Abstract
We exploit supply-driven heterogeneity in the expansion of cable television across Norwegian municipalities to identify developmental effects of commercial television exposure during childhood. We find that higher exposure to commercial television reduces cognitive ability and high school graduation rates for boys. The effects appear to be driven by consumption of light television entertainment crowding out more cognitively stimulating activities. Point estimates suggest that the effects are most negative for boys from more educated families. We find no effect on high school completion for girls, pointing to the growth of noneducational media as a factor in the widening educational gender gap.
I. Introduction
Since television was introduced to a large audience around the mid-20th century, its effects have been debated. A widespread concern has been that television encourages a particularly passive form of engagement and thus may be damaging to intellectual development. This view has been argued by social commentators (for example, Postman 1985), but has also received support in professional circles: American pediatricians have concluded that television affects children negatively and recommend limiting children’s television time (American Academy of Pediatrics 2001; Strasburger, Jordan, and Donnerstein 2010). The complete opposite view has also found supporters. Johnson (2006), for example, argues that popular culture, including television, has become more complex and intellectually demanding over time and that this gives beneficial cognitive payoffs.
Cognitive skills are essential for individual (Griliches and Mason 1972; Cunha and Heckman 2007) as well as aggregate economic outcomes (Hanushek and Woessmann 2008). After a long period with generally rising IQ scores in industrialized countries, recent empirical evidence indicates that this development may now have gone into reverse; see Dutton and Lynn (2015) for a recent overview of the literature.1 Since its introduction, television has spread all over the world, so its effects are also likely to be ubiquitous. If television viewing can indeed be harmful to the development of cognitive skills, this may be one explanation why IQ scores have stagnated or even begun to decline in many countries. If so, this is critical knowledge for policymakers and families alike.
The existing empirical evidence is not conclusive, however. In a review of the pediatrics literature, Jolin and Weller (2011) point out that statistical evaluations are often based on cross-sectional data, and they lament the lack of longitudinal research designs. There are some experimental studies focusing on short-term impacts, but the external validity of such studies is questionable; see Thakkar, Garrison, and Christiakis (2006) for a review. An influential study within the economics literature is Gentzkow and Shapiro (2008), who analyze the effect of the introduction of television in the United States in the 1940s and 1950s on standardized test scores. They find that, contrary to popular worries, television exposure during preschool age not only failed to lower test scores, but in fact raised reading and general knowledge scores for socially disadvantaged groups. On the basis of this effect heterogeneity, they conclude that “the cognitive effects of television exposure depend critically on the educational value of the alternative activities that it crowds out” (Gentzkow and Shapiro 2008, p. 282). As most societies have undergone large changes in their educational and home environments since the early postwar era, this makes it pertinent to reexamine the issue. The impacts of television exposure also clearly depend on the contents of the programs that children are watching. Kearney and Levine (2015) examine the effects of the introduction of an explicitly educational children’s show—Sesame Street—into a television programming space dominated by entertainment shows. They find support for the positive educational effect of Sesame Street, in particular for boys and children from disadvantaged backgrounds.
In this paper, we examine how a geographically staggered expansion of access to commercial cable television in Norway during the 1980s and 1990s affected children’s cognitive skill developments and subsequent high school graduation rates. As we explain in more detail below, the introduction of cable television in Norway implied access to a wide range of channels broadcast by satellites, and it turned out to have huge impacts both on children’s overall exposure to television and on the type of programs they watched. Whereas the state broadcaster, which held a legal monopoly on television broadcasts until 1981, offered only a single TV channel based on a clear educational mandate, the new commercial channels were almost entirely dominated by light entertainment without any educational content. We provide empirical support for the view that the expansion of commercial cable television indeed had negative effects on men’s cognitive skills—as measured by intelligence test scores at the time of military service enrollment—and that it also reduced their high school completion rates. For women, we find no significant effects on high school completion rates. Data on intelligence test scores exist only for men.
When television was deregulated in Norway in the beginning of the 1980s, it became legal to forward television signals broadcast by satellites in local cable networks, and a large rollout of cable networks was initiated. We argue that the growth in cable networks was a supply-led development driven by geographical factors and settlement patterns, and the key explanatory variable in our analysis is the time-varying cable television coverage rate at the municipality level. Because residence decisions may be endogenous to cable television coverage, we either include family fixed effects in our regression models or use the future coverage in an individual’s municipality of birth to instrument for actual exposure.
