The Effects of Children TV: Evidence from the Italian Transition to Digital TV (Preliminary - Please do not cite or circulate outside)

Luca Facchinello?

Abstract

When Italy transitioned from analog to Digital Terrestrial Television (DTT), several children-specific channels became freely available to the whole population, marking the start of public Children Television. In this paper I study how exposure to the new television regime affected students’ educational achievement, non-cognitive skills and behavior. To identify the reduced-form effects of exposure and differential exposure to Children TV, I exploit the staggered implementation of the DTT reform, carried out from 2008 to 2012, in a difference-in-difference analysis. My data includes administrative school information, standardized test scores and survey responses for the universe of Italian students attending compulsory school from 2010 to 2017. Preliminary results show that exposure to Children TV does not affect educational performance, school motivation or preferences for education of Italian students, at any grade. Students exposed to Children Television instead experience a reduction in physical bullying in elementary school, possibly explained by reduced exposure to violence in general TV. Future analysis will explore heterogeneity in the treatment effect, and exploit commercial TV data to investigate consumption patterns and estimate actual treatment effects.

JEL codes: I21; I28; J24 Keywords: Educational outcomes, non-cognitive skills, Early interventions, Television, Media

————————————————————— ?School of Economics, Singapore Management University. Email: [email protected] Acknowledgements: I thank prof. Costas Meghir and seminar participants at SMU Macro labor and Applied Micro Workshop for useful feedback. Financial support from SMU MOE Tier 1-A Internal Grant 18-C244-SMU-003 enabled data collection for this project. The usual disclaimers apply.

1 1 Introduction

Between 2008 and 2012 Italy transitioned from analogue to Digital Terrestrial Television (DTT). The technical upgrade turned into a substantial reform of the television system: hundreds of new channels were added to the existing TV programming. The switch to digital TV was in particular consequential for children. Several children-specific channels became freely available to all the population, marking the start of public Children . In this paper I study how exposure to the new television regime affected educational achievement, non-cognitive skills and behavior of Italian students. There are two main differences between this paper and the previous literature. First, the main focus of this study is on the effects of TV on non-cognitive outcomes and children behavior. The existing economic literature only considers effects on academic performance.1 Second, my setup and data allows me to identify the effect of Children Television, high- quality programming suitable to children, rather than an expansion of general TV. This paper thus also contributes to the literature on early child interventions.2 Before the reform took place, children programming aired in two of the six national public TV channels during specific hours of the day. These programs were, with a few exceptions, targeted to grade school children and teenagers. Children-specific channels meant for narrow demographics (including pre-schoolers), similar to those introduced after the switch to DTT, were only available on Pay-TV. The DTT reform basically extended high quality Children TV to the whole population. The full transition to DTT happened with the switch-off (termination) of analogue signal transmission. The switch-off schedule was mandated by the central government, and de- pended on technical factors like existing antenna infrastructure and geography. Switch-offs were generally implemented at the regional level: the first regions switched-off in 2008, the last regions did so in 2012. The time variation in switch-offs translates into differences in ex- posure to Children Television of at most four years between different (sometimes bordering) regions. I exploit this setup in a difference-in-difference design. My analysis uses INVALSI data, which records from 2010 to 2017 administrative school information and standardized test scores for the universe of Italian students attending specific grades in compulsory school. In grades 5 and 6, students responded to surveys inquiring about their school motivation, preferences for schooling, bullying frequency, time usage and

1See for instance Gentzkow & Shapiro (2008) and, more recently, Hernæs et al (2019). 2My treatment can be thought of as a scaled up version of what Kearney & Levine (2015) study.

2 parental inputs, outcomes in my analysis. I complement my analysis with commercial data on television viewership from 2008 to 2018. The data is disaggregated by gender, age, socio- economic status and region, which allows me to precisely measure television consumption for relevant demographic profiles. Given the reform implementation, the data allows me to identify two treatment effects. Using data up to 2012, before all regions had switched to DTT, I identify the effect of being exposed to Children TV for up to three years. Using data from 2013, after all regions had switched to DTT, I instead identify the effect of additional exposure, up to eight years, to Children TV. Both effects are intention-to-treat effects. In future analysis I will use Children TV consumption data to measure the actual treatment effects. The potential size of the treatment depends on how much TV children watched at the time of the reform. One year after the full implementation of the reform, children aged 4-14 were watching on average 2 hours and 49 minutes of TV per day, a substantial amount of time.3 The actual size of the treatment depends instead on whether children switched to Children TV once it became available. OssCom documented in 2013 a change in consumption patterns for children, who were increasingly substituting Children Television for general TV after the DTT transition.4 While compliance to the treatment appears to be high, it is likely to be heterogeneous. Children with Pay-TV had few reasons to switch to public Children TV, since they could already access (possibly better) Children TV. Poor families might have higher TV consumption, but less TV sets available for children. TV consumption data will allow me to disentangle consumption patterns, while the large dataset allows to explore heterogeneity in the treatment effect. There are different channels through which Children Television can affect the outcomes I consider. First, Children Television programming might convey educational content and include better Italian than general TV, which could affect cognitive outcomes. Second, Chil- dren Television might portray pro-social behavior and positive attitudes towards schooling, which could affect students’ behavior and preferences for schooling. Last but not least, if time spent on Children Television translates into lower exposure to general television, children will be less exposed to violence and inappropriate content. The relevance of these mechanisms hinges on family background, so it will be important to study heterogeneity in treatment effects. Preliminary results show that exposure to Children TV does not affect students’ educa-

3“Il Sole 24 Ore”, May 9th 2013. 4OssCom is the Research Center on Media and Communication at Cattolica University.

3 tional performance and deep parameters like school motivation and preferences for education, at any grade. Students exposed to Children Television instead experience a reduction in physical bullying in elementary school. Children TV seems to work “negatively” as a buffer against violence, rather than “positively” by fostering pro-social behavior. Finally, students exposed to Children TV report more books at home in Elementary school. In the absence of effects on preferences for schooling, I attribute this result to targeted advertisement on Children TV. The literature on the effects of television on educational outcomes was started by Gentzkow & Shapiro (2008), who analyzed in their seminal paper how introduction of television in the US in the 1940s and 1950s affected standardized test scores. They found that television exposure during pre-school age raised reading and general knowledge scores for socially dis- advantaged groups. Recent literature on the effects of TV exploited expansions of the cable TV infrastructure (Hernæs et al, 2019) or, like this paper, the transition to DTT (Nieto, 2017). These papers identify the broad effect of a change in the supply of TV programs on educational outcomes (and some behavioral outcomes, in the last case). Hernæs et al (2019) studies the effect of the expansion of TV in Norway from one high quality channel to multiple entertainment channels, and find that students exposed at ages 0-18, especially if boys and high SES, display lower levels of ability, and are less likely to graduate from high school. Nieto (2017) uses the transition to DTT in the UK, where cable TV was already widespread, to study how passing from five national channels to many DTT channels affects children scores and risky habits. He finds that exposure to television improves test scores at age 11 and reduces the probability of drinking later on. The literature is still open: in- troduction or expansion of TV is found to have both positive (Gentzkow & Shapiro, 2008; Kearney & Levine, 2015; Nieto, 2017) and negative effects (Hernæs et al, 2019) on academic achievement. Finally, it is worth noting that the economic literature already exploited the implementation of the reform I consider: Barone et al. (2015) analyze the effects of the re- formed television system on electoral outcomes, while Minale & Mastrorocco (2018) consider effects on media outcomes. The latter show in particular that the timing of the reform pro- vides a valid quasi-experiment to identify the effect of DTT on crime reporting and electoral behavior in Italy. My paper contributes to the literature on the effects of television in the following ways. First, I provide credible estimates of the effect of Children TV on educational, non-cognitive, and behavioral outcomes. Most of the previous literature focused on the effects of increased general TV consumption, and only considered educational outcomes. Second, I estimate the

4 effect of Children TV on Children TV consumption. These estimates allow me to gauge the actual treatment effect of exposure to Children Television on the main outcomes. Previous literature only identified reduced-form effects. Third, I investigate the intensive margin effect of exposure to Digital TV. I study how children reacted to the additional supply of TV channels, which included Children TV itself, educational programs, and low quality channels. Fourth, I measure the contemporaneous effect of Children Television on students cognitive and non-cognitive skills. Previous literature only identified medium- to long-run effects of increased exposure to television. Fifth, I estimate the effect of Children TV on students exposed to the new television regime at different ages, including pre-schoolers. This paper thus also contributes to the literature on early child interventions. Previous literature did not differentiate effects of TV by age. Finally, population data allows me to precisely detect heterogeneous effects, which is not possible in previous studies based on survey data.

2 Educational Setup

2.1 Education system

In this section I provide a brief description of the Italian education system, focusing in particular on compulsory school during 2010-2018, the period I consider in my analysis. Kindergarten runs from age three to six. Even if not compulsory, it is attended by most students.5 Kindergarten focuses on educational goals, on top of basic physical, social and emotional child development. Hence it can be considered as a pre-school stage. Compulsory education starts the year children turn 6 and ends the year they turn 16. Tuition is free in state schools, attended by 95.1% of students, while fees can be charged in private schools.6 Compulsory school is co-educational and includes three levels: elementary school (grades 1-5), junior high (grades 6 to 8), and the first part of high school (grades 9 and 10). At each level students move to a different building, and hence interact only with similarly aged peers. Admission to junior high and high school is not competitive. If oversubscribed, schools are however allowed to prioritize students living closer to the school. Ability tracking and multi-grade classes are not allowed. The maximum number of students per class is regulated

5In my sample, about 93% of the students attended at least one year. Monthly fees are generally subsidized, and currently around 300 EUR/month. 6All educational materials are however paid by children’s families. Data on private school attendance is taken from ISTAT (2017) and refers to school year 2014/2015.

