Perceptions of Candidate Viability: Media Eects During the Presidential Nomination Process Philip Paolino The University of Texas at Austin Abstract One way that the media inuence the presidential nomination process is through focusing upon the horse-race (e.g. Patterson 1980; Robinson and Sheehan 1983). A number of studies have shown that voters' preferences are aected by their perception of candidate viability (e.g. Abramson, Aldrich, Paolino and Rohde 1992; Bartels 1988; Brady and Johnston 1987). This means that the media's ability to communicate information about candidate viability will have a great eect upon the outcome of the nomination process. Accordingly, we need to know just how strongly the media inuence voters' perceptions about candidate viability in order to better understand the media's eect upon the nomination process. In this paper, I will examine how the media inuence voters' perceptions of candidate viability with respect to both the direction and the clarity of perceptions. The importance of this is research is to help our understanding of the factors that help fuel candidate momentum during the nomination process. Paper prepared for presentation at the Annual Meeting of the Midwest Political Science Association, Chicago, IL, April 1996. Any comments can be sent to me at the Department of Government, Burdine Hall 536, The University of Texas at Austin, Austin, TX 78712-1087 or e-mailed to ppaolino@jeeves.la.utexas.edu. This paper is also available at http://pc11.gov.utexas.edu. In the literature on the nomination process, there is widespread agreement that candidate viability is critical to candidates' ability to win the nomination (e.g. Aldrich 1980; Bartels 1988; Brady and Johnston 1987). Candidates who the public perceives as doing well, are more likely to attract votes (e.g. Abramson et al. 1992), media attention (e.g. Robinson and Sheehan 1983), and nancial contributions (e.g. Paolino 1995). Without question, the media play a crucial role in disseminating information about candidate viability. Consequently, we need to understand how the media aect the mass public's perceptions in order to understand, rst, the inuence of the media in the nomination process, and, second, the factors that help shape candidate viability and momentum during the process. For the rst part, there are two ways that we can estimate the media's impact. First, the media can inuence the direction of the public's perceptions. The media, however, can also inuence the certainty with which the public holds those perceptions. The direction is probably the more important (and certainly the focus of much work on the media in the nomination process), but the clarity with which the public receives these signals from the media is also critical because it might tell us something about the stability of voters' perceptions. Whatever we can learn on this rst question will surely help us better understand the nature of momentum during the primaries. The signicance of the media's role in providing information about candidate viability is that we can better understand the process by which candidates gain and lose momentum. The media can clearly inuence momentum by providing signals about the direction in which a candidate is going in trying to win the nomination. The clarity of such signals, however, is also important because momentum swings may be smaller if voters are more certain about the direction in which candidates are moving. By contrast, if the media do not provide strong signals that reduce voters' uncertainty about candidates' likely success, then relatively small shifts in candidate performance can produce much larger swings in momentum that would otherwise by the case. Without question, the strength of the media's signals on both dimensions is going to vary across candidates. Where the media are not clear on a candidate's chances of receiving the nomination, we should not expect that the media will aect public's perceptions of candidate viability with respect to either the direction or certainty of candidate viability. In other instances, the media's signal should be loud and clear, but only to members of the public who are attentive to the media's campaign coverage. Finally, it is possible that the media, in some instances, send a message that is so strong that even casual attention is sucient to pick up the signal. The public's awareness of the nomination process, however, is probably not suciently high that casual attention will be enough to take in the media's signals.1 The major focus of this paper, then, is to examine the degree to which the media inuence the direction and clarity of the public's perceptions about candidate viability in the nomination campaign. In the rst part of this paper, I'll briey review the literature on the media's role in the nomination process. Then, I'll make an argument about the relationship between the media's role in the campaign and how they shape public perceptions of viability. Third, I'll present models showing the eect of the news media on both the direction and the clarity of the public's perceptions of candidate viability. Finally, I'll discuss the implications of these ndings for the way that we view the role of the media in our understanding of the nomination process. 1 Data from the 1988 NES Super Tuesday study reveals that slightly more than 40% of the respondents who were asked could correctly identify the winner of the Republican contests in Iowa and New Hampshire and fewer than 35% could correctly name the winners of the Democratic contests. 1 The News Media in the Presidential Nomination Process There is a fairly well-developed literature on the role of the news media in the presidential nomination process (e.g. Arterton 1984; Robinson and Sheehan 1983). This literature clearly shows that the press's treatment of the candidates is not equal. Robinson and Sheehan (1983) argue that the media divide the nomination eld into three tiers: the likely, the plausible, and the hopeless candidates. The likely and plausible candidates are the ones who stand the best and some chance, respectively, of winning the nomination. These candidates receive the most space in both the print and electronic media. Robinson and Sheehan (1983) write that the coverage extended to the third class of candidates, the hopeless, can be described as sparse and gloomy. The characteristic feature of the coverage that the hopeless candidates receive is called the \death watch." The death watch is the coverage that chronicles these candidates' imminent exit from the eld. Arterton (1984) argues that journalists have incentives to use these expectations about likely candidate success to structure their coverage accordingly. In this, the likely and plausible candidates, to use Robinson and Sheehan's (1983) terminology, are more likely to have reporters assigned to follow their campaign, a method that Arterton (1984) calls \man-to-man" coverage, while the hopeless candidates are, at best, able to pick up only random coverage from the reporter covering the state (\zone coverage" (Arterton 1984)) or from the local media in Iowa and New Hampshire (Buell 1987). The consequence of using such a means of coverage is that the likely and plausible candidates get much more coverage than the hopeless candidates. Elsewhere (Paolino 1995), I have argued that the relationship between the media's allocation of coverage based upon candidates' standing in the polls increases in the time leading up to the rst contests. This work indicates that the media rely heavily on the polls for allocating candidate coverage during the primary and pre-primary seasons.2 The way that the media allocate coverage, naturally, has a strong eect upon the amount of information that the public has about the candidates (Paolino 1995). By itself, this means that the media have an important eect upon the nomination campaign because they mostly control the ow of information about the candidates.3 What the literature on the media do not address, however, is the degree to which the public's perceptions of candidate viability are inuenced by the process of the media's coverage. It is one thing to say that the media both rely and focus upon the horse-race for making decisions about the allocation and subject of coverage. It is another, however, to say that the public are attentive to these signals from the