¹ Key, V. O. Jr., “A Theory of Critical Elections,” Journal of Politics, 17 (02, 1955), 3–18 CrossRefGoogle Scholar. The notion of critical elections has been broadened to include a whole typology of elections. See Campbell, Angus, “A Classification of the Presidential Elections,” in Campbell, Angus et al., Elections and the Political Order (New York: John Wiley and Sons, Inc., 1966), pp. 63–77 Google Scholar.

² Philip Converse, “The Concept of a Normal Vote,” in Campbell, Ibid., p. 13.

³ The images of the parties and candidates in the 1952 election are discussed in Campbell, Angus, Gurin, Gerald, and Miller, Warren E., The Voter Decides (Evanston: Row, Peterson and Co., 1954), pp. 41–68 Google Scholar.

⁴ This argument, of course, is the one made in Campbell, Angus et al., The American Voter (New York: John Wiley and Sons, Inc., 1960), chapter 2Google Scholar.

⁵ Campbell, The Voter Decides.

⁶ Since the 1952 election the multiple correlations of these predictors have varied between .72 and .75. See Stokes, Donald E., “Some Dynamic Elements of Contests for the Presidency,” American Political Science Review, 60 (03, 1966), 19–28 CrossRefGoogle Scholar.

⁷ Lazarsfeld, Paul, Berelson, Bernard, and Gaudet, Hazel, The People's Choice (2nd ed.; New York: Columbia University Press, 1948), p. 56 Google Scholar. For a discussion of cross-pressures that places the concept into balance theory, see Pool, Ithiel de Sola, Abelson, Robert, and Popkin, Samuel, Candidates, Issues, and Strategies: A Computer Simulation of the 1960 and 1964 Presidential Elections (Rev. ed.; Cambridge: The M.I.T. Press, 1965), pp. 9–15 Google Scholar.

⁸ Lazarsfeld, op. cit., pp. 52–64.

⁹ Berelson, Bernard, Lazarsfeld, Paul, and McPhee, Wiliam, Voting (Chicago: The University of Chicago Press, 1954), pp. 129–32, 230–31, and 283–84Google Scholar.

¹⁰ Campbell, , The Voter Decides, p. 87 Google Scholar.

¹¹ Campbell, , The American Voter, p. 141 Google Scholar. The prediction is successfully tested, p. 142.

¹² Hyman, Herbert H., Political Socialization (Glencoe: The Free Press, 1959), p. 46 Google Scholar.

¹³ Converse, Philip, “On the Possibility of Major Political Realignment in the South,” in Campbell, , Elections and the Political Order, pp. 212–244 Google Scholar.

¹⁴ A section on specific measurement procedures follows.

¹⁵ Statistical interaction is a change in the relationship between two variables over a range of a third. If the relationship is constant, it is said to be additive. See the Appendix for a lengthy discussion of the meaning of interaction and a method of measuring it.

₁₆ For an explicit statement of the proposition of interaction among attitudes, see Rokeach, Milton and Rothman, Gilbert, “The Principle of Belief Congruence and the Congruity Principle as Models of Cognitive Interaction,” Psychological Review, 72 (01, 1965), 128–142 CrossRefGoogle ScholarPubMed.

¹⁷ This does not mean that empirically one's method of analysis will always reveal an inconsistency effect as statistical interaction. The Appendix discusses the method of analysis.

¹⁸ Lipset, Seymour Martin and Bendix, Reinhard, Social Mobility in Industrial Society (Berkeley: University of California Press, 1964), p. 69 Google Scholar.

¹⁹ Lenski, Gerhard, “Status Crystallization: A Non-Vertical Dimension of Social Status,” American Sociological Review, 19 (08, 1954), 405–413 CrossRefGoogle Scholar. See also Lenski, Gerhard, “Status Consistency and the Vote: A Four Nation Test,” American Sociological Review, 32 (04, 1967), 298–301 CrossRefGoogle Scholar. Kelly, K. Dennis and Chambliss, William J., “Status Consistency and Political Attitudes,” American Sociological Review, 31 (06, 1966), 375–382 CrossRefGoogle Scholar. For a dissenting view, see Kenkel, William F., “The Relationship Between Status Consistency and Politico-Economic Attitudes,” American Sociological Review, 21 (06, 1956), 365–368 Google Scholar.

