Accessible Published by De Gruyter January 31, 2017

Can News Draw Blood? The Impact of Media Coverage on the Number and Severity of Terror Attacks

Klaus B. Beckmann, Ralf Dewenter and Tobias Thomas


Using a new data set that captures the share of reporting on terrorism, we explore the nexus between terrorist attacks and the news. It turns out that terrorism mainly influences news reports through the number of incidents. Regarding the reverse causality, we provide evidence that the share of the news devoted to terrorism Granger-causes further terrorist activities. However, short-run and medium-run effects differ: media coverage on terror only impact in the severity of terror in the short run (up to two months). From the third to the tenth months, it causes an increase in the number as well as in the severity of the attack. These observations are consistent with the idea of competition between terrorist groups.

JEL Classification: D74; B25

1 Introduction

The hypothesis that terrorist attacks lead up to news coverage cannot surprise anyone. On second thought, one might even wonder whether at some point habituation effects kick in, as in a world plagued with terrorism new attacks are hardly news unless they strike close at home and (or) exceed the customary level of violence. The reverse causality, however, appears much more interesting and politically relevant: Is there a feedback mechanism through which news reporting induces additional terrorism?

There exists a small theoretical literature on the latter issue, in particular Frey and Rohner (2007) and Pfeiffer (2012b). This strand of literature typically focusses on the share of news reporting devoted to terrorism. The reason is that the models view optimizing networks and newspapers as being constrained by a given volume of news broadcasts or a maximum number of pages available in a daily publication. We believe this constraint to be a plausible one, which merits further study, in particular in an empirical manner.

Over the past two decades, a rich literature has emerged dealing with both the economic modeling, and the econometric estimation, of terrorism. [1] The pertinent empirical literature using newsfeeds, [2] however, typically relies on the number of Google searches (Pfeiffer, 2012a), the number of articles in a newspaper (Jetter, 2014) and other absolute measures of news coverage. The present paper differs from the existing literature in that it exploits a new dataset allowing for a reliable measure of relative coverage. We also add to the literature in that we focus on television reports rather than the Internet or newspapers, using a selection of media from three Western countries. Finally, by aggregating time series at a weekly and monthly level, we use a finer granularity than Pfeiffer (2012a) while still dealing primarily with the overall media attention to terrorism instead of the effects of individual news items.

We begin with a short dynamic model that introduces a feedback loop from news reports to terrorist activity, broadly construed (Section 2). Section 3 describes our new data set, and Section 4 presents our estimation strategy and results. Section 5 concludes.

2 A simple model

Our model of the interaction between the media and terrorist groups is similar in spirit to Frey and Rohner (2007) and Pfeiffer (2012b). However, their models are static – although Frey and Rohner (2007) consider a dynamic extension in the appendix – and formulated in terms of controls rather than states. We also explicitly introduce a third sector, namely sponsors of terrorism, to motivate our hypothesis that there exists a feedback loop between the share of media capacity devoted to terrorism and terrorist activity. On the other hand, the model remains ad hoc as we specify the behavior of agents in our model parametrically rather than derive it from optimization. To do the latter, one would need to solve a differential game between terrorists (and the media sector) that contains non-linearities. Beckmann and Reimer (2014) review the theoretical problems this entails and the consequent limitations for formal conflict economics.

The model contains three state variables, viz. the resources rt available to the terrorist group, the public attention at for terrorism and for other news items (bt ). Suppressing time indices, the latter two are supposed to change over time according to the following linear differential equations:

(1) a ˙ = ϵ s r p δ 1 a

(2) b ˙ = n ˜ δ 2 b

In equation (1), s represents the share of available resources that terrorist groups devote to carrying out attacks, p are the costs of a single attack, and ϵ measures the effectiveness of terrorist attacks, e.g. the average number of casualties per attack.

Public interest in a news item wanes as time goes by. We capture this by including depreciation of a and b at the rates δ1 and δ2, respectively. “Other” news is assumed to arrive at a random rate n ˜ ; in numerical simulation, we typically assume this to be uniformly distributed over some range [0, nmax].

