VIEWS: 22 PAGES: 34 POSTED ON: 5/30/2009
The Presidential Primary Vote: What Predictors Matter? Wayne P. Steger Department of Political Science DePaul University 990 W. Fullerton, Suite 2200 Chicago, IL 60614 firstname.lastname@example.org Abstract: This paper assesses the relationships between a number of theoretically significant variables as empirical predictors of the Democratic and Republican primary vote from 1976 through 2004. Previous forecasting models have found that the main predictors of the aggregate presidential vote are candidates standing in pre-primary national polls (Mayer 1996, 2003) and money on hand at the beginning of the primary season (Steger 2000; Adkins and Dowdle 2000, 2001). While these factors have been significant, primary forecasting studies have not looked at the effects of presidential endorsements by party elites or the effects of electability. Numerous studies have argued for a resurgence of the political party establishments since the 1970s, yet no study has assessed the effects of party elites on in forecasting models of the presidential primary vote. Several also have argued that nominations are likely to be influenced by how candidates are expected to fare in the second stage of the election process (Brady and Johnson 1987; Abramson, Aldrich, Paolino and Rhode 1992). This study looks at a variety of factors as predictors of the aggregate presidential primary vote from 1976 through 2004. Paper prepared for the Southern Political Science Association Meeting, New Orleans, LA, January 6-8, 2005 1 This paper assesses the relationships between a number of theoretically significant variables as empirical predictors of the Democratic and Republican primary vote from 1976 through 2004. Existing forecasting models have found that the main predictors of the aggregate presidential vote are candidates standing in pre-primary national polls (Mayer 1996, 2003) and money on hand at the beginning of the primary season (Adkins and Dowdle 2000, 2001; Steger 2000). The fundamental problem is that forecasting models using these variables get it wrong in 2004 (Dowdle, Adkins and Steger 2004b). John Kerry’s unexpected victories in the Iowa caucuses and the New Hampshire primary gave him a powerful surge of momentum while the campaign of Howard Dean, the pre-primary front-runner collapsed. Kerry’s surge is not picked up in national polls conducted before the Iowa caucuses. Like Jimmy Carter in 1976, John Kerry’s national poll figures were in the single digits in early Iowa when he ranked third or fourth in the candidate field. Within a week of the Iowa Caucuses, his national poll figures dwarfed all of his rivals. While Kerry went on to raise more money than any prior Democratic party presidential nominee, Kerry trailed Howard Dean by over $10 million going into the Iowa caucuses and had only about one-third of the money that Howard Dean had on hand at the end of January. The existing forecasting models fail to predict a Kerry victory, and indeed, place him second or third in the primary vote (Dowdle, Adkins and Steger 2004). The failure of the forecasting models suggest that either, something rare and atypical occurred in the 2004 primaries, or the forecasting models are missing something. This paper presumes the later is the case, and that we need to go back to the drawing board to look beyond the relative support and resources of the candidates to develop more robust and accurate presidential primary vote forecasts. Studies on presidential nominations identify three theoretical factors that are likely candidates to improve the forecasting models. First, Steven Brams (1978) uses a spatial model of primary competition to argue that nominations are the result of 2 competitions among factions of each party. Understanding the ideological competition within the parties may be a factor that can be used to forecast more effectively the presidential primary vote. Second, numerous studies have argued for a resurgence of the political party establishments since the 1970s, yet no study has assessed the effects of party elites on in forecasting models of the presidential primary vote. Cohen, Noel and Zaller (2003, 2004) and Steger (2002, 2003) have argued that patterns of endorsement by party elites are a factor influencing the presidential primary vote. Third, several studies have argued that nominations are likely to be influenced by how candidates are expected to fare in the second stage of the election process (Brady and Johnson 1987; Abramson, Aldrich, Paolino and Rhode 1992; Stone, Rapoport and Abramowitz 1992). A small percentage of sophisticated voters in each party’s primary constituencies may be influenced in their nominee preferences by whether or not a candidate is expected to win during the fall election. This study looks at a variety of factors as predictors of the aggregate presidential primary vote from 1976 through 2004. Forecasting presidential nominations has become serious business. As Mayer, Adkins and Dowdle, Steger and others have argued, the main story in presidential nominations is no longer what happens during the primaries, but what is happening the year before the primaries in the invisible primary competition for support, money, and media coverage. Given the importance of the pre-primary season for the competition during the primaries, it is worthwhile to further develop our theoretical and empirical understanding of presidential nominations. Literature Review Those studies most relevant to the current study are those forecasting the presidential primary vote and those positing theoretically important factors that might be usable in presidential nomination forecasts. 3 The earliest models of voting of the presidential primaries focused on spending in each state (Goldstein 1978; Grush 1980). Bartels (1985, 1987, 1988) developed models to explain the dynamics of the campaign during the primaries. Rather than focusing on campaign spending, Bartels focused on changes in public opinion across the primaries. Bartels found that prior expectations and support for a candidate had a strong effect on primary voters’ preferences for candidates, and that voters bandwagon behind candidates who beat expectations for performance in the early caucuses and primaries. Norrander (1993) analyzed the influence of campaign spending, past performance, the number of candidates remaining in the race, and whether the candidate was a “favorite son” of the state on state primary and caucuses votes. She found these variables were better predictors of candidates’ vote shares, whereas momentum was primarily a factor for candidates who emerged as the main challenger to the nominee. Haynes, Gurian, and Nichols (1997) built on her work by tracing the influence of variables such as campaign spending, delegates pledged to the candidate, incumbency, whether the candidate was a “favorite son” of the state on the 1980 and 1988 contests. More recently, Cohen, Noel and Zaller (2001, 2003) blend the resource-based approach with the public opinion/momentum approach. They analyze both dynamic and static factors (ideological positioning, money, media and elite endorsements). They conclude that momentum continues to exist, but has changed since the 1970s, now favoring insider candidates whereas previously it was relative dark-horse candidates who benefited from momentum during the primaries. The apparent decline of momentum during the primaries led to a series of studies that focused on predicting the primary vote with information from before the primaries. If momentum effects faded or simply returned the pre-primary favorite to his or her status ante primaries, then the primary vote should be predictable using information from the pre-primary period. Mayer (1996a; 2003a, 2003b) uses the last national Gallup poll taken prior to the Iowa caucuses and 4 funds raised during the year prior to the primaries to predict the aggregate primary vote (APV) in all nominations between 1980 and 2000. Mayer finds that national poll status is a powerful predictor of the aggregate primary vote, while funds raised do not significantly predict the primary vote. A central theme in Mayer’s work, elaborated in his (1996b) book, is that the Democratic Party is less unified and therefore has more divisive nomination campaigns. According to Mayer, these divisions are embodied in the national poll results. Adkins and Dowdle (2000; 2001a) analyzed nomination contests without an incumbent president.1 Their model includes Mayer’s variables, but also the cash reserves of each candidate as of December 31 of the year prior to the election. They also craft a second model that includes two variables