media. To answer this question, there is a body of work that examines the eects of the media upon public perceptions of candidate viability. Patterson (1980) argues that his Erie and Los Angeles data show a relationship between media usage and a change in the respondents' perception of candidates' viability in the \correct"direction in the 1976 nomination campaign. Similarly, Brady and Johnston (1987) argue that average viability scores move in concert with the coverage that candidates receive in their dataset of stories from the UPI wire. Finally, Bartels (1988, 50) shows that there were times in the 1984 campaign, specically during the period right before the New Hampshire primary, when high users of the media picked up on Hart's improving chances more quickly than low media users.4 There are some diculties, however, with these data for our understanding of how the media inuence the public's perceptions of candidate viability. In the cases of Patterson (1980) and Brady and Johnston (1987), the relationships are measured by means that are, at best, indirect. The change in the public's perceptions of candidates' viability may move with the media's coverage of I dene the pre-primary season as the 5-month period preceding the New Hampshire primary. One exception to this is candidates who have millions of dollars to spend and are not bound by spending limits. Bartels (1988) points out, however, that media usage does not explain the bulk of respondents perceptions, and his results in Table A.8 show that media usage does not have a statistically signicant eect upon voters' evaluations of Hart's chances at winning the nomination. 2 3 4 2 the candidates' viability, but these eects may be working only through changes in the public's attitudes toward the candidates. For instance, we know that the media's overall allocation of candidate coverage moves in concert with the candidates' performance in the polls. As the public obtains more information about these candidates, their preferences may shift in favor of these candidates just because of simple increases in candidate name recognition. Bartels (1988) shows quite clearly, however, that people's preferences toward candidates strongly inuence their perceptions of candidates' viability. So, these correlations between the media's coverage of the candidates' viability and the public's perceptions of candidate viability may be mediated through preferences. Bartels (1988) avoids this problem by estimating the eect of media usage in a model that includes an estimate of the respondents' preferences for the candidates. This means of analysis, however, still ignores the degree to which the the media inuence respondents' certainty about the candidates' viability. Furthermore, the 1984 data upon which Bartels's (1988) analysis rests does not include a variable measuring whether or not the respondent is aware of which candidates won the Iowa caucuses and New Hampshire primary. Changes in the public's perceptions of viability may be more strongly related to candidates' actual performance. The knowledge of this will almost certainly be related to media usage, but the implications for the formation of perceptions are dierent. The media may highlight certain features of an outcome in ways that provides dierent information than the outcome itself. For instance, a voter who sees that Mondale defeated Hart 45-16% in Iowa may conclude that Mondale won a stunning victory, while another voter paying attention to the media's interpretation of the outcome may conclude that Hart did surprisingly well. This is not to say, of course, that the media's interpretation of the results will necessarily deviate from an \unbiased" view. This paper lls in these gaps and explores the hypothesis that the media's coverage of the nomination campaign sends dierent signals based upon a candidate's status in the eld. I expect that the media signals will be positive and clear for the likely candidates, less clear for plausible candidates, and negative and clear for the hopeless candidates. Examining the Clarity of the Media's Signals In considering the clarity of the media's signals to the public, we have to understand the indicators we have for clarity. In the most fundamental sense, clarity can be examined by looking at the variance of the model for candidate viability. If the media have an eect upon the formation of perceptions of candidate viability, then we should expect to see heterogeneity in these models related to media usage. It may not, however, be this simple. If the media send a clear and unambiguous signal about candidate viability, we might expect that everyone in the population, even people with the most casual contact with the media, to be able to pick up this signal. In this case, there is no reason to expect any relationship between media exposure and the variance of the model. The same would be true in cases where the media send absolutely no signal about candidate viability. In fact, imagine if the media didn't have any coverage of candidate viability. In this case, we would also predict that there would not be any relationship between media exposure and the variance of voters' perceptions. There is a fundamental diculty, therefore, in distinguishing between these two cases with respect to evaluating the media's inuence. In the nomination process, it is unlikely that the lack of a relationship between the media usage and the variance of candidate viability reects either the complete lack or utter pervasiveness of media signals. The attention that the media pay to the horse-race is well-documented (e.g. Brady and Johnston 1987; Popkin 1991; Robinson and Sheehan 1983). For example, in a content analysis of newspaper stories during the 1988 pre-primary period (cited in Buell 1991), 229 of the 1062 stories on George Bush dealt with Bush's position in the horse-race. Even a less well-covered candidate 3 like Kemp received 59 of 621 stories on his place in the horse-race. At the same time, it is not likely that attention to the media is so pervasive among voters that even trace attention is sucient to determine with clarity a candidate's chances of winning the nomination. As noted above, fewer than 50% of the respondents interviewed in the Super Tuesday study after the Iowa caucuses were able to correctly identify Dole as the winner. There are additional reasons to suspect that casual media users could obtain a clear signal regarding candidates' viabilities. For example, an examination of Buell (1991) data, which includes a measure of the stories' evaluation of the candidates' national standing in polls (scaled from 1 to 7), shows that the standard deviation for most candidates is greater than 1, suggesting a non-trivial amount of variation in the media's assessment of candidates' viability. Only for George Bush did all of the stories agree that he was the front-runner. And, this dataset, unlike the one that I use in this paper's analysis, does not include media coverage following the Iowa caucuses, when evaluations of Bush's viability may have become less certain. Beyond the question of how to evaluate the direction and clarity of the media's inuence, another consideration how to measure individuals' perceptions of candidate viability. The 1988 NES Super Tuesday study provides 100-point thermometer scores of responses to the question, Now, thinking about these nominating conventions, who do you think is likely to win the Democratic nomination for President. We will be using a scale which runs from 0 to 100, where 0 represents no chance for the nomination, 50 represents an even chance, and 100 represents certain victory. You may use any number between one and one hundred. What do you think ALBERT GORE's chances are? There are, of course, similar questions for all 13 candidates in the Democratic and Republican elds. A potential problem in using this measure is that it is likely that there are scaling dierences across respondents. Some people may take \no chance" and \certain victory" as meaning quite dierent things. And while models of sophisticated voting may depend upon knowing the dierences in a voters' assessment of candidates' probability of winning, it probably suces that voters can place candidates in the likely order of nish. This means that there could be some substantive problems in using the raw 100-point score for this analysis. The