The Political Legacy of News-Free Television: Evidence from the Rise of Berlusconi

Transcription

1 The Political Legacy of News-Free Television: Evidence from the Rise of Berlusconi Ruben Durante Paolo Pinotti Andrea Tesei February 22, 2013 ABSTRACT We investigate the impact of news-free commercial television on political and social behavior in Italy exploiting differences across municipalities in the early availability of Silvio Berlusconi s private TV network, Mediaset, largely attributable to geomorphological factors. We document that municipalities that had access to Mediaset prior to 1985 displayed significantly higher electoral support for Berlusconi s party in 1994, when he first ran for office. This effect, quantified between 1 and 2 percentage points, tends to persist in the following three elections and is very significant and robust to several specifications and falsification tests. In particular, we find that exposure to Mediaset before 1985 is uncorrelated with electoral support for any other party in any election between 1976 and Based on the timing of the introduction of news programs on Mediaset channels and on additional evidence from survey data, we can exclude that the documented effect is due to differential exposure to partisan news bias. Instead, we present evidence that earlier exposure is associated with a substantial decline in social capital consistent with the diffusion of a culture of individualism and civic disengagement that characterized Mediaset programs and which favored Berlusconi s political message. We thank Brian Knight, David Weil, and Ekaterina Zhuravskaya for very helpful comments and seminar participants at Bocconi, CREI, NYU, MIT, Sciences Po, Brown, Dartmouth, Paris 1, WZB, Surrey and Queen Mary for helpful discussion. We thank Nicola D Amelio and Giuseppe Piraino for their assistance with the collection of electoral data and Laura Litvine for her outstanding help with the digitalization of the transmitters data. Ruben Durante thanks the Sciences Po Research Board for financial support. Sciences Po; contact: Ruben.Durante@sciences-po.org (corresponding author). Bocconi University; contact: paolo.pinotti@unibocconi.it. Queen Mary University; contact: a.tesei@qmul.ac.uk. 1

2 1. INTRODUCTION As of February 2013, Silvio Berlusconi is running for the sixth time at the Italian national elections as the leader of the center-right coalition. This extraordinary political longevity is intimately related to his dominant position in Italian TV industry, thanks to the control over the private network Mediaset (Durante and Knight, forthcoming). A burgeoning literature in economics suggests that politically-oriented media may influence electoral results by exposing voters to partisan bias in news and information programs. DellaVigna and Kaplan (2007) and Enikolopov, Petrova, and Zhuravskaya (2011) show that exposure to Fox News (a conservative network) and NTV (an independent news channel) had significant effects on electoral participation and voting choices in the U.S. and Russia, respectively. Another strand of literature looks at the effects of media along other dimensions of individual behavior. Path-breaking work by Robert Putnam (2000) blames television, and in particular light entertainment shows, for civic dis-engagement in the U.S. during the postwar period. More recently, Olken (2009) documents a negative relationship between the number of TV channels available across different areas of Indonesia and several aspects of social capital (trust, participation in social organizations and control of corruption). On related matters, exposure to soap operas reduces fertility and increases divorce rates in Brazil (Chong, Duryea, and La Ferrara, 2008; Chong and La Ferrara, 2009), satellite and cable TV contribute to women empowerment in India (Jensen and Oster, 2009) and exposure to Western TV drove consumers in former East Germany toward goods that were more intensively advertised before the re-unification (Bursztyn and Cantoni, 2012). This body of work points at an important role of television for the diffusion of cultural models that, once adopted, may have significant and persistent effects on individual behavior. So far, these two strands of literature have evolved in parallel. However, adherence to new cultural models may well be reflected into different voting behavior. Put differently, even news-free television may have significant, and perhaps subtler, political consequences. This paper shows that this is exactly what happened in Italy over the last 30 years: differential exposure to non-informative, light-entertainment TV shows had significant and long-lasting effects on electoral outcomes over the following decades. Our empirical analysis compares voting behavior, and particularly electoral support for Berlusconi after he entered politics in 1994, between Italian municipalities that were exposed earlier to his TV network, Mediaset, and municipalities that were exposed only later. The data combine unique information on the location and technical characteristics of Mediaset transmitters in 1985 and electoral results across Italian municipalities over the period

3 Our identification strategy takes advantage of the staggered introduction of Mediaset across geographical areas and exploits exogenous variation in signal reception due to idiosyncratic geomorphological factors. After controlling for average topography and for local fixed effects, residual variation in signal is indeed uncorrelated with voting for any party in the elections held during the pre-exposure period ( ). Turning to our main results, we uncover a significant electoral advantage for the personal party of Berlusconi, Forza Italia, in municipalities that were exposed to Mediaset before The effect - between 1 and 2 percentage points - is very precisely estimated and persists until 2006 (the last election before Forza Italia merged with former coalition allies into a new party). The estimated coefficient does not depend on variation across a few large cities (if anything, it increases across sub-samples of small cities characterized by extreme variation in signal reception) and is higher for areas where people spend more time watching TV. We also pool together the data for all elections over the period and estimate several difference-in-differences specifications in which we interact exposure before 1985 with dummy variables for different sub-periods. Between the advent of Mediaset and the creation of Forza Italia ( ), early exposure drives lower voting turnout and votes swings away from the traditional mass parties (the Christian Democrats and the Communists alike) and toward a few minor (and often short-lived) parties. After 1994, the rise of the newborn Forza Italia in municipalities that were earlier exposed to Mediaset is paralleled by a reversal in these trends, as electoral participation increases and voting for other minor parties decreases in such municipalities. Therefore, the political success of Berlusconi may have ultimately intercepted the changes in voting behavior triggered, a few years before, by the expansion of his own television network. However, any electoral effect of exposure to Mediaset before 1985 can hardly be attributed to explicit partisan bias in news. The programming schedule at that time was devoted exclusively to action dramas, soap operas, quiz and gossip shows. The first newscast would arrive only in 1991, at which time Mediaset was available to the entire Italian population (other information programs were also introduced in later years). We also exclude that earlier exposure increased access to Mediaset newscasts after they were introduced (for instance, because of habit formation or attachment to the network), as individual-level survey data for the period do not reveal any systematic difference in this respect between individuals that were differentially exposed. Interestingly, opinions about the honesty and ability of Berlusconi are also similar. Therefore, greater support for Forza Italia among the voters that were exposed first (confirmed also in the individual-level regressions) is not explained by access to different sources of information in 1994, nor by different beliefs about Berlusconi.

4 An alternative explanation, which is consistent with our results, is that voters filter the same information and beliefs through the lens of different values and preferences, which were shaped in turn by differential exposure to Mediaset. The latter doubled the number of TV channels (from three to six) and expanded dramatically the range of TV shows to light entertainment (as opposed to the educational vocation of public television at that time). According to Putnam (2000), both effects go in the direction of a decrease in the civic engagement of individuals that were exposed earlier, which could also be reflected into a different voting behavior. Consistent with this hypothesis, we observe a strong differential decline, between 1981 and 1991, in the incidence of voluntarily associations (one of the measures of civic engagement used by Putnam, 2000) in municipalities that were exposed before 1985 relative to municipalities that were exposed only later. Participation to voluntarily associations and other measures of civic engagement are in turn negatively associated with voting for Forza Italia, both at the municipality- and at the individual-level. Although we are extremely careful in attaching any causal interpretation to these last findings, they are consistent with the diffusion of a cultural model of individualism and civic dis-engagement, on the part of Mediaset, which ultimately paved the way for the political success of Berlusconi. These results parallel recent work by Barone, D Acunto, and Narciso (2012). Using electoral data for a local election in an Italian region, they document a decline in voting for the centreright coalition in western provinces, which had switched to digital TV before the elections, relative to eastern provinces, which switched to digital TV only afterward. Since digital TV diluted Mediaset s share in TV news, they attribute such effect to partisan bias in Mediaset. We complement these findings with a throughout investigation of the role of Mediaset in the recent political history of Italy. Based on this evidence, we conclude that political effects reflect wider social trends triggered by early exposure to media-fueled cultural models. The remainder of the paper is organized as follows. Section 2 provides more information on the evolution of Italy s political system and broadcast television industry over the period of interest. In section 3 we illustrate the data and the identification strategy. Section 4 presents the main findings. In section 5 we discuss alternative interpretations of such results and we provide additional evidence to help distinguishing between them. Section 6 concludes.

