The 2000 presidential election in historical perspective

Political Science Quarterly

Volumn 116, Issue 1, 2001, Pages 29-52

  Erikson, Robert S ^a  

^a NONE

Author keywords

[No Author keywords available]

Indexed keywords

EID: 0042406523     PISSN: 00323195     EISSN: None     Source Type: Journal    
DOI: 10.2307/2657819     Document Type: Article

References (30)

1. To understand how close the election in Florida was, consider the likely outcome if the approximately six million voters all were flipping coins. Over multiple trials of six-million coin flips, the median lead for heads or tails would be about 600 votes.
2. For 1989, 1993, and 1997, the late Everett Carll Ladd wrote these election postmortems. I am grateful for being asked to try to follow in his footsteps. See Everett Carll Ladd, "The 1988 Election: Continuity of the Post-New Deal System," Political Science Quarterly 104 (Spring 1989); 1-18: Everett Carll Ladd, "The 1992 Vote for President Clinton: Another Brittle Mandate?" 108 (Spring 1993): 1-28; Everett Carll Ladd, "1996 Vote: The 'No Majority' Realignment Continues," Political Science Quarterly 112 (Spring 1997): 1-28.
3. For 1989, 1993, and 1997, the late Everett Carll Ladd wrote these election postmortems. I am grateful for being asked to try to follow in his footsteps. See Everett Carll Ladd, "The 1988 Election: Continuity of the Post-New Deal System," Political Science Quarterly 104 (Spring 1989); 1-18: Everett Carll Ladd, "The 1992 Vote for President Clinton: Another Brittle Mandate?" 108 (Spring 1993): 1-28; Everett Carll Ladd, "1996 Vote: The 'No Majority' Realignment Continues," Political Science Quarterly 112 (Spring 1997): 1-28.
4. For 1989, 1993, and 1997, the late Everett Carll Ladd wrote these election postmortems. I am grateful for being asked to try to follow in his footsteps. See Everett Carll Ladd, "The 1988 Election: Continuity of the Post-New Deal System," Political Science Quarterly 104 (Spring 1989); 1-18: Everett Carll Ladd, "The 1992 Vote for President Clinton: Another Brittle Mandate?" 108 (Spring 1993): 1-28; Everett Carll Ladd, "1996 Vote: The 'No Majority' Realignment Continues," Political Science Quarterly 112 (Spring 1997): 1-28.
5. This article goes to press in March 2001, only four months after the election and three months after the final determination of its outcome. Thus it runs the risk of being premature. Future analysts will have information about the election and its voters not yet available in the election's immediate aftermath. This will include the rich survey data of the 2000 National Election Study, which is normally disseminated about six months after the election. And the exit polls will reveal more of their riches after their individual-level data are released a year after the election. The exit poll information available on the web immediately after the election consists of reported bivariate relationships between certain predictors (social characteristics, attitudes, policy preferences) and the vote. Valuable though this information is, it allows for no multivariate analysis. At their most rigorous, voting studies examine the effects of several predictors on the vote while controlling for each other. This is possible only with individual-level data.
6. We could add 1964 and 1972 to this list of landslide reelections under economic prosperity. Those elections are more remembered for the ideological "extremism" of losers Barry Goldwater (1964) and George McGovern (1972) than for being conducted under the aura of prosperity.
7. See, for instance, the consensus of very accurate forecasts of the 1996 presidential election in James Campbell and James Garand, eds., Before the Vote: Forecasting American National Elections (Thousand Hills, CA: Sage, 2000).
8. Milton Lodge, Marco Steenbergen, and Shawn Brau, "The Responsive Voter: Campaign Information and the Dynamics of Candidate Evaluation," American Political Science Review 89 (May 1995): 309-26; Daron R. Shaw, "A Study of Campaign Even Effects from 1952 to 1992," Journal of Politics 61 (May 1999): 387-422.
9. Milton Lodge, Marco Steenbergen, and Shawn Brau, "The Responsive Voter: Campaign Information and the Dynamics of Candidate Evaluation," American Political Science Review 89 (May 1995): 309-26; Daron R. Shaw, "A Study of Campaign Even Effects from 1952 to 1992," Journal of Politics 61 (May 1999): 387-422.
10. Andrew Gelman and Gary King, "Why Are American Presidential Election Polls so Variable When Votes Are so Predictable?" British Journal of Political Science 23 (October 1993): 409-519; Thomas Holbrook, Do Campaigns Matter? (Thousand Oaks, CA: Sage, 1996); James E. Campbell, The American Campaign: U.S. Presidential Campaigns and the National Vote (College Station: Texas A&M Press, 2000).
