Media slant against foreign owners: Downsizing

Journal of Public Economics 120 (2014) 97–106

Contents lists available at ScienceDirect

Journal of Public Economics journal homepage: www.elsevier.com/locate/jpube

Media slant against foreign owners: Downsizing Guido Friebel a,b,c,⁎, Matthias Heinz d a

Goethe University Frankfurt, Grueneburgplatz 1, 60323 Frankfurt, Germany CEPR, United Kingdom c IZA, Germany d University of Cologne, Germany b

a r t i c l e

i n f o

Article history: Received 29 January 2013 Received in revised form 21 August 2014 Accepted 3 September 2014 Available online 28 September 2014 JEL classification: L82 L33 L10

a b s t r a c t Using a unique data set from nationally distributed quality newspapers in Germany, we find evidence for both quantitative and qualitative media slant against foreign firms. A downsizing foreign firm receives almost twice as much attention as a domestic firm, and the tone of media reports is more negative. Media slant is a measure for economic xenophobia directed against foreign owners, which constitutes an obstacle to foreign direct investment. © 2014 Elsevier B.V. All rights reserved.

Keywords: Media economics Globalization Economic xenophobia Multi-national enterprises Foreign direct investment

Using a unique data set of nationally distributed quality newspapers in Germany, we establish the existence of a strong media slant against foreign owners. On average, firms that are controlled by foreign blockholders attract almost twice the media attention for each job shed compared to domestic firms. More articles are written about downsizing when the owners are foreigners, and, in an average article, more of the words written concern downsizing rather than other topics, such as firm performance, and products. Quantitative slant is accompanied by qualitative slant; that is, newspapers report more negatively about downsizing in foreign-owned firms. Our estimates are likely to be a conservative measure of media slant against foreign owners, because our sources are quality newspapers. Moreover, Germany is a leading export country, and Germans tend to be very positive about globalization.1 Nonetheless, we find evidence in line with Scheve and Slaughter's (2006) observation that “people

perceive an asymmetric distribution of the benefits of globalization: more for consumers and corporations, but less for workers”. Media slant may affect foreign direct investment (FDI), because bad press creates a risk of being penalized by consumers. In Section 2, we provide illustrative examples of consumer reactions to downsizing news that were harmful for foreign firms. Foreign investors will price in this risk, leading to FDI discounts, and lower investment in consumption good industries to avoid visibility to consumers. Foreign investors may also refrain from buying firms that are candidates for downsizing. Media slant may hence constitute a behavioral obstacle to FDI, which according to some observers may potentially be equally important as the formal obstacles dictated by governments (OECD, 2003). We use three data sets to establish and investigate the slant against foreign owners. Our first data set is based on 5394 articles from Germany's newspaper Die Welt on a total of 651 downsizing events in Germany between December 2000 and September 2008.2 Following Gentzkow and Shapiro's (2006) definition of slant, we focus on the

⁎ Corresponding author at: Goethe University Frankfurt, Grueneburgplatz 1, 60323 Frankfurt, Germany. E-mail addresses: [email protected] (G. Friebel), [email protected] (M. Heinz). 1 On a level above that of the U.S., cf. Mayda and Rodrik (2005).

2 We end with September 2008 because the collapse of Lehman Brothers triggered a wave of downsizing and state interventions, and we are interested in normal, rather than special, times.

1. Introduction

http://dx.doi.org/10.1016/j.jpubeco.2014.09.001 0047-2727/© 2014 Elsevier B.V. All rights reserved.

98

G. Friebel, M. Heinz / Journal of Public Economics 120 (2014) 97–106

varying intensity of reporting a fact, measured by the number of words written per job shed in instances of downsizing by foreign-owned employers relative to domestic ones. A second data set from the media consultancy Media Tenor measures the different qualitative evaluations involved. We also collected a third data set that includes all articles on downsizing during five randomly chosen months from all nationally distributed quality newspapers in Germany. To build the first data set, we developed an algorithm aiming to reduce the probability of making two types of errors: including “false” downsizing cases, and failing to include “real” ones (see Section 3). The regressions (Section 4) show a strong media slant against foreignowned firms controlling for the magnitude of the downsizing event, the industry, and the size of the firm measured in employment. The results are unchanged when more of the downsizing event's specificities are taken into account, like the downsizing magnitude relative to firm size, establishment closures, regional unemployment, and the speed of downsizing. Propensity score methods deal with foreign firms' potential self-selection into specific regions and industries. Investigating potential spurious correlations, our regressions show that the results are robust to the inclusion of controls for the reason for downsizing and the type of ownership (see Section 5). We also find no evidence that specific groups of countries3 drive the slant: rather, the magnitude of the slant is quite similar across country groups. There is no tendency to report more about foreign firms in general, or their job creation.4 Other omitted variables can be controlled for to some extent. For a subset of large firms, we are able to match our data set with financial and accounting information from the Amadeus data set (Bureau van Dijk, 2011). Despite a smaller sample, the slant is statistically significant, while covariates such as assets, employment costs per capita or return on equity before downsizing are not. We also carried out a number of additional analyses to investigate what hypotheses could explain the slant, some of which are briefly presented in Section 6. On the supply side, we find no evidence that the bias5 of certain journalists can explain newspapers' reports,6 or that ownership changes affect the slant. Looking at other stakeholders, we do not find significant advertiser influence. Political economy explanations like Besley and Prat's (2006) theory do not apply either, because in our period of observation, Germany was run by the globalizationfriendly Schröder and Merkel governments who had little to gain by influencing the media against foreign owners. Inspired by Puglisi's (2011) analysis of changes in the agenda-setting behavior of the New York Times during presidential campaigns, we also compared the coverage of downsizing events before German elections with no-campaign periods and find no systematic differences (as shown in Table E in Web Appendix II).7 Our analysis is mainly based on a data set of reports from Die Welt, raising the question of whether these results are specific to one 3 Distinctive features of such country groups could be cultural or linguistic distance to Germany, or rather general perceptions of “free” market-based economic systems like those of the U.K. or the U.S. compared to the continental European model of a “social” market economy. 4 The slant is also unrelated to the impact of newspaper technology on the timing of the news, as in Soltes (2009), and to the size of standard articles in different newspaper sections, as shown in Web Appendix I. 5 We have so far used the term “slant” rather than “bias”, as the literature tends to associate bias with behavior that is not in line with profit maximization (Gentzkow and Shapiro, 2010). Here, we prefer using the word “bias”, as it describes individual behavior that is not necessarily in line with profit maximization. 6 Baron (2006) argues that a newspaper may allow journalists to report in a biased way because this makes it possible to retain the journalist and/or to cut wage costs. While we cannot exclude that all or most journalists are biased against foreign owners, we have investigated whether the slant may be driven by those journalists who are most actively writing on downsizing and have not found any supporting evidence (see Tables C and D in Web Appendix II). 7 There is a noteworthy exception. The leader of the SPD, the German social democrats, conducted a short campaign against what he called the “locusts”, foreign private equity investors. We do find a slight increase in the slant during that period, but the slant is present across the entire period.

