This paper uses analyses of commercial polls alongside content-analytic measures of sentiment in the content of nine newspapers to explore the relationship between voter preferences and the tone of news coverage in the 2016 presidential election campaign. Both media coverage and voter preferences reflected the effects of certain campaign events—the conventions and the initial Comey intrusion—and there also is evidence of a relationship between the two. Indeed, it appears that the media both led and followed public preferences throughout much of the campaign, though evidence of followership actually is more robust; and the final weeks of the campaign show little to no media effects at all. Results speak to the importance of considering media not just as a driver, but also a follower of public sentiment.
This is a preview of subscription content, log in to check access.
Buy single article
Instant access to the full article PDF.
Price includes VAT for USA
Subscribe to journal
Immediate online access to all issues from 2019. Subscription will auto renew annually.
This is the net price. Taxes to be calculated in checkout.
The fundamentals come in two varieties, one “internal” to voters, such as party identification, and the other “external,” such as the national economy (Erikson and Wlezien 2012).
For a useful review of work on media in election campaigns, see Graber and Dunaway (2017).
There is a good deal of work focused on contingencies in media effects, after all. Some citizens are more open to media effects than others, due for instance to differences in preferences and exposure, in terms of both volume and selectivity (e.g., Eveland et al. 2003; Hillygus and Jackman 2003). A central finding in the last thirty years of research in political communication is that there is heterogeneity in media effects, across both issues and people.
Polling focuses on representative samples, and takes several days to gather; media coverage focuses on hyper-attentive journalists, and can be gathered nearly instantly. It seems reasonable to expect that a measure of news coverage might lead a polls-based measure for these reasons alone.
Note that a view of news coverage as being responsive rather than uni-directionally causal is relatively common in work focuses on media-elite relationships (for a review, see Cohen 2009), and it is evident in work focused on the role of media in policy agenda-setting as well (for a review, see Wolfe et al. 2013). We are not focused on an elite audience, however, and it may be more difficult to imagine a leading role for regular citizens.
There is of course a good deal of work suggesting the potential impact of campaign ads (see, e.g., Ansolabehere and Iyengar 1997; Gilens et al. 2007; Shaw 1999b), though the overall influence of ads given increases in micro-targetting may not be the ads themselves so much as media coverage of those ads (Ridout and Smith 2008).
This seems like a distinct possibility given the considerable literature on biases in attentiveness to news (e.g., Soroka 2014).
For surveys in the field for an even number of days, the fractional midpoint is rounded up to the following day. There is a good amount of variation in the number of days surveys are in the field: The mean number of days is 4.5; the standard deviation is 1.8 days.
Restricting to houses with fewer polls makes little statistical difference to the resulting portrait of preferences or the analyses that follow.
In the analysis, we adjust the number of respondents for the NBC/Survey Monkey polls, which were comparatively massive, ranging from 6000 to 70,000 and averaging nearly 20,000. In order to prevent those polls from carry extraordinary weight in our analysis, we used the mean N of 1389 for all other pre-election polls in 2016.
The CNN estimate is virtually indistinguishable from that for Fox—they differ by 0.01 of a point—so choosing one or the other make no perceptible difference. Although centering on the median survey house may seem reasonable, it may not be quite right; the problem is that we cannot tell for sure. It may be tempting to use the house that best predicted the final outcome, though this is even more tenuous given that all polls contain error, i.e., getting it right at the end is as much the result of good luck as it is good survey design.
Note that our approach here differs from that in previous research, which used coefficients for the survey date variables in the ANOVA as estimates of electoral preferences (Erikson and Wlezien 1999). Our approach differs only slightly—the correlation between the two series of estimates is .91—but allows greater comparability with the “pooling” of polls based on each date they are in the field instead of the middle date of the reported polling period.
This focus is understandable for a number of reasons. First, we know that conventions and debates are very visible, where large numbers of people watch on televisions and/or acquire information about them in other ways. Second, we can anticipate these events, so our interpretation of their effects is not subject to the after-the-fact reduction that characterizes interpretations of the seeming effects of many other campaign events. Third, there already is evidence that they matter a lot more than other events, or at least that they can.