measuring the influence of the New Hampshire primary—each candidate’s share of New Hampshire vote and a dummy variable for the winner representing the additional media coverage and increased viability, or “bounce,” that the a winner of the Granite State’s primary usually receives. They find that both prior public support and cash reserves are significant predictors of the presidential primary vote. Steger (2000) adds to Mayer’s model by including TV news coverage during the year preceding the election. Steger focuses on differences in Democratic and Republican Party primaries between 1976 and 1996. Steger concludes that candidates’ cash reserves at the end of January significantly effect Democratic nominations while Gallup poll standings do not; while Republican candidates are differentiated by their prior public support but not by cash reserves. Steger’s assertion buttress Mayer’s (1996b) findings that the Democratic Party presidential nomination electorate is often sharply divided in exhibition season polls as to which candidate to 1 A major difference between the model of Mayer (1996a, 2003a) and those of Steger (2000) and Adkins and Dowdle (2000, 2001a) is the inclusion of presidential renomination races in the model. Mayer includes all races, where as the others exclude these races, arguing that incumbent presidents have not been beaten. Including these races in the models has the effects of distorting the estimates of other variables in the models. 5 support, while Republican Party identifiers usually coalesce around a clear front-runner before the primaries commence. Several Republican hopefuls typically start the primary season with a large war chest, while Democratic contests have tended to have a clear fundraising front-runner. These forecasting models are consistent with both the public opinion/support approach of Bartels and the resource-based models of the state-by-state primary vote. Candidates who have developed substantial support prior to the presidential primaries tend to do well in the primaries. Prior public support is a better indicator of Republican nomination success than it is for Democrats as Mayer (1996b) and (Steger 2000) show. Republican identifiers tend to rally around a clear front-runner early in the pre-primary year, and their support remains stable through the primaries. Polls of Democratic candidate fields are, by comparison, highly volatile (see below). Adkins and Dowdle (2000) and Steger (2000) find that how much money candidates raise is not as important as how much they have on hand when the primaries begin. The funds on hand constitute most of what candidates will have available to campaign during the primaries. One effect of front-loading is to severely constrain the time frame for candidates to take advantage of momentum (should they gain it). Dark horse candidates lack the time to raise enough money and convert it into usable campaign appeals.2 Candidates who have a large bank can take advantage of momentum should the get it (e.g., Clinton in 1992) or beat back a challenger if they do not do well in the early primaries (e.g., Mondale in 1984 or Bush in 2004). These forecasting models, however, fail to predict correctly the winner of the 2004 Democratic nominee. Dowdle, Adkins and Steger (2004) show that the Mayer model as well as their own models predict Howard Dean as the winner. The reason is that Dean led in pre-primary national Gallup polls, had raised almost $11 million more than John Kerry, and had about $3 2 Though the internet increases the speed with which candidates can raise money, they still need time to produce and disseminate direct mail or television advertisements; and they pay premium rates for media buys since they typically must push off previously scheduled ads. 6 million in the bank to Kerry’s $1 million at the start of the primary season. While the models may remain valid in Republican races, the inability to predict correctly the winner is troubling. This is a weak barometer of the efficacy of a model. The failure of the model could owe to momentum during the caucuses or primaries, or to a deficiency in the model. Cohen, Noel and Zaller (2004) attribute Kerry’s success in the primaries to “instantaneous momentum” following Iowa and New Hampshire—suggesting the continued import of momentum. The Iowa Political Stock Market, however, shows that Dean’s support/chances of success collapsed during the 12 days preceding the Iowa caucuses. The pre-Iowa national polls were conducted too early to measure this drop.3 Dean’s market value dropped from $.648 on January 11th, 2004 to an average of $.134 on January 23rd—the day before the Iowa caucuses. Kerry’s market value increased from $.033 to $.552 over the same time frame.4 The Iowa Political Stock Market has generated the most accurate measures of candidate support over the past four presidential nomination cycles of any poll or market-type measure (Berg, et al. 2000, 2003). Together with the results of the Iowa Caucus, it would appear that the momentum and collapse and Kerry’s momentum occurred before the nomination voting began. The question is how can a model of the nomination competition be constructed that enables us to figure out which factors systematically and accurately forecast the presidential primary vote. Strictly speaking, the IPSM only measures candidate chances at a given point, and would be a useful dependent variable rather than an independent variable. 3 The same problem occurs in 1976, when Henry Jackson appeared to fade prior to Iowa (according to newspaper reports, but not indicated in available national polls). 4 Results are from: http://18.104.22.168/pricehistory/PriceHistory_GetData.cfm. The IPSM involves people trading in shares of stock. Candidates have shares valuing from a fraction of a penny to $1. Shares of the candidate winning the nomination pay out as a dollar. Candidate shares can be interpreted as the market’s valuation of the candidates’ chances of winning the nomination. 7 One way to improve forecasting models will be to include the effects of the party elite establishments. As early as 1984, Price argued that the “parties are back” in his analysis of political competition in the United States. Steger and Davis (2000), Steger (2002b), and Cohen, Noel and Zaller (2003, 2004) argue that patterns of endorsement by party elites influence the presidential primary vote. Endorsements have the potential to affect the primary vote in several ways (Steger 2002b). One, endorsements serve as cues to party activists, contributors and the media as to who are the viable and desirable candidates. Such cues may be important when voters cannot use party labels to differentiate candidates, as is the case in nomination campaigns. Second, endorsing elites often are involved in the campaign as surrogates at events, attacking rivals and defending the candidate in the media. Third, in a few cases, endorsing candidates actively aid the campaign through their own organizations and fundraising networks. There is reason to believe that elite endorsements are more important in the Republican Party than in the Democratic Party (Steger 2002b). In part, this may be a lingering effect of the Hunt Commission creation of super-delegates.5 Democratic governors, senators and representatives endorse presidential candidates, on average six to seven months after their Republican counter-parts and ultimately, far fewer of these officials on the Democratic side endorse a candidate prior to the primaries (Steger 2000). Further, like Democratic identifiers, Democratic office-holders are highly divided in their preferences for candidates, dividing their endorsements more evenly across presidential candidates (Steger and Davis 2002). The media, contributors and activists have less indication as to whom they should be supporting prior to the caucuses and primaries. Another way forecasts can be improved is to include information on candidates’ electability. Brady and Johnston (1987) find that electability was a significant factors in Walter Mondale’s victory over Gary Hart in 1984. Abramson, Aldrich and Rhode (1992) find that at 5 As part of the deal creating super-delegates, those serving the party in this capacity were to refrain from endorsing candidates before primary voters had the opportunity to decide. 