Institutional Maintenance and Construction Guidelines Division of
Transcript of Institutional Maintenance and Construction Guidelines Division of
THE CONVENTION BUMP
James E. CampbellLynna L. CherryKenneth A. Wink
This paper appeared in the American Politics Quarterly, v.20, n.3 (July 1992) pp.287-307. An update ofthe analysis was presented in Campbell’s The American Campaign (Texas A&M University Press, 2000)pp.145-51. Following the original article’s tables, Table 1 of the paper is updated through the 2000election.
ABSTRACT
THE CONVENTION BUMP
Do the national conventions of the Democratic and Republican parties affect the poll standings of
the presidential candidates they nominate? This study investigates whether the poll standings of
presidential candidates are bumped upwards following their parties' conventions. The convention bump is
examined with Gallup and Harris time series data of presidential trial-heats throughout the course of the
seven campaigns from 1964 to 1988. We find: (1.) that with few exceptions, there is a convention bump,
(2.) that the effects of convention bumps do not quickly disappear during the campaign, (3.) that the first
convention in the campaign sequence, the convention held by the out-party, generates a larger convention
bump, (4.) that convention bumps are greatest in harmonious conventions following a contentious
nomination struggle and (5.) that convention bumps have not become smaller since the reform of the
nomination process in the 1970s.
THE CONVENTION BUMP
Given the '76 experiences it was clear that Reagan should and would get a very big boostin July at his convention.... We knew that once we had our convention there was going tobe a bounce back for the incumbent [Carter], as there had been in '76.... It was predictable.
-- Patrick J. Caddell1
Over the years, the national party conventions have lost many of whatever deliberative functions
they once had. Although they still write platforms, officially bestow the party nomination on a previously
determined nominee, and provide an audience for the announcement of the vice-presidential nominee,
they have not served as a forum in which the presidential nomination is actually decided for some time
(see, Carleton, 1964; Parris, 1972, and Marshall, 1981). The last major party national convention to have
gone to a second presidential nomination ballot was the 1952 Democratic convention which took three
ballots to nominate Adlai Stevenson (Congressional Quarterly, 1985). Even in the close nomination
fights of Republicans in 1976 and the Democrats in 1980, battles over delegate votes were decided before
the delegates ever assembled at their parties' conventions. Yet while conventions no longer provide an
assembly in which actively decides the parties' nominees, they continue to serve several functions for the
parties.
Perhaps the most important remaining function of party conventions is what David, Goldman and
Bain (1964: 339) refer to as the "rally function." Conventions mark an important transition in campaigns
and set the tone for the parties' fall campaigns (see, Kessel, 1988). As Crotty and Jackson more recently
put it:
The party is well positioned for the race if the convention has been successful in creatingenthusiasm for the candidate and in creating or ratifying a consensus; if the party hasadopted positions that promise to be attractive to the voters; and if the party hassuccessfully avoided alienating its activists and voters. If problems remain evident afterthe convention, or if the problems are actually exacerbated by the events of the conventionitself, then the nominee and his party are likely to be in trouble in November (Crotty andJackson, 1985: 206).
The impact of the national convention as the campaign kick-off or rally appears in the "trial-heat" polls
now commonly conducted throughout the course of the campaign. Pollsters and political commentators
have observed what has become known as "the convention bump" in these polls (see, Breglio and
Harrison 1989; Caddell and Wirthlin, 1981; Hart and Wirthlin, 1985 and Cook, 1988). The conventional
wisdom is that a nominee's poll standing improves a bit, is bumped upwards, following his party's
convention. This paper systematically assesses the "convention bump."
REASONS FOR THE CONVENTION BUMP
There are several possible reasons why candidates might benefit directly from their conventions.
(1.) The convention may help to heal internal party divisions. Supporters of candidates who did not win
the party's nomination may feel uncomfortable immediately casting their support with the nominated
candidate that they had opposed just a short-time ago. They may resent that candidate. Many may
initially indicate indecision about their general election vote decision. For a time, some may even
indicate a decision to vote for the opposition party. While some of the wounds of internal party battles
may heal with time alone, the convention may speed along the process. Conventions allow factional
leaders to come together in a show of unity, sending the message that differences with the opposing party
outweigh any differences remaining within. As a result, though some disgruntled and disappointed
partisans may sit out the election or even bolt to the opposition, the convention encourages many who
might have contemplated these options to return to the fold.2
(2.) Related to their possible healing effects, national conventions may also give an extra push to
their nominee's bandwagon.3 The official investing of a candidate as the party's standard-bearer may draw
less attentive voters to declare their support. Once nominated, a candidate may also gain greater respect
from the more wary partisans who had held-off committing to any candidate.
(3.) The convention bump may also reflect the generally favorable publicity for the party
generated by its convention. Certainly conventions focus a good deal of media attention on the party.
Moreover, most of this attention is likely to be favorable to the party. Convention speakers and the
usually warm to enthusiastic receptions they receive from the delegates creates positive images of the
party. As Richard Wirthlin, Reagan's campaign advisor, noted following the 1980 campaign: "We viewed
the convention as the single best opportunity to present, almost unencumbered, our candidate to a very
wide voter group (Caddell and Wirthlin, 1981: 4)."
Although voters may react differently to the content of the party's message, the message is usually
unrebutted and the atmosphere surrounding its delivery is almost always favorable. With the notorious
exception of the 1968 Democratic National Convention in Chicago, parties generally have control over
much of the information reported and use this control to their advantage. Many convention activities are
now intentionally orchestrated for "external consumption" to place the party and its nominee in the most
attractive light. With the withering of the deliberative functions of the convention, they have become
even more carefully choreographed for their public relations value.4 In this light, we should not be
3
surprised that undecideds or those with weakly held preferences are swung by the convention in favor of
the party's nominee.