5 2. BACKGROUND INFORMATION 2.1. THE RISE OF COMMERCIAL TV IN ITALY The state-owned TV company (RAI) was established in For over twenty years it maintained the public monopoly of broadcasting over the Italian territory. Despite some attempts to obtain the concession of frequencies, private networks were prohibited to broadcast over the air, on the consideration that the State protected and guaranteed better than the private the conditions of impartiality, objectivity and completeness of the television service (judgment 59/1960 of the Constitutional Court). At the beginning of the 1970s, however, an increasing number of local televisions started broadcasting via cable, exploiting the legislative vacuum concerning cable emission. Subject to increasing public pressure, in 1976 the Constitutional Court allowed private televisions to broadcast at the local level, both via cable and over the air (judgment 202/1976 of the Constitutional Court). The sector took off in the years immediately after the sentence: in 1977 there were already 256 private televisions, increasing to 400 one year later (Menduni, 2002). Together with local entrepreneurs, larger business groups also became interested in private television, attracted by the potential profits from advertising. They soon established broadcast syndications of several (formally) independent local televisions that transmitted the same show simultaneously across different areas, effectively circumventing the ban to broadcast beyond the local level. One such syndication, Canale 5, was created by Berlusconi in 1980; the other important ones were Prima Rete, Italia 1 and Rete 4, held by Rizzoli, Rusconi and Mondadori, respectively. Most syndications were however short-lived, due to stiff competition over the same products (movies and TV shows) and a climate of legal uncertainty. In 1981 the Constitutional Court reiterated the ban to expand beyond the local level, leading Prima Rete to leave the market and convincing Rusconi and Mondadori that antitrust legislation was on its way. Only Berlusconi was prepared to sail closest to the wind (?). Risking illegality, he extended his network and explicitly grouped the stations under the common logo of Canale 5. In the absence of any further political intervention and thanks to the soaring profits from advertising, he then negotiated from a vantage point the acquisition of his main competitors, Italia 1 (in 1982) and Rete 4 (in 1984). Canale 5, Italia 1 and Rete 4 were incorporated into the holding Fininvest, later renamed Mediaset. 1 By 1984 Berlusconi had thus established a 1 In the rest of the paper, we simply refer to Mediaset to denote the network of commercial televisions owned by Berlusconi.

6 near monopoly in private broadcasting, controlling the same number of channels as the State television RAI. The TV shows launched by Mediaset in the early 1980s represented an absolute novelty in the Italian television landscape, so much that it profoundly influenced the lifestyle models over the following decades (??). Most of the airtime was devoted to foreign TV series, especially action dramas and soap operas, and the rest was filled with internal productions: quiz, gossip and entertainment shows. The latter type of programs adhered to a cultural model centered on individualism, materialism and hedonism. The yuppies and playboys that populated these shows exemplify well the new ideal types, while women were relegated to the role of sexual bodies (Zanardo, 2010; Benini, 2012). At the opposite, Mediaset did not have any newscast until the 1990s. Italia 1 introduced news at the time of the first Iraqi war, in 1991, followed the year after by Canale 5 and Rete 4. Also, the three private channels never engaged (until very recently) in educational television. This stood in sheer contrast with the pedagogical nature of public channels at that time, entirely devoted to documentaries, newscasts and family films. This revolutionary approach to television-making proved indeed extremely successful. According to the independent technical assessment included in Constitutional Court (1988) and based on a Nielsen survey conducted in 1987, the share of viewing time for the three Mediaset channels between 7:00am and 2:00am reached 44.7%, slightly below that of the three RAI channels at 48.3%. In terms of advertising (the main source of revenues for commercial televisions) the turnover of Mediaset even exceeded that of RAI. Reactions to Berlusconi s concentration in the television industry arrived soon. On October 16th 1984 the attorneys of Turin, Pescara and Rome ordered the switch off of the Mediaset transmitters, which were openly in breach of the ruling 202/1976. A few days later, however, the government led by Bettino Craxi issued a law decree suspending the ban for private companies to broadcast at the national level. The decree, initially rejected by the Parliament, was forcefully reiterated and finally approved on February 4th, The so-called Berlusconi Decree represents a landmark in the evolution of the Italian television system, as it made apparent to everybody that the ban to broadcast at the national level would remain dead letter. For this reason, after 1985 Mediaset invested massively in the acquisition of new transmitters, reaching within a few years an almost complete coverage of the national territory. Already in 1987, 3,800 transmitters irradiated the signal of Canale 5, Italia 1 and Rete 4 to 92%, 86% and 80% of the national population, respectively (Constitutional Court, 1988). Such figures were comparable to those for the three public channels (98%, 97% and 80%, respectively). In 1990 the Mammì Law merely acknowledged the

7 existence of this public-private duopoly in national broadcasting. Until 1985, however, the uncertain legal prospects faced by Mediaset had delayed the geographical expansion of the network. According to our data, Mediaset could count at that time on just 1,700 transmitters inherited from the members of the broadcast syndication. In fact, Mediaset never built its own antennas, finding it cheaper to use those of the small local televisions that were progressively incorporated into the network. These had been conceived to reach a local audience and lacked the power of RAI transmitters. Based on the radio propagation software presented in Section 3, Mediaset could guarantee just half of the population with a perfect reception of its channels. In our empirical analysis we will exploit this heterogeneity in signal reception across different geographical areas to estimate the impact of earlier exposure to Mediaset channels on the support for Berlusconi, after he first run for election ten years later. Figure 1 reports the timeline of the events just described THE POLITICAL SYSTEM AND THE CREATION OF FORZA ITALIA Berlusconi had no intention to pursue a political career until a few months before the 1994 national elections. His decision came after the collapse of the First Italian Republic in and the disgrace of his long-time protector, the powerful leader of the Italian Socialist Party (PSI) Bettino Craxi. The PSI was an historical leftist party founded in 1892, later embracing social democracy and eventually joining the coalition formed by the Christian Democrats and other minor parties in While the dominant socialist leadership had always accommodated the existence of a liberal faction inside the party, the latter prevailed definitely when Craxi was elected secretary in The new leader cut the residual links with the left and enthusiastically embraced market-oriented values, overturning the Christian Democrats at the center-right of the political spectrum. In this way he acquired a pivotal position in the political arena, obtaining the post of Prime Minister in the new coalition government sworn in after the election of Far from being a profound political thinker, Craxi built most of the success of the PSI upon a renewed public image. In the words of?, Craxi offered a modernizing veneer which Italy s two major parties, the Christian Democrats and Communists, both lacked (...) under his leadership, politics were to be personalized and simplified, they were to have a strong showbiz element, their principal medium was to be television. On these premises, Craxi naturally became the protector and a very close friend of Berlusconi, to the point of being the godfather of his daughter. The political fortunes of the PSI came to an abrupt end with the collapse of First Republic