alternatives, however, are not without their own problems. One possibility is to transform the thermometer scores into probabilities by taking the raw thermometer score for each candidate and dividing that score by the sum of the scores for all candidates in that party. In other words, Viabilityi;j = Pkthermometer scorei;j (1) j =1 thermometer scorei;j for the thermometer scores of only the candidates that a respondent can place. This measure has a value in that a respondent's answers are eectively translated into the probabilities, which reect the intent of the question, but may not be reected in respondents' answers. The problem of using such a measure in this case, however, is that the transformed probability is likely to get smaller as the respondent can place more candidates on the viability scales. Part of this has to do with respondents not thinking in probabilistic terms, but could occur even if respondents did think perfectly in these terms. Imagine, for instance, a respondent who thought that Bush had an all-but-certain chance of winning the nomination, and so placed him at 99 on the scale. If that same respondent placed only one other candidate, who he thought had almost no chance, at 1, this scheme would work out (as would using the raw scores). Another respondent with the same beliefs, however, would have several other candidates that he also placed at 1. Using this kind of normalized score would then be partly a function of the number of candidates a respondent could rate. In other words, the addition of 4 candidates who are perceived to have little chance of winning will eectively lower the scores of the more viable candidates as compared to the scores of the same candidates for respondents who don't have enough information to evaluate the viability of these candidates with little chance of winning. Another means of normalizing the viability scores would be to give the highest placed candidate(s) a score of 1, the lowest placed candidate(s) a score of 0, and place the other candidates accordingly in between. This measure would be as follows: thermometer scorei;j , thermometer scorei;min Viabilityi;j = thermometer (2) scorei;max , thermometer scorei;min Unfortunately, this measure is also vulnerable to problems related to the number of candidates that a respondent can rate. Unless a respondent can place at least three candidates at three dierent places on the thermometer, then the more highly rated candidates receive a score of 1 and the less highly rated candidates receive a score of 0. Respondents not able to do this are likely to show increased variance because the measure places their assessments at the extremes. Furthermore, this measure is susceptible to problems with respondents who place a small number of candidates fairly closely together as small dierences can become magnied greatly. Because of the various diculties with the dierent measures, the analysis will be done using just the raw measure. While using this measure made add variance related to error resulting from inter-respondent scaling dierences, this type of error should be random and, therefore, is preferable to error that is related to some of the primary independent variables, political information and media usage. Estimating Media Eects Estimating the eects of media exposure upon the direction and clarity of public perceptions of viability is not a dicult task, but it is one with which many people may not be familiar. If we were only estimating the direction of the media's eect upon perceptions, we could simply use ordinary least squares, where the viability score was a function of some set of independent variables, which included media usage. Estimating the variance function is simply a matter of specifying the \error" as the function of some other set of independent variables and estimating the model through maximum likelihood using the normal distribution.5 Using King's (1989) notation, fn (yj; 2 ) says that y is a normally-distributed variable with mean, , and variance, 2. In estimating the model, = Xi , simply the linear specication one would use for OLS, and 2 = exp(Xi ), where the exponential function is used to insure values greater than 0. Specically, the \direction" portion of the model to be estimated is as follows: d i;j + 2 Best in Statei;j + E (Viabilityi;j ) = i = + 1 Preference 3 Political Informationi + 4 Television Exposurei + 5 Newspaper Exposure i + 6 Interviewed Before Iowai : (3) \Preference," as denoted by the hat, is the predicted feeling thermometer score for each candidate. Bartels (1985) shows quite clearly that there is a reciprocal relationship between candidate preference and candidate expectations. Using the predicted values of preference from instrumental variables simply deals with the simultaneity problem.6 \Best in State" is a dichotomous variable indicating 5 In least squares terminology, the model simply assumes that the model has heteroskedastic errors that can be modeled as the function of independent variables. 6 The estimates generated for each candidates from instrumental variable models are available from the author upon request. The general model, as well as the construction of all of the independent variables, is presented in the appendix. 5 whether or not the respondent felt that candidatej was the candidate most likely to win his party's contest in the respondent's state. Both of these variables reect the eect of individuals' candidate assessments upon individual assessments of candidate viability. \Political Information" is a 5-point scale that is a summary index of each respondent's ability to accurately place Ronald Reagan, George Bush, Gary Hart, and Jesse Jackson on ideological scales (cf. Luskin 1987).7 A major problem with the 1988 Super Tuesday survey is a total absence of variables for measuring respondents' political information. The only objective measures of political information are whether or not the respondent can correctly identify each party's winners in the Iowa and New Hampshire contests. Unfortunately, the direct measures of this information is available for only part of the sample, and these measures are also domain specic to the question being investigated. The scale that I have chosen to use is not without problems because three of the four political gures are candidates in the 1988 race { and, so this measure may not be totally free of domain specicity { but, I think this problem is minimized to the extent that all three candidates were well-recognized going into the 1988 race. As such, information about these candidates is more likely to reect a general level of political information than any other measure available in the survey. Political information is expected to be related to viability for some candidates in that people with a more sophisticated view of the political process may be better able to see that some candidates are more viable than others, apart from these candidates' current positions in public opinion polls. For instance, while Gary Hart re-entered the 1988 race with high levels of expressed support, more sophisticated observers of politics should have been more skeptical of his long-term viability. Certainly, a similar argument should hold for Pat Robertson, even after his strong showing in Iowa. \Television Exposure" and \Newspaper Exposure" are the primary variables of interest in this study. Both variables are 7-point scales that take into account a respondent's frequency of media use and his interest in politics during exposure. As mentioned above, both variables are expected to have positive signs for the likely candidates, in 1988, Bush, Dole, and to some extent, Dukakis, while they should have negative signs for the hopeless candidates, Jackson, Robertson, and Hart. Similar expectations hold for both variables, as well as for the political information variable. \Interviewed Before Iowa" is a dichotomous variable that indicates whether or not the respondent was interviewed before or after the results of the Iowa caucuses were known. On one level, respondents interviewed later in the season should have more information about the candidates with which to make a more accurate judgment about viability. In this case, the direction of viability should be related along the same lines as for media use and political information. Furthermore, this variable provides a means of examining how actual, \hard" results aect voters' perceptions as compared with voters who have only \soft" indicators with which to make assessments of viability. The \clarity" portion of the model is specied as follows: V (Viabilityi;j ) = i2 = exp( + 7 Political Informationi + 8 Television Exposure i + 9 Newspaper Exposurei + 10 Interviewed Before Iowai + 11Viability Dierence i): (4) The variables in equation 4 are the same as those in equation 3, with the exception of \viability dierence." This variable is simply the dierence between the score for the candidate in each party who the respondent thinks is the most and least likely to win. This variable is included to try to account for inter-respondent scaling dierences. It is expected that the variance will be positively related to this dierence. 