Our results indicate that one year of living in a municipality with full coverage of cable television during childhood and adolescence lowers ability test scores of young men by 1.2 percent standard deviations, corresponding to 0.18 IQ points. It reduces the high school completion rate for males by 0.4 percentage points. These effects are not huge, yet far from negligible. They appear to be driven by consumption of light television entertainment crowding out more cognitively stimulating activities such as reading. Point estimates suggest that the effects are most negative for boys from more educated families. For women, we do not find convincing evidence of a negative effect on high school completion; thus, a net effect of commercial TV expansion was to widen the educational gender gap. A policy lesson that potentially applies to other forms of light media entertainment as well is that although consumption of such entertainment is not necessarily harmful in itself, one should be alert about what activities it substitutes for, in particular for boys.
Our paper adds to a rapidly growing literature in economics on the effects of media. Television in particular has been found to impact outcomes as diverse as voter turnout in the United States (Gentzkow 2006), party choice in the United States and Russia (DellaVigna and Kaplan 2007; Enikolopov, Petrova, and Zhuravskaya 2011; Martin and Yurukoglu 2014), social capital in Indonesia (Olken 2009), fertility, women’s status and school enrollment in India (Jensen and Oster 2009), and divorce and fertility in Brazil (Chong and Ferrara 2009; La Ferrara, Chong, and Duryea 2012).
Our study also contributes to the literature on the educational gender gap. In most developed nations, the traditional male education premium has been completely reversed, such that girls now have higher high school graduation rates and higher tertiary education than boys (OECD 2014, 2015). In the United States, girls have had higher high school graduation rates than boys for most of the twentieth century (Goldin and Katz 2008), and from 1970 the gap has widened (DiPrete and Buchmann 2013; Murnane 2013). In the same period, girls also made gains relative to boys on standardized math and reading tests (Goldin, Katz, and Kuziemko 2006; Cho 2007) and grades (Fortin, Oreopoulos, and Phipps 2015). Previous research has found evidence that boys are more sensitive than girls with respect to both parental (Bertrand and Pan 2013; Riphahn and Schwientek 2015) and school/preschool inputs (Krueger 1999; Machin and McNally 2008; Chetty et al. 2011; Havnes and Mogstad 2011). This is supported by evidence showing that boys benefit more strongly than girls from early childhood interventions (Campbell et al. 2014; Conti, Heckman, and Pinto 2016) and are more impatient (Bettinger and Slonim 2007; Dohmen et al. 2010; Castillo et al. 2011; Golsteyn, Grönqvist, and Lindahl 2014).
II. Data and Institutional Background
Television was introduced in Norway in the 1960s. Until 1981, the state-controlled Norwegian Broadcasting Corporation held a legal monopoly on broadcasting in the country, and for most Norwegians, only a single TV channel was available. In December 1981, the newly elected government announced that 30 other agents would obtain broadcasting licenses the following year, thereby breaking the public monopoly. It then became legal to forward television signals broadcast by satellites in local cable networks. All such local cable networks had to register with the Post- and Telecommunications Authority. The legalization in 1981 initiated a large-scale rollout of local cable networks. Because of economies of scale in laying the necessary cables, the rollout took place primarily in densely populated areas (Norwegian Ministry of Culture 1995). Mandatory registration continued until 2004, at which point 40 percent of the population was covered. The aggregate evolution of the number of households covered is shown in Figure 1.
Share of Households with Cable Television, 1970–2005
The television data consist of the universe of Norwegian local cable networks up to 2004. They contain more than 11,000 unique networks, each with the number of households covered, the first date of operation, and the municipality. We combine this with data on the number of households in a municipality to obtain the yearly coverage rate in each municipality up to 2005. The maps in Figure 2 show three snapshots of coverage rates across the country between 1985 and 2005. They illustrate the considerable geographical disparities in the rollout process, with cable networks first established in the Oslo area and then expanded to other densely populated areas throughout the country. To arrive at individualized exposed-to-TV variables, we construct two variables for each person born between 1974 and 1987: the cumulative cable TV coverage over their first 18 years both in their municipality of actual residence at any time and in their municipality of birth. The resultant TV coverage variables thus vary from zero (for individuals who grew up in municipalities were no one had access to cable TV before they turned 18) to 18 (for individuals who grew up in municipalities with full coverage already from birth).
Cable Television Coverage by Municipality, 1985–2005
Unfortunately, we do not have the same kind of data on commercial television coverage by parabolic antennas. This is a minor problem in the present context, however, because parabolic antennas were providing only a small amount of television coverage in this period. When Statistics Norway started their media survey in 1991, only 5 percent of the households had a parabolic antenna. Since parabolic antennas and cable television were substitutes for each other, the presence of parabolic antennas will nevertheless tend to attenuate our estimated effects of commercial television access slightly, as we implicitly will assume that cable networks are the only providers of these channels.