(3) r ˙ = a w a + b δ 3 r s r

The terrorist sector receives funding from sponsoring nations, groups and individuals, whose willingness to pay depends on the parameter w and the share of reporting on terrorist events in the media (Pfeiffer, 2012b). The fraction a a + b represents the familiar ratio conflict success function (Hirshleifer, 2001), where the scale parameter has been set to unity. Terrorist resources can be spent on attacks (sr) or saved, in which case they depreciate at a rate δ3. This assumption reflects the hypothesis that terrorist groups find it difficult to put their capital to productive use in the official economy, and that some resources will be destroyed by anti-terror efforts.

There is no stationary state proper in this model due to the random stream of general news n ˜ . However, if n were a constant and equal to its expectation, a stationary state could be computed. Letting a ˙ = b ˙ = r ˙ = 0 , two solutions to our system of equations can be obtained. The first one is a corner solution where both the number of attacks and the terrorist resources are zero, and the second one is

(4) a = ϵ s w δ 1 p ( δ 3 + s ) E n ˜ δ 2

(5) b = E n ˜ δ 2

(6) r = w s + δ 3 δ 1 p E n ˜ δ 2 ϵ s

Besides imposing obvious restrictions on the values of our parameters required to ensure that the stationary state lies in the positive orthant, these results do not offer any surprises. In order to determine the stability properties of the system, we compute its Jacobian

(7) J = ( δ 1 0 ϵ s p 0 δ 2 0 b w ( a + b ) 2 a w ( a + b ) 2 δ 3 s )

the eigenvalues of which are λ1=–δ2, λ 2 = A 2 + 4 b ϵ s w ( a + b ) p ( δ 3 + s + δ 1 ) 2 ( a + b ) p and λ 3 = A 2 + 4 b ϵ s w ( a + b ) p ( δ 3 + s + δ 1 ) 2 ( a + b ) p , where A represents the expression A = ( a + b ) p ( δ 3 + s δ 1 ) . As A2+4bϵw>0, all three eigenvalues will be elements of the real line. Furthermore, it is obvious that λ1<0 and λ3<0. The sign of λ2 is indeterminate, although numerical experimentation shows that it is positive only for small values of a and b. Consequently, the stationary point – if it exists at all in the positive orthant –, is either asymptotically stable or saddle point stable, with the former case being more likely. Figure 1 presents a representative simulation run of the model. [3]

Figure 1: An example simulation of the theoretical model. (A) Vector field, (B) time path of media attention and the number of incidents.

Figure 1:

An example simulation of the theoretical model. (A) Vector field, (B) time path of media attention and the number of incidents.

As the simulated time series in Figure 1 illustrates, the model predicts terrorist acts and their share of reporting to move together. All three stocks are endogenous, in particular, there is a causal link from terrorism to media reports as well as the reverse causality. We therefore state the following two hypotheses:

Hypothesis 1.Terrorist activity Granger-causes the proportion of media reports devoted to terrorism.

Hypothesis 2.The intensity of reporting in the media Granger-causes the volume of terrorist activity.

3 Data

The present study draws on two different data sources: Data on terrorist incidents are taken from the Global Terrorism Database (GTD) (START, n.d), which has become a staple of the empirical literature on the economics of terrorism (see, e.g. Kis-Katos, Liebert, & Schulze, 2011). GTD captures a variety of information for incidents (if available), including – but not limited to – date, geographical location, perpetrators, targets, weapons and methods used, number of victims killed and wounded, and an estimate of material damage (for more on GTD, see LaFree, Dugan, & Miller, 2015). From this data, we construct separate time series containing the total number of incidents per day and the total number of casualties and fatalities for international, domestic and all terrorism, respectively. [4] The models presented in the results section (4) use the number of fatalities to measure the severity of attacks. We ran the main models using all casualties as an alternative measure of severity, but found no qualitative difference. We therefore omit the number of wounded victims from our estimations.

