1 The reader may wish to refer to some of the literature on voting behavior which contains these divergent interpretations. I have selected a few prominent examples of studies that have emphasized (1) party identification, (2) candidates, (3) issues, or (4) ideology as the major explanatory factor in voting decisions. The selected studies have all been published within the last ten years, and all are based on survey rather than aggregate data.
(1) Party identification emphasis: Converse, Philip E., Miller, Warren E., Rusk, Jerrold G., and Wolfe, Arthur C., “Continuity and Change in American Politics: Parties and Issues in the 1968 Election,” American Political Science Review 63 (12, 1969), see especially pp. 1096–1099. (2) Emphasis on candidates: Stokes, Donald E., “Some Dynamic Elements in Contests for the Presidency,” American Political Science Review 60 (03, 1966), see especially p. 27. Kelley, Stanley Jr. and Mirer, Thad W., “The Simple Act of Voting,” American Political Science Review 68 (06, 1974), 572–591. (3) Emphasis on issues: Key, V. O. Jr., The Responsible Electorate (Cambridge: Harvard University Press, 1966). Boyd, Richard W., “Popular Control of Public Policy: A Normal Vote Analysis of the 1968 Election,” American Political Science Review 66 (06, 1972), 429–449. Others, such as Pomper, have studied issues, but have not related issues to voting behavior. See Pomper, Gerald M., “From Confusion to Clarity: Issues and American Voters, 1956–1968,” American Political Science Review 66 (06, 1972), 415–428. (4) The ideological factor: Nie, Norman H. with Andersen, Kristi. “Mass Belief Systems Revisited: Political Change and Attitude Structure,” Journal of Politics 36 (08, 1974), see especially pp. 580–583. Field, John Osgood and Anderson, Ronald E., “Ideology in the Public's Conceptualization of the 1964 Election,” Public Opinion Quarterly 33 (Fall, 1969), 380–398. Pierce, John C., “Party Identification and the Changing Role of Ideology in American Politics,” Midwest Journal of Political Science 14 (02, 1970), 25–42.
2 Converse, Philip E., “The Nature of Belief Systems in Mass Publics,” in Ideology and Discontent, ed. Apter, David E. (New York: Free Press, 1964), pp. 206–261.
3 These data were made available through the Inter-University Consortium for Political Research. Computer time was provided by The University of Connecticut Computer Center. The analysis and interpretation of the data are my own.
4 The Most Salient Issue (MSI) variable in Equations 2, 3, and 4 is based on an open-ended question that asked, “What do you personally feel are the most important problems the government should try to take care of?” In 1972, the wording of this question was changed to read: “What do you think are the most important problems facing the country?”
The party identification measure (PI) is based on the standard SRC-CPS question that has been asked since 1952. See Miller et al. footnote 2 for the exact wording.
The measure of attitude toward each of the candidates in 1960, 1964, and 1968 was developed from open-ended questions that asked respondents' likes and dislikes about the two (or in 1968, the three) candidates. The measure is the sum of the positive comments about a candidate minus the sum of the negative comments. In 1972, the measure of candidate attitude based on open-ended comments was not available in conjunction with the MSI measure. (Two separate samples with two different questionnaires had been used in the 1972 election study.) Therefore, the measure of candidate attitude in Equation 1 had to be based on a pre-election feeling “thermometer” question. A careful check was made to see if the “thermometer” measure could validly be used as a close surrogate for the measure based on open-ended questions. I found that it could, with only slight adjustments necessary to make it comparable with previous election year measures. (Detailed information on the testing procedure and adjustments can be obtained from the author.)
For the 1968 equation (Equation 2), I used a dichotomous dependent variable (V) with both a Wallace vote and a Nixon vote scored as a non-Democratic vote. Alternative methods of dealing with the three-candidate race in 1968 were explored and this method was found to be the best.
The number of cases that each equation (1 through 4) is based upon is the total number of voters in each election study. Respondents who did not have an attitude on a particular variable were placed at the mid or neutral point of the attitude measure.
5 See my article “Issue Salience and Party Choice,” American Political Science Review 65 (06, 1971), 389–400, for a description of the Strength of Issue Partisanship measure. The measure was based on degree of concern and party preference on a number of issues. Since the 1972 election study recorded party preference on only one issue (the one problem the respondent thought was most important), the Strength of Issue Partisanship measure could not be built for 1972. The measure could not be calculated for 1960 either because the 1960 election study did not probe for degree of concern. To maintain comparability from one equation to the next, therefore, the MSI measure had to be based on only one issue—the single most important problem—in each election study.
I was able to build the Strength of Issue Partisanship measure for 1964 and 1968. When this multi-point measure, based on a more complete set of issue concerns, was substituted for the MSI measure in Equations 2 and 3, the beta-weight of the issue variable increased by about .03 in each equation.