8 least some voters in the 1988 primaries exhibited “sophisticated voting” in which they considered both a candidates’ appeal on issues and his/her chances of winning the election in the fall. Using a rational decision-making model, Abramson, et al. argue that the presidential nomination choice can be thought of as selecting the candidate with the greatest expected utility. The expected utility of supporting a candidate is a function of the value of the candidate to a person (determined by policy agreement, for example) weighted by the probability that the candidate wins the nomination. Expanding this argument to a two-stage lottery choice, the value of a candidate is itself, a function of policy agreement with the individual weighted by the probability that the candidate wins in the second stage of the election process (Steger 2003). As such, primary voters acting in accord with the rational choice model should take into account both candidate appeal (the extent to which a candidate satisfies a voter’s preferences for policy and candidate attributes such as race, character, etc.) as weighted by the probability that a candidate will win the general election. Thus, we have the basis for a forecasting model of presidential primary vote, using information from the period prior to the primaries: candidates’ relative standing in national public opinion polls, cash on hand, endorsements and electability. Unfortunately, at this moment, the data on spatial distribution of policy in the political parties is not in a form that can be used in a forecasting model. The analysis of this data will be qualitative. Measurement of Data National Poll Standing One of the best, if not the strongest, predictor of nomination outcomes has been national preference polls of partisan public opinion (Mutz, 1995; Mayer, 1996a; Hinckley and Green, 1996; Adkins and Dowdle, 2000, 2001a, 2001b; Norrander, 2000a, 200b; Steger, 2000, Steger, Dowdle, and Adkins, 2004). National poll results (I use Gallup polls) have a direct and an 9 indirect effect on the process. As a direct effect, polling data represents the level of rank-and-file partisan support for a particular candidate. The indirect effect is demonstrated the polls influence on media attention and support from contributors, activists and the public (Mutz, 1995; Hinckley and Green, 1996; Adkins and Dowdle, 2002). The measure used is the candidates’ average percentage of national Gallup polls during January (prior to the Iowa caucuses) reported in monthly editions of The Gallup Report or annual editions of The Gallup Poll from 1976 to 2004. Candidates’ shares are those reported in the polls. Some polls include results for candidates not actually in the candidate fields (e.g., Kennedy in 1976). Poll support for non-candidates is treated as undecided or uncommitted or none of the above. We do not, as some have done, proportionately redistribute non-candidate support among the other candidates. Fund-raising, campaign spending and money on hand The exact nature of the relationship between money raised by a campaign during the exhibition season and future success in primary races is a complex one. All the Republican winners of the “money primary” since 1980 to win the Republican nomination. Until 2004, no nominee had lost the money primary since Jimmy Carter in 1976 (Adkins and Dowdle, 2002). Previous forecasting models relied on a measure of total fundraising for the entire exhibition season that is highly collinear with the Gallup poll variable as measured by the Lewis-Beck (1980) standard. Separating total fundraising into two variables, campaign expenditures and cash reserves, solves this problem (Adkins and Dowdle, forthcoming). This study uses two financial variables: funds raised at the end of January of the election year, and money on hand at the end of January of the election year.6 6 Technically, the January report includes funds raised and expenditures that occur after the Iowa caucuses for 2000 and 2004, and for four days after the New Hampshire primary in 2004. 10 Money spent or disbursements are the amount of money in millions of dollars spent by each candidate spent during a quarter or month. The data for each quarter or monthly observation comes from Line 9 “Total Disbursements This Period” of an individual presidential candidate “Reports of Receipts and Disbursements” (form 3P) for that particular quarter or month.7 The variable representing cash reserves is calculated as the unspent money in millions of dollars that each candidate has available at the end of a quarter or month. There is a problem with using the raw data since campaigns have become more expensive, creating a heteroskedasticity problem that increases autocorrelation in the models. The problem of growing costs of campaigns is resolved by dividing the cumulative sum of disbursed funds by each candidate by the cumulative disbursed funds of all candidates running in a nomination campaign. This creates a measure of each candidates’ percentage share of campaign funds spent in each nomination cycle, and which becomes comparable across time. The “cash disbursements” variable is calculated as disbursements of a given candidate as a percentage of the total disbursements made by all of the candidates in a given quarter. Using the percentage of disbursements has the advantage of controlling for the ever increasing sums of money in presidential nomination campaigns. As such, the variable effectively standardizes money across presidential nomination campaigns from the 1970s through 2004. Network News Coverage Numerous studies hold that presidential nomination campaigns are mass media campaigns (e.g., Patterson 1980; Robinson and Sheehan 1982; Arterton 1984; Lichter and Lichter 1986, 1988). These studies recognize that candidates may be able to take advantage of free exposure on national network news to generate name recognition and support among potential primary voters. Shrewd candidates may be able to compete, partially making up for 7 I thank Randall Adkins and Andrew Dowdle for providing me with these data. 11 what they lack in funds, by getting the attention of potential voters through the network news programs--where most voters gain their information about candidates. News coverage has been shown to play a major role in the dynamics of the primary season (Bartels 1985; 1988). Media coverage gives candidates visibility, name recognition, and prestige (Peabody, Ornstein, and Rhode 1976, 243-34). Greater media coverage increases name recognition—especially for lesserknown candidates, which may increase perceptions of a candidate’s viability and ability to attract supporters and raise campaign funds (Bartels 1988; Abramson, et al. 1992; Mutz 1997). A second way of thinking about candidate coverage can be found in media agenda-setting studies. In this perspective, the media do not tell voters what to think, but what to think about (Cohen 1963). This is accomplished through the volume of attention given by media outlets to certain topics (McCombs and Shaw, 1972; Iyengar and Kinder 1987). With respect to policy issues, when news coverage focuses more on a particular issue, people are more likely to cite that issue as the most important concern facing the nation (Iyengar, et al. 1982). The same principle should apply to candidates. Given the centrality of the horse race in determining the volume of coverage given to candidates, the volume of news coverage will reflect the relative salience of the candidates in the nomination campaign. Further, the media’s coverage of candidates will reflect not only their relative position in the race, but also certain intangible aspects like perceptions of candidates’ character, experience, etc. Since most voters get their information from the mass media, the volume of exposure will influence public attention to the candidates. Candidates' campaign coverage is measured as the frequency of candidate appearances or mentions in campaign stories on nightly national network news programs.8 The variable 8 Ideally, a measure of candidates' campaign coverage would include a measure of the tone coverage. Limitations of the Vanderbilt television archives, however, preclude obtaining a reliable measure of coverage tone. The network news abstracts can be obtained from the website at http://tvnews.vanderbilt.edu 12 excludes network news stories relating to candidates’ governing activities.9 Such coverage is uncorrelated with candidates’ standing in the polls or their performance in the polls (Steger 2002). Only that portion of candidates’ network news coverage relating to the campaign correlates strongly with candidate standing in the polls and in the primaries. The Vanderbilt Television Archives were used to generate an event-count of nightly network news stories that referred to or mentioned candidates campaigning for the presidential nomination of one or the other major political parties. These event counts were aggregated to get a daily summary count for each candidate. For instance, if