Existing contributions on the impact of media coverage on perceptions and behavior in a political and economic context often use data from various media outlets in a certain country. [5] When it comes to the global impact of the media on perception and behavior, however, the international reach and accessibility of the media play an important role. Because TV newshows have the highest reach inside a country and are broadcast internationally by satellite and the internet, Helmenstein, Krabb, and Thomas (2016) use media coverage on Austria in 21 international TV newsshows as a proxy for the international perception of Austria as a business location to analyse investment activities. TV data also allows to look at the domestic perception of international events, especially when pictures play an important role. In this vein, Eisensee and Strömberg (2007) use data of four major US TV networks to estimate the impact of media coverage on natural catastrophes on US disaster relief.

For the purposes of the present paper, it therefore seems reasonable to use media which have internationally and domestically the best accessibility and the highest reach. Additionally, pictures play an important role when it comes to the news coverage of terrorism (in much the same way that they do for natural catastrophes). And, finally, the existing literature on the media impact of terrorism underutilizes TV data.

The media side of our data set is based on media content analysis by Media Tenor International. [6]All news items broadcast by selected television programs in three countries [7] were captured and analysed by human experts. As computer linguistic approaches are still unable to produce reliable results when it comes to content and sentiment analysis in the political context even in the analysis of printed texts, an automated analysis of TV newsshows is simply not feasible from a technical perspective so far.

In total 1,797,181 news items from nine TV news broadcasts were processed to generate our data set. Available variables include date, source country, name of medium, style of the item, category of main topic, main topic, main actor referenced, region or country referenced, and an evaluation. News items are standardized pieces of information provided by a media report. For example, if one terrorist incident is covered, each time the protagonist, such as the terrorist or the victims, or the source, such as the anchorperson or terror expert or a politician, changes, this will be added as a separate item on the same incident. As we know the number and type of all news items broadcast in every single TV broadcast, we are in a position to construct a fair measure of the share of reporting an incident receives.

Note that this measure is content-based rather than time-based: it does not reflect the share of broadcasting time devoted to terrorism, but rather the share of different perspectives and content as evaluated by a human researcher.

The four panels in Figure 2 trace key variables in our data set over time. The top left panel (2a) shows the share of news items, newsshare, on terrorism. We immediately see the outlier that was 9/11 and to a lesser degree the attack on the Boston Marathon in 2013, but we also recognize that the portion of news broadcasts devoted to terrorism is fairly low (below 1%) on most days. At the top right, we show (panel 2b) the time series of incidents and victims killed (kills) over time. Again, 9/11 stands out, but the second outlier in the number of fatalities time series reflects ISIS activities in Tikrit, Badush and Sinjar. We also find some indication that terrorist incidents have grown deadlier over time.

Figure 2: Plots of selected time series from our data set. (A) The share of news devoted to terrorism, (B) terrorist events and victims killed, (C) number of items, (D) international terrorist events.

Figure 2:

Plots of selected time series from our data set. (A) The share of news devoted to terrorism, (B) terrorist events and victims killed, (C) number of items, (D) international terrorist events.

On the bottom row of Figure 2, the left-hand panel shows the time series of the number of items reported on all our media channels, while the right panel (2d) plots the number of terrorist acts classified as international by the GTD over time. It is apparent that international terrorism is on the rise in absolute terms, and a visual comparison to panel (2b) suggests that this form of terrorism has also become more prevalent in relative terms. But note the scale on the y axis: Regarding the number of incidents, localized terrorism still dwarves the international variant.

4 Results

To analyse the implications of the theoretical model, we aggregate our time series on a weekly basis, pooling international and regional terrorism. At a first step, we run Phillips-Perron unit root tests on incidents, kills and newsshare in order to determine possible nonstationarities (see Table 1 below). [8] Overall, all variables are found to be I(0). The same finding obtains if we shorten our time series by dropping the years 2001 and 2013 in order to eliminate the two major outliers identified in Section 3.

Table 1:

Philips-Perron unit root tests (H0: no stationarity).

