Hostname: page-component-7c8c6479df-27gpq Total loading time: 0 Render date: 2024-03-27T17:28:07.465Z Has data issue: false hasContentIssue false

The Electoral Impact of Congressional Roll Call Voting

Published online by Cambridge University Press:  01 August 2014

Robert S. Erickson*
Affiliation:
Florida State University

Abstract

This paper presents evidence that candidate issue positions have a measurable impact on elections for the U.S. House of Representatives. For eight election years, electoral margins of Northern incumbent congressional candidates were examined to test the proposition that “moderates” within each party are better vote getters than those whose roll call records reflect their party's ideological extreme. The effects of roll call positions on election results were estimated by examining the relationships between roll call “extremism” and vote margins with district presidential voting held constant as a control for normal constituency voting habits. Although no strong support was found for the proposition that Democratic Representatives lose electoral support when they take extremely liberal roll call positions, a clear pattern emerged for Republicans: the Republican Congressmen who are the best vote getters tend to be the relative moderates and liberals who avoid the extreme conservative end of the political spectrum. An analysis of survey data suggests that the small group of voters whose electoral decisions are influenced by their Republican Congressman's roll call performance are found within the ranks of a select group who are both free of strong partisan motivations and highly politically informed.

Type
Articles
Copyright
Copyright © American Political Science Association 1971

Access options

Get access to the full version of this content by using one of the access options below. (Log in options will check for institutional or personal access. Content may require purchase if you do not have access.)

References

1 Miller, Warren E. and Stokes, Donald E., “Constituency Influence in Congress,” American Political Science Review, 57 (1963), 4556CrossRefGoogle Scholar; Stokes, Donald E. and Miller, Warren E., “Party Government and the Saliency of Congress,” Public Opinion Quarterly, 26 (Winter, 1962), 531546CrossRefGoogle Scholar.

2 Miller and Stokes, op. cit., p. 54.

3 Miller and Stokes, op. cit., p. 55.

4 Downs, Anthony, An Economic Theory of Democracy (New York: Harper and Row, 1957)Google Scholar, Chapter 8.

5 It is impossible to estimate the exact degree to which uncontrolled confounding influence works against the hypothesized roll call effect in congressional elections. Relevant but unknown factors are the degree to which Congressmen act as electoral opportunists rather than ideologues, the degree to which “opportunists” react to their expected electoral margins as opposed to district interests, and the degree to which Congressmen are sensitive to those causes of their vote margins that are not held constant by the control for district presidential voting. An additional consideration is that the vote for Congress may have a feedback effect on presidential voting. That is, congressional candidates' “coattails” may have some slight effect on the district's presidential vote division. Any such tendency, however, would simply be one additional reason why the roll call effect may be underestimated rather than overestimated. This is because if both the congressional coattail effect and roll call effect are at work, a Congressman's “extreme” roll call position would slightly lower his party's presidential vote as well as his own vote.

6 Before each Congressional election, The New Republic presents a “box-score” of each Representative's roll call performance in the previous Congress on what this liberal journal considers to be important and representative issues. A Congressman's New Republic rating is scored here as the frequency of his support for the liberal New Republic position on these selected items. (Failure of the Congressman to take a position on an issue is scored as half a vote for each alternative.) The number of roll calls included on this scale for a particular Congress ranges from nine to sixteen and includes a cross section of social welfare, civil rights, and foreign policy issues. For 1960, because The New Republic failed that year to present roll call records for the outgoing Congress, liberalism scores given by the Americans for Democratic Action are substituted.

7 This pattern of correlations is consistent with the relationships between district voting and roll call performance that have been previously reported. Negligible correlations between roll call “extremism” and measures of district “safeness” that are based on congressional election results have been reported in Shannon, Wayne, “Electoral Margins and Voting Behavior in the House of Representatives: The Case of the Eighty-Sixth and Eighty-Eighth Congresses,” Journal of Politics, 30 (November, 1968), 10281045CrossRefGoogle Scholar; Waldman, Loren K., “Liberalism of Congressmen and the Presidential Vote in their Districts,” Midwest Journal of Political Science, 11 (February, 1967), 7385CrossRefGoogle Scholar; and Froman, Lewis A. Jr., Congressmen and their Constituencies (Chicago: Rand McNally, 1963), pp. 9295Google Scholar. (Although Froman claims to have found an appreciable relationship, Shannon has shown that Froman's data do not support this conclusion.) A positive relationship between Democratic presidential voting and roll call liberalism has been reported in Cummings, Milton C. Jr., Congressmen and the Electorate, (New York: The Free Press, 1966), pp. 114116Google Scholar; in Waldman, op. cit.; and (for the South) in Flinn, Thomas A. and Wohnan, Harold, “Constituency and Roll Call Voting: The Case of Southern Democratic Congressmen,” Midwest Journal of Political Science, 10 (March, 1966), 192199CrossRefGoogle Scholar.