Coinciding with the expansion of commercial television in Norway, there has been a marked change in the way young people spend their leisure time. Figure 3 illustrates some key findings from a sequence of time use surveys provided by Statistics Norway. Unsurprisingly, time spent watching television increased considerably for both boys and girls during the 1980s and 1990s. To a large extent, this substituted for reading, particularly among boys aged 16–24, for which time spent reading dropped by as much as 74 percent, from 39 minutes per day in 1980 to 10 minutes per day in 2000.2 The reason why we focus on those aged 16–24 here is that there is no time use data collected for younger persons. However, from 1991, there have been media surveys also covering those aged 9–18. While the media surveys do not allow us to examine the longer term trends, they confirm the picture of a considerable rise in time spent television watching and a corresponding decline in the time spent reading. We return to these data in Section VI, where we examine the relationship between cable TV access and media consumption more closely.
Minutes Spent during an Average Day on Television Watching (Panel A) and Reading (Panel B) by Boys (Solid Lines) and Girls (Dotted Lines)
Sources: For ages 16–24: Statistics Norway, time use surveys collected in 1970, 1980, 1990, 2000, and 2010. For ages 9–18: Statistics Norway, media surveys collected annually 1991–2004
The data we use in the statistical analysis of the effects of cable TV exposure are collected from different administrative registers and cover the complete population. The outcomes we focus on are cognitive ability score at age 18 for young men and a dichotomous variable indicating high school completion by age 21 (two years after the “normal” graduation age) for both men and women. Our measure of cognitive ability is based on data from the Norwegian Armed Forces. Before compulsory military service, all Norwegian men undergo an assessment of suitability. From 1969–1970, this assessment has included intelligence testing. All test-takers receive a score that is a composite of three tests, on arithmetic, word similarities and pattern recognition. In this paper we have standardized the test scores such that point estimates will be directly interpretable as percentage points of a standard deviation, though for expositional purposes we sometimes refer to the corresponding IQ scores.3
Figure 4 (Panel A) shows that there has indeed been a negative development on this test since the early 1990s, in line with previous findings reported by Sundet, Barlaug, and Torjussen (2004). Another notable phenomenon was a widening of the gender gap in high school completion, which increased by more than six percentage points (Panel B).
IQ of Norwegian Conscripts, 1992–2005 (Panel A) and Female–Male Gender Gap in High School Completion at 21 (Percentage Points) (Panel B)
Notes: The IQ test was normed on the test cohort of 1980, which provides the reference point (that is, 1980 mean = 100).
We match the two outcomes with data on parental characteristics and on the municipality of residence from the year of birth and onwards. The first cohort in our sample is the cohort born in 1974, which is the first year for which we have data on residence at birth. We include all subsequent cohorts until and including the one born in 1987, which turned 18 years old in 2005. The reason we stop at the 1987 birth cohort is that we do not have the data needed to compute total cable television coverage for later cohorts, as the mandatory registration of networks ended in 2004. To sum up, our sample consists of the birth cohorts 1974–1987, with outcomes measured in 1992–2005 (ability test scores) and 1995–2008 (high school completion). To allay worries concerning an increasing share of people with immigrant background taking the test, we include only individuals born in Norway to Norwegian-born parents.
Table 1 presents an overview of the data we use in this paper, with descriptive statistics for the two outcomes and for the various control variables we are going to use in the statistical analysis. In total, we have around 311,000 male and 294,000 female observations. On average, the persons in our data set had access to cable TV in only 3.45 years during their childhood and youth, but TV exposure increased considerably over time, from less than two exposure years on average for our first cohorts to more than five for our last (not shown in the table). Figure 5 provides a more detailed description of how the TV coverage variable is distributed within our analysis population. It shows that the typical coverage rates have been quite moderate, with approximately one-half of the population being attributed less than three years of exposure. Note that in our data, a given level of exposure can come about through different combinations of years with exposure and exposure intensity. For example, a cable TV coverage of 0.8 can come about by living one year in a municipality with 80 percent coverage, or by living four years in a municipality with 20 percent coverage.
Descriptive Statistics
The Distribution of Cable TV Exposure, Age 0–18
Notes: The first bar shows the fraction with no cable TV exposure, the second bar shows the fraction with zero to one coverage-year exposure, the third bar the fraction with one to two coverage-year exposure, and so on.
The control variables are defined at the individual or at the municipality × cohort levels, in both cases measured no later than the time of birth. In addition, we will in some cases use municipality characteristics measured in 1980 (prelegalization) interacted with time indicators. Individual characteristics comprise the educational attainment and earnings levels of both parents, whereas municipality characteristics cover socioeconomic factors, such as average education, employment, and earnings; see Table 1 for details.