5 (DPR 81/2009) at 26, 27, and 30 students respectively for elementary school, junior high and high school. When students enroll in high school, they choose between three types of track. Vocational tracks are meant to prepare students for specific jobs. General tracks provide technical skills, and also prepare for college. Academic tracks only provide theoretical skills and offer the best preparation for college. High school tracks last five years (grades 9 to 13), but students on vocational tracks might end their education at grade 11 with a vocational diploma. School assessment is continuous, and based on criterion-referenced grades on a 0 to 10 scale (6 is the passing grade). Subject-specific school grades, computed as the average for each semester in which the academic year is formally divided, are assigned in report cards. If at the end of the academic year students fail a certain number of subjects, they are retained. Grade retention is minimal in elementary school, where it must be approved by parents, increases in junior high, and is much higher in high school.7 At the end of junior high (grade 8) students take national exams. The final grade is largely determined by exam scores, and in minor part by school grades. Since 2010, students in grades 2, 5, 6, 8, and 10 take INVALSI standardized tests, meant to evaluate the education system. The tests are low-stakes to the students, with the exception of grade 8 tests, which up to academic year 2016/2017 concurred with the final exam score to determine junior high’s final grade. Non-compulsory education starts on grade 11, when high school curricula change, gen- erally becoming more specialized. At the end of high school students take again national exams. Their final GPA is largely determined by exam scores, and in minor part by school grades. Italian students apply for specific majors at specific universities. Universities are mostly public, heavily subsidized (in particular for poorer households), and adhere to the EU harmonized system (Bologna Process). High school GPA is in general ignored when students apply for, or directly enroll into, a major. For many majors there is indeed no requirement to enroll, apart from having a 5-year high school diploma. Universities can and increasingly do restrict enrollment to specific majors via admission tests, or condition second-year enrollment to passing specific first-year exams.

7In my sample students who are enrolled in a grade lower than the one they should attend are 2% by grade 5, 10% by grade 8, and 20% by grade 10.

6 2.2 Data

The main data I use in my analysis comes from INVALSI, the National Institute for the Evaluation of the Education System. The institute operates directly under the Ministry of Education. It is responsible for the administration of the national tests used to assess the Italian education system and the performance of individual schools. As mentioned in the previous section, INVALSI has been running national tests on the universe of students attending grades 2, 5, 6, 8, and 10 since 2010. The tests measure students’ academic proficiency (similarly to school grades) and skills (similarly to OECD PISA tests) in Italian and math.8 The tests are taken on the second week of May, near the end of the school year, and last 60 minutes per subject.9 All schools, both public or private, are subject to the tests, and only students with serious disability are exempted from taking them. After taking the tests, students in grades 5, 6, and 10 fill in a 30-minute survey. The detailed surveys allow me to measure school motivation, preferences for schooling, bully- ing frequency, time usage and parental inputs (number of books of books, availability of encyclopedia), which I use as outcomes in my analysis. INVALSI datasets include a fairly complete set of demographics: gender, foreign back- ground, day care and kindergarten attendance, parental education, and parental occupation. Class and school indicators allow me to measure class size, grade size (a proxy for the level of urbanization), percentage of foreign and disadvantaged students in class, useful as additional controls and to inspect heterogeneity of treatment effects. The obvious choice to measure educational achievement would be the INVALSI test scores. Unfortunately the tests are known to be affected by teacher cheating, prevalent in southern administrative Regions and explained by teachers shirking when marking the tests.10 The level of cheating might have changed over time, so these measures of achievement could be problematic if used in a difference in difference analysis. To measure educational achievement I instead use math and Italian school grades, reported to students and their families on first semester report cards and included in the INVALSI administrative data. While these grades are not standardized, any time-invariant regional difference (grading

8For instance the Italian test is made of a grammar section (academic knowledge) and a reading comprehension section (applied skill). An English test was added since 2018. 9Grade 2 tests last half an hour. 10This is documented by Angrist et al. (2017). See Bertoni et al. (2013) and Battistin et al. (2017) for related literature on cheating in INVALSI tests. Notice that for a 5% random sample external monitors were hired to invigilate tests and report results. No student or teacher cheating is possible for this subsample.

7 strictness) and region-invariant time difference (grade inflation) is removed in the difference- in-difference specification I use. Methodologically there should thus be little difference using these grades instead of standardized grades. A remaining substantive difference is that school grades are known by students and their families, while INVALSI standardized test scores are only reveled to teachers and schools.11 Finally not all information mentioned above is available in all grades and years. Table1 summarizes data availability in my sample.

Grade 2 Grade 5 Grade 6 Grade 8 Grade 10 (age 8) (age 11) (age 12) (age 14) (age 16) Grades, tests, Grades, tests, A/Y 2010 Grades, tests Tests survey survey Grades, tests, Grades, tests, Grades, tests, A/Y 2011 Grades, tests Tests survey survey survey Grades, tests, Grades, tests, Grades, tests, A/Y 2012 Grades, tests Tests survey survey survey Grades, tests, Grades, tests, Grades, tests, A/Y 2013 Grades, tests Tests survey survey survey Grades, tests, Grades, tests, A/Y 2014 Grades, tests Tests survey survey Grades, tests, Grades, tests, A/Y 2015 Grades, tests Tests survey survey Grades, tests, Grades, tests, A/Y 2016 Grades, tests Tests (survey) (survey) Grades, tests, Grades, tests, A/Y 2017 Grades, tests Tests (survey) (survey) Grades, tests, Grades, tests, A/Y 2018 Grades, tests Tests (survey) (survey)

Table 1: Summary of education data available from INVALSI

11Schools can advertise aggregate performance in INVALSI tests. Notice that from 2018 tests are taken on computers, and students see their test scores.

8 3 Children Television

3.1 The Television Market

The Italian television market was historically dominated by two incumbents offering non- specialized channels. The state-owned incumbent RAI, mostly funded by public contribu- tion, used to operate three national channels: , and (the latter also includes regional programming). The company had to meet specific societal goals set by the parlia- ment, but also competed with , the private company controlled by former Italian prime minister Silvio Berlusconi. Mediaset used to offer three national channels, mostly geared towards entertainment: , , and . Two more channels ( and MTV), originally owned by telecommunications incumbent Telecom, completed the list of national channels. On top of the national channels, Italians had access to low-budget regional networks, whose programming mostly focused on tele-marketing and, to a less ex- tent, local news. The market share of RAI and Mediaset decreased from 2003, with the entry of satellite pay-TV broadcaster Sky. With the transition to Digital Terrestrial TV (DTT), new operators entered the free television market. The most important, in terms of audience, are Cairo Communication, Sky, Discovery, WarnerMedia, and GEDI (Gruppo Editoriale L’Espresso). The two incumbents also launched new channels on DTT: Rai and Mediaset added respectively nine and ten channels, and both operated two channels entirely devoted to children programming. The number of freely available national channels increased from eight to about one hundred, further reducing the incumbents’ market share (see picture 1). As of 2017, Rai’s market share was stable at 37.2%, Mediaset’s decreased to 30.3%, and Sky’s increased to 8.5%. The remaining 24% was shared by the plethora of new DTT chan- nels.12 Even with these developments, the market is still an oligopoly, with two companies controlling two thirds of the market.

12Data taken from AGCOM “2018 Annual report”.

9 Figure 1: Trends in television audience (1990-3013).

Note: Extracted from AGCOM 2014 report “The communications sector in Italy”.

3.2 Regulation of Children TV

Children programming in Italian TV is affected by several layers of legislation, mandating both negative protection from harmful content and positive provision of content suitable to children. The first level of legislation is international law. On 27 May 1991 (legge n. 176) Italy ratified the 1989 United Nations Convention on the Rights of the Child.13 In the spirit of the UN convention, parliament passed on 29 November 2002 the “Self-regulation law on children and minors”, enforced from 2004. The law banned from TV content not suitable for children between 7 am and 10.30 pm, limited violent and sexual scenes in both news and advertisement, and forced broadcasters to mark content for adults. The law also mandated an improvement in the quality of programs for children, which had to foster civic and personal development, promoting collaboration with schools and experts in pedagogy. Importantly, it compelled the two dominant companies to offer in at least one of their channels content suitable to children during prime time. It also created a protected time window running from 4 to 7 pm. During this time children protection standards, applying to all networks, became stricter, and the two incumbents had to broadcast children programming. More stringent provisions applied to the public incumbent, including the mandate to create two channels for children, including one for pre-schoolers. Reports from AGCOM, the communication sector

13This treaty protects three types of children’s rights: the right of access to the media, the right to information promoting child development, and the right to protection from harmful material.

10 regulator and competition authority, and CORECOM, its regional counterpart, stressed that enforcement of these laws, in particular the positive provision part, was weak, and several violations were reported over time. Children programming was also affected by EU legislation. The “Audiovisual Media Ser- vices” directive (2007/65/CE) was adopted in the 2005 Radiotelevision Code (D.Lgs 31 July 2005, “Television and children”).14 The law forced all national broadcasters (including those on pay-TV) to reserve at least 50% of broadcasting time to broadly defined EU productions and at least 10% (15% for the public broadcaster) of net earnings to production or purchase of independent EU productions. Within the EU quota, at least 6% of programming had to be devoted to programs for children, and at least 20% of programming had to be suitable to children. All in all, the regulated Italian television system offered basic protection and suitable content to children. This was especially true for children whose families provided guidance and restrictions. Children without proper supervision watching TV outside the 4 to 7 pm time window could easily be exposed to non-suitable content.

3.3 Supply and Demand of Children TV

Before the transition to DTT took place, the supply of free children programming was con- centrated in two national channels, in specific hours of the day. In 2010, RAI broadcast about 1688 hours of children programming, 68% on Rai 2 and the rest on Rai 3. These programs were mostly targeted to grade school children, and included some programming for pre- schoolers, in line with the social mandate of the state-owned provider. Mediaset broadcast about 1339 hours of children programming, virtually all on Italia 1.15 Mediaset’s “children programming” was instead mostly targeted to teenagers and young adults. Children-specific channels meant for narrow demographics, similar to those introduced after the switch to DTT, were only available on Sky and, later on, Mediaset Pay-TV.