One previous examination of the relationship between congressional election outcomes and roll call liberalism with district presidential voting held constant is a brief analysis by Cummings (op. cit., pp. 118–119) of the 1948 election. Cummings grouped Republican incumbent candidates on the basis of similarity of district presidential voting and found an intriguing tendency for the average liberalism score (defined as ADA ratings) of the winners to be higher than that of their counterparts in the same group who were defeated. Further evidence of a roll call effect can be found in Schoenberger, Robert A., “Campaign Strategy and Party Loyalty: The Electoral Relevance of Candidate Decision-Making in the 1964 Congressional Elections,” American Political Science Review, 63 (June, 1969), 515520CrossRefGoogle Scholar. Schoenberger's data indicate that Republican Congressmen who supported Goldwater prior to the 1964 convention (presumably the most conservative) represented districts that were more Republican in their presidential voting but less Republican in their congressional voting than the districts of their counterparts who did not support Goldwater at an early date.

8 The “North” is defined so as to exclude the Border States of Kentucky, Maryland, Missouri, Oklahoma, and West Virginia, in addition to the former Confederate States.

9 For the 1962 equations, a few districts in Connecticut, Michigan, and Wisconsin were excluded from analysis because the 1964 presidential vote was unavailable. This was due to redistricting between 1962 and 1964 and the drawing of district boundaries in a way that did not conform to county lines. Redistricting in Washington between 1956 and 1958 caused the exclusion of four constituencies in that state from the 1958 analysis. The extensive reshuffling of district lines between 1964 and 1968 was the reason why no analysis was made of the 1966 election.

10 Of course no equations with the residual vote as an independent variable were calculated for 1952 and 1968 because there was no measure of the residuals from the immediately prior 1950 and 1966 election years. When the residual vote was controlled, the number of cases was slightly reduced because of the exclusion of districts with freshman Congressmen and Congressmen who faced no major party opposition in the previous election. In the equations from which the residual vote was derived, the measure of roll call performance that is employed is the Congressman's New Republic rating. Two independent variables were included in the equations from which the residual vote values were derived, but were dropped from the equations from which the roll call effect estimates in this paper were taken. These two variables are the Congressman's seniority and an index of state forces (the coattail effects of statewide contests for Governor or U.S. Senator). Seniority, as measured by the number of years served in Congress, was found to have neither a strong nor a consistent “effect” on congressional election margins. Estimated from the average of the regression coefficients for seniority, one decade of seniority offers the Congressman no more than an additional one per cent increase in his share of the two-party vote. The index of state forces is the set of residuals from equations with the vote for statewide office regressed on the state's presidential vote. When the index of state forces is included as an independent variable in the equations predicting the district congressional vote, the regression coefficients for this variable, although tending in the expected direction, are quite erratic. The estimates of the roll call effect when either or both of these rejected variables are controlled are virtually identical to those without these controls that are reported in this paper.

11 See Campbell, Angus and Miller, Warren, “The Motivational Basis of Straight and Split Ticket Voting,” American Political Science Review, 51 (June, 1957), 293312CrossRefGoogle Scholar.

12 For districts with Republican incumbent candidates, the median correlations between the presidential and congressional vote divisions over the five presidential elections is + .51 for single choice districts but only + .28 for multiple choice districts. A less dramatic difference in median correlations of + .88 vs. + .83 is found for districts with Democratic incumbent candidates. (The presidential vote division for 1968 is the Humphrey percentage of the three-party vote.) A similar ballot effect is found on the correlations between congressional voting in midterm elections and the presidential vote in the surrounding presidential years. These correlational differences suggest that candidates' characteristics—including roll call performances—may have a greater impact on congressional election results when the multiple choice ballot is used. But these correlational differences also indicate that the multiple choice ballot lowers the adequacy of presidential voting as an indicator of the Congressman's expected vote margin. Consequently the least confounding influence and the most accurate estimates of the roll call effect should be found for single choice districts.

13 For presidential years, the average slopes (b's) of the regression of the congressional vote on the presidential vote are (with New Republic ratings held constant) as follows: for districts with Republican incumbent candidates, +.76 with a single choice ballot and +.46 with a multiple choice ballot; for districts with Democratic incumbent candidates, +.80 with a single choice ballot and + .72 with a multiple choice ballot The tendency is similar for midterm years. There appears to be no consistent additive effect of the ballot form, since the points of intersection of the single choice and multiple choice slopes do not deviate systematically from the intersections of the two mean vote divisions for districts with incumbent candidates of the same party.

