¹ In using this criterion of explanatory value, we are following C. G. Hempel who observes: “… it may be said that an explanation … is not complete unless it might as well have functioned as a prediction: If the final event can be derived from the initial conditions and universal hypotheses stated in the explanation, then it might as well have been predicted, before it actually happened, on the basis of a knowledge of the initial conditions and the general laws.” ( Hempel, Carl G., Aspects of Scientific Explanation [The Free Press: New York, 1965] p. 234.Google Scholar)

² See Campbell, Angus, Converse, Philip E., Miller, Warren E., and Stokes, Donald E., The American Voter (John Wiley and Sons, Inc.: New York, 1960), pp. 66–68 Google Scholar. See also Campbell, Angus and Stokes, Donald E., “Partisan Attitudes and the Presidential Vote,” pp. 353–371 Google Scholar, in American Voting Behavior, Burdick, Eugene and Brodbeck, Arthur J., eds. (The Free Press: Glencoe, 1959)CrossRefGoogle Scholar; Stokes, Donald E., Campbell, Angus, and Miller, Warren E., “Components of Electoral Decision,” The American Political Science Review 52 (06, 1958), 367–387 CrossRefGoogle Scholar; and Stokes, Donald E., “Some Dynamic Elements of Contests for the Presidency,” The American Political Science Review 60 (03, 1966) 19–28 CrossRefGoogle Scholar.

³ See Stokes, , Campbell, , and Miller, , “Components of Electoral Decision,” pp. 380–385 Google Scholar, and Stokes, Donald E., “Some Dynamic Elements of Contests for the Presidency,” p. 20 Google Scholar.

⁴ The questions about parties were worded as follows: “I'd like to ask you what you think are the good and bad points about the two parties. Is there anything in particular that you (like, don't like) about the (Democratic, Republican) Party? What is that?” The questions about candidates were, “Now I'd like to ask you about the good and bad points of the (two, three) candidates for President. Is there anything in particular about (name of candidate) that might make you want to vote (for him, against him)? What is that?” The variable numbers for these questions in the code books of the Inter-University Consortium for Political Research are as follows: 1952, 0018–0021; 1956, 0015–0022; 1960, 0020–0027; 1964, 0021–0028; 1968, 0028–0037.

⁵ For a larger sampling of responses to these questions, see Campbell, Angus et al., The American Voter, pp. 224–229 Google Scholar.

⁶ Cf. Downs, Anthony, An Economic Theory of Democracy (Harper and Brothers: New York, 1957), pp. 260–274 Google Scholar.

⁷ For a precise description of the procedure employed in predicting how voters would vote on the basis of the Voter's Decision Rule, see Appendix.

It may be useful here to note some of the other decision rules we formulated and tested against data for the 1964 election. There were twenty-five such rules applied to data of the kind to which we applied the Rule, but they may be thought of as members of two families. The first of these comprised rules that differed from each other in regard to the responses to which they were applied but were alike in giving equal weight to all responses treated as considerations. Thus, in the case of each rule, voters were predicted to vote for the candidate toward whom they had the greatest net number of favorable attitudes. Some of the different sets of responses treated as considerations were those to questions about candidates, in which one or both were associated with favorable or unfavorable developments in some named area of public policy; about parties, in which one or both were associated with favorable or unfavorable developments in the general course of public policy, etc.

The thirteen rules of the second family were more complex. Their application involved four distinct steps: (a) Responses to the questions described in f.n. 4 were grouped into mutually exclusive subsets (as few as two, as many as nine), (b) These subsets of responses were assigned an order of importance (e.g., responses involving specific issues of policy might be treated as of greatest importance to the voter, those concerned with the general course of public policy as of next most importance, etc.). (c) The responses (considerations) in each subset were summed, each response weighing equally. Then, (d) in the case of each rule, voters were predicted to vote for the candidate for whom they had the greatest net number of favorable attitudes in the subset of attitudes designated as most important, if there was such; if not, for the candidate for whom they had the greatest net number of favorable attitudes in the subset of attitudes designated as next most important, if there was such, etc. The rules of this general sort differed from each other in the order of importance they assigned to different subsets of considerations, and they were applied to subsets of responses that were defined in several different ways.

At a very early point in our research we formulated still other rules, and these were applied to responses other than those to which we applied the Rule and the rules just described.