8 amidst massive corruption scandals. The so-called Tangentopoli (Italian for bribeville ) started indeed with the arrest of a local administrator affiliated to the party and eventually brought to the bar the whole leadership of the PSI. Craxi himself was sentenced to ten years and could avoid the prison just by escaping to Tunisia, where he died a few years later. The Christian Democrats and the other members of the center coalition, which had governed the country for thirty years, shared a similar destiny. In the wake of the emergency, a technical government led the transition to the Second Republic by calling for new elections in March 1994, after a popular referendum had changed the electoral rule from a fully proportional to a mixed majoritarian-proportional system. Against this background, the Berlusconi group had lost its former political allies and the electoral polls were predicting a large victory for the Democratic Party (the descendant of the Communist Party, the only one unscathed by the scandals). At the same time, the dissolution of the DC and PSI had left a vacuum on the centre-right of the political arena. Berlusconi decided to turn this situation to his advantage and founded Forza Italia in December 1993, just three months before the elections. The new party brought to the extreme consequences the personalization and spectacularization of politics initiated with Craxi and the PSI, leading Seisselberg (1996) to define Forza Italia a media-mediated personality-party. The announcement that Berlusconi was taking the field was filmed at his home and aired simultaneously on the three Mediaset channels. The party coordinators and many of the parliamentary candidates were drawn directly from the ranks of the company and among the presenters of many popular prime-time shows, after receiving an intensive training to grasp the basic notions of politics. The training was administered by the branch of Mediaset specialized in commercial advertising, Publitalia, which also directed the selection of candidates (?). Such an aggressive communication strategy proved indeed extremely successful and turned the polls around within just a couple of months. At the elections in March 1994 Forza Italia gained the relative majority of votes (20%, about the same result as the Democratic Party and twice as much the votes of the centrist Popular Party), the coalition with the autonomist Northern Liga and the post-fascist Alleanza Nazionale obtained the absolute majority in the parliament. Berlusconi was appointed Prime Minister on May 10, The first Berlusconi government was short-lived, as very soon the Northern Liga left the centre-right coalition and the country transited to yet another technical government on January Nevertheless, the electoral blitzkrieg of 1994 had long-lasting consequences on the Italian political system. Almost two decades later, Berlusconi still leads the centre-right coalition through the same type of spectacular campaigns, managing to win the elections

9 again in 2001 and Also, many of his opponents have come to imitate his same approach to politics, although with much less success. The enduring primacy of Berlusconi in the media-driven politics of the Italian Second Republic depends in fact related to his long-time ownership of the main private TV network of the country. 3. DATA AND VARIABLE CONSTRUCTION We aim to estimate the long-term impact of the introduction of commercial TV in Italy on political behavior and electoral competition by looking at differences across Italian municipalities. To do so, we need detailed information on access to Mediaset in the early stages of the network s diffusion (when geographic differences in coverage were still large), and on electoral outcomes, both before and after the advent of Mediaset. Information on the distribution of Mediaset viewers in the early 1980s is unfortunately not available, and in any case, actual viewership would measure an equilibrium outcome - possibly correlated with a range of socio-economic confounds - rather than an exogenous source of variation. Instead, we construct a measure of Mediaset channels signal intensity based on the location and technical characteristics of the network s transmitters, and focus on the variation attributable to differences in geomorphological factors across municipalities within small areas. With regard to political outcomes, we use municipal-level data on turnout and voting outcomes for all the national parliamentary elections held between 1976 and In what follows we describe the data and explain how the variables used in the empirical analysis are constructed, with particular regard for the measure of signal intensity. We then illustrate our empirical strategy DATA: SIGNAL INTENSITY Broadcast television signal is transmitted over the air according to the laws of physics for electromagnetic propagation. In the free space, signal strength decreases in the square of the distance from the transmitter; in practice, however, the patterns of decay are much more complex due to diffraction caused by mountains and other obstacles. We compute the intensity of Mediaset s signal in early 1985 using a professional engineer-developed software that simulates signal propagation according to the Longley-Rice Irregular Terrain Model (ITM) algorithm. The ITM was originally developed by the US government for frequency-planning purposes and allows to formulate extremely accurate predictions of signal strength for narrow geographical cells (Phillips, Sicker, and Grunwald, 2011). The version used in this paper is described in Hufford (2002) and has been previously used by Olken (2009),Yanagizawa- Drott (2010), Enikolopov, Petrova, and Zhuravskaya (2011) and DellaVigna, Enikolopov,

10 Mironova, Petrova, and Zhuravskaya (2012). To implement the ITM algorithm we combine information on transmitters location and power with a high-resolution geo-orographic map of Italy. Detailed data on the location and technical characteristics of the 1,700 Mediaset transmitters operating in 1985 was obtained directly from the Mediaset group. For each transmitter the documentation includes one technical report sheet indicating the latitude, longitude and altitude of the transmitter s location, the height of the transmitter as well as its transmitting power and frequency (a sample technical report sheet is reported in Figure?? of the Appendix). Using the ITM algorithm we compute Mediaset signal intensity in decibels (db) at the centroid of each one of the 8,100 Italian municipalities (comune). Municipalities are Italy s lowest administrative level and are fairly small both in terms of surface (mean of 37.2 km 2, median of 21.8 km 2 ) and population (mean of 7,010 people and median of 2,296 individuals). 2 Descriptive statistics for our complete sample are reported in Table 1. Figure 2 reports the distribution of signal intensity across Italian municipalities. While signal intensity ranges from -114 db to +40 db, most municipalities display values around zero, that is, the value above which signal reception is perfect. As signal strength turns negative, reception deteriorates considerably; however, in the absence of data on the number of Mediaset viewers in 1985, the precise relationship between signal and reception can only be inferred from previous studies. Using survey data on viewership of 11 TV channels in Indonesia, for example, Olken (2009) finds that for values of signal intensity below -50 the share of individuals able to watch a given channel is close to zero; this share increases as signal gets larger and reaches 100% when signal becomes positive. Bursztyn and Cantoni (2012) perform a similar exercise and find that the share of individuals able to receive Western TV in various areas of East Germany increased from about 0 to 80% when signal intensity passed from db to db, thus pointing at better reception at relatively lower levels of signal intensity than Olken (2009) s findings. Finally, Enikolopov, Petrova, and Zhuravskaya (2011) estimate that a unit increase in signal strength of the independent Russian network NTV is associated with an average increase in the share of viewers of 0.3 percentage points; however, the authors do not distinguish between areas with positive and negative signal intensity. Taken together, this evidence confirms that in areas with positive signal intensity all people are potentially exposed, while reception is poorer (and possibly nil) in areas with negative predicted signal. Although the relationship between signal and viewership is not stable across countries and/or periods, it is reasonable to expect that most of the variation in exposure should occur at intermediate values of signal intensity, while even large differences in signal strength at both extremes of the distribution should have little or no effect on the quality of reception. To focus on the relevant source of variation, in 2 Figures were very similar in 1981, around the time commercial television was introduced in Italy.

11 most of our analysis we will exclude municipalities in the top and bottom 2.5% of the signal distribution (see Figure 2) EMPIRICAL STRATEGY In the absence of data on actual Mediaset viewers in 1985, we focus on the direct effect of Mediaset signal intensity. Our approach is the same used by Yanagizawa-Drott (2010) and differs from the one used by Olken (2009) and Enikolopov, Petrova, and Zhuravskaya (2011), who have information on the number of viewers and use signal intensity as an instrument for viewership rates in a two-stage-least-square (2SLS) framework. 4. Hence, we estimate the reduced form coefficient of exposure to Mediaset or, equivalently, the intention-to-treat effect. Our identification strategy exploits variation in signal intensity between otherwise similar neighboring municipalities. To account for the potentially endogenous location of the transmitters, we simulate what would be the hypothetical signal intensity given the true distribution of the transmitters but in the absence of any geomorphological obstacles, i.e. assuming terrain is flat. Comparing the the latter with the actual signal intensity, within relatively small areas, allows us to disentangle what portion of signal is explained by the potentially endogenous presence of transmitters, and what part by the presence of geographical factors that thwart the transmission. The geographic distribution of the two measures is shown in Figure 3. Differences between the two maps are primarily due to the presence of major mountain ranges both in the Northern (Alps) and in the Center-Southern part of the country (Apennines). Our empirical analysis combines a standard OLS estimation with a neighbor pair-matching technique. In what follows we discuss the results obtained with both methodologies OLS ESTIMATION Following Olken (2009), in our baseline OLS specifications we regress our outcomes of interest on signal intensity (Signal) controlling for signal intensity under the flat terrain hypothesis (SignalFree). The underlying idea is that, keeping SignalFree constant, the coefficient of Signal is identified by the residual variation due to idiosyncratic differences in topography, rather than by the (potentially endogenous) location and power of the transmitters. 3 All results remain qualitatively unchanged when including all observations 4 Bursztyn and Cantoni (2012) employ yet another approach, dividing German municipalities into treatment and control group, based on signal intensity being above or below the one available in a particular location (Dresden) presumably close to the signal intensity discontinuity

