This is a sequel to an earlier article in the Journal by the same authors (‘Policy Directions and Presidential Leadership: Alternative Interpretations of the 1980 Presidential Election’, British Journal of Political Science, 12(1982), 357–74), and is based on a paper presented at the 1985 Annual Meeting of the American Political Science Association. Some sections of the 1985 paper have been expanded or clarified, but the present article retains the basic argument and conclusions from the original APSA essay. The authors are currently extending these analyses to include comparisons of the Reagan elections with similar results from 1988 and earlier decades, as well as comparisons between different groups or sectors of the American electorate. An analysis of the 1988 presidential election result will appear in a forthcoming issue of the Journal.

The survey data analysed in this article were collected by the National Election Studies, based on a continuing grant from the National Science Foundation to the Center for Political Studies at the University of Michigan. Neither the Center nor the Foundation is responsible for any of our results or conclusions. In preparing this essay, we have benefited from support provided by the Departments of Political Science at Arizona State University and the Berkeley campus of the University of California, and the Berkeley Survey Research Center. Research and clerical assistance have been provided by Chong-Min Park, Cathi Walton, Ruth Wren, Maureen Jurkowski, Michelle Skinner and David Malin. We are also grateful to several colleagues who took the time to comment on earlier drafts of this paper, including Herb McClosky, Chris Achen, Paul Sniderman, Jack Citrin, Richard Brody and William Flanagan. These individuals, however, are absolved from responsibility for any remaining errors or confusion.

¹ Berelson, Bernard R., Lazarsfeld, Paul F. and McPhee, William N., Voting (Chicago: University of Chicago Press, 1954), pp. 182–99.Google Scholar

² Stokes, Donald E., ‘Spatial Models of Party Competition’, American Political Science Review, 57 (1963), 368–77.CrossRefGoogle Scholar

³ See Miller, and Shanks, , ‘Policy Directions and Presidential Leadership’Google Scholar. For other contributions to this debate, see Miller, Arthur H. and Wattenberg, Martin P., ‘Throwing the Rascals Out: Policy and Performance Evaluations of Presidential Candidates, 1952–1980’, American Political Science Review, 79 (1985), 359–72CrossRefGoogle Scholar, and Markus, Gregory B., ‘Political Attitudes during an Election Year: A Report on the 1980 NES Panel Study’, American Political Science Review, 76 (1982), 538–60.CrossRefGoogle Scholar

⁴ Miller, and Shanks, , ‘Policy Decisions and Presidential Leadership’, p. 301Google Scholar. Several political scientists have distinguished between alternative criteria for ‘importance’, but the following three discussions have been influential in shaping our own work: Stokes, Donald E., Campbell, Angus and Miller, Warren E., ‘Components of Electoral Decision’, American Political Science Review, 52 (1958), 367–87CrossRefGoogle Scholar, presented the first regression analysis based on survey data which decomposed an electoral outcome into ‘contributions’ from different attitudes; Miller, Warren E. and Levitin, Teresa E., Leadership and Change: The New Politics and the American Electorate (Cambridge, Mass.: Winthrop, 1976)Google Scholar, distinguished analytic from political importance in much the same sense we intend in this present discussion; and Achen, Christopher, Interpreting and Using Regression, Sage Series, No. 29: Quantitative Applications in the Social Sciences (Beverly Hills, Calif: Sage Publications: 1982)CrossRefGoogle Scholar, provided a more general discussion in statistical terms.

⁵ Lazarsfeld, Paul F., Bereisen, Bernard R. and Gaudet, Hazel, The People's Choice (New York: Columbia University Press, 1944)Google Scholar and Janowitz, M. and Miller, Warren E., ‘The Index of Political Predisposition in the 1948 Election’, Journal of Politics, 14 (1952), 710–27.CrossRefGoogle Scholar

⁶ For views emphasizing the magnitude of intra-campaign volatility, see Weisberg, Herbert F., ‘The Electoral Kaleidoscope: Political Change in the Polarizing Elections of 1984’, paper presented at the 1985 Annual Meeting of the American Political Science Association at New OrleansGoogle Scholar, and Allsop, Dee and Weisberg, Herbert F., ‘Measuring Change in Party Identification in an Election Campaign’, paper presented at the 1986 Annual Meeting of the Midwest Political Science Association at Chicago.Google Scholar

⁷ The classic presentation of the logic of causal analysis of survey data was made by Rosenberg, Morris, The Logic of Survey Analysis (New York: Basic Books, 1968)Google Scholar. A more recent and succinct precis of this general line of argument explains its relationship to multiple equation models, in Davies, James A., The Logic of Causal Order, No. 55 in the Quantitative Applications in the Social Sciences series (Beverly Hills, Calif.: Sage Publications, 1985).CrossRefGoogle Scholar