Variable Model # lags Z(t) p-Value
incidents constant 6 –2.503 0.11
incidents constant and trend 6 –6.577 0.00
kills constant 6 –4.575 0.00
kills constant and trend 6 –13.056 0.00
newsshare constant 6 –22.558 0.00
newsshare constant and trend 6 –22.637 0.00

Next, we tackle hypothesis 1, running Granger causality test using incidents and fatalities on the one and newsshare on the other hand. Although Bayesian as well as Akaike information criteria suggest the use of six lags, we include up to a dozen lags in our regressions. By doing so, we are able to cover as much as a quarter. As can be seen from Table 2, the number of incidents Granger cause the newsshare using up to 12 lags (i.e. weeks). The expected short-run effect through the frequency of attacks channel can therefore be regarded as established. [9] However, using the number of kills instead of the number of incidents apparently leads to a quite different result: the null that the number of kills do not Granger cause coverage on terrorism can not be rejected in any case. Taking logs of the number of incidents and fatalities to reflect that our left-hand variable is a proportion leads to similar results.

Table 2:

Granger causality tests of terrorism causing media reporting (weekly data, 2001–2013).

H0 # of lags χ2-statistics Prob>χ2
incidents↛newsshare 1 0.63 0.42
incidents↛newsshare 2 6.45 0.04
incidents↛newsshare 3 10.12 0.02
incidents↛newsshare 4 10.53 0.03
incidents↛newsshare 5 10.87 0.05
incidents↛newsshare 6 11.84 0.06
incidents↛newsshare 7 17.82 0.01
incidents↛newsshare 8 18.22 0.02
incidents↛newsshare 9 18.70 0.03
incidents↛newsshare 10 19.05 0.04
incidents↛newsshare 11 19.73 0.05
incidents↛newsshare 12 19.96 0.07
kills↛newsshare 1 0.97 0.32
kills↛newsshare 2 0.96 0.62
kills↛newsshare 3 1.90 0.59
kills↛newsshare 4 1.91 0.75
kills↛newsshare 5 6.41 0.27
kills↛newsshare 6 8.02 0.23
kills↛newsshare 7 8.45 0.29
kills↛newsshare 8 8.33 0.40
kills↛newsshare 9 8.36 0.50
kills↛newsshare 10 8.54 0.57
kills↛newsshare 11 8.80 0.64
kills↛newsshare 12 8.95 0.70

However, the second part of this result hinges on the presence of the outliers. Restricting our sample to the period of 2002:1–2012 :52, and therefore excluding terrorist attacks on 9/11 as well as Boston Marathon bombing, we are able to derive different results. Now, both the number of incidents and the number of people killed are found to be Granger-causal to media coverage. This reveals that the extraordinary events led to biased estimates. (See Table 3 below. We suppress the unchanged results.) [10]

Table 3:

Granger causality tests of terrorism causing media reporting (weekly data, 2002–2012).

H0 # of lags χ2-statistics Prob>χ2
kills↛newsshare 1 5.70 0.02
kills↛newsshare 2 5.75 0.06
kills↛newsshare 3 8.64 0.03
kills↛newsshare 4 8.58 0.07
kills↛newsshare 5 8.52 0.12
kills↛newsshare 6 8.88 0.18
kills↛newsshare 7 14.61 0.04
kills↛newsshare 8 15.00 0.06
kills↛newsshare 9 16.56 0.06
kills↛newsshare 10 15.62 0.11
kills↛newsshare 11 17.38 0.10
kills↛newsshare 12 19.27 0.08

Let us now turn to the second hypothesis. To test this, we run Granger causality test under the null that newsshare does not Granger cause the number of incidents or number of fatalities, respectively. Table 4 summarizes the results from both tests regarding a rather short-run perspective. Again, even though information criteria suggest six lags (weeks), we admit up to a dozen lags to account for at least a 3 months period. Interestingly, while the share of reports on terrorist attacks fail as a predictor of the number of incidents, the null that newsshare does not Granger cause the number of kills can be rejected. This finding is consistent with the type of model discussed in Section 2, although terrorists appear to compete through the intensity of effort rather than through its quantity. This time, restricting the sample in order to exclude the outliers fails to make a difference, and we do not show the respective table.

Table 4:

Granger causality tests of media reporting causing terrorism (weekly data, 2001–2013).