12 Of course, terrain characteristics could potentially affect the socio-economic environment in other ways; for example terrain ruggedness could affect population density and the type and intensity of economic activity. To address this concern, in our main specification we control for a range of additional geographic variables, including the municipality s area and its square, average altitude and its square, and average terrain ruggedness. We also include two sets of fixed effects: for electoral districts (EDs) and local labor markets (LLM). Single-member electoral districts represent the level at which direct elections for the majoritarian portion of the mixed electoral system in place between 1994 and 2001 used to be held. Overall, there are 475 electoral district which generally include multiple adjacent municipalities within a given province. LLMs, instead, represent areas with homogeneous economic characteristics defined by the Italian statistical institute (ISTAT) on the basis of workers commuting patterns. Overall there are 686 LLMs which, unlike EDs, can include adjacent municipalities belonging to different provinces or regions. 5 Hence, the effect of early exposure to Mediaset is identified from residual differences between municipalities with analogous political and economic conditions; a test that is especially demanding in light of the fact that EDs and LLMs tend to be rather small, much smaller than regions (20) or provinces (110). The following estimating equation summarizes our empirical strategy: y m = βsignal m + γsignalfree m + δ T m + D i(m) + L j(m) + ε m (1) where y m represents the outcome of interest in municipality m, Signal m and SignalFree m represent respectively Mediaset s actual signal intensity in 1985, and the hypothetical signal intensity assuming flat terrain, and T m is a vector of municipal characteristics including area and its square, altitude and its square, and average terrain ruggedness index, D i(m) and L j(m) represent respectively the electoral district and local labor market fixed effects, and ε m is an error term. In all regressions standard errors are clustered by electoral district, although all results are analogous when clustering by LLM or by electoral district-llm cells. Under the assumption that Signal is independent of ε m, the OLS coefficient β consistently estimates the causal effect of Signal on y. While such assumption is fundamentally untestable, in Table 2 we show the correlation between Signal and a battery of municipal socio-economic controls, measured at the beginning of the 1980s or, if not available for that time, at the earliest available date. To facilitate the interpretation of the results, we normalized Signal dividing it by its standard deviation. Although the univariate correlation coefficients (in column 5 Figure?? in the Appendix provides an example of the distribution of EDs and LLMs in the mid-sized region of Abruzzo.

13 1) generally differ from zero, most of this correlation is absorbed by other variables on the right-hand side of equation (see column 2). Comparing the R 2 coefficients in columns (1) and (2), the joint variation in SignalFree, topography, and fixed effects explains between 60 and 90 percent of the overall variation for most variables. Once these additional variables are controlled for, Signal is no longer correlated with population (levels, density and growth), labor market conditions, and the number of firms per capita (also by class size). Signal continues to be correlated, instead, with both educational attainment and income per capita; in light of this, we will include both these variables as additional controls in our specification MATCHING NEIGHBOR MUNICIPALITIES We complement the OLS analysis with an alternative strategy which allows us to relax the functional form assumptions implicit in the regression-based approach. In the spirit of the pair-matching methodology implemented by Acemoglu, García-Jimeno, and Robinson (2012), we restrict our analysis to pairs of adjacent municipalities such that one has positive signal intensity (i.e. perfect reception) and the other one has negative signal intensity (i.e. imperfect or no reception). Moreover, to compare municipalities that are more and more similar in other relevant dimensions, we can further restrict the sample to pairs of neighbors with increasingly similar values of SignalFree. By doing so, we approximate the ideal experiment of comparing two identical municipalities that are equally distant from a transmitter so that any difference in reception is solely due to idiosyncratic geomorphological factors. Overall we identify 3,021 such eighbor-pairs. In column (3) of Table 2 we report the simple mean difference in socio-economic characteristics between the two groups. While the picture looks fairly balanced with respect to most variables, there are some significant differences with regard to size, educational attainment and number of firms. However, these differences disappear when restricting the comparison to 811 pairs of neighbor municipalities with difference in SignalFree smaller than 1 db. We also implement a regression version of this approach by estimating the following equation: y mp = β 1(SIGNAL m > 0) + µ X m + F p + ν mp (2) where y mp is the outcome in municipality m of neighbor-pair p, 1( ) is an indicator variable for Signal m greater than 0, X m is a vector of municipal controls, F p is a pair-specific fixed effect, and ν mp is the residual error term. By including neighbor-pair fixed effects, in particular, we can control for any observable and unobservable characteristics common to each

14 two adjacent municipalities. Therefore, to estimate the coefficient of interest β we exploit variation between municipalities with perfect and imperfect (or nil) reception within each pair, accounting for a range of other observable municipal characteristics and controlling for any other factor that is common to each two neighbors. We cluster standard errors by municipality to allow for non-zero correlation in the error term for all observations including a given municipality. Overall, the data provide support for our main identification assumption that - after controlling for a few additional variables (in the OLS approach), or when comparing neighbors with minor differences in SignalFree (in the matching approach) - Signal is uncorrelated with the error term in equations (1) and (2) (see Table 2). This conclusion is strengthened by placebo estimates of the effect of Mediaset signal intensity in 1985 on electoral outcomes in the years before the introduction of commercial TV, discussed in the next section. 4. RESULTS In what follows we estimate the effect of early exposure to Mediaset on electoral support for Berlusconi s party, Forza Italia, and for the other main Italian parties. Municipal level data on electoral turnout and vote share for each party in all national elections held between 1976 and 2006 were obtained from the Ministry of Interior. We first focus on the evolution of Forza Italia s vote share. As mentioned, Berlusconi s party was founded at the beginning of 1994 and run in all national elections held since then, hence in 1994, 1996, 2001, and Table 1 reports the average and median vote share of Forza Italia across Italian municipalities in each of these elections. We focus on vote share in the elections for the lower house (Camera) rather than for the upper house (Senate) since electoral rules for the latter encourage parties to form coalitions making it impossible to assess the relative electoral strength of each individual party. Furthermore, since these coalitions are formed at the regional level, a party may be allied with different partners in different regions, 7 complicating the interpretation of the results even further. 6 We exclude the 2008 national elections because, shortly before those elections, FI merged with its historical ally, National Alliance (AN), into the People of Freedom party (PDL). Since FI and AN traditionally catered to rather different segments of the electorate, and since both parties accounted for a non-negligible fraction of combined votes, it would be rather inaccurate to consider the PDL s electoral consensus in 2008 as a mere extension of FI s electoral support in previous elections. 7 This was indeed the case of Berlusconi s party in the 1994 elections in which FI was allied to the separatist Northern League party in the North, and to the nationalist National Alliance party in the Center and in the South.

