Lost in the Crowd: The Effect of Volatile Fields on Presidential Primaries

Abstract

All candidates attempt to convey agendas through the media, but presidential primary contenders face distinctive conditions. Do these conditions affect candidates’ ability to transmit their agendas through the media? Drawing on theories of journalism norms, I argue that media responsiveness to a candidate’s agenda will, on average, decrease as the primary campaign goes on and candidates’ messages become less timely. Expansions in the size of the field should create more complicated electoral races, leading to declining contextual simplicity and less overlap between candidate and media agendas. I test these hypotheses using a novel dataset of primary candidates’ speeches and news articles about those candidates. I show that the entrance of new candidates is correlated with a decline in convergence. Contrary to expectations, there is no decline in convergence over time. The results have implications for how nomination contests should be structured in the era of media-oriented campaigns.

Keywords

primaries campaigns and elections mass media agenda building journalism norms

“Imagine a sport where you spend 1 year [hopping] on one leg and not using the other, then suddenly having to run a marathon.¹” – Zach Weinersmith, cartoonist

Presidential primaries are radically different from the general elections that follow. They are long, sequential contests in which candidates often jump in the race nearly a year before the first in a concatenation of caucuses and primaries. They are multicandidate, leading to volatility in the field as candidates hop in and out of the race. They are intraparty, removing the party moniker cue and often featuring candidates with similar issue positions. And the electorate is radically different in size and political interest.

But there is one important consistency between presidential primaries and general elections. They are both oriented primarily through the media (Azari, 2016; Polsby, 1983). From the candidates’ perspective, media-centric primaries are essential in the post-reform era as they are the most cost- and time-efficient means of contacting a large base of potential supporters across the country. Voters are also advantaged by the media-centric primary system as it helps organize a regularly chaotic process that would otherwise be time-consuming to comprehend. The media serves as a conduit between the candidates’ desire to cheaply reach a broad audience and voters’ preference to learn about politics in as efficient a manner as possible.

And yet the media does not operate to maximize the utility of candidates or voters. The media has its own incentive structures and preferences that affect coverage (Zaller, 1999). While we know how those preferences shape coverage in concert with the electoral structure of general elections (e.g., Hayes, 2010; Vavreck, 2009), we don’t yet know how the structure of American presidential primary races affects how the media covers primaries. Obtaining such knowledge is important because the agenda-setting literature (McCombs & Shaw, 1972) suggests that what the media covers affects public opinion, because candidates actively seek to convey agendas through the media, and because parties should want to ensure that the candidate who emerges from this process is well positioned for the general election.

This paper seeks to explain the relationship between the structure of primaries and media coverage, specifically in terms of what issues the media talks about, via a theory of journalism norms. Journalists are socialized via their profession into prioritizing certain traits in potential news stories. These traits include timeliness and simplicity.

I argue that the structure of the primary system should lead to variation in the timeliness and simplicity journalists perceive in the race. Given how long primaries are, reporters closely following each campaign should gradually find candidates’ messages less timely as their talking points and stump speeches become old news. The multicandidate attribute of primaries allows for volatility in the size of the field. The entrance of new candidates should lead to decreasing contextual simplicity as the introduction of more candidates creates a more complex messaging environment, thereby decreasing the newsworthiness of candidate messages. A winnowing of the field, in contrast, should have a clarifying effect for journalists, increasing contextual simplicity and boosting the newsworthiness of the remaining contenders’ agendas.

I test these theoretical expectations using a novel dataset of candidate speeches and media coverage called the Presidential Primary Communication Corpus. I measured the issues candidates and the media discussed using an unsupervised topic modeling technique called Structural Topic Modeling (Roberts et al., 2014). I measured the overlap between the agenda of the candidates and of the media by calculating convergence scores (Sigelman & Buell, 2004). I then assessed these convergence scores descriptively for change over time and regressed the convergence scores on a measure of aggregate change in the size of the field.

Briefly, there was strong and consistent support for the hypothesis that expansions in the field of candidates will be correlated with decreases in agenda overlap. Across a range of model specifications, an increase in the number of candidates seeking the nomination was correlated with a significant decrease in convergence between candidate and media agendas. This effect is experienced primarily among candidates performing poorly in the polls. The result was robust even when accounting for the media’s tendency to engage in horserace coverage. There was no support for the hypothesis that agenda overlap would decrease over time.

Overall, the length of presidential primaries does not seem to create a circumstance that clashes with the media’s professional preferences. By contrast, the multicandidate nature of primaries conflict with the norms of the media in a manner that leads to deviations from candidates’ preferred issue focus. These results emphasize the powerful influence that the media’s newsmaking norms have in shaping coverage of presidential primary candidates. They also raise questions for candidates and parties about how best to structure nomination races so the media’s professional proclivities do not affect the efficacy of the contests. I conclude by discussing such questions.