7 This variable, like all others in the analysis, with the exception of the predictedevaluation score for each candidate, was re-scaled to a 0 to 1 range. 6 Results Contrary to expectations, the results of the model, Tables 1 and 2, do not provide very strong evidence for media eects upon either the direction or the clarity of the public's perceptions of candidate viability. The results in Table 1 indicate that media exposure aected both dimensions of perceptions for Gary Hart. Newspaper exposure was also signicant for both dimensions for Jesse Jackson. Controlling for other factors, the highest level of of media expsoure placed Hart more than 13 points lower than the lowest level of media users. Looking at the variance function, the the variance for someone with the highest level of media usage, holding the other variables at their mean, is .027, as compared to a respondent with mean levels of media usage, .034 { a 21% reduction in the variance. Otherwise, there is no other candidate, including the Republican candidates, for whom media usage has a comparable eect upon the variance. Looking at the other variables, we see that candidate evaluation has a large eect upon the direction of respondents' perceptions of candidates' viability, corroborating Bartels's (1985) ndings. Similarly, if the respondent thinks that the candidate is most likely to win the primary in the respondent's state, then the respondent's perception of the candidate's viability also increases.8 Political information generally has the expected signs, although the statistical signicance of these eects are limited to a few candidates. Admittedly, these results may also bolster the media eects hypotheses because some authors have found that the best predictor of media exposure is political information (Price and Zaller 1990), but even if this is so, the results are not so strong as to counterbalance the limited eects for the media use variables. The parameter in Tables 1 and 2 that stand out are the estimates for time of interview, especially as concerns the direction of respondents' perceptions. In Table`1, respondents interviewed after the Iowa caucuses (and, many of them the New Hampshire primary), perceived the average chances of a Dukakis and Gephardt nomination more likely and the chances of a Simon and Hart nomination less likely than respondents interviewed before either of the two contests. Gephardt beneted particularly, with a dierence of more than 12 points for the two sets of respondents. The results are in the expected direction for the three candidates who had the most riding on the rst two contests, Gephardt, Simon, and Dukakis. The results for Hart probably indicate a drop in viability related to respondents learning about other viable alternatives as much as Hart's lack of success in either of the early contests. It is also interesting to note that there was no statistically signicant dierence between the two sets of respondents for Gore and Jackson, the two candidates who postponed their \debut" until Super Tuesday. Finally, the certainty concerning Dukakis's, Gephardt's, and Hart's chances also increased in the period following the rst contests, with the eect statistically signicant for the latter two. For Gephardt, the reduction in variance, holding all other variables at their means, was from .046 among respondents interviewed prior to the Iowa caucuses to .032 among respondents interviewed after Iowa, a 30% reduction in the uncertainty regarding his prospects attributable to voters simply having some \hard" data to go on. In other words, voters interviewed after Iowa felt more positive and more certain about Gephardt's chances of winning the nomination than respondents interviewed prior to Iowa. For Hart, losing in Iowa and New Hampshire also increased certainty about his chances, but in a negative direction. For the Republicans, the results are quite similar to those for the Democratic candidates. With respect to media eects, the results in Table 2 show that television exposure had some eect upon the public's perceptions of Dole and Robertson, but contrary to expectations, newspaper exposure only increased uncertainty about Robertson's chances of winning the nomination. Political information 8 Undoubtedly, some of this eect is indirectly related to candidate evaluation since it seems likely that people who prefer one candidate to the others are more likely to think that the candidate will win the respondent's state. 7 had some eect upon respondent's perceptions, as more politically sophisticated respondents were more likely to downplay both Kemp's and Robertson's chances of winning the nomination, and the more politically sophisticated were also more certain on Bush's chances of winning the nomination than the less sophisticated. Again, the variable that stands out in Table 2 is the time of interview. The direction of respondent's perceptions about candidate viability changed in the expected direction, with Dole and Robertson gaining from their performances in Iowa, Bush losing slightly from his third-place showing, and Kemp losing somewhat substantially from his lack of success. It is worth noting that the eect on the direction of perceptions of the before and after Iowa interviewees is smallest for Bush, and the variable is signicant only at p < :1. Unquestionably, some of this limited eect might be related to the fact that he \went 1-1" in the rst two contests, but Bush was probably also helped that, as the well-known front-runner in this race, expectations about his performance were less likely to be strongly tied to performance in either of the earlier contests. In other words, if Bush had done well, that would merely serve to conrm his front-runner status, while it would take more than just one loss to fatally wound him. In other words, it seems as though well-known front-runners get more of a \cushion" in the earlier races than lesser-known candidates. In races without any well-known front-runner, the eect of \hard results" is much greater because voters' expectations about who is likely to win are less well formed. Consequently, the eects of momentum are likely to be greater in these kinds of races than in ones with a well-known frontrunner. This result is consistent with arguments I have made elsewhere (Paolino 1995). Admittedly, there are several objections that could be made to these conclusions. First, saying that there is a dierence between respondents interviewed before and those interviewed after the Iowa caucuses does not mean that the results from these contests had any direct eect upon voters. Rather, in line with my critiques of Patterson's (1980) and Brady and Johnston's (1987) conclusions, these eects may be simple artifacts of the increased coverage that the successful candidates received and the reduced coverage for the less successful candidates. I will address this objection in the next section. A second objection goes back to the idea that the media can have very strong eects upon people's perceptions, such that even casual media users can detect the signal. Along these same lines, it is also possible that using the raw scores creates measurement error that attenuates these results. To address this concern, I propose that we look at respondents' perceptions of Bush's viability with respect to the candidate a respondent rated as the most viable. Certainly, if the media have any eect in this manner, it should occur with Bush who, as I pointed out earlier using Buell's (1991) data, was clearly the consensus front-runner in the print media.9 This analysis indicates that 83% of the respondents in the Super Tuesday study interviewed before Iowa (when we have data on the print media's coverage) who could place at least 2 Republican candidates felt that Bush had the best chance of winning the nomination. This is certainly a large percentage, but not overwhelming, considering the strength of the print media's assessment of his position in the eld. Furthermore, of the respondents who scored less than .33 on both indicators of media usage and who could place Bush and at least one other Republican candidate on the viability thermometers, 90% placed Bush in the lead (N=68), compared with 83% of the respondents who scored highest on at least one measure of media usage and could place Bush and at least one other Republican on the viability thermometers (N=217). These results do not provide support for the criticism that the lack of media eects is simply a product of a media that sent such strong signals that even casual users of the media were able to pick up such signals. In fact, high media users were slightly less likely to accept the media's evaluation of Bush's front-runner status than casual users. 