Content uploaded by Christopher Wlezien
Author content
All content in this area was uploaded by Christopher Wlezien on Feb 06, 2017
Content may be subject to copyright.
A Cross-National Analysis of the Causes
and Consequences of Economic News*
Christopher Wlezien Stuart Soroka Dominik Stecula
University of Texas at Austin University of Michigan University of British Columbia
* Direct all correspondence to Christopher Wlezien, Department of Government, Batts Hall,
University of Texas at Austin, Austin, TX 78712, Wlezien@austin.utexas.edu. A previous version of
this paper was presented at “The New World of Comparative Political Communication”
Conference, Texas A&M University, May 2016. We are grateful to the organizers – Keith Gaddie,
Kirby Goidel, and Guy Whitten – as well as the other participants for comments, especially Paul
Kellstedt and Sofie Marien.
Abstract
Objective: Work on economic news argues that US coverage focuses primarily on changes
rather than levels of future economic conditions; it also both affects and reflects public
economic sentiment. Given that economic perceptions are related to policy preferences and
government support, this is of consequence for politics. This paper explores the generalizability
of these findings.
Methods: Using nearly 100,000 stories over 30 years in the US, UK, and Canada, we compare
media tone, public opinion and economic conditions.
Result: Results demonstrate that media tone and public opinion follow future economic change
in all three countries. Media and opinion are also related, but the effect mostly runs from the
public to the media, not the other way around.
Conclusion: These results confirm the generalizability of prior findings, and the importance of
considering more than a simple uni-directional link between media coverage and public
economic sentiment.
2
Public perceptions of the state of the economy play an important role in politics, both in the
United States and elsewhere. Yet relatively little is known about the sources of these attitudes.
Economic sentiment cannot be based solely on the real national economy, since most individuals
have direct experience with only a small part of it. At least part of what citizens know about the
economy seems likely to come from the mass media, then; and there is a long-standing and
growing body of work finding a connection between media coverage and public attitudes about
the economy (e.g., Behr and Iyengar 1985; DeBoef and Kellstedt 2004; Goidel and Langley
1995; Ju 2008; Nadeau et al. 1999; Soroka 2006). The nature of coverage of economic news
nevertheless remains underexplored, especially in the comparative context. In this paper, we aim
to broaden our understanding of the causes and consequences of economic news coverage, and
its relationship to public perceptions about the economy, in three countries: Canada, the UK, and
the US.
We build specifically on our previous work (Soroka, Stecula and Wlezien 2015) that finds
evidence that US media coverage of the economy tends to (a) focus on future rather than the
current or past economic conditions, and (b) react mainly to changes in rather than levels of
those conditions. This appears to be true not just for media coverage, but public economic
sentiment as well, which both responds to and affects coverage. The possibility that media
coverage is driven by public sentiment has received little attention in past work – the tendency is
simply to regard media as a driver of public sentiment.1 But there are good reasons to expect a
reciprocal relationship. Journalists regularly report on the state of public sentiment, after all, as it
is captured in poll reports, but also more generally in economic coverage. Indeed, media
organizations have been responsible for many public opinion polls, and presumably for a reason.
3
So it should not be surprising that we found strong evidence that public opinion matters to
economic reporting, at least in the US.
The finding of a reciprocal connection between media coverage and public sentiment, alongside
results suggesting the significance of prospective changes in conditions, are of some significance
given the established importance of economic sentiment on policy and spending preferences
(e.g., Durr 1993; Stevenson 2001; Wlezien 1995; Erikson, MacKuen, and Stimson 2002; Soroka
and Wlezien 2010), as well as government approval and election outcomes (e.g., Hibbs 1987;
Erikson 1989; Erikson and Wlezien 2012; Nadeau et al. 1999; Sanders, Marsh, and Ward 1993;
for reviews see Lewis-Beck and Stegmaier 2000, 2007). The political significance of economic
sentiment is by no means restricted to the US either – there is a vast body of work, across a wide
range of countries, establishing connections between economic sentiment on political behavior
(e.g., Nadeau et al. 2012; Duch and Stevenson 2008; van der Brug et al. 2007).
Whether prior findings regarding the nature of economic news coverage generalize beyond the
US is nevertheless unclear. There is a small, but growing, body of work considering the nature
of economic news outside the US (e.g., Kalogeropoulos et al. 2015; van Dalen et al 2015; de
Vreese et al. 2001; Ju 2008). Rarely are there direct comparisons across countries, however. A
small collection of research provides some hints about whether and when we should expect the
nature of economic news to be different across countries. Lischka (2014) suggests that economic
news content varies with the revenue incentives of news organizations, for instance, a domain-
specific illustration of more general arguments about the differences between commercial and
public broadcasters (e.g., Soroka et al. 2013).2 Other research exploring role conceptions and
practices amongst journalists suggests the possibility of both similarity and difference. On the
one hand, the literature points to convergence across nations towards a global journalist culture
4
rooted in similar notions of impartiality and critical reporting. On the other hand, work relying
on surveys of journalists in particular serves to highlight some potentially important cross-
national differences in approach. (The literature is considerable and growing, but see, e.g.,
Hallin and Mancini 2004; van Dalen et al. 2012; Hanitzch et al. 2011; Hanitzsch and Mellado
2011; Waheed et al. 2013).
We thus do not yet have a good sense for whether we should expect cross-national differences
but we have some preliminary interests. Building in part on the work cited above, we are
interested in, for example, the possibility that more competitive media markets produce different
– perhaps more critical, or sensationalistic – approaches to economic news, although we are
aware that the three countries we examine here are all part of the same “liberal” media model in
the Hallin and Mancini’s typology. We wonder whether institutional forces and varying
journalistic norms will produce different approaches to leading or following public opinion. We
also wonder whether the competitiveness of elections matters. We have argued that increased
political competition may produce clearer signals about policy change (Soroka and Wlezien
2010), and the same may be true for the economy. Political systems that impede or enhance
accountability may also give different weight to retrospective versus prospective economic
evaluations, in media content and/or public opinion.