An explanation for the ballot form's impact on the regression of the congressional vote on the presidential vote can be drawn from the ballot form's impact on the propensity toward split ticket voting. Consider the following hypothetical example in which the ballot form is assumed to have the maximum possible impact. Suppose that in single choice districts the individual voter's congressional vote is perfectly dependent on his presidential choice (i.e., a perfectly operating presidential coattail effect). If so, at the district level there would be a one to one correspondence between the vote division for President and the vote division for Congress. The regression coefficient for this hypothetical relationship would be a steep 1.0. Now suppose that the multiple choice ballot encourages perfect split ticket voting so that there is absolutely no relationship at the individual level between the voter's presidential and congressional choices. If so, there would be no relationship at the district level (the expected congressional vote division in each district would be the mean) and the slope would be zero. The same logic applies to midterm elections also, with the presidential vote divisions in surrounding presidential years indicating the distribution of district partisanship and congressional voting at the individual level assumed to be more consistent with party identification when the single ballot is employed.

14 An alternate means of controlling for the ballot form that would not involve separate equations for single choice and multiple choice districts would be the inclusion of the ballot form and an interaction term (the product of the ballot form and the presidential vote) as variables in the equations. On this method of controlling for interaction in regression analysis, see Draper, Norman and Smith, Harry, Applied Regression Analysis (New York: John Wiley and Sons, 1966), pp. 140141Google Scholar. The hypothesized ballot impact on both the roll call effect and the degree to which confounding influence is controlled (see footnote 12, above) was the major reason this method of control was not employed.

15 As a rule, regression coefficients should be reported in unstandardized form unless the metric of one or both of the variables lacks substantive meaning. Since the units of roll call indices applied to different Congresses are not directly comparable (because the selected items are different), unstandardized coefficients would seem inappropriate. However, the standarization of the units of the roll call indices, but not those of the easily interpretable dependent variable, overcomes this difficulty. Assuming that the spread of a party's ideological distribution in Congress is constant over time, standardization of the roll call index makes the slopes derived for different election years directly comparable. This procedure also makes the slopes using alternate roll call indices for the same election year comparable. On choosing whether or not to standardize the slopes, see Blalock, Hubert M., “Causal Inferences, Closed Populations, and Measures of Association,” American Political Science Review, 61 (March, 1967), 130136CrossRefGoogle Scholar.

16 Scores on these two indices were computed as the frequency of party voting or opposition to the Democratic Administration on only those relevant roll calls on which the Congressman took a position. This method leaves the scores unaffected by rates of absenteeism.

17 Over eight Congresses, the median correlation between Party Unity scores and New Republic ratings is +.78. For the last three Congresses analyzed, the median correlations between the indices are: New Republic-Party Unity, +.87; New Republic-Administration Opposition, +.78; and Party Unity-Administration Opposition, +.92. Unlike the other two indices, which have distributions that are somewhat negatively skewed, Administration Opposition scores approximate a normal distribution, and for that reason may offer the preferred index for those years it can be employed.

18 Over all eight elections, the average of the sixteen slopes using the Party Unity index is −1.5, which is the same as the average slope using the New Republic index without the residual vote control. Since alternate indices that are imperfectly correlated with each other produce equivalent slopes, each index would appear to have about the same amount of measurement error. Since random measurement error in the independent variable may depress the magnitudes of regression coefficients, imperfection in the measure of “conservetism” is a further reason, in addition to the problem of confounding influence, why the roll call effect may be underestimated.

19 For 1968, the vote for President is measured as Nixon's percentage of the three-party vote. For the midterm elections, the vote for President is measured as the average of the two-party vote divisions in the two surrounding presidential years. When the presidential vote in only one of the surrounding presidential years is available for one or both of the matched cases, only the vote for the one year is used.

20 To be matched, two districts with Republican Congressmen at least one standard deviation apart on the appropriate roll call index had to meet the following minimum criteria:

1. They are within the same state and either adjacent or both entirely within the same SMSA

2. They are within 5 per cent of each other in the two-party division of the presidential vote (with the presidential vote measured by the criteria indicated in footnote 19)

3. The total two-party vote in each district totals at least 97% of the total votes cast.

When more than one possible combination of matched pairs of districts were possible in the same state, the following criteria were applied, in descending order, until the tie was broken.