15 4.1. EXPOSURE TO MEDIASET AND VOTING FOR FORZA ITALIA In table 3 we report the OLS estimates of the effect of pre-1985 exposure to Mediaset on voting for FI in 1994, the first election in which Berlusconi ran for office. The univariate regression in column 1, documents a positive relationship between actual intensity of Mediaset signal and vote share for FI; the effect is statistically significant at the 1% level and rather large, a one standard deviation increase in Signal corresponding to a 3-percentage point increase in FI vote share. In column 2 we control for signal intensity under the flat terrain hypothesis (SignalFree) and in column 3 we add the geomorphological controls; the fact that the coefficient of Signal remains unaffected is reassuring of the fact that endogeneity in the location and power of Mediaset transmitters is not driving the result. This is confirmed by the fact that the univariate regression of FI vote share on SignalFree, i.e. the component of signal strength that reflects just the location and power of the transmitters, the estimated coefficient is not significantly different than zero (coefficient , standard error 0.483). In the following two columns we add electoral district and local labor market fixed effects, respectively. The point estimate on Signal decreases to slightly less than 1 percentage point; the coefficient remains virtually unaffected in column 6 when we control for the number of eligible voters, income per capita (log), and education (the only variables that were statistically significant in the balance test of Table 2). In Table 4, we estimate the latter specification for voting for FI in all subsequent elections (1996, 2001, and 2006). Interestingly, the coefficient on Signal remains very stable over time. In the last column we pool observations for the four elections together and include year fixed-effects; when doing so we find that on average, over the period , one standard deviation increase in signal strength is associated to an increase in electoral support for FI of 0.75 percentage points. As discussed above, the estimated coefficients on Signal in Tables 3 and 4 capture the reduced form (or the intention-to-treat) effect of early exposure to Mediaset on voting, while, ideally, we would like to estimate the effect of signal-driven changes in actual viewership. Although, as mentioned, municipal data on the latter are not available for the early 1980s, we know from Olken (2009) that potential viewership is increasing in signal strength, when this is negative, and reaches 100% when signal equals zero. To test whether the documented effect of signal strength on voting is consistent with such a pattern, we estimate a local linear regression (LLR) of FI vote share on Signal; the results, summarized graphically in the left graph in Figure 4, confirm that, in fact, FI vote share increases with signal strength for values of signal strength below zero while it remains rather stable for values of signal strength above zero. Interestingly, if we estimate the LLR separately for negative and positive values of Signal (right graph of Figure 4), we observe a discontinuous increase of about 2 percentage points at Signal = 0, the signal intensity corresponding to perfect reception. This probably represents

16 an underestimate of the true effect of having perfect reception since the zero threshold is only an approximation of the signal intensity that allows perfect municipalities with signal close to zero acquire perfect reception. For this reason, we do not pursue the regression discontinuity design any further. Instead, we implement the matching strategy described in Section 3.2.2, restricting our analysis to pairs of neighboring municipalities with signal strength respectively below and above zero (table 5). We start by considering all 3,021 pairs of neighbors that satisfy this condition; the first column reports a series of mean comparisons tests between municipality with signal intensity below and above zero. The results suggest that municipalities with perfect reception of Mediaset channels display significantly higher vote share for FI than those with imperfect or no reception in all four elections; furthermore we find that neighbors with signal intensity above and below zero are not systematically different from each other in terms of any key geomorphological factors (i.e. area, altitude, or ruggedness). In the second and third columns we replicate the analysis focusing on pairs of neighboring municipalities with similar levels of SignalFree, i.e. difference smaller than 1 db and 0.5 db respectively. Interestingly, the difference in vote share of FI between municipalities with Signal bigger and smaller than 0 is larger when comparing municipalities that are more similar to each other in terms of SignalFree (0.8 and 0.95 percentage points respectively as opposed to 0.65 in the baseline comparison). In the last three columns we report the results of the regressions with municipal controls and neighbor-pair fixed effects which largely confirm those based on simple mean comparisons. Given the similarity of the results of OLS and matching analysis, in the reminder of the paper we will confine our discussion to the OLS results which refer to the entire sample of municipalities. One potential issue with the OLS estimates concerns the weighting scheme. While this is essential to guarantee the representativeness of the estimated coefficient at the national level, one may worry that the overall effect might be driven by the relationship between exposure to Mediaset and voting in few large cities. To address this concern, in Table 6 we re-estimate our baseline specification first using the unweighted observations (column 1), and then restricting the sample to less populated municipalities (columns 2 to 5). 8 The coefficient on Signal in the unweighted regression is slightly higher than the weighted one; most importantly, the point estimate is very similar for municipalities with different population size, and increases strongly when focusing on the very small ones (less than 500 inhabitants). These results further confirm that the effect of signal strength is capturing variation across a multitude of 8 All regressions in Table 6 use the specification in the last column of Table 4, pooling together all elections in the period and including year fixed-effects. We obtain analagous results when running separate regressions for each election year.

17 small cities with scattered access to Mediaset and is not driven by a handful of large cities. We are also interested in examining whether early exposure to Mediaset has a similar impact on voting for FI in different areas of the country. To this end, in columns 6 to 8 we estimate our baseline specification separately for municipalities in the North, Center and South. We find evidence of a considerable heterogeneity across macro-regions, with no significant effect in the North, a very large and significant effect in the South, and an intermediate effect in the Center, very similar in magnitude to the national average. It is instructive to relate the (dis)similarities in the effect of signal strength across municipalities of different size and across geographical areas to (dis)similarities in TV consumption habits prior to the advent of Mediaset. Indeed, for a given level of signal strength, one would expect a larger effect on voting where people watched more TV, since, in these areas, individuals were presumably more likely to be exposed to all channels, including Mediaset s when available. To explore this aspect, in the bottom part of table 6 we report information on the average number of hours spent watching TV in municipalities of different size and in different macro-regions based on responses to a survey on media consumption conducted on a representative sample of the Italian population by ISTAT in 1983 (ISTAT, 1985). In line with our findings, there are not large differences in average TV viewership between municipalities of different size (no information is unfortunately available for municipalities with less than 500 inhabitants); there are, however, noticeable geographical differences in TV consumption, with respondents in the North reporting to watch fewer hours of TV than those in the Center and, especially, in the South. A graphical representation of the relation between effect of early exposure to Mediaset and pre-existing TV consumption patterns is provided in Figure (5). Taken together, these results provide strong support for the existence of a positive relationship between early expsosure to Mediaset and propensity to vote for Berlusconi s party. This effect appears to be quite sizeable, very persistent over time, robust to different specifications and empirical tests, and, reassuringly, also consistent with pre-existing differential patterns in TV consumption VOTING FOR OTHER PARTIES In this section we investigate the impact of early exposure to Mediaset on voting for parties other than Forza Italia. This analysis potentially serves two purposes. On the one hand, examining whether access to Mediaset in the early 1980s was related to voting patterns in elections held before the introduction of Mediaset (i.e. in the 1970s) provides us with a precious opportunity to indirectly test our key identification assumption, namely that (con-

18 ditional on observables and fixed effects) Mediaset signal strength is uncorrelated with other determinants of voting behavior that may enter the error term in equations (1) and (2). On the other hand, assessing the impact of early exposure to Mediaset in elections held after the network was introduced - and both before and after the advent of Berlusconi s party - can elucidate what parties attracted the consensus of Mediaset early viewers prior to the emergence of FI, and, in and after 1994, what parties were penalized by FI s electoral success. Shedding light on these aspects can inform our understanding of the channel(s) through which the effect documented in the previous section may be operating, an issue that we discuss more in length in the next section. As discussed in the background section, over the past decades the Italian political landscape has undergone major changes and has been characterized by a high degree of instability and fragmentation, with the frequent formation and dissolution of parties and coalitions. The most important transformation occurred between 1992 and 1994, when, following a series of corruption scandals, most of the political parties that had traditionally dominated Italian politics collapsed, and new political actors, including Berlusconi s party, emerged. For the purpose of our analysis we divide the years between 1976 and 2006 into two periods: 1976 to 1992, the so-called first republic, and 1994 to 2006, the second republic. We are able to reconstruct the evolution of electoral support for eight parties, for the first period, and eight parties (FI and seven others), for the second one. 9 For each party in each election, we estimate our main specification regressing vote share on Signal, SignalFree, the vector of geographic and socio-economic controls, electoral districts and local labor market fixed effects. The results - limited to the coefficients on Signal, our regressor of interest - are shown in table 7 which also reports the coefficients for similar regressions having, as dependent variable, voting turnout, vote share for minor parties, and share of spoiled ballots over total ballots cast. In the first two rows of the top panel, we look at the effect of Mediaset signal strength in 1985 on party vote share in the 1976 and 1979 elections, hence in the pre- Mediaset era. We find no evidence of a systematic correlation between Signal and the vote share of any party in both years which is further reassuring of the fact that Mediaset early coverage is not related to pre-existing electoral patterns. We then turn to the period between 1983 and 1992, that is, after the introduction of Mediaset and before the advent of the second republic (rows 3 to 5 of the top panel); here we are exploring the impact of differential early exposure to Mediaset on the electoral performance of traditional political parties both in the ruling five-party majority (DC, PSI, PSDI, PRI, PLI) and in the opposition (PCI, MSI, PR). The only notable change with respect to the previous period is represented by a significant increase in electoral support for the Italian Socialist Party in municipalities with better re- 9 A brief description of each party is reported in the Appendix.

