¹ For a survey and introduction to these theories, see Riker, William and Ordeshook, Peter, An Introduction to Positive Political Theory (Englewood Cliffs, N.J.: Prentice Hall, 1974)Google Scholar, or Frohlich, Norman and Oppenheimer, Joe A., Modern Political Economy (Englewood Cliffs, N.J.: Prentice Hall, 1978).Google Scholar

² Some efforts have been made to test these theories. Many of these tests are referenced in the footnotes of Frohlich and Oppenheimer, Modern Political Economy, and contained in the bibliography of Riker and Ordeshook, An Introduction.

³ A review of most of these is contained in Frohlich and Oppenheimer, Modern Political Economy, Ch. 5.

⁴ A comprehensive summary of these hypotheses is contained in Riker and Ordeshook, An Introduction. Also see Ordeshook, Peter, “The Spatial Theory of Elections: A Review and a Critique,” in Party Identification and Beyond, Rudge, Ian, et al., ed. (London: Wiley, 1976), pp. 285–314Google Scholar.

⁵ For relevant examples consult RePass, David E., “Issue Salience and Party Choice,” American Political Science Review, 65 (June 1971), 389–400CrossRefGoogle Scholar, and Jackson, Joh E., “The Importance of Issues and Issue Importance in Presidential Elections: A Test of a ‘Rational’ Model,” paper prepared for the annual meeting of the Public Choice Society, Pittsburgh, Pennsylvania, May 1972Google Scholar.

⁶ Shaffer, William R., Computer Simulations of Voting Behavior (London: Oxford University Press, 1972)Google Scholar.

⁷ Downs, p. 7.

⁸ Downs, p. 36.

⁹ Ibid., Ch. 9 contains a brief sketch of the kinds of changes that would be required to extend the analysis to multisided races.

¹⁰ Ibid., p. 39. In this formulation, t + 1 is the period of incumbency of the government to be elected, E(U) symbolizes the expected value of U, and U^A and U^B are the utility streams flowing from the alternative governments.

¹¹ Ibid., pp. 40–45.

¹² Downs, p. 244.

¹³ Ibid., p. 85.

¹⁴ Ibid., p. 100.

¹⁵ Downs, p. 270.

¹⁶ Ibid.

¹⁷ There is one exception to this conclusion: when individuals have a large enough long-run participation value to make their vote value greater than their estimated cost of voting, though they perceive no significant differences between the parties. Downs assumes that such individuals will actually vote, and he includes a detailed discussion of tie-breaking procedures for such voters. Given our operationahzations, the number of individuals perceiving no differences between the parties is extremely small in the sample we use. In addition, it is not possible successfully to operationalize the tie-breaking procedures suggested by Downs on the basis of the SRC data. Therefore, we have chosen to ignore this issue in the work reported in this essay.

¹⁸ Shaffer, , Computer Simulations, pp. 58, 77, and 88Google Scholar.

¹⁹ Ibid., p. 84.

²⁰ Ibid., pp. 88–94.

²¹ Ibid., p. 70.

²² Downs, in contrast, asserts, “There can be no simple identification of ‘acting for one's own greatest benefit’ with selfishness in the narrow sense because self-denying charity is often a great source of benefit to oneself. Thus our model leaves room for altruism in spite of its basic reliance upon the self-interest axiom” (p. 37).

²³ Hempel, Carl G., “The Theoretician's Dilemma,” in Aspects of Scientific Explanation (New York: Free Press, 1965), esp. pp. 187–89Google Scholar.

²⁴ An unfortunate failing of the SRC survey instrument is that respondents are not asked which candidate, rather than which party, is preferable on each issue. Recent research indicates that the agreement between preferred candidate and preferred party is not so great as to render the issue trivial. See the discussion of party and candidate images in Popkin, Samuel L., Gorman, John W., Phillips, Charles and Smith, Jeffrey A., “comment: What Have You Done for Me Lately?: Toward an Investment Theory of Voting,” American Political Science Review, 70 (September 1976), 779–805CrossRefGoogle Scholar.

²⁵ Alternative operationalizations of this effect are possible. It would be feasible to increase the weight of issues to reflect increased information or to combine this approach with an information discount on party identification. Although we focus exclusively on the discount of party identification in our basic model, we have undertaken some tests employing alternative operationalizations for this factor.

²⁶ Actually, to eliminate an indeterminate prediction with respect to that one individual, the cost of voting is set just above that individual's vote value and below that of the next higher individual in the rank ordering.

²⁷ The sample size here is larger than that used in predictions of party preference; in our analysis of party preference, we dropped individuals with D = O for whom we could not make predictions. Other fluctuations in sample size stem from similar considerations.