newspaper. Although Die Welt is owned by the Axel Springer Publishing House8 (a close ally of the business-friendly CDU party in government since 2005), the readership of Die Welt is conservative, and conservatism may trigger some bias against foreign owners. We therefore set up a third data set including five randomly drawn months of reports from the six other leading national newspapers. We find that all newspapers slant their reports in the same direction, with some indications of an increasing slant when moving from right to left along the political spectrum.9 These results are consistent with the presence of a general bias among the readership of all newspapers. Because the newspapers we look at have a relatively well-educated readership compared to the main tabloids or the regional newspapers (Jandura and Brosius, 2011), we would expect the slant in other media to be even stronger. We contribute to a growing empirical literature on media economics, which includes studies about the ideological position of media (e.g. Ansolabehere et al., 2006; Durante and Knight, 2012; Gentzkow and Shapiro, 2010; Groseclose and Milyo, 2005; Larcinese et al., 2011; Puglisi and Snyder, 2011, forthcoming), and the impact of the media on voting turnout (e.g. Gentzkow et al., 2011; Oberholzer-Gee and Waldfogel, 2009) and electoral outcomes (e.g. Chiang and Knight, 2011; DellaVigna and Kaplan, 2007; Enikolopov et al., 2011). Other studies analyze the impact of media on political outcomes (e.g. Besley and Burgess, 2002; Eisensee and Strömberg, 2007; Snyder and Strömberg, 2010; Strömberg, 2004) and household decisions (e.g. Olken, 2009). Our focus is on the functioning of media markets,10 and our results lend themselves to the interpretation that the slant is demand-driven rather than the result of agenda-setting. The empirical paper most closely related to ours is Gentzkow and Shapiro (2010) who find that newspapers tailor their slant to the beliefs of their potential readers. We are also close to Mullainathan and Shleifer (2005) who assume that newspapers receive an identical signal about the truth, but can slant their stories by omitting some of the information. When readers have a preference for news that is consistent with their initial beliefs, oligopolistic newspapers can charge higher prices by differentiating themselves through slanted reports. Because we find slant against foreign owners in all newspapers, and no slant in favor of foreign owners, it is likely that the underlying beliefs of the population are biased in a homogeneous way (similar to the beliefs about foreign policy, see Mullainathan and Shleifer (2005)). Gentzkow and Shapiro (2006) generate slant in a theory in which consumers think that newspapers that share their perspectives are more reliable. In their model, slant arises as a natural consequence of newspapers' desire to build a reputation for accuracy. Anderson and McLaren (2012) show that similar mechanisms prevail when consumers are fully rational. Our paper also contributes to a better understanding of the intricacies of corporate social responsibility (CSR) and its significance for globally operating businesses. Kitzmueller and Shimshack (2012) provide 8 The Springer family owned the majority of the publicly listed Axel Springer Publishing House throughout the entire period. In the meanwhile the media entrepreneur Leo Kirch, the Deutsche Bank, the private equity firm Hellman & Friedman hold larger minority stakes in the company. 9 Representative data about the political orientation of Die Welt's or other nationally distributed newspapers' readership is not readily available. Jandura and Brosius (2011) provide some evidence for the right-of-center orientation of Die Welt, and in Web Appendix I, we summarize the German political science community's perception about the newspapers' orientation. Further support for these perceptions is lent by a survey we carried out among students of Goethe University. A group of 102 participants in a lab experiment unrelated to the topic of this paper were asked about their political orientation on a scale from 1 (left) to 10 (right). We also asked them how frequently they are reading the national newspapers on a scale from 1 (never) to 10 (every day). Participants who are frequently reading (with values of 8 or more) the Frankfurter Allgemeine Zeitung have an average political orientation of 5.23 (s.d. 1.80; n = 39), Die Welt 4.93 (s.d. 1.68; n = 12), the Süddeutsche Zeitung 4.60 (s.d. 1.85; n = 20), the Frankfurter Rundschau 4.37 (s.d. 1.56; n = 30) and Die Tageszeitung 3.40 (s.d. 1.67; n = 5). 10 Related studies include Sweeting (2007, 2010) on product positioning in radio stations, George (2007) on the effect of concentration on news variety, and Myers (2005) on racial diversity and discrimination in competitive media markets.

G. Friebel, M. Heinz / Journal of Public Economics 120 (2014) 97–106

numerous examples of the scale and scope of CSR activities including public statements, costly certifications and investments. Labor policies, in particular, downsizing have great impact on a company's reputation (Karake, 1998), and media play a particular role in channeling information that is CSR relevant (Couttenier and Hatte, 2013). Our paper adds to these insights by showing that the identity of a company (foreign vs. domestic) is important for the differential impact of such activities on the company's reputation. The additional negative attention foreign firms attract in the media provides a concrete measure for the concept of “liability of foreignness” in the management literature (Zaheer, 1995).

2. Case studies on media slant and FDI To the extent that media slant may reinforce readers' negative beliefs about foreign owners, it may trigger or contribute to penalties imposed by consumers. Systematic evidence is hard to generate due to a lack of data, but the following three cases are illustrative of what we have in mind. When the Finnish multinational Nokia announced the shutdown of its mobile-phone plant in Bochum, Germany (a loss of 2300 jobs) on January, 16th 2008, the decision attracted overwhelmingly negative media attention. Nokia lost 8 percentage points in the German mobile phone market in the subsequent six months, while in the rest of Europe, Nokia's market share remained constant. Meanwhile, Nokia's global market share actually increased from 36 to 39 percentage points,11 suggesting that the loss in Germany cannot be explained by the entry of new technologies like the iPhone. We estimate that Nokia lost 220 million Euro of sales after the downsizing, based on the following back-of-the-envelope calculation. In 2007, Nokia sold 12 million mobile phones in Germany, representing a 44% market share. At an average price of 110€ per mobile phone, a loss of 8 percentage points of market share (i.e. 2 million sales) implies a fall in turnover of 220 million Euro in 2008. As shown in Table 1, Nokia also suffered enormous reputational damage as documented by the market research institute YouGov.12 Average quality ratings of Nokia mobile phones fell from +50 two weeks before to + 1 two weeks after the announcement (on a scale from + 100 to − 100). The willingness to recommend Nokia products, and the perceived price-performance ratio, also decreased. Nokia's ratings recovered slightly in the following months, but nine months after the announcement they had not reached half the pre-announcement level.13 The white-goods manufacturers Miele, Bosch Siemens HHG (both German) and Swedish Electrolux provide a second illustrative case. In 2005, Miele announced the shedding of 1078 jobs in Germany, BSH 420, and Electrolux 1750. Electrolux attracted four times more articles as Miele and BSH. While Miele and BSH increased their joint market share in Germany during 2006 by 5.2 percentage points, Electrolux lost 4 percentage points, moreover, its market share was constant in the rest of Europe.14 A third case illustrates that consumer penalties may even differ across regions. Since the mid-2008, there has been an ongoing discussion about GM Opel's plan to close its plant in Bochum in 2014, attracting much 11 Figures come from Gartner and GfK, market research institutes, and are documented in several newspapers and magazines, e.g. New York Times (http://www.nytimes.com/2008/ 04/17/technology/17iht-17nokia.12089585.html?_r=0). 12 These data are not publicly available, but can be provided by the authors upon request. 13 In comparison, BMW shed 7500 jobs at the same time, mainly in Leipzig, Eastern Germany, a much larger job loss of an equally well-known brand, in a region plagued by a higher unemployment rate. BMW, controlled by a German block-holder, received little media attention, the YouGov ratings remained constant and BMW gained in 2008 the largest market share in their history. 14 These figures also come from GfK and are documented in several newspapers and magazines, e.g. Nürnberger Nachrichten (http://www.nordbayern.de/nuernbergernachrichten/wirtschaft/electrolux-machte-mehr-gewinn-1.761846).

99

Table 1 Average rating of Nokia products by consumers around the downsizing announcement date.

