¹ Two noted critics are Mayhew, David, Congress: The Electoral Connection (New Haven, Conn.: Yale University Press, 1974)Google Scholar, and Fiorina, Morris, Congress: Keystone of the Washington Establishment (New Haven, Conn.: Yale University Press, 1977).Google Scholar

² For example, see Lowi, Theodore, The End of Liberalism, 2nd edn (New York: Norton, 1979)Google Scholar, and Burns, James MacGregor, Leadership (New York: Harper and Row, 1978), especially Chaps. 12 and 13Google Scholar. A recent empirical analysis of a collective model of representation in the House is Schwarz, John E., Fenmore, Barton and Volgy, Thomas J., ‘Liberal and Conservative Voting in the House of Representatives: A National Model of Representation’, British Journal of Political Science, X (1980), 317–39.CrossRefGoogle Scholar

³ Mayhew, , Congress, pp. 176–7.Google Scholar

⁴ The two dimensions are common in the literature on representation. Two classic examples are Eulau, Heinz, Wahlke, John C., Buchanan, William and Ferguson, Leroy C., ‘The Role of the Representative: Some Empirical Observations on the Theory of Edmund Burke’, American Political Science Review, LIII (1959), 742–56CrossRefGoogle Scholar; and Miller, Warren E. and Stokes, Donald E., ‘Constituency Influence in Congress’, American Political Science Review, LVII (1963), 45–56.CrossRefGoogle Scholar

⁵ MacRae, Duncan, ‘The Relation between Roll Call Votes and Constituencies in the Massachusetts House of Representatives’, American Political Science Review, XLVI (1952), 1046–55CrossRefGoogle Scholar; Turner, Julius and Schneir, Edward, Party and Constituency: Pressures on Congress (Baltimore: The Johns Hopkins University Press, 1970)Google Scholar. Fiorina cogently criticizes the tendency of early studies to confound party and constituency in this way. See Fiorina, Morris P., Representatives, Roll Calls and Constituencies (Lexington, Mass: D. C. Heath, 1974), Chap. 1.Google Scholar

⁶ The principal findings are reported in Pomper, Gerald, ‘From Confusion to Clarity: Issues and the American Voters, 1956–1968’, American Political Science Review, LXVI (1972), 415–28CrossRefGoogle Scholar. The findings are updated to 1972 in Pomper, Gerald, Voters' Choice (New York: Dodd, Mead, 1975), Chap. 8.Google Scholar

⁷ Pomper's work has been criticized by Margolis, Michael, ‘From Confusion to Confusion’, American Political Science Review, LXXI (1977), 31–43Google Scholar, and by Piereson, James, ‘Issue Alignment and the American Party System, 1956–1976’, American Politics Quarterly, VI (1978), 275–307CrossRefGoogle Scholar. There is also a debate in the literature over the comparability of survey opinion measures through the SRC/CPS time series. See Sullivan, John L., Piereson, James E. and Marcus, George E., ‘Ideological Constraint in the Mass Public: A Methodological Critique and Some New Findings’, American Journal of Political Science, XXII (1978), 233–49CrossRefGoogle Scholar, and Bishop, George F., Tuchfarber, Alfred J. and Oldendick, Richard W., ‘Change in the Structure of American Political Attitudes: The Nagging Question of Question Wording’, American Journal of Political Science, XXII (1978), 250–69CrossRefGoogle Scholar. I shall discuss the applicability of these criticisms to my work once I have presented my evidence (see fn. 19 below).

⁸ Pomper, , Voters' Choice, p. 168Google Scholar. A tendency for Democrats in the electorate to adopt the liberal position on issues, and Republicans to adopt the conservative position – termed ‘polarization’ in the text – is reflected in a positive correlation between party identification and issue position. Pomper's data show an increased average gamma correlation between party identification and four issues relating to domestic welfare: education/taxation, medical care, government guarantee of jobs, and fair employment/aid to minorities, up to 1964. In 1956–60 the average gamma is 0·15 or less. It jumps to 0·33 in 1964, then declines to 0·30 in 1968 and 0·21 in 1972. This shows that polarization between the parties in the national electorate was clearly at a peak in 1964 on these issues.