9 Of course, I cannot make any denitive conclusions if this held for the electronic media, but Arterton (1984) has pointed out that the electronic media often takes cues from the \prestige press" sources that Buell (1991) examines. 8 If anything, the media did a better job of creating doubts about Bush's chances than of portraying him as invincible. Another objection is that the model that I am using to estimate the eect of the media could be biased because of selection biases. Certainly, high media users are more likely to be able to place the candidates on viability scales than low media users. Unless that selection process is taken into consideration, the results may understate the eect of the media upon the formation of the public's perceptions about candidate viability. There is a certain amount of diculty in specifying an identied model to account for selection biases because many of the same factors that are likely to inuence information about perceptions are the same factors that are likely to inuence whether or not respondents have any information with which to place the candidates on the viability thermometers. I believe that it is because of this identication problem that I was only able to obtain estimates for four of the candidates, Bush, Dole, Dukakis, and Gephardt.10 The results of this analysis, Table 3, indicate that selection biases do not appear to have a great eect upon the conclusions from earlier in the paper.11 In the cases of Dole and Dukakis, there is not any evidence to indicate that selection bias was any problem. For Bush and Gephardt, even though the bivariate correlation was signicant, there were not any instances where the parameter values were greatly aected. Certainly, this is not to say that such biases may not be present and, under certain circumstances, do not have any eect upon the results. There is no evidence presently, however, to lead us to suspect that the results in Tables 1 and 2 are biased. The Eects of \Hard Data" on Candidate Perceptions Having seen in the last section that the respondents interviewed after the Iowa caucuses had assessments of candidate viability that reected the results of the Iowa and New Hampshire contests, it is time to look at these respondents more closely. The model to be estimated in this section parallel the one specied in equations 3 and 4, with the exception that two dichotomous variables have been included to account for respondents who could correctly identify the winners of the Iowa and New Hampshire in each party. The hypothesis in this section is that voters use \hard data" to change and sharpen their perceptions of candidates' viability. The results are presented in Tables 4 and 5. First of all, the results show that, once information about the winners of the early contests is controlled for, the media have little eect upon either the direction or clarity of individuals' perceptions of candidate viability. But, as expected, respondents who knew which candidates won the Iowa and New Hampshire contests adjusted their perceptions of candidate viability. Starting with Table 4, the rst thing to notice here is that, consistent with the results in Table 1, assessments of Al Gore's viability were completely unaected by the outcomes of the Iowa caucuses and the New Hampshire primary, consistent with the fact that he did not compete in either state. Second, the New Hampshire primary seemed to have more eect upon respondents' perceptions than the Iowa caucuses, a nding that also evidenced in the Republican race (Table 5. This may reect a recency eect, but could also reect the increased general importance of New Hampshire. Candidates who have not been successful in either of the rst two races may be fatally wounded as the race moves to the rest of the country. More specically, Gephardt gained a huge increase in perceived viability from winning in Iowa, but lost some of his momentum from nishing second in New Hampshire. By contrast, Dukakis 10 With Dole, I was not able to get standard errors. For the question of bias, however, we are not necessarily concerned with the standard errors. 11 For brevity sake, the results of the selection model are not shown, but all of the independent variables, political information, television exposure, newspaper exposure, party identication, and time of interview, all had large and statistically signicant eects upon respondent's ability to place the candidates on viability thermometers. 9 lost a slight amount by not winning in Iowa, but gained much more by winning in New Hampshire. Unexpectedly, individuals' perceptions of Simon's viability was not hurt by losing to Gephardt in Iowa, although it was certainly the more important of the rst two contests for Simon. Perhaps, this reects a feeling that a potentially competitive candidate is not mortally wounded until he's lost in both of the rst two contests. Finally, decreased perceptions of Jesse Jackson's viability are seen among people who knew which candidate won Iowa, not New Hampshire. This result suggests that perceptions of Jackson's viability were related to his name recognition advantage over the other \dwarfs" (which, included Bruce Babbitt and Joe Biden), but that advantage faded once voters had one candidate who distinguished himself from the rest of the eld. The eects of knowing who won the early contests were not as strong on the variance function as on the expected value of candidates' viability. Certainly, respondents who knew who won both of the early contests knew that Gary Hart was dead and buried. Otherwise, the only eect of knowledge of the early contests that reduced respondents' uncertainty was Michael Dukakis. With the exception of Hart, Dukakis was also the only candidate for whom media usage inuenced certainty about perceptions of viability, but the eects for television and the print media worked in opposite directions. Taking just the eect of knowledge that Dukakis won in New Hampshire, we can see that, holding the other variables at their means, the eect of New Hampshire was to reduce the variance for Dukakis from .035 to .024, a 31% reduction in uncertainty related to this \hard data." If we examine the eects of media usage, both signicant at p < :1, the reduction in variance using the same settings, only changing media usage from the mean to the maximum values, the reduction in variance is still from .035 to .024. The media values taken together, therefore, did not have much eect. If we look at the results for the Republicans in Table 5, we see that the results generally corroborate those for the Democrats. Again, media usage did not have much eect on respondents' perceptions, with the sole exception of Pat Robertson, who may have beneted greatly in the media, with however little real justication, from his surprising performance in Iowa. But again, the results here for the media are consistent with those in Table 4. There are some important dierences, however. The eects of knowing which candidates won the early contests are much smaller for the Republicans than for the Democrats. This probably reects the idea that the presence of a well-known front-runner, in this case, Bush, makes for a more stable contest, where results of the early contests do not have as much eect upon people's