each of the three networks mentioned George W. Bush, his daily score would be a three. Since a news story may refer to multiple candidates, the number of candidate-mentions exceeds the actual number of network news stories mentioning candidates in each nomination campaign. The daily scores for each candidate are then summed to the quarterly level to get a measure of quarterly campaign coverage comparable to the money variables. The campaign news variable also is calculated as the volume of coverage received by a candidate as percentage of the total coverage received by all candidates in a given quarter. Candidate Endorsements by Party Elites Data on candidate endorsements were obtained from three sources. For the years 1988 through 2004, endorsements are measured from a content analysis of newspapers in all 50 states. Articles were identified through a Lexis-Nexis search of newspapers in four regions from January 1, 1987 through March 31, 2004. A Lexis-Nexis search identified thousands of articles in newspapers in 50 states explicitly focusing on or referring to the endorsement of candidates for the Democratic and Republican presidential nominations. Articles were identified through the Academic Universe program of Lexis-Nexis using candidate names. Each article was then read 9 Campaign coverage was defined as a news story that mentioned the candidate in the context of any aspect of the campaign. Candidate views on a policy matter, in which he or she is not directly involved in the unit of government making decisions, were coded as campaign stories. 13 thoroughly, coding the following variables: endorsed candidate, date of story, name of newspaper, date of endorsement (if discernible), endorsing individual or group, position of endorsing individual, state of endorsing individual, and a brief description of the endorsement or endorser. Each endorsement was counted as a separate unit of analysis. Endorser positions were then coded for governor, US Senator, statewide party or elected official, US Representative, state senator, state representative, and local officials, on a scale of one to seven. Local officials were dropped from the analyses because of inconsistencies in reporting of local official endorsements from state to state and newspaper to newspaper. Additionally, endorsements of former party and elected officials were coded as a second variable using the same coding scheme. These also are dropped from this analysis. This method was used for searching in the years 1996 and 2000. For 1988 and 2004, a streamlined approach was used to reduce the volume of repeat articles. For 1988 and 2004, endorsements were searched using candidate names (usually just last name), the name of the endorsing official, and the word “endors!” Lexis-Nexis has several limitations as a measuring instrument. First, the method may miss endorsements if the reporting of endorsements used words like “backed” or “support” if the article did not also include variations on the word endorse. This problem is most acute for state and local officials. The problem is less for top-level officials--governors, US senators and representatives since these endorsements generally receive coverage in multiple newspapers in different regions of the country. Another limitation is that the date of an endorsement often could not be discerned unless the event itself was reported. Endorsements were dated on the date of an event if reported or a month was mentioned. Otherwise, an endorsement was dated as the month in which the endorsement was reported. These dates were then collapsed into quarters, beginning in January of the year prior to the election. There are some errors in the dating of specific endorsements. The limited data available on the timing of endorsements limits the extent to 14 which temporal sequencing between endorsements and fund-raising and candidate poll-position can be discerned. Few endorsements were reported outside of the quarter in which they were reported, with exceptions occurring mostly in the first quarter of the pre-primary year when press reports include some endorsements that had been made in the previous year. The measure presented is just a count of the number of governors and senators and US representatives endorsing a candidate prior to the Iowa caucuses. The most significant limitation of measuring endorsements through newspaper reporting is that reporting varies considerably from newspaper to newspaper and from state to state.10 Newspapers of different states do not give similar coverage to campaign endorsements. The mountain states of Idaho, Montana, Nevada, Utah, Wyoming, North and South Dakota, Alaska and Hawaii appear to be under-represented in the sample. Some papers (e.g., Cincinnati Enquirer) reported on local endorsements down to city commissioners. For other states, the endorsements of state legislators and county officials may not be reported at all. Often, a news article would report an imprecise number like “most state legislators have endorsed George Bush." Such imprecision precludes reliable measurement of state and local party and elected official endorsements. This further limits the range of state-level comparisons of the effects of endorsements on fund-raising and organizational development. The bias matters because lower tier candidates tend to build organizations around local and county level officials. Leading candidates like George Bush generally went after federal and statewide officials in their organizational efforts. Non-front-runners like Bill Bradley had relatively more success with state Unfortunately, it is not possible to obtain reliable lists of endorsements for past elections. Nor is it reliable to use the websites of candidates for the 2000 election. Candidates appear to selectively advertise their endorsements on state-by-state web pages. Further, some candidates like Bill Bradley did not advertise many of their endorsements as they sought to cultivate a “nonestablishment” or “outsider” image. 10 15 legislators. The newspapers appear to be most consistent reporting higher level officials— Governors, statewide party and elected officials, US Senators and Representatives. The value of measuring endorsements through newspaper reports of endorsements is that they are public announcements of support. This is important since one of the major theorized effects is cue giving. The level of exposure of a given endorsement varies with the status of the endorsing individual. Gubernatorial, senatorial, and congressional endorsements generally receive widespread attention—repeatedly in the newspapers in a state, and repeatedly in papers in other states. Endorsements by local officials and state legislators are more likely to be missed. Because of this unreliability, state legislators and representatives are dropped from the analyses. For years prior to 1988, I use the measure of endorsements generated by Cohen, Noel and Zaller.11 These data differ in several respects. One, there are fewer endorsements recorded in these years. This may be due to either fewer endorsements being made, fewer news reports of endorsements, or to a difference in methodology. As a matter of comparison, I correlated the Cohen, Noel and Zaller measures of endorsements for the years 1988 to 2000 and found the aggregate counts of endorsements to be very highly correlated: r = 1.0 for governors, r = .98 for senators, and r = .91 for US Representatives. These datasets differ mainly in the dating of endorsements which is irrelevant for this study. To account for differences in the frequency of endorsements across nominations, each candidate’’ endorsements were converted to a percentage of the endorsements of all candidates running in a particular nomination campaign. Candidate electability As noted earlier, at least some primary voters have been found to make sophisticated voting decisions (Brady and Johnston 1987; Abramson, Aldrich and Rhode 1992). Sophisticated voters consider candidates’ appeal on policy issues and/or character, and his chances of winning 11 I am exceptionally grateful to Hans Noel for providing me with his data. 16 the election in the fall. The idea is that primary voters may select the candidate with the greatest expected utility. The expected utility in a presidential nomination campaign is a two-step process. First, the voter calculates the expected utility of the candidate, which can be represented as: Uc1= P(Uc1|win) + (1-P)( Uc1|loses), where Uc1 represents candidate one, P represents the probability the candidate wins the nomination, (1-P) represents the probability the candidate loses, (Uc1|win) represents the utility of the candidate