Quality of products Willingness to recommend products Perceived priceperformance ratio

January 03–15, 2008

January 16–28, 2008

June 2008

September 2008

+50 +42

+1 −15

+14 +4

+19 +11

+31

−6

+2

+9

interest and negative attention from the German media. Between 2007 and 2012, Opel's market share decreased by 41% (from 15.1 to 8.9 percentage points) in the Ruhr valley, while n the whole of Germany it “only” decreased by 26% (from 9.1 to 6.8 percentage points). The German BILD tabloid called the consumer reaction a “message sent to the hardhearted GM-management in Detroit”.15

3. Data 3.1. Identifying articles about downsizing There are no data on downsizing in Germany, because the Labor Agency does not distinguish announcements from actual downsizing decisions. We identify downsizing through LexisNexis including only articles for which both firm identity and the actual number of jobs shed are known.16 Our algorithm is described in detail in Web Appendix I. We first identified a list of German synonyms for downsizing, ensuring intercoder reliability for the synonyms through the paid assistance of twenty students from different fields. An additional step was inspired by Eisensee and Strömberg (2007): after identifying all articles in the data base in which synonyms for downsizing appeared, we searched all articles on the 498 firms mentioned in these articles, thus investigating 40,000 articles. All told, we found 5394 articles on downsizing, with a total of 42417 companies and 651 downsizing cases. Finally, another 24 students received packages of articles either related or unrelated to downsizing. Their classifications were highly congruent with ours. We then counted the words in the paragraphs in which at least one synonym for downsizing appeared. Sometimes, all paragraphs in an article are on downsizing, but there are also articles in which the bulk of paragraphs refer to e.g. a new strategy or the CEO, and downsizing only appears as a minor issue. The descriptive statistics show that the proportion of an article, which is devoted to downsizing, constitutes an important distinction between the reports on domestic versus foreign firms (see Section 3.2.2). We only recorded articles on foreign firms' downsizing when we could verify that German locations were indeed affected by the downsizing of the company. In most downsizing events, the total number of jobs shed are mentioned in Die Welt's articles. For consistency, we cross-checked other quality media, agency reports like Reuters and DPA, and information from the company itself including annual reports and press releases. If there remained doubt about the total number of jobs shed, we omitted the entire downsizing case. In our data set, one observation is one downsizing event (or “case”) involving one firm.

15

http://www.uni-due.de/~hk0378/publikationen/2013/20130330_WAZ.pdf. In Web Appendix III, we provide examples of articles that did and did not qualify to be included in our data base. 17 For an additional 74 firms, we found reports of rumors, or announcements of downsizing, but no information regarding the number of jobs shed. These cases were not entered into the data set. 16

100

G. Friebel, M. Heinz / Journal of Public Economics 120 (2014) 97–106

Table 2 Break-down of firms (Panel A) and downsizing cases by characteristics (Panel B), both by origin of ownership. Panel A: Characteristics of firms

Panel B: Characteristics of downsizing cases

Number of employees in Germany Mean (SD) Domestic Foreign t-test (p-value)

≤750

Magnitude (total number of jobs shed)

751–2500 2501–7500 N7500 Mean (SD) ≤150

29,272 63 67 (56,510) (62%) (58%) 5083 39 49 (6713) (38%) (42%) 0.000

57 (54%) 49 (46%)

74 (74%) 26 (26%)

963.71 (2474.72) 557.15 (847.19) 0.021

Mean downsizing Establishment speed (in months; SD)

151–750 N750

115 198 (68%) (64%) 53 113 (32%) (36%)

113 14.03 (72%) (10.63) 43 12.75 (28%) (8.85) 0.247

Mean unemployment rate (SD)

Closure No closure 143 (63%) 83 (37%)

169 (64%) 94 (36%)

11.26% (3.88) 11.60% (3.85) 0.388

In Panel A, one observation is one firm. Crucial for the classification of firms is the total number of employees prior to the first downsizing event that we recorded between December 2000 and September 2008. Firms with changing ownership between different downsizing events are recorded under both categories. Column (1a) shows the mean number of employees in Germany, broken down by ownership. The corresponding standard deviations are in parenthesis. Column (2a) provides the total number of firms that employ 750 or less workers, column (3a) the total number of firms that employ between 751 and 2500 workers, column (4a) between 2501 and 7500 employees, column (5a) more than 7500. In Panel B, one observation is one downsizing case. Mean downsizing speed is the speed of downsizing as reported in the newspaper. Establishment captures whether a firm closes one or more establishments. Mean unemployment rate is the average quarterly unemployment rate in the cities or municipalities where the jobs are lost (according to Die Welt). As our sample includes 47 young hightech firms that tend to locate in booming cities like Munich (which have low regional unemployment rates), we calculate the same descriptive statistics omitting all firms listed in NASDAQ and NEMAX/TecDAX. The main qualitative results are the same. Note that we do not have information about downsizing speed, establishment closure and the unemployment rate for all downsizing cases.

3.2. Descriptive statistics 3.2.1. Firms, and downsizing events Our classification of domestic versus foreign firms is based on the prescriptions of German law that gives shareholders with 25% or more of the equity special control rights. In 426 cases, foreign blockholders hold less than 25% of the equity of the firms, in 209 cases between 45% and 100%. The first group of firms is treated as domestic, the second as foreign. For the remaining 16 downsizing cases, classifying firms was less clear-cut. Excluding those cases from our regressions, our final data set contains 5172 articles from 412 companies and 635 downsizing cases. In Web Appendix I, we provide more information about the classifications, the firms that we excluded, and a number of robustness checks to ensure the reliability of our results. In Web Appendix II (Table K) we also provide a summary of the downsizing cases broken down by two-digit industry classification and by ownership (domestic versus foreign).18 Except for seven out of 39 industries, the data set contains at least one domestic and one foreign firm in each industry. Panel A of Table 2 provides an overview of the size of firms, measured by the number of German employees before downsizing. Foreign firms are under-represented in the class of very large firms. In all regressions, we control for size and industry, and we later use propensity score techniques to control for self-selection into size, region and industry. Other covariates, such as assets or employment costs per capita, might differ between domestic and foreign firms. To control for those variables, we have matched a subset of large firms (80 domestic, 24 foreign) from our data set with financial and accounting information from the Amadeus data set (Bureau van Dijk, 2011). Table 3 provides an overview of sales, assets, debt, employment costs and profitability of domestic and foreign firms in 2001 (Panel B) and in the year before downsizing (Panel C). Only for one variable (tangible fixed assets, in Panel B) do we find statistically significant differences in a two-sided t-test, and controlling for it does not change the main results (Section 5). In Panel B of Table 2, we summarize several characteristics of the downsizing events: the magnitude of downsizing, downsizing speed (in months), the number of cases with establishment closures, and the quarterly local unemployment rate in the towns/districts where the downsizing occurs. With the exception of fewer large downsizing 18 Table L in Web Appendix II provides an overview using a one-digit industry classification. In all regressions, we include sector dummies based on two-digit industry classifications. Using one-digit industry dummies does not change the results (Web Appendix II, Table M).

events by foreign firms, we do not find great differences between domestic and foreign firms' downsizing events.19 3.2.2. Media slant: dependent variables Our approach to measuring newspaper slant is similar to the method used by Eisensee and Strömberg (2007), who compare the media coverage of natural disasters, depending on the number of persons killed and the region in which it occurs. As dependent variables we use: (i) the words per job lost (the total number of words about downsizing across all articles reporting on a given downsizing case, divided by the number of jobs shed in this event); (ii) the number of articles about downsizing for a given downsizing case, divided by the number of jobs shed in this event, scaled down by 1/1000. Panel A in Table 4 compares the newspaper coverage per job shed of foreign versus domestic firms' downsizing. The coverage (in words per job shed) of foreign firms that downsize is more than twice as large as that of domestic firms. Over 50% more downsizing articles per job shed are written on foreign firms. Panel B compares the size of articles, and the number of words devoted to downsizing (versus other news related to the firm). The size of the typical article is the same for domestic and foreign firms (388.96 compared to 388.61 words), but 21% more words are devoted to downsizing by foreign firms. Hence, media slant operates through two channels: more articles are written on downsizing by foreign firms, and in each article, more words are devoted to downsizing. The dependent variable words per job shed incorporates both these channels, because it accounts for the total slant. Hence, we focus on the total words written on a downsizing event, and report regressions with articles as the dependent variable in the Web Appendix II. The coefficients are always positive, but not always statistically significant.20 19 A concern could be that foreign and domestic multi-national enterprises (MNEs) may be downsizing more or in different ways than domestic ones. The literature on FDI finds no evidence for this (see e.g. Borrmann et al., 2003; Braconier and Ekholm, 2000; Buch and Lipponer, 2010; Konings and Murphy, 2006; Marin, 2004; Navaretti and Venables, 2004). A notable exception is Becker and Muendler (2010) who document that German MNEs tend to adjust at the intensive margin at home, but at the extensive margin abroad. Looking at the closure of establishments (the extensive margin), we find no descriptive differences between foreign and domestic firms in our sample. In our regressions in Section 4.1 we find no impact of closures on the media slant. 20 An additional interesting observation relates to firms that change ownership from foreign to domestic (or vice versa) and downsize under both owners. For the 13 such firms in our data set, the descriptive statistics show that when owned by domestic owners, the newspaper prints 1.84 (s.d. 1.31) words per job shed, and 15.41 (s.d. 13.10) articles per 1000 jobs shed. If the same firm is owned by foreign investors, media attention increases to 7.37 (s.d. 6.29), and 33.64 (s.d. 26.90) respectively.