Note also that results in Table 1, as well as those in subsequent tables, do not change significantly if we shift the starting point for the latter campaign period closer to the conventions, up to 3 weeks before Labour Day.
The pattern of correlations is what we might expect of a so-called combined process, where some effects last indefinitely and others decay (Wlezien 2000).
The average number of articles per day, by month, is as follows: Junuary, 63; February, 79; March, 92; April, 75; May, 81; June, 83; July, 119; August, 85; September, 97; October, 127; November (-8th), 185. Note that there are several days for which articles were not returned in the Nexis search, and this accounts for slightly different Ns in Tables 3, 4, and 5.
Druckman (2005) finds, for instance, that while the quantity of news is far greater in newspapers, the content is not very different. Also see footnote 25.
The dictionary includes roughly 4567 entries in total, so full details are available via lexicoder.com. Even so, by way of example, negative words include, “cheat,” “harm,” and “reckless,” and positive words include, “dependable,” “honest,” and “resourceful.”
Note that there are other possible options, including using positive and negative words in (a) paragraphs that mention parties, or (b) articles than mention parties. These larger “windows” tend to capture tone that is not necessarily related to the parties themselves. Postive words about one party at the start of an article/paragraph should not in most circumstances be counted as tone directed at a party mentioned at the end of an article/paragraph, for instance. This intuition is confirmed by our own reading of texts; and sentence-level coding is in line with some previous work on sentiment, particularly but not exclusively in the campaign context (e.g., van Atteveldt et al. 2008; Belanger and Soroka 2012; Haselmayer and Jenny 2017).
Note that over-time variation in the volume of news (see footnote 19), as well as the possibility that the amount of candidate coverage may be imbalanced at certain times, points to the potential advantages of volume-weighted measures of news content. Multiplying our existing measure by the number of stories each day produces a measure that straightforwardly deals with the amount of coverage. Results using such a (volume-weighted) measure do not differ significantly from what we present below, though in some cases they are slightly stronger (see footnote 33 for details). But weighting by volume does complicate analyses of media as the dependent variable, as we expect levels of coverage to reflect events more than preferences, and tone of coverage to reflect preferences. Given our emphasis on the reciprocal relationship between news and opinion, we rely on the straightforward measure of tone here. Although we rely also on a combined “Clinton minus Trump” measures of tone, we also report below (in footnote 33) the results using candidate-specific ones.
Previous tests suggest that this measure is very highly correlated with others, with r > 0.95.
Note that this finding is similar to what was found in work by Media Tenor, reported by the Shorenstein Center (https://shorensteincenter.org/news-coverage-2016-general-election/). They find roughly similar proportions of negative content for Clinton and Trump in the final months of the campaign, but more negative content for Trump from January to November, the period analysed here. Given that Media Tenor data rely on different sources (including television news), and a different approach to content analysis, we take the similarity in results as a further indication of the validity of our measure.
Not surprisingly, the correlations for the pre-Labor Day period actually are larger—0.39 and 0.41, respectively, for media tone at day t − 2 and the associated 3-day moving average—than they are for the full year.
Note that this first-order ADL (1, 1) specification is a specific from of vector autoregression (VAR). We rely on the ADL model here because we are interested in the effects of numerous political events (in Tables 4 and 5), and using a general VAR there would require estimating a very large number of parameters, and so would be (radically) inefficient. That said, estimation of basic models (as in Table 3) using third-order VAR confirms a symmetric effect between electoral preferences and media coverage, i.e., that the latter both lead and follow the former. (It also implies that the effect of preferences may be very short-lived.) Note that we are interested in how the bi-directional relationship between polls and the media plays out over longer stretches of time, and how this may differ over the course of the campaign, and explicitly consider this below (see Table 6).