At the start of the transition to DTT, in 2008, Italian families had immediate access to four channels with programming exclusively targeting children. By the end of the transition, in 2012, the number of such channels increased to seven (see Table2), making Italy the country with the highest number of free children television channels in the EU.16

14Later on amended in the “Audiovisual and radiophonic media services code” 2010 law (D.Lgs 15 March 2010). 15Statistics taken from AGCOM “White Book on Media and Minors” (2018). 16See the description of the European market in Osscom (2013).

11 Channel Company Country Target age On DTT since Boing Mediaset/Warner Italy/US 03-14 11/2004 RAI Italy 08-14 06/2007 Rai Yoyo RAI Italy 04-07 07/2009 K2 Discovery US 04-14 07/2009 Frisbee Discovery US 04-14 07/2010 Mediaset/Warner Italy/US 03-06 08/2011 Super! De Agostini/Viacom Italy/US 10-14 03/2012 PoP Sony Japan 06-09 05/2017

Table 2: Children TV channels freely available on Digital Television

With the transition to DTT both Rai and Mediaset shifted part of their children pro- gramming from traditional to thematic channels. In 2016 Rai aired only 619 hours of children programming, mostly on Rai Due (a 63% decrease). Mediaset reduced children programming on Italia 1 to 1092 hours (an 18% decrease). Some of these changes happened already from 2010, when some regions had not yet transitioned to DTT. This implies that children in these regions saw the amount of children programming decrease before the switch to DTT.17 Let’s consider demand for children programming. First of all, the fraction of chil- dren watching TV (AMR%) increased from 2012 to 2016 among the 4-7 demographic, re- mained stable among the 8-13 demographic, and decreased among the 14-17 (teenagers) demographic.18 In 2012, close to the end of the transition to DTT, Children TV reached a 39% share among children aged 4-7, and a 25.8% share among children 8-14.19 Table3 reports the relative share of the 15 top channels seen by children in 2016, four years after the DTT transition. The share of Children TV grew among the intended 4-13 target, especially among the youngest. Teenagers were mostly unaffected by Children TV. They either stuck to traditional channels, or switched to the new thematic channels brought by DTT (some of which cater to young adults). A 2017 study on children 3 to 16, based on Auditel television consumption data, reported that the share of free Children TV had increased by 42% in five years. The share of private Children TV, mostly offered on Sky, had instead dropped by 43% in the same period. The

17See AGCOM “White Book on Media and Minors” (2014, 2018). 18AGCOM “White Book on Media and Minors” (2018). 19OssCom “Television and Children” (2013)

12 report explains this as a consequence of the dire economic situation in Italy. This inter- pretation is actually not supported by the data: the number of Sky subscriptions did not decrease, and as of 2018 Sky is the first company in terms of earnings in the television mar- ket.20 An alternative interpretation is that Italians with pay TV subscriptions (around 33% of the population) deemed public Children Television a valid alternative to Sky’s Children television package, which would have cost them 5 additional euros per month. The switch to DTT had basically brought high quality Children TV to the whole population. The same report stressed the attractiveness of the market for children, which generates each year in Italy more than 100 million euros in ads revenues. The entry of another Children Television provider on free DTT in 2017 appears to support this analysis.21

Age Children TV Traditional TV Thematic DTT 04-07 59.66% 34.82% 5.53% 08-13 40.52% 52.04% 7.44% 14-17 8.54% 75.29% 16.17%

Table 3: Most watched channels: shares by category in 2016

Note: The data is based on the shares of the top 15 programs watched by each demographic, reported in AGCOM “White Book on Media and Minors” (2018). The aggregate shares in the table are computed summing the shares of the channels in each category over the sum of all the shares. Children TV channels are those reported in table2. Traditional channels are Rai 1, Rai 2, Rai 3, Rete 4, Canale 5, Italia 1, La7 and MTV. Notice that these channels might also contain programming for children. Thematic DTT channels are the free channels available on DTT and not belonging to the two other categories.

20See AGCOM “2018 Yearly Report”. 21Results are discussed in Catholic newspaper Avvenire on the 9 Dec 2017 issue (https://www.avvenire.it/agora/pagine/tv-dei-ragazzi). Italian Catholic organizations appear to be particularly active in monitoring Italian Television, especially with concern to children protection.

13 4 The reform

4.1 EU Framework

EU institutions first recommended and later on mandated that member countries switch from analogue to digital television (from now on DTT) by 2012.22 There were many reasons behind the reform. Digital TV allows the broadcast of more channels, allowing to increase competition and plurality in the media market. As described in Section 3.3, this is precisely what happened in the Italian case. Moreover, since DTT uses a narrower range of frequen- cies than analog television, the switch frees up bandwidth that can be exploited for other uses.23 Frequencies could for instance be used for wireless internet, to cover households living in remote or rural parts, thus meeting another objective of the European Union. This did not happen in Italy, where freed frequencies were not auctioned. Finally Digital television was seen as a technological improvement over its analog counterpart. First, for every ana- log channel, it is possible to broadcast six to eight digital channels, since digital signal is compressed. Second, DTT offers better reception and distribution of the signal, allowing to cover areas that were previously not reached due to geographic reasons. Third, digital signal allows to broadcast programs in high definition, an increase in quality over analog signal. Finally, DTT supposedly produces less electro-magnetic pollution.

4.2 The Transition to DTT

Transmission of DTT started in Italy around 2003, and increased rapidly from 2004, when the number of broadcast channels increased. From 2004 the government provided subsides covering the full cost of a DTT decoder, the device needed to receive digital signal on older TV sets. Between 2005 and 2007 both decoder sales and the number of digital channels sharply increased. Different areas of Italy transitioned to DTT on assigned switch-off dates, when analog signal was turned off and digital signal was boosted. From that day onwards, households without a recent television, or a DTT decoder connected to an older TV set, were no longer able to watch television. The first switch-off was performed in Sardinia in the fall of 2008, the last was carried out in Sicily in the summer of 2012. The regions that switched-off before 2010 went through an intermediate phase called

22European Council communication 2005/204 and European Commission recommendation 2009/848. 23This is generally referred to as digital dividend.

14 switch-over, which could last from a few months to more than a year. During this period, both analog and digital terrestrial signals co-existed. Due to expected interference and bandwidth saturation issues, it was not possible to broadcast all analog TV channels alongside the digital channels. The government thus decided to stop analogue broadcast of one channel per incumbent, namely Rai 2 and Rete 4.24 Sardinia and Aosta Valley were the first regions to go through switch-over, in the spring of 2007. They were also the regions in which this phase lasted the longest, possibly due to expected difficulties related to the mountainous terrain. All regions switching-off from 2010, apart from Emilia-Romagna and Lombardy, skipped the switch-over phase and went instead through sharp switch-offs: on the same day digital signal started to be broadcast, analogue signal was turned off. The central government scheduled the switch-off dates from 2008 to 2012, and the sched- ule depended on availability of antenna infrastructure and geographical factors.25 The switch- off areas generally corresponded to the administrative Italian Regions (see Picture2), with four exceptions. In the north-west region of Piedmont and Lazio, the region hosting the capital, different Provinces switched-off at different times. In the northern Region of Emilia- Romagna and insular Sardinia different Provinces switched-over at different times, while switched-off happened at the same time.26 Picture3 shows the year in which each transition area (from now on DTT-Region) switched-off. Regions from both the South and North of Italy, including highly populated Campania and Lazio, switched-off before 2008, followed by the rest of the Northern regions in 2010. Central Italy followed in 2011, while the remaining Southern regions switched-off in 2012. The differential timing of the reform translates into 1 to 4 years difference in expo- sure to the new television system between (sometimes bordering) regions. This provides the source of exogenous variation that I will exploit to answer my research question, and that will be discussed in Section 5.3.

24Notice that Rai 2 is one of the two traditional channels offering programs for children. 25Legge 20 marzo 2001, n. 66. 26This is not visible in the maps, which refer instead to switch-off times.

15 Figure 2: Switch-off time by province Figure 3: Switch-off year by province

Areas transitioning to Digital Terrestrial Televi- Purple areas switched-off before 2009, red areas sion in the same month are marked with the same switched-off in 2010, orange areas switched-off in color. 2011 and yellow areas switched-off in 2012.

Note: Thick and subtle lines mark respectively regional and provincial borders.

16 5 Treatment and Identification

5.1 Mechanisms

Provision of Children Television can affect both the amount of Television children watch (extensive margin effect), and the quality of the content they are exposed to (intensive margin effect). In the following I separately discuss these two effects. I expect the extensive margin effect to be generally negative. Increased TV watching can translate into less time spent on homework, sports, social and cultural activities. More contact time with TV might also lead children to spend more time on video-games. Finally, Children Television might partly substitute for care-takers effort. In the case of particularly disadvantaged families, these alternative activities might actually be detrimental, so more TV consumption might actually be positive for these children. I expect the intensive margin effect to be generally positive instead. First, Children Television programming might include educational content. It also generally features better Italian than general TV, often characterized by regionalized or incorrect language. This might affect educational achievement, and, in particular, Italian grades. Second, Children Television is explicitly regulated (see Section 3.2) to foster interaction and child development, often portraying positive attitudes towards schooling and stressing social values (friendship, respect, diversity). This could affect behavior in school, school motivation and preferences for schooling. Last but not least, if time spent on Children Television translates into lower exposure to general television, children will be mechanically less exposed to violence and inappropriate content. The relevance of these mechanisms strongly depends on family background. More ed- ucated families might already provide educational inputs, speak good Italian, and protect their children from inappropriate content. It is thus likely that Children Television will affect more strongly students from disadvantaged families or with immigrant background.