⁸ It is appropriate to acknowledge here some lines of thought akin to those that we brought to the present study. Anthony Downs in An Economic Theory of Democracy has made the notion of a decision rule for voters a central element in his theory of voting, as do a number of other writers, including Davis, Otto A., Hinich, Melvin J., and Ordeshook, Peter in “An Expository Development of a Mathematical Model of the Electoral Process,” American Political Science Review 64 (06, 1970), 426–448 CrossRefGoogle Scholar. Donald E. Stokes, Angus Campbell, and Warren E. Miller (in “Components of Electoral Decision”) have used a summated scale, not interpreted as a decision rule, to indicate the preferences of voters with regard to each of the “components of the electoral decision” that they identify, and Walter F. Murphy and Joseph Tannenhaus (again, among others) have used such a scale to measure specific support for the Supreme Court. (See their “Public Opinion and the United States Supreme Court,” Law and Society Review 2 [05, 1968], pp. 357–384.CrossRefGoogle Scholar) Very similar to our own conception of voting is that of Richard Brody and Benjamin I. Page in “Indifference, Alienation, and Rational Decisions: The Effects of Candidate Evaluations on Turnout and the Vote” (unpublished ms.). The same may be said about the conception of voting underlying one of the models tested by Shaffer, William R. in Computer Simulations of Voting Behavior (Oxford University Press: New York, 1972), pp. 65–68 Google Scholar and that developed by Kovenock, David M., Beardsley, Philip L., and Prothro, James W. in “Status, Party, Ideology, Issues and Candidate Choice: A Preliminary Theory-Relevant Analysis of the 1968 American Presidential Election” (a working paper prepared for Specialist Meeting B:XI, Eighth World Congress of the International Political Science Association, 08 31-09 5, 1970)Google Scholar.

⁹ There are errors involved in all the estimates presented in Table 3 (particularly large ones in 1964). Since the three procedures for estimating the division of the vote have been applied to the same data for any given election, however, these sources of error have been (for our purposes) held constant.

¹⁰ We are using the term, “net score,” here and elsewhere to refer to the quantity yielded by subtracting the net number of favorable attitudes that a voter has toward the Republican candidate from his net number of favorable attitudes toward the Democratc candidate. See Appendix.

¹¹ Note that Figure 2 is really two figures: The average number of voters with each net score is given by the full height of the bars, while the average number of nonvoters is given by the lower portions of the bars.

¹² William R. Shaffer tried to simulate decisions of voters to vote and voters' choices among candidates in accordance with rules inspired by the analysis of the “components of electoral decision” by Campbell, Converse, Stokes, and Miller. His decision rule for choice among candidates was nearly the same as ours, and he tested the accuracy of his simulation against data from the SRC survey of the 1964 presidential election that was only slightly different from that with which we worked in analyzing that election. His rule for decisions about whether to vote did not work well, and he concludes that “abstention does not appear to be rooted in the same preferential decision-making process” ( Shaffer, , Computer Simulations of Voting Behavior, p. 133 Google Scholar). We think there is good reason to believe that some of the same considerations enter both the decision to vote and the decision about how to vote, but there is equally good reason to think that each is determined by a different set of considerations. For instance, once in the polling booth, considerations of convenience will play no part in a choice among candidates, since it is equally convenient to vote for any of them, but convenience does have an important bearing on the decision about whether to vote.

¹³ This figure holds, of course, not for all Catholic voters, but for those Catholic voters (a) whose votes were predicted on the basis of their attitudes, and (b) who expressed a preference for a candidate in their pre-election interview.

¹⁴ We thought that differentiated weighting of considerations by respondents in two other groups—Southerners in 1964 and blacks in 1968—might be reflected in the error rates of our predictions for them. The error rate in predictions for Southern respondents voting in the 1964 election turned out to be virtually identical to that for all respondents—9.1 per cent as against 9.3 per cent. The case of black respondents in 1968 is astounding, but it makes no better evidence for unequal weighting of considerations than the case of Catholic voters in 1960. Our predictions for black respondents voting in the election of 1968 were correct in 97 per cent of the cases, and we made no errors at all in predicting that blacks would vote Democratic. This phenomenal accuracy might seem to be evidence that black respondents were tied to the Democrats by a consideration or considerations of more than average weight. That could be true, but two facts count against this interpretation of the data: (1) In fewer than 20 per cent of the cases did the Rule yield a net score of + 3 or less for those black respondents predicted to vote for Humphrey; (2) among those black respondents whose votes were predicted from attitudes and who expressed a preference for a candidate in their pre-election interviews, our predictions squared with the preference in 98 per cent of the cases.

¹⁵ Probably the best way to find out the extent to which errors in predictions based on the Rule derive from the unequal weighting of considerations by respondents would be to modify the set of questions described in f.n. 4 so that respondents could say whether any particular like or dislike was of unusual or decisive importance. This kind of data, which we would have liked very much to have had, was unavailable to us.