H0 # of lags χ2-statistics Prob>χ2
newsshare↛incidents 1 0.02 0.87
newsshare↛incidents 2 0.27 0.87
newsshare↛incidents 3 0.96 0.81
newsshare↛incidents 4 2.22 0.69
newsshare↛incidents 5 2.78 0.73
newsshare↛incidents 6 3.75 0.71
newsshare↛incidents 7 4.56 0.71
newsshare↛incidents 8 4.77 0.78
newsshare↛incidents 9 6.15 0.72
newsshare↛incidents 10 6.46 0.77
newsshare↛incidents 11 7.10 0.79
newsshare↛incidents 12 7.17 0.84
newsshare↛kills 1 105.00 0.00
newsshare↛kills 2 82.19 0.00
newsshare↛kills 3 69.93 0.00
Newsshare↛kills 4 59.72 0.00
newsshare↛kills 5 56.10 0.00
newsshare↛kills 6 53.69 0.00
newsshare↛kills 7 51.14 0.00
newsshare↛kills 8 49.83 0.00
newsshare↛kills 9 49.02 0.00
newsshare↛kills 10 50.90 0.00
newsshare↛kills 11 51.33 0.00
newsshare↛kills 12 51.01 0.00

As the planning of terrorist acts – depending on the type of attack – may require a considerable amount of time, we now aggregate the data on a monthly basis and rerun our Granger causality tests. Again, the outliers make a difference. If we use the full sample, we fail to find long-term effects of terrorism on news reporting. Neither the number of kills nor the number of incidents does Granger-cause media coverage on a monthly basis. This changes if we consider a subsample of 2002–2012, neglecting 9/11 and the Boston Marathon Bombing (see Table 5).

Table 5:

Granger causality tests of terrorism causing media reporting (monthly data, 2002–2012).

H0 # of lags χ2-statistics Prob>χ2
incidents↛newsshare 1 11.65 0.00
incidents↛newsshare 2 12.60 0.01
incidents↛newsshare 3 11.94 0.01
incidents↛newsshare 4 15.77 0.00
incidents↛newsshare 5 17.97 0.00
incidents↛newsshare 6 20.03 0.00
incidents↛newsshare 7 21.02 0.00
incidents↛newsshare 8 24.66 0.00
incidents↛newsshare 9 28.09 0.006
incidents↛newsshare 10 24.08 0.01
incidents↛newsshare 11 25.53 0.01
incidents↛newsshare 12 33.63 0.01
kills↛newsshare 1 6.02 0.01
kills↛newsshare 2 6.83 0.03
kills↛newsshare 3 7.90 0.04
kills↛newsshare 4 8.14 0.09
kills↛newsshare 5 8.77 0.12
kills↛newsshare 6 9.57 0.14
kills↛newsshare 7 11.67 0.11
kills↛newsshare 8 19.99 0.01
kills↛newsshare 9 21.54 0.01
kills↛newsshare 10 18.30 0.05
kills↛newsshare 11 17.09 0.11
kills↛newsshare 12 17.88 0.12

Table 6 shows our results for the reverse causation. Again, we find that using monthly rather than weekly data does not make much of a difference. While newsshare is found to Granger cause kills up to a lag length of 10 months, the number of incidents shows different results. Relative media coverage is not causal for the number of incidents during the first 2 months. However, the null can be rejected for lags more than two and less than 10.

Table 6:

Granger causality tests of media reporting causing terrorism (monthly data, 2001–2013).

H0 # of lags χ2-statistics Prob>χ2
newsshare↛incidents 1 0.29 0.58
newsshare↛incidents 2 2.01 0.36
newsshare↛incidents 3 7.66 0.05
newsshare↛incidents 4 8.33 0.08
newsshare↛incidents 5 9.79 0.08
newsshare↛incidents 6 11.30 0.08
newsshare↛incidents 7 11.81 0.10
newsshare↛incidents 8 13.52 0.09
newsshare↛incidents 9 15.10 0.09
newsshare↛incidents 10 13.74 0.18
newsshare↛incidents 11 12.97 0.29
newsshare↛incidents 12 15.11 0.23
newsshare↛kills 1 31.07 0.00
newsshare↛kills 2 26.51 0.00
newsshare↛kills 3 25.33 0.00
newsshare↛kills 4 25.68 0.00
newsshare↛kills 5 27.51 0.00
newsshare↛kills 6 27.16 0.00
newsshare↛kills 7 27.94 0.00
newsshare↛kills 8 31.01 0.00
newsshare↛kills 9 32.90 0.00
newsshare↛kills 10 16.30 0.09
newsshare↛kills 11 17.05 0.11
newsshare↛kills 12 14.63 0.26