Recall that polls on a particular day, regardless of how we date them, almost always are conducted over multiple days. (See footnote 10.) Measured vote intentions thus reflect responses during a window in time, e.g., measured intentions on day t from a 4-day poll will be based on interviews on days t − 2, t − 1, t, and t + 1. And our approach to “pooling” also introduces temporal dependence. This complicates our analysis in two conflicting ways: (1) since intentions on day t − 1 are not completely prior to media coverage on day t, it is harder to credit effects of the former on the latter; (2) since vote intentions on day t − 1 are to some extent registered prior to media coverage on day t − 1, it is harder to detect effects of the former on day t coverage, i.e., those effects are partly captured in lagged tone. Similar issues pertain to estimating the effects of media tone on vote intentions. Although we think the analyses are informative, we are cautious when interpreting the results.
In the polls equation, the Democratic Convention variable is not lagged but is measured currently, at time t, as this substantially increases the size and significance of the estimated coefficient. Using the lagged specification, the estimate is small and not close to statistically significant.
In the polls equation, the lagged July Comey Announcement and First Debate variables are themselves lagged, as this maximizes the estimated coefficient size and model fit.
This only partly reflects our approach to pooling polls, as the coefficient (0.41) on lagged polls is larger even when using polls aggregated based on the middle date of the reported polling period.
It is worth noting two important diagnostics. First, weighting tone by the volume of coverage produces similar results in analysis of poll results. The volume-weighted measure does more reliably predict the polls in basic ADL models, e.g., Table 3, but this is not the case when events are included. Second, estimating the separate effects of Clinton and Trump tone on polls also produces similar results. The former matters and the latter does not in models excluding events; neither are significant predictors when including events.
Of course, it may be that media tone did not actually influence voter preferences even for these events, and that both the media and the public responded independently. While possible, this is difficult to credit given the reliance on media coverage, particularly for the Comey announcement and even the nominating conventions.
While the effect of the Comey letter does not meet conventional standards of statistical significance, the estimate is sizable, about half that of the July Comey announcement. Given the late timing of the event and the persistence of poll shares over time, the intrusion may have impacted the final result. Indeed, given the closeness of the race in Pennsylvania, Michigan, and Wisconsin, it may have been decisive.
Note that this finding is in line with other recent work, e.g., Enns et al. (2017).
This may appear to be less than we might expect given the variation in results from different firms and previous elections (see, e.g., Wlezien 2003), which partly reflects the focus on firms with five or more polls in the field during the year. When including all polling organizations regardless of the number of polls they conducted, house effects are quite a bit larger, approaching 6.8 points.
AAPOR. (2017). An evaluation of 2016 election polls in the United States. A report by the American Association for Public Opinion Research Ad Hoc Committee on 2016 election polling.
Alvarez, R. M. (1997). Information and elections. Ann Arbor: University of Michigan Press.
Ansolabehere, S., & Iyengar, S. (1997). Going negative: How political advertisements shrink and polarize the electorate. New York: The Free Press.
Bartels, L. M. (1993). Messages received: The political impact of media exposure. American Political Science Review,87(2), 267–285.
Belanger, E., & Soroka, S. (2012). Campaigns and the prediction of election outcomes: Can historical and campaign-period prediction models be combined? Electoral Studies,31, 702–714.
Brady, H., & Johnston, R. (Eds.). (2006). Capturing campaign effects. Ann Arbor, MI: University of Michigan Press.
Campbell, J. E. (2000). The American campaign: US presidential campaigns and the national vote. College Station, TX: Texas A&M University Press.
Campbell, J. E. (2017). A recap of the 2016 election forecasts. PS: Political Science and Politics, 50(2), 331–332.
Cohen, J. E. (2009). The presidency and the mass media. Oxford Handbooks Online. Retrieved June 28, 2017, from http://www.oxfordhandbooks.com/view/10.1093/oxfordhb/9780199238859.001.0001/oxfordhb-9780199238859-e-12. Accessed Dec 2016.
Converse, P. E., & Traugott, M. W. (1986). Assessing the accuracy of polls and surveys. Science,234, 1094–1098.
Crespi, I. (1988). Pre-election polling: Sources of accuracy and error. New York: Russell Sage.