11. Andrew Gelman and Gary King, "Why Are American Presidential Election Polls so Variable When Votes Are so Predictable?" British Journal of Political Science 23 (October 1993): 409-519; Thomas Holbrook, Do Campaigns Matter? (Thousand Oaks, CA: Sage, 1996); James E. Campbell, The American Campaign: U.S. Presidential Campaigns and the National Vote (College Station: Texas A&M Press, 2000).
12. Andrew Gelman and Gary King, "Why Are American Presidential Election Polls so Variable When Votes Are so Predictable?" British Journal of Political Science 23 (October 1993): 409-519; Thomas Holbrook, Do Campaigns Matter? (Thousand Oaks, CA: Sage, 1996); James E. Campbell, The American Campaign: U.S. Presidential Campaigns and the National Vote (College Station: Texas A&M Press, 2000).
13. This forecasting equations is IPV = 33.49 + 7.56PCIG + 0.30PA (3.05) (2.71) ((0.07) where IPV = the incumbent presidential party's percent of the two-party vote, PCIG = quarterly per capita income growth through quarter 15 of the administration, weighted so each quarter counts 1.25 times the previous one, and PA = the president's Gallup approval rating in quarter 15. Standard errors are in parentheses. Adjusted R squared =.82; N = 12.
14. This graph is created from the polls reported at the website pollingreport.com. Each national presidential poll was coded by the middle date of its reported polling period. Where a poll was reported as conducted over an even number of dates, the later of the two middle dates was used. When, as often was the case, multiple polls were centered on a particular date, the date's observation represents the average - a poll of the polls. Where a particular poll series represented a daily tracking poll as a moving average over two, three, or four days, the daily samples were discounted by the number of polling dates. For example, the three-day Gallup tracking polls would each be counted, but each weighted by one-third.
15. See W. S. Cleveland, "Robust Locally Weighted Regression and Smoothing Scatterplots," Journal of American Statistical Association 74 (December 1979): 829-36. For political science discussions, see William G. Jacoby, Statistical Graphics for Univariate and Bivariate Data (Thousand Oaks, CA: Sage, 1997); and Nathaniel Beck and Simon Jackman, "Beyond Linearity by Default: Generalized Additive Models," American Journal of Political Science 42 (April 1998): 596-627. For an application of LOWESS to the 1996 campaign, see Robert S. Erikson and Christopher Wlezien, "Presidential Polls as a Time Series: The Case of 1996," Public Opinion Quarterly 63 (Summer 1999): 163-178.
16. See W. S. Cleveland, "Robust Locally Weighted Regression and Smoothing Scatterplots," Journal of American Statistical Association 74 (December 1979): 829-36. For political science discussions, see William G. Jacoby, Statistical Graphics for Univariate and Bivariate Data (Thousand Oaks, CA: Sage, 1997); and Nathaniel Beck and Simon Jackman, "Beyond Linearity by Default: Generalized Additive Models," American Journal of Political Science 42 (April 1998): 596-627. For an application of LOWESS to the 1996 campaign, see Robert S. Erikson and Christopher Wlezien, "Presidential Polls as a Time Series: The Case of 1996," Public Opinion Quarterly 63 (Summer 1999): 163-178.
17. See W. S. Cleveland, "Robust Locally Weighted Regression and Smoothing Scatterplots," Journal of American Statistical Association 74 (December 1979): 829-36. For political science discussions, see William G. Jacoby, Statistical Graphics for Univariate and Bivariate Data (Thousand Oaks, CA: Sage, 1997); and Nathaniel Beck and Simon Jackman, "Beyond Linearity by Default: Generalized Additive Models," American Journal of Political Science 42 (April 1998): 596-627. For an application of LOWESS to the 1996 campaign, see Robert S. Erikson and Christopher Wlezien, "Presidential Polls as a Time Series: The Case of 1996," Public Opinion Quarterly 63 (Summer 1999): 163-178.
18. See W. S. Cleveland, "Robust Locally Weighted Regression and Smoothing Scatterplots," Journal of American Statistical Association 74 (December 1979): 829-36. For political science discussions, see William G. Jacoby, Statistical Graphics for Univariate and Bivariate Data (Thousand Oaks, CA: Sage, 1997); and Nathaniel Beck and Simon Jackman, "Beyond Linearity by Default: Generalized Additive Models," American Journal of Political Science 42 (April 1998): 596-627. For an application of LOWESS to the 1996 campaign, see Robert S. Erikson and Christopher Wlezien, "Presidential Polls as a Time Series: The Case of 1996," Public Opinion Quarterly 63 (Summer 1999): 163-178.