This is all just speculation, however. As this volume makes clear, there is very little directly
comparable work in political communication. All we can reliably say at this point is that the
existing literature does not point towards clear expectations where the change-oriented,
prospective emphasis of news coverage, or the reciprocal (rather than uni-directional)
relationship between news content and public sentiment are concerned. We thus take a first step
down this path below, offering what we hope will be a preliminary, research-stimulating
5
exploration of the degree to which these US findings are generalizable to two other Anglo-
American countries: the UK and Canada. Our interest in these countries is both pragmatic and
substantive. Pragmatically speaking, our use of automated content analysis to derive sentiment in
news coverage depends on using English-speaking media; and a comparison of this content with
public opinion depends on long-term trends in public economic expectations. Substantively
speaking, there are established literatures in each of these countries on both economic voting and
media influence on political and economic attitudes. (On the UK, see, e.g., Clarke and Stewart
1995; Nadeau, Niemi, and Amato 1996; Price and Sanders 1993; Sanders et al. 1993; on Canada,
see, e.g., Happy 1986; Belanger and Soroka 2012.) Just as for the US, then, we know that
economic news coverage has significant political implications in the UK and Canada.
Given the paucity of work on the elements of economic news coverage investigated here, we
have no strong expectations about whether prior findings will be reflected in the UK and Canada.
There is some variation in both media competitiveness and journalistic cultures, even within this
entirely Anglo-American set. Even within the developed world, however, what we have here is a
study of more similar rather than different systems. Does economic news coverage exhibit the
same systematic tendencies across all three countries? This is the focus of the analyses that
follow, alongside diagnostic work comparing results across three different approaches to
measuring sentiment in economic news coverage. In a concluding discussion, we reconsider our
findings in light of the literatures on economic news and journalistic approaches in more broadly
comparative contexts.
Methods
This paper builds on our previous work and follows much of the same methodological choices
(Soroka, Stecula and Wlezien 2015). Our analyses rely on measures of (a) the economy, (b) the
6
tone of economic news coverage, and (c) public economic sentiment. We use composite
indicators as measures of the economy. The most straightforward measure is the Composite
Leading Indicators (CLI) series from the OECD, available for a wide range of countries.3
Previously we were able to compare models using past, coincident and leading indicators since
we were dealing with the US alone (Soroka, Stecula and Wlezien 2015) . Unfortunately, it is not
possible to do so outside the US – there is no directly comparative lagging or coincident
indicators series available for all three countries. Therefore, the “future” emphasis of media
content cannot be tested across all three countries.4 That said, we can explore the extent to which
media focus on changes rather than levels, and across multiple indicators. In doing so, we can
speak also to the duration of the impact of economy on media tone. Put differently, we can
explore the extent to which media tone shifts, quickly and/or over the longer term, to economic
change. The top panels in Figure 1 show trends for the OECD Composite Leading Indicators
(CLI) series in each country. (Note that some data is missing during 2008 in Canada.) Here we
can see a lot of fluctuation over time, much of which is common across countries; indeed, the
average bivariate correlation between the three series is 0.73.
[Figure 1 about here]
Our measures of media content are based on a database of news coverage over a 30-year period
in each of the US, UK and Canada. Articles are extracted from Lexis-Nexis, using the same
search criteria we employed in previous work (Soroka Stecula and Wlezien 2015).5 Appendix
Table 1 shows a breakdown of stories across each of 6 newspapers, annually. We chose two high-
circulation, and widely considered to be influential, daily newspapers from each country that are
available in the Lexis-Nexis archive: New York Times and Washington Post in the US, The Times
and The Guardian in the UK, as well as The Globe and Mail and Toronto Star in Canada. Our
7
dataset misses newspapers in two years for each of the UK and Canada – a result of the absence
of topic coding in the Lexis-Nexis database. And because the Times and the Toronto Star come
and go in two different years (1996-97, and 1990-91 respectively, we do not estimate models
including article counts as either a dependent or independent variable, just article tone. Note that
missing data is not the only reason for this decision, however: past work indicates a stronger link
between the economy and newspaper sentiment than between the economy and simple article
counts (Soroka Stecula and Wlezien 2015).
The second and third rows in Figure 1 depict our measures of sentiment in news content. The
second row shows the first, “net tone” based on the Lexicoder Sentiment Dictionary (LSD). The
LSD was designed as a general purpose sentiment dictionary, and is described in some detail in
Young and Soroka (2011). It is a relatively large sentiment dictionary, with roughly 3,000
negative and 3,000 positive words. The application of the dictionary here represents what is
typically referred to as a “bag-of-words” approach – we count the number of words in an
established dictionary. Note, however, that the creation and testing of the dictionary incorporates
elements that are sometimes regarded as falling mainly under “supervised learning” approaches.
To be clear, we do not rely here on a measure that is derived only by algorithms applied to the
existing corpus. Rather, we rely on a dictionary that was built from a careful cross-tabulation of
three very large existing dictionaries, in order to both expand coverage and remove potentially
ambiguous words; and tested again both human coders, alongside eight other pre-existing
dictionaries (Young and Soroka 2011). The results here thus depend on a good deal of prior
testing. We rely here on a simple measure of net tone: ((# positive words - # negative words) /
total word count) * 100. The resulting measure captures both the direction and magnitude of
article tone.