given the candidate wins the nomination, and (Uc1|loses) represents the utility of the candidate given the candidate loses the nomination campaign. The value of (Uc1|win) depends on the outcome of the general election, which is a second utility equation like the first but using the probabilities of winning and losing the general election. A rational voter would thus take into consideration the probabilities of each candidate winning in the general election. The utility of each candidate is a function of the value of the candidate to a person (determined by policy agreement or voter preferences for candidate characteristics such as race, experience, or other character quality) (Steger 2003). It seems likely that factors influencing the utility estimations of different candidates may vary across elections. Judging from the kinds of questions asked in various national surveys in different nomination campaigns, the relative value of particular candidate characteristics seems to vary across nominations. For example, candidates’ integrity seems to have been an important consideration in 1976, leadership ability seems to have been particularly important in 1984, and integrity emerges again in polls in 2000 and 20004. While the relative salience of certain candidate characteristics varies across elections, those characteristics would be relatively constant in comparisons of candidates within a given nomination campaign. The same principle holds for policies. Analyzing these multitudinous factors would lead us into the realm of description, which is not the purpose of this paper. Instead, the point is to analyze the effects of systematic factors that affect all candidates across all elections. Electability is one such factor. 17 Measuring electability, however, is exceptionally difficult and there are a host of methodological issues. There are no consistent questions asked by any polling organization across all of the nomination campaigns. This limits the comparability of measures created. Perhaps the best questions asked are trial heat questions that ask voters their vote preference for a candidate of one party versus a candidate of the opposing party. For example, Gallup has used a variant of “Suppose the (year) presidential election were being held TODAY. If (candidate X) were the (Party A ) candidate and (Candidate Y) were the (Party B) candidate, which would you vote for?” Such questions are not asked in every nomination campaign prior to the Iowa caucuses, and generally are not asked of the all the candidates of a party vis-à-vis the eventual nominee of the opposing party. Another methodological issue is that surveys vary in the time between the survey and the Iowa caucuses. As noted in the discussion about polling, timing matters because public opinion varies—especially for candidates in the Democratic Party. Another methodological problem is that even when surveys are conducted with usable questions, the sample sets differ between adult population, registered voters, party identifiers, party identifiers and independents, and party identifiers and independents who lean toward that party. These factors decrease the comparability of various possible measures of candidate electability. After reviewing well over 1,000 survey questions asked during January and early February across all of the nomination campaigns between 1976 and 2004, there seem to be three types of questions that seem to measure electability. Most preferable are trial heat comparisons. Such questions were available from different survey organizations for at least two candidates for every year between 1976 and 2004. At a minimum, the questions involved candidates the polling organizations considered viable for the nomination. A second type of question asks which candidate in the party’s candidate field is most electable. For example, in 1988 a CBS News/New York Times poll asked registered primary voters of each party, “Regardless of which 18 candidate you support for the 1988 presidential nomination, which of the following Democratic (Republican) candidates do you think would have the best chance of winning the election in November if he were nominated?” Several other organizations had similar questions in different years. A third type of question used records favorable or unfavorable opinion of candidate. Using the combination of these questions, the first variable created is a dummy variable using a trial heat question that identifies the “most electable” candidate. For “ELECT1” the most electable candidate is scored as a one and all other candidates are scored as a zero. This is the narrowest definition of electability since it only scores one candidate and misses the nuances. Note that in no year was the coding decision made based on just one poll or type of question or limited to a comparison of one or two candidates. While the least desirable from a measurement standpoint, this is likely the most reliable of the electability measures. A second electability variable, ELECT2, is coded as the number of responses indicating a particular candidate in the various trial heat questions asked in a given poll in a given year. This variable uses all of the trial heat information available, and codes candidates according to the responses in the poll. Candidates for which no question is asked are coded as a 10—the lowest level recorded in any of the polls identified for a given candidate in a trial heat question. Because of the arbitrary decision regarding the coding of candidates not included in a trial heat question, this measure is also less reliable than the first measure. It does, however, use more of the information available and may be more accurate. Importantly, this measure is superior to ELECT1 in differentiating among candidates by providing a measure of relative electability. The third electability variable, ELECT3, is the most inclusive with nearly all candidates being coded with a score from a question asked in a single survey in a given election year. The two years with the least number of candidates being questioned were 1976 and 1988. For these 19 two years, I used different questions for the values in ELECT3. The 1988 question from a poll by Times Mirror conducted by Gallup, Jan. 8-17 of a national adult sample is: “I am going to read you a list of names. Please tell me how likely you are to vote for each person in November (1988) if he is the nominee of the Democratic (Republican) Party. Would you be very likely, somewhat likely, not too likely or not at all likely to vote for him” Candidates scores for ELECT3 were scored as the combined number of very likely or somewhat likely categories. For 1976, ELECT3 responses are coded from a poll by CBS News/New York Times that asked the following question: “Do you have a favorable or unfavorable opinion about … “candidate name.” The candidates’ score for ELECT3 is the percentage responding favorable. For determining the most electable candidate in ELECT1, I used the difference between favorable and unfavorable responses. A summary of sources, dates of surveys, question type, sample type, and number of candidates for whom responses were identified are presented in Table 1. Table 1: Sources of information for coding electability of candidates. Year 2004 2000 1996 1992 1988 1988 1984 1980 source Newsweek/PSRA Gallup/CNN/USA Today Newsweek/PSRA NBC News/ Wall Street Journal Times Mirror/Gallup Times Mirror/Gallup Date Jan 8-9, Jan. 17-19 Jan. 25-26 Jan. 17-21 Jan. 8-17 Jan. 8-17 Type Trial heat Trial heat Trial heat Trial heat Trial heat Likely to vote* Trial heat Trial heat sample Adults Registered Registered Registered Adults Adults # of candidates 6 of 8 Dem. 2 of 2 Dem. 2 of 6 Rep. 3 of 8 Rep. 5 of 5 Dem. 3 of 8 Dem 3 of 6 GOP 8 of 8 Dem 6 of 6 Dem 4 of 8 Dem 3 of 7 Rep. Gallup Jan. 13-16 Registered Time/Yankelovich, Skelly Jan. 23-24 Registered & White 1976 CBS News/NYT Feb. 2-8 Trial heat Adult 2 of 13 Dem. 1976 CBS News/NYT Feb. 2-8 Favorable# Adult 9 of 13 Dem Survey results used here were obtained from searches of the iPoll Databank provided by the Roper Center for Public Opinion Research, University of Connecticut 20 Analysis and Discussion The following analyses will be at the bivariate level. There is high multicollinearity between these variables. The worst culprit is collinearity between national Gallup poll measures and the TV news variable and candidate endorsements. These three variables are highly interrelated, exceeding r > .80, which Lewis-Beck (1980) identifies as the level at which multicollinearity distorts the results of multivariate regression models. After considerable efforts to find alternative strategies, mainly