Literature Review

Presidential primaries, much like presidential general elections, are contests run primarily via the news media (Azari, 2016; Polsby, 1983). While candidates do have other means of reaching voters directly, traditional news coverage remains the cornerstone of the national profile that candidates need to compete effectively for their party’s nomination. Primary candidates, therefore, do their best to get their agenda reported in the media. In this way, primaries are just like general election campaigns (Vavreck, 2009). But unlike presidential general elections, primaries are lengthy, multicandidate affairs (Norrander, 2010). Candidates often enter the race nearly a full year before the Iowa caucus while there is typically only 3 or 4 months between the party convention and Election Day. And while general elections typically involve only two major candidates, primary fields now routinely crack double digits.² These differences could, and likely do, create a unique communication environment.

And yet we know little about the ways the media interacts with this campaign structure. Students of primary campaigns have noted the important role the media plays (Bartels, 1988) and documented the extensive amount of horserace content that dominates media coverage (Patterson, 2016). Others have studied the relationship between the issues candidates emphasize and those that saturate media coverage (Conway et al., 2015; Conway-Silva et al., 2018; Flowers et al., 2003; Kendall, 2000; Miller et al., 1998; Vinson & Moore, 2007). But this work does not tell us what effect the unique structure of presidential nomination contests has on shaping media coverage of those races.

This is not to say that scholars of campaigns have completely overlooked the role that the structure of primaries plays. Political scientists working on understanding electoral outcomes (Aldrich, 1980; Bartels, 1988) have contended with modeling either theoretically or statistically the influence of long campaigns. The same can be said of scholars examining candidate attrition (Haynes et al., 2004; Norrander, 2000). And scholars studying candidate strategies have considered how multicandidate fields restructure messaging incentives (Haynes et al., 2002; Haynes & Rhine, 1998; Ridout & Holland, 2010). But the role these features play in shaping what issues the media talks about has not been sufficiently addressed.

Doing so is important for three reasons. First, experimental evidence shows that the issues the media emphasizes affect what issues the public thinks are most important (Feezell, 2018; Iyengar & Kinder, 1987). Second, candidates seek to convey agendas through the media (Pfetsch, 1999; Sanders et al., 2011). Because that is an explicit goal, understanding the role that contextual factors outside the candidates’ control play is important. Third, the parties should want the nominee to finish the primary having cemented an agenda in the minds of voters so he or she has established a national profile. Therefore, understanding how the structure of primaries affects the ability of candidates to get messages across should be of vital interest to parties trying to devise a primary system that leaves their nominee well positioned for the general election.

Structure-Induced Differences in Newsworthiness

How do the features of the primary process itself affect how the media covers candidates’ agendas? To answer that it is first important to understand how the media goes about making the news. Newsmaking is best understood as the product of a confluence of different forces (Boydstun, 2013). The individual preferences of the journalists, editors, and owners exert a powerful, sometimes subconscious influence (e.g., Donsbach, 2004; Dunaway, 2008). The for-profit American media also follows the economic laws of supply and demand (Hamilton, 2004). And journalism as a profession has standards of shared values that shape collective perceptions of what constitutes “good” and “bad” journalism (Zaller, 1999). These shared values can contribute of a tendency in some media outlets to influence the coverage for others via intermedia agenda setting (Roberts & McCombs, 1994).

While I do not preclude the possibility that other forces play a role in some form, the primary theoretical thrust of this paper concerns professional norms. The scholarly literature on journalism’s professional norms goes back for decades and is broad and nuanced but, to simplify, it can be understood as organized into two camps. First, there are studies that derive norms by studying media content, often in comparison to some sort of objective, real-world standard (e.g., Harcup & O’Neill, 2001; Lee, 2009; Soroka, 2012). Second, there are ethnographic studies that document norms by directly observing journalist behavior including what news events they encounter, which they chose to follow up on, and then how the coverage of those few selected topics is structured (e.g., Gans, 1979; Tuchman, 1978; Usher, 2014).

As with any long-standing and diverse literature, consensus is elusive. But there are common threads. A significant number of studies note a general prioritization of news stories that are timely and simple. These would appear to be traits that journalists covet when they are trying to make the news.

The professional newsmaking procedures of journalists can shape media content in primaries (Flowers et al., 2003). Often, this is articulated via the tendency of media outlets to engage in horserace coverage of the campaigns (Dunaway & Lawrence, 2015; Patterson, 2016; Searles & Banda, 2019). But the media must also decide what issues to cover via newsworthiness values. How does the structure of primaries intersect with these professional newsworthiness decisions? I argue that the lengthy, multicandidate nature of primaries should create circumstances that systematically vary the timeliness and simplicity journalists perceive the race, and therefore the candidates, as possessing. This, in turn, will affect how closely the media hews to the candidates’ agendas. If the structure of the primary is creating a highly newsworthy (i.e., timely and simple) environment, then the media will see candidate messages as more newsworthy and converge more toward the candidate’s agenda. If the context is depressing newsworthiness than the media will be more likely to wander off toward other topics while covering the race, leading to divergence between the candidate and media agendas.

Starting with an application of timeliness, I expect that candidates will attempt to stay “on-message” in their agenda to maximize the chances that a potential supporter will encounter their core message. This should conflict with the journalistic value of timeliness as the repetition of the message will naturally diminish its timeliness relative to other, more zeitgeist-y approaches to discussing the race. Media responsiveness to a candidate’s agenda should, therefore, decrease as the campaign goes on and the candidate’s agenda becomes old news.