8
Although we have confidence in our LSD-based measure, we want to ensure that results are not
dependent on the use of this dictionary. Our interest is motivated in part by recent work
comparing the performance of both dictionary and supervised learning approaches – especially
Barbera et al. (2016), who focus on the coding of sentiment in economic news. Note that the
appendix to Soroka, Stecula and Wlezien (2015) includes tests with an R-word Index, and an
Angst Index. We extend that work here, using results two other dictionaries, each of which was
designed specifically for use with economic news content. The first of these was developed by
De Boef and Kellstedt (2004). This dictionary takes a somewhat different approach than the
more broadly-aimed LSD – it is designed to count co-occurrences of economic keywords, e.g.,
unemployment, inflation, alongside directional and/or valence keywords, e.g., upward,
downward, good, and bad. De Boef and Kellstedt (2004) use co-occurrences within the same
paragraph; we narrow this to within-sentence co-occurrences. Otherwise, we implement the
dictionary in the same way as De Boef and Kellstedt (2004): positive and negative mentions are
counts of a specific, narrowly-focused set of within-sentence co-occurrences; and overall tone is
measured by subtracting the total number of negative mentions from the total number of positive
mentions.
We also produce a measure of sentiment in news coverage using a dictionary built by Loughran
and McDonald (2011) to capture sentiment specifically in financial texts. This dictionary is, like
the LSD, a simple word count; although it estimates negativity only, not positivity. For the sake
of comparison, we also generate a comparable measure from the LSD focused just on negativity.
For both the Loughran-McDonald (LM) and LSD versions of negativity, the measure is: (#
negative words/total word count)*100; that is, the percent of words in an article that are
categorized as negative.
9
[Table 1 about here]
Measures of net tone are shown in the second row of Figure 1, while measures of negativity are
shown in the third row. There clearly are links between the different operationalizations, and
Table 1 shows basic bivariate correlations between all of them. For the US, we include
correlations with the measure of net tone used in in our previous article as well. That measure is
based on the same dictionary as the one used here, but word counts were estimated in an older
version of Lexicoder, the software we use for text analysis.6 These are correlated at 0.96.
Correlations between net tone estimated using the LSD and the De Boef-Kellstedt dictionary are
positive and statistically significant, but relatively low in magnitude – across the three countries,
the mean correlation is 0.48. Correlations between LSD net tone and LSD negativity are of
course relatively high, given that the negativity dictionary is one half of the LSD net tone
measure; the mean across all three countries is -0.71. Correlations between net tone and the LM
negativity measure are slightly lower, on average, -0.55. In sum, the various approaches to
measuring the tone of economic news content produce measures that show both similarity and
difference. Analyses below demonstrate whether and how the differences matter.
The final row of Figure 1 shows our measures of sociotropic economic evaluations. We do not
have measures of retrospective evaluations in all countries, so we focus only on prospective
evaluations here. Note that our prospective measures differ somewhat across countries. In the
US, the measure is from the University of Michigan’s Survey of Consumers, and we focus here
on responses to the question, “And how about a year from now, do you expect that in the country
as a whole, business conditions will be better, or worse than they are at present, or just about the
same?” UK data are from Eurostat’s European Sentiment Indicator, based on the question,
“How do you expect the general economic situation in this country to develop over the next 12
10
months?” Canadian data are from the Conference Board of Canada, based on the question,
“How do you feel the job situation and overall employment will be in this community six months
from now?” The final indicator in each case is the percentage saying “better” minus the
percentage saying “worse.”
There are differences between these questions, to be sure, the most significant being the focus on
jobs, on region rather than country, and on a six-month time horizon in the Canadian case. It thus
is important not to compare levels across countries. Insofar as each series captures an element of
prospective evaluations, however, we expect them to exhibit similar relationships with the
economy and media content over time.
Appendix Table 2 includes unit root tests, specifically, Augmented Dickey-Fuller tests with both
one and three lags, across all series used in the analyses.7 In no case do we fail to reject the null
hypothesis of a unit root, though for leading economic indicators we come close. This simplifies
empirical analysis; since none of the variables are integrated, it means that cointegration is not a
concern, so we can use more standard econometric approaches. This is true when including
leading economic indicators in our analysis, which are long-memoried, what are sometimes
referred to as “near integrated” (DeBoef and Granato 1999). Even for those concerned that those
economic indicators really are integrated, they are not dependent variables in any of our analyses
and we use error correction models (ECM) to analyze media content and economic perceptions.
There thus is little risk of spurious results (Banerjee, et al 1993; Granger, et al 2001).
Results
Table 2 shows the basic ECMs relating different measures of media coverage and leading
economic indicators. The first important finding is that all of the measures of media tone respond
to economic indicators, in both changes and levels, and in expected ways: net tone is positively
11
related to leading indicators, and negativity is appropriately negatively related to leading
indicators. This is true in all three countries. Results also suggest that the LSD net measure
performs better than all other measures, again in all countries. This interpretation is based on the
model R-squareds, which suggest that the LSD produces a measure more in line with economic
indicators than does the DeBoef-Kellstedt dictionary, and also that both net tone measures
outperform the negativity-only measures. We take the improvement in model fit as an indication
that including positive words is important to capturing the nature of economic news coverage.
Even using negativity-only dictionaries, the LSD-based measure better reflects conditions than
that produced using the Loughran-McDonald dictionary.
[Table 2 about here]
These comparisons across measures are of some significance for those interested in the accurate
estimation of sentiment in economic news. One concern about the LSD is that it is a general-
purpose dictionary, intended to apply across a wide range of topics. It thus includes words that
may not be relevant, or may have a different meaning, in an economic context (e.g., “liability”).
This concern – not about the LSD specially, but about general-purpose dictionaries generally –
was part of the motivation behind the Loughran-McDonald dictionary. But the narrow
dictionaries rely on smaller sets of words, and the narrower focus appears to miss relevant words
in economic news coverage. Our supposition is that when journalists use the word “sad” in an
economic news story, it tells us something about the economy, even though the word itself is not
especially economic. As a result, the broader LSD produces a measure that follows the economy
more closely than do the other measures, and likely offers a better indication of the “media
signal” that readers get as well. All subsequent results thus focus just on LSD-estimated net tone
(though note that our substantive findings are no different when using the other measures).