6 I want to stress that I am talking here of the direct effect of party identification in explaining vote. The indirect effects of party identification in influencing attitudes toward candidates and perceptions of party differences on issue concerns is not examined in this comment.
7 In order to keep these issue measures as similar to the MSI measure as possible, a party proximity measure was calculated for each preformulated issue used.
8 This reiterates an observation that was first made in my article “Issue Salience and Party Choice.”
9 In all fairness, I should point out that in the original version of the 1972 election analysis by Miller et al., as reported in a paper presented at the 1973 American Political Science Association meeting, the candidate factor was more clearly identified and discussed than in their present article. About all that remains of that original discussion and analysis is a sentence near the end of the present article that says: “After the affective ratings of the candidates, the next most important short-term [sic] factor accounting for the voter's decision was a liberal-conservative ideology ….” But even in the original, the authors placed heavy interpretive emphasis on factors other than the candidates.
For reasons explained in my discussion of the candidate proximity measure, the beta-weights for the candidates in Miller's multiple regression equations were not as large as they should have been. Had the beta-weights for candidates in their analysis been at the high level that I found in my Equation 1, Miller et al. would probably have given the candidate factor more importance in their interpretation.
10 Brody, Richard A. and Page, Benjamin I., “Comment: The Assessment of Policy Voting,” American Political Science Review 66 (06, 1972), 455.
11 Net attitude toward candidates was measured by simultaneously observing respondents' affect toward both candidates. (Affect was measured by the pre-election feeling “thermometer” question.) Those who were relatively warm toward Nixon and cool toward McGovern were put in the “Nixon More Favorably Perceived than McGovern” group. To be included in this group, there had to be at least a 30° difference between the candidates in thermometer ratings. Similarly, the “McGovern More Favorably Perceived than Nixon” group was composed of those who were substantially warmer toward McGovern than Nixon. The “Little Difference in Feelings or Neutral” group were those who placed both candidates within about 10° of each other on the “thermometer,” or who placed both candidates at the 50° or neutral mark.
12 There are certain difficulties with Boyd's S measure. (See Boyd, Richard W., “Popular Control of Public Policy,” American Political Science Review 66 [06, 1972], 448–449.) One problem stems from the use of absolute values in the computational formula. As a result, Boyd's S is unable to detect whether the observed vote line moves away from the expected vote line in an appropriate direction. For the data represented by my Figure 1b I have adjusted Boyd's measure to compensate for this shortcoming.
I should also point out that Boyd's S cannot be used as a measure to compare the overall impact of one issue versus another issue in an election. It lacks a standard base. Miller et al. do not try to use Boyd's S in this way. Boyd and others have attempted, however, to rank order the impact of issues in an election, using the S value as the indicator of impact. This is incorrect.
I have been working on modifications of Boyd's S to remedy these difficulties and will be happy to furnish further information on request.
13 This conclusion should not be interpreted to mean that issues were unimportant in the 1972 election. The reader should bear in mind that we are working here with responses to a single item in a survey and that these issue items are measured by preformulated fixed-choice questions. The importance or salience of the issue to the respondent is unknown. Miller et al. refer several times to the issues represented in their Figure 2 as “salient” and they assume that, taken alone, these preformulated questions are a measure of “attitude.” In their footnote 20 they point to the fact that a “saliency screen” is used with these questions. (The screen question was: “Or haven't you thought much about this?”) Such screening questions, which try to discourage respondents who do not have an attitude from answering the question, have been used in SRC-CPS studies since the 1950's. Yet these screens have not kept respondents who have only very casual or fleeting feelings about the questions from giving a response. The fact that in the normal vote analysis just presented (with candidate attitude controlled) responses to such preformulated issue questions were not strongly related to vote is evidence, once again, that salience is not detected by these preformulated questions.
14 Problems of projection and rationalization specifically related to these new issue scale questions have been discussed by Richard Brody and Benjamin Page in “Comment: The Assessment of Policy Voting” and also in their article “Policy Voting and the Electoral Process: The Vietnam War Issue,” American Political Science Review 66 (09, 1972), 986–989. Finch, Gerald worked extensively with proximity measures in his dissertation “Policy and Candidate Choice in the 1968 American Presidential Election” (Ph.D. dissertation, The University of Minnesota, 1973), see especially Chap. 3.
15 See footnote 20 in Miller et al. for the exact method of calculating candidate proximity. Whenever I use proximity measures, I follow the same method of calculation.
16 The candidate proximity measure for each of several different issues was correlated with the attitude toward candidate variable; the correlation varied to some extent depending on which issue was used. The resulting coefficients were averaged to arrive at the .350 figure.
Correlations were also calculated between respondents' party identification and party proximity measures for several issues. This yielded an average tau b of .250—a full point below the candidate attitude-candidate proximity link. Thus, although psychological adjustment takes place in locating party positions, the party proximity measure does not create as severe a distortion problem as the candidate proximity measure.