5.2 Dynamics

The effects discussed above, especially those related to child development, might take time to materialize, hence it is possible that Children TV have dynamic effects. There are two more reasons why treatment effects might be dynamic. First, since I estimate an intention-to-treat effect, compliance becomes part of the dynamics. As discussed in Section 3.3, children did switch to Children TV, but this happened over time. Second, students who were exposed for

17 more years to Children Television were also exposed at a younger age. If younger children are more receptive to TV content, effects are going to be stronger for them. Finally, notice that compliance itself is going to be heterogeneous. Wealthier families might already had access to high quality Children Television via pay TV. Provision of Public Children Television might have been virtually irrelevant to them. On the other hand, poorer households might have less television sets, which implies lower exposure to Children Television. All in all it is reasonable to expect children TV to have both dynamic and heterogeneous effects. In future analysis, I will use television consumption data to better disentangle con- sumption patterns. Furthermore, the large dataset will allow me to explore heterogeneity in the treatment effect.

5.3 Identification

The implementation of the DTT reform provides exogenous variation in exposure to Children Television by DTT Region, illustrated in Table4. For the bulk of the sample, the reform creates differences in exposure to Children Television of one to three years between DTT Regions, which I exploit in a difference-in-difference analysis. DTT Regions exposed to Children Television last act as the control group, and allow to identify outcome trends in the absence of treatment until 2012. Deviations from these trends in DTT Regions exposed one to three years earlier to Children Television are interpreted as the effect of exposure to Children Television for one to three years. The setup allows to identify the effect of exposure to Children TV for up to eight years. However, starting from 2013 all DTT Regions are exposed to Children Television (see Section 4.2). When estimating the effect of exposure to Children TV for four to eight years, effects are detected off a “control” group already exposed to Children TV. So the setup only allows to identify the effect of additional exposure to Children Television for lags four to eight. If for instance all effects of Children TV appear from the first year, these lags should be insignificant.27 Right now both effects I estimate are intention-to-treat effects: students might or might not have been exposed to Children TV when this became available. In future analysis I will use Children TV consumption data to measure actual treatment effects.

27The issue clearly relates to the point raised by Borusyak et al (2017) and Goodman-Bacon (2018). If all units are at some point treated, and effects are assumed to be dynamic, it is not possible to fully identify dynamic treatment effects. The reason is that, unless specific assumptions are made, it is not possible to identify a counterfactual trend out of the control group when all units are treated.

18 AppendixesA toE provide, for each grade, more information on identification. Subsec- tion 1 illustrates treatment variation: for how many years students were exposed to Children Television and at which age they were first exposed. Subsection 2 discusses in detail identi- fying variation and shows the final specification. One final point that should be highlighted from these tables is that grade 10 students are hardly affected by Children TV. As can be seen in Table E.2, most students were exposed to Children Television when they already were teenagers, while Children TV specifically targets children aged 3 to 14. When interpreting results for this grade, effects are best attributed to the new thematic channels some teenagers started to watch. These channels mostly consist of light entertainment TV (see the discussion in Section 3.3).

5.4 Identifying assumptions

The main identifying assumptions are the following. First, the parallel trends assumption, in this specification with non-parametric trend, requires that differences in outcomes between DTT Regions remain constant through the years. The assumption is testable in my setup for regions that switched to DTT in 2011 and 2012. If outcomes evolved in the same fashion in these regions between 2010 and 2011, I should observe β0 = 0. Second, treatment effect is independent of calendar year: (βt, δt) ⊥ αy. If this was not the case, it would not be possible to identify δt. Another set of identifying assumptions is more easily discussed reinterpreting the differences- in-difference setup as an IV.28 In my case the DTT reform acts as an instrument for the treatment, Children Television. First, the instrument should be excluded, that is, the reform should affect my outcomes only through Children Television. This is true to the extent that children switched to DTT, and not to other programs, on the reformed television system. This is an assumption I can test with television consumption data. It is also possible that the switch to DTT directly affected households. For instance, a decoder was required to watch digital television. However DTT decoders were fully subsidized by the government, and in any case were cheap. So I don’t expect any significant effects on children from this channel. Households might instead have purchased new TV sets after the switch-off, as new TV sets included a digital TV tuner. These households might have sustained a large-ticket expen- diture, while at the same time improving the quality of their television sets. It is however not clear how big an effect this could have for students on the outcomes I consider. Second,

28For a discussion see Angrist & Pischke (2009).

19 treatment should not affect outcomes in the control group (SUTVA). As discussed before, children programming was switched from traditional TV to Children TV, but this happened after the reform. A few programs on Rai 3 were actually shifted to Children TV before switch-off, but these programs represent a small fraction of traditional television children programming, hence should not undermine identification.

20 Table 4: Exposure to DTT by Region and Year

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 1 1 1 1 1 1 1 1 Piedmont(West) 1 1 1 1 1 1 1 1 Piedmont(East) 0 1 1 1 1 1 1 1 Lombardy 0 1 1 1 1 1 1 1 Trentino 1 1 1 1 1 1 1 1 Veneto 0 0 1 1 1 1 1 1 Friuli-Venezia Giulia 0 0 1 1 1 1 1 1 Emilia-Romagna (West) 0 1 1 1 1 1 1 1 Emilia-Romagna (East) 0 0 1 1 1 1 1 1 Liguria 0 0 1 1 1 1 1 1 Tuscany 0 0 1 1 1 1 1 1 Umbria 0 0 1 1 1 1 1 1 Marche 0 0 0 1 1 1 1 1

South: Lazio(North) 0 0 1 1 1 1 1 1 Lazio(South) 1 1 1 1 1 1 1 1 Abruzzo 0 0 0 1 1 1 1 1 Molise 0 0 0 1 1 1 1 1 Campania 0 1 1 1 1 1 1 1 Basilicata 0 0 0 1 1 1 1 1 Apulia 0 0 0 1 1 1 1 1 Calabria 0 0 0 1 1 1 1 1 Sicily 0 0 0 1 1 1 1 1 Sardinia(North) 1 1 1 1 1 1 1 1 Sardinia(South) 1 1 1 1 1 1 1 1

Note: The amount of time each DTT Region is exposed to DTT is measured in years, and corresponds to the difference between May 15 of each year and the switch-off date. For DTT Regions going through switch-over, the transition date is the earliest between four months from switch-over and switch-off date.

21 6 Results and Conclusion

The model I estimate is the following:

3 T¯ X X Yigry = αr + αy + β0 T (0) + βt T (t) + δt T (t) + ∆ Xigry + igry (1) t=1 t=4

where i indexes the student, g the grade, r the DTT region and y the year. αr and αy are respectively DTT Region and Year fixed effects. The treatment is T , the number of years t, ranging from 0 to T¯, students were exposed to Children TV. The coefficients of interest are βt , the effect of exposure to Children Television, and δt, the effect of additional exposure to Children Television. Xigry is a vector of controls including demographics (gender, foreign status), socioeconomic background (parental education and parental occupation), class and school characteristics (class size, grade size, fraction of boys in class, fraction of foreigners in class, fraction of low educated). Standard errors are clustered at the DTT Region.29 Section 3 in AppendixesA toE illustrate results. For each outcome, I provide a picture where I display point estimates and 95% confidence interval for β0, βt and δt in Equation 1 in the upper panel, and trends in the outcome variable (Yigry) by switch-off year in the lower panel. Dashed lines indicate periods before transition to DTT, while solid lines mark periods after DTT transition. This allows to visually inspect whether outcomes are trending in a similar way over time, and in particular if this is the case before treatment. Overall the identification strategy seems to work: outcomes appear to move in tandem in DTT Regions exposed to Children Television in different years, and formal tests for parallel trends pass. This is true despite the large sample size, and resulting high test precision.30 The main result in this preliminary analysis is that students exposed, or differentially exposed, to Children Television do not perform differently in school, at any grade. Effects are insignificant at the 5% level, despite the high precision of the estimates.31 These results seems to be at odds with the economic literature on the effects of TV, which found either positive (Kearney & Levine, 2015; Nieto, 2017) or negative effects (Hernæs et al, 2019). While estimates are 0 and largely insignificant for grades 8 to 10, they tend to be negative, and sometimes significant at the 10% level, for grades 2 to 6. This seems to be particularly the case for math. Future analysis will delve deeper into those patterns, and examine whether

29Grade 8 specifications do not control for socioeconomic background and class fraction with low educated parents, since these data are not available. 30The specifications for educational outcomes use more than 3 million observations. 31I can detect effects as low as 3.5% of a standard deviation.

22 specific categories of students were negatively affected by the television reform, as found in the literature. I consider an extensive set of outcomes related to school motivation and preferences for schooling for grades 5 and 6, and similarly do not find any effect. The only exception is a small increase, 0.92% over sample average, in the frequency of homework in grade 5 (Figure B.5). All in all, it seems that cognitive and non cognitive skills of students are not responding to Children Television. Survey information from grades 5 and 6 allows me to measure the effect of Children Television on the amount of hours spent on TV. I find no effect for grade 5 children, and a small increase, 1.5% over sample average, for grade 6 students. Overall there seems to be no extensive margin effect. Consistently, there is no effect on the time students devote to video-games, sports or cultural activities. The largest effect I find relates to bullying. Grade 5 students are less likely to report harassment episodes (a 5.9% decrease over sample average, see Picture B.10), and physical assault (a 8.35% decrease over sample average, see Picture B.12) when exposed to more years of Children Television.32 Finally, I find that students exposed to more years of Children Television report more books at home in grade 5 (a 12.6% increase over sample average), and, to less extent, in grade 6 (a 2.2% increase over sample average), where effects are however barely significant. When interpreting results together, it appears that Children TV does not affect deep pa- rameters like preferences for schooling, motivation, and, likely, pro-social behavior. The de- crease in bullying might reflect a reduction in exposure to violence in general TV, rather than the pro-social effect of Children TV. In this sense, Children TV seems to work “negatively” as a buffer against inappropriate content, rather than “positively” by shaping pro-social behavior. In the absence of effects on educational achievement, motivation or preferences for schooling, I attribute the effects I find on books availability to targeted advertisement. Most of children programming indeed features a licensing model where characters intellectual property is licensed to other industries, most notably publishing and child entertainment.33

32 Grade 6 estimates instead reveal no effect on physical bullying (Picture C.10), an increase in harassment (a 9.2% increase over sample average, see Picture C.8) and in social exclusion episodes (a 3.9% increase over sample average, see Picture C.9). However tests for parallel trends in this grade tend to fail at standard significance levels, biasing upwards estimates (bullying was decreasing faster for late switchers than for early switchers). Further analysis will try to better assess the robustness of these results.