²⁰ Goffman, Irwin W., “Status Consistency and Preference for Change in Power Distribution,” American Sociological Review 22 (06, 1957), 275–281 CrossRefGoogle Scholar.

²¹ Bell, Daniel (ed.), The Radical Right (Books, Anchor ed.; Garden City: Doubleday and Co., Inc., 1964), pp. 39–61, 63–80, and 259–312 Google Scholar.

²² Rush, Gary B., “Status Consistency and Right-Wing Extremism,” American Sociological Review, 32 (02, 1967), 86–92 CrossRefGoogle Scholar.

²³ For example, Blau notes that mobility affects a whole range of behavior, including voting. “… the upwardly mobile are more likely to vote Republican than people who have remained workers and less likely to do so than those who have originated in the middle class,” Blau, Peter M., “Social Mobility and Interpersonal Relations,” American Sociological Review, 21 (06, 1956), 291 CrossRefGoogle Scholar. Presuming that there is not a commensurate change in the disposition among the mobile to switch party identification, then the mobile would be more likely to be numbered among the defectors that the non-mobile.

²⁴ Analogously, to the degree that a man gains or loses in the status race during his mature years, attitudes formed in one social situation are likely to conflict with attitudes formed in another. Thus, if adequate measures were available in the SRC data, we would also test the hypothesis that intra-generational mobility increases cross-pressures.

²⁵ Boyd, Richard W., A Theory of Voting Defection: Attitudinal Cross-Pressures and Political Alienation (Unpublished Ph.D. Dissertation, Department of Government, Indiana University, 1967), pp. 35–43 Google Scholar.

²⁶ The three elections are chosen because they are recent and because they cover the elections which serve as the tests of the hypotheses. The debt that this measure of voting defection owes to the concept of the normal vote should be obvious. See Converse, “The Concept of a Normal Vote,” op. cit. In brief, the rationale for the averaging procedure is the assumption that across elections the balance of partisan forces changes, not the pressure required to make a voter defect. What is being assumed, then, is that individuals have threshold points beyond which they cannot resist partisan forces counter to their identification. Furthermore, all men within a given category of identifiers have approximately the same threshold point; i.e., they have a similar degree of allegiance to their party. Different percentages of defectors over different elections are assumed to reflect changes in the strength of the attacks that partisan forces mount against men's threshold points. While men may change their identification, the threshold remains the same—for the category which they leave and for the one into which they move. Within any given category the reason some defect while others do not is simply that some individuals are exposed to more intense partisan forces running counter to their identification than are others. Using the average of the three elections for the measure keeps this threshold point constant and allows the partisan forces of the day to change.

The formula D = 2 (P — 50) is itself somewhat arbitrary. The multiplication by 2 allows the theoretical range of the variable to extend from 0 to 1. The multiplication has no effect on the analysis.

²⁷ Independent Republicans and Independent Democrats are excluded along with Independents.

²⁸ Matthews, Donald R. and Prothro, James W., “The Concept of Party Image and its Importance for the Southern Electorate,” in Jennings, M. Kent and Zeigler, L. Harmon (eds.), The Electoral Process (Englewood Cliffs: Prentice-Hall, Inc., 1966), p. 159 Google Scholar.

²⁹ It does not matter whether or not all these questions elicit responses which fall upon a single liberal-conservative continuum. All that is important is that the questions probe beliefs that are relevant to the voting decision. To the extent that issues ever affect electoral outcome, the questions that make up this index are important questions.

³⁰ For a description of the unbiased correlation ratio, see Blalock, Hubert M. Jr., Social Statistics (New York: McGraw-Hill Book Co., 1960), p. 267 Google Scholar. These ratios are based on a series of one-way analyses of variance. The estimate of the percentage of variance explained is obtained by multiplying the ratio by 100. For example, party compatibility explains 5.8 percent of the variance in voting defection in 1956.

³¹ We might note in this regard the rather low amount of variance in defection explained by policy compatibility in 1964 relative to candidate and party compatibility. Perhaps the direct impact of issues upon voting in 1964 was not as great as is popularly supposed.

³² In any year the three pairs are party and candidate, party and policy, and candidate and policy.

³³ The F ratio for interaction is the interaction mean squares divided by the total mean squares.