²⁸ A technique called Engel Curve analysis has been developed in econometrics to deal with decision problems analogous to the problem of whether or not to vote. The economic problem for which this technique was developed is the prediction of household purchases of consumer durables. The household's decision is assumed to be a function of household income. Each household is assumed to have some income threshhold level for the purchase of the durable. If the household's income exceeds the income threshhold level, the household is predicted to purchase the durable. This technique permits the formulation of probabilistic predictions regarding the household's purchase of the item. For a given probabilistic distribution of income level thresholds, the probability that a given household will purchase the item is calculated as the probability that its income is above the income threshold level. Empirically it has been found that the best predictions are obtained when income threshold levels are assumed to be distributed lognormally and independently of income. This assumption yields the prediction that the relationship between the log of income and probability of purchase is linear.

With respect to the question of voter turnout, we can adapt the assumptions of this model by letting cost play the role of income threshold levels and letting vote value play the role of income. For a good discussion of the nature of Engel Curve analysis see Kramer, J. S., Empirical Econometrics (New York: North Holland, 1971), Ch. 3Google Scholar.

²⁹ The high correlation obtained here suggests that it would be desirable to conduct independent tests of the assumption that voting costs are distributed lognormally.

³⁰ See, for example, Kelley, Stanley Jr., and Mirer, Thad W., “The Simple Act of Voting,” American Political Science Review, 68 (June 1974), 572–91CrossRefGoogle Scholar.

³¹ To see this consider the following illustration. There is no a priori justification for assigning a stronger explanatory role to one voter's response that (s)he is a “strong party identifier” than to another's response that (s)he is a “weak party identifier.” In fact, there are significant variations in language usage among ethnic and other subgroups in American society. Members of some subgroups may be prone to overstatements and the use of strong language to describe what are actually rather weak preferences (the stereotype of the brash Texan comes to mind in this connection). Members of other groups may characteristically understate strong preferences (the stereotype of the taciturn Vermonter constitutes an example). Such phenomena may be nonrandomly distributed with respect to the variable labeled party identification in our model. To take an extreme example, if all and only Texans were in reality weak party identifiers and all and only Vermonters were actually strong party identifiers, then weightings on this variable derived from the statements of respondents would be in exactly the wrong order. To the extent that weaker versions of such phenomena are present in our data set the ordinal properties of the weights assigned to the variables in the model are suspect.

³² Ferejohn, John A. and Fiorina, Morris P., “Closeness Counts Only in Horseshoes and Dancing,” American Political Science Review, 69 (September 1975), 920–25CrossRefGoogle Scholar.

³³ The SRC's research on voting began without any clearly specified model of individual voting behavior, though there was a commitment to a research strategy. Specifically, the SRC analysts set out to examine the role of psychological variables in addition to social characteristics. For an excellent review of the evolution of the SRC's model over time, as well as a perceptive aitique of the model, see Natchez, Peter B., “Images of Voting: The Social Psychologists,” Public Policy, 18 (Summer 1970), pp. 553–88Google Scholar.

³⁴ Campbell, Angus, Converse, Philip, Miller, Warren, and Stokes, Donald, The American Voter: An Abridgement (New York: Wiley, 1964), p. 13Google Scholar.

³⁵ The equation estimated is:

p(R) = β₀ + β_1×1 + β_2×2 + β_3×3 + β_4×4 + β_5×5 + β_6×6 + U

where p(R) is the probability of voting Republican, the X_i's are the scores on the six attitudinal dimensions, U is a random error term, and the β_i's are the regression coefficients.

³⁶ Stokes, Donald, “Some Dynamic Elements of Contests for the Presidency,” American Political Science Review, 60 (March 1966), 28CrossRefGoogle Scholar.

³⁷ The relative importance of the six attitude dimensions over a series of elections is examined using this procedure in Stokes and in Natchez.

³⁸ This argument is formulated and supported by data using an independent measure of intensity of party preference in Campbell et al., Ch. 4.

³⁹ We estimated the SRC model for the subgroup of voters alone to compare the results with the total sample including both voters and nonvoters. Although the multiple correlation of the six factors with vote was smaller in the latter case, the difference was slight (approximately .03) and the coefficients of the two models were quite similar. A further modification of the sample resulted from the exclusion of nonregis tered respondents, to facilitate comparison with the sample used in analyzing the Downsian model. The difference in final sample size for the two models is attributable to the exclusion of different cases due to missing data.

In addition, it should be noted that the factor scores and predictions for the six-factor model reported herein are peculiar to the particular coding scheme employed for translating responses to the open-ended questions into scores for the six factors. Coding schemes other than the one employed here may be equally justifiable. We believe, however, that the (probably marginal) differences which would result would not significantly alter the conclusions reached in this section. See, for example, Kagay, Michael R. and Caldeira, Greg A., “‘I Like the Looks of His Face’: Elements of Electoral Choice, 1952–1972,” paper prepared for annual meeting of the American Political Science Association, San Francisco, September 2–5 , 1975Google Scholar. These researchers report 89.5 percent accuracy in party predictions for 1964 (voters only), as compared to 89.1 percent reported in the text.