12
[Table 3 about here]
The test of dictionaries here is secondary to our interest in whether prior US results generalize to
the UK and Canada. Table 3 provides that test. It shows the estimated short- and long-term
impacts of the economy on media content, drawn from results in Table 2, comparing across the
three countries. In line with previous work, the short-term impact always outweighs the long-
term impact, and by a lot, in all three countries. In Canada, for instance, a one unit change in the
CLI is associated with a short-term increase in net tone of 0.12 in the current month.8 The long-
term impact, by contrast, is just under 0.05, and the error correction rate (-0.46) suggests that the
remaining disequilibrium is corrected quite quickly, by about one half each month. In short, the
immediate impact of the economy on media tone is substantial and mostly short-lived. Indeed,
the Canadian case is the only one in which the long-run effects make up roughly one-third of the
total impact – in both the US and UK, nearly 90% of the impact of economic conditions is
immediate. And it is important to keep in mind that these multipliers actually overstate the true
long-term effects of economic shocks, which technically are not permanent. Recall that our
analysis of stationarity (see Appendix Table 2) indicates that while shocks diminish very slowly,
they do not last indefinitely. Thus, an economic impulse will tend to decay, and this will have
corresponding (decreasing) effects on media content.
[Table 4 about here]
Findings in Tables 2 and 3 make clear that the media focus on current change in prospective
conditions is not exclusive to the US. But, are there differences in the relationship between
media coverage and public expectations? We want to know whether public evaluations reflect
media coverage and also whether that coverage reflects economic evaluations. To begin with, the
top panel of Table 4 incorporates economic evaluations into the models of media content.
13
Specifically, it includes the current (time t) changes in evaluations as well as their lagged (t-1)
levels, the latter of which are of special importance to us given that they presumably are
exogenous. (While the current changes in evaluations may be endogeneous to current changes in
media coverage, this is not true for lagged levels.) And there is evidence in Table 4 that media
respond in part to public sentiment. Most importantly, the effect of lagged evaluations is positive
and significant in each country, though especially the US and, to a lesser extent, the UK. The
coefficients in the three countries are quite similar – between 0.03 and 0.05 – and this is of real
consequence given that the standard deviation in those evaluations (and media tone) also are
quite similar.9 On average, a one-standard-deviation change in lagged economic prospections
produces a 0.13-standard deviation change in media tone, controlling for both leading economic
indicators and current changes in prospections.
In all three countries, then, results suggest that media content responds to public sentiment above
and beyond the impact of the economy itself. This is an important finding particularly given that
most research assumes that the causality runs in the other direction. (The literature exploring
the uni-directional impact of media coverage on public sentiment is extensive, but consider, e.g.,
De Boef and Kellstedt 2004; Goidel and Langley 1995; Hester and Gibson 2003; Nadeau et al.
1999; Soroka 2006.) Now, let us consider the effect of media coverage on public evaluations.
The bottom panel of Table 4 shows results from estimating an error correction model of
prospective evaluations in the three countries. Here we can see that the public’s economic
expectations do follow leading economic indicators, though most of the effects are short-lived,
particularly in the UK. Results further suggest that evaluations also may respond to changes in
media tone – though the strongest evidence of this is in Canada, the only country for which
lagged levels of media tone are a significant predictor of evaluations. This also is an important
14
finding. While there may be a reciprocal relationship between media content and public
sentiment, the effect appears to run primarily from the latter to the former, especially in the US
and UK.10 This has important implications for the way in which we interpret the substance of
media coverage, as we discuss further, below.
Discussion
This paper offers a first comparative exploration of the relationships between the economy, news
coverage, and public sentiment in three Anglo-American democracies. Results suggest
remarkable similarities across the countries. In each case, media coverage follows economic
conditions; it focuses more on change in the economy than on levels; and the impact of change
appears to be primarily current, that is, the effect of economic change is reflected mainly in
current media tone, and dissipates relatively quickly thereafter.
The reactivity of media coverage to change in economic conditions is illustrated in Figure 2,
which replicates a figure from Soroka, Stecula and Wlezien (2015) across all three countries,
focusing on the period surrounding the Great Recession. The top left panel shows US net tone
and levels of the CLI; net tone is plotted alongside changes in CLI on the right. The first panel
shows what appears to be a relatively weak correspondence between the two series; the second
illustrates a powerful concurrent relationship. Indeed, the correlation between media tone and
levels of the CLI in the US over this period is -0.02 (p=.88), while the correlation between tone
and changes in the CLI is 0.61 (p<.01). This is as we have seen in prior work (albeit with a new
CLI measure here), and highlights the degree to which economic news coverage responds
primarily to change.
[Figure 2 about here]
15
Preceding results suggest a similar dynamic in the UK and Canada, so the second and third rows
of Figure 2 illustrate the same quantities for these other countries. Results are strikingly similar.
Changing from levels to changes in the CLI shifts the correlation with media tone in the UK
from 0.20 (p=.11) to 0.66 (p<.01). In Canada, the change is more muted, from 0.29 (p=.02) to
0.47 (p<.01). The differences in correlations, across both levels and changes, may well tell us
something about differences in economic reporting from one country to the next, though we do
not wish to read too much into these differences without further investigation. For the time
being, we take these as evidence that mass media coverage of the economy in all three countries
focuses primarily on change.
Recall that in each case media coverage also reflects public sentiment itself; as the public
becomes more optimistic or pessimistic about the future, economic news follows. The reverse is
not consistently true, however, since media content reliably influences the public’s economic
expectations only in Canada. To be clear: while a considerable body of work finds evidence of
media effects on public economic sentiment, we find that media coverage is more likely to
reflect the nature of public sentiment than it is to affect it. This finding challenges conventional
characterizations of the media-public relationship, which clearly is not a one-way street. It
highlights potential consequences as well, particularly to the degree perceptions and economic
reality do not match. The nature of media coverage might reflect tendencies in the way in which
publics think about the economy. Most importantly, public perceptions may have a potentially
distorting effect on media content, which has implications for understanding media coverage in
the time of hyper-partisan politics.