G. Friebel, M. Heinz / Journal of Public Economics 120 (2014) 97–106

101

Table 3 Characteristics of the largest domestic and foreign downsizing firms. Domestic Panel A: Descriptive statistics for subsample of firms Total number of firms 80 Total number of downsizing cases 163 Words per job shed (mean, SD) 1.52 (1.90)

Foreign

t-test (p-values)

24 40 2.33 (1.67)

0.014

Panel B: Firm characteristics (year 2001) Obs. 63 60 75 64 74 70 60 73 76 61

Mean (SD) 8421 (25,418) 20.37 (1.97) 20.05 (2.03) 15.19 (2.72) 19.77 (1.93) 18.31 (1.91) 11.21 (0.971) 25.77 (7.50) 6.48 (9.73) 6.18 (3.21)

Obs. 11 17 19 18 19 16 9 19 19 14

Mean (SD) 4198 (9456) 20.18 (1.36) 18.74 (2.06) 14.94 (3.36) 19.22 (1.66) 17.98 (1.56) 11.25 (0.51) 18.15 (6.05) 5.34 (9.14) 10.31 (10.85)

0.590 0.707 0.014 0.738 0.253 0.521 0.909 0.614 0.645 0.549

Panel C: Firm characteristics (year before first downsizing) Employees 42 Ln (Sales) 39 Ln (Tangible fixed assets) 47 Ln (Intangible fixed assets) 43 Ln (Long term debts) 47 Ln (Total employment costs) 44 Ln (Employment costs per capita) 41 Return on equity 45 Return on assets 47 EBITDA margin 39

6050 (23,726) 19.75 (2.24) 19.84 (1.91) 15.26 (2.05) 19.26 (2.00) 17.65 (1.98) 11.31 (0.69) 10.00 (24.52) 3.51 (11.24) 7.38 (26.10)

10 9 12 11 13 10 8 11 12 9

1380 (2157) 20.10 (1.18) 18.84 (1.99) 15.00 (3.15) 19.11 (2.32) 17.55 (1.36) 11.42 (0.57) 21.44 (51.25) 6.83 (10.94) 7.71 (9.77)

0.540 0.661 0.113 0.742 0.818 0.886 0.660 0.281 0.364 0.515

Employees Ln (Sales) Ln (Tangible fixed assets) Ln (Intangible fixed assets) Ln (Long term debts) Ln (Total employment costs) Ln (Employment costs per capita) Return on equity Return on assets EBITDA margin

We selected the 500 largest German firms (based on turnover) and matched our data with firm-level financial and accounting information from the Amadeus data set (Bureau van Dijk, 2011). Panel A provides an overview on the total number of firms and downsizing cases, and the media coverage of those cases for our subset of firms, broken down by ownership. Data are from 2001–2007. Panel B shows the mean sales, mean tangible fixed assets, mean long term debt, mean total employment costs, mean employment costs per capita, mean return on equity, mean return on assets and mean EBITDA margin of the firms in the year 2001. The corresponding standard deviations are in parenthesis. Panel C provides the statistics for the year before the firm first sheds some jobs in our period of observation. Output and capital variables are deflated using deflators obtained from EU-KLEMS database. Note that we lose some observations in each row as we do not have the information for all downsizing firms. In column (5) we present the p-values of two-sided t-tests.

4. Empirical specification and basic results 4.1. Quantitative media slant against foreign firms Our baseline OLS regression is as follows: yi ¼ β0 þ β1 foreignit þ β2 X it þ εi;t

ð1Þ

with the dependent variable words per job shed. Ownership is captured by a foreign dummy variable, and controls include industry and time21 dummies and employment. Standard errors are robust, and clustering on the sector level does not change the results. The first two specifications of Table 5 present the results with and without controls. Media coverage of foreign firms is roughly two times higher. In specifications (3) and (4) we use words per job shed (+1) in logs as dependent variable, finding a media slant against foreign firms that is highly significant albeit of smaller size. Specifications

21 There are several ways to control for seasonal differences in reporting and time trends, because different articles on a downsizing event appear at different points of time. In most cases, the temporal structure of news reports is as follows. First, one or a few articles refer to rumors and advance notices, followed by an official announcement and background reports some days later. A final article or series of articles appears when downsizing is completed, usually some months later. The bulk of articles on a downsizing case appear in a short window of time: in 450 of the downsizing cases at least 75% of the articles and words about downsizing appear in one month; in 120 cases in two, in 43 cases in three months, and in 38 in four or more months. In the baseline regression, we include monthly and yearly time dummies. The dummies are set to one for the months (years) during which at least 75% of the words and articles have appeared, and zero otherwise. Other specifications (e.g. dummies set to one for the month/year where the first article appears) do not affect the results in a qualitative way (see Table N, Web Appendix II).

(5) to (6) use the logs of the raw numbers of words, and regress these in addition on the log of the magnitude of downsizing to allow for non-linear effects, for instance, because there is a minimum size of an article. For log of words, the size is similar to the ones in the baseline regressions. The coefficient of the interaction term between foreign and magnitude indicates that the slant is larger for firms that shed fewer jobs. Notice that specification (6) should be taken with a grain of salt because of the multicollinearity of employment and magnitude of downsizing.22 Table 6 presents the results of an augmented regression that takes into account additional characteristics of the downsizing event. We now include the magnitude of the downsizing, the magnitude relative to firm size, whether or not there was an establishment closure, regional unemployment, and, in a second specification, the speed of downsizing.23 The log of magnitude is the only statistically significant variable, indicating that smaller downsizing events receive relatively more attention. The foreign dummy remains stable. 24 Unemployment in the regional market does not seem to matter for media reports, which is surprising as local labor markets have quite

22 In Figure A in Web Appendix II we also present a scatterplot of words and the number of jobs lost within domestic and foreign firms which shows that the functional forms looks quite linear. 23 We also run propensity score estimations using the same characteristics of the downsizing events as covariates as in the OLS regression. The estimated coefficients of the average treatment on the treated are similar to our OLS baseline regressions. Results are provided in Web Appendix II, Table P. 24 We also run our baseline regression including a dummy for firms which, according to the newspaper reports, had received subsidies. However we only had 18 such cases and the results did not change much. The inclusion of an interaction term between foreign and subsidies did not affect the results.