19. Erikson and Wlezien, "Temporal Horizons."
20. The number 200 is approximate, with tracking polls weighted downward to reflect their rolling averages.
21. The statistical technique employed is pooled cross-section time-series analysis using Least Squares Dummy Variable analysis (LSDV), pooling all surveys-national or state-conducted in September and October. Each date is a dummy variable. Similarly, each state is a dummy variable, with the base category being national polls. With the final date in October serving as the base category, the date coefficients represent differences from the horse-race on 31 October. Each state coefficient represents the difference between the particular state and the national polls. The key assumption underlying this use of state data to analyze national trends is that the movement of electoral preferences is uniform across states. For instance, if a candidate gains two points nationally, the assumption is that he gains the same two points in every state, or at least that national gains on average are reflected uniformly across states.
22. November data is not included for this figure.
23. The Electoral College has two sources of bias that cancel out. Small states gain extra representation from the fact that each state starts with two votes for each senator, and small states tend to vote Republican. Large states gain from their winner-take-all feature, and large states tend to vote Democratic. If the two votes per senator were taken away, the partisan advantage would go to the Democrats. If electoral votes were assigned proportionately rather than at-large, the advantage would go to the Republicans.
24. Angus Campbell, Philip E. Converse, Warren E. Miller, and Donald E. Stokes, The American Voter (New York: Wiley, 1960).
25. For two recent accounts, see Larry Bartels, "Partisanship and Voting Behavior, 1952-1996," American Journal of Political Science 44 (January 2000): 35-50; and John H. Aldrich and Richard G. Niemi, "The Sixth Party System: Electoral Change 1952-1992" in Stephen C. Craig, ed., Broken Contract? Changing Relationships between Americans and their Government (Boulder, CO: Westview, 1996).
26. For two recent accounts, see Larry Bartels, "Partisanship and Voting Behavior, 1952-1996," American Journal of Political Science 44 (January 2000): 35-50; and John H. Aldrich and Richard G. Niemi, "The Sixth Party System: Electoral Change 1952-1992" in Stephen C. Craig, ed., Broken Contract? Changing Relationships between Americans and their Government (Boulder, CO: Westview, 1996).
27. The gap sizes in parentheses are presented as variable, because the gaps in the Gore vote are not precisely the same as the gaps in the Bush vote.
28. See Gerald C. Wright, John P. Mclver, Robert S. Erikson, and David B. Hoolian, "Stability and Change in State Electorates, Carter through Clinton" (paper delivered at the meeting of the Midwest Political Science Association, Chicago, April, 2000).
29. The exact OLS equation predicting the states' 2000 vote from exit poll partisanship and ideology is: Percent Gore = 0.83 + 0.57 (percent Democrats) + 0.43 (percent Liberals) where Percent Gore = Gore percent of the two-party vote in the state, percent Democrat = the Democratic percent of combined Republican and Democratic identifiers and percent Liberal = the liberal percent of combined liberal and conservative identifiers.
30. For one version of balance theory, see Alberto Alesina and Howard Rosenthal, Partisan Politics, Divided Government, and the Economy (Cambridge, UK: Cambridge University Press, 1995). Balance theory competes with the traditional thesis that withdrawn coattails contribute to the presidential party's midterm loss. There were no coattails in 2000 to withdraw in 2002. * I am greatly indebted to several coauthors on different projects. Without these joint intellectual enterprises, the current article could not have been written. The discussions of campaign dynamics and statistical models of elections borrow extensively from my collaboration with Christopher Wlezien. The discussions of the Electoral College and state-level campaign polls borrow from my collaboration with Karl Sigman. The discussion of rival statistical models borrows from joint work with Michael B. MacKuen and James A. Stimson. The discussion of state-level partisanship and ideology borrow from my collaboration with Gerald C. Wright, John P. McIver, and David Hoolian.