33The relationship between Children TV and publishing is analyzed in OssCom (2013).

23 I conclude by examining results in grade 10. As discussed in Section 5.3, in this grade effects are best attributed to the new thematic channels, rather than Children TV. Being exposed to more years of DTT programming does not affect educational choices, educational performance, school motivation, and preferences for education. It does however affect the number of books students have at home, where I observe decreases up to 9.5% of the sample average. This effect seems to amplify over time, but does not increase with further exposure. A possible interpretation is that digital entertainment is a substitute for traditional forms of entertainment like books.

24 References

AGCOM, 2014. “The Communications Sector in Italy”. Annual report on the activity carried out and on the work programmes 2014.

AGCOM, 2014. “White Book on Media and Minors,” jointly produced by AGCOM and CENSIS, September 2014.

AGCOM, 2018. “RELAZIONE ANNUALE 2018”. Annual report on the activity carried out and on the work programmes 2018.

AGCOM, 2018. “White Book on Media and Minors 2.0,” produced by AGCOM, 16 January 2018.

Joshua D. Angrist, Erich Battistin, and Daniela Vuri, 2017. “In a Small Moment: Class Size and Moral Hazard in the Italian Mezzogiorno.” American Economic Journal: Applied Economics, 9 (4): 216-49.

Joshua D. Angrist & Jörn-Steffen Pischke, 2009. "Mostly Harmless Econometrics: An Em- piricist’s Companion," Economics Books, Princeton University Press, edition 1, number 8769.

Piermarco Aroldi & Nicoletta Vittadini (editors), 2013. “Televisione e Infanzia. Rapporto sull’offerta televisiva per bambini in Italia,” jointly produced by OssCom – Media and Re- search Communication Center at Università Cattolica – and Fondazione per la Sussidiarietà, Milano 2013: 183.

Anne Austin & Jonathan Barnard & Hutcheon, 2016. “Media Consumption Forecasts 2016,” Zenith, SSN 2059-2620, June 2016.

Guglielmo Barone & Francesco D’Acunto & Gaia Narciso, 2015. “Telecracy: Testing for Channels of Persuasion,” American Economic Journal: Economic Policy, American Eco- nomic Association, vol. 7(2), pages 30-60, May.

Erich Battistin & Michele De Nadai & Daniela Vuri, 2017. “Counting rotten apples: Student achievement and score manipulation in Italian elementary Schools,” Journal of Econometrics, Elsevier, vol. 200(2), pages 344-362.

Marco Bertoni & Giorgio Brunello & Lorenzo Rocco, 2013. “When the cat is near, the mice won’t play: The effect of external examiners in Italian schools,” Journal of Public Economics, Elsevier, vol. 104(C), pages 65-77.

25 Gordon Dahl & Stefano DellaVigna, 2009. “Does Movie Violence Increase Violent Crime?,” The Quarterly Journal of Economics, Oxford University Press, vol. 124(2), pages 677-734.

Ruben Durante & Paolo Pinotti & Andrea Tesei, 2017. “The Political Legacy of Entertain- ment TV,” Forthcoming in American Economic Review.

Giacomo Gambassi, 2017. “L’inchiesta tv. È boom dei canali per i ragazzi. Ma crollano le reti a pagamento,” Avvenire © Online Edition, 9 December 2017 (link).

Matthew Gentzkow & Jesse M. Shapiro, 2008. “Preschool Television Viewing and Adoles- cent Test Scores: Historical Evidence from the Coleman Study,” The Quarterly Journal of Economics, Oxford University Press, vol. 123(1), pages 279-323.

Øystein Hernaes & Simen Markussen & Knut Røed, 2019. “Television, Cognitive Ability, and High School Completion,” Journal of Human Resources, vol. 54 no. 2 371-400.

Martin Kanzler & Susan Newman-Baudais, 2011. “Audiovisual Media for Children in Europe: Television for Children,” European Audiovisual Observatory, Amsterdam, 18 & 19 October 2011.

Johanna Karsenty & Claire Mitchell & Léa Besson, 2014. “Kids’ TV Consumption. How much do they watch?,” Eurodata TV.

Melissa S. Kearney & Phillip B. Levine, 2019. “Early Childhood Education by Television: Lessons from Sesame Street,” American Economic Journal: Applied Economics, American Economic Association, vol. 11(1), pages 318-350, January.

Mastrorocco, Nicola & Minale, Luigi, 2018. “News media and crime perceptions: Evidence from a natural experiment,” Journal of Public Economics, Elsevier, vol. 165(C), pages 230- 255.

Adrian Nieto, 2017. “Digital Television: Impact on Cognitive Development, Educational Inequality and Habits,” Granger Centre Discussion Working Paper. 29(2), pages 301 - 348

Ofcom, 2007. “The Future of Children’s Television Programming,” Ofcom Office of Commu- nications, Research Report 3 October 2007.

Francesco Siliato, 2013. “I bambini per 3,5 ore davanti alla Tv,”Il Sole 24 Ore © Online Edition, 9 May 2013 (link).

26 A Grade 2

A.1 Treatment variation

Table A.1: Years of Exposure to DTT by Region and Year Grade 2 (Age 8)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 3 4 5 5 5 5 5 5 Piedmont(West) 1 2 3 4 5 5 5 5 Piedmont(East) 0 1 2 3 4 5 5 5 Lombardy 0 1 2 3 4 5 5 5 Trentino 1 2 3 4 5 5 5 5 Veneto −1 0 1 2 3 4 5 5 Friuli-Venezia Giulia −1 0 1 2 3 4 5 5 Emilia-Romagna (West) 0 1 2 3 4 5 5 5 Emilia-Romagna (East) −1 0 1 2 3 4 5 5 Liguria −1 0 1 2 3 4 5 5 Tuscany −1 0 1 2 3 4 5 5 Umbria −1 0 1 2 3 4 5 6 Marche −2 −1 0 1 2 3 4 5

South: Lazio(North) −1 0 1 2 3 4 5 6 Lazio(South) 1 2 3 4 5 5 5 5 Abruzzo −2 −1 0 1 2 3 4 5 Molise −2 −1 0 1 2 3 4 5 Campania 0 1 2 3 4 5 5 5 Basilicata −2 −1 0 1 2 3 4 5 Apulia −2 −1 0 1 2 3 4 5 Calabria −2 −1 0 1 2 3 4 5 Sicily −2 −1 0 1 2 3 4 5 Sardinia(North) 2 3 4 5 5 5 5 5 Sardinia(South) 3 4 5 5 5 5 5 5

27 Table A.2: Age of Exposure to DTT by Region and Year: Grade 2 (Age 8)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 5 4 3 2 1 0 −1 −2 Piedmont(West) 7 6 5 4 3 2 1 0 Piedmont(East) 8 7 6 5 4 3 2 1 Lombardy 8 7 6 5 4 3 2 1 Trentino 7 6 5 4 3 2 1 0 Veneto 9 8 7 6 5 4 3 2 Friuli-Venezia Giulia 9 8 7 6 5 4 3 2 Emilia-Romagna (West) 8 7 6 5 4 3 2 1 Emilia-Romagna (East) 9 8 7 6 5 4 3 2 Liguria 9 8 7 6 5 4 3 2 Tuscany 10 9 8 7 6 5 3 2 Umbria 10 9 8 7 6 5 4 3 Marche 10 9 8 7 6 5 4 3

South: Lazio(North) 10 9 8 7 6 5 4 3 Lazio(South) 7 6 5 4 3 2 1 0 Abruzzo 10 9 8 7 6 5 4 3 Molise 10 9 8 7 6 5 4 3 Campania 8 7 6 5 4 3 2 1 Basilicata 10 9 8 7 6 5 4 3 Apulia 10 9 8 7 6 5 4 3 Calabria 10 9 8 7 6 5 4 3 Sicily 10 9 8 7 6 5 4 3 Sardinia(North) 6 5 4 3 2 1 0 −1 Sardinia(South) 5 4 3 2 1 0 −1 −2

28 A.2 Identification

Results in this section are based on the following specification:

Yi2ry = αr + αy + β0 T (0) + β1 T (1) + β2 T (2)+β3 T (3) + δ4 T (4) + δ5 T (5) + ∆ Xi2ry + i2ry (2)

Table B.3 shows identifying variation for grade 2 students, supposed to be 8, including age of first exposure to DTT (A) and years of exposure to DTT (T). Green cells denote control observations, used to run a placebo test corresponding to β0 in equation2. Red cells identify the last lag, δ8, in specification2.