H1: Agenda similarity between candidate messages and media coverage will decrease over time.

Second, I consider how the multicandidate nature of primaries and the volatility it introduces interacts with media preference for simplicity. As the field expands to include more contenders, evaluating the messages of each candidate becomes more complex. It is harder for journalists to divine the most salient fault lines between candidates and contextualize their agendas for two reasons. First, more candidates means more messages, each offered with an argument that this is the most compelling campaign news of the day. Journalists reporting on the primary race will be aware of these myriad messages which will lead to uncertainty about the relative newsworthiness of any one candidate’s agenda. Second, the introduction of new candidates creates more permutations of pairwise comparisons between the candidates, making the race appear more complicated as there are more ways of analyzing it. Over time, journalists will adapt to the field and so it is not the number of candidates per se that will lead to reductions in contextual simplicity but the shock of an expansion in the field, shifts that are only possible because of the volatility multicandidate races allow, that will affect newsworthiness.

In contrast, contractions in the size of the field should have an opposite, clarifying effect. Winnowing (Haynes et al., 2004; Norrander, 2000, 2006; Steger et al., 2002) should make the messages of the remaining candidates simpler for journalists because the agendas will stand out more clearly against a less noisy electoral context of alternative messages. Put simply, an expanding primary field reduces the contextual simplicity of the contest while a retracting primary field simplifies the context. This leads to my second hypothesis:

H2: Agenda similarity between candidate messages and media coverage will decrease when the field expands.

Data and Methods

To test these hypotheses, I rely on a novel database of presidential primary communication: the Presidential Primary Communication Corpus. I now turn to a description of how the corpus was constructed with the caveat that a more extensive description can be found in the accompanying supplementary materials.

Collecting the Corpus

I limited analysis to candidates for major party nominations who actively campaigned from 2000 to 2016.³ I used The New York Times and the Washington Post as stand-ins for the overall media agenda.⁴

I utilized candidate speeches to measure candidates’ issue agendas. The text of these speeches came primarily from closed captioning of the C-SPAN video archive.⁵ The corpus contains 2611 candidate speeches obtained via this method.

I retrieved the media coverage of each candidate from Nexis Uni. The articles were then individually assessed for relevance to the candidate to remove articles that mentioned the candidate only in an off-hand manner. Once the textual wheat had been separated from the chaff, I retained a corpus of 29,336 news articles.⁶

Identifying Agendas

This extensive quantity of text precludes the possibility of a detailed content analysis. Of the myriad methods of computational methods of textual analysis (Grimmer & Stewart, 2013), an unsupervised model is most appropriate given the specifics of the corpus and lack of a pre-defined topic list. Unsupervised methods classify documents into topics without researcher input on what the topics are or what the words that comprise that topic should be. Given the wide time range considered in this corpus and the natural tendency for topic saliency to fluctuate over time, this is more appropriate than attempting to fit documents from 2016 and 2000 to the same topic typology via a supervised topic classification or dictionary method.

Specifically, I utilized a Structural Topic Model (STM). What sets STM apart from other unsupervised methods is that it allows for the inclusion of covariates in the topic model itself, allowing the researcher to account for ways the use of topics might vary systematically across documents (Roberts et al., 2014). STM can be guided based on prior theory but remains flexible enough to work with a diverse corpus.⁷

The Lee and Mimno (2014) algorithmic method suggested that an 82-topic model was a particularly good fit for the corpus. A broader survey of model fit statistics with varied topic parameters (available in the appendix) supported this estimate. While model fit is important, a well-fit model can still produce nonsensical results. As such, it is important to validate the identified topics. I did so via a close reading of several excerpts that were highly associated with each topic. That, in concert with surveying the high frequency words, assisted me in identifying a label for each topic. The results of this validation effort can be found in the appendix.

Calculating Convergence

Each document can be expressed as a combination of the identified topics. How prominent each topic is in the individual document can be found in the theta matrix output. This matrix includes a measure, the theta score, for how closely each of the documents is matched to each of the topics. I use this matrix to compare candidate and media agendas. I do so by calculating a convergence score (Sigelman & Buell, 2004). Convergence scores take two separate sets of text classified into topics and compare the proportions attributed to those topics to generate a single, continuous measure of agenda similarity. The formula for convergence scores is as follows:

I = 1 - \frac{| c_{1} - m_{1} | + | c_{2} - m_{2} | + ... | c_{j} - m_{j} |}{2}

Where I denotes the issue convergence score, c refers to the proportion of the first actor’s (in this case the candidate’s) agenda that was dedicated to issue j, and m refers to the proportion of the second actor’s (in this case the media’s) agenda that was dedicated to issue j. I ranges from 0 (perfect divergence) to 1 (perfect convergence).

To calculate convergence scores, I aggregated the thetas for candidates and the media separately for each month of the campaign. This time period was chosen due to data availability as a shorter unit of time, for example a week, led to a great deal of missingness in the series.⁸ The dependent variable is thus convergence scores calculated for each candidate-media-month.