¹⁶ The figures from which this average was computed were, for 1952, 73.5 per cent; for 1956, 66 per cent; for 1960, 71.9 per cent; and for 1964, 80.4 per cent.

¹⁷ In raw form our data consisted of the number of respondents interviewed on a given date and the number of these whose votes had been predicted incorrectly. If we had treated all respondents interviewed at the same time as the unit of observation, we would have given equal weight to cases involving only one respondent (with a consequent error rate of either 100 per cent or zero per cent) and cases involving many more. It was to avoid this weighting problem that we bunched respondents into equal groups of 50.

¹⁸ Joseph Napolitan, the well-known campaign consultant, observes: “Political time buying on television and radio follows a traditional curve; you usually start about three weeks before the election and build up to a heavy concentration the final week. We've even evolved a reasonable formula: spend one sixth of your money the third week before the election, two sixths the second week, and three sixths the final week” ( The Election Game: [Doubleday and Co., Inc.: Garden City, N.Y., 1972], pp. 143–144 Google Scholar).

¹⁹ See Stokes, Donald E., “Some Dynamic Elements of Contests for the Presidency,” The American Political Science Review 60 (03, 1966), pp. 19–28 CrossRefGoogle Scholar.

²⁰ Clausen, Aage R., “Response Validity: Vote Report,” Public Opinion Quarterly 32 (Winter, 1968–1969), 602 CrossRefGoogle Scholar.

²¹ Ibid., p. 603.

²² A more precise description of the procedure we followed in adjusting samples is in order here. In each case the number of respondents removed from our samples of voters was equal to 10 per cent of the corresponding population sample. Those respondents removed all had net scores between ± 4 and had expressed no more than four likes and dislikes of parties and candidates. Because the number of respondents to be removed was in every case fewer than the total number of respondents with these scores, respondents were removed from the sample in this order: (0), (0); (±0), (0); (±0), (2); (±0), (4); (±1), (1); (±1), (3); (±2), (4); (±3), (3); (±4), (4); where (±0), (0) indicates respondents with net scores of ±0) who expressed no likes or dislikes of parties and candidates, (±0), (2) indicates respondents with net scores of (±0) who expressed two likes or dislikes of parties and candidates, etc. When the total number of respondents removed would have been too large if we had removed all those in the last set of respondents to figure in our calculations, we kept within bounds by taking a smaller number proportionately from respondents with plus and minus net scores.

²³ Politically sophisticated readers may find it difficult to believe that anyone could forget how he voted for President. If so, they should try to recall for sure how they voted for freeholder or registrar of deeds at the last election; for some voters, a vote for president may carry with it no greater emotional involvement than the sophisticate's vote for one of these lesser officers of government.

²⁴ Working with data from postelection surveys of the 1968 election, Brody and Page attempted to predict how respondents had voted from their responses to a request to rate each of the presidential candidates on a scale ranging from 0 (very unfavorable) to 100 (very favorable). Predictions were made in terms of a “decision rule model,” which postulated that “if a voter favors one candidate even slightly more than others, he will vote for that candidate” and (an amendment) “party identifications break ties.” When this rule was applied to respondents' ratings of candidates, respondents' votes were correctly predicted in 90 per cent of the cases. These results are not directly comparable to ours, since our predictions were based on data taken from pre-election surveys. The conception of voting arrived at by Brody and Page, however, is obviously quite similar to our own. They have also suggested that their decision rules might be applied “to candidate evaluations in previous elections—computed from the S.R.C. open-ended items, lacking ‘thermometer’ ratings.…” That, of course, is very nearly what we have done. Brody and Page, “Indifference, Alienation and Rational Decisions.”

²⁵ Stokes, Donald E., Campbell, Angus, and Miller, Warren E., “Components of Electoral Decision,” pp. 368–369. Emphasis in the originalGoogle Scholar.

²⁶ That term can of course be construed differently. One may speak of a person as “rationalizing” if, when explaining or justifying an action, he cites only such actual motives as are socially acceptable—he tells the truth, but not the whole truth. To call the responses at issue here “rationalizations” in this sense, however, is to admit that they are expressions of genuine motivating attitudes.

²⁷ Converse, Philip E., “The Nature of Belief Systems in Mass Publics,” p. 249 Google Scholar, in Ideology and Discontent, ed. Apter, David E. (The Free Press: Glencoe, 1964)Google Scholar.

²⁸ Cf Brody and Page.