PREVIOUS RESEARCH
Two recent studies find evidence of a convention bump by comparing polls before and after
conventions. Shafer (1988: 232-235) examined Gallup trial-heat polls before and after each of the
conventions in six campaigns from 1964 to 1984. He found fairly consistent evidence of the bump in
these campaigns. The nominee's standing in the polls improved after his convention in all but two cases.
The improvement was typically on the order of about 5 percentage points. Two candidates however,
McGovern in 1972 and Johnson in 1964, registered no gains following their nominating conventions. In
Johnson's case, his lead over Goldwater may have been so large before the convention that there was
little room to add to his lead (Shafer, 1988: 233). The results of Wayne's analysis of the 1976, 1980 and
1984 conventions, also based on Gallup results, indicated convention bumps of from 4 to 10 percentage
points in a candidate's standing (Wayne, 1988: 140).
In addition to the Shafer and Wayne studies, there is also circumstantial evidence of a convention
bump. First, a sizeable number of voters report making up their minds about their vote choice by the end
of the parties' conventions. While better than a third of the public normally report that they reached their
decision prior to the conventions, another quarter of the electorate say that they decided how to vote at
the time of the conventions (Davis, 1983: 196). Second, presidential election forecasting models have
been found to be much more accurate when based on polling data immediately following the conventions
rather than prior to the conventions (Lewis-Beck, 1985; Campbell and Wink, 1990). Presumably
something happens to public opinion between these pre-convention and post-convention polls that makes
the post-convention polls substantially more reliable predictors of the November vote.
THE CONVENTION BUMP QUESTIONS
There are several questions to be addressed about the bump. The first concerns the regularity and
magnitude of the bump. Is there solid evidence of a systematic convention bump? Do presidential
candidates regularly receive a boost in their share of supporters following their conventions? How large
is this boost typically and does it reflect the effects of the political convention or would it occur even in
the absence of a convention? Secondly, is the convention bump of any lasting consequence in the
campaign? Is it merely a temporary reaction to the convention that is soon forgotten or does it leave
4
voters with impressions that stay with them through the course of the campaign? Thirdly, what affects the
magnitude of the convention bump? Presumably, not all conventions are equally beneficial to
candidates. Some conventions are more unifying events than others. Certainly the Democrats might have
expected a bigger boost from relatively united conventions like their 1964 Atlantic City convention or
their 1988 Atlanta convention than from their 1968 debacle or their 1948 convention in which both the
left and right wings of the party bolted to run their own candidates.
DATA
The data to examine the convention bump are drawn from the results of trial-heat polls conducted
by the Gallup Poll and the Harris organizations in the seven elections from 1964 to 1988. These trial-heat
questions ask respondents during the campaign which of the presidential candidates they would vote for
if the convention were held at the time of the poll. From these polls we constructed the dependent
variable of this analysis: the Democratic presidential candidate's share of major-party supporters.
There were several obstacles to the analysis that we should note. Prior to 1964 polls were asked
too infrequently to be of use in evaluating convention effects. Moreover, prior to 1964 it was especially
rare to find polls conducted in the three to five weeks between the two national conventions. Between-
convention polls as well as pre and post-convention polls are necessary in order to distinguish the effects
of one convention from the other.5
Even after limiting the analysis to the seven campaigns from 1964 to 1988, not all the polls in
these campaigns are appropriate for examining convention effects. Three criteria are used to ensure that
polls are appropriate for our purposes. (1) Because of the notorious volatility of early polls, we decided
not to include any poll conducted prior to March of the election year in this analysis. (2) Polls conducted
within a few days of the official convening of the convention or actually during the convention are
excluded since it is unclear whether they may have been influenced by the conventions. These polls
occupy a "no-man's land," neither clearly pre nor post convention. Since the inclusion of these polls as
either pre or post might have clouded estimates of convention effects, we decided simply to exclude
them.6 (3) Finally, some polls are excluded because the candidate options offered in the trial-heat
question differed from those generally offered in that particular campaign. In most campaigns, polls
explicitly offering a third-party candidate option were excluded. However, in the 1968 and 1980
campaigns, third-party trial-heat options were the rule rather than the exception. In 1968 most post-
5
March polls included George Wallace as an option for respondents and in 1980 most post-March polls
included John Anderson as an option. In these two elections, for the sake of consistency and
comparability, polls are excluded if they offered only the two major party candidates as the choice.
The application of these criteria leaves between 15 and 40 polls available in each of the seven
presidential campaigns. In each of these trial-heat series, there are at least three polls prior to the first
convention, at least one poll (and usually three to six polls) between the two conventions and at least 8
and as many as 23 polls following the second convention.
METHODS
The effects of party conventions are examined in two ways. Like the previous studies of Shafer
and Wayne, we examine the change in trial-heat poll standings from before to after each party's
respective convention. Polls conducted between 14 and 6 days prior to a convention are compared to
polls in the week following the convention or the first available post-convention poll.
Beyond examining simple differences between pre-convention and post-convention polls, we
conduct a time series regression analysis on each campaign's trial-heat polls. This places the bump within
the context of the full campaign and allows us to distinguish convention effects from more general trends
of a campaign.7 The series includes a time variable to control for any long-term trend over the campaign
and the lagged poll results to take autocorrelation into account.8 The regressions are conducted with four
different measures of the convention interventions -- a simple pulse taking a value of 1 immediately after
the convention and zero beyond that, a coding specifying an effect decaying after 30 days, a coding
specifying an effect decaying after 60 days, and one specifying no decay in the convention's effects.9
The analysis is conducted for each year separately and also after pooling the data across the seven
elections.