17 The reader should also be aware that when Miller uses path analysis to demonstrate the combined effect of party identification and issues on the vote—concluding that “the total effect of party identification was .51 whereas the issue effect was .55”—he leaves the candidate variable out of the model entirely. Miller's footnote 29, however, does show a causal model that includes the candidate variable. Thus, the picture we see (the arrow diagram in footnote 29) and the discussion in the text are based on quite different models.
In addition, Miller's measure of “Total Effects” in his footnote 29 is questionable. The usual method of measuring the contribution of each path in a model of this sort is as follows:
Miller apparently ascribes the variance in these four paths to the effects of party identification (PI) alone despite the fact that three of the paths include issues or candidates or both.
Miller's Issues effect (.58) and Candidate effect (.45) are apparently based on a different, three-variable model:
This three-variable model drops the party identification variable, yet the only way that this variable could properly be eliminated is if the coefficient for the direct path between party identification and vote (Path 4) were close to zero. Obviously this is not the case. See Stokes, Donald E., “Compound Paths in Political Analysis” in Mathematical Applications in Political Science, V, ed. Herndon, James F. and Bernd, Joseph L. (Charlottesville: University Press of Virginia, 1971), p. 82.
Finally, the direction of the arrow between Issues and Candidates in Miller's model should probably be reversed so that it goes from Candidates to Issues. The Issues measure is a compound of a number of candidate proximity measures, and we have seen how candidate attitude affects proximity.
18 Among pro-Nixon respondents, the distribution of responses to the Vietnam question were the same in the pre- and the postelection question. Perhaps with negotiations under way and the outcome uncertain, Nixon's position was not perceived as clearly as McGovern's; thus rationalizing responses would not tend to move toward a definite position on the scale.
19 I am referring here to the economics of the business cycle, not the cluster of issues that Miller et al. labeled “economic.” Their “economic” dimension had to do with questions of general social welfare.
20 If net feeling toward candidates is measured by the thermometer questions and is used in this correlation instead of net attitude toward candidates as determined by open-ended questions, the Pearsonian r goes up by only .04.
21 Prior to this, when discussing comments made about candidates, I have given proportions based on the number of comments. Here I am considering the proportion of respondents who made certain references. There were 1119 respondents in this sample.
22 Converse, , “Nature of Belief Systems,” p. 207.
23 Ibid., p. 214.
24 The exact format of the liberal-conservative scale question is as follows:
We hear a lot of talk these days about liberals and conservatives. I'm going to show you a seven-point scale on which the political views that people might hold are arranged from extremely liberal to extremely conservative.
Where would you place yourself on this scale, or haven't you thought much about this?
Where would you place Richard Nixon?
Where would you place George McGovern?
25 We do know from previous survey materials that people often bring meanings to the terms “liberal” or “conservative” that fall short of the concepts political scientists would supply to those terms. For example, Converse found a Stratum II that contained a rather large number of people who defined liberalism-conservatism in terms of a simple “spend-save” dimension. Others equated conservative with “conservation” and some even talked of conserving natural resources. (See Converse, , “Nature of Belief Systems,” pp. 222–223.)
George Gallup has also asked a national sample to state what “comes to your mind when you think of someone who is a (liberal) (conservative)?” “Spending too freely,” “generous,” “permissive”; “saves,” “cautious,” “careful” are some of the terms used in Gallup's codes to describe many of the responses to his question. (Incidentally, over 35 percent in the Gallup survey could not even begin to supply a meaning to the term “liberal” or the term “conservative.”) See The Gallup Opinion Index, Report No. 59 (May. 1970), pp. 6–7.
26 In his footnote 28, Miller says that he uses the “case deletion” procedure when performing computer analysis. This would eliminate any respondent who did not answer a question. Not only were the many respondents who did not place themselves on the liberal-conservative scale left out by this procedure, but others who failed to answer any one of the many issue questions examined by Miller were also dropped. Thus, the multiple regression analysis upon which he bases conclusions about the entire electorate contains only that subset of respondents who gave an opinion on all issues presented to them in the interview. Miller should have reported the number of cases remaining in his analysis and generalized only about that subset.
27 Pierce, John, “Changing Role of Ideology in American Politics,” p. 30.
28 A few respondents may have been cold toward both “liberals” and “conservatives” not because of indifference but because they have a strong ideology which is neither liberal nor conservative. (They may be “moderates,” “radicals,” “socialists,” etc.) This would not create a problem for our analysis here, however; it is the change in the proportions of “liberals” and “conservatives” over the past decade that is being investigated.