102

G. Friebel, M. Heinz / Journal of Public Economics 120 (2014) 97–106

Table 4 Coverage per job shed (Panel A), mean number of words in the entire articles and the downsizing parts of the articles (Panel B), by origin of ownership. Panel A: Coverage per job shed

Panel B: Coverage per article

Ownership

Observations

Mean number of words per job shed

Mean number of articles per 1000 jobs shed

Ownership

Observations

Mean number of words in the whole article

Mean number of words reporting on downsizing per article

All cases Domestic Foreign

635 426 209

2.38 (3.58) 1.71 (2.19) 3.76 (5.13)

18.10 (22.23) 15.43 (18.11) 23.55 (28.14)

All firms Domestic Foreign

5172 3288 1884

388.83 (295.97) 388.96 (287.87) 388.61 (309.67)

148.45 (128.37) 137.75 (114.48) 167.12 (147.75)

In Panel A, one observation is one downsizing case. Column (2a) provides the mean number of words per job shed in all downsizing cases, broken down by origin of ownership, that is, we have summed up all words in the downsizing parts of articles of a given downsizing case and divided the number of words by the number of jobs shed. Column (3a) is the mean number of articles per 1000 jobs shed in Germany in each downsizing case. The corresponding standard deviations are in parenthesis. In Panel B, one article about downsizing is one observation. Column (1b) is the number of observations, broken down by ownership. Column (2b) is the mean number of words in the entire article, while column (3b) is the mean number of words reporting on downsizing per article, again broken down by origin of ownership. The corresponding standard deviations are in parenthesis.

Table 5 Determinants of coverage, OLS.

Table 6 Regression results, controlling for characteristics of the downsizing event, OLS.

Words per job shed

Ln (1 + words per job shed)

Ln (words)

Specifications

(1)

(2)

(3)

(4)

(5)

(6)

Constant

1.708*** (0.106) 2.050*** (0.370)

0.774 (0.713) 1.757*** (0.438)

0.818*** (0.026) 0.433*** (0.056)

0.451*** (0.115) 0.331*** (0.059)

1.633*** (0.214) 1.310*** (0.428) 0.722*** (0.037) −0.111

1.001*** (0.258) 1.226*** (0.391) 0.604*** (0.040) −0.123*

(0.075) 0.439 635 No No No

(0.067) 0.645 635 Yes Yes Yes

Foreign Ln (magnitude) Ln (magnitude) *Foreign 2

R Sample size Industry Employment Time

Words per job shed Specifications

(1)

(2)

Constant

5.760*** (1.377) 1.455*** (0.386) −1.190*** (0.193) −1.422 (0.928) 0.036 (0.455) 0.036 (0.455)

7.782*** (2.005) 1.347** (0.564) −1.146*** (0.265) 0.144 (1.174) 0.308 (0.600) −0.008 (0.069) −0.026 (0.031) 0.558 207 Yes Yes Yes

Foreign Ln(magnitude) Relative magnitude Establishment closure

0.073 635 No No No

0.300 635 Yes Yes Yes

0.102 635 No No No

0.395 635 Yes Yes Yes

One observation is one downsizing case. Dependent variables are the number of words per job shed (column (1) and (2)) and (1 + words per job shed) in logs (column (3) and (4)). In column (5) and (6), dependent variables are the total number of words reporting on downsizing. Foreign is a dummy variable which is set to 1 for foreign ownership. Ln(magnitude) indicates the total number of jobs shed (in logs) in the firm in Germany. In specification (2), (4) and (6) we include dummy variables for the months and years in which at least 75% of the articles/words about the downsizing event had appeared. We furthermore include the total employment by the firm in Germany before downsizing and dummy variables for industries that follow the two-digit classification from Destatis (2008). * p b 0.1, ** p b 0.05, *** p b 0.01; robust standard errors are in parenthesis.

different degrees of unemployment that develop differently over time. Also, German workers are relatively immobile compared to U.S. workers (David et al., 2010). Hence, one should expect media attention for downsizing to depend on the local unemployment rate.25 To look further into this issue, we split the sample of firms by quintiles of the unemployment rate, see Table 7. We find that domestic and foreign firms are distributed similarly across quintiles. This suggests that there is no self-selection of foreign or domestic firms in regions plagued by high unemployment rates. Fig. 1 offers an interesting observation: while there is no pattern regarding the correlation of unemployment quintile and media coverage per job shed for domestic firms, for foreign firms, when moving from the first to the fifth quintile, media coverage per job shed increases from 3.07 to 4.63. Thus, there is a higher media concern for unemployment when a foreign firm is the downsizing one. To deal with potential problems of self-selection of foreign owners we use propensity score methods (Becker and Ichino, 2002; Rosenbaum and Rubin, 1983).26 Foreign owners may systematically select firms that are different from domestic ones; for instance, they may sort into different

25 For example, Jacobson et al. (1993) have shown that workers' costs of losing their jobs are higher when the local unemployment rate is high, which would provide a rationale for the media to report more about this. 26 Here we use the Nearest-Neighbor method as our matching procedure. The main results are the same when using a Kernel method.

Unemployment Downsizing speed R2 Sample size Industry Employment Time

0.482 364 Yes Yes Yes

Relative magnitude indicates the number of jobs shed divided by the total number of workers employed by the firm in Germany before downsizing. Establishment closure is a dummy set to one if the firm closes an establishment, Unemployment the average quarterly unemployment rate in the cities/municipalities where the jobs are lost. Downsizing speed is the speed of downsizing as reported in the newspaper. Note that we lose some observations as we do not know whether an establishment gets closed, the regions where the jobs are lost and the downsizing speed for all downsizing cases. * p b 0.1, ** p b 0.05, *** p b 0.01; robust standard errors are in parenthesis.

industries or different regions. If these characteristics attract more interest from the media in general, they also do so when firms downsize, and our estimates would be biased. The results of the propensity score regressions are shown in Web Appendix II, Table R. The estimated coefficients of the average treatment on the treated are similar to our OLS baseline regression.27 Given our controls, we find no evidence that foreign firms in our sample have characteristics that systematically differ from domestic companies.

4.2. Qualitative media slant against foreign firms To investigate whether Die Welt describes the downsizing of domestic and foreign firms in different ways, one needs measures of qualitative aspects of reporting. While these are usually hard to measure, we received access to data by Media Tenor, a Swiss media consulting company.

27 Running estimations where we exclude all observations with a propensity score less than 0.1 or greater than 0.9 (Table R, Panel B), as suggested by Crump et al. (2009), leads to similar results.

G. Friebel, M. Heinz / Journal of Public Economics 120 (2014) 97–106

Domestic

Foreign

54 51 50 46 51

29 32 33 37 32

2.29 (3.24) 2.08 (2.71) 1.31 (1.21) 1.34 (1.08) 2.18 (2.73)

3.07 (2.68) 3.31 (2.44) 3.53 (4.09) 4.22 (5.80) 4.63 (4.21)

The table shows the mean number of words per job shed by quintiles of the mean local unemployment rate. Mean local unemployment rate is the average quarterly unemployment rate in the cities or municipalities where the jobs are getting lost (according to Die Welt). Standard deviations are in parenthesis. Note that we lose some observations as we do not know the regions where the jobs are lost for all downsizing cases. As our sample includes 47 young high-tech firms that tend to locate in booming cities like Munich (which have low regional unemployment rates), we calculate the same descriptive statistics omitting all firms listed in NASDAQ and NEMAX/TecDAX. The main qualitative results are the same.

.4

Domestic

.3

Foreign

.2

Domestic

.1

Mean number of words per job shed

0

3.8–7.9% 7.94–10.1% 10.1–11.7% 11.7–14.21% 14.29–22%

Observations

Density

1 2 3 4 5

Mean local unemployment rate

Foreign

.5

Table 7 Coverage per job shed, by quintiles of the mean unemployment rate. Quintile

103

-1

0

1

-1

0

1

Evaluations (articles about downsizing) Fig. 2. Distribution of the articles about downsizing, by ownership and evaluation.

1

2

3

4

5

Foreign

0

Mean number of words per job shed

Domestic

1

2

3

4

5

1

2

3

4

5

Fig. 1. Mean number of words per job shed across unemployment rate quintiles, lowest (1) to highest (5), by ownership.

