Published online by Cambridge University Press: 27 January 2009
The availability of rich survey data, and concerns over the ecological fallacy, have led voting researchers to focus on the explanation of individual voting decisions at the expense of accounting for patterns of aggregate election outcomes. This has skewed our understanding of the relative importance of various factors in the electoral process. A framework for analysis of elections at multiple levels is developed and applied using data from twenty-three exit polls from the US Senate elections. Comparable parameters for a simple voting model are estimated for individual voting and for election outcomes. Election-level factors, especially candidates' issue strategies and incumbency, are substantially more important in accounting for election outcomes than in explaining individual voting decisions. Finally, working with election outcomes permits an estimate of a path model of Senate election outcomes that shows key relationships that are not accessible from individual level data.
2 See, for example, Crewe, Ivor and Payne, Clive, ‘Another Game with Nature: An Ecological Regression Model of the British Two-party Vote Ratio in 1970’, British Journal of Political Science, 6 (1976), 43–81CrossRefGoogle Scholar; Goodman, Leo A., ‘Some Alternatives to Ecological Correlation’, American Journal of Sociology, 64 (1959), 610–25CrossRefGoogle Scholar; Hanushek, Eric A., Jackson, John E. and Kain, John F., ‘Model Specification, Use of Aggregate Data, and the Ecological Correlation Fallacy’, Political Methodology, 1 (1974), 89–107Google Scholar; Langbein, Laura I. and Lichtman, Allan J., Ecological Inference (Beverly Hills, Calif.: Sage Publications, 1976)Google Scholar; Shively, W. Phillips, ‘Ecological Inference: The Use of Aggregate Data to Study Individuals’, American Political Science Review, 63 (1969), 1183–96Google Scholar among many others.
3 Schmidt, Otto and van Dijk, Tibert, ‘Ecological Inference: An Empirical Analysis of Dutch Electoral Data’Google Scholar, and Tomaselli, Verna, ‘The Comparison of Conventionalist Goodman's Fluxes and True Fluxes of Voting and Non-voting’Google Scholar. Both papers were presented at the Joint Sessions of the European Consortium for Political Research, Rimini, April 1988.
5 Campbell, Angus E., Converse, Phillip E., Miller, Warren E. and Stokes, Donald E., The American Voter (New York: Wiley, 1960).Google Scholar
6 Stokes, Donald E., ‘Some Dynamic Elements of Contests for the Presidency’, American Political Science Review, 60 (1966), 19–28CrossRefGoogle Scholar; Miller, Warren E. and Shanks, Merrill, ‘Policy Directions and Presidential Leadership: Alternative Interpretations of the 1980 Presidential Election’, British Journal of Political Science, 12 (1982), 299–356.CrossRefGoogle Scholar
7 Campbell, Angus E., Converse, Phillip E., Miller, Warren E. and Stokes, Donald E., Elections and the Political Order (New York: Wiley, 1966).Google Scholar
8 Kinder, Donald R. and Kiewiet, D. Roderick, ‘Sociotropic Politics: The American Case’, British Journal of Political Science, 9 (1979), 129–61Google Scholar and Kinder, Donald R. and Kiewiel, D. Roderick, ‘Economic Discontent and Political Behavior: The Role of Personal Grievances and Collective Economic Judgments in Congressional Voting’, American Journal of Political Science, 23 (1979), 495–527.CrossRefGoogle Scholar
10 Ranney, Austin, ‘The Utility and Limitations of Aggregate Data in the Study of Electoral Behavior’, in Ranney, A., ed., Essays on the Behavioral Study of Politics (Champagne: University of Illinois Press, 1962), 91–102, at p. 99.Google Scholar
11 Key, V. O. Jr., ‘The Politically Relevant in Surveys’, Public Opinion Quarterly, 24 (1960), p. 55.Google Scholar
13 Elsewhere I have argued that we should use slopes to compare between populations and betas to compare variables within a single population. This perspective fails to consider that changes in beta weights across populations, or here across levels of analysis, can be of theoretical interest. See Wright, Gerald C., ‘Linear Models for Evaluating Conditional Relationships’, American Journal of Political Science, 61 (1976), 130–6Google Scholar; see also Blalock, Hubert M., ‘Causal Inference, Closed Populations and Measures of Association’, American Political Science Review, 61 (1967), 130–6.CrossRefGoogle Scholar