simultaneous equation models, I simply present the results of a series of bivariate regression equations between all of the variables identified above and the aggregate primary vote (APV). While this amounts to a willful misspecification that results in biased coefficient estimates, the models are still informative as to the relative weights of each variable as a predictor of the APV. The main statistic of interest is the r2 for each model. The coefficients for the constant term and the relevant variable are presented for readers’ benefit. Bivariate models of the various predictors on the APV of both political parties’ presidential candidates are presented in table 2. These models indicate that all of the commonly identified variables are significant and substantially related to the APV. The January Gallup figures produce the largest r2 followed by endorsements by governors, senators and representatives. January TV news coverage of the candidates, and candidate endorsements by senators and governors produce the next highest r2. The two financial variables and the three electability measures produce relatively lower goodness-of-fit statistics. The r2 of these models are sufficiently low, as to be likely that the effects wash out when combined with other variables in multivariate analyses. Although not shown, all of the financial variables—funds disbursed and cash on-hand at the end of January are not significant in a multivariate model with any one of the three variables—TV news coverage in January, January Gallup results, or either measure of candidate endorsements. This is notable because these variables are not as highly correlated. 21 Interestingly, the different bivariate models of the APV produce different mixes of outliers in the residuals. Outliers represent extreme over- or under-predictions of a candidate’s vote share. For example, in the Gallup model, all of the candidates who have been identified in previous studies as having momentum during the primary season are residuals—Jimmy Carter ’76, George Bush ’80, Gary Hart ’84, and John Kerry ’04, are significant and substantial outliers with vote shares being under-predicted by more than 2 standard deviations. In the model using cash on hand, Carter ’76, Reagan ’80, Bush ’88, Gore ’00, and Kerry ’04 are all substantially underestimated (residuals greater than 2 st. dev.), while Henry Jackson is over-predicted by this amount. In the model using campaign disbursements, Clinton ’92, Dole ’96 and Kerry ’04 are substantial under-predicted (by more than 2 st. dev.). In the model with TV news coverage of the candidates, Reagan ’80, Hart ’84, Dukakis ’88, Bush ’88, and Kerry ’04 are under-predicted while Bradley ’00 is over-predicted (by more than 2 st. dev.). The electability variables produce a different set of candidates’ vote shares being over- or under- predicted. In the ELECT1 model, Hart ’88, Dean and Gephardt ’04 are over-predicted while Kerry’s ’04 vote share is underpredicted (by more than 2 st. dev.). In the somewhat more inclusive trial heat electability variable, ELECT2, the vote shares of Bush ’88, Clinton ’92, Dole ’96, Bush ’00, and Kerry ’04 are under-predicted (by more than 1.8 st. dev.). The most inclusive measure of electability, ELECT3, produces under-predictions of Reagan ’80, Bush ’88, Clinton ’92, Gore ’00, and Kerry ’04 (by more than 1.8 s. d.). The models with the endorsement variables indicate that the vote shares of Representatives (Udall ‘76 and Gephardt ’88 and ‘04) and Senators (Bentsen ’76, Baker ’80, Glenn ’84, Dole ‘88) are over-predicted, while “outsider” or maverick candidates Carter, Reagan, Hart, Dukakis, and McCain are all under-predicted (by more than 1.8 s.d.). Assessing the residuals suggests that the various variables may tapping different aspects of the nomination process. The Gallup, TV and financial variables would appear to be similar in their 22 effects, though with some differences. The endorsement variables may be tapping a Washington insider/outsider dimension. The electability measures differ in that most of the over- and underpredictions occur on the Democratic side, with ELECT 2 and ELECT3 producing the greatest under-predictions for 5 candidates each—for of whom went on to win the plurality of the vote in the general election. By this standard, the ELECT1 variable seems to produce the least serious errors and all on the Democratic side. Table 2: Bivariate forecasting models of the Aggregate Vote in combined Democratic and Republican primaries, 1976-2004. Aggregate Primary Vote Gallup (mean of pre-Iowa January polls) r2 Constant (st. error) Est. b coeff. (st. error) Money on hand at end of January Constant (st. error) Est. b coeff. (st. error) Money spent through January Constant (st. error) Est. b coeff. (st. error) Candidate’s share of Jan. TV news volume Constant (st. error) Est. b coeff. (st. error) r2 .61 1.87 (1.87) 1.05 (.10) .38 5.83 (2.24) .58 (.09) .32 3.71 (2.67) .82 (.15) .48 2.66 (2.70) 1.17 (.15) .26 9.64 (2.21) 27.38 (5.58) .32 -5.03 (3.92) .75 (.14) .35 -5.51 (3.73) .82 (.14) .51 5.06 (1.97) .62 (.07) .43 6.04 (2.09) .55 (.08) P <.05 * * * r2 * r2 * * * Electability (most electable dummy) r2 Constant (st. error) Est. b coeff. (st. error) Electability (% elect or acceptable) r2 Constant (st. error) Est. b coeff. (st. error) Electabilty (% elect/accept/favorability) Constant (st. error) Est. b coeff. (st. error) r2 * * * * * * Endorsements (gov., sen., rep.) r Constant (st. error) Est. b coeff. (st. error) 2 Endorsements (gov & sen. Only) r2 Constant (st. error) Est. b coeff. (st. error) 23 In addition to the combined party models, I estimated separate models for the aggregate primary vote of each political party, presented in table 3. The models indicate the similar variables are significant predictors of the APV at the bivariate level, although all of the models account for a much higher percentage of variance explained in the Republican primaries compared to the Democratic primaries. Usually finding similar variables significant and in the same direct would indicate a single, combined-party model would be fine. The differences in the variance explained however, combined with information about individual coefficients suggests that different things are going on in the two parties’ presidential primaries. In almost no model are the over- or under-predictions of individual candidates’ vote shares as severe in the separate party models as they are in the combined party models. The differences between the parties APV in the model using national Gallup polls reinforces what Mayer, Adkins, Dowdle and Steger found previously. Republican nomination outcomes do not change much from the pre-primary to the primary. Analysis of the residuals for the two models indicates under-estimation of the vote shares of Carter, Hart and Kerry (by more than 2 st. dev.). None of the Republicans are similarly under-estimated, though Baker ’80, Connally ’80, and Dole ’88 are overestimated (by more than 1.5 st. dev.). Money spent or disbursements produces nearly identical errors to the equivalent model in the combined party APV, and these models account for the least variance explained of those analyzed. The TV model under-predicts (by more than 2 st. dev.) Hart’s vote share in 1984, but that is the only substantial outlier in either party’s model. It would appear that TV coverage is a relatively good predictor both party’s candidates. The differences between the endorsement models for the two parties are similar to the traditional variable models. Endorsements explain a great deal of the variance in the APV in 24 Republican nominations but relatively little in the models of the Democratic APV. The patterns of residual error are similar to those noted for the full models. Table 3: Bivariate forecasting models of the Aggregate Vote (APV) in Democratic and Republican primaries, 1976-2004. Democratic APV Gallup (mean of pre-Iowa Jan. polls) r2 Constant (st. error) Est. b coeff. (st. error) Money on hand at end of January r Constant (st. error) Est. b coeff. (st. error) Money spent through January r2 Constant (st. error) Est. b coeff. (st. error) Candidate’s % of Jan. TV news volume r Constant (st. error) Est. b coeff. (st. error) Electability (most electable dummy) r2 Constant (st. error) Est. b coeff. (st. error) Electability (elect or acceptable %) r2 Constant (st. error) Est. b coeff. (st. error) Electabilty (elect/accept/favorability) r Constant (st. error) Est. b coeff. (st. error) Endorsements (gov., sen., rep.) r2 Constant (st. error) Est. b coeff. (st. error) Endorsements (gov & sen. Only) r Constant (st. error) Est. b coeff. (st. error) 2 2 2 2 Republican