Matching the data that Media Tenor recorded between August 2001 and September 2007 with our database, we have qualitative evaluations for 2109 articles. Each article is classified as positive (+1), neutral (0) or negative (−1).28 Fig. 2 provides an overview of the distribution of the articles, by ownership and evaluation. In total, the 1346 articles reporting about downsizing of domestic firms receive an average evaluation of − 0.215 (s.d. 0.678). However, the 763 downsizing articles reporting about foreign firms receive an average evaluation of −0.307 (s.d. 0.644). This difference (roughly 50%) is significant according to a t-test (two-sided, p-value: 0.001), and of a similar magnitude as the quantitative effects. To ensure that we are capturing a slant against foreign firms that downsize, and not one against foreign firms in general, we collected all articles on the 424 foreign and domestic firms that downsized in our data set. Then, we checked Media Tenor's qualitative evaluations of those articles on the firms, which are not related to downsizing. In

28 The company employs more than 120 employees who read several newspapers each day and record all articles where first, the name of any company is mentioned, and second, five or more sentences report about the company. The evaluation of each article is based on the following algorithm. First, the adjectives used are recorded and whether these are positive or negative. Second, it is checked whether the context of an article is consistent with the adjective-based evaluation. In case of contradiction, the evaluation is based on the context. To make sure that all employees of Media Tenor evaluate the articles in the same way, they receive three months of training. Once per month, Media Tenor's management randomly selects some articles and checks whether they have been coded in the right way. Employees' wages are based on the result of this check.

total there are 34,554 articles reporting about our domestic firms and 10,351 articles about our foreign firms. Articles about domestic firms receive an average evaluation of 0.012 (s.d. 0.572), foreign firms of 0.015 (s.d. 0.590). As indicated in Fig. 3 this difference is not significant (twosided t-test, p-value: 0.667). An interesting additional fact is that domestic firms receive more media attention in general (by a factor of three). This makes our result that Die Welt publishes more articles about foreign firms' downsizing even more striking.29 We have identified an additional fact in line with our story on slant against foreign firms that downsize. We examined all reports in LexisNexis on job creation, or “upsizing”, distinguishing foreign versus domestic firms. Using a similar algorithm to that for downsizing, we ended up with 451 articles about 76 companies and 100 upsizing events.30 A share of 67% of the upsizing cases report on domestic firms. Comparing the number of words in the paragraphs in which at least one synonym for upsizing appeared, we find no statistically significant difference in reporting on foreign versus domestic firms (134.66 to 125.27, with a two-sided t-test, p-value: 0.428). The average number of articles per upsizing case is higher for domestic compared to foreign firms (5.34 to 2.85), but the difference is not significant (two sided ttest, p-value: 0.445). The number of words per new job created (1.29 to 1.49, two sided t-test, p-value: 0.656) and articles per 1000 new jobs created (10.82 to 10.43, two sided t-test, p-value: 0.916) is roughly the same.31 5. Robustness checks In this section, we primarily address potential omitted-variable biases. All regression tables for this section are provided in Web Appendix II (Tables V–Z). First, we classify foreign and domestic companies into five categories: publicly listed, privately owned, private-equity, government owned and multiple/other owners. Interacting these classifications with domestic and foreign ownership, we find that the respective coefficients for

29 Using LexisNexis, we also counted all articles that mention the three leading domestic firms in each sector in other contexts, such as strategy, new products, and new management. We compared their media coverage with that of the three leading foreign-owned competitors, for instance, Siemens versus Alstom, VW versus Ford and found that 2.7 times more articles were written on the three largest German firms in each sector than on their foreign counterparts (see Web Appendix II, Table T). 30 The upsizing company with the lowest share of a foreign block-holder is N3 Engine Overhaul Services, a joint venture owned 50:50 by the German Lufthansa and the British Rolls-Royce. 31 In OLS regressions similar to the ones used for downsizing, with the dependent variables words per job created, and articles per 1000 jobs created, the foreign dummy is far from significant in any specification we tried (Web Appendix II, Table U).

104

G. Friebel, M. Heinz / Journal of Public Economics 120 (2014) 97–106

Foreign

6. Channels of the slant

.8

Domestic

.4 0

.2

Density

.6

6.1. News process, perception

-1

0

1

-1

0

1

Evaluations (articles not related to downsizing)

An event becomes news through the following stages: a firm's downsizing action or press release is received by the Deutsche Presse Agentur (DPA), the largest press agency in Germany, which then circulates the news to all journals. Thus, as argued by Gentzkow and Shapiro (2006) and Mullainathan and Shleifer (2005), all newspapers observe the same signal delivered by DPA, and then decide whether, how and how much to report. We checked for a subsample DPA's reports on 50 downsizing cases (sorted in descending alphabetical order) in our data set. The agency writes 2.71 (s.d. 2.56) words per job shed when the firm is owned by a German blockholder, and 1.95 (s.d. 2.40) for foreign-owned firms. DPA actually reports less on downsizing by foreign firms, which makes the slant more significant.33

Fig. 3. Distribution of the articles not related to downsizing, by ownership and evaluation.

6.2. Advertiser influence most of the foreign, but for none of the domestic interaction terms are significant. Hence, certain ownership forms receive more attention when the owner is foreign. It is also possible that the slant is against owners from specific countries. For instance, Anglo-Saxon countries are perceived to represent a “free market economy” which may have negative connotations, compared to the “social” (or Rhineland) market economies of Germany or France. Due to small sample sizes, we cannot control for each country separately. Rather, we use the following regional variables: Continental Europe, U.K./Ireland, North America, Japan, others. The estimated coefficients in this regression are all positive, significant, and of similar magnitude. We also conduct a number of regressions where we include controls like sales, assets, debt, employment costs and profitability for a subset of larger firms. The data are financial and accounting data from the Amadeus data set (Bureau van Dijk, 2011). The foreign coefficient is in most specifications statistically significant, but smaller than in the baseline regressions for large firms and, as discussed before, larger for smaller firms. Foreign firms may tend to shed jobs for reasons that differ from those of domestic firms. We augment our baseline model with the reasons given for downsizing in the media and press releases of the company. The list includes state intervention, offshoring, insolvency, M&A, other reasons, and multiple reasons. The regressions show a similar effect of foreign as before. The coefficient on state intervention is positive and significant; none of the other variables are statistically significant. Distinguishing M&A with acquiring firms that are headquartered in Germany from the ones headquartered in foreign countries shows another interesting pattern. M&A domestic is insignificant in the regression, but M&A foreign is positive and significant, i.e. downsizing related to mergers and acquisitions attracts more media attention if the acquiring firms are foreign owned.32 Foreign firms might not communicate their downsizing programs in the same way as German firms do, in particular, they may use PR differently. To investigate this point, one can compare domestic and foreign SMEs only, because SMEs tend not to invest a lot in PR, as shown by Solomon (2012). Running our baseline regressions on firms employing 500 or less workers in Germany we find that the foreign coefficient is positive, significant and substantially larger than in the baseline regression with the full sample. If anything, foreign firms seem to use PR to dampen media interest.

32 Examples include the heated public discussions in Germany when British Vodafone Airtouch acquired German Mannesmann in 2001, or when Spanish ACS acquired German Hochtief in 2011.