⁸ Were our goal to model long-term processes, rather than those associated with a single election, we would have to attempt to ‘unpack’ the history or experiences that give political meaning to the demographic attributes. For an important statement on this topic see Mebane, Walter R. Jr, ‘The Dubious Exogeneity of Demographics in Survey Cross-sections’, paper presented at the 1988 Annual Meeting of the Midwest Political Science Association at Chicago.Google Scholar

⁹ Converse, P. E. and Markus, Gregory, ‘A Dynamic Simultaneous Equation Model of Electoral Choice’, American Political Science Review, 73 (1979), 1055–70Google Scholar; Fiorina, Morris P., Retrospective Voting in American National Elections (New Haven, Conn.: Yale University Press, 1981)Google Scholar; Jackson, John, ‘Issues, Party Choices and Presidential Votes’, American Journal of Political Science, 19 (1975), 161–85CrossRefGoogle Scholar; Green, Donald and Palmquist, Bradley, ‘Of Artifacts and Partisan Instability’, paper presented at the 1988 Annual Meeting of the Midwest Political Science Association at Chicago.Google Scholar

¹⁰ Miller, and Shanks, , ‘Policy Decisions and Presidential Leadership’, pp. 327–31 and 355–6Google Scholar. For an illuminating discussion of research on the topics of persuasion and projection as influences on perceptions of candidate issue preference see Krosnick, Jon A., ‘Psychological Perspective in Political Candidate Perception: A Review of Research on the Projection Hypothesis’, paper presented at the 1988 Annual Meeting of the Midwest Political Science Association at Chicago.Google Scholar

¹¹ The data underlying our analyses of the vote (Table 15 on) are based on calculations involving only validated voters. Our treatment of political opinion in this section, up to Table 14, follows the tradition of describing the electoral context with data pertaining to the entire eligible electorate, not just those who ultimately voted in the election.

¹² Although the evidence requires more intense scrutiny than it has thus far received, NES data from the first quarter pre-election sample (the first two weeks of September) suggest a small Carter lead among those NES respondents who were subsequently validated as having voted in November. That lead disappeared in late September and was replaced with a modest 4–6 point margin for Reagan, a lead which persisted through the last two quarters of the pre-election interviewing. The post-election reports from NES data reflect the actual margin of some 11 points, suggesting that there was indeed some last minute shift of voters to Reagan.

¹³ This measurement of ideological self-placement, subsequently used in our analyses of the 1980 and 1984 elections, was first used by NES in 1972. However, a strongly correlated measure based on ‘thermometer’ assessment of liberals and conservatives extends back to 1964. That measure also reflects great stability over the years with virtually no change from the clear conservative pluralities of 1966 and 1968.

¹⁴ Although virtually all of the major public opinion polls (CBS, Time, Roper and Gallup) showed Democratic pluralities in 1984 that were comparable to the NES data, some of them, notably NBC and Gallup, showed the Democratic edge evaporating by September 1984. Most polls, however, matched the NES figures; CBS, Time and Roper all showed 10 point reductions in the Democratic margin. See Public Opinion (10–11 1985), p. 50.Google Scholar

¹⁵ For a more complete account of the 1980–84 party realignment, see Miller, Warren E., ‘Party Identification and Political Belief Systems: Changes in Partisanship in the United States, 1980–84’, Electoral Studies, 5 (1986), 101–21CrossRefGoogle Scholar. For a somewhat broader account of changes in partisanship and policy preferences between 1980 and 1984, see Miller, Warren E., ‘A New Context for Presidential Politics: The Reagan Legacy’, Political Behavior, 9 (1987), 91–113.CrossRefGoogle Scholar

¹⁶ For a discussion of the relative stability of policy preferences, see Converse, and Markus, , ‘A Dynamic Simultaneous Equation Model of Electoral Choice’.Google Scholar

¹⁷ This perspective has been well argued by colleagues such as Page, Benjamin I. and Jones, Calvin C., ‘Reciprocal Effects of Policy Preferences, Party Loyalties and the Vote’, American Political Science Review, 73 (1979), 1071–90CrossRefGoogle Scholar; Fiorina, , Retrospective Voting in American National ElectionsGoogle Scholar, and Erickson, R., Mackuen, M. B. and Stimson, J. A., ‘On the Importance of Economic Experience and Expectation for Political Evaluation’, paper presented at the 1988 Annual Meeting of the American Political Science Association, Washington, DC.Google Scholar At the same time, the statistical evidence which suggests the influence of short-term forces on partisanship has been challenged by the work of Green, and Palmquist, , ‘Of Artifacts and Partisan Instability’Google Scholar, and the effects minimized by Converse, and Markus, , ‘A Dynamic Simultaneous Equation Model of Electoral Choice’.Google Scholar

¹⁸ These results accord with the observations and injunctions offered by King, Gary in ‘How Not to Lie with Statistics: Avoiding Common Mistakes in Quantitative Political Science’, American Journal of Political Science, 30 (1986), 666–85.CrossRefGoogle ScholarPubMed

¹⁹ This decision to ignore three social characteristics that are frequently crucial in the analysis of mass electoral behaviour reflects two interconnected facts: all three variables are most relevant for contemporary analyses of individual or micro-level changes in partisanship and participation, but none of them differentiated Reagan voters from Mondale voters in 1984 after other stable social characteristics were held constant.