This still left some missingness in the data. I addressed this missingness via some limited logical interpolation. For those months with media coverage but no candidate speeches, I used the candidate agenda from the preceding month. So, if the corpus contained no candidate speech for the month of November, the assumption was made that the candidate’s agenda did not change from October. For those months with no media coverage at all, I interpolated a convergence score of 0. A candidate who got no media coverage necessarily did not get any of their agenda across. Given the cross-sectional time series nature of the data, these assumptions prevent significant loss of data from listwise deletion while still being reasonable.⁹

Modeling

The data are cross-sectional time series. Each unit of the data is a candidate-media-month. Convergence scores are a continuous variable warranting OLS.

Given the serial nature of primaries, candidates are incentivized to remain stable in their agendas as not all potential voters are paying attention at one time. In addition, previous work on news agendas establishes that what the media wants to cover is itself strongly influenced by what it has covered in the past (Boydstun, 2013). As such, from both the candidate and media perspective it is extremely likely that convergence represents a dynamic temporal process where what has happened in the past needs to be accounted for in modeling the present (Keele & Kelly, 2006). Therefore, I utilize the Koyck distributed lag model.¹⁰

As H1 establishes, I expect a downward trajectory for convergence over time. This hypothesis is tested by a descriptive analysis of the pooled convergence series.

To test H2 on the effect of volatility created by the multicandidate nature of primaries on media coverage, I use the change in the number of candidates competing in the race in that month. For example, the 2016 GOP primary saw two candidates jump into the field in April, 2015 (Marco Rubio and Rand Paul) and is therefore +2. This variable neatly captures the exogenous shock to convergence that H2 theorizes.

The data generation process outlined above likely created two sources of measurement error that need to be addressed. First, I control for whether or not the candidate agenda was interpolated in that time unit.¹¹ Second, because there is so much more media content than candidate speeches, there are more opportunities for a wide array of issues to be invoked by the media in the data. Therefore, the media agenda is more likely to be widely dispersed not because the media talks about more issues but because it has more opportunities in the corpus to talk about more issues. To account for this, I control for the extent of topic dispersion in the candidate’s agenda in that time unit, measured via Shannon’s H Entropy (Boydstun, 2013).

I also control for contextual factors that might matter. First, there are candidate-level controls. These include the party, gender, and race of the candidate. Prior literature suggests that the media may respond differently to political elites based on such features (Niven, 2004). I also include a measure of standing in the polls.¹² We might expect that candidates who are performing better in the polls will be more interesting to journalists given their heightened odds of assuming significant political influence (Zaller, 1999).

Studies of general election campaigns suggest that the competitiveness of the race should affect convergence (Hayes, 2010). Measuring competitiveness of multicandidate races is difficult as the gap between two candidates is partially conditional on the number of other candidates in the race. To measure the competitiveness of the race, I utilized a modified Hirschman-Herfindahl index (Steger et al., 2002). This is a concentration metric which captures the number of “effective” candidates that are in the race based on the distribution of poll shares. If there are five candidates in a race, each polling at approximately 20%, there are approximately five effective candidates. If a race features two dominant candidates and three fringe contenders, the number of effective candidates will be closer to two. Prior literature suggests that more effective candidates in a race should correlate with lower convergence scores.¹³

Results

Figure 1 plots the mean convergence scores per month throughout the primary. Given the cross-sectional time series nature of the data, analyzing means provides a more parsimonious assessment of H1 than examining the 76, separate series (although plots of those series are available in the appendix). The top panel plots the mean convergence score by month, while the bottom panel applies a LOESS smoothing function and calculates confidence intervals.

Figure 1.

Mean agenda convergence by month in presidential primaries.

The case for an overall negative trend through the duration of the primary is nonexistent. There does appear to be a noticeable initial decline over the first 3 months, more obvious in the top panel, but this reverses over the subsequent months. H1 therefore sees no support. In fact, average convergence later in the primary seems to be higher than in the early months.

There is another possibility, however. Most campaigns end well short of a year. Perhaps this apparent positive trend stems from which candidates remain so deep into the election season, not how the media is reacting to candidate messages late into the cycle.

To test this possibility, I plotted the convergence scores per month of the two most prominent candidates for each primary race (Figure 2). When looking at just the top contenders, who typically campaign the longest, there does not appear to be a clear pattern of increasing convergence over time. Some series do display such a trend, most notably Santorum-2012 and McCain-2008, but the vast majority are better described as temporally stable. In addition, while the average convergence score only crosses 30 during the trend’s highest peaks, many of these series approach or exceed that threshold routinely. Overall, the apparent increase in mean convergence over time stems from the fact that the candidates still campaigning at month 17, 18, and 19 are those already best able to elicit media coverage of their agenda. Regardless, the timeliness hypothesis is unsupported.

Figure 2.

Convergence throughout primary by campaign, major candidates.

Effect of Change in Size of the Primary Field on Convergence

Moving on, Table 1 presents several models designed to test H2.¹⁴ The hypothesis stated that an increase in the size of the primary field should be associated with a decrease in convergence. The first column provides a test of this hypothesis while also accounting for the quirks of the data generation process, candidate-level controls, and election-year fixed effects.