We do not, as of yet, have clear expectations for a more broadly comparative study. Our sense
from these analyses is that while some past work points towards cross-national differences in the
16
tone of coverage, or the emphasis on one indicator or another, or the political-ideological bias in
economic news coverage, the emphasis on change in economic conditions may be broadly
generalizable. The priority given to news that is both new, and salient to news consumers’
political and economic decision-making, would seem to make change in prospective conditions
especially relevant across most, if not all, contexts. This should be true regardless of whether
media, or the public, are leading. Of course, there may also be differences in the direction of the
media-public relationship, both across countries and over time. Furthermore, whether these
results generalize to other popular news mediums, such as television, or the Internet, remains to
be seen. For example, social media – including news from traditional producers designed for
distribution through social media like Twitter or Facebook – may behave differently than
newspapers or broadcast news. We cannot explore the many possibilities with the limited dataset
used here. But results above clearly highlight the importance of considering more than a simple
uni-directional link between media coverage and public economic sentiment.
Future work on this subject might also focus on the broader implications of the nature of
economic news. There is already work on some of the political consequences of that coverage,
particularly for government support and electoral outcomes. The relationship between economic
news and other political and policy preferences has been barely addressed, however. So too has
the impact that economic news may have on the economy itself. In short, these potentially
broadly generalizable findings may be of real significance to a host of political and economic
phenomena. Exploring these relationships in a context that looks beyond the United States is
therefore an important goal for future research.
Bibliography
17
Banerjee, Anindya, Juan Dolado, John W. Galbraith and David F. Hendry. 1993. Co-Integration,
Error Correction, and the Econometric Analysis of Non-Stationary Data. Oxford: Oxford
University Press.
Barbera, Pablo, Amber Boydstun, Suzanna Linn, Ryan McMahon and Jonathan Nagler. 2016.
“Methodological Challenges in Estimating Tone: Application to News Coverage of the US
Economy.” Paper presented at the Annual Meeting of the Midwest Political Science
Association, Chicago IL.
Behr, Roy L. and Shanto Iyengar. 1985. “Television News, Real-World Cues, and Changes in the
Public Agenda.” Public Opinion Quarterly 49(1): 38-57.
Clarke, Harold D. and Marianne C. Stewart. 1995. “Economic Evaluations, Prime Ministerial
Approval and Governing Party Support: Rival Models Reconsidered.” British Journal of
Political Science 25(2):145–70.
Croteau, D., & Hoynes, W. 2001. The business of media: Corporate media and the public
interest. Thousand Oaks, CA: Pine Forge Press.
De Vreese, Claes, Jochen Peter, Holli A. Semetko. 2001. “Framing Politics at the Launch of the
Euro: A Cross-National Comparative Study of Frames in the News.” Political Communication
18(1): 107-122.
De Boef, Suzanna and James Granato. 1997. “Near-Integrated Data and the Analysis of Political
Relationships.” American Journal of Political Science 41:619-640.
De Boef, Suzanna, and Paul M. Kellstedt. 2004. “The Political (and Economic) Origins of
Consumer Confidence.” American Journal of Political Science 48(4): 633–649.
18
Deuze, Mark. 2002. “National News Cultures: A Comparison of Dutch, German, British,
Australian and US Journalists.” Journalism and Mass Communication Quarterly 79(1): 134-
149.
Duch, R. M., & Stevenson, R. T. 2008. The economic vote: How political and economic
institutions condition election results. New York, NY: Cambridge University Press.
Durr, Robert H. 1993. “What Moves Policy Sentiment?” American Political Science Review 87
(1):158–70.
Erikson, Robert S. 1989. “Economic Conditions and the Presidential Vote.” American Political
Science Review 83:567-573.
Erikson, Robert S., Michael B. MacKuen, and James A. Stimson. 2002. The Macro Polity.
Cambridge: Cambridge University Press.
Erikson, Robert S. and Christopher Wlezien. 2012. The Timeline of Presidential Elections.
Chicago: University of Chicago Press.
Goidel, Robert K., and Ronald E. Langley. 1995. “Media Coverage of the Economy and
Aggregate Economic Evaluations: Uncovering Evidence of Indirect Media Effects.” Political
Research Quarterly 48(2): 313–328.
Granger, Clive W.J., Namwon Hyung and Yongil Jeon. 2001. “Spurious Regressions with
Stationary Series." Applied Economics 33:899-904.
Hallin, D. C., & Mancini, P. 2004. “Americanization, Globalization, and Secularization:
Understanding the Convergence of Media Systems and Political Communication.” In F. Esser
& B. Pfetsch (ed.), Comparing Political Communication. Theories, Cases, and Challenges.
Cambridge: Cambridge University Press, pp. 25–44.
19
Hanitzsch, Thomas, Folker Hanusch , Claudia Mellado , Maria Anikina , Rosa Berganza , Incilay
Cangoz , Mihai Coman , Basyouni Hamada , María Elena Hernández , Christopher D.
Karadjov, Sonia Virginia Moreira , Peter G. Mwesige , Patrick Lee Plaisance , Zvi Reich , Josef
Seethaler, Elizabeth A. Skewes , Dani Vardiansyah Noor & Edgar Kee Wang Yuen. 2012.
“Mapping Journalism Cultures Across Nations.” Journalism Studies 12(3): 273-293.
Hanitzsch, Thomas and Claudia Mellado. 2011. “What Shapes the News around the World? How
Journalists in Eighteen Countries Perceive Influences on Their Work.” International Journal of
Press and Politics 16(3): 404-426.