19 ception of Mediaset channels early on: the coefficient on Signal increases from 0.36 in 1979 (non-statistically significant) to 0.64 in 1992 (significant at the 5% level). This finding is, in many aspects, not surprising. As mentioned above, Berlusconi was linked to the PSI s leader Bettino Craxi by a close personal relation. More important, as discussed by Ginsborg (2004), the PSI s political message was more compatible with some of the message of renovation and nonconformist values put forth by commercial TV than traditional mass parties, namely the Christian Democrats and the Communists, whose rhetoric was rather centered around working class instances and family values respectively. Turning to the post-1992 period, after the collapse of the PSI (alongside the other majority parties), Berlusconi s newly formed party is the only one to benefit from significantly higher electoral support in areas that were exposed to Mediaset prior to 1985 in the 1994 elections (first column bottom panel). Furthermore, as discussed above, this effect tends to persist with similar magnitude in the three following elections. Interestingly the same pattern does not apply to other parties on the center-to-the-right side of the political spectrum: while we find no significant impact on electoral support for National Alliance, Forza Italia s right-wing ally, we find evidence of a negative and significant effect on the vote share of the regionalist and populist Northern League, Berlusconi s traditional electoral partner in the North; this effect, which holds in all elections, is especially pronounced in 1996 (-0.795, significant at the 1% level), the only election in which the League, after causing the fall of Berlusconi s first government, did not run as part of Berlusconi s coalition. To further explore what electoral basins Berlusconi s party support was coming from, it is useful to look separately at the results in the North and in the Center and South part of the country where, in virtually all electoral districts, the League was absent from the ballot. The two panels of table 8 report the results for the North (top) and for the Center-South (bottom). In addition to the negative effect on the League, we find that in the North municipalities with pre-1985 exposure to Mediaset displayed significantly higher electoral turnout, which suggests that part of Forza Italia electoral support might have come from individuals that were previously abstaining. With regard to the Center and Southern regions, we find the positive effect of signal intensity on vote share for Forza Italia to be much larger and more significant than in the North for all four elections; furthermore, this result is associated with significantly lower electoral support for parties left-to-the-center of the political spectrum, namely the Democratic Party of the Left and the Communist Refoundation party (always significant at the 5% level or more). This finding is in line with the similarly negative effect documented for the Northern League which also caters primarily to working class voters (Tambini, 2001). Taken together these results point towards the existence of a strong and rather specific link between early exposure to Mediaset content and receptiveness to Berlusconi s political message that operated since

20 the earliest stage of his political experience and that, to a large extent, transcended from traditional ideological, class, and party lines. Figure 6 summarizes the results discussed above. The graph reports, on the horizontal axis, the coefficient on Signal estimated using our baseline specification in individual regressions of vote share for all parties in all elections, and, on the vertical axis, the corresponding t- statistics; we distinguish the results for the pre-1983 elections (marked with Xs), for the period (hollow circles), and for the post-1992 elections (solid circles). While no party displays a significant coefficient prior to the introduction of Mediaset, and only in three cases (PSI) a significant coefficient between 1983 and 1992, Forza Italia s coefficients for all four post-1992 elections (black solid circles) are abnormally large in terms of both magnitude and statistical significance. Taken together these results point towards the existence of a strong and rather specific link between early exposure to Mediaset content and receptiveness to Berlusconi s political message that operated since the earliest stage of his political experience and that, to a large extent, transcended from traditional ideological, class, and party lines DIFFERENCE-IN-DIFFERENCE ESTIMATES To shed further light on the long-term political impact of exposure to Mediaset, we compare changes in voting behavior before and after the advent of Mediaset between municipalities that had access to Mediaset prior to 1985 and those that did not. To do so, we pool together data from all elections between 1976 and 2006, and estimate a difference-indifference specification interacting Signal with dummy variables for different sub-periods. Exploiting longitudinal variation across municipalities that were differentially exposed to Mediaset, allows us to control for municipality and year fixed-effects and thus to account for the effect of any time-invariant municipal characteristics and any common time-varying factor. Unfortunately, given the transformations occurred in the Italian party system over the past decades, we are not in the position to examine the evolution of electoral support for the same parties over the entire thirty-year period. In the first columns of table 9 we start by looking at the evolution of vote share for first-republic parties between 1976 and The results document a significant decline in the electoral support for the two major parties - the Christian Democrats (majority) and the Communist Party (opposition) - in areas that were exposed to Mediaset earlier (-0.7 and -1 percentage points respectively); the negative effect on these parties was partly compensated by a significant increase in vote share of the PSI (+0.3 percentage points), and partly by increased support for a few minor (and often shortlived) protest parties and an increase in abstention (columns 5 and 6). Since data on the vote

21 share of minor parties and turnout are available for the whole period, in both specifications we include another interaction term for the elections in and after Interestingly, the transition to the second republic and the advent of Forza Italia is associated with a reversal of both these patterns with a significant increase in turnout and an equally significant decline in the vote share of minor parties in more exposed municipalities relative to the others. The results from the diff-in-diff estimations dovetail nicely with the other findings presented above and, taken together, suggest that the political impact of commercial TV started unraveling since the 1980s - well before the advent of Forza Italia - by favoring voters disenchantment with mainstream parties and demobilization. This tendency was similarly detrimental to both the main majority and the main opposition parties (DC and PCI), and, despite the improved performance of the PSI, did not result in substantial changes in the balance of power between left and right. The situation changed, however, with the collapse of traditional parties and the emergence in 1994 of Forza Italia which benefitted disproportionally from the electoral support of early Mediaset viewers at the expense of competing popular parties both on the right and the left side of the political spectrum. In the following section we will attempt to shed light on the possible channel(s) through which the documented political effect of commercial television may have operated. 5. CHANNELS The evidence presented so far points at significant effects of early exposure to Mediaset on political behavior. In particular, it appears that individuals in municipalities with access to Mediaset prior to 1985 were much more likely to vote for Berlusconi s party ten years later, when he first run for office, as well as in subsequent elections. Previous research has documented that exposure to media can have important effect on individual voting behavior and, ultimately, on electoral competition (DellaVigna and Kaplan, 2007; Enikolopov, Petrova, and Zhuravskaya, 2011). These contributions have focused exclusively on the impact of exposure to partisan bias in political news coverage. However, as discussed in section XXXXXXX, no newscasts were broadcast on Mediaset until 1991, when news services on the three channels were established. Other informative programs, e.g. political talk shows or investigative journalism reports, were similarly absent from Mediaset channels whose offer consisted primarily of light entertainment programs (e.g. soap operas, quiz shows, etc.). Figure?? compares the share of airtime devoted to different types of programs respectively on Mediaset and RAI channels, as reported by the statistics on culture of the Italian statistical institute (ISTAT, 1987). As of the 1987, the first year for which such data are available, the share of airtime devoted to all informative programs combined was