Table 1.

Effect of Change in Number of Candidates on Convergence in Primaries.

	Convergence
	(1)	(2)	(3)	(4)
Convergence, lagged	0.123* (0.038)	0.115* (0.038)	0.113* (0.038)
Convergence, lagged (horserace omitted)				0.116* (0.032)
Change in # of candidates	−0.006* (0.002)	−0.009* (0.002)	−0.013* (0.003)	−0.006* (0.002)
Inverse Shannon H entropy	−0.325* (0.051)	−0.311* (0.051)	−0.300* (0.051)	−0.291* (0.042)
Interpolated candidate Agenda	−0.075* (0.012)	−0.073* (0.012)	−0.073* (0.012)	−0.014 (0.012)
Poll	0.001* (0.0003)	0.002* (0.0003)	0.002* (0.0003)	0.001* (0.0003)
Horserace coverage				−0.082* (0.026)
GOP	−0.012 (0.011)	−0.032* (0.013)	−0.030* (0.013)	0.005 (0.011)
Woman	−0.032* (0.014)	−0.033* (0.014)	−0.035* (0.014)	−0.015 (0.011)
Nonwhite	0.007 (0.011)	0.004 (0.011)	0.005 (0.011)	0.001 (0.009)
# of Effective candidates		0.009* (0.003)	0.009* (0.003)	0.003 (0.003)
Change in # of candidates * poll			0.0004* (0.0002)	0.0003* (0.0002)
Constant	0.335* (0.029)	0.322* (0.030)	0.313* (0.030)	0.386* (0.026)
Observations	648	648	648	585
R ²	0.260	0.270	0.276	0.177
Breusch-Godfrey test statistic	0.178	0.785	1.262	0.314

Note. *Denotes p < .05, one-tailed. OLS regression. All models include election-year fixed effects not presented to conserve space. Convergence score in Model 4 calculated by omitting all horserace topics. All Breusch-Godfrey test statistics statistically insignificant at conventional levels.

The results support the hypothesis. Each additional candidate entering the race is correlated with a reduction in the convergence score of each candidate by .006 on average. The largest change in the data, −7 candidates in the GOP race in February, 2016, would therefore project to a .042 increase for each Republican candidate in the race at that time, on average.¹⁵

The controls perform as expected. Candidate agendas more narrowly tailored to a smaller number of topics are correlated with lower convergence scores, as expected given the difference in opportunity to invoke an issue provided by the difference in quantity of documents. And units where the candidate agenda was interpolated using a previous candidate-month agenda are correlated with substantially lower convergence scores, on average.¹⁶

Of the candidate-level controls, two stand out. First, the candidate’s standing in the polls is a powerful positive correlate with convergence. Each additional percentage point increase is associated with an increase of .001. This also helps contextualize the substantive power of the change in the number of candidates. Figure 3 presents the predicted effect of seven candidates dropping out of the race against the predicted difference of a 30-percentage-point increase in poll standing. The difference predicted by the change in the size of the field is statistically indistinguishable from, and even slightly substantively larger than, the difference between a footnote in the race and a viable frontrunner.

Figure 3.

Comparison of predicted substantive magnitudes.

Second, women running for party nominations receive lower agenda convergence on average and by a significant degree. This finding serves as an aggregate reinforcement to the literature on gender bias in elections (e.g., Heldman et al., 2005). Neither the race nor party of the candidate are not correlated with substantively notable differences in agenda convergence in this model.

Model 2 accounts for the competitiveness of the field. Including a measure of the number of effective candidates in the race does not dampen the substantive or statistical significance of changes in the size of the field. The number of effective candidates is indeed strongly correlated with convergence, but in the opposite of the expected direction. While Hayes (2010) finds that as general elections get more competitive the media diverges more from the candidates’ agendas, I find that as the number of effective candidates increases, which should capture the uncertainty of the outcome, the media hews closer to candidates’ agendas.

So far, the results consistently support the simplicity hypothesis. Expansions in the size of the field are correlated with reductions in convergence. But it is possible that this effect is not experienced equivalently among candidates. Perhaps frontrunners and marginal candidates experience the effect differently. Model 3 accounts for this possibility by introducing an interaction term between the change in the number of candidates and poll standing. The change in the number of candidates remains a statistically significant negative correlate with convergence while poll standing remains a statistically significant positive correlate with convergence. But the interaction term between them is positive and significant, implying that some of the decline in convergence is offset for frontrunners.

Because interaction terms are frequently unintuitive, especially when dealing with two continuous variables, I plotted the predicted relationship between change in the number of candidates and convergence for three different values of poll standing: 0, 12, and 27 percentage points (Figure 4). For a candidate polling at 27%, a change in the number of candidates has essentially no effect on convergence. Among candidates who are polling worse, expansions in the field are correlated with significant decreases in convergence. This suggests that it is lower-tier candidates who are disadvantaged by the effect expanding fields have on the media agenda.

Figure 4.

Effect of change in size of primary field on convergence by poll standing.