Happy, J.R. 1986. “Voter Sensitivity to Economic Conditions: A Canadian-American
Comparison.” Comparative Politics 19 (1): 45–56.
Hester, Joe Bob, and Rhonda Gibson. 2003. “The Economy and Second-Level Agenda Setting: A
Time-Series Analysis of Economic News and Public Opinion about the Economy.” Journalism
& Mass Communication Quarterly 80(1): 73–90.
Hibbs, Douglas A., Jr. 1987. The American Political Economy. Cambridge: Harvard University
Press.
Ju, Youngkee. 2008. “The asymmetry in economic news coverage and its impact on public
perception in South Korea.” International Journal of Public Opinion Research, 20(2): 237-249.
Kalogeropoulos, Antonis, Helle Molgarrd Svensson, Arjen van Dalen, Claes de Vreese and Erik
Albaek. 2015. “Are watchdogs doing their business? Media coverage of economic news.”
Journalism 16(8): 993-1009.
Lewis-Beck, Michael. S. and Mary Stegmaier. 2000. “Economic Determinants of Electoral
Outcomes.” Annual Review of Political Science 3: 183-219.
20
——. 2007. “Economic Models of Voting.” in Russell Dalton and Hans-Dieter Klingemann, eds.,
The Oxford Handbook of Political Behavior, (Oxford: Oxford University Press):518-537
Lischka, Juliane A. 2014. “Different revenue incentives, different content? Comparing economic
news before and during the financial crisis in German public and commercial news outlets over
time.” European Journal of Communication 29(5): 549-566.
Loughran, Tim and Bill McDonald. 2011. “When is a Liability not a Liability? Textual Analysis,
Dictionaries, and 10-Ks.” The Journal of Finance 66(1): 35-65.
Nadeau, Richard, Richard G. Niemi and Timothy Amato. 1996. “Prospective and Comparative or
Retrospective and Individual? Party Leaders and Party Support in Great Britain.” British
Journal of Political Science 26 (2): 245–58.
Nadeau, Richard, Michael Lewis-Beck and Eric Belanger. 2012, “Economics and Elections
Revisited.” Comparative Political Studies 46(5): 551-573.
Nadeau, Richard, Richard G. Niemi, David P. Fan, and Timothy Amato. 1999. “Elite Economic
Forecasts, Economic News, Mass Economic Judgments, and Presidential Approval.” Journal
of Politics 61 (1): 109–35.
Price, Simon and David Sanders. 1993. “Modeling Government Popularity in Postwar Britain: A
Methodological Example.” American Journal of Political Science 37 (1): 317–34.
Sanders, David. 1996. “Economic Performance, Management Competence and the Outcome of
the Next General Election.” Political Studies 44 (2): 203–31.
Sanders, David, David Marsh, and Hugh Ward. 1993. “The Electoral Impact of Press Coverage
of the British Economy, 1979–87.” British Journal of Political Science 23 (2): 175–210.
Soroka, Stuart. 2014. Negativity in Democratic Politics: Causes and Consequences. Cambridge:
Cambridge University Press.
21
———. 2006. “Good News and Bad News: Asymmetric Responses to Economic Information.”
The Journal of Politics 68(2): 372-385.
———. 2002. Agenda-Setting Dynamics in Canada. Vancouver: University of British Columbia
Press.
Soroka, Stuart and Christopher Wlezien. 2010. Degrees of Democracy: Politics, Public Opinion
and Policy. Cambridge: Cambridge University Press.
Soroka, Stuart N., Dominik Stecula and Christopher Wlezien. 2015. “It's (Change in) the
(Future) Economy, Stupid: Economic Indicators, the Media, and Public Opinion.” American
Journal of Political Science 59(2): 457-474.
Soroka, Stuart, Blake Andrew, Toril Aalberg, Shanto Iyengar, James Curran et al. 2013. “Auntie
Knows Best? Public Broadcasters and Current Affairs Knowledge.” British Journal of Political
Science 43(4): 719-739.
Stevenson, Randolph T. 2001. “The Economy and Policy Mood: A Fundamental Dynamic of
Democratic Politics?” American Journal of Political Science 45 (3): 620–33.
Uscinski, Joseph E. 2009. “When Does the Public’s Issue Agenda Affect the Media’s Issue
Agenda (and Vice-Versa)? Developing a Framework for Media-Public Influence*.” Social
Science Quarterly 90(4): 796–815.
van Dalen, Arjen, Claes H. de Vreese and Erik Albaek. 2012. “Different roles, different content?
A four-country comparison of the role conceptions and reporting style of political journalists.”
Journalism 13(7); 903-922.
van Dalen, Arjen, Claes H. de Vreese and Erik Albaek. 2015. “Economic news through the
magnifying glass. How the media cover economic boom and bust.” Journalism Studies DOI:
10.1080/1461670X.2015.1089183.
22
van der Brug, W., van der Eijk, C., & Franklin, M. N. 2007. The economy and the vote:
Economic conditions and elections in fifteen countries. New York, NY: Cambridge University
Press.
Waheed, Moniza, Andreas R.T. Schuck, Peter C. Neijens and Claes H. de Vreese. 3013. “Values
in the News: Comparing the news coverage of political speeches in developed and developing
countries.” Journalism Studies 14(4): 618-634.
Weaver, David H. and Lars Willnat, eds. 2012. The Global Journalist in the 21st Century. New
York: Routledge
Wlezien, Christopher. 1995. “The Public as Thermostat: Dynamics of Preferences for Spending.”
American Journal of Political Science 39:981-1000.