Region switched to DTT in: 2009 2010 2011 2012 A/Y 2010 → A: 7; T: 1 A: 8; T: 0 A: 9; T: -1 A:10; T: -2 A/Y 2011 → A: 6; T: 2 A: 7; T: 1 A: 8; T: 0 A:9; T: -1 A/Y 2012 → A: 5; T: 3 A: 6; T: 2 A: 7; T: 1 A:8; T: 0 A/Y 2013 → A: 4; T: 4 A: 5; T: 3 A: 6; T: 2 A: 7; T: 1 A/Y 2014 → A:3; T: 5 A: 4; T: 4 A: 5; T: 3 A: 6; T: 2 A/Y 2015 → A:2; T: 5 A: 3; T: 5 A: 4; T: 4 A: 5; T: 3 A/Y 2016 → A:1; T: 5 A: 2; T: 5 A: 3; T: 5 A: 4; T: 4 A/Y 2017 → A:0; T: 5 A: 1; T: 5 A: 2; T: 5 A: 3; T: 5 Table A.3: Identifying Treatment Variation (Grade 2) Note: Children below 3 were not exposed to Children Television, therefore A and T sum to 8 only for students who were above 3 when first exposed to Children Television.

29 A.3 Results

DiD estimates and trends: Italian school grade .05 0 -.05 -.1 0 1 2 3 4 5 8.2 8.1 8 7.9 7.8 7.7 2010 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure A.1: Effect of Children Television on Cognitive Skills - Italian Grades (grade 2) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 2, including one lead (β0) and five lags (βt and δt). Standard errors are clustered at the DTT Region.

30 DiD estimates and trends: Math school grade .05 0 -.05 -.1

0 1 2 3 4 5 8.3 8.2 8.1 8 7.9 7.8 2010 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure A.2: Effect of Children Television on Cognitive Skills - Italian Grades (grade 2) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 2, including one lead (β0) and five lags (βt and δt). Standard errors are clustered at the DTT Region.

31 B Grade 5

B.1 Treatment variation

Table B.1: Years of Exposure to DTT by Region and Year: Grade 5 (Age 11)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 3 4 5 6 7 8 8 8 Piedmont(West) 1 2 3 4 5 6 7 8 Piedmont(East) 0 1 2 3 4 5 6 7 Lombardy 0 1 2 3 4 5 6 7 Trentino 1 2 3 4 5 6 7 8 Veneto −1 0 1 2 3 4 5 6 Friuli-Venezia Giulia −1 0 1 2 3 4 5 6 Emilia-Romagna (West) 0 1 2 3 4 5 6 7 Emilia-Romagna (East) −1 0 1 2 3 4 5 6 Liguria −1 0 1 2 3 4 5 6 Tuscany −1 0 1 2 3 4 5 6 Umbria −1 0 1 2 3 4 5 6 Marche −2 −1 0 1 2 3 4 5

South: Lazio(North) −1 0 1 2 3 4 5 6 Lazio(South) 1 2 3 4 5 6 7 8 Abruzzo −2 −1 0 1 2 3 4 5 Molise −2 −1 0 1 2 3 4 5 Campania 0 1 2 3 4 5 6 7 Basilicata −2 −1 0 1 2 3 4 5 Apulia −2 −1 0 1 2 3 4 5 Calabria −2 −1 0 1 2 3 4 5 Sicily −2 −1 0 1 2 3 4 5 Sardinia(North) 2 3 4 5 6 7 8 8 Sardinia(South) 3 4 5 6 7 8 8 8

32 Table B.2: Age of Exposure to DTT by Region and Year: Grade 5 (Age 11)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 8 7 6 5 4 3 2 1 Piedmont(West) 10 9 8 7 6 5 4 3 Piedmont(East) 11 10 9 8 7 6 5 4 Lombardy 11 10 9 8 7 6 5 4 Trentino 10 9 8 7 6 5 4 3 Veneto 12 11 10 9 8 7 6 5 Friuli-Venezia Giulia 12 11 10 9 8 7 6 5 Emilia-Romagna (West) 11 10 9 8 7 6 5 4 Emilia-Romagna (East) 12 11 10 9 8 7 6 5 Liguria 12 11 10 9 8 7 6 5 Tuscany 13 12 11 10 9 8 6 5 Umbria 13 12 11 10 9 8 7 6 Marche 13 12 11 10 9 8 7 6

South: Lazio(North) 13 12 11 10 9 8 7 6 Lazio(South) 10 9 8 7 6 5 4 3 Abruzzo 13 12 11 10 9 8 7 6 Molise 13 12 11 10 9 8 7 6 Campania 11 10 9 8 7 6 5 4 Basilicata 13 12 11 10 9 8 7 6 Apulia 13 12 11 10 9 8 7 6 Calabria 13 12 11 10 9 8 7 6 Sicily 13 12 11 10 9 8 7 6 Sardinia(North) 9 8 7 6 5 4 3 2 Sardinia(South) 8 7 6 5 4 3 2 1

33 B.2 Identification

Results in this section are based on the following specification:

Yi5ry = αr + αy + β0 T (0) + β1 T (1) + β2 T (2) + β3 T (3) + δ4 T (4) (3)

+δ5 T (5)+δ6 T (6) + δ7 T (7) + δ8 T (8) + ∆ Xi5ry + i5ry

Table B.3 shows identifying variation for grade 5 students, who are supposed to be 11, including age of first exposure to DTT (A) and years of exposure to DTT (T). Green cells denote control observations, used to run a placebo test corresponding to β0 in equation3. Red cells identify the last lag, δ8, in specification3.

Region switched to DTT in: 2009 2010 2011 2012 A/Y 2010 → A: 10; T: 1 A: 11; T: 0 A: 12; T: -1 A: 13; T: -2 A/Y 2011 → A: 09; T: 2 A: 10; T: 1 A: 11; T: 0 A: 12; T: -1 A/Y 2012 → A: 08; T: 3 A: 09; T: 2 A: 10; T: 1 A: 11; T: 0 A/Y 2013 → A: 07; T: 4 A: 08; T: 3 A: 09; T: 2 A: 10; T: 1 A/Y 2014 → A: 06; T: 5 A: 07; T: 4 A: 08; T: 3 A: 09; T: 2 A/Y 2015 → A: 05; T: 6 A: 06; T: 5 A: 07; T: 4 A: 08; T: 3 A/Y 2016 → A: 04; T: 7 A: 05; T: 6 A: 06; T: 5 A: 07; T: 4 A/Y 2017 → A: 03; T: 8 A: 04; T: 7 A: 05; T: 6 A: 06; T: 5 Table B.3: Identifying Treatment Variation (Grade 5)

34 B.3 Results

DiD estimates and trends: Italian school grade .2 .1 0 -.1 -.2 -.3

0 1 2 3 4 5 6 7 8 8 7.8 7.6 7.4 2010 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.1: Effect of Children Television on Cognitive Skills - Italian Grades (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

35 DiD estimates and trends: Math school grade .1 0 -.1 -.2 -.3 -.4

0 1 2 3 4 5 6 7 8 8 7.9 7.8 7.7 7.6 7.5 2010 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.2: Effect of Children Television on Cognitive Skills - Math Grades (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

36 DiD estimates and trends: Individual motivation for studying (0-1) .03 .02 .01 0 -.01 -.02 0 1 2 3 4 5 6 .85 .8 .75 .7

2010 2011 2012 2013 2014 2015

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.3: Effect of Children Television on Non-cognitive Skills - Motivation (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

37 DiD estimates and trends: Social motivation for studying (0-1) .02 .01 0 -.01

0 1 2 3 4 5 6 .55 .5 .45 .4 .35 .3

2010 2011 2012 2013 2014 2015

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.4: Effect of Children Television on Non-cognitive Skills - Motivation (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

38 DiD estimates and trends: Homework: frequency per week .1 .05 0 -.05 0 1 2 3 4 5 4.5 4

2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.5: Effect of Children Television on Time Use - Homework (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

39 DiD estimates and trends: Daily hours spent on TV .04 .02 0 -.02

0 1 2 3 4 1.26 1.24 1.22 1.2 1.18 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.6: Effect of Children Television on Time Use - TV (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

40 DiD estimates and trends: Daily hours spent on PC/videogames .04 .02 0 -.02 -.04 0 1 2 3 4 1.3 1.25 1.2 1.15 1.1 1.05 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.7: Effect of Children Television on Time Use - Gaming (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

41 DiD estimates and trends: Culture: frequency per week .05 0 -.05

0 1 2 3 4 1 .9 .8 .7 .6

2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.8: Effect of Children Television on Time Use - Culture (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

42 DiD estimates and trends: Sports: frequency per week .1 .05 0 -.05 -.1 0 1 2 3 4 2.35 2.3 2.25 2.2 2.15 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.9: Effect of Children Television on Time Use - Sports (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

43 DiD estimates and trends: Bullying: Harassment .005 0 -.005 -.01 -.015

0 1 2 3 4 5 6 .2 .15 .1

2010 2011 2012 2013 2014 2015

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.10: Effect of Children Television on Behavior - Harassment (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

44 DiD estimates and trends: Bullying: Social exclusion .06 .04 .02 0 -.02

0 1 2 3 4 5 6 .5 .4 .3 .2

2010 2011 2012 2013 2014 2015

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.11: Effect of Children Television on Behavior - Exclusion (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

45 DiD estimates and trends: Bullying: Assault .01 0 -.01 -.02 -.03 0 1 2 3 4 5 6 .25 .2 .15 .1 .05 2010 2011 2012 2013 2014 2015

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.12: Effect of Children Television on Behavior - Assault (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

46 DiD estimates and trends: Family input: Encyclopedia .05 0 -.05 -.1 0 1 2 3 4 5 6 7 8 .75 .7 .65 .6 .55 .5

2010 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.13: Effect of Children Television on Educational Inputs - Encyclopedia (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

47 DiD estimates and trends: Family input: Number of books 20 15 10 5 0

0 1 2 3 4 5 6 7 8 100 90 80 70 60 2010 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure B.14: Effect of Children Television on Educational Inputs - Books (grade 5) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 3, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