Another prominent finding in the academic literature on media coverage of campaigns is the predominance of horserace coverage (Patterson, 2016; Searles & Banda, 2019). Some of the topics identified by the STM include what appears to be horserace talk on polls, staffing, and fundraising. Implicitly, the measure of convergence assumes that that horserace is a legitimate part of an agenda. But perhaps, if horserace is best conceptualized as entirely separate from other topics, trends in horserace coverage underpins the observational results documented so far.

This alternative theoretical explanation goes as follows: Candidates enter the race when the election itself is still far off. Because the media does not immediately foresee a need to perform its duty of informing the public, it instead settles comfortably into reporting on the horserace, which is less frequently invoked by candidates. As the first caucuses and primaries near, however, the media begins to provide more substantive coverage to make sure voters are informed. This shift in media behavior coincides with the winnowing of the field. As such, any change in convergence is entirely tied to the media moving from horserace to substantive coverage as opposed to a reaction to the candidates’ agendas themselves.

To account for this alternative explanation, I replicate the model with two changes. First, I calculate a convergence score that omits horserace-oriented topics. Second, I control for the total proportion of the media’s agenda in that month that was horserace topics. This replication (Model 4) therefore checks if the effect of changes in the number of the candidates in the race is independent from the media’s tendency to engage in horserace coverage.

The change in the number of candidates remains a statistically significant, negative correlate with convergence even when horserace topics are omitted. Even when ignoring the horserace, this contextual factor still matters. The media’s tendency to cover the horserace does appear to come at the expense of converging toward the candidate’s agenda. When the media engages in more horserace coverage, the media’s non-horserace agenda is less reflective of the candidate’s agenda. But it does not erase the effect of change in the number of candidates.¹⁷

Conclusion

In this paper I address the question of how the lengthy, multicandidate nature of presidential primaries affects the media’s coverage of candidate agendas. I argue that this atypically long campaign and the especially large fields of candidates should create natural dissonance with the preference of journalists for stories that are timely and simple. I show that changes in the size of the field are correlated with candidate-media agenda similarity. An expansion in the primary field is correlated with a significant decrease in agenda convergence. These results line up with the theoretical argument I offer which notes that expansions of primary fields should lead to increasing complexity in the race as new electoral fault lines muddle the campaign messaging environment. This complexity comes into conflict with media preferences for simplicity in coverage, causing this decline. Further analysis suggested that this effect is primarily experienced by candidates performing poorly in the polls and is independent from the media’s tendency to engage in horserace coverage. That the results so consistently support the contextual simplicity hypothesis highlights the vital need to appreciate the role of journalism’s professional practices when contemplating how the media covers presidential primaries.

While I argued that a professional interest in timely stories should lead to declining agenda convergence over the course of the primary as the candidate’s agenda fades into old news, the results did not bear that out. The overall tendency of agenda convergence in primaries appears to be more akin to temporal stability than declining timeliness.

I demonstrated these results on a novel corpus of more than 2600 speeches by presidential primary candidates from 2000 to 2016 and more than 29,000 news articles by The New York Times and the Washington Post covering those candidates. The methods used were rigorous applications of text analysis techniques necessary to handle such a large and diverse corpus.

These findings have important implications. It is likely that the degree of success a candidate faces in conveying an agenda through the media has repercussions on public opinion (McCombs & Shaw, 1972). Additionally, candidates want to get their messages out via the media and these results suggest that a field that is not consistently expanding best positions them to do so. But it also appears only those candidates who have middling or weak support are disadvantaged and are therefore incentivized to strongly prefer more stable fields. While the frontrunners may still be able to get their messages out via the media even as the race is being shaken up, most candidates will not be in such an advantageous situation. Regardless, the candidates themselves are poorly positioned to affect the strategic incentives of their competitors. They also face a collective action problem: While competing in a crowded primary field is not ideal, the utility of candidacy trumps not entering the race for most.

Candidates need to rely on parties to structure primaries to keep the fields small and stable. I argue that parties should do so. A key duty of American political parties is facilitating the selection of nominees who are best situated to win general elections. They should thus want to be sure that the candidate who is still standing at the end of the primary process is indeed the one whose message appeals to the party’s base.

With this in mind, there does not appear to be a significant drawback, in terms of media coverage, to having a lengthy primary season. There may be other downsides, for example depletion of resources that could be reserved for the general election or voter fatigue, but looking exclusively at the effects on media coverage suggests that parties should not be overly concerned about candidates announcing that they’re running 12 months before a caucus or primary.

Yet parties should think carefully about how they can structure their nomination contests to constrain the field from consistently expanding once the most viable candidates who represent distinct perspectives within the party have entered the race. This should mean setting high standards for the avenues of legitimization, like participation in the debates, to de-incentivize marginal candidates from entering the race in hopes of catching on after some grand moment of political spectacle. By limiting the race to only viable candidates the parties can improve the odds that the nominee is chosen after clear contrasts are debated by the primary electorate.

By weakening formal control over the nomination process the parties have, perhaps unintentionally, foisted a great deal of responsibility onto the media as conduits for the selection of the parties’ standard-bearers. That does not mean the parties have absolved themselves of a role in the process. The multicandidate structure of primaries and the contextual volatility that creates meaningfully interacts with the media’s preference for simple stories. The parties should consider that in any attempt to restructure their nomination contests to be more efficacious.