perceptions of candidate viability because the eld has a less uid structure. Looking at the estimates in the variance function, increased usage of the print media actually increased the uncertainty around Pat Robertson's prospects for winning the nomination. The results for George Bush, however, are more signicant. People who knew that Bush won in New Hampshire had higher perceptions of Bush's probability of winning the nomination and more certainty that he was ahead in the race. Holding other values at their mean, the variance of viability perceptions for Bush dropped from .028, for respondents who did not know that Bush won in New Hampshire, to .022 for respondents aware of who won New Hampshire, a 21% reduction in the variance. These results can be compared with the results for the other New Hampshire winner. For Dukakis, winning in New Hampshire put him in front, by increasing respondents' perceptions of his viability and reducing the uncertainty, while for Bush the results conrmed has status as the front-runner and likely nominee.12 So far, there has been little support for the hypothesis that media usage has an eect upon perceptions of candidate viability. One last attempt to nd some eects comes from looking at the respondents who were aware of which candidate won the Iowa caucuses in each party. The results in 12 The interpretation that Bush and Dukakis were in dierent positions comes partly from knowledge about their relative advantages over their competitors in the race and from a comparison of the intercepts in the models, where Bush has a much bigger lead over his competition than Dukakis. 10 Tables 6 and 7. Again, the results here do not provide much support for media eects with respect to either the expected value or variance. These results show that we need to focus our attention more on how people use \hard data" to assess candidate viability more than they might use the media's own interpretations of candidate viability. At this point, I should also note that little evidence for media eects were found when just using respondents interviewed before the Iowa caucuses { including estimates for Babbitt, DuPont, and Haig, who all dropped out, and whose measure of viability were dropped from the survey, after Iowa. It does not appear, therefore, that people are much inuenced by the media's assessments in the absence of \hard data." Discussion To say that my big nding in this paper is that voters' perceptions about candidates' viability is aected by the outcomes of the primary contests is likely to create as much excitement as one might get from watching grass grow. But, that's my nding. And, I hope to convince you in this section that it is more exciting for our understanding of the nomination process than watching grass grow. To say that media usage does not have an eect upon voters' perceptions of candidates' viability is not to say that the media do not have any aect upon the nomination process. Rather, if we take it as granted that the media do have an aect upon the nomination process, then this nding tells us more about how the media inuence the process. Without question, media use is related to the ability to correctly identify the winners of the early contests.13 And, without question, the media's allocation of candidate coverage is related to the polls. But, if the media's reliance on the polls to allocate coverage is reected in their analysis of the race, this reection does not appear in voters' assessments of candidate viability. Instead, I think that these results speak to the event-orientation of the media. Robinson and Sheehan (1983) point out that the media can use the polls and the horse-race to maintain a degree of objectivity, but that same desire to maintain objectivity may lead the media to play down the advantages of the front-runners and play up the potential of dark horses. It may be that members of the media have become so wary of dismissing the chances of the next Gary Hart (of 1984, not 1988) that they are reluctant to express their true beliefs about such candidates' realistic chances of winning the nomination. This same event-orientation may also lead the media to place such strong emphasis on discrete events, such as caucuses and primaries, that these events become removed from the larger process. As a result, individual voters base their assessments of candidates' viability on their own wishful thinking and what they observe in these events. Clearly, we know that individual members of the media may occasionally voice clear opinions about which candidates they think will do well in the primary process, but the collective signals from members of the media do not appear to be very strong in this regard before the process begins. It appears that this situation reverses quite dramatically once the process begins. The media send a very strong signal about what the results of individual contests mean, while individual opinions about the larger process disappear. If the media focus upon the results of individual contests, it is easy to understand how momentum can appear from nothing and disappear as suddenly. A media that focuses on the \event-of-the-day" can create great uncertainty among voters about who's ahead. To take the 1996 campaign as an example, what are voters to think when the media present the prospect of a loss in South Carolina as fatal to Bob Dole's campaign, while a victory is the prelude to a sweep of the remaining contests? Was Dole really as close to being out of the race as the media portrayed it? The results in Tables 4 and 5 also indicate that the momentum swings arising from this event-orientation are probably 13 The author will provide the evidence upon request for anyone doubting this. 11 greater in some types of races, those without an established front-runner, than others, those with a well-known front-runner, but that the swings are still produced in both cases. Finally, these results say a great deal about the dierent ways that the media inuence the nomination process. On the one hand, the media provide a disadvantage to some candidates during the pre-primary period by making it dicult for them to get the coverage they need to move up in the polls. On the other hand, the media's heavy emphasis on the early contests provides the conditions under which these same disadvantaged candidates, if they can nd a way to move up in the polls in these states (perhaps by using the local media (Buell 1987)), have a chance to make up for the lack of organization and money caused, in part, by the media's lack of early coverage. This type of inuence probably aects perceptions of viability through the public's preferences for some candidates { particularly, the ones about which they receive the most information. I don't think that this is any conscious move on the part of the media. It would be hard to imagine that thousands of journalists could coordinate this. Rather, it speaks to the media's changing incentives across the course of the nomination campaign. Before the rst contests when the public's interest in the campaign is low, the media have little incentive to devote limited space to candidates they believe have little chance of winning. Once the contest starts and the public's interest is greater, the media's incentives shift to placing all attention on the winners, while winnowing the losers from the eld in what Robinson and Sheehan (1983) call the \death watch." And, whether or not this event-orientation is good for the process or not, it certainly helps us understand how the media have greater inuence in a nomination process that allows momentum than one without, which Bartels (1988) produces quite dierent outcomes. But, that's not to say that the latter system removes the media's inuence over the process. Instead, that inuence just takes a dierent form { focusing upon candidates' resources rather than their performance. If the current trend of frontloading primaries continues, momentum may have less eect upon the process and, correspondingly, the media's inuence will come more in this latter form. Appendix: Measures Used in the Analysis Candidate Preference is the predicted candidate evaluation from the following instrumental vari- ables: education (v541, recoded to less than high school (1 and 2), high school graduate (3), and more than 12 years (4, 5, and 6). The recoded variable is scaled from .33 to 