22 close to zero, and given the trend, this share was presumably even lower in previous years. Hence, based on the timing of the introduction of news programs on Mediaset channels, it seems unlikely that the documented effect of early access to Mediaset can be attributed to (differential) exposure to partisan news bias before Yet, it is possible that early Mediaset viewers developed some form of attachment to the network that made them more likely to watch any Mediaset program, including newscasts once these were introduced. In what follows we investigate this possibility using individuallevel survey data on political behavior and media consumption from various waves of the Italian National Election Studies (ITANES) PARTISAN NEWS BIAS The Italian National Election Study (ITANES) is a series of pre- and post-electoral surveys conducted on a representative sample of the Italian population in coincidence with each national election. The survey has been conducted since 1972; we focus on the four waves carried out since In each round, ITANES interviews between 2000 to 3000 individuals. The survey reports detailed information on individual characteristics of the respondents, such as age, gender, education, employment and familiar status. Crucially, the survey also reports the code of the municipality where the respondent is resident. Individuals are asked their political vote, opinion on political leaders, degree of political participations, where they form political opinion,given the relative stability of the effect of Signal on post-1994 electoral outcomes documented in the municipal regressions, to increase the number of observations, in the individual analysis we pool data from four separate ITANES waves conducted for the 1994, 1996, 2001, and 2006 national elections. In table 10 we examine the relation between pre-1985 access to Mediaset and several individual outcome. We start by replicating our main result with individual data. In column 1 we first regress a dummy variable for voting for Forza Italia on Signal, SignalFree, all municipal controls included in our baseline municipal regressions (area, area squared, altitude, altitude squared, ruggedness, income, education), a range of individual controls (gender, age, education, employment, marital status, and household size), and election year fixed effects. The results confirm that individuals living in areas that had access to Mediaset prior to 1985 are more likely to vote for Berlusconi s party (point estimate of significant at the 1% level). Since many electoral districts and local labor markets only have a few respondents, in the first regression we do not include the full set of fixed effects used in our municipal regressions. Instead, in column 2 we include fixed effects for 110 provinces (the administrative level just above municipalities) and find that the effect of Signal on the propensity to vote for Forza Italia remains positive and signif-

23 icant (though only at the 10% level). According to the estimated coefficient, a one standard deviation increase in pre-1985 signal strength is associated with an increase in the probability of voting for Forza Italia of 2 percentage points, a magnitude that is comparable to that estimated in the municipal regressions. We then examine to what extent early access to Mediaset is associated with viewership of Mediaset news programs in and after To do so, we construct two dummy variables that equal one for respondents reporting to watch news most frequently on Mediaset channels and public channels respectively, and use them as dependent variables in our baseline specification (columns 3 and 4). Interestingly, we find no significant effect of early exposure to Mediaset on the probability of watching news on either Mediaset or RAI. As an indirect test of the news bias channel, we finally look at differences in beliefs about Berlusconi between early and late Mediaset viewers. Indeed, if early viewers were exposed to more favorable coverage of Forza Italia and its leader, one would expect to observe systematic differences in the evaluation of Berlusconi between the two groups. Luckily some questions in the ITANES survey can be used for this purpose. In particular in some waves, respondents were asked whether they believe Berlusconi to be honest (in 2001 and 2006), qualified (2001 and 2006), and sincere (only in 2001). In the last three columns we estimate our baseline specification using as dependent variable a dummy for positive response on each attribute. Again, we find no significant difference between early and late Mediaset viewers in their evaluation of Berlusconi along any of these dimensions. Overall, we find little evidence that voters that had access to Mediaset earlier on were more exposed (or vulnerable) to pro-berlusconi news bias later on, nor that they formed a more positive opinion of Berlusconi as a person and political leader. An alternative interpretation is that early Mediaset viewers are more likely to vote for Berlusconi not because they receive different information about him (or his political platform), but, rather, because they evaluate this information through a different set of values, whose diffusion was favored by exposure to commercial TV. In the following section we discuss this alternative explanation in greater detail It is worth noting that we are by no mean trying to dismiss the potential impact of partisan news bias on voting in the context of Italy (discussed in Durante and Knight, forthcoming, and, more recently, by Barone, D Acunto, and Narciso, 2012). We are simply arguing that, given the specific typology of content available on Mediaset channels in the early stage of the network diffusion, an explananation based solely on exposure to partisan news bias appears insufficient to explain the documented differences in voting behavior between individuals exposed to Mediaset prior to 1985 and others.

24 5.2. NON-INFORMATIVE CONTENT AND VALUES In his seminal work on the decline of social capital in the United States in the post-war period, Putnam (2000) dedicates an entire chapter to discussing the role of television and its negative impact on social capital. According to Putnam, both the quantity of TV watched and the quality of content to which people are exposed matter in this sense. Regarding quantity, Putnam intuitive argument is that the more time individuals spend watching TV, the less they will devote to various forms of civic engagement (e.g. participation in association, attendance of public meeting, etc.). Regarding quality, however, Putnam draws a marked distinction between informative and non-informative content: if, on the one hand, by facilitating access to news and educational programs television can foster political participation and civic engagement, on the other hand, in Putnam s view, light entertainment programs are generally detrimental to social capital. Putnam even classifies specific program categories in terms of their pro-civic potential: at the top of the pro-civic hierarchy are news programs and educational television (...) at the other hand of the scale fell action dramas (exemplified in an earlier era by Hazzard and Miami Vice), soap-operas (such as Dallas and Melrose Place) and so-called reality TV. Interestingly enough for our analysis, all the least pro-civic programs mentioned by Putnam were broadcast in Italy by Mediaset. More in general, as shown in figure??, which reports the distribution of different program categories on both Mediaset and RAI channels, action dramas, soap operas and entertainment accounted for more than 70% of Mediaset total airtime in the 1980s. Following Putnam s argument, the advent of Mediaset might have favored a decline in civic engagement both by expanding the quantity and variety of content available to viewers, and by favoring the diffusion of the least pro-civic programs. This pattern, one would expect, would have been more pronounced in areas that had access to Mediaset programs earlier rather than later. To test this prediction, in what follows we examine the relation between Mediaset signal intensity prior to 1985 and the evolution of civic engagement over the period between , proxied by the number civic associations per capita in a municipality. We choose this measure because it is the only one, of those originally used by Putnam, for which municipal level data dating back to the 1980s are available. The data come from the national census and are only available at ten-year intervals (1981, 1991, 2001). In Table 12 we examine the relation between the number of voluntary associations and signal intensity in 1985 by estimating our baseline specification. While in the top panel the dependent variable is the ratio between number of voluntary associations and population, in the bottom panel we divide the number of voluntary associations (which the census classify as economic organizations) by the total number of firms. Reassuringly, signal strength in 1985 does not appear to be correlated with the number of voluntarily associations in 1981, i.e. before the

25 expansion of Mediaset (column 1). Instead, the correlation between the two variables becomes negative and statistically significant after the introduction of Mediaset, in both 1991 and 2001 (columns 2 and 3); this pattern is suggestive of a more pronounced decline in civic engagement - both between 1981 and 1991 and 1991 and 2001, in municipalities that were exposed to Mediaset earlier than others. In column (4) we estimate such differential effect by pooling together observations for all census years and interacting signal strength in 1985 with a dummy for the observations after that year (1991 and 2001). The estimated difference-in-differences is negative and strongly statistically significant: a one standard deviation increase in signal strength is associated with a decline in civic engagement of 3.5 voluntarily associations per 10,000 inhabitants or, equivalently, 6.2 associations per 1,000 firms. These are sizable effects, corresponding respectively to 40 and 36 percent of the standard deviation in the number of voluntarily associations per capita and per firm in The estimated effects of interest remain remarkably similar when we include municipality fixed effects, thus dropping Signal and exploiting only differential changes over time across municipalities (column 5). Overall, the results in Table 12 suggest that municipalities that were exposed to Mediaset before 1985 experienced a decline in civic engagement over the following years, relative to municipalities that were exposed later on. Although we are not in the position to distinguish what part of the documented effect may be due to an increase in TV consumption rather than to changes in the typology of content, either of these two mechanisms would be consistent with Putnam s argument VALUES AND VOTING Why the diffusion of a culture of individualism and civic disengagement, favored by the advent of commercial TV, would result in increased electoral support for Berlusconi s party? Considering the specificity of Berlusconi s political rethoric and his contentious views on issues such as corruption, tax evasion, or gender equality, it is not unreasonable to believe that individuals subscribing to pro-civic values would find his political message ununattractive, and viceversa. Indeed, extensive anecdotal evidence attest to substantially different views between Berlusconi s political supporters and opponents on a range of culturally charged issues which often transcend, and in some cases go against, the traditional right-left ideological divide. The ITANES data allow us to explore this issue more systematically; the survey includes a range of questions that allows for the construction of individual measures of civic engagement analogous to those used by Putnam (2000). In particular, respondents are asked whether