²⁰ See Feldman, Stanley, ‘Report on Values in the 1983 Pilot Study’, Reports to the NES Board of Overseers (10 1983)Google Scholar, and Conover, Pamela Johnston and Feldman, Stanley, ‘Morality Items in the 1985 Pilot Study’, Reports to the NES Board of Overseers (05 1986)Google Scholar; ‘Pilot Study Measures of Civic Obligation’, Reports to the NES Board of Overseers (05 1986)Google Scholar; and ‘Measuring Patriotism and Nationalism’, Reports to the NES Board of Overseers (09 1987)Google Scholar, for discussions of NES-related measurement development concerning a variety of basic values. Also see Stroker, Laura L., ‘Morality and Politics: Conduct and Control’, Pilot Study Report to the NES Board of Overseers (09 1987).Google Scholar

²¹ A recent measurement-related assessment of the ‘endogeneity’ of party identification has produced results congenial to the analytic decisions discussed in this section. The work of Green and Palmquist, which is now being revised, suggests that error in the measurement of party identification is responsible for the apparent influence of short-term forces (like policy distances or performance evaluations) on partisanship. By taking measurement error into account through quasi-Markov simplex models, Green, and Palmquist, conclude: ‘It is not far-fetched to assume party identification to be exogenous with respect to variables such as voting behavior, candidate evaluations, issue proximities, and retrospective performance evaluations’Google Scholar. See Green, and Palmquist, , ‘Of Artifacts and Partisan Instability’, p. 18.Google Scholar

²² Given the seven-point nature of the issue questions involved and the scores for both respondent and ‘current federal policy’, each issue could in principle yield a score from −6 (for a maximum preferred change in the liberal direction) to a + 6 (for a maximum change in the conservative direction). As a result, scores on our summary measure of policy change could vary (in principle) from −42 to +42. In fact, the range of actual scores in the 1984 NES pre-election sample was −39 to +24. The resulting measure is predictably related to both ideological self-designation and party identification, with zero-order correlations of 0.43 and 0.41 with those two predispositions.

²³ The wording for the twelve specific traits which have been combined into our four summary measures is as follows: for leadership (Commands respect, Inspiring, Provides strong leadership); for competence (Hardworking, Intelligent, Knowledgeable); for integrity (Decent, Moral, Sets a good example); and for empathy (Compassionate, Kind, Really cares about people like you). For a discussion of the measures, see Kinder, Donald R., ‘Presidential Traits’, Pilot Study Reports to the NES Board of Overseers (10 1983).Google Scholar

²⁴ The explanatory variables we have classified as ‘stable social characteristics’ and ‘policy-related predispositions’ have been excluded from this table and from all assessments of ‘contributions’ to Reagan's aggregate victory because of the absence of a clear metric that would provide a ‘neutral’ point (a score of zero) for those characteristics. Our disaggregation of the aggregate (Reagan) victory is therefore limited to variables with an explicitly political character.

²⁵ In order to capture all of the impact on the vote of individual variables which will ultimately be grouped together in our summary accounts, total effects have been estimated in a ‘block recursive’ fashion by introducing all of the variables in each group into the analyses at the same time, based on the sequence of stages discussed above. Thus, for example, the effect attributed to defence was estimated in an equation that included all of the respondent's policy preferences as well as all of the variables which precede policy preferences in our model, and the total effect for the other six policy preferences was estimated in the same equation. This procedure deviates from the sequential introduction of individual variables in other blocks, which we used in our earlier analysis of the 1980 elections, but the two procedures yield identical results for the combined ‘importance’ of each group (or block) of variables.

²⁶ These calculations yield the same value as can be produced if we actually construct each of the residuals involved and report the standardized regression coefficient for that variable when all of the variables in its group are added as the ‘last’ (or most proximate) group in the vote predictive equation. The residuals involved in any such group may be correlated with each other, so that these standardized regression coefficients, when squared, do not represent that portion of the variance (in the dependent variable) that can be uniquely attributed to that (specific) residual. This deficiency, however, is eliminated when we combine all of the variables in each explanatory theme into a single group or block.