This study is not without its limits. A month is a particularly blunt unit of time and its use prevents analysis that could get at the way the serial nature of primaries structures media coverage. Primaries and caucuses generally occur at weekly intervals and would therefore require daily estimates of convergence to study how temporal proximity to an electoral context affects media reactions to candidate agendas. The Presidential Primary Communication Corpus does not include enough candidate speeches to make such an analysis feasible.

Additionally, The New York Times and the Washington Post are two critical news outlets in the American political context but they do not capture the array of broadcast, cable, partisan, and public media that also make up the news ecosystem. As such, the results of this study cannot shed light on the different ways different media might react to candidate messages. Furthermore, newspaper journalists operating under the same newsworthiness values might still experience presidential primaries differently. Prior research suggests national and state newspapers differ in how they cover primaries (Flowers et al., 2003; Vinson & Moore, 2007). Perhaps the timeliness of candidate messages will indeed decline for those reporters working in newsrooms of states that elicit the most candidate attention: Iowa and New Hampshire.

This is not the first, nor will it be the last, study to note that more could be gleaned from having more data. These limitations should not obscure the contribution that this study makes. The evidence suggests an expanding primary field leads the media to diverge from candidate agendas. One need only look at the constantly shifting field of candidates that competed for the 2020 Democratic nomination to see that this can present a disruptive force in campaign politics.

Supplemental Material

LitC_supp_materials_apr_V2 – Supplemental material for Lost in the Crowd: The Effect of Volatile Fields on Presidential Primaries

Supplemental material, LitC_supp_materials_apr_V2 for Lost in the Crowd: The Effect of Volatile Fields on Presidential Primaries by Zachary Scott in American Politics Research

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Zachary Scott

Supplemental Material

Supplemental material for this article is available online.

Notes

Author Biography

Zachary Scott is a visiting assistant professor of Political Science at Bryn Mawr College. His research interests include political communication, mass media, and campaigns and elections.

References

Aldrich

J. H.

(1980). A dynamic model of presidential nomination campaigns. American Political Science Review, 74(3), 651–669.

Azari

J. R.

(2016). How the news media helped to nominate trump. Political Communication, 33(4), 677–680.

Bartels

L. M.

(1988). Presidential primaries and the dynamics of public choice. Princeton University Press.

Boydstun

A. E.

(2013). Making the news: Politics, the media, and agenda setting. Chicago University Press.

Conway

B. A.

Kate

(2015). The rise of twitter in the political campaign: Searching for intermedia agenda-setting effects in the presidential primary. Journal of Computer-Mediated Communication, 20(4), 363–380.

Conway-Silva

B. A.

Filer

C.R.

Kenski

Tsetsi

(2018). Reassessing twitter’s agenda-building power: An analysis of intermedia agenda-setting effects during the 2016 presidential primary season. Social Science Computer Review, 36(4), 469–483.

Donsbach

(2004). Psychology of news decisions: Factors behind journalists’ professional behavior. Journalism, 5(2), 131–157.

Dunaway

(2008). Markets, ownership, and the quality of campaign news coverage. The Journal of Politics, 70(4), 1193–1202.

Dunaway

Lawrence

R. G.

(2015). What predicts the game frame? Media ownership, electoral context, and campaign news. Political Communication, 32(1), 43–60.

10.

Feezell

J. T.

(2018). Agenda setting through social media: The importance of incidental news exposure and social filtering in the digital era. Political Research Quarterly, 71(2), 482–494.

11.

Flowers

J. F.

Haynes

A. A.

Crespin

M. H.

(2003). The media, the campaign, and the message. American Journal of Political Science, 47(2), 259–273.

12.

Gans

H. J.

(1979). Deciding what’s news: A study of CBS evening news, NBC nightly news, newsweek, and time. Random House, Inc.

13.

Grimmer

Stewart

B. M.

(2013). Text as data: The promise and pitfalls of automatic content analysis methods for political texts. Political Analysis, 21(3), 267–297.

14.

Groseclose

Milyo

(2005). A measure of media bias. The Quarterly Journal of Economics, 120(4), 1191–1237.

15.

Hamilton

J. T.

(2004). All the news that’s fit to sell: How the market transforms information into news. Princeton University Press.

16.

Harcup

O’Neill

(2001). What is news? Galtung and ruge revisited. Journalism Studies, 2(1), 261–280.

17.

Hayes

(2010). The dynamics of agenda convergence and the paradox of competitiveness in presidential campaigns. Political Research Quarterly, 63(3), 594–611.

18.

Haynes

A. A.

Flowers

J. F.

Gurian

P. H.

(2002). Getting the message out: Candidate communication strategy during the invisible primary. Political Research Quarterly, 55(3), 633–652.

19.

Haynes

A. A.

Gurian

P. H.

Crespin

M. H.

Zorn

(2004). The calculus of concession: Media coverage and the dynamics of winnowing in presidential nominations. American Politics Research, 32(3), 310–337.

20.