26 they belong to any voluntary association, whether they are interested in politics, sign petitions, attend public meetings and demonstrations. Based on the response to each question we construct a set of dummy variables and use them to characterize Forza Italia s voters along these dimensions of civic engagement. Specifically, in table 11, we regress a dummy for voting for Forza Italia on various measures of political and social engagement (both separately and jointly), controlling for baseline individual characteristics, and electoral year fixed effects; since in this case we are exploiting variation at the individual level, we also include municipality fixed effects. The results suggest that, relative to the average individual in the sample, Forza Italia voters are significantly less likely to participate in voluntary associations (column 1 and 7), and tend to be significantly less interested and actively engaged in politics (columns 2 to 6). Although these findings can hardly be given a causal interpretation, the evidence is consistent with the view that Forza Italia s is especially popular among segments of the population characterized by low levels of civic engagement, and that may have hence benefited from from the decline in social capital favored by the diffusion of Mediaset s non-informative content. To further corroborate these findings, we finally examine the relationship between civic engagement and voting for Forza Italia at the municipal level; to do so we pool data on electoral outcomes for the elections between 1994 and 2006 and combine them with census data on the number of civic associations between 1981, 1991, In the first row of table xxx we regress average vote share for each main party on the number of voluntary associations per capita in the closest census year 11 and our baseline set of regressors which includes: Signal, SignalFree, municipal geographic and socio-economic controls, electoral district and local labor market fixed effects. The results confirm that Berlusconi s party enjoys significantly lower electoral support in areas with relatively low levels of civic engagement; interestingly Forza Italia displays the largest coefficient among all parties, most of which, both on the right and on the left, tend to receive lower support in areas with low civic engagement. Furthermore, the relation between civic engagement and voting for Forza Italia seems quite general and not driven by exposure to Mediaset; to test this, in the second row we replicate our analysis restricting the sample to municipalities with pre-1985 signal intensity above 0, that is, municipalities that should not differ much in exposure to Mediaset. When doing so we find that the qualitative results remain very similar and, if anything, the pattern in row become even more extreme. To make sure that the results are not driven by differences in unobservable municipal characteristics, in the lower panel of table xxx, we replicate our analysis including municipal fixed effects, thus identifying the effect only from variation in voting and civic engagement over time in the same municipality. When using 11 We use the number of voluntary associations reported in the 1991 census for the elections in 1994 and 1996, and that reported in the 2001 census fr the elections of 2001 and 2006.

27 this approach the qualitative results remain unchanged while the magnitude of all coefficients increases. 6. CONCLUSION

28 REFERENCES ACEMOGLU, D., C. GARCÍA-JIMENO, AND J. ROBINSON (2012): Finding Eldorado: Slavery and Long-run Development in Colombia, Journal of Comparative Economics. BARONE, G., F. D ACUNTO, AND G. NARCISO (2012): Telecracy: Testing for Channels of Persuasion, Available at SSRN BENINI, S. (2012): Televised bodies: Berlusconi and the body of italian women, Journal of Italian Cinema and Media Studies, 1(1), BURSZTYN, L., AND D. CANTONI (2012): A Tear in the Iron Curtain: The Impact of Western Television on Consumption Behavior, CEPR Discussion Papers 9101, C.E.P.R. Discussion Papers. CHONG, A., S. DURYEA, AND E. LA FERRARA (2008): Soap Operas and Fertility: Evidence from Brazil, CEPR Discussion Papers, CHONG, A., AND E. LA FERRARA (2009): Television and divorce: Evidence from Brazilian novelas, Journal of the European Economic Association, 7(2-3), CONSTITUTIONAL COURT (1988): Sentenza n. 826/1988 del Luglio 1988, Discussion paper. DELLAVIGNA, S., R. ENIKOLOPOV, V. MIRONOVA, M. PETROVA, AND E. ZHU- RAVSKAYA (2012): Cross-border media and nationalism: Evidence from Serbian radio in Croatia?, Discussion paper, CEPR Discussion Papers. DELLAVIGNA, S., AND E. KAPLAN (2007): The Fox News Effect: Media Bias and Voting, The Quarterly Journal of Economics, 122(3), DURANTE, R., AND B. KNIGHT (forthcoming): Partisan control, media bias, and viewer responses: Evidence from Berlusconi s Italy, Journal of the European Economic Association. ENIKOLOPOV, R., M. PETROVA, AND E. ZHURAVSKAYA (2011): Media and political persuasion: Evidence from Russia, The American Economic Review, 101(7), HUFFORD, G. (2002): The ITS Irregular Terrain Model, version The Algorithm, Institute for Telecommunication Sciences, National Telecommunications and Information Administration, US Department of Commerce. its. bldrdoc. gov/itm. html. ISTAT (1985): Indagine sulle strutture ed i comportamenti familiari,.

29 (1987): Statistiche culturali,. JENSEN, R., AND E. OSTER (2009): The Power of TV: Cable Television and Women s Status in India, The Quarterly Journal of Economics, 124(3), MENDUNI, E. (2002): Televisione e società italiana, Bompiani. OLKEN, B. (2009): Do television and radio destroy social capital? Evidence from Indonesian villages, American Economic Journal: Applied Economics, 1(4), PHILLIPS, C., D. SICKER, AND D. GRUNWALD (2011): The Stability of The Longley-Rice Irregular Terrain Model for Typical Problems, arxiv preprint arxiv: PUTNAM, R. D. (2000): Bowling Alone. New York: Simon & Schuster. SEISSELBERG, J. (1996): Conditions of success and political problems of a mediamediated personality-party: The case of Forza Italia, West European Politics, 19(4), TAMBINI, D. (2001): Nationalism in Italian politics: the stories of the Northern League, , no. 3. Routledge. YANAGIZAWA-DROTT, D. (2010): Propaganda and conflict: Theory and evidence from the rwandan genocide, Discussion paper, Working Paper, Harvard University. ZANARDO, L. (2010): Il corpo delle donne, vol Feltrinelli.

30 FIGURES AND TABLES Figure 1: Development of Mediaset and Berlusconi s political career ( ) Jan-Mar 1994 time The figure provides a graphical representation of the timing of the development of Mediaset group, with particular regard to the expansion of its signal coverage, as well as some key political events in the Italian political landscape between 1980 and B1, B2 and B3 label the first, second and third Mediaset channel (corresponding respectively to Canale 5, Italia 1 and Rete 4

31 Figure 2: Distribution of Mediaset signal intensity in 1985 Density The figure reports the distribution of 1985 signal intensity across Italian municipalities. The dashed red lines indicate the top and bottom 2.5% of the distribution.

32 Figure 3: Geographic distribution of Mediaset signal intensity in 1985 (a) Signal (b) Signal free The maps represent the geographic distribution of the simulated intensity of Mediaset s signal in 1985 under real conditions (left), and in the absence of geo-morphological obstacles respectively (right).

33 Figure 4: Mediaset signal intensity in 1985 and voting for Forza Italia in 1994 (local linear regressions) % votes' share of Forza Italia in intensity of Mediaset TV signal in 1985 % votes' share of Forza Italia in intensity of Mediaset TV signal in 1985 The graph on the left summarizes the result of the estimation of a local linear regression of Forza Italia s vote share on Mediaset signal intensity in 1985 (controlling for all baseline controls). The graph on the right summarizes the results of a similar exercise; in this case however we estimate the local linear regression separately for negative and positive values of signal intensity. Figure 5: Mediaset signal intensity, heterogeneity in hours of TV watched and voting for Forza Italia The graphs above report the estimated coefficients (and respective confidence intervals) of a series of regressions of Forza Italia s vote share on Mediaset signal intensity in 1985 (controlling for all baseline controls) for different samples of municipalities divided by size of the population (<500K, <50k, <5k) and macro-geographic areas (North, Center, South). For each sample, the corresponding columns report the share of the population watching TV for less than 2 hours per day (dark grey), between 3 and 4 hours per day (light grey), and 5 or more hours per day (white).