⁹ Pomper links his findings on the national electorate to a potential for a system of representation characterized by stronger parties and popular influence over public policy: ‘At appropriate times, [voters] align their party loyalty to their issue preferences. A link has been forged that can promote a… system in which the parties put forth programs, receive popular approval of these programs, and then carry out the popular mandate.’ (Pomper, , Voters' Choice, p. 182.)Google Scholar

¹⁰ Excluding 1962 and 1966 for lack of items on domestic welfare. The questions employed dealt with the federal government's guarantee of jobs (1956, '58, '60, '64, '68, '72), federal subsidization of medical care costs (1956, '58, '60, '64, '68, '70, '72), federal subsidization of local schools (1956, '60, '64, '68), a general assessment of the power of the federal government (1964, '68, '70), and federal action to remedy inflation (1970, '72).

¹¹ I selected items from the surveys based on whether they appeared to relate to domestic welfare. Once the questions were chosen on face criteria, I performed a principal components analysis (rotated to a varimax solution) of these and other issue items not included in this analysis. The principal components analysis was used only as a check, and in no case was an item placed on a dimension based solely on the principal components analysis. Once the items were identified, individuals were scored and aggregated to the constituency level. The mean score is the aggregate measure employed. See Stone, Walter J., ‘Measuring Constituency-Representative Linkages: Problems and Prospects’, Legislative Studies Quarterly, IV (1979), 623–40CrossRefGoogle Scholar for a more complete discussion of the methods employed.

¹² The method for selecting roll calls was similar to that utilized with the survey data. Roll calls were chosen for their similar content to the dimension identified in the mass data. Principal components analyses were performed until a set of roll calls was identified in each Congress that matched in content as closely as possible the survey items from the election studies. The roll calls dealt with a wide range of issues including the school milk programme, Department of Labor and HEW appropriations, food stamps, unemployment, housing and health care, and the minimum wage. A complete description of the roll calls and the final solutions of the principal components analyses are available from the author on request.

¹³ Measurement equivalence through time and between the survey items and roll calls is obviously a difficult problem. Weissberg, Robert (‘Assessing Legislator–Constituency Policy Agreement’, Legislative Studies Quarterly, IV (1979), 605–22)CrossRefGoogle Scholar has argued that equivalence demands that representatives and constituents be asked identical questions. Such a concept of equivalence misses the fact that electorates and their representatives necessarily operate in different worlds. We create complex legislative institutions, which most voters do not fully understand, we pay people to become professional lawmakers (itself a ‘non-equivalence’), and we do this because the nature of policy making demands a sustained effort that most people cannot afford. The roll calls I have included were surrounded by complex strategies and embedded in legislative minutiae which even most members of Congress did not comprehend fully. The survey items tap a similar core policy problem, but they in no way confront the respondent with the identical stimulus that confronts the member of Congress. Indeed, I suspect that it is impossible even with identical question wording since legislators would bring to their answers a set of experiences unavailable to the typical citizen.

¹⁴ The average sample size for the district majority (those who share the partisan affiliation of the incumbent) is nine interviews. The district minority consists of respondents identifying with the party opposite to that of the incumbent, and the average minority sample size is 5·6. The small district sample size results in sampling error confounding the district opinion measures. This means the results I report are conservative or attenuated estimates of the population parameter we would estimate if we had census opinion data from each district. Several important factors work against the objection that the district samples are too small to sustain the analysis. Firstly, my purpose is never to say anything about a single district, but rather to analyse reasonably large numbers of districts. This smooths the effect of the sampling error which would be prohibitively large if the purpose were to make conclusions about a single district. A reasonable assumption for this analysis is the more the small sample error dominates our estimates, the closer the expected value of statistical relationships employing the district estimates will be to zero. The coefficients might fluctuate substantially, but their expected value would approach zero. The coefficients I report depart substantially from zero and certainly do not appear to hover randomly around zero. Secondly, my purpose is not so much to identify the population parameter as it is to compare equally attenuated results. Thirdly, I employ a district weight which corrects for differences in district populations, and is proportional to the district sample size. Thus the analysis weights the larger district samples more heavily than the smaller district samples. Finally, the analysis I present is not based upon a single year where random fluctuations would be very difficult to see, but on seven election years. While a certain healthy scepticism is no doubt appropriate, I believe the results are internally consistent and conform with plausible theoretical expectations.