FINDINGS
The simple computation of the convention bump confirms the earlier findings of Shafer (1988)
and Wayne (1988). Almost invariably, presidential candidates get a boost in the polls with their parties'
national conventions. The change in the trial-heat standing of presidential nominees between the pre-
convention and post-convention polls are presented in Table 1. As these differences indicate, a
presidential candidate can expect typically to receive a boost in the polls of about six percentage points.
6
However, it is not at all unusual for the convention bump to be in excess of ten percentage points.
Candidates received a double-digit boosts following four of the fourteen conventions held since 1964.
/Table 1 about here/
As the previous studies found, among recent elections, there are only two instances in which
candidates failed to improve their standing after their party's convention. In 1964, Lyndon Johnson's
standing was unchanged after the Democratic convention. As Shafer suggested, the Johnson exception
may be due to a ceiling effect on the support for a then very popular and well known incumbent
president. The second exception is McGovern. McGovern actually dropped two points in the polls
following his 1972 Democratic nomination. The 1972 Democratic convention was apparently unusually
ineffective in pulling disaffected partisans back into the Democratic fold. That convention, the first
Democratic convention conducted under the McGovern-Fraser Commission reforms, was highly
controversial, contentious and disorganized.10 If that were not enough, it was capped off by the
controversial nomination of McGovern's initial running-mate, Senator Thomas Eagleton.11 Whether a
result of organizational difficulties, the Eagleton fiasco, or the apparently unpopular turn to the left by
McGovern Democrats, the usual benefits of the convention bump simply did not materialize for the
Democrats in 1972.
The results from the time series analyses are generally in accord with those of the simple before
and after convention differences. The results of the regressions with the Democratic presidential
candidate's share of the two-party trial-heats as the dependent variable are presented in Table 2 for each
of the seven campaigns. Table 3 presents the regressions on the entire pooled series.
/Tables 2 and 3 about here/
As indicated earlier, four different specifications of convention effects are examined for each
party in each election. Each differs how rapidly or whether convention effects decay in the weeks
following the respective conventions. Given that the dependent variable is the extent of support for the
Democratic candidate, we should expect positive coefficients for the Democratic convention variables
and negative coefficients for the Republican convention variables. In most cases these expectations are
met. They are met in each instance of the pooled analysis in Table 3. Regardless of how the deterioration
of the bump was specified, its estimated effect was in the expected direction and statistically significant.
Given that convention effects survived the inclusion of the trend variable, it would appear that the
7
analysis not only confirms the simple difference findings but also suggests that these effects are not
simply extensions of prior campaign trends.
Of the 56 estimated individual campaign bump coefficients in Table 2, 45 (80%) are in the
expected direction. Moreover, most deviations from expected effects involved the same conventions in
which the simple difference analysis also failed to find a positive bump: the 1964 and 1972 Democratic
conventions. Beyond these two exceptions, the time series also suggests, in most of its variations, that
there was no bump following the raucous 1968 Democratic convention.
As to the various specifications of the bump, an examination of the four different time series
estimates does not unambiguously indicate a single preferable specification of convention effects.
Judging by either the stability or the magnitude of these estimates, there is no single consistently
strongest specification of the decay of convention effects for all conventions. While inconclusive as to
which specification of the bump is most appropriate, the analysis does suggest that the effects of
conventions on the candidates' poll standings are not strictly temporary. The specification of an
immediate decay in convention effects in most cases was not the strongest convention variable.12
Generally speaking, conventions appear to be of some lasting consequence to the candidates' standing
with the public. While some convention effects appear fleeting, most have staying power and, as such,
may ultimately affect the election results.
VARIATION IN CONVENTION BUMPS
While the analysis indicates the existence of a convention bump, it also suggests that there is
considerable variation in its magnitude. There are bumps and then there are bumps. Although candidates
generally operate in the gain column following their conventions, a few do not, a few receive more of a
blip than a bump and others receive a rather considerable boost. While there are undoubtedly many
idiosyncracies in any convention (eg. the conflict on the streets of Chicago during the 1968 Democratic
convention) and while our set of conventions is small in number, we consider three possible systematic
sources of variability in the convention bump: the order of the convention in the campaign, the nature of
conflict in the nomination contest leading up to the convention, and the broad intra-party democratizing
reforms of the national party and nominating process in the early 1970s.13
(1.) The convention bump may vary according to the sequence of the convention. First
conventions in a campaign differ in two potentially important respects from second conventions. Most
8
obviously, they reach voters earlier in the decision process when voters might be more open to influence.
Also, since the 1936 campaign, the out-party has held its convention before the incumbent's party.
(2.) The convention bump may also vary according to the level of internal party conflict. We
might expect a larger bump for parties that had a divisive nomination contest but a harmonious, healing
convention. In such a case, many partisans may be disgruntled during the nomination battle but may also
be won back to support the party's nominee. Under these circumstances, there is the potential for
relatively large convention gains.
(3.) Given the substantial reform of the nomination process since the 1968 campaign, we might
expect a difference in the convention bump between the pre-reform and post-reform eras. Party reforms
have been blamed for weakening parties as effective vote-getting institutions (see, Kirkpatrick, 1978 and
Ceaser, 1982). If true, we might expect to see weaker convention bumps in post-reform campaigns.