We interpret these results as evidence for an impact of media coverage on terrorist activities. We consistently find a “forward” causation from attacks on news reporting using both weekly and monthly aggregation levels. The reverse causality also exists, although media coverage seems to have a delayed effect on terrorist acts. This is plausible as attacks usually require planning and preparation. We also posit that the number of kills is a measure of quality rather than quantity. Therefore mimicking terror acts may result in a higher “quality” of attacks or, put differently, in more violent acts. However, media coverage also seems to have a longer-term “reverse” causal effect. TV reporting Granger causes the number of incidents when the time considered is measured in months rather than weeks. Our evidence thus supports both hypothesis 1 and hypothesis 2, although there are interesting nuances and qualifications.

As a final robustness check, we also consider regional effects of terrorism. The previous analysis bunched all forms of terrorism – both regional and international, irrespective of target and origin – together and also used media data from all three countries. Clearly, it is likely that a country’s media (and the public) react differently to terrorism affecting the home country (or citizens and national interest abroad). One might be particularly wary of our result that the number of fatalities does not cause the share of news reporting.

Unfortunately, GTD contains no information on the nationality of victims with the exception of US citizens. We therefore replicate our first model, using only the number of incidents affecting the USA (either taking place on its territory or being directed at a US target) and the number of US victims. Media data is restricted to the three US channels in our data set. Table 7 below displays the results, confirming that the same pattern holds if we look through a “national lens”.

Table 7:

Granger causality tests of the number of US victims causing reporting in the US media (weekly data, 2001–2013).

H0 # of lags χ2-statistics Prob>χ2
kills↮newsshare 1 6.30 0.01 0.91 0.34
kills↮newsshare 2 5.66 0.06 1.27 0.53
kills↮newsshare 3 6.84 0.07 1.63 0.65
kills↮newsshare 4 8.76 0.07 1.95 0.75
kills↮newsshare 5 9.99 0.08 2.01 0.84
kills↮newsshare 6 36.40 0.00 2.31 0.88
kills↮newsshare 7 47.22 0.00 2.50 0.92
kills↮newsshare 8 43.77 0.00 2.59 0.95
kills↮newsshare 9 43.42 0.00 2.81 0.97
kills↮newsshare 10 44.60 0.00 3.34 0.97
kills↮newsshare 11 44.92 0.00 3.41 0.98
kills↮newsshare 12 45.90 0.00 7.65 0.81

5 Conclusion

This paper uses a new dataset uniquely containing a measure of the share of television news devoted to terrorism in order to investigate two hypotheses derived from a stylised three-sector dynamic model of terrorism, its financial sponsors, and the media: (a) terror acts cause media coverage and (b) media coverage causes further terrorist acts. It turns out that the empirical evidence confirms both the main theoretical hypotheses, sort of. News can indeed draw blood, but there are qualifications.

We do find the expected short-run effect of terrorism on media coverage (hypothesis 1). This is straightforward as terror attacks are of certain news value for the media. The reverse causality implied by the theoretical model seems to be present as well (hypothesis 2). However, we also present evidence that short-run and medium-run mechanisms differ.

In the very short run (2 months), media coverage on terror has an impact on the severity, but not on the quantity, of terror attacks. This is plausible because terror attacks need some time to plan and organize, and so terror sponsors can in the short run just “invest” in already planned attacks. On the other hand, the medium run adds another margin of choice in that the number of attacks can also be varied to attract sponsorship. As a result the number of terror incidents increases as well. These effects last up to ten months.