Daku, M., Soroka, S., & Young, L. (2015). Lexicoder, version 3.0. www.lexicoder.com. Accessed Dec 2016.
de Vreese, C. H., & Semetko, H. A. (2004). News matters: influences on the vote in the Danish 2000 Euro referendum campaign. European Journal of Political Research,43, 699–722.
Druckman, J. N. (2004). Priming the vote: Campaign effects in a U.S. Senate election. Political Psychology,25, 577–594.
Druckman, J. N. (2005). Media matter: How newspapers and television news cover campaigns and influence voters. Political Communication,22(4), 463–481.
Enns, P. K., Lagodny, J., & Schuldt, J. P. (2017). Understanding the 2016 US presidential polls: The importance of hidden trump supporters. Statistics, Politics and Policy,8(1), 41–63.
Erikson, R. S., & Sigelman, L. (1995). Poll-based forecasts of midterm congressional elections: Do the pollsters get it right? Public Opinion Quarterly,59, 589–605.
Erikson, R. S., & Sigelman, L. (1996). Poll-based forecasts of the house vote in presidential election years. American Politics Quarterly,24, 52–531.
Erikson, R. S., & Wlezien, C. (1999). Presidential polls as a time series: The case of 1996. Public Opinion Quarterly,63, 163–177.
Erikson, R. S., & Wlezien, C. (2012). The timeline of presidential elections. Chicago: University of Chicago Press.
Eveland, W. P., Shah, D. V., & Kwak, N. (2003). Assessing causality in the cognitive mediation model. Communication Research,30(4), 359–386.
Fan, D. P. (1988). Predictions of public opinion from the mass media: Computer content analysis and mathematical modeling. New York: Greenwood.
Gelman, A., & King, G. (1993). Why are American presidential election polls so variable when votes are so predictable? British Journal of Political Science,23, 409–519.
Gentzkow, M., & Shapiro, J. M. (2010). What drives media slant? Evidence from US daily newspapers. Econometrica,78(1), 35–71.
Gilens, M., Vavreck, L., & Cohen, M. (2007). The mass media and the public’s assessments of presidential candidates, 1952–2000. Journal of Politics,69(4), 1160–1175.
Graber, D. A., & Dunaway, J. (2017). Mass media and american politics (10th ed.). Washington, DC: CQ Press.
Groves, R. M. (1989). Survey errors and survey costs. New York: Wiley.
Hamilton, J. (2004). All the news that’s fit to sell: How the market transforms information into news. Princeton NJ: Princeton University Press.
Hansen, K. M., & Pedersen, R. T. (2014). Campaigns matter: How voters become knowledgeable and efficacious during election campaigns. Political Communication,31(2), 303–324.
Hardy, B. W., & Scheufele, D. A. (2009). Presidential campaign dynamics and the Ebb and flow of talk as a moderator: Media exposure, knowledge, and political discussion. Communication Theory,19(1), 89–101.
Haselmayer, M., & Jenny, M. (2017). Sentiment analysis of political communication: Combining a dictionary approach with crowdcoding. Quality & Quantity,51(6), 2623–2646.
Heise, D. R. (1969). Separating reliability and stability in test-retest correlations. American Sociological Review,34, 93–101.
Hillygus, D. Sunshine. (2005). Campaign effects and the dynamics of turnout intention in election 2000. Journal of Politics,67(1), 50–68.
Hillygus, D. Sunshine., & Jackman, S. (2003). Voter decision making in election 2000: Campaign effects, partisan activation, and the clinton legacy. American Journal of Political Science,47(4), 583–596.
Holbrook, T. (1996). Do campaigns matter?. Thousand Oaks, CA: Sage.
Hopmann, D. N., Vliegenthart, R., de Vreese, C., & Albaek, E. (2010). Effects of election news coverage: How visibility and tone influence party choice. Political Communication,27, 389–405.
Iyengar, S., Norpoth, H., & Hahn, K. S. (2004). Consumer demand for election news: The horserace sells. Journal of Politics,66(1), 157–175.