Convention Sequence. Although we have a very limited number of campaigns in which to discern
systematic differences, the evidence is fairly persuasive that a convention's order in a campaign matters.
The bump from the first convention of a campaign is generally greater than from the second convention.
The presidential candidate of the out-party, traditionally holding the earlier party convention, obtains a
bigger boost from his national convention than the presidential candidate of the incumbent party
nominated at the campaign's second convention. As measured by simple before and after differences, first
conventions typically provide their candidates about twice the boost in the polls of the second
convention. The time series analysis also confirms stronger first convention effects (first conventions are
set in boxes in Table 2). The single exception to the pattern of stronger first convention effects is 1972.
As has already been noted, there are a variety of reasons why McGovern's 1972 Democratic presidential
nomination convention is an exception to the rule. Finally, with the sole exception of the "no decay"
Democratic bump specification, the interaction terms of the first convention sequence variable (1 if first
and 0 if second) and the convention effect variable in Table 3 (equations 1b, 2b and 3b) are consistently
strong, of the expected sign and statistically significant.
There are three plausible reasons for the bigger first convention bump. (1.) The first convention
may reach more voters while they are making up their minds. By the time of the second convention, more
voters may have firmed up their vote decision. (2.) Some of the difference may be artificial. Following a
lengthy nomination battle, it may just take some time for the wounds of disgruntled partisans to heal. A
9
good deal of this healing in both parties may take place during the first convention and before the second
convention. If true, much of the natural return to the party fold with time might be attributed to the first
convention. (3.) Perhaps the most plausible reason for the bigger first bump is that the out-party holds
the first convention. Generally speaking, voters are less familiar with the out-party's presidential
candidate and the convention can relieve uncertainties about supporting that candidate. Voter
impressions about the in-party's candidate may be more fully formed through evaluations of that party's
recent performance in office, especially if the incumbent is seeking reelection. It is less likely that a
convention would affect the decisions of voters who had fairly definite ideas about the incumbent party
before the convention.
Intra-Party Conflict and Conventions. A good portion of the convention bump is presumably
based on the convention's healing of internal party divisions. Two conditions are implied in the healing
of these divisions: that there have been significant internal party conflicts that require healing and that
the convention serves as a forum for reconciliation. These two conditions suggest three types of
conventions: (1.) those in which neither the nomination nor the convention were conflictual, (2.) those in
which both the nomination and convention were conflictual and (3.) those in which the nomination was a
matter of serious conflict ending prior to the convention. We expected a greater bump following this
third situation, what might be called "healing conventions". While the assignment of conventions to the
three categories is admittedly impressionistic and the differences not startling, the convention bump was
typically greater following these healing conventions. The average bump from a healing convention was
about 8 percentage points compared to about a 5 percent bump in other convention circumstances.14
Post-Reform Conventions. Following their 1968 convention debacle, the Democratic party
reformed the presidential nomination process in a variety of ways. While not all of the post-1968
reforms pursued the same goals of internal party democracy, the general effect (intended or not) of the
reforms was to open up the nomination process by increasing its reliance on presidential primaries and by
awarding delegates on a more proportional basis.15 Many argued that this was harmful to the party's
major responsibility of attracting support for its candidates. While their are a wide variety of ways in
which the electoral consequences of party reform might be exhibited, one potential way is in a decline of
the convention bump. If the reformed process produces dissension rather than unity in the party and
10
leads to the nomination of candidates supported by ideological activists but distasteful to the rank-and-
file party voter, we might see smaller convention bumps in the post-1968 conventions.
The estimated convention effects do not indicate that party reforms have reduced convention
bumps at all. While we have only four cases of pre-reform conventions, the available evidence does not
suggest smaller convention bumps in the post-reform era. There may be several reasons for this. One
explanation may be that with the reforms has come more pre-convention party conflict to settle at the
convention. With the opening of the nomination process there have been more candidates for the
nomination, dividing partisans into more factions. The effects of recent conventions may look strong
because they are pulling together more splintered parties.16 Also, while the reformed process has opened
the party to more internal conflict over the nomination, it is also widely noted to be an increasingly
"front-end loaded" system. The importance of the early Iowa caucus, New Hampshire primary and the
predominantly Southern "Super Tuesday" primaries may settle party conflict earlier and provide time to
prepare the way for reconciliation both before and at the convention (see, Marshall, 1981). With the
nomination now usually settled before the convention, the party can give greater emphasis to using the
convention for general election campaign purposes than airing internal party disputes.17
SUMMARY
There are five principal findings regarding the convention bump. First, it exists. In most cases,
conventions continue to fulfill their "rally function" for the political parties and their presidential
candidates. Presidential candidates typically increase their poll standings following their party's
convention. The analysis of the simple difference of pre-convention and post-convention polls as well as
the time series analyses generally confirms the earlier analyses of Shafer (1988) and Wayne (1988). Like
these previous studies, we found the Democratic conventions of 1964 and 1968 to be exceptions to the
usual convention gains. Unlike the analysis of simple differences, however, the time series analysis also
indicated that the 1968 Democratic convention was also an exception. These exceptions suggest that
while candidates are generally strengthened by their conventions, these gains cannot be taken for granted.
They are by no means automatic. Conventions present a unique opportunity to reunite the party and
attract uncommitted voters. It is up to the party to exploit this opportunity.
Second, the time series analysis suggests that the effects of conventions are not strictly short-lived.
The convention bump is not merely a convention blip. In most cases, the effects of the convention carry
11
well into the campaign. The way a candidate comes out of his convention is of some real consequence to
the ultimate election outcome.