1. N=2052, mean=0.7917, std dev=0.2418); race (v564, recoded to 0 if v564 ne 2, and 1 if v564 eq 2. N=2117, mean=0.0973, std dev=0.2965); ideological dierence (the absolute value of the dierence between the respondent and the candidate on the 7-point ideological scale); the log of the respondent's age (v18); union membership (v547, recoded to non-union member (5) and union member (1). The recoded variable takes on values of 0 or 1. N=2047, mean=0.1079, std dev=0.3104); candidate character is the average of all the trait measures on which a person could place each candidate; born-again (v558, recoded to 1 if \yes" and 0 otherwise. N=2075, mean=0.4381, std dev=0.4954); partyid (v265, the standard 7-point scale. N=2039, mean=0.4774, std dev=0.3468); a series of four dummy variables for respondents living in the south (which means all respondents in the Super Tuesday study not living in Massachusetts, Maryland, Missouri, and Oklahoma), Massachusetts, Missouri, and Tennessee. Best in State is a dummy variable coded 1 if the respondent thought that candidatei would do the best in the respondent's state (v201 for Democrats and v203 for Republicans). Respondents who did not name a candidate from the correct party were coded as missing. Political Information is a 5-pt variable, rescaled from 0 to 1, where a respondent gets one point for each of the following: placing Jesse Jackson to the left of Ronald Reagan on an ideological scale, placing Jackson to the left of Gary Hart, placing Hart to the left of Reagan, and placing Jackson to the left of George Bush. N=2117, mean=0.3900, std dev=0.4014. 12 Television and Newspaper Exposure were both created in similar ways. The frequency of exposure (v255 for newspaper and v258 for television) wa recoded into 0 for 0 days, 1 for 1 or 2 days a week, 2 for 3 or 4 days a week, and 3 for 5 to 7 days each week. The interest in the exposure (v257 for newspaper and v259 for television) was recoded 1 for very little or none, 2 for some, and 3 for quite a bit and a great deal. The frequency and interest variables were multiplied, and scores of 6 were recoded to 5 and scores of 9 were recoded to 6. The nal variable was, then, rescaled from 0 to 1, with 0 reecting someone who reported not watching the news on TV or reading about it in the paper. For television exposure, N=2105, mean=0.6539, std dev=0.3460. For newspaper exposure, N=2105, mean=0.5065, std dev=0.3503. Interviewed Before Iowa is scored 1 if the respondent's interview (v22) was completed before February 8, 0 otherwise. Thermometer Dierence is the dierence between the viability thermometer score for the respondent's most highly rated candidate and least highly rated candidate in each party. N=1897, mean=0.2960, std dev=0.2434 for the Democratic candidates and N=1921, mean=0.2454, std dev= 0.1933 for the Republican candidates. Knows Who Won Iowa/New Hampshire are dichotomous variables scored 1 if the respondent correctly identies the winner of the Iowa (v205 and v207) and New Hampshires (v209 and v211) contests for the respective parties. References Abramson, Paul R., John H. Aldrich, Phil Paolino and David W. Rohde. 1992. \Sophisticated Voting in the 1988 Presidential Primaries." American Political Science Review 86(1):55{69. Aldrich, John H. 1980. Before the Convention. Chicago: University of Chicago. Arterton, F. Christopher. 1984. Media Politics: The News Strategies of Presidential Campaigns. Lexington, MA: Lexington Books. Bartels, Larry. 1985. \Expectations and Preferences in Presidential Nominating Campaigns." American Political Science Review 79:804{15. Bartels, Larry. 1988. Presidential Primaries. Princeton: Princeton University. Brady, Henry E. and Richard Johnston. 1987. What's the Primary Message: Horse Race or Issue Journalism. In Media and Momentum: The New Hampshire Primary and Nomination Politics, ed. Gary R. Orren and Nelson W. Polsby. Chatham, NJ: Chatham House Publishers, Inc. Buell, Emmett H. Jr. 1987. Locals' and `Cosmopolitans': National, Regional, and State Newspaper Coverage of the New Hampshire Primary. In Media and Momentum: The New Hampshire Primary and Nomination Politics, ed. Gary R. Orren and Nelson W. Polsby. Chatham, NJ: Chatham House Publishers, Inc. Buell, Emmett H. Jr. 1991. Meeting Expectations? Major Newspaper Coverage of Candidates During the 1988 Exhibition Season. In Nominating the President, ed. Jr. Emmett H. Buell and Lee Sigelman. Knoxville: University of Tennessee Press. King, Gary. 1989. Unifying Political Methodology: The Likelihood Theory of Statistical Inference. Cambridge: Cambridge University Press. 13 Luskin, Robert C. 1987. \Measuring Political Sophistication." American Journal of Political Science pp. 856{99. Paolino, Philip O. 1995. Candidate Name Recognition and the Dynamics of the Pre-Primary Period of the Presidential Nomination Process PhD thesis Duke University. Patterson, Thomas. 1980. The Mass Media Election: How Americans Choose Their President. New York: Praeger Publishers. Popkin, Samuel L. 1991. The Reasoning Voter: Communication and Persuasion in Presidential Campaigns. Chicago: University of Chicago Press. Price, Vincent and John Zaller. 1990. \Measuring Individual Dierences in Likelihood of News Reception." Paper presented at the 1990 Annual Meeting of the American Political Science Association. Robinson, Michael J. and Margaret A. Sheehan. 1983. Over the Wire and On TV: CBS and UPI in Campaign '80. New York: Russell Sage Foundation. 14 Table 1: Respondents' Placement of Candidate Viability { All Respondents Linear-Normal MLE Models Independent Dukakis Gephardt Gore Hart Jackson Simon Variable Expected Value 0.3227** 0.0735 0.1203** 0.2005** (0.0736) (0.0709) (0.0341) (0.0379) 0.4500** 0.6022** 0.5114** 0.4151** (0.1004) (0.0987) (0.0599) (0.0444) 0.1188** 0.1093** 0.3100** 0.1379** (0.0237) (0.0227) (0.0419) (0.0204) 0.0158 -0.0305 -0.0743** -0.0838** (0.0276) (0.0442) (0.0269) (0.0267) -0.0092 0.0120 -0.0754** 0.0333 (0.1480) (0.0613) (0.0248) (0.0277) 0.0015 0.0166 -0.0623** -0.0705** (0.0792) (0.0364) (0.0237) (0.0265) -0.1213** -0.0034 0.1719** 0.0178 (0.0212) (0.0230) (0.0190) (0.0169) 0.1237** (0.0639) 0.5102** (0.0939) 0.1586** (0.0493) -0.0480 (0.0351) 0.0543 (0.0373) -0.0449 (0.0333) 0.0804** (0.0213) -3.7696** -4.4064** -3.5479** -2.4348** -2.9586** (0.2588) (0.5063) (0.2537) (0.2515) (0.1560) Political -0.1590 -0.2071 0.0002 -0.3939** -0.0393 Information (0.2076) (0.3314) (0.0710) (0.1630) (0.1840) Television News 0.2822 0.2238 -0.3206 -0.3137* -0.0085 Exposure (0.2022) (0.2242) (0.2774) (0.1901) (0.0216) Newspaper -0.2417 0.0716 0.1584 -0.4006** -0.2668* Exposure (0.1916) (0.8300) (0.2317) (0.1842) (0.1584) Interviewed 0.1631 0.3508** -0.0843 0.4284** -0.0788 Before Iowa (0.1341) (0.1585) (0.1484) (0.1333) (0.1028) Thermometer 1.2032** 3.0716** 1.9089** -0.9806** Dierence (0.3534) (0.4962) (0.4510) (0.4141) N 519 476 373 682 686 Source: 1988 NES Super Tuesday Survey. Standard errors in parentheses. p < :1 p < :05 -3.6629** (0.3616) -0.2578 (0.2041) 0.0875 (0.2862) 0.0161 (0.1859) -0.0537 (0.1274) 2.6025 (0.4702) 444 Candidate Preference Candidate Best in State Political Information Television News Exposure Newspaper Exposure Interviewed Before Iowa Variance 0.3592** (.0544) 0.3911** (0.0681) 0.1041** (0.0176) 0.0090 (0.0334) 0.0121 (0.0304) 0.0086 (0.0287) -0.0699** (0.0188) 15 Table 2: Respondents' Placement of Candidate Viability { All Respondents Linear-Normal MLE Models Independent Bush Dole Kemp Robertson Variable Expected Value Candidate Attitude Candidate Best in State Political Information Television News Exposure Newspaper Exposure Interviewed Before Iowa Variance 0.5358** 0.3915** 0.0684 0.1814** (0.0507) (0.0364) (0.0617) (0.0349) 0.2001** 0.3641** 0.5514** 0.5461** (0.0297) (0.0492) (0.0753) (0.0422) 0.0780** 0.0929** 0.2381** 0.1116** (0.0115) (0.0119) (0.1175) (0.0275) 0.0202 0.0012 -0.0663** -0.0434** (0.0123) (0.0242) (0.0264) (0.0217) 0.0123 0.0458** 0.0056 0.0669** (0.0795) (0.0181) (0.0334) (0.0231) -0.0062 0.0235 0.0410 0.0019 (0.0712) (0.0178) (0.0281) (0.0181) 0.0219* -0.0328** 0.0535** -0.0661** (0.0129) (0.0121) (0.0172) (0.0147) -3.8080** -4.3172** -3.9603** (0.1706) (0.1487) (0.2714) Political -0.2993** 0.1214 -0.0331 Information (0.1385) (0.1674) (0.1931) Television News 0.1379 0.0449 0.0806 Exposure (0.3480) (0.1931) (0.2004) Newspaper -0.2877 -0.0973 0.1760 Exposure (0.2851) (0.1510) (0.1584) Interviewed -0.0200 -0.0637 -0.2667** Before Iowa (0.4560) (0.0998) (0.1126) Thermometer 1.6437** 2.3772** 2.1169** Dierence (0.3040) (0.2490) (0.5807) N 955 848 548 Source: 1988 NES Super Tuesday Survey. Standard parentheses. p < :1 p < :05 16 -3.9514** (0.2010) -0.1927 (0.1397) -0.0985 (0.1723) 0.3701** (0.1495) 0.0480 (0.0972) 2.2200** (0.3652) 835 errors in Table 3: Respondents' Placement of Candidate Viability { All Respondents Bivariate Normal Selection Model { MLE Independent Bush Dole Dukakis Gephardt Variable Expected Value Candidate Attitude Candidate Best in State Political Information Television News Exposure Newspaper Exposure Interviewed Before Iowa Variance 0.2629** 0.3682 0.16430 (0.1308) (0.1660) 0.2475** 0.3703 0.4164** (0.0395) (0.0723) 0.0780** 0.0936 0.1034** (0.0113) (0.0176) 0.2423** 0.0129 0.1061 (0.1044) (0.0798) 0.0558* 0.0506 0.0328 (0.0288) (0.0351) 0.0538 0.0295 0.0635 (0.0361) (0.0533) 0.0008 -0.0338 -0.0943** (0.0157) (0.0286) 0.5540** (0.1411) 0.4042** (0.1002) 0.1272** (0.0209) -0.0921 (0.0648) -0.0441 (0.0346) -0.0500 (0.0405) -0.0852** (0.0259) -3.5657** -4.3262 -3.6920** -4.1971** (0.2620) (0.3014) (0.3702) Political -0.2432* 0.1266 -0.1289 -0.1956 Information (0.1459) (0.1979) (0.1955) Television News 0.1508 0.0493 0.2838 0.1799 Exposure (0.1395) (0.2024) (0.2016) Newspaper -0.2420 -0.0877 -0.2532 0.0448 Exposure (0.1556) (0.1896) (0.1729) Interviewed -0.0341 -0.0675 0.1890 0.3083** Before Iowa (0.1015) (0.1319) (0.1445) Thermometer 1.5446** 2.3467 1.1613** 2.9334** Dierence (0.1813) (0.3772) (0.4260) 0.5042** 0.0489 0.2955 -0.3413** (0.1813) (0.2131) (0.1594) N 2022 2022 2022 2022 Source: 1988 NES Super Tuesday Survey. Standard errors in parentheses. p < :1 p < :05 17 Table 4: Respondents' Placement of Candidate Viability { Post-Iowa Linear-Normal MLE Models Independent Dukakis Gephardt Gore Hart Jackson Simon Variable Expected Value 0.3585** 0.2925** 0.1040 0.0785* 0.2756 ** (.0579) (0.0702) (0.0796) (0.0477) (0.0687) 0.3532** 0.5296** 0.5687** 0.1590* 0.3957** (0.0805) (0.1136) (0.1162) (0.0862) (0.0928) 0.0859** 0.0689** 0.1080** 0.5788** 0.1008** (0.0197) (0.0217) (0.0295) (0.0727) (0.0246) 0.0205 -0.0141 0.0110 -0.0057 -0.0857** (0.0279) (0.0281) (0.0423) (0.0204) (0.0342) -0.0109 -0.0444 0.0092 -0.0162 0.0215 (0.0270) (0.0285) (0.0414) (0.0245) (0.0566) 0.0150 -0.0179 -0.0192 -0.0236 -0.0680 (0.0295) (0.0293) (0.0363) (0.0300) (0.0527) -0.0268 0.1431** -0.0226 -0.0192 -0.0578** (0.0196) (0.0187) (0.0274) (0.0192) (0.0202) 0.1018** -0.0382** -0.0104 -0.0462** -0.0203 (0.0182) (0.0189) (0.0312) (0.0170) (0.0202) 0.1630** (0.0715) 0.4745** (0.1063) 0.2173** (0.0439) -0.0320 (0.0402) 0.0449 (0.0396) -0.0428 (0.0378) 0.0244 (0.0257) -0.1155** (0.0251) -3.8250** -4.4368** -3.7504** -0.8618 -3.5686** (0.2865) (0.3585) (0.3800) (0.7104) (0.8485) Political -0.0176 -0.1499 0.0892 -0.3556 0.1144 Information (0.1216) (0.2565) (0.3393) (0.2610) (0.6563) Television News 0.4335* 0.1937 -0.1632 -0.2051 0.0405 Exposure (0.2505) (0.2360) (0.2779) (0.3348) (1.0542) Newspaper -0.3546* -0.1230 0.1291 -0.8259** -0.3522 Exposure (0.2040) (0.2183) (0.2693) (0.2630) (0.4667) Knows Who Won 0.0687 -0.1023 0.0252 -0.5994** 0.2477 Iowa (0.1427) (0.1428) (0.1445) (0.2187) (0.2030) Knows Who Won -0.3673** 0.1396 0.3292* -0.4471* -0.1065 New Hampshire (0.1446) (0.1351) (0.1694) (0.2094) (0.1960) Thermometer 1.1210** 3.0627** 1.3785** -4.4463** 1.2219** Dierence (0.4066) (0.4729) (0.5750) (1.2916) (0.3590) N 381 358 274 472 482 Source: 1988 NES Super Tuesday Survey. Standard errors in parentheses. p < :1 p < :05 -3.2374** (0.3451) -0.4448* (0.2310) -0.1402 (0.2549) -0.1520 (0.2407) 0.4117** (0.1324) -0.0922 (0.1324) 1.7266** (0.5174) 325 Candidate Attitude Candidate Best in State Political Information Television News Exposure Newspaper Exposure Knows Who Won Iowa Knows Who Won New Hampshire Variance 18 Table 5: Respondents' Placement of Candidate Viability { Post-Iowa Linear-Normal MLE Models Independent Bush Dole Kemp Robertson Variable Expected Value 0.5439** (.0300) Candidate 0.1968** Attitude (0.0344) Candidate Best 0.0714** in State (0.0136) Political 0.0211 Information (0.0203) Television News 0.0297 Exposure (0.0209) Newspaper -0.0210 Exposure (0.0218) Knows Who Won -0.0378** Iowa (0.0148) Knows Who Won 0.0444** New Hampshire (0.0136) Variance 0.4281** (0.0428) 0.2949** (0.0609) 0.0881** (0.0147) 0.0139 (0.0215) 0.0365* (0.0208) -0.0120 (0.0218) 0.0594** (0.0173) -0.0241** (0.0157) 0.1162* 0.2178** (0.0681) (0.0421) 0.5459** 0.5107** (0.0868) (0.0485) 0.5838** 0.0889** (0.0355) (0.0282) -0.0672* -0.0332 (0.0358) (0.0265) -0.0093 0.0694** (0.0342) (0.0269) 0.0018 0.0028 (0.0391) (0.0214) -0.0155 -0.0049 (0.0277) (0.0168) -0.0485** -0.0529** (0.0235) (0.0169) -3.8537** -4.2945** -3.2595** (0.1893) (0.2113) (0.3702) Political -0.1730 0.1463 -0.1365 Information (0.1599) (0.2009) (0.2200) Television News 0.1096 0.0601 -0.0005 Exposure (0.1501) (0.1780) (0.0620) Newspaper -0.1243 -0.0655 0.0738 Exposure (0.1739) (0.1736) (0.2114) Knows Who Won 0.0313 -0.1488 -0.0850 Iowa (0.0952) (0.1736) (0.1584) Knows Who Won -0.2574** -0.0240 -0.0949 New Hampshire (0.1127) (0.1542) (0.1222) Thermometer 1.4321** 2.4161** 0.7075 Dierence (0.3098) (0.2958) (1.0497) N 653 581 375 Source: 1988 NES Super Tuesday Survey. Standard parentheses. p < :1 p < :05 19 -3.9248** (0.2585) -0.3002* (0.1627) -0.0836 (0.2285) 0.3790** (0.1885) -0.0670 (0.1303) -0.0668 (0.1391) 2.4961** (0.4138) 579 errors in Table 6: Respondents' Placement of Candidate Viability { Post-Iowa Respondents Correctly Identifying Iowa Winner Linear-Normal MLE Models Independent Dukakis Gephardt Gore Hart Jackson Simon Variable Expected Value -0.0011 0.3708 ** (0.0274) (0.0828) 0.0138 0.2985** (0.0288) (0.0719) 0.6717** 0.1359** (0.0185) (0.0437) -0.0093 -0.1190** (0.0140) (0.0569) 0.0042 -0.0211 (0.0206) (0.0519) 0.0111 -0.1290** (0.0090) (0.0494) 0.1300 (0.1063) 0.4492** (0.1521) 0.0901** (0.0508) -0.0984 (0.0690) 0.0559 (0.0647) 0.0338 (0.0627) -4.8191** -4.5906** -3.0422** -0.0235 -3.4488** (0.5594) (0.5819) (0.4450) (0.2283) (0.5810) Political -0.2580 -0.2814 -0.2725 0.2721 -0.1768 Information (0.3063) (0.3248) (0.3862) (0.4978) (0.4341) Television News 1.1726** 1.0130** 0.0658 -1.4677** -0.1607 Exposure (0.3862) (0.3876) (0.2972) (0.6073) (0.3034) Newspaper -0.4360 -0.4475 0.1322 -0.2940 -0.2202 Exposure (0.3066) (0.2764) (0.3410) (0.6019) (0.2914) Thermometer 2.3590** 2.2955** -9.3977** 1.7936** Dierence (0.7091) (0.6817) (1.1905) (0.5727) N 195 206 149 220 223 Source: 1988 NES Super Tuesday Survey. Standard errors in parentheses. p < :1 p < :05 -3.3115** (0.4254) -0.5959** (0.2938) 0.4161 (0.3319) -0.1599 (0.3093) 2.2173** (0.8013) 181 Candidate Attitude Candidate Best in State Political Information Television News Exposure Newspaper Exposure Variance 0.2735** (.0796) 0.4376** (0.1019) 0.0647** (0.0249) 0.0514 (0.0389) 0.0692 (0.0423) 0.0054 (0.0515) 0.5736** (0.0897) 0.2708** (0.1300) 0.0676** (0.0245) -0.0416 (0.0444) -0.0317 (0.0358) -0.0256 (0.0457) -0.0135 (0.1440) 0.6874** (0.1766) 0.0856** (0.0377) -0.0471 (0.0824) 0.0146 (0.0711) 0.0547 (0.0597) 20 Table 7: Respondents' Placement of Candidate Viability { Post-Iowa Respondents Correctly Identifying Iowa Winner Linear-Normal MLE Models Independent Bush Dole Kemp Robertson Variable Expected Value Candidate Attitude Candidate Best in State Political Information Television News Exposure Newspaper Exposure Variance 0.4986** (.0393) 0.2096** (0.0431) 0.0792** (0.0161) 0.0467* (0.0255) 0.0377 (0.0246) -0.0269 (0.0268) 0.5395** (0.0552) 0.2048** (0.0688) 0.1066** (0.0170) -0.0049 (0.0251) 0.0410 (0.0250) -0.0286 (0.0259) 0.0052 (0.0512) 0.6194** (0.0731) NA -0.0486 (0.0375) -0.0100 (0.0384) 0.0248 (0.0449) -3.8746** -3.8031** -3.4009** (0.2673) (0.1753) (0.4714) Political -0.2643 0.1864 -0.1661 Information (0.1972) (0.2097) (0.2207) Television News 0.0753 0.0639 -0.1425 Exposure (0.2322) (0.1963) (0.2233) Newspaper -0.1673 0.0439 0.3614 Exposure (0.3024) (0.1700) (0.2202) Thermometer 1.5904** 2.4850** 0.6275 Dierence (0.3928) (0.5033) (1.1342) N 424 410 289 Source: 1988 NES Super Tuesday Survey. Standard parentheses. p < :1 p < :05 21 0.1874** (0.0500) 0.5639** (0.0571) 0.0975** (0.0328) -0.0431 (0.0329) 0.0563* (0.0326) -0.0006 (0.0245) -4.2089** (0.2644) -0.2923 (0.2002) 0.0644 (0.2026) 0.3527* (0.2061) 2.8673** (0.5029) 388 errors in