14 Hanushek, et al. , ‘Model Specification, Use of Aggregate Data, and the Ecological Correlation Fallacy’, pp. 90–5.Google Scholar
15 Cartwright, Desmond, ‘Ecological Variables’ in Borgatta, Edgar F., ed., Sociological Methodology 1969 (San Francisco: Jossey-Bass, 1969), pp. 205–12.Google Scholar
16 Cain, Bruce, Ferejohn, John and Fiorina, Morris, The Personal Vote (Cambridge, Mass.: Harvard University Press, 1987), pp. 130–4CrossRefGoogle Scholar, use analysis of variance terminology to make this and several of the points made here in their discussion of problems of relating congressmen's service and advertising behaviour to citizen evaluations of incumbents. By implication, the conclusions offered below can be extended to many instances in which the analyst wishes to assess elite behaviour or system characteristics (such as economic performance) using mass survey data. Two promising instances are the study of representation and voter perceptions of party positions. These are explicated very nicely in chapters that came to the author's attention after this article was written. See Converse, Philip E., ‘Popular Representation and the Distribution of Information’Google Scholar, and Stimson, James A., ‘A Macro Theory of Information Flow’, both in Ferejohn, John A. and Kuklinski, James, eds, Information and Democratic Processes (Urbana: University of Illinois Press, forthcoming).Google Scholar
17 The exact wording and scoring (in parentheses) used throughout are as follows: Party Identification: ‘Do you usually think of yourself as a: Democrat (– 1), Independent (0), Republican (1)?’ Voter Ideology: ‘On most political matters do you consider yourself Liberal (– 1), Moderate (0), Conservative (1)?’ Presidential Popularity: ‘Do you approve or disapprove of the way Ronald Reagan is handling his job as President?’ Approve (1), Disapprove(– 1) (undecided was volunteered by some respondents, 0); Family Finances: ‘Compared to a year ago, is your family's financial situation better today (1), worse today (– 1), about the same (0)?’
18 Candidates and incumbents were asked if they favour or oppose constitutional amendments to: (1) allow individual states to prohibit abortions; (2) permit organized prayer in the public schools; and (3) require a balanced budget. In addition, they were asked about positions on (4) the Equal Rights Amendment; (5) a mutual nuclear freeze with the Soviet Union; (6) domestic content legislation for foreign cars sold in the United States; (7) cancelling the July 1983 tax cut; (8) cutting back increases in military spending; (9) additional reductions in domestic social programs; and (10) regulation of air pollution.
19 The use of the midpoint as a summary of candidates' issue positions is adapted from Robert Erikson's ideas on measuring party elite preferences in comparative state politics. See Erikson, Robert S., Wright, Gerald C. and McIver, John P., ‘Political Parties, Public Opinion, and State Policy’, American Political Science Review (forthcoming).Google Scholar
20 One generally needs to be concerned with heteroscadascity when using OLS regression with a dichotomous dependent variable. Regression creates no distortions in these data, however. All the individual-level equations were run with probit and the same conclusions resulted. Regression is used at both levels here to achieve comparability of the estimated coefficients.
21 Barone, Michael and Ujifusa, Grant, The Almanac of American Politics 1984 (Washington, DC: The National Journal, 1984), pp. 781–8, at p. 785.Google Scholar
22 Wright, Gerald C. and Berkman, Michael, ‘Candidates and Policy in United States Senate Elections’, American Political Science Review, 80 (1986), 567–88CrossRefGoogle Scholar; Wright, Gerald C. and Berkman, Michael. ‘Do US Senators Moderate Strategically?’, American Political Science Review, 82 (1988), 242–4.Google Scholar
23 In this analysis we use only Democratic incumbency because Republican incumbency had no electoral payoff in 1982. Changing the variable of incumbency to reflect values for Republican in cumbency (i.e., 1,0, – 1 for Republican incumbents, open seats and Democratic incumbents) does not affect the results reported.