Third, convention effects are more substantial for first conventions in a campaign. Whether
because the out-party traditionally holds their convention first or because first conventions reach voters
when they are more impressionable, on average, presidential candidates nominated in the first convention
of a campaign receive about twice the boost in the polls that candidates nominated in the second
convention receive.
Fourth, convention effects also are somewhat larger when parties have been divided in the
nomination campaign but manage to hold a conciliatory convention. If their are wounds to heal and the
battle has clearly ended, conventions can be especially effective in reuniting a party for the fall
campaign.
Finally, though there are numerous charges against party reforms for their impact on the effective
performance of parties in elections, there is no evidence that convention bumps are smaller in the post-
reform era. This may be due to the increased initial competition for the party's nomination combined
with the front-end loading of the post-reform system. As noted with respect to party divisiveness,
conventions may have their greatest effects when there are internal divisions that are also resolvable.
12
REFERENCES
Bartels, Larry. 1988. Presidential Primaries and the Dynamics of Public Choice. Princeton, NJ:Princeton University Press.
Breglio, Vincent J. and Irwin "Tubby" Harrison. 1989. "Pollsters on the Polls." Public Opinion,January/February 1989: 4-7 and 50-60.
Caddell, Patrick and Richard Wirthlin. 1981. "A Conversation with the Presidents' Pollsters." PublicOpinion, December/January 1981: 2-12.
Campbell, James E. and Kenneth A. Wink. 1990. "Trial-Heats Forecasts of the Presidential Vote,” American Politics Quarterly, 18: 251-269.
Carleton, William G. 1964. "The Revolution in the Presidential Nominating Convention," in FrankMunger and Douglas Price (ed.) Political Parties and Pressure Groups. New York: Crowell, p.282-295.
Ceaser, James W. 1979. Presidential Selection: Theory and Development. Princeton, NJ: PrincetonUniversity Press.
Ceaser, James W. 1982. Reforming the Reforms. Cambridge, MA: Ballinger.
Congressional Quarterly, 1985. Guide to U.S. Elections, Second Edition. Washington, D.C.,Congressional Quarterly.
Cook, Rhodes. 1988. "Topsy-Turvy Polls: Medium Is the Message," Congressional Quarterly WeeklyReport, September 17, 1988: 2559-2562.
Crespi, Irving. 1988. Pre-Election Polling: Sources of Accuracy and Error. New York: Russell SageFoundation.
Crotty, William and John S. Jackson III. 1985. Presidential Primaries and Nominations. Washington,D.C.: Congressional Quarterly Press.
David, Paul T., Ralph M. Goldman, Richard C. Bain, 1964. The Politics of National Party Conventions,Revised Edition. Washington, D.C.: The Brookings Institution, Reprinted by Peter Smith, Gloucester,Mass., 1972.
Davis, James W. 1983. National Conventions in an Age of Party Reform. Westport, Connecticut:Greenwood Press.
Hart, Peter and Richard Wirthlin. 1985. "Moving Right Along? Campaign '84's Lessons for 1988." PublicOpinion, December/ January 1989: 8-11 and 59-63.
Kenny, Patrick J. and Tom W. Rice. 1988. "Presidential Prenomination Preferences and CandidateEvaluations." American Political Science Review, 82: 1309-1320.
Kessel, John H. 1988. Presidential Campaign Politics: Coalition Strategies and Citizen Response, ThirdEdition. Homewood, IL: Dorsey Press.
Lewis-Beck, Michael S. 1985. "Election Forecasts in 1984: How Accurate Were They?" PS, 18: 53-62.
Marshall, Thomas R. 1981. Presidential Nominations in a Reform Age. New York: Praeger.
Parris, Judith H. 1972. The Convention Problem. Washington, D.C.: The Brookings Institution.
Pomper, Gerald. 1966. Nominating the President: The Politics of Convention Choice. New York: Norton.
13
Ranney, Austin and Willmore Kendall. 1956 Democracy and the American Party System New York:Harcourt, Brace and Company.
Rossiter, Clinton L. 1960. Parties and Politics in America. Ithaca, N.Y.: Cornell.
Sorauf, Frank J. and Paul Allen Beck. 1988. Party Politics in America. Glenview, IL: Scott, Foresmanand Co.
Shafer, Byron E. 1988. Bifurcated Politics: Evolution and Reform in the National Party Convention.Cambridge, MA: Harvard University Press.
Stone, Walter J. 1986. "The Carryover Effect in Presidential Elections." American Political ScienceReview, 80: 271-279.
Sullivan, Denis G., Jeffrey L. Pressman, and F. Christopher Arterton. 1976. Explorations in ConventionDecision Making: The Democratic Party in the 1970s. San Francisco: W.H. Freeman.
Wayne, Stephen J. 1988. The Road to the White House, 3rd Edition. New York: St. Martin's Press.