Jamieson, K. H. (1996). Packaging the presidency: A history and criticism of presidential campaign advertising (3rd ed.). New York: Oxford University Press.
Jennings, W., & Wlezien, C. (2016). The timeline of elections: A comparative perspective. American Journal of Political Science,60(1), 219–233.
Johnston, R., Blais, A., Brady, H. E., & Crete, J. (1992). Letting the people decide: Dynamics of a Canadian election. Kingston, Canada: McGill-Queen’s Press.
Johnston, R., Hagen, M. G., & Jamieson, K. H. (2004). The 2000 presidential election and the foundations of party politics. Cambridge, MA: Cambridge University Press.
Just, M. R., Crigler, A. N., Alger, D. E., Cook, T. E., Kern, M., & West, D. M. (1996). Crosstalk: Citizens, candidates, and the media in a presidential campaign. Chicago, IL: University of Chicago Press.
Kennedy, C., Blumenthal, M., Clement, S., Clinton, J. D., Durand, C., Franklin, C., et al. (2018). An evaluation of the 2016 polls in the United States. Public Opinion Quarterly,82(1), 1–33.
Lau, R. (1994). An analysis of the accuracy of ‘trial heat’ polls during the 1992 presidential election. Public Opinion Quarterly,58, 2–20.
Lewis-Beck, M. (1988). Economics and elections: The major western democracies. Ann Arbor: University of Michigan Press.
Lowe, W., Benoit, K., Mikhaylov, S., & Laver, M. (2011). Scaling policy preferences from coded political texts. Legislative Studies Quarterly,36(1), 123–155.
Mendelsohn, M., & Nadeau, R. (1999). The rise and fall of candidates in Canadian election campaigns. International Journal of Press/Politics,4, 63–76.
Nadeau, R., & Lewis-Beck, M. (2012). Does a presidential candidate’s campaign affect the election outcome? Foresight,24, 15–18.
Nadeau, R., Nevitte, N., Gidengil, E., & Blais, A. (2008). Election campaigns as information campaigns: Who learns what and does it matter? Political Communication,25, 229–248.
Norris, P., Curtice, J., Sanders, D., Scammell, M., & Semetko, H. A. (1999). On message: Communicating the campaign. London: Sage.
Plasser, F., & Plasser, G. (2002). Global political campaigning: A worldwide analysis of campaign professionals and their practices. Westport, CT: Greenwood.
Proksch, S.-O., Lowe, W., & Soroka, S. (2016). Multilingual sentiment analysis: A new approach to measuring conflict in parliamentary speeches. Paper presented at the Annual Meeting of the American Political Science Association, Philadelphia PA.
Reuning, K., & Dietrich, N. (2016). Media coverage, public interest, and support in primary elections. Working paper, SSRN. https://ssrn.com/abstract=2709208 or http://dx.doi.org/10.2139/ssrn.2709208. Accessed Dec 2016.
Ridout, T. N., & Smith, G. R. (2008). Free advertising: How the media amplify campaign messages. Political Research Quarterly,61(4), 598–608.
Schudson, M. (1989). The sociology of news production. Media, Culture and Society,11, 263–282.
Shaw, D. R. (1999a). A study of presidential campaign event effects from 1952 to 1992. Journal of Politics,61, 387–422.
Shaw, D. R. (1999b). The effect of TV ads and candidate appearances on statewide presidential votes, 1988-96. The American Political Science Review,93(2), 345–361.
Shaw, D. R. (1999c). The impact of news media favorability and candidate events in presidential campaigns. Political Communication,16(2), 183–202.
Sides, J., & Vavreck, L. (2014). The gamble: Choice and chance in the 2012 presidential election. Princeton, NJ: Princeton University Press.
Soroka, S. (2014). Negativity in democratic politics: Causes and consequences. Cambridge Studies in Public Opinion and Political Psychology, Cambridge University Press.
Soroka, S., Bodet, M. A., Young, L., & Andrew, B. (2009). Campaign news and vote intentions. Journal of Elections, Public Opinion and Parties,19(4), 359–376.