¹⁵ Stone, Walter J., ‘The Dynamics of Constituency Electoral Control in the House’, American Politics Quarterly, VIII (1980), 399–424.CrossRefGoogle Scholar

¹⁶ Greatest design discontinuities occur in 1960 since that year's study was based on a sampling frame different from the 1964 study. 1972 presents a similar problem and so is not included.

¹⁷ Stone, , ‘Dynamics’Google Scholar. This conclusion is substantiated by analysis of the civil rights dimension as well as the domestic welfare issues analysed here. Interestingly, the results reported here may result from opinion leadership of constituency opinion. Individual representatives may provide some of that leadership, but highly visible national leaders such as presidential candidates may stimulate considerable changes of opinion in the electorate.

¹⁸ Nie, Norman H. with Andersen, Kristi, ‘Mass Belief Systems Revisited: Political Change and Attitude Structure’, Journal of Politics, XXXVI (1974), 540–91CrossRefGoogle Scholar, and Nie, Norman H., Verba, Sidney and Petrocik, John R., The Changing American Voter (Cambridge, Mass: Harvard University Press, 1976).Google Scholar

¹⁹ Sullivan, et al. , in ‘Ideological Constraint’Google Scholar, present evidence that changed question wording accounts for most of the increase in the constraint correlations on which Nie's conclusions were based. By implication, any work on the time series using the attitude questions where wording has changed, is open to question. Without examining the full debate (for a response to Sullivan et al., see Nie, Norman H. and Rabjohn, James N., ‘Revisiting Mass Belief Systems Revisited: Or, Doing Research Is Like Watching a Tennis Match’, American Journal of Political Science, XXIII (1979), 139–75CrossRefGoogle Scholar), the evidence in this paper cannot be explained away by the Sullivan hypothesis. In the first place, I show changed levels of congruence with the 1960 election before the survey items were altered. Secondly, the predicted decline in congruence levels occurs following 1964, with the new question wording. Thirdly, the predicted pattern of congruence (and other constituency-based relationships) matches very closely the party by roll-call series that does not make use of the survey data (cf. Figures 1 and 2 above).

Pomper's analysis has also been subjected to criticism. Margolis, , ‘From Confusion to Confusion’Google Scholar, reanalysed Pomper's data and included in the base from which percentages were calculated those respondents with no opinion (so that he had a fair characterization of the entire electorate). However, even Margolis's data show 1964 to be a peak partisan year on domestic welfare issues, with a fall-off analogous to that in Pomper's data. Piereson's critique in ‘Issue Alignment’ does not apply to this paper because he examines the data for partisan issue alignment across policy dimensions, where I have analysed change within the domestic welfare domain.

²⁰ See Cover, Albert D., ‘One Good Term Deserves Another: The Advantage of Incumbency in Congressional Elections’, American Journal of Political Science, XXI (1977), 523–42CrossRefGoogle Scholar; Ferejohn, John A., ‘On the Decline of Competition in Congressional Elections’, American Political Science Review, LXXI (1977), 166–76CrossRefGoogle Scholar; Mann, Thomas E. and Wolfinger, Raymond, ‘Candidates and Parties in Congressional Elections’, American Political Science Review, LXXIV (1980), 617–32CrossRefGoogle Scholar; Hinckley, Barbara, Congressional Elections (Washington, DC: Congressional Quarterly Press, 1981).Google Scholar