³⁴ When one is testing a proposition about interaction in specific cells, measures of association or F tests should not be a final determinant of acceptance or rejection in any case. These measures of association and F tests are not specific to certain cells. The impact of interaction in one cell is diluted by additivity in another. Thus, the variance explained by interaction may be small in spite of a large impact upon individuals in the interaction cell. This dilution is especially critical when, as is often the case, relatively few people occupy the cell manifesting the interaction effect. For this reason, we assign as much importance to a pattern the interaction displays as to either the variance the interaction explains or its statistical significance.

³⁵ The four dimensions of status upon which Hypothesis 3 is tested are racial-ethnic prestige, occupational prestige, family income, and education.

³⁶ Because the tables yield negative results they are omitted. They do appear in Boyd, op. cit., Appendix C.

³⁷ Jackson, Elton F. and Burke, Peter J., “Status and Symptoms of Stress: Additive and Interaction Effects,” American Sociological. Review, 30 (08, 1965), 556–564 CrossRefGoogle ScholarPubMed. See also Jackson, Elton F., “Status Consistency and Symptoms, of Stress,” American Sociological Review, 27 (08, 1962), 469–480 CrossRefGoogle Scholar.

³⁸ The same is also true for the relationship of mobility to voting defection in the three elections. The non-mobile defect more than the mobile in all three elections, though, again the magnitude of the differences is not great.

³⁹ See Table 4. In the 1964 election, however, the rank-order of the importance of the three sets of orientations as judged by their ability to statistically explain voting defection was candidate, party, and policy orientation. The impact of party attitudes is not discussed in relation to its possible impact upon campaign strategy, because there seems little a party could do directly to alter its image with the public. Changes in attitudes toward the parties probably result from candidates and issues.

⁴⁰ Quite likely, the type of issue necessary to affect a realignment is a “position issue” rather than a “valence issue.” For the distinction between the two, see Stokes, Donald E., “Spatial Models of Party Competition,” in Campbell, , Elections and the Political Order, pp. 171–172 Google Scholar.

⁴¹ For an illuminating discussion of the subject, complete with graphic examples, see Morgan, James N. and Sonquist, John A., “Problems in the Analysis of Survey Data, and a Proposal,” Journal of the American Statistical Association, 58 (06, 1963), 415–434 CrossRefGoogle Scholar.

⁴² For just a few of the many examples of social phenomena explicitly or implicitly conceptualized in terms of interaction, see the following article and the notes cited therein. Blalock, Hubert M. Jr., “Status Inconsistency, Social Mobility, Status Integration and Structural Effects,” American Sociological Review, 32 (10, 1967), 790–801 CrossRefGoogle Scholar.

⁴³ Blalock, , Social Statistics, p. 332 Google Scholar.

⁴⁴ As in the case of interaction, problems also crop up when multicollinearity is present. See Blalock, Hubert M. Jr., “Correlated Independent Variables: The Problem of Multicollinearity,” Social Forces, 42 (12, 1963), 233–237 CrossRefGoogle Scholar.

⁴⁵ The remainder of the discussion applies only to multiple-way analysis of variance. One-way analysis of variance, which is used to test some hypotheses, does not require an adjustment of unequal cell sizes.

⁴⁶ Winer, , Statistical Principles in Experimental Design, p. 224 Google Scholar. The method has been around a long time. See Yates, F., “The Analysis of Multiple Classifications with Unequal Numbers in the Different Classes,” Journal of the American Statistical Association, 29 (03, 1934), 51–66 CrossRefGoogle Scholar.

⁴⁷ Duncan, Otis Dudley, “Residential Areas and Differential Fertility,” Eugenics Quarterly, 11 (06, 1964), 82–89 CrossRefGoogle ScholarPubMed.

⁴⁸ Elton F. Jackson and Peter J. Burke, loc. cit.

⁴⁹ Treiman, Donald J., “Status Discrepancy and Prejudice,” American Journal of Sociology, 71 (05, 1966), 651–664 CrossRefGoogle ScholarPubMed.

⁵⁰ Suits, Daniel B., “Use of Dummy Variables in Regression Equations,” Journal of the American Statistical Association, 52 (12, 1957), 548–551 CrossRefGoogle Scholar.