Soroka, S. N., Stecula, D. A., & Wlezien, C. (2015). It’s (change in) the (future) economy, stupid: Economic indicators, the media, and public opinion. American Journal of Political Science,59, 457–474.
Sumpter, R. S. (2000). Daily newspaper editors’ audience construction routines: A case study. Critical Studies in Media Communication,17(3), 334–346.
Trussler, M., & Soroka, S. (2014). Consumer demand for cynical and negative news frames. International Journal of Press and Politics, 19(3), 360–379.
van Atteveldt, W., Kleinnijenhuis, J., Ruigrok, N., & Schlobach, S. (2008). Good news or bad news? Conducting sentiment analysis on Dutch text to distinguish between positive and negative relations. Journal of Information Technology & Politics,5(1), 73–94.
van der Meer, T. W. G., Walter, A., & Van Aelst, P. (2016). The contingency of voter learning: How election debates influenced voters’ ability and accuracy to position parties in the 2010 dutch election campaign. Political Communication,33(1), 136–157.
Vavreck, L. (2009). The message matters. Princeton, NJ: Princeton University Press.
Weaver, D. H. (1996). What voters learn from media. Annals of the American Academy of Political and Social Science,546, 34–47.
West, D. M., Kern, M., Alger, D., & Goggin, J. M. (1995). Ad buys in presidential campaigns: The strategies of electoral appeal. Political Communication,12(3), 275–290.
Wlezien, C. (2000). An essay on ‘combined’ time series processes. Electoral Studies,19(1), 77–93.
Wlezien, C. (2003). Presidential election polls in 2000: A study in dynamics. Presidential Studies Quarterly,33(1), 172–186.
Wlezien, C., & Erikson, R. S. (2001). Campaign effects in theory and practice. American Politics Research,29, 419–437.
Wlezien, C., & Erikson, R. S. (2002). The timeline of presidential election campaigns. Journal of Politics,64(4), 969–993.
Wlezien, C., Jennings, W., Fisher, S., Ford, R., & Pickup, M. (2013). Polls and the vote in Britain. Political Studies,61(1), 66–91.
Wlezien, C., & Morris, G. E. (2017). Dynamics of (national) electoral during the 2016 US presidential race. In A. Cavari, R. Powell, & K. Mayer (Eds.), The 2016 presidential election: The causes and consequences of an electoral earthquake. Lanham, MD: Lexington Books.
Wolfe, M., Jones, B. D., & Baumgartner, F. R. (2013). A failure to communicate: agenda setting in media and policy studies. Political Communication,30, 175–192.
Young, L., & Soroka, S. (2012). Affective news: The automated coding of sentiment in political texts. Political Communication,29, 205–231.
Table 7 contains the results of analysis of polling universe and house effects. In the table, we can see that the general polling universe did not meaningfully affect poll results during 2016, controlling for survey house and date. This is not entirely surprising, as the same was true of polls in previous election years (see Erikson and Wlezien 1999; Wlezien 2003). The table also shows that survey house did matter in 2016. For the full election year, the range of the house effects is just less than 4.0 percentage points.Footnote 38 These are big differences and ones that are difficult to fully explain given the available information about the practices of different survey houses. There is reason to suppose that the observed differences across houses largely reflect underlying differences in design (Wlezien and Erikson 2001), though we cannot be sure. Whatever the source, the differences have consequences for our portrait of preferences during 2016, as poll results will differ from day-to-day merely because different houses report on different days. Also notice in Table 7 that survey date effects easily meet conventional levels of statistical significance (p < .001) during the full election year. Separate analysis (not reported here) shows that date effects are less apparent (p = .09) during the post-Labor Day period, which implies that electoral preferences changed mostly over the course of the long campaign.
About this article
Cite this article
Wlezien, C., Soroka, S. Mass Media and Electoral Preferences During the 2016 US Presidential Race. Polit Behav 41, 945–970 (2019). https://doi.org/10.1007/s11109-018-9478-0
- Electoral preferences
- Campaign effects
- Mass media