48 C Grade 6

C.1 Treatment variation

Table C.1: Years of Exposure to DTT by Region and Year: Grade 6 (Age 12)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 3 4 5 6 7 8 9 9 Piedmont(West) 1 2 3 4 5 6 7 8 Piedmont(East) 0 1 2 3 4 5 6 7 Lombardy 0 1 2 3 4 5 6 7 Trentino 1 2 3 4 5 6 7 8 Veneto −1 0 1 2 3 4 5 6 Friuli-Venezia Giulia −1 0 1 2 3 4 5 6 Emilia-Romagna (West) 0 1 2 3 4 5 6 7 Emilia-Romagna (East) −1 0 1 2 3 4 5 6 Liguria −1 0 1 2 3 4 5 6 Tuscany −1 0 1 2 3 4 5 6 Umbria −1 0 1 2 3 4 5 6 Marche −2 −1 0 1 2 3 4 5

South: Lazio(North) −1 0 1 2 3 4 5 6 Lazio(South) 1 2 3 4 5 6 7 8 Abruzzo −2 −1 0 1 2 3 4 5 Molise −2 −1 0 1 2 3 4 5 Campania 0 1 2 3 4 5 6 7 Basilicata −2 −1 0 1 2 3 4 5 Apulia −2 −1 0 1 2 3 4 5 Calabria −2 −1 0 1 2 3 4 5 Sicily −2 −1 0 1 2 3 4 5 Sardinia(North) 2 3 4 5 6 7 8 9 Sardinia(South) 3 4 5 6 7 8 9 9

49 Table C.2: Age of Exposure to DTT by Region and Year: Grade 6 (Age 12)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 9 8 7 6 5 4 3 2 Piedmont(West) 11 10 9 8 7 6 5 4 Piedmont(East) 12 11 10 9 8 7 6 5 Lombardy 12 11 10 9 8 7 6 5 Trentino 11 10 9 8 7 6 5 4 Veneto 13 12 11 10 9 8 7 6 Friuli-Venezia Giulia 13 12 11 10 9 8 7 6 Emilia-Romagna (West) 12 11 10 9 8 7 6 5 Emilia-Romagna (East) 13 12 11 10 9 8 7 6 Liguria 13 12 11 10 9 8 7 6 Tuscany 14 13 12 11 10 9 7 6 Umbria 14 13 12 11 10 9 8 7 Marche 14 13 12 11 10 9 8 7

South: Lazio(North) 14 13 12 11 10 9 8 7 Lazio(South) 11 10 9 8 7 6 5 4 Abruzzo 14 13 12 11 10 9 8 7 Molise 14 13 12 11 10 9 8 7 Campania 12 11 10 9 8 7 6 5 Basilicata 14 13 12 11 10 9 8 7 Apulia 14 13 12 11 10 9 8 7 Calabria 14 13 12 11 10 9 8 7 Sicily 14 13 12 11 10 9 8 7 Sardinia(North) 10 9 8 7 6 5 4 3 Sardinia(South) 9 8 7 6 5 4 3 2

50 C.2 Identification

Results in this section are based on the following specification:

Yi6ry = αr + αy + β0 T (0) + β1 T (1) + β2 T (2)+β3 T (3) + δ4 T (4) + ∆ Xi6ry + i6ry (4)

Table C.3 shows identifying variation for grade 6 students, who are supposed to be 12, including age of first exposure to DTT (A) and years of exposure to DTT (T). Green cells denote control observations, used to run a placebo test corresponding to β0 in equation4. 34 Red cells identify the last lag, δ4, in specification4.

Region switched to DTT in: 2009 2010 2011 2012 A/Y 2010 → A: 11; T: 1 A: 12; T: 0 A: 13; T: -1 A: 14; T: -2 A/Y 2011 → A: 10; T: 2 A: 11; T: 1 A: 12; T: 0 A: 13; T: -1 A/Y 2012 → A: 09; T: 3 A: 10; T: 2 A: 11; T: 1 A: 12; T: 0 A/Y 2013 → A: 08; T: 4 A: 09; T: 3 A: 10; T: 2 A: 11; T: 1 Table C.3: Identifying Treatment Variation (Grade 6)

34Due to data availability, it is only possible to estimate up to 4 lags for this grade.

51 C.3 Results

DiD estimates and trends: Italian school grade .04 .02 0 -.02 -.04 0 1 2 3 4 6.8 6.7 6.6 6.5 6.4 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.1: Effect of Children Television on Cognitive Skills - Italian Grades (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

52 DiD estimates and trends: Math school grade .04 .02 0 -.02 -.04 0 1 2 3 4 7 6.8 6.6 6.4 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.2: Effect of Children Television on Cognitive Skills - Math Grades (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

53 DiD estimates and trends: Social motivation for studying (0-1) .01 .005 0 -.005 0 1 2 3 4 .32 .3 .28 .26 .24 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.3: Effect of Children Television on Non-cognitive Skills - Motivation (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

54 DiD estimates and trends: Daily hours spent on TV .06 .04 .02 0 -.02 0 1 2 3 4 1.4 1.38 1.36 1.34 1.32 1.3 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.4: Effect of Children Television on Time Use - TV (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

55 DiD estimates and trends: Daily hours spent on PC/videogames .02 0 -.02 -.04

0 1 2 3 4 1.5 1.4 1.3 1.2 1.1 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.5: Effect of Children Television on Time Use - Gaming (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

56 DiD estimates and trends: Culture: frequency per week .06 .04 .02 0 -.02 -.04 0 1 2 3 4 1 .95 .9 .85 .8 .75

2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.6: Effect of Children Television on Time Use - Culture (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

57 DiD estimates and trends: Sports: frequency per week .05 0 -.05 -.1 0 1 2 3 4 2.35 2.3 2.25 2.2 2.15 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.7: Effect of Children Television on Time Use - Sports (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

58 DiD estimates and trends: Bullying: Harassment .01 .005 0

0 1 2 3 .065 .06 .055 .05 .045 .04 2010 2011 2012

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.8: Effect of Children Television on Behavior - Harassment (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

59 DiD estimates and trends: Bullying: Social exclusion .015 .01 .005 0 -.005 0 1 2 3 .22 .21 .2 .19 .18 .17 2010 2011 2012

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.9: Effect of Children Television on Behavior - Exclusion (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

60 DiD estimates and trends: Bullying: Assault .006 .004 .002 0 -.002 0 1 2 3 .085 .08 .075 .07 .065 2010 2011 2012

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.10: Effect of Children Television on Behavior - Assault (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

61 DiD estimates and trends: Family input: Encyclopedia .01 0 -.01 -.02 0 1 2 3 4 .82 .8 .78 .76 .74

2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.11: Effect of Children Television on Educational Inputs - Encyclopedia (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

62 DiD estimates and trends: Family input: Number of books 6 4 2 0 -2

0 1 2 3 4 110 100 90 80 70 2010 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure C.12: Effect of Children Television on Educational Inputs - Books (grade 6) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 4, including one lead (β0) and four lags (βt and δt). Standard errors are clustered at the DTT Region.

63 64 D Grade 8

D.1 Treatment variation

Table D.1: Years of Exposure to DTT by Region and Year: Grade 8 (Age 14)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 3 4 5 6 7 8 9 10 Piedmont(West) 1 2 3 4 5 6 7 8 Piedmont(East) 0 1 2 3 4 5 6 7 Lombardy 0 1 2 3 4 5 6 7 Trentino 1 2 3 4 5 6 7 8 Veneto −1 0 1 2 3 4 5 6 Friuli-Venezia Giulia −1 0 1 2 3 4 5 6 Emilia-Romagna (West) 0 1 2 3 4 5 6 7 Emilia-Romagna (East) −1 0 1 2 3 4 5 6 Liguria −1 0 1 2 3 4 5 6 Tuscany −1 0 1 2 3 4 5 6 Umbria −1 0 1 2 3 4 5 6 Marche −2 −1 0 1 2 3 4 5

South: Lazio(North) −1 0 1 2 3 4 5 6 Lazio(South) 1 2 3 4 5 6 7 8 Abruzzo −2 −1 0 1 2 3 4 5 Molise −2 −1 0 1 2 3 4 5 Campania 0 1 2 3 4 5 6 7 Basilicata −2 −1 0 1 2 3 4 5 Apulia −2 −1 0 1 2 3 4 5 Calabria −2 −1 0 1 2 3 4 5 Sicily −2 −1 0 1 2 3 4 5 Sardinia(North) 2 3 4 5 6 7 8 9 Sardinia(South) 3 4 5 6 7 8 9 10

65 Table D.2: Age of Exposure to DTT by Region and Year: Grade 8 (Age 14)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 11 10 9 8 7 6 5 4 Piedmont(West) 13 12 11 10 9 8 7 6 Piedmont(East) 14 13 12 11 10 9 8 7 Lombardy 14 13 12 11 10 9 8 7 Trentino 13 12 11 10 9 8 7 6 Veneto 15 14 13 12 11 10 9 8 Friuli-Venezia Giulia 15 14 13 12 11 10 9 8 Emilia-Romagna (West) 14 13 12 11 10 9 8 7 Emilia-Romagna (East) 15 14 13 12 11 10 9 8 Liguria 15 14 13 12 11 10 9 8 Tuscany 16 15 14 13 12 11 9 8 Umbria 16 15 14 13 12 11 10 9 Marche 16 15 14 13 12 11 10 9

South: Lazio(North) 16 15 14 13 12 11 10 9 Lazio(South) 13 12 11 10 9 8 7 6 Abruzzo 16 15 14 13 12 11 10 9 Molise 16 15 14 13 12 11 10 9 Campania 14 13 12 11 10 9 8 7 Basilicata 16 15 14 13 12 11 10 9 Apulia 16 15 14 13 12 11 10 9 Calabria 16 15 14 13 12 11 10 9 Sicily 16 15 14 13 12 11 10 9 Sardinia(North) 12 11 10 9 8 7 6 5 Sardinia(South) 11 10 9 8 7 6 5 4

66 D.2 Identification

Results in this section are based on the following specification:

Yi8ry = αr + αy + β0 T (0) + β1 T (1) + β2 T (2) + β3 T (3) + δ4 T (4) (5)

+δ5 T (5)+δ6 T (6) + δ7 T (7) + δ8 T (8) + ∆ Xi8ry + i8ry

Table B.3 shows identifying variation for grade 8 students, who are supposed to be 14, including age of first exposure to DTT (A) and years of exposure to DTT (T). Green cells denote control observations, used to run a placebo test corresponding to β0 in equation5. Red cells identify the last lag, δ8, in specification5.