These observations are consistent with an extension of our model that emphasizes competition between terrorist groups (instead of the monolithic terrorist sector we posited earlier).


We are indebted to two anonymous referees as well as to the editor of the Journal for their most helpful comments and feedback on the draft version of this paper. The usual caveat applies.


Baron, D. P. (2005). Persistent media bias. Journal of Public Economics, 90, 1–36. Search in Google Scholar

Beckmann, K. B., & Reimer, L. (2014). Dynamics of military conflict: An economics perspective. Review of Economics, 65, 265–285. Search in Google Scholar

Dewenter, R., Heimeshoff, U., & Thomas, T. (2016). Media coverage and car manufacturers’ sales. Economics Bulletin, 36, 976–982. Search in Google Scholar

Eisensee, T., & Strömberg, D. (2007). News droughts, news floods, and U.S. disaster relief. The Quarterly Journal of Economics, 122(2), 693–728. Search in Google Scholar

Frey, B. S., & Rohner, D. (2007). Blood and Ink! The common-interest-game between terrorists and the media. Public Choice, 133, 129–145. Search in Google Scholar

Garfinkel, M. R., & Skaperdas, G. (2012). The Oxford handbook of the economics of peace and conflict. Oxford: Oxford UP. Search in Google Scholar

Garz, M. (2012). Job insecurity perceptions and media coverage of labor market policy. Journal of Labor Research, 33, 528–544. Search in Google Scholar

Garz, M. (2013). Unemployment expectations, excessive pessimism, and news coverage. Journal of Economic Psychology, 34, 156–168. Search in Google Scholar

Garz, M. (2014). Good news and bad news: Evidence of media bias in unemployment reports. Public Choice, 161(3), 499–515. Search in Google Scholar

Gentzkow, M. A., Shapiro, J. M., & Sinkinson, M. (2011). The effect of newspaper entry and exit on electoral politics. American Economic Review, 101, 2980–3018. Search in Google Scholar

Haucap, J., Thomas, T., & Wagner, G.G. (2015). What influence do researchers have in the media and in economic policy? Wirtschaftsdienst,95, 68–75. Search in Google Scholar

Helmenstein, C., Krabb, P., & Thomas, T. (2016). Location-related sentiment as determinant of investment activity – The explanatory power of international media coverage and national perception for locational choice. Wirtschaftspolitische Blätter, 63, 171–187. Search in Google Scholar

Hirshleifer, J. (2001). The dark side of the force. Cambridge: Cambridge UP. Search in Google Scholar

Jetter, M. (2014). Terrorism and the media. IZA DP 8497. Search in Google Scholar

Kholodilin, K., Thomas, T., & Ulbricht, D. (2016). Do media data help to predict German industrial production? Journal of Forecasting, published online first: 3 November 2016. DOI: 10.1002/for.2449. Search in Google Scholar

Kis-Katos, K., Liebert, H., & Schulze, G. G. (2011). On the origin of domestic and international terrorism. European Journal of Political Economy, 27(1), S17–S36. Special Issue: Terrorism. Search in Google Scholar

LaFree, G., Dugan, L., & Miller, E. (2015). Putting terrorism in context: Lessons from the global terrorism database. New York, NY: Routledge. Search in Google Scholar

Lamla, M., & Dräger, L. (2013). Updating inflation expectations: Evidence from micro data. Economics Letters, 117, 807–810. Search in Google Scholar

Pfeiffer, C. P. (2012a). Causalities and casualties: Media attention and terrorism, 1970–2010. Diskussionspapier, Helmut-Schmidt-Universität, Fächergruppe Volkswirtschaftslehre 127. Search in Google Scholar

Pfeiffer, C. P. (2012b). Terrorism and its oxygen: A game-theoretic perspective on terrorism and the media. Behavioral Sciences of Terrorism and Political Aggression, 4(3), 212–228. Search in Google Scholar

START. National consortium for the study of terrorism responses to terrorism: Global terrorism database [data file]. Available at: Search in Google Scholar

Published Online: 2017-1-31
Published in Print: 2017-1-1

©2017 Walter de Gruyter GmbH, Berlin/Boston