14
Table 1. Change in the Two-Candidate Trial-Heat Standings ofPresidential Nominees between their Pre-Convention and the Post-Convention Polls, 1964-1988
44444444444444444444444444444444444444444444444444444444444444444 Trial-Heat Poll Standing for Presidential Candidates in Pre and Post-Convention Polls (% of Two-Candidate Responses) ))))))))))))))))))))))))))))))))))))))))))))))))))))))))) Democrat Republican )))))))))))))))))))))))))) )))))))))))))))))))))))))))Year Pre Post Change Pre Post Change ))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))1964 69.1% 69.1% 0.0% 20.8% 33.7% +12.9%*1968(t) 39.2% 41.9% +2.7% 46.8% 60.8% +14.1%*1972 41.3% 39.3% -2.0%* 64.8% 65.5% +.7%1976 59.6% 68.1% +8.6%* 35.2% 42.5% +7.3%1980(t) 36.0% 48.4% +12.4% 54.9% 67.1% +12.2%*1984 44.3% 50.0% +5.7%* 56.0% 57.9% +1.8%1988 53.4% 59.3% +5.9%* 46.5% 52.7% +6.2%))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))Mean Percentage Change:
All Conventions 6.3% First Conventions 8.2%Second Conventions 4.4%Democratic Conventions 4.8%Republican Conventions 7.9%
Percent with Gains 86% (12 of 14)Median Percentage Change 6.1%4444444444444444444444444444444444444444444444444444444444444444444444444Note: The percentages are based on only supporters of the two majorparty candidates. The percentage point gains are the differencesbetween the percentage of respondents indicating a preference for theparty's candidate in the last pre-convention and post-conventionpolls. Pre-convention polls were completed at least 6 days prior tothe convention. In cases in which more than one poll was conducted between 14 and 6 days before the convention, poll results wereaveraged. The post-convention poll was the first post-convention pollor the average of post-convention polls if there were more than onepoll in the week following the convention. The specific dates of thepolls are presented in the Appendix.
(t) indicates that original polls included an explicit option of athird-party candidate.
* indicates the party had the first convention in the campaign.
15
Table 2. GLS Estimates of Democratic and Republican Party Convention Bump Effects, 1964-1988.
444444444444444444444444444444444444444444444444444444444444444444444444444444444444Dependent Variable: Democratic Percentage of Two-Candidate Trial-Heat Support
Equation 1 Equation 2 )))))))))))))))))) ))))))))))))))))))Independent Variables Estimate SE Estimate SE))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))Constant 52.148* .647 51.216* .645 Democratic Convention Bump 3.192* .986 5.211* 1.086Democratic Bump in 1964, 1968 & 1972 -- -- -6.767* 1.785Days Since Democratic Convention -.038 .036 -.096* .038 Republican Convention Bump -7.430* 1.372 -6.698* 1.332Days Since Republican Convention .052 .035 .113* .037))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))Number of Cases 176 176R2 (Adjusted R2) .80 (.78) .82 (.80)Standard Error of Estimate 3.03 2.91Durbin-Watson 1.98 2.12
444444444444444444444444444444444444444444444444444444444444444444444444444444444444* p < .01 (one-tailed). Note:The "bump" variables are scored 1 for polls following the convention and 0 prior to theconvention. "Days since the convention" are the number of days since the last day of therespective convention. The equation also includes a set of dummy variables for theelection years and interactions of the these dummy variables with the number of days beforethe first convention of a year to control for preconvention trends, whenever these trendswere statistically significant in the initial saturated OLS estimate of the model (includedfor 1964, 1968, 1976 and 1980). The coefficient of the first-order autocorrelation used incomputing "generalized differences" in the GLS estimate was .296. The first case in eachyear was dropped in computing the generalized differences (Wonnacott and Wonnacott, 1970:142). Two cases identified by diagnostic statistics as outliers with high "leverage" werealso dropped from the final estimates.
16
Table 3. GLS Estimates of Convention Sequence Effects on Democratic andRepublican Convention Bumps, 1964-1988.444444444444444444444444444444444444444444444444444444444444444444444444444444444444Dependent Variable: Democratic Percentage of Two-Candidate Trial-Heat Support
Equation 1 Equation 2 ))))))))))))))))))) )))))))))))))))))))Independent Variables Estimate SE Estimate SE))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))Constant 50.239** .666 51.501** .622
Democratic Convention Bump 9.598** 2.324 2.260* 1.131Democratic First Convention Bump -7.839 3.785 -- -- Democratic First Convention Blip -- -- 3.339** 1.373Days Since Democratic Convention -.057 .068 -.063* .034Democratic Bump in 1964, 1968 & 1972 -7.759** 1.729 -4.824** 1.517
Republican Convention Bump -5.816** 1.834 -4.870** 1.222Republican First Convention Bump -6.324* 4.005 -- --Republican First Convention Blip -- -- -8.621** 2.218Days Since Republican Convention .059 .069 .075* .033))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))Number of Cases 177 173R2 (Adjusted R2) .86 (.84) .87 (.86)Standard Error of Estimate 2.87 2.76Durbin-Watson 2.09 2.00 444444444444444444444444444444444444444444444444444444444444444444444444444444444444*p < .06, **p < .01 (one-tailed).Note: See Table 2 for descriptions of the "bump" and "days since convention" variables. The "first bump" variables are scored like the "bump" variables, but only when a party'sconvention is held first in a campaign. The first convention "blip" variables are scored 1for polls in the week following a campaign's first convention and 0 otherwise. Theseequations also include election dummy variables and controls for preconvention trends (for1964, 1968, 1976 and 1980). The first-order autocorrelation coefficients used in computinggeneralized differences in the GLS was .238 in Equation 1 and .142 in Equation 2. Casesidentified by diagnostic statistics as outliers with high "leverage" was dropped from thefinal estimates (1 case from Equation 1 and 5 cases from Equation 2).