APV .85 -.70 (2.13) 1.19 (.10) .59 2.83 (3.4) .83 (.14) .40 -2.60 (5.38) 1.21 (.30) .64 -6.95 (4.19) 1.46 (.22) .86 6.33 (1.76) 56.43 (4.56) .48 -8.41 (5.77) .88 (.18) .55 -9.15 (5.0) 1.05 (.18) .82 3.09 (2.13) .78 (.07) .60 4.77 (3.17) .67 (.11) .43 3.84 (2.76) .91 (.16) .27 7.41 (2.87) .44 (.11) .29 5.79 (3.01) .67 (.16) .37 .05 (3.53) .96 (.20) .02 11.75 (3.0) 10.56 (7.44) .17 -1.54 (5.53) .62 (.20) .18 -1.62 (5.35) .63 (.20) .28 6.93 (2.89) .47 (.12) .31 7.10 (2.78) .46 (.11) * * * * * * * * * * * * * * * * * * - * * * The main differences between the separate models occur on the electability models. ELECT1, the narrowest definition of electability, accounts for 86% of the variance explained in the Republican APV, but is not significant on the Democratic side. The main outliers in the Republican model are those of runners-up (Bush ’80, Dole ’88, Buchanan ’96, and McCain ’00) 25 and these are all by less than 2 standard. deviations. The ELECT2 and ELECT3 models are slightly better predictors of the Democratic APV but are notably weaker predictors of the Republican APV—though the errors of over- and under-prediction are reduced in these models compared to the ELECT1 model. In general, these results suggest that electability has not been a major factor in Democratic primary voting, but has been in Republican primary voting. One reason Democrats may be losing presidential elections is that they have not been as concerned with the electability of their presidential candidates. Notably, the errors of prediction are relatively small (less than one st. dev.) for both Carter and Clinton in these models—the two candidates who ranked most electable among the Democratic candidates who became the nominees of their party! That electability does not account for much of the APV in Democratic primaries while the ELECT1 model accounts for even more variance explained than the Gallup model in Republican primaries is an important result. Conclusions The main story here seems to be that Democratic and Republican primary voters do consider different factors. We just don’t know exactly what most Democratic primary voters do consider. The results here indicate what Democrats do not seem to consider, or at least do not seem to consider as much as Republicans—endorsements and electability of the candidates. All of these things inter-relate to create distinct dynamics in Democratic and Republican presidential nomination campaigns. Democratic office-holders tend to delay endorsing candidates (except for their friends/colleagues) while Republican office-holders are much more strategic in their endorsement patterns—rallying around a single front-runner early. This sends a signal to the media, to party activists and to campaign contributors as to who are the serious presidential candidates on the Republican side. In every case since 1980, the Republicans have gone on to nominate the candidate who was perceived as being the most electable of those in the 26 race. Democrats did so in 1976, 1984, and 1992, but not in 1988, 2000 or 2004. Democrats won two of the three presidential elections in which they did select the candidate deemed most electable about one month before the Iowa Caucuses. The two parties differ in the dynamics indicated in polling results for their respective presidential candidate fields during the pre-primary season. Among Republican candidates, a front-runner typically emerges early in the pre-primary season (usually by the second quarter) and remains atop the field until the convention. Further, this candidate usually has two to three times the support of his rivals in national Gallup polls—also providing a clear indication of who the most viable candidate is.12 Democratic battles, on the other hand, are often less stable, and it is not unusual for the eventual nominee to emerge as the leader in public opinion until the beginning of the election year (Mayer, 1996b; Newport, 1999). Democratic polls are less stable for several reasons. One reason identified by Mayer (1996), that the Democratic Party identifiers are more diverse and divided in their preferences for presidential candidates. There typically is not a candidate who attracts a majority of support. Only in 1984 and 2000, when a former and current vice president sought the nomination, did a Democratic candidate attract as much as 40% of support in national Gallup polls. In the other post-reform years (’76, ’88, ’92 and ’04), the front-runners typically attracted less than 30% of respondents in national polls. Effectively, there is no leader in such a situation, and the media, contributors and activists have no clear indication as to the viability of candidates. This may be one reason Democratic office holders refrain from endorsing candidates until the primaries begin. The Democratic candidate fields also tend to remain unstable until the primaries begin. In 1976, national polls indicated that Hubert Humphrey and Ted Kennedy attracted more support than any of the candidates in the race. Kennedy and Humphrey announced their non-candidacy Viability here, refers to a candidates chances of winning the nomination. Electability refers to the chances of winning the general election. 12 27 in November and December of 1975, respectively. At this point, those Democrats hoping for a declaration of candidacy have to begin looking elsewhere—creating considerable volatility in polls. In 1988, the main source of instability was Gary Hart’s on, off, on-again candidacy created instability in pre-primary polls. In 1992, the campaign was effectively frozen in place by speculation about whether Mario Cuomo would run. His declaration of non-candidacy in late October of 1991 created considerable fluidity in late pre-primary polls as his “supporters” began to look for other candidates. In 2004, some instability was created by Clark’s late entry into the race in September. The problems with having a divided party and volatile candidate fields is that interested parties—the media, activists and contributors have little indication as to who is viable and can win. The effect is to lessen the predictive power of polling results on the aggregate primary vote. Prescriptively, it would appear that Republicans are doing it right. Republican officeholders jump into presidential nominations early with endorsements of one main candidate; that candidate gains the greatest lead in national opinion polls; that candidate is perceived as the most electable; receives the most network news coverage (though not with as decisive a lead as in the other categories), raises the most money by the end of January and has the most cash on hand to compete in the primaries; and wins the most votes in the primaries. I do not mean to imply a causal sequence in this paragraph, and note that all of these variables are interacting at the quarterly level in the year prior to the primaries. Democrats, in contrast, experience very little coordinated interaction---and that is one of their big problems. 28 BIBLIOGRAPHY Aldrich, John. 1980a. Before the Convention: Strategies and Choices in Presidential Nominations. Chicago, IL: University of Chicago. Aldrich, John. 1980b. A Dynamic Model of Presidential Nomination Campaigns. American Political Science Review 74:651-669. Adkins, Randall. 2000. “Progressive Ambition and the Presidency: Reexamining the Benefit of Running for Higher Office.” Paper presented at the annual meeting of the Midwest Political Science Association. Chicago, IL. Adkins, Randall E., and Andrew J. Dowdle. 2000. Break Out the Mint Juleps in New Hampshire? Is New Hampshire the ‘Primary’ Culprit Limiting Presidential Nomination Forecasts? American Politics Quarterly 28:251-269. Adkins, Randall E., and Andrew J. Dowdle. 2001a. How Important Are Iowa and New Hampshire to Winning Post-Reform Presidential Nominations? Political Research Quarterly. 54:431-444. Adkins, Randall E., and Andrew J. Dowdle. 2001b. Is the Exhibition Season Becoming More Important to Forecasting Presidential Nominations? American Politics Research. 29:283288. Adkins, Randall E., and Andrew J. Dowdle. 2002. The Money Primary: What Influences the Outcome of Presidential Nomination Fundraising? Presidential Studies Quarterly. 32: 256-275. Adkins, Randall E., and Andrew J. Dowdle. Forthcoming. Overcoming Pitfalls in Forecasting Presidential Nominations Atkeson, Lonna and Cherie Maestas. 2003. Racing to the Front: The Effect of Frontloading on Presidential Primary Turnout. Paper presented at the Annual Meeting of the American Political Science Association. Philadelphia, PA. Arterton, Christopher F. 1984. Media Politics: New Strategies of Presidential Campaigns. Lexington, MA: D.C. Heath. Bartels, Larry M. 1985. Expectations and Preferences in Presidential Nominating Campaigns. American Political Science Review 79:804-815. Bartels, Larry M. 1988. Presidential Primaries and the Dynamics of Public Choice. Princeton, NJ: Princeton University. Beachler, Donald. 