as both the news media and public officials often imply that people on welfare are black. However, these associations do not appear as frequently in public debate when the topic turns to more deserving subsets of the needy (Gilens 1999). Building on this perspective and on scattered bits of evidence on racialized public opinion, I posited that whites view welfare programs as distinct from other social programs. My analysis suggests that spending attitudes are divided along these lines and that racial stereotypes only impact welfare opinion. In contrast, Professor Jacoby (2008 [this issue], 1994) finds evidence of a unidimensional structure in which all spending attitudes are grounded in racial predispositions. Can we say that one of these empirical representations of spending opinion is closer to the truth? As Professor Jacoby notes, dimensional analysis is a subjective business. Hence, in a situation where we have competing models of the data, well-accepted scientific criteria should be invoked to assess the rival claims. Professor Jacoby stresses parsimony above all else, but we can, of course, use other scientific criteria to assess propositions, such as accuracy, theoretical utility, innovation, and so on (Gerring 2001).1 Thus, the appropriate question to ask is not whether the unidimensional representation is preferable to the two-dimensional representation on the grounds of parsimony but, rather, how they compare across a broader set of germane scientific criteria. I will focus the comparison on a few criteria. To begin with accuracy, I believe the two-dimensional model provides a better representation of spending opinion than its one-dimensional rival. I concur that it is literally impossible to say that any particular dimensionality is the single 'correct' specification (Jacoby 2008,158). But insofar as we can estimate and directly compare theoretically informed models of attitude structure on statistical grounds, it seems fair to say that some approximations of it are more plausible than others. …