17
Table 4. GLS Estimates of Nomination Divisiveneses Effectson Democratic and Republican Convention Bumps, 1964-1988.444444444444444444444444444444444444444444444444444444444444444444444444444444444444Dependent Variable: Democratic Percentage of Two-Candidate Trial-Heat Support
Equation 1 Equation 2 ))))))))))))))))))) )))))))))))))))))))Independent Variables Estimate SE Estimate SE))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))Constant 52.959*** .755 52.455*** .763
Democratic Convention Bump 5.913** 2.762 3.223* 2.433Interaction with Nomination Divisiveness -1.711 1.546 .564 1.334Democratic Bump in 1964 -- -- -11.199*** 3.505Days Since Democratic Convention .040 .042 -.072* .044
Republican Convention Bump -5.145** 2.594 -4.385** 2.516Interaction with Nomination Divisiveness -2.066* 1.605 -2.063* 1.452Days Since Republican Convention -.034 .042 .097** .045))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))Number of Cases 175 177R2 (Adjusted R2) .82 (.81) .84 (.83)Standard Error of Estimate 2.97 2.97Durbin-Watson 2.12 2.01 444444444444444444444444444444444444444444444444444444444444444444444444444444444444*p < .10, **p < .05, ***p < .01 (one-tailed).Note: The "nomination divisiveness" index was scored 1 for high conflict conventions, 2 forconventions following less severely divisive nomination battles and 3 for conventionsfollowing settled nominations after a divisive nomination battle (a "healing" convention). The codings of individual conventions are indicated in the text. See Tables 2 and 3 fordescriptions of the other variables. These equations also include election dummy variablesand controls for preconvention trends (for 1964, 1968, 1976 and 1980). The coefficients ofthe first-order autocorrelation used in computing generalized differences in the GLS was.289 in Equation 1 and .247 in Equation 2. Three cases in Equation 1 and one case inEquation 2 were identified by diagnostic statistics as outliers with high "leverage" weredropped from the final estimates of the respective equations.
18
UPDATE of TABLE 1 THROUGH 2000Table 7.3. Change in the Two-Candidate Preference Poll Standings of Presidential Nominees between their Preconventionand Postconvention Polls, 1964-2000
Democratic Party Republican Party_____________________________________ ___________________________________
Year In-Party Pre (%) Post (%) Change (%) Pre (%) Post (%) Change (%)1964 Democrats 69.1 69.1 0.0 20.8 33.7 + 12.91968 Democrats 39.2 41.9 + 2.7 46.8 60.8 + 14.11972 Republicans 41.3 39.3 – 2.0 64.8 65.5 + 0.71976 Republicans 59.6 68.1 + 8.6 35.2 42.5 + 7.31980 Democrats 36.0 48.4 + 12.4 54.9 67.1 + 12.21984 Republicans 44.3 50.0 + 5.7 56.0 57.9 + 1.81988 Republicans 53.4 59.3 + 5.9 46.5 52.7 + 6.21992 Republicans 46.5 62.9 + 16.4 40.4 42.4 + 2.01996 Democrats 56.8 61.8 + 5.0 36.6 46.1 + 9.52000 Democrats 41.5 50.5 + 9.1 56.2 59.3 + 3.2
Positive Postconvention Poll Changes (Convention Bumps): 90% (18 of 20)Mean Percentage Change:
All Conventions 6.7 Democratic Conventions 6.4 First Conventions (Out-Party) 8.7Republican Conventions 7.0 Second Conventions (In-Party) 4.7
Frontrunning Candidates' Conventions 4.7Trailing Candidates' Conventions 8.0
(with 1968 & 1972 excluded) (9.6)
Estimated portion of net campaign bump surviving to election day: 36%
Note: The out-party traditionally holds its conventions first in the campaign year. Frontrunners are in bold. The Net Campaign Bump isthe difference between the two parties' bumps in the same year. The portion of the net bump surviving to the election is based on aregression of the in-party vote. The vote is explained in terms of the net bump, the pre-convention poll percentage for the in-partycandidate (before either party’s convention), and the first half growth rate in the GDP (with half credit for successor candidates). Theadjusted R2 is .85. The net bump coefficient is .36 (p<.02, one-tailed). The equation includes the ten elections from 1964 to 2000.Comparable polling data is not available to calculate net convention bumps in elections before 1964.
19
1. The epigraph is taken from a Public Opinion's joint interview with the top-level campaignadvisors to Carter's and Reagan's 1980 presidential campaign (Caddell and Wirthlin, 1981: 5). Dukakis' pollster in 1988, "Tubby" Harrison, also acknowledged the effects of the conventionbump in a post-election Public Opinion interview with Ben Wattenberg in the followingexchange with Wattenberg about Dukakis' big lead in the polls following the Democraticconvention:
Wattenberg Q: "So the plus-seventeen was a halo effect from the Democraticconvention?" Harrison A: "Yes, and the campaign so understood it....(Breglio and Harrison, 1989:5).
2. As research by Southwell (1986), Stone (1986), and Kenny and Rice (1988) indicate,convention effects notwithstanding, many partisans who had supported candidates not receivingthe party's nomination refuse to return to the fold for the general election.
3. Bartels (1988) offers an excellent analysis of the effects of bandwagons, campaign momentumor candidate viability in presidential primaries. Although he finds evidence of early bandwagoneffects among primary voters, we might expect later bandwagon effects from less attentive non-primary voters who may yet vote in the general election.
4.Ironically, the greater orchestration of conventions for public consumption may have lead to thedeemphasis of their coverage by the major television networks. All three networks previouslyoffered viewers "gavel-to-gavel" coverage when conventions were more unpredictable andpotentially more conflictual. More recently, however, the networks have reduced the on-aircoverage, most probably as a consequence of stricter convention organization and anunwillingness to let the convention organizers dictate what receives media time.