1. Maximize the number of pairs

2. Maximize the minimum difference in roll call performance

3. Minimize the maximum difference in presidential voting.

From the set of matched pairs obtained by this method, we excluded those in which the Republican presidential vote in the district with the more conservative Congressmen exceeded that in its matched district by more than 3.5 per cent. This made the mean Republican percentages of the two-party presidential vote in the two sets of districts equal (summed over all cases) to within 0.1 per cent of each other. Before this correction, the presidential vote in the districts with “conservatives” was over one per cent more Republican than that in the districts with “liberals.” Since the Congressman's lead over his presidential ticket is inversely related to his party's share of the presidential vote, the correction was necessary to insure against a slight spurious inflation of the estimated roll call effect.

21 For the 1963–1964 Congress, Party Unity and Administration Support Scores of Northern Democrats were found to be correlated at a weak +.38 and + .47, respectively, with New Republic ratings. Their correlations with the Liberal Issue scores (discussed below) are an even less substantial +.18 and +.17.

22 The median correlation between Liberal Issue scores and New Republic ratings is +.64. The number of roll calls included in the Liberal index are: 1961–1962, 12; 1963–1964, 16; 1967–1968, 62. The large increase in the number of includable roll calls for the 1967–1968 Congress appears to reflect the increased division within the Northern wing of the Democratic party. A list of the roll calls included in the Liberal Issue index is available from the author upon request.

23 For the matched Democratic Congressman for 1968, the presidential vote is measured as the Humphrey percentage of the three-party vote. Otherwise the procedure is the same as for the matching of Republicans. Exclusion of the originally matched pairs in which the liberal's district is more than 3.5% more Democratic produces a virtual equal balance in the mean presidential voting 6f the two groups, as this correction did in the matching of Republicans.

24 For a detailed demonstration of this partisan difference, see Mayhew, David, Party Loyalty Among Congressmen (Cambridge: Harvard University Press, 1966)CrossRefGoogle Scholar.

25 Correlations will be larger with larger units (e.g., districts rather than individuals) because through the use of larger units, various disturbing influences are controlled. See Blalock, Hubert M. Jr., Causal Inferences in Non-Experimental Research (Chapel Hill: University of North Carolina Press, 1964). pp. 95114Google Scholar.

26 In a supplementary exercise, the Congressman's Administration Opposition score was included as an independent variable in a regression analysis of the individual data predicting the vote of the respondents who were Johnson voters with Republican Congressmen. Control variables were the respondent's party identification, the ballot form, an index of the respondent's issue position based on five domestic policy items, and the district two-party vote for President (to control for a slight effect of one-party dominance on congressional voting). The slope for the Congressman's roll call position (−4.9) indicated that an increase of one standard deviation in the Congressman's conservatism decreased the probability of a Johnson supporter's voting for him by 4.9 per cent. A similar regression analysis for Goldwater voters with Republican Congressmen produced the slightly positive slope of +.0.4. Taking into account the fact that 60 per cent of the survey's congressional election voters in districts with Republican Congressmen reported that they voted for Johnson, the overall estimate of the roll call effect for Republican Congressmen derived from the regression analysis of survey data is a loss of 2.8 per cent of the vote for each standard deviation of conservatism, or:

Given that the partial correlation coefficients accompanying the two slopes are a microscopic −.10 and +.01, this estimate is surprisingly close to the estimates of the 1964 roll call effect for Republicans from the aggregate analysis.

27 General political information is, as one would expect, correlated with recognition of the Congressman. In the sample of Johnson supporters with Republican Congressmen, 75 per cent of the “high recognition” group but only 47 per cent of the remainder could identify the Republican party as the most “conservative.” Sixty-eight per cent of the “high recognition” group but only 55 per cent of the remainder monitored at least three types of mass media during the campaign.

28 Even with only fifteen cases, this result is sufficiently strong to be statistically significant at the .05 level, based on Fisher's exact probability test The subsample appears to be reasonably “representative,” since the fifteen respondents are drawn from eleven separate constituencies.

29 A rough estimate of the proportion of the voting electorate that possesses sufficiently high levels of recognition and involvement to be influenced by the roll call performance of their Republican Congressman, given sufficient crosspressures, is eighteen or nineteen per cent. Eighteen per cent is the proportion of Johnson voters voting in congressional elections with Republican Congressmen who were in both the “high recognition” and “high information” category. Nineteen per cent is the proportion of Republicans and Independents for Johnson who scored high on both recognition and information. These figures, even if somewhat exaggerated, contrast sharply with the reported proportion of voters who in open-ended questions offer appraisals of their Congressman's policy position. Stokes and Miller (op. cit., p. 543) report that “by the most reasonable count, references to current legislative issues comprised not more than a thirtieth part of what the constituents had to say about their Congressmen.”