Haynes

A. A.

Rhine

S. L.

(1998). Attack politics in presidential nomination campaigns: An examination of the frequency and determinants of intermediated negative messages against opponents. Political Research Quarterly, 51(3), 691–721.

21.

Heldman

Carroll

S. J.

Olson

(2005). ‘She brought only a skirt’: Print media coverage of elizabeth dole’s bid for the republican presidential nomination. Political Communication, 22(3), 315–335.

22.

Iyengar

Kinder

(1987). News that matters: Television and American opinion. University of Chicago Press.

23.

Keele

Kelly

N. J.

(2006). Dynamic models for dynamic theories: The ins and outs of lagged dependent variables. Political Analysis, 14(2), 186–205.

24.

Kendall

K. E.

(2000). Communication in the presidential primaries: Candidates and the media, 1912-2000. Greenwood Publishing Group.

25.

Kaid

L. L.

Pfau

(2003). The impact of web site campaigning on traditional news media and public information processing. Journalism & Mass Communication Quarterly, 80(3), 528–547.

26.

Lee

J. H.

(2009). News values, media coverage, and audience attention: An analysis of direct and mediated causal relationships. Journalism & Mass Communication Quarterly, 86(1), 175–190.

27.

Lee

Mimno

(2014). Low-dimensional embeddings for interpretable anchor-based topic inference. In Proceedings of the 2014 conference on empirical methods in natural language processing, 1319–1328.

28.

McCombs

M. E.

Shaw

D. L.

(1972). The agenda-setting function of mass media. The Public Opinion Quarterly, 36(2), 176–187.

29.

Miller

M. M.

Andsager

J. L.

Riechert

B. P.

(1998). Framing the candidates in presidential primaries: Issues and images in press releases and news coverage. Journalism & Mass Communication Quarterly, 75(2), 312–324.

30.

Niven

(2004). A fair test of media bias: Party, race, and gender in coverage of the 1992 house banking scandal. Polity, 36(4), 637–649.

31.

Norrander

(2000). The end game in post-reform presidential nominations. The Journal of Politics, 62(4), 999–1013.

32.

Norrander

(2006). The attrition game: Initial resources, initial contests and the exit of candidates during U.S. presidential primary season. British Journal of Political Science, 36(3), 487–507.

33.

Norrander

(2010). The imperfect primary: Oddities, biases, and strengths of U.S. presidential nomination politics. Routledge.

34.

Patterson

T. E.

(2016). News coverage of the 2016 presidential primaries: Horse race reporting has consequences. https://shorensteincenter.org/news-coverage-2016-presidential-primaries/

35.

Peters

Woolley

J. T.

(2019). The American presidency project. http://www.presidency.ucsb.edu/

36.

Pfetsch

(1999). Government news management - strategic communication in comparative perspective. Science Center Berlin for Social Research.

37.

Polsby

N. W.

(1983). Consequences of party reform. Oxford University Press.

38.

Protess

McCombs

(1991). Agenda setting: Readings on media, public opinion, and policymaking. Lawrence Earlbaum.

39.

Ridout

T. N.

Holland

J. L.

(2010). Candidate strategies in the presidential nomination campaign. Presidential Studies Quarterly, 40(4), 611–630.

40.

Roberts

M. E.

Stewart

B. M.

Tingley

Lucas

Leder-Luis

Gadarian

S. K.

Albertson

Rand

D. G.

(2014). Structural topic models for open-ended survey responses. American Journal of Political Science, 58(4), 1064–1082.

41.

Roberts

McCombs

(1994). “Agenda setting and political advertising: Origins of the news agenda. Political Communication, 11(3), 249–262.

42.

Sanders

Crespo

M. J. C.

Holtz-Bacha

(2011). Communicating governments: A three-country comparison of how governments communicate with citizens. The International Journal of Press/Politics, 16(4), 523–547.

43.

Searles

Banda

K. K.

(2019). But her emails! How journalistic preferences shaped election coverage in 2016. Journalism, 20(8), 1052–1069.

44.

Sigelman

Buell

E. H.

(2004). Avoidance or engagement? Issue convergence in U.S. presidential campaigns, 1960–2000. American Journal of Political Science, 48(4), 650–661.

45.

Soroka

S. N.

(2012). The gatekeeping function: Distributions of information in media and the real world. The Journal of Politics, 74(2), 514–528.

46.

Steger

W. P.

Hickman

Yohn

(2002). Candidate competition and attrition in presidential primaries, 1912-2000. American Politics Research, 30(5), 528–554.

47.

Tuchman

(1978). Making news: A study in the constructing of reality. The Free Press.

48.

Usher

(2014). Making news at the New York Times. University of Michigan Press.

49.

Vavreck

(2009). The message matters: The economy and presidential campaigns. Princeton University Press.

50.

Vinson

C. D.

Moore

W. V.

(2007). The campaign disconnect: Media coverage of the 2000 South Carolina presidential primary. Political Communication, 24(4), 393–413.

51.

Zaller

J. R.

(1999). A theory of media politics: How the interests of politicians, journalists, and citizens shape the news. (Unpublished manuscript).

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.59 MB