Ellman and Germano (2009), Gambaro and Puglisi (2010) and Reuter and Zitzewitz (2006) argue that advertisers' influence can cause slant. Domestic firms in our dataset could be placing more advertising than foreign ones. As a result, domestic firms could put pressure on editors or journalists to provide favorable coverage (e.g. write less about downsizing activities).34 However, as already shown, the media slant against foreign firms is even larger for SMEs. Because neither domestic nor foreign SMEs invest a lot in advertising in main national newspapers, this fact is not in line with a story that advertisers are the primary force explaining the slant. As a second analysis, we randomly chose five months (November 2002, July 2005, September 2006, May 2007, May 2008), and recorded all firms that placed advertisements in Die Welt in this period (290 advertisers). During those five months, a total of 147 downsizing cases are in our data set, with a total of 133 firms downsizing. 42 out of those firms were advertisers in Die Welt. As shown in Table AB in Web Appendix II, neither the inclusion of a dummy capturing whether a downsizing firm is placing advertising nor an interaction term between Advertiser and foreign/domestic affects the results. 6.3. Differentiation: evidence from six other newspapers A number of theoretical papers, in particular, Mullainathan and Shleifer (2005), and Anderson and McLaren (2012) have argued that newspapers are catering to the beliefs of their readership, and Gentzkow and Shapiro (2010) have presented empirical evidence. But, why would a business-friendly newspaper like Die Welt slant reports about foreign owners? There are two possible interpretations. One is that a conservative business-friendly readership may be xenophobic for reasons beyond a simple economic cost–benefit appraisal. Another interpretation is that in terms of beliefs about foreign owners everyone may be biased, and thus, newspapers with otherwise quite different readerships may report in similar ways. We look at six newspapers with a total of 1.56 million copies sold in 2008 who together with Die Welt represent 90% of the national quality newspaper market in Germany: Handelsblatt and Financial Times Deutschland (FTD), the Frankfurter Allgemeine Zeitung (FAZ), Süddeutsche Zeitung (SZ), Frankfurter Rundschau (FR) and Die Tageszeitung (TAZ). 33 To exclude the possibility that journalists write more about downsizing in Hamburg and Berlin (headquarters of Die Welt and the Axel Springer Publishing House), as a large portion of them (and large parts of the readership) might live there, we run our baseline including a dummy for jobs shed in the two towns. As shown in Table AA in Web Appendix II, the main qualitative results are the same. 34 This is an argument similar to political economy explanations of media bias, where journalists report favorably about certain policies, because they need to maintain a good relationship to politicians who provide them with privileged material.

G. Friebel, M. Heinz / Journal of Public Economics 120 (2014) 97–106

105

Table 8 Downsizing reports in the seven leading national newspapers for five randomly selected months, by origin of owner. Taz

FR

SZ

FAZ

Die Welt

Handelsblatt

FTD

(left)

(left-center)

(left-center)

(right-center)

(right-center)

(business)

(business)

Panel A: Total number of articles about downsizing All firms 86 129 Domestic 63 84 Foreign 23 45

143 100 43

Panel B: Mean number of words in the downsizing part of the articles All firms 201.7 126.0 185.3 (217.3) (84.6) (145.7) Domestic 188.2 116.0 172.7 (222.7) (75.4) (121.9) Foreign 238.7 144.5 214.8 (201.9) (97.8) (188.2)

150 112 38

217.8 (145.9) 208.6 (138.6) 244.7 (164.4)

351 246 105

141.6 (105.7) 135.5 (101.1) 155.9 (155.2)

192 128 64

308.2 (240.1) 293.8 (250.2) 336.9 (217.4)

209 138 71

124.8 (88.8) 121.7 (77.1) 130.8 (108.3)

Panel A reports the total number of articles reporting about downsizing in November 2002, July 2005, September 2006, May 2007 and May 2008 for the TAZ, SZ, FAZ, Die Welt, Handelsblatt and FTD, broken down by ownership. Panel B reports the mean number of words reporting on downsizing per articles for TAZ, SZ, FAZ, Die Welt, Handelsblatt and FTD. Standard deviations are in parenthesis. Panel A and B also report the same data for FR, but only for July 2005, September 2006, May 2007 and May 2008 as the data for November 2002 were not available in LexisNexis.

To investigate whether the slant found in Die Welt is idiosyncratic or may be a general phenomenon of these quality newspapers in Germany, we randomly chose five months35 from our period of observation and carried out the same algorithm as the one used for the initial data set, allowing us to measure whether the average number of words on foreign versus domestic firms shows patterns across newspapers.36 The descriptive results (see Panel B of Table 8) show that all newspapers slant their news in the same direction.37 This observation is in line with Mullainathan and Shleifer (2005) who argue that people have heterogeneous beliefs on some and homogeneous beliefs on other topics. They argue that foreign policy is a good example of the latter. Our data indicate that negative beliefs about foreign owners' impact on employment may be similarly homogeneous in the population. The magnitude of the slant, however, is quite different. The business newspaper FTD spends on average 1.07, and Handelsblatt 1.15 times, more words per article on foreign firms' downsizing. The ratio of Die Welt is 1.15, consistent with the initial dataset. FAZ is considered a right-of-center newspaper, and has a ratio of 1.17. SZ and FR, generally seen as left-of-center, have ratios of 1.24 and 1.25, respectively. TAZ with its distinct left-wing orientation has the strongest slant with a ratio of 1.27.

6.4. Owner influence There is another possible explanation for the slant: newspaper owners could be biased (as studied by Gentzkow and Shapiro (2010), who find no evidence, while Larcinese et al., (2011), do). The newspapers we look at have different types of owners: foundations, cooperatives and families. We can also use the fact that three newspapers changed their owners in our period of observation, all because of severe financial difficulties. The FR was majority-owned by a foundation until 05/2004 (Karl-Gerold-Stiftung), by the social-democratic party SPD until 07/2006, and afterwards by the DuMont Schauberg family. The SZ was owned by several families, and in 03/2008 acquired by the entrepreneur Dieter Straub and several other families. FTD initially was a 50– 50 joint venture of Bertelsmann and the U.K. based Pearson Group; but in January 2008 Bertelsmann acquired Pearson's 50% share. We expand our dataset from the previous section by randomly chosen months until we have at least three months of observations for each owner of each 35

November 2002, July 2005, September 2006, May 2007, May 2008. We cannot investigate whether during those five months, the newspapers published more articles, because the five months are disjunct, and the publishing sequence of articles may differ between newspapers. 37 The nature of this (third) data set does not lend itself to statistic tests or regressions and should hence be taken with a grain of salt. 36

newspaper. Table AC in Web Appendix II shows that owner changes have no substantial impact on any newspaper's slant. 7. Concluding remarks We have established a robust fact about media reports on downsizing by foreign owners. In Germany, for any job shed, a foreign firm receives twice as much media attention when it downsizes than a domestic one. Germany is a leading export country with a population inclined to globalization, and the effect is both robust against the inclusion of relevant controls, and against potential self-selection of foreign owners into regions or industries that could attract negative attention. The slant is present in all national quality newspapers, indicating that there may be homogenous, and biased beliefs in the respective readerships. The media play an important role in transmitting news to people, and this news and the information it contains shape economic decisions. The slant we identify, may have severe economic consequences and cause substantial obstacles to FDI. To the extent that a potentially negative a-priori belief of the population about the costs of globalization is strengthened by the media, foreign companies may fear punishment for activities that domestic firms would barely be noticed for. Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.jpubeco.2014.09.001. References Anderson, Simon P., McLaren, John, 2012. Media mergers and media bias with rational consumers. J. Eur. Econ. Assoc. 10 (4), 831–859. Ansolabehere, Stephen, Lessem, Rebecca, Snyder Jr., James M., 2006. The orientation of newspaper endorsement in U.S. elections. Q. J. Polit. Sci. 1 (3), 393–404. Baron, David P., 2006. Persistent media bias. J. Public Econ. 90 (1–2), 1–36. Becker, Sascha, Ichino, Andrea, 2002. Estimation of average treatment effects based on propensity score. Stata J. 2 (4), 358–377. Becker, Sascha, Muendler, Marc-Andreas, 2010. Margins of multinational labor substitution. Am. Econ. Rev. 100 (5), 1999–2030. Besley, Timothy, Burgess, Robin S.L., 2002. The political economy of government responsiveness: theory and evidence from India. Q. J. Econ. 117 (4), 1415–1451. Besley, Timothy, Prat, Andrea, 2006. Handcuffs of the grabbing hand? Media capture and government accountability. Am. Econ. Rev. 96 (3), 720–736. Borrmann, Christine, Jungnickel, Rolf, Keller, Dietmar, 2003. Auslandskontrollierte Unternehmen – ein Gewinn für den nationalen Arbeitsmarkt? Nomos Verlag, Baden-Baden. Braconier, Henrik, Ekholm, Karolina, 2000. Swedish multinationals and competition from high- and low-wage locations. Rev. Int. Econ. 8 (3), 448–461. Buch, Claudia M., Lipponer, Alexander, 2010. Volatile multinationals? Evidence from the labor demand of German firms. Labor Econ. 17 (2), 345–353. Bureau van Dijk, 2011. Amadeus. A Database of Comparable Financial Information for Public and Private Companies across Europe, Database.