29 When 1972 respondents who had placed themselves on Miller's liberal-conservative scale were compared with those I identified as liberals or conservatives through the use of the thermometer questions (reported in Table 3), it was found that about 11 per cent more placed themselves as liberals or conservatives on Miller's scale. Interestingly, of those who were “Inconsistent or Neutral” as measured by the thermometer procedure, 17 per cent placed themselves as liberals, and 18 per cent as conservatives on Miller's scale—a 50-50 division that suggests randomness.
30 Other studies, beside Miller's, may have led political scientists to conclude that there has been an increase in the number of ideologically oriented people within the mass public. In particular, Norman Nie's article “Mass Belief Systems Revisited: Political Change and Attitude Structure” lends support to that conclusion. The reader is advised to take a careful look at the change in content of the questions Nie used in that analysis, particularly the marked difference in items that make up the “Cold War” dimension before 1964 and those used from 1964 onward.
In “Party Identification and the Changing Role of Ideology in American Politics” John Pierce found no increase in the proportion of “informational” ideologues in 1964. It was the “informational” ideologue that Converse was searching for in his “Nature of Belief Systems in Mass Publics.”
31 At one point in his “Nature of Belief Systems” article Converse did present matrices of interitem correlations. The purpose of that analysis was to compare a political elite with the general public. This use of the matrix method by Converse was ancillary and was not his principal method. See Converse, pp. 227–231.
32 To be considered consistently “liberal,” the respondent had to give “liberal” responses (scale positions 1, 2, or 3) to at least three out of the four questions. Failure to answer one question, or responding in the middle (4 position) on one question, was accepted, but the respondent could not have given a “conservative” response and still be considered consistent. The same criteria were used to determine consistent “conservatives,” except of course here we search for responses on the other side of the scale (positions 5, 6, and 7).
It is interesting that in the “social” cluster in Miller's Table 2, the average correlation was .39. With that degree of intercorrelation, only 30 per cent of the sample had “constrained” or consistent responses across the four items.
33 See Kessel, John, “Comment: The Issues in Issue Voting,” American Political Science Review 66 (06, 1972), 463.
34 Converse explicitly prescribed a “relatively wide range” of idea-elements in his concept of a belief system. “Nature of Belief Systems,” pp. 208–209.
35 Converse, , “Nature of Belief Systems,” pp. 234–238.
36 Twenty-two per cent were consistent “conservatives” and 14 per cent were consistent “liberals” in 1970.
37 The two questions had to do with the government in Washington seeing to it that Negroes get fair treatment in jobs and housing, and the government in Washington staying out of school desegregation.
38 About 40 per cent of the 1968 sample gave consistent responses to the following four items (about half were “liberal” and half “conservative”): government in Washington seeing to it that Negroes get fair treatment in jobs; government seeing to it that white and Negro children are allowed to go to the same schools; government support of the right of Negroes to go to any hotel or restaurant they can afford; and general favorable or unfavorable attitudes toward desegregation. An index built from these items strongly correlated with respondents' reactions to the speed of the civil rights movement, and the degree of violence they perceived in the movement.
39 Fourteen per cent of the respondents mentioned inflation or high prices as the most important problem in 1972, and an additional 9 per cent referred to unemployment or general economic conditions as their top concern. Altogether these economic issues were second only to Vietnam—which 25 per cent mentioned as the most important issue.
40 If Miller et al. had used just the pre-election responses to the Vietnam question in Table 1, they would have been less likely to have incorporated rationalized positioning on this issue item. Looking at pre-election responses only, the Democrats who voted for McGovern were 63 per cent dove and 15 per cent hawk—a 48-percentage-point spread rather than the 57-point difference shown in Miller's Table 1. Democrats who voted for Nixon were 32 per cent dove and 37 per cent hawk if only pre-election responders are included—a 5-percentage-point difference rather than the 9-point margin shown in Miller's Table 1.
41 When those with consistent “liberal” and “conservative” attitudes on Social Disruption were removed from Miller's Table 1, differences in the percentages of “left” and “right” responses between Democrats who had voted for McGovern and those who had voted for Nixon disappeared on four items—marijuana, campus unrest, aid to minorities, and busing. Differences between “loyalists” and defectors also vanished on two other issues—Vietnam and amnesty—when those who had a core attitude in the Military domain were removed from Miller's Table 1.
42 This is the same conclusion I reached in my article “Issue Salience and Party Choice.” There I pointed out that “candidate image emerges as the most important factor in individual voting choice in 1964, but issues had a strong independent effect as well” (p. 400).
43 For a further discussion of this point see my paper “Levels of Rationality Among the American Electorate” presented at the annual meeting of the American Political Science Association (1974), especially pp. 49–55.
Email your librarian or administrator to recommend adding this journal to your organisation's collection.
Email your librarian or administrator to recommend adding this journal to your organisation's collection.
* Views captured on Cambridge Core between <date>. This data will be updated every 24 hours.
Usage data cannot currently be displayed