23
Figure 1. Time Series
Figure 2. Leading Indicators and Media Tone, 2007-2011
Table 1. Correlations between Measures of Sentiment
US AJPS Net
Tone LSD Net Tone DK Net
Tone LSD
Negativity
LSD Net Tone 0.96
DK Net Tone 0.56 0.56
LSD Negativity -0.71 -0.71 -0.45
LM Negativity -0.61 -0.59 -0.48 0.78
UK LSD Net Tone DK Net
Tone LSD
Negativity
DK Net Tone 0.40
LSD Negativity -0.75 -0.22
LM Negativity -0.55 -0.07 0.78
CA LSD Net Tone DK Net
Tone LSD
Negativity
DK Net Tone 0.49
LSD Negativity -0.66 -0.43
LM Negativity -0.50 -0.42 0.75
Cells contains Pearson’s correlation coefficients. All correlations are
significant at p < .05.
Table 2. Responsiveness of Media to Leading Indicators
US DV: ∆ in…
LSD Tone DK Tone LM Neg LSD Neg
DV t-1 -0.574*** -0.563*** -0.351*** -0.450***
(0.045) (0.046) (0.039) (0.043)
∆ LEI t 0.194*** 0.535*** -0.077*** -0.105***
(0.031) (0.104) (0.023) (0.028)
LEI t-1 0.014*** 0.046*** -0.014*** -0.014***
(0.005) (0.017) (0.004) (0.005)
Constant -1.198** -4.280** 2.307*** 2.506***
(0.511) (1.725) (0.469) (0.535)
N 383 383 383 383
Rsq 0.311 0.286 0.188 0.237
UK DV: ∆ in…
LSD Tone DK Tone LM Neg LSD Neg
DV t-1 -0.634*** -0.555*** -0.458*** -0.510***
(0.051) (0.050) (0.047) (0.048)
∆ LEI t 0.398*** 0.678*** -0.159*** -0.185***
(0.069) (0.183) (0.048) (0.054)
LEI t-1 0.030*** -0.042 -0.032*** -0.034***
(0.011) (0.029) (0.009) (0.010)
Constant -2.970*** 4.266 4.435*** 4.827***
(1.146) (2.946) (0.959) (1.045)
N 329 329 329 329
Rsq 0.324 0.281 0.236 0.259
CA DV: ∆ in…
LSD Tone DK Tone LM Neg LSD Neg
DV t-1 -0.462*** -0.427*** -0.244*** -0.353***
(0.043) (0.042) (0.034) (0.040)
∆ LEI t 0.116*** 0.337*** -0.044** -0.057***
(0.025) (0.068) (0.020) (0.021)
LEI t-1 0.021*** 0.021* -0.013*** -0.014***
(0.005) (0.012) (0.004) (0.004)
Constant -1.875*** -1.812 1.838*** 2.207***
(0.472) (1.187) (0.418) (0.442)
N 380 380 380 380
Rsq 0.248 0.225 0.137 0.187
* p < .10; ** p < .05; *** p < .01. Cells contain OLS coefficients with
standard errors in parentheses.
Table 3. The Short- and Long-Term Impacts of the Economy on Media Tone
US UK CA
Rate of Error correction ( )Φ-0.574 -0.634 -0.462
Short-term effect (β0) 0.194 0.398 0.116
Long-run multiplier ( )γ0.024 0.047 0.045
Based on models for LSD Tone in Table 4.
Table 4. Responsiveness of Media Tone to Prospective Economic Evaluations (and the
Economy), and Vice Versa
DV: ∆ in LSD Tone
US UK CA
DV t-1 -0.601*** -
0.657*** -0.482***
(0.047) (0.052) (0.045)
∆ LEI t 0.151*** 0.321*** 0.104***
(0.033) (0.076) (0.030)
LEI t-1 0.014*** 0.022* 0.014**
(0.005) (0.013) (0.007)
∆ Pros Evalst0.008*** 0.009** 0.014***
(0.002) (0.004) (0.005)
Pros Evalst-1 0.004*** 0.005** 0.003*
(0.001) (0.003) (0.002)
Constant -1.290** -2.136* -1.193*
(0.508) (1.273) (0.652)
N 380 323 354
Rsq 0.334 0.344 0.268
DV: ∆ Prospective Evaluations
US UK CA
DV t-1 -0.192*** -0.209*** -0.072***
(0.028) (0.033) (0.018)
∆ LEI t 2.995*** 5.406*** 2.632***
(0.730) (1.033) (0.318)
LEI t-1 -0.201* 0.127 0.178**
(0.115) (0.173) (0.073)
∆ LSD Tone t 3.814*** 1.665** 1.734***
(1.137) (0.765) (0.598)
LSD Tone t-1 1.940 1.272 1.950***
(1.247) (0.876) (0.578)
Constant 21.278* -15.199 -19.108***
(11.375) (17.523) (7.327)
N 380 323 354
Rsq 0.174 0.179 0.296
* p < .10; ** p < .05; *** p < .01. Cells contain OLS coefficients with
standard errors in parentheses.