⁵¹ Kempthorne, Oscar, The Design and Analysis of Experiments (New York: John Wiley and Sons, Inc., 1952), pp. 79–87 Google Scholar; Snedecor, George W., Statistical Methods Applied to Experiments in Agriculture and Biology (5th ed.; Ames, Iowa: The Iowa State College Press, 1956), pp. 338–391 Google Scholar; and Anderson, R. L. and Bancroft, T. A., Statistical Theory in Research (New York: McGraw-Hill Book Co., 1952), pp. 278–284 Google Scholar.

⁵² The assumptions which permit the determination of the coefficients are the following: (1) The sum of squares due to interaction shall be minimized.

(2) The sum of the interaction terms (the deviations from an additive model), weighted by the cell frequencies, shall sum to zero over each row and each column.

In short, the additive pattern in each table is defined as that pattern which minimizes the sum of squares attributable to interaction. The only difference between the procedure used in this paper and the more familiar method of analysis of variance is that, in this procedure, the distribution of cell frequencies is allowed for theoretical reasons to influence the estimates of interaction. The program that computed the coefficients and the adjusted sums of squares was written jointly by C. Richard Hofstetter of Ohio State University and this author.

⁵³ The use of a least squares technique to estimate the location of interaction is not uncontroversial. In the first place, so long as one's cell means conform to an additive model, the pattern of cell frequencies has no effect on the size of the regression coefficients. However, if one's cell means do not conform to an additive model, the pattern of cell frequencies will affect the coefficients and, thus, the magnitude of the deviations from an additive model. Furthermore, because the technique is based on least squares, the largest deviations will tend to be found in the cells with the smallest cell sizes. To guard against the possibility that my results are an artifact of the method I used, I also ran the analysis, substituting equal cell sizes while retaining the original cell means. The same results obtained. In the second hypothesis, the direction of the deviations in the 27 critical cells remained the same in every instance. The size of the deviations did not vary greatly from those in Table 6.

⁵⁴ Blalock, Hubert M. Jr., “Tests of Status Inconsistency Theory: A Note of Caution,” Pacific Sociological Review, 10 (Fall, 1967), 69–74 CrossRefGoogle Scholar.

⁵⁵ See the following works by Blalock, , “Review Symposium,” American Sociological Review, 33 (04, 1968), 296–297 CrossRefGoogle Scholar. “Status Inconsistency and Interaction: Some Alternate Models,” The American Journal of Sociology, 73 (11, 1967), 305–315 CrossRefGoogle Scholar. “The Identification Problem and Theory Building: The Case of Status Inconsistency,” American Sociological Review, 31 (02, 1966), 52–61 CrossRefGoogle Scholar.

⁵⁶ A few attempts have been made to partition chi-square into additive and interactive effects as in analysis of variance, but the technique seems as yet undeveloped. For example, diochotomization of the dependent variable is required. Wilson, K. V., “A Distribution-Free Test of Analysis of Variance Hypotheses,” Psychological Bulletin, 53 (1965), 96–101 CrossRefGoogle Scholar.

⁵⁷ Johnston, J., Econometric Methods (New York: McGraw-Hill Book Co., 1963), pp. 201–207 Google Scholar.

⁵⁸ Dixon, W. J. (ed.), Biomedical Computer Programs (Berkeley: University of California Press, 1967), pp. 258–275 Google Scholar.

⁵⁹ ICPR Codebooks, Survey Research Center, University of Michigan, Occupation Code.

⁶⁰ For example, see Jackson, op. cit., p. 471. See also Reiss, Albert J. Jr., Occupations and Social Status (New York: The Free Press of Glencoe, 1961), pp. 109–161 Google Scholar and Hodge, Robert W., Siegel, Paul M., and Rossi, Peter H., “Occupational Prestige in the United States, 1925–63,” American Journal of Sociology, 70 (11, 1964), 286–302 CrossRefGoogle Scholar.

⁶¹ Jackson, loc cit. For 1964, see ICPR Codebook, Deck 7, Col. 12, Deck 8, Col. 33, Deck 9, Cols. 11–13, 15–17, 21.

⁶² For 1956, Ibid., Deck 6, Cols. 10, 20–22, 25, Deck 5, Col. 27.

⁶³ For 1964, Ibid., Deck 9, Col. 44.

⁶⁴ For 1964, Ibid., Deck 7, Cols., 27–28.

⁶⁵ Jackson, loc. cit.