1996. “The South and the Democratic Presidential Nomination, 1972-1992.” Presidential Studies Quarterly, 26: 402-414. Berg, Joyce, Robert Forsyth, Forrest Nelson and Thomas Rietz. 2000. “Results from a Dozen years of Election Futures Markets Research,” working paper. 29 Berg, Joyce, Forrest Nelson and Thomas Rietz. 2003. “Accuracy and Forecast Standard Error of Prediction Markets.” working paper. Brady, Henry E. and Richard Johnston. 1987. What’s the Primary Message: Horse Race or Issue Journalism. In Media and Momentum, edited by Gary R. Orren and Nelson W. Polsby, 104-126. Chatham, NJ: Chatham House. Brams, Steven J. 1978. The Presidential Election Game. New Haven: Yale University Press. Burden, Barry C. 2002. "United States Senators as Presidential Candidates." Political Science Quarterly. 117: 81-102. Cohen, Bernard C. 1963. The Press and Foreign Policy. Princeton, NJ: Princeton University Press. Cohen, Marty, David Karol, Hans Noel, and John Zaller. 2002. Beating Reform: The Resurgence of Parties in Presidential Nominations, 1980-2000. Paper presented at the Annual Meeting of the American Political Science Association. Boston, MA. Cohen, Marty, Hans Noel and John Zaller. 2003. Constant Rules, Changing Game: a General Model of Presidential Primaries, 1972 to 2000. Paper presented at the Annual Meeting of the American Political Science Association. Philadelphia, PA. Damore, David F. 1997. A Dynamic Model of Candidate Fundraising: The Case of Presidential Nomination Campaigns. Political Research Quarterly 50(2): 343-364. Dowdle, Andrew J., Randall Adkins and Wayne P. Steger. 2004. Gaining Traction: How Do Frontrunners Surface Before the Presidential Primaries? Paper presented at the annual meeting of the American Political Science Association, Chicago, IL Goldstein, Joel. H. 1978. The Influence of Money in the Pre-Nomination Stage of the Presidential Selection Process: The Case of the 1976 Election. Presidential Studies Quarterly 8:164-179. Grush, Joseph E. 1980. Impact of Candidate Expenditures, Regionality, and Prior Outcomes on the 1976 Democratic Presidential Primaries. Journal of Personality and Social Psychology 38(2):337-347. Gurian, Paul-Henri. 1993. Candidate Behavior in Presidential Nomination Campaigns: A Dynamic Model. Journal of Politics 55(1):115-140. Gurian, Paul-Henri., 1986. Resource Allocation Strategies in Presidential Nomination Campaigns. American Journal of Political Science 30:802-821. Gurian, Paul-Henri. 1990. The Influence of Nomination Rules on the Financial Allocations of Presidential Candidates. Western Political Quarterly 43:661-687 30 Gurian, Paul-Henri. 1996. Winnowing: The Strategy of Eliminating Opponents in Presidential Primaries. Paper presented at the Annual Meeting of the American Political Science Association, San Francisco, CA, August 28-September 1. Gurian, Paul-Henri and Audrey A. Haynes. 1993. Campaign Strategy in Presidential Primaries, 1976-1988. American Journal of Political Science 37:335-341. Hanson, Gordon S. 2000. Out of Cash: U.S. Presidential Prenominations. Party Politics 6:47-59. Haynes, Audrey A., Paul-Henri Gurian, and Stephen M. Nichols. 1997. The Role of Candidate Spending in Presidential Nomination Campaigns. Journal of Politics 59:213-225. Haynes, Audrey A., Paul-Henri Gurian, Michael H. Crespin, and Christopher Zorn. 2004. The Calculus of Concession: Media Coverage and the Dynamics of Winnowing in Presidential Nominations. American Politics Review 32(3):310-337. Iyengar, Shanto, Mark D. Peters, and Donald Kinder. 1982. Experimental Demonstrations of the Not-So-Minimal Political Consequences of Mass Media. American Political Science Review. 76: 848-858. Iyengard, Shanto, and Donald Kinder. 1987. News that Matters: Television and American Opinion. Chicago: University of Chicago Press. Kiewiet, D. Roderick, and Langche Zeng. 1993. "An Analysis of Congressional Career Decisions." American Political Science Review. 87: 928-941. Lichter S. Robert, Stanley Rothman, and Linda S. Lichter. 1986. The Media Elite: America’s New Power Brokers. Bethesda, MD: Adler and Adler. Lichter S. Robert, Daniel Amundson, and Richard Noyes. 1988. The Video Campaign: Network Coverage of the 1988 Primaries. Washington DC: American Enterprise Institute. Mayer, William G. 1996a. Forecasting Nominations. in In Pursuit of the White House: How We Choose Our Presidential Nominees, ed. William G. Mayer. Chatham, NJ: Chatham House. pp. 44-71. Mayer, William G. 1996b. The Divided Democrats: Ideological Unity, Party Reform, and Presidential Elections. Boulder, CO: Westview. Mayer, William G. 2003a. Forecasting Presidential Nominations or, My Model Worked Just Fine, Thank You. PS: Political Science and Politics 36(2):153-157 Mayer, William G. 2003b. The Basic Dynamics of the Presidential Nomination Process: Putting the 2000 Races in Perspective. Presidential Studies Quarterly 33(1):72-100 Mayer, William G. and Andrew E. Busch. 2004. The Front-Loading Problem in Presidential Nominations. Washington, DC: Brookings. 31 McCombs, Maxwell E. and Donald L. Shaw. 1972. The Agenda-Setting Function of Mass Media. Public Opinion Quarterly. 36: 176-87. McCombs, Maxwell E., and Donald L. Shaw. 1993. The Evolution of Agenda-Setting Research: Twenty-five Years in the Marketplace of Ideas. Journal of Communication. 43: 58. McCombs, Maxwell E., and Tamara Bell 1996. The Agenda-Setting Role of Mass Communication. In Michael B. Salwen and Don W. Stacks (eds.). An Integrated Approach to Communication Theory and Research. Hillsdale, NJ: Erlebaum. Norrander, Barbara. 1993. Nomination Choices: Caucus and Primary Outcomes, 1976-1988. American Journal of Political Science 37:343-364. Norrander, Barbara. 2000. The End Game in Post-Reform Presidential Nominations. Journal of Politics 62:999-1013. Patterson, Thomas E. 1980. The Mass Media Election. New York: Praeger. Patterson, Thomas E. 1993. Out of Order. New York: Alfred A. Knopf. Peabody, Robert L., and Eve Lubalin. 1975. "The Making of Presidential Candidates." In Charles W. Dunn, (ed.). The Future of the American Presidency. Morristown, NJ: General Learning Press. pp. 26-65. Peabody, Robert L., Norman J. Ornstein, and David W. Rhode. 1976. "The United States Senate as a Presidential Incubator: Many are Called but Few are Chosen." Political Science Quarterly. 91: 237-258. Poole and Rosenthal. 1997. Congress: a Political Economic History of Roll Call Voting. Poole, Keith, Howard Rosenthal, and Nolan McCarty. 1997. Income Redistribution and the Realignment of American Politics. Washington DC: AEI Press. Robinson, Michael J., and Kevin R. Appel. 1979. "Network News Coverage of Congress. Political Science Quarterly. 94: 407-418. Schlesinger, Joseph A. 1966. Ambition and Politics: Political Careers in the United States. Chicago: Rand McNally. Schlesinger, Joseph A. 1975. "The Primary Goals of Political Parties: a Clarification of Positive Theory." American Political Science Review. 69: 840-49. Steger, Wayne P. 2000. Do Primary Voters Draw from a Stacked Deck? Presidential Nominations in an Era of Candidate-Centered Campaigns. Presidential Studies Quarterly 30:727-753. Steger, Wayne P. 2000b. Political Parties Still Matter: Endorsements in Presidential Nominating Campaigns. MWPSA. Chicago. April, 2000. 32 Steger, Wayne, P. 2002a. A Quarter Century of Network News Coverage of Candidates in Presidential Nomination Campaigns. Journal of Political Marketing. 2002. 1(1): 91-116 Steger, Wayne P. 2002b. “King-Makers in Presidential Nomination Campaigns, 1992-2000,” Southern Political Science Association, Savannah, GA. Steger, Wayne P. 2003. Presidential Renomination Challenges in the 20th Century. Presidential Studies Quarterly. 33: 827-852. Steger, Wayne P., n.d., The US Senate: Stepping Stone or Tombstone for Presidential Ambition? Steger, Wayne P., and Anne Davis. 2000. “Candidate Endorsements in the 2000 Presidential Nominating Campaign.” Paper presented at the Southern Political Science Association. Atlanta. Steger, Wayne P. 2003. Presidential Renomination Challenges in the 20th Century. Presidential Studies Quarterly. 33: 827-852. Steger, Wayne P., Andrew J. Dowdle, and Randall E. Adkins. 2004. The New Hampshire Effect in Presidential Nominations. Political Research Quarterly. 57 (3):375-390. Zaller, John. 1992. The Nature and Origin of Mass Opinion. Cambridge, England: Cambridge University Press. 33
"The Presidential Primary Vote"