Online Appendix
Appendix Table 1. Dataset by Newspaper, Annually
Year NYT WPost Times Guardian G&M TStar
1980 448 291 0 0 673 0
1981 855 206 0 0 698 0
1982 1228 369 0 0 1338 0
1983 727 307 0 0 904 0
1984 563 230 0 114 818 0
1985 440 236 85 259 496 0
1986 415 225 195 224 316 407
1987 515 286 173 153 372 460
1988 484 241 157 192 506 560
1989 488 210 276 240 623 722
1990 547 347 255 367 865 1375
1991 858 534 425 692 1140 0
1992 826 473 434 655 857 0
1993 560 354 342 488 638 1088
1994 553 317 286 423 589 1393
1995 459 253 235 267 559 865
1996 380 230 0 197 372 401
1997 397 194 0 233 609 725
1998 423 271 94 436 594 778
1999 391 233 165 220 521 260
2000 400 202 153 167 520 284
2001 772 441 291 308 804 464
2002 503 350 280 205 638 414
2003 423 293 287 172 531 154
2004 336 247 150 132 577 136
2005 257 185 205 148 459 235
2006 304 183 249 153 489 245
2007 263 120 314 195 581 310
2008 598 457 1061 604 660 415
2009 788 575 3108 1521 1099 642
2010 526 402 1594 1017 623 393
2011 445 358 637 384 544 255
Appendix Table 2. Tests of Stationarity
Variable Country 1 Lag 3 Lags
ADF Prob B ADF Prob B
CLI US -12.82 0.01 -0.03 -6.66 0.01 -0.01
CLI UK -13.58 0.01 -0.03 -5.26 0.01 -0.01
CLI CA -8.23 0.01 -0.04 -5.9 0.01 -0.03
LSD Net Tone US -8.29 0.01 -0.40 -6.69 0.01 -0.38
LSD Net Tone UK -6.59 0.01 -0.35 -4.24 0.01 -0.25
LSD Net Tone CA -7.1 0.01 -0.31 -5.72 0.01 -0.28
D-K Net Tone US -7.39 0.01 -0.35 -5.71 0.01 -0.30
D-K Net Tone UK -7.5 0.01 -0.41 -4.9 0.01 -0.30
D-K Net Tone CA -5.99 0.01 -0.25 -4.48 0.01 -0.20
LSD Negativity US -6.59 0.01 -0.29 -5.63 0.01 -0.27
LSD Negativity UK -6.17 0.01 -0.30 -3.44 0.01 -0.18
LSD Negativity CA -6.41 0.01 -0.26 -4.86 0.01 -0.22
L-M Negativity US -5.88 0.01 -0.23 -4.33 0.01 -0.18
L-M Negativity UK -5.91 0.01 -0.27 -3.87 0.01 -0.19
L-M Negativity CA -4.8 0.01 -0.15 -3.98 0.01 -0.14
Prospective Evals US -4.94 0.01 -0.14 -5.21 0.01 -0.15
Prospective Evals UK -4.25 0.01 -0.13 -3.44 0.01 -0.11
Prospective Evals CA -4.46 0.01 -0.06 -4.71 0.01 -0.07
1 There are some exceptions in the agenda-setting literature, which has more actively entertained
the possibility of bi-directional causality. See, e.g., Behr and Iyengar 1985; Soroka 2002;
Uscinski 2009. This work is focused on issue salience, however, not on that content or tone of
public sentiment and/or media coverage.
2 It may also be that media competition matters – a commercial media environment in which there
is less competition for consumers may produce different new coverage, and thus different
relationships between economic conditions, media content, and public sentiment. Consider the
following possibility: an emphasis on prospective conditions may be greatest in more competitive
media environments, while less competitive environments may facilitate a combination of
prospective and retrospective reporting. This would be in line with work suggesting that the
quality of journalism, and in particular the depth of reporting, suffers in highly commercialized
environments (e.g., Croteau and Hoynes 2001.)
3 These data and other economic data used here are readily available online through OECD.Stat.
Note that we rely on the “amplitude adjusted” series here, but the other available series produce
comparable results. Note also that OECD CLI series are provided by national statistical agencies,
and thus vary in composition from one country to the next. This makes good sense – the
economies of different countries should be best predicted by somewhat different factors.
4 It is possible to run our analysis with directly comparative measures for the harmonized
unemployment rate (HUR) and inflation rate (CPI), also drawn from the OECD. Results suggest
that media coverage responds little to the CPI, but responds to the HUR in a way that is very
similar to what we see with the CPI below. We see HUR results as useful supporting evidence for
what we show here; results are available upon request.
5 Indeed, the US data used here are identical, with the exception that we do not include 2012 data
since our comparative data are updated only to 2011. The search is based on a set of Lexis-Nexis
subject categories which, based on manual testing, most reliably return results pertaining to the
national economy. (For a comparison of these results with a broader text-based search, see
Barbera et al. 2016.) The final search captured stories for which any of the following terms were
listed as “Relevancy: Major Terms only”— under (a) “Economic Conditions”: Deflation,
Economic Decline, Economic Depression, Economic Growth, and Economic Recovery, Inflation
and Recession; under (b) “Economic Indicators”: Average Earnings, Consumer Credit, Consumer
Prices, Consumer Spending, Employment Rates, Existing Home Sales, Money Supply, New
Home Sales, Productivity, Retail Trade Figures, Unemployment Rates, Wholesale Prices.
6 The newer version of Lexicoder (3.0) is redesigned to reduce processing time for large datasets,
and to facilitate integration in R. Because it deals with word counts and suffixes in a slightly
different way than the older version, we compare the new and old estimates here.
7 This number of lags is based on empirical analysis of statistical significance of the lagged
differences; in most cases, no more than one lag is significant, in most other cases no more than
three lags are so, and only in a handful are additional lags significant, specifically, for leading
economic indicators, and including them makes no difference to the time-serial diagnosis.
8 Given that the standard deviations in CLI and net tone are 3.3 and 0.34 respectively, these
estimates suggest, roughly speaking, that a one-third standard deviation shift in CLI produces a
one-third standard deviation shift in net tone.
9 Specifically, standard deviations in prospective evaluations (media tone) range from 13.35 (0.37)
in Canada to 11.10 (0.47) in the UK and 12.80 (0.36) in the US.
10 Note that this finding is in line with Soroka, Stecula and Wlezien (2015), where Granger
causality tests indicated a stronger impact running from prospective evaluations to media tone
than the reverse. This was not the case for retrospective evaluations, which showed stronger bi-
directional causality; but we cannot test retrospective evaluations cross-nationally here. Note
also that Granger tests using these data produce findings similar to those in previous work, though
across three countries: while public evaluations Granger-cause media tone in all three countries,
the reverse is not true. That said, we do not focus on these Granger results, since they are not ideal
indications of causality – they are simple bivariate tests in what we know is a multivariate
environment.