Region switched to DTT in: 2009 2010 2011 2012 A/Y 2010 → A: 13; T: 1 A: 14; T: 0 A: 15; T: -1 A: 16; T: -2 A/Y 2011 → A: 12; T: 2 A: 13; T: 1 A: 14; T: 0 A: 15; T: -1 A/Y 2012 → A: 11; T: 3 A: 12; T: 2 A: 13; T: 1 A: 14; T: 0 A/Y 2013 → A: 10; T: 4 A: 11; T: 3 A: 12; T: 2 A: 13; T: 1 A/Y 2014 → A: 9; T: 5 A: 10; T: 4 A: 11; T: 3 A: 12; T: 2 A/Y 2015 → A: 8; T: 6 A: 09; T: 5 A: 10; T: 4 A: 11; T: 3 A/Y 2016 → A: 7; T: 7 A: 08; T: 6 A: 09; T: 5 A: 10; T: 4 A/Y 2017 → A: 6; T: 8 A: 07; T: 7 A: 08; T: 6 A: 9; T: 5 Table D.3: Identifying Treatment Variation (Grade 8)

67 D.3 Results

DiD estimates and trends: Score in Italian (raw) 1 .5 0 -.5 -1

0 1 2 3 4 5 6 7 8 75 70 65 60

2010 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure D.1: Effect of Children Television on Cognitive Skills - Italian Test Scores (grade 8) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 5, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

68 DiD estimates and trends: Score in math (raw) 2 1 0 -1 -2

0 1 2 3 4 5 6 7 8 65 60 55 50 2010 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure D.2: Effect of Children Television on Cognitive Skills - Math Test Scores (grade 8) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 5, including one lead (β0) and eight lags (βt and δt). Standard errors are clustered at the DTT Region.

69 E Grade 10

E.1 Treatment variation

Table E.1: Years of Exposure to DTT by Region and Year: Grade 10 (Age 16)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 3 4 5 6 7 8 9 10 Piedmont(West) 1 2 3 4 5 6 7 8 Piedmont(East) 0 1 2 3 4 5 6 7 Lombardy 0 1 2 3 4 5 6 7 Trentino 1 2 3 4 5 6 7 8 Veneto −1 0 1 2 3 4 5 6 Friuli-Venezia Giulia −1 0 1 2 3 4 5 6 Emilia-Romagna (West) 0 1 2 3 4 5 6 7 Emilia-Romagna (East) −1 0 1 2 3 4 5 6 Liguria −1 0 1 2 3 4 5 6 Tuscany −1 0 1 2 3 4 5 6 Umbria −1 0 1 2 3 4 5 6 Marche −2 −1 0 1 2 3 4 5

South: Lazio(North) −1 0 1 2 3 4 5 6 Lazio(South) 1 2 3 4 5 6 7 8 Abruzzo −2 −1 0 1 2 3 4 5 Molise −2 −1 0 1 2 3 4 5 Campania 0 1 2 3 4 5 6 7 Basilicata −2 −1 0 1 2 3 4 5 Apulia −2 −1 0 1 2 3 4 5 Calabria −2 −1 0 1 2 3 4 5 Sicily −2 −1 0 1 2 3 4 5 Sardinia(North) 2 3 4 5 6 7 8 9 Sardinia(South) 3 4 5 6 7 8 9 10

70 Table E.2: Age of Exposure to DTT by Region and Year: Grade 10 (Age 16)

2010 2011 2012 2013 2014 2015 2016 2017 North: Aosta Valley 13 12 11 10 9 8 7 6 Piedmont(West) 15 14 13 12 11 10 9 8 Piedmont(East) 16 15 14 13 12 11 10 9 Lombardy 16 15 14 13 12 11 10 9 Trentino 15 14 13 12 11 10 9 8 Veneto 17 16 15 14 13 12 11 10 Friuli-Venezia Giulia 17 16 15 14 13 12 11 10 Emilia-Romagna (West) 16 15 14 13 12 11 10 9 Emilia-Romagna (East) 17 16 15 14 13 12 11 10 Liguria 17 16 15 14 13 12 11 10 Tuscany 18 17 16 15 14 13 11 10 Umbria 18 17 16 15 14 13 12 11 Marche 18 17 16 15 14 13 12 11

South: Lazio(North) 18 17 16 15 14 13 12 11 Lazio(South) 15 14 13 12 11 10 9 8 Abruzzo 18 17 16 15 14 13 12 11 Molise 18 17 16 15 14 13 12 11 Campania 16 15 14 13 12 11 10 9 Basilicata 18 17 16 15 14 13 12 11 Apulia 18 17 16 15 14 13 12 11 Calabria 18 17 16 15 14 13 12 11 Sicily 18 17 16 15 14 13 12 11 Sardinia(North) 14 13 12 11 10 9 8 7 Sardinia(South) 13 12 11 10 9 8 7 6

71 E.2 Identification

Results in this section are based on the following specification:

Yi10ry = αr + αy + β0 T (0) + β1 T (1) + β2 T (2) + β3 T (3) + δ4 T (4) (6)

+δ5 T (5)+δ6 T (6) + δ7 T (7) + δ8 T (8) + ∆ Xi10ry + i10ry

Table E.3 shows identifying variation for grade 10 students, who are supposed to be 16, including age of first exposure to DTT (A) and years of exposure to DTT (T). Green cells denote control observations, used to run a placebo test corresponding to β0 in equation6. Red cells identify the last lag, δ8, in specification6.

Region switched to DTT in: 2009 2010 2011 2012 A/Y 2011 → A: 14; T: 2 A: 15; T: 1 A: 16; T: 0 A: 17; T: -1 A/Y 2012 → A: 13; T: 3 A: 14; T: 2 A: 15; T: 1 A: 16; T: 0 A/Y 2013 → A: 12; T: 4 A: 13; T: 3 A: 14; T: 2 A: 15; T: 1 A/Y 2014 → A: 11; T: 5 A: 12; T: 4 A: 13; T: 3 A: 14; T: 2 A/Y 2015 → A: 10; T: 6 A: 11; T: 5 A: 12; T: 4 A: 13; T: 3 A/Y 2016 → A: 09; T: 7 A: 10; T: 6 A: 11; T: 5 A: 12; T: 4 A/Y 2017 → A: 08; T: 8 A: 09; T: 7 A: 10; T: 6 A: 11; T: 5 Table E.3: Identifying Treatment Variation (Grade 10)

72 E.3 Results

DiD estimates and trends: Academic track .04 .02 0 -.02

1 2 3 4 5 6 7 .6 .55 .5 .45 .4

2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure E.1: Effect of Children Television on Educational Choices - Academic Track Enrollment (grade 10) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 6, including eight lags (βt and δt). Standard errors are clustered at the DTT Region.

73 DiD estimates and trends: Italian school grade .1 .05 0 -.05 1 2 3 4 5 6 7 6.4 6.2 6 5.8 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure E.2: Effect of Children Television on Cognitive Skills - Italian Grades (grade 10) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 6, including eight lags (βt and δt). Standard errors are clustered at the DTT Region.

74 DiD estimates and trends: Math school grade .1 .05 0 -.05 1 2 3 4 5 6 7 6.2 6 5.8 5.6 5.4 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure E.3: Effect of Children Television on Cognitive Skills - Math Grades (grade 10) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 6, including eight lags (βt and δt). Standard errors are clustered at the DTT Region.

75 DiD estimates and trends: Individual motivation for studying (0-1) .01 0 -.01 -.02 1 2 3 4 5 .66 .64 .62 .6 .58 2011 2012 2013 2014 2015

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure E.4: Effect of Children Television on Non-cognitive Skills - Motivation (grade 10) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 6, including eight lags (βt and δt). Standard errors are clustered at the DTT Region.

76 DiD estimates and trends: Usefulness of Italian .002 .001 0 -.001 -.002 -.003

1 2 3 .8 .79 .78 .77 .76 2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure E.5: Effect of Children Television on Non-cognitive Skills - Usefulness of Italian (grade 10) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 6, including eight lags (βt and δt). Standard errors are clustered at the DTT Region.

77 DiD estimates and trends: Usefulness of math .005 0 -.005 -.01 -.015 1 2 3 .52 .51 .5 .49 .48

2011 2012 2013

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure E.6: Effect of Children Television on Non-cognitive Skills - Usefulness of Math (grade 10) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 6, including eight lags (βt and δt). Standard errors are clustered at the DTT Region.

78 DiD estimates and trends: Family input: Encyclopedia .01 .005 0 -.005 -.01 -.015 1 2 3 4 5 6 7 .9 .85 .8 .75 .7

2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure E.7: Effect of Children Television on Educational Inputs - Encyclopedia (grade 10) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 6, including eight lags (βt and δt). Standard errors are clustered at the DTT Region.

79 DiD estimates and trends: Family input: Number of books 10 0 -10 -20 1 2 3 4 5 6 7 180 160 140 120 100 80 2011 2012 2013 2014 2015 2016 2017

Switch 2009 Switch 2010 Switch 2011 Switch 2012

Figure E.8: Effect of Children Television on Educational Inputs - Books (grade 10) Note: The lower panel shows trends in the outcome variable by switch-off year. Dashed lines indicate periods before transition to DTT. The upper panel reports point estimates and 95% confidence interval from specification 6, including eight lags (βt and δt). Standard errors are clustered at the DTT Region.

80