106

G. Friebel, M. Heinz / Journal of Public Economics 120 (2014) 97–106

Chiang, Chun-Fang, Knight, Brian, 2011. Media bias and influence: evidence from newspaper endorsements. Rev. Econ. Stud. 78 (3), 795–820. Couttenier, Mathieu, Hatte, Sophie, 2013. Mass media effects on the production of information: evidence from Non-Governmental Organization (NGO) Reports. Working Paper. Crump, Richard K., Hotz, Joseph, Imbens, Guido, Mitnik, Oscar A., 2009. Dealing with limited overlap in estimation of average treatment effects. Biometrika 96 (1), 187–199. David, Quentin, Janiak, Alexandre, Wasmer, Etienne, 2010. Local social capital and geographical mobility. J. Urban Econ. 68 (2), 191–204. DellaVigna, Stefano, Kaplan, Ethan, 2007. The Fox News effect: media bias and voting. Q. J. Econ. 122 (3), 1187–1234. Durante, Ruben, Knight, Brian, 2012. Partisan control, media bias, and viewer responses: evidence from Berlusconi's Italy. J. Eur. Econ. Assoc. 10 (3), 451–481. Eisensee, Thomas, Strömberg, David, 2007. News droughts, news floods, and U.S. disaster relief. Q. J. Econ. 122 (2), 693–728. Ellman, Matthew, Germano, Fabrizio, 2009. What do the papers sell? A model of advertising and media bias. Econ. J. 119 (537), 680–704. Enikolopov, Ruben, Petrova, Maria, Zhuravskaya, Ekaterina V., 2011. Media and political persuasion: evidence from Russia. Am. Econ. Rev. 101 (7), 3253–3285. Gambaro, Marco, Puglisi, Riccardo, 2010. What do ads buy? Daily coverage of listed companies on the Italian Press. Working paper. Gentzkow, Matthew, Shapiro, Jesse M., 2006. Media bias and reputation. J. Polit. Econ. 114 (2), 280–316. Gentzkow, Matthew, Shapiro, Jesse M., 2010. What drives media slant? Evidence from U.S. Daily Newspaper. Econometrica 78 (1), 35–71. Gentzkow, Matthew, Shapiro, Jesse M., Sinkinson, Michael, 2011. The effect of newspaper entry and exit on electoral politics. Am. Econ. Rev. 101 (7), 2980–3018. George, Lisa, 2007. What's fit to print: the effect of ownership concentration on product variety in daily newspaper markets. Inf. Econ. Policy 19 (3–4), 285–303. Groseclose, Timonthy, Milyo, Jeffrey, 2005. A measure of media bias. Q. J. Econ. 120 (4), 1191–1237. Jacobson, Louis S., LaLonde, Robert J., Sullivan, Daniel G., 1993. Earnings losses of displaced workers. Am. Econ. Rev. 83 (4), 685–709. Jandura, Olaf, Brosius, Hans-Bernd, 2011. “Wer liest sie (noch)? Das Publikum der Qualitätszeitungen”. In: Blum, Roger, Bonfadelli, Heinz, Imhof, Kurt, Jarren, Otfried (Eds.), Krise der Leuchttürme öffentlicher Kommunikation. VS Verlag, Wiesbaden. Karake, Zeina A., 1998. An examination of the impact of organizational downsizing and discrimination activities on corporate social responsibility as measured by a company's reputation index. Manag. Decis. 36 (3), 206–216. Kitzmueller, Markus, Shimshack, Jay, 2012. Economic perspective on corporate social responsibility. J. Econ. Lit. 50 (1), 51–84. Konings, Jozef, Murphy, Alan Patrick, 2006. Do multinational enterprises relocate employment to low-wage regions? Evidence from European multinationals. Rev. World Econ. 142 (2), 267–286. Larcinese, Valentino, Puglisi, Riccardo, Snyder Jr., James M., 2011. Partisan bias in economic news: evidence on the agenda-setting behavior of U.S. newspapers. J. Public Econ. 95 (9-10), 1178–1189.

Marin, Dalia, 2004. ‘A nation of poets and thinkers’ — Less so with eastern enlargement? Austria and Germany. CEPR Discussion Paper 4358. Mayda, Anna-Maria, Rodrik, Dani, 2005. Why are some people (and countries) more protectionist than others? Eur. Econ. Rev. 49 (6), 1393–1430. Mullainathan, Sendhil, Shleifer, Andrei, 2005. The market for news. Am. Econ. Rev. 95 (4), 1031–1053. Myers, Caitlin Knowles, 2005. Discrimination as a competitive device: the case of local television news. IZA Discussion Paper No. 1802. Navaretti, Giorgio Barba, Venables, Anthony J., 2004. Multinational Firms in the World Economy. Princeton University Press, Princeton and Oxford. Oberholzer-Gee, Felix, Waldfogel, Joel, 2009. Media markets and localism: does local news en Espanol boots hispanic voter turnout? Am. Econ. Rev. 99 (5), 2120–2128. OECD, 2003. VII. Foreign Direct Investment Restrictions in OECD Countries, Paris. Olken, Benjamin A., 2009. Do television and radio destroy social capital? Evidence from Indonesian villages. Am. Econ. J. Appl. Econ. 1 (4), 1–33. Puglisi, Riccardo, 2011. “Being The New York Times: the political behaviour of a newspaper”, The B.E. J. Econ. Anal. Policy 11 (1), 1–20. Puglisi, Riccardo, Snyder Jr., James M., 2011. Newspaper coverage of political scandals. J. Polit. 73 (3), 1–20. Puglisi, Riccardo, Snyder Jr., James M., 2014. (forthcoming) J. Eur. Econ. Assoc. (“The Balanced U.S. Press”). Reuter, Jonathan, Zitzewitz, Eric, 2006. Do ads influence editors? Advertising and bias in the financial media. Q. J. Econ. 121 (1), 197–227. Rosenbaum, Paul R., Rubin, Donald B., 1983. The central role of propensity score in observational studies for causal effects. Biometrika 70 (1), 41–55. Scheve, Kenneth F., Slaughter, Matthew J., 2006. Public opinion, international economic integration, and the welfare state. In: Bardhan, Pranab, Bowles, Samuel, Wallerstein, Michael (Eds.), Globalization and Egalitarian Redistribution. Russell Sage Foundation, New York, pp. 217–260. Snyder, James M., Strömberg, David, 2010. Press coverage and political accountability. J. Polit. Econ. 118 (2), 355–408. Solomon, David, 2012. Selective publicity and stock prices. J. Financ. 67 (2), 599–638. Soltes, Eugene, 2009. News Dissemination and the Impact of the Business Press. The University of Chicago Graduate School of Business (Ph.D. diss). Strömberg, David, 2004. Radio's impact on public spending. Q. J. Econ. 119 (1), 189–221. Sweeting, Andrew, 2007. Dynamic product repositioning in differentiated product markets: the case of format switching in the commercial radio industry. NBER Working Paper 13522. Sweeting, Andrew, 2010. The effects of mergers on product positioning: evidence from the music radio industry. RAND J. Econ. 41 (2), 372–397. Zaheer, Srilata, 1995. Overcoming the liability of foreignness. Acad. Manag. J. 38 (2), 341–363.