5. The 1936 Gallup series, for instance, lacked a pre-convention trial-heat poll and, therefore, isnot useful for the purposes here. In 1940 there were no trial-heat polls prior to the Republicanconvention in June but the first trial-heat poll was conducted within a week of the start of theDemocratic convention in July. Since this rules out analysis of the 1940 Republican conventionbump and because the July trial-heat reading was so close to the Democratic convention, the1940 case is excluded from the analysis. The requirement of a between convention poll rules outthe remaining trial-heat series prior to the 1964 campaign.
6. The explicit criteria was to exclude any poll completing its field work within five days of theconvention. Eleven polls were excluded on these grounds. Most of these excluded polls wereconducted either during the convention or within three days of the opening of the convention.
7. The analysis initially examined two versions of the trial-heat variables, one excludedundecided and those indicating a preference for a third-party candidate (the major-party divisionof support) and one including these respondents. The results were quite similar for the twoanalyses.
8. The lagged trial-heat variable was included after examining first-order autocorrelations. Ineach election, autocorrelation was positive and in each election except 1984 it was in excess of.45. First-differencing was examined. However, first-differences of trial-heat ratings leftsubstantial negative autocorrelation, at or close to statistical significance in four of the sevencampaigns. The option of including the lagged trial-heat variable appears to most effectivecontrol of autocorrelation and the Durbin-Watson statistics are reported.
9. As a point of comparison, the conventions since 1964 have taken place anywhere from 68 to113 days (measured from the last day of the convention) before election day. The final day of
NOTES
20
each convention and the number of days before election day are: 1964, R July 16 (110) & D August 27 (68); 1968, R August 8 (89) & D August 29 (68); 1972, D July 13 (113) & R August 23 (76); 1976, D July 15 (110) & R August 19 (75); 1980, R July 17 (110) & D August 14 (82); 1984, D July 19 (110) & R August 23 (75); 1988, D July 21 (110) & R August 18 (82). The 30 and 60 day decay codings were calculated based on the number of days from the end ofthe convention to election day (convention), the number of days from the final day of the poll toelection day (poll) and the specified period of decay (period). The formula used to compute thevariables is: (poll - convention + period)/period. The variable has a maximum of one for a pollconducted at the end of the convention and a minimum of zero for a poll conducted on the lastday of the period, 30 or 60 days after the convention.
10. Several bitterly fought delegate credentials disputes combined with the decidedly non-centristplatform planks, contributed to the poor image many Americans carried away from thisconvention. Probably the most graphic evidence of the convention's disarray is the fact thatMcGovern gave his nomination acceptance speech in the wee-hours of morning, long after thelarge prime-time television audience had gone to bed.
11. Public concern about Eagleton's mental-health history eventually led to his taking himself offthe ticket a little more than two-weeks following the convention. McGovern's image may havealso been damaged at the convention by the widely publicized refusal of Senator EdwardKennedy to accept the vice-presidential nomination.
12. Actually even the "immediate" decay specification of convention effects (1 immediatelyfollowing the convention and 0 thereafter) is not really immediate since some residual effectsshould be felt through the lagged trial-heat variable included in the equation.
13. We also examined two other convention differences: partisan differences and the front-runnerdifferences. (1.) While convention bumps for the Republican party were slightly larger than forthe Democratic party, the differences were not consistent. In three of the seven elections, theDemocrats achieved greater bumps than the Republicans (Table 1). (2.) While Johnson's veryhigh standing in the 1964 pre-convention polls may explain why he did not improve on thatstanding after his convention, we did not find evidence that the front-runner generally receivedless of a bump than his opponent (see Table 1 or 2).
14. The conventions were grouped as follows: not seriously divided nomination -- 1964D,1968R, 1972R and 1984R; conflictual conventions -- 1968D, 1972D, 1976R, 1980D and 1980R;and healing conventions -- 1964R, 1976D, 1984D, 1988D and 1988R. Several of thesecategorizations may be open to question. For instance, the 1968 Republican convention may beclassified as a healing convention since Rockefeller and Reagan challenged Nixon, though Nixoneasily won the nomination on the first ballot (Congressional Quarterly, 1985: 117). However, inthis instance, reclassifying the 1968 convention would only strengthen the bump differencebetween "healing" and other types of conventions.
15. The reforms begun by the Democratic Party's McGovern-Fraser Commission for the 1972campaign were continued by a series of following commissions. Some extended the originaldemocratizing efforts and others attempted to limit democratization (eg. the creation of so-called"super-delegates"). While there was some diversity in these reforms, the general consequenceswere to increase internal-party democracy. The growth in the number of state presidentialprimaries also had the unintended effect of "reforming" the presidential nomination process ofthe Republican party.
21
16. In essence, this argument is that the gains of recent conventions are more of a consequence ofstarting from a lower pre-convention poll position rather than boosting the party's nomination toa higher post-convention standing. Ceaser, however, argues that reforms increase not only theodds of internal party conflict but of that conflict spilling over into the convention itself. Whileseveral post-reform conventions appear to fit this mold (1976 Republican and 1980 Democratic),the front-end loading of the process appears to counteract extended nomination battles.
17. While the hypothesis suggested a negative effect of party reforms, the pre-reform conventionfocus on internal disputes might lead to the contrary hypothesis that convention bumps haveemerged in the post-reform era. For an excellent discussion of convention decision-making inthe pre-reform era see Pomper (1966). Earlier reforms, such as the 1936 elimination of the two-thirds rule in Democratic conventions, may also have had the effect of shifting conventionpurposes away from internal matters and toward preparing the way for the general electioncampaign (see Ranney and Kendall, 1956).