III. Empirical Strategy
In order to identify the causal impacts of cable network expansion on the ability test score and schooling outcomes, we use the municipality × cohort specific coverage rate described in Section II as the key explanatory variable. The identifying assumption is then that the geographical rollout of cable TV was as good as randomly assigned with respect to other factors that could have generated local variations in the developments of cognitive ability and high school completion rates, conditional on the control variables we use in the analysis. We argue that we can rule out reverse causation in this case, as there is evidence that the cable network expansion during the 1980s and 1990s was solely supply side driven. Building cable networks required heavy investment in infrastructure and was only profitable in densely populated areas. Given that there was a large excess demand for cable TV everywhere, the actual expansion pattern was determined by economies of scale and physical/topological constraints. Participants from the supply side of the cable television market in this period have confirmed that it is hard to see any factors other than suppliers’ capacity and population density that had an impact on where networks were built, and the deregulation suddenly allowed suppliers to cater to a demand that had been present for a long time.4
In 1999, a government white paper concluded that “significant development of cable networks beyond today’s level will most likely not be profitable (Norwegian Ministry of Culture 1999, Ch. 2.2)” and that “one does not expect significant further development of cable utilities beyond today’s coverage of around 38 percent (Norwegian Ministry of Culture 1999, Ch. 2.3).” From Figure 1 we see that this assessment proved correct. The report cites topographical and physical barriers as reasons for why full coverage would not be possible.
As an empirical check on the supply-driven expansion pattern, we run a regression where we use the potential years of coverage (in the municipality of birth) as the dependent variable and a number of pre-expansion (1980) municipality population characteristics as explanatory variables, including their changes from 1970 to 1980. As can be seen from the results reported in Table 2, the variable that most clearly stands out as having affected cable television is the level of population density at the start of the expansion period. There does not appear to have been any clear-cut relationships with other municipality variables, such as average education and income. In particular, we find no indication whatsoever that the pattern of cable TV expansion was statistically associated with local levels and trends in cognitive ability up to 1980.
TV Exposure in Birth Municipality as Dependent Variable. Regression Results (OLS)
Although demand did not play a significant role for the rollout of cable TV in Norway, we cannot a priori rule out that some families have taken cable TV coverage into account in their residential decisions. Neither can we rule out that the (typically densely populated) areas selected for early expansion have been characterized by different postexpansion developments in ability scores and schooling outcomes than other areas for reasons unrelated to the cable television expansion. Such differential developments may have come about through selective migration of families to and from densely populated areas or through different developments of the learning environments in rural and urban districts. Hence, an important element of our empirical strategy is to control for geographically differentiated developments along these lines, and also to examine the robustness of our results with respect to the exact way in which this is done.
To empirically assess the impacts of cable TV coverage, we use linear regression models with the two outcomes (cognitive ability score and a dichotomous indicator for high school completion) as dependent variables and the number of years with cable TV coverage until age 18 as the central explanatory variable. Throughout this paper our identification strategy will be based on two alternative approaches. The first is a control function strategy, where we use a wide range of control variables to achieve identification. To account for selective migration decisions, this approach relies on an instrumental variables strategy, where we use the TV coverage in the municipality of birth as an instrument for actual TV coverage. The second approach instead relies on family fixed effects, implying that we essentially compare brothers and sisters who to varying degrees have been exposed to cable TV. For this model, we use actual TV coverage as the central explanatory variable, as selective migration decisions will have been equally selective for all offspring belonging to the same family.
We now describe the baseline versions of the two models in more detail. We will return to a number of extensions/modifications of these models later on in order to examine robustness and mechanisms. Let i be a subscript for individuals, m for municipality of birth, and t for year of birth. We can then write the second step regression in our instrumental variables (IV) model as:
(1)
where yi is an individual outcome measured at age 18 or age 21, 
is predicted cumulative cable TV exposure during the first 18 living years, δm is a birth-municipality fixed effect, θt is a birth-year fixed effects, and pi is a vector of parental characteristics. The predicted cable TV exposure is the linear prediction from a first-stage equation where actual TV exposure (based on correctly assigned municipality in every year) is the left-hand-side variable, and the corresponding TV exposure in the municipality of birth is the right-hand-side variable, together with the other controls included in Equation 1. An important point to note is that all the explanatory variables in Equation 1 are predetermined at birth or earlier. The birth-municipality fixed effects are included to control for any time-invariant factors that vary between municipalities in a way that correlates with cable TV expansion. The birth-year fixed effects are included to control for any common trends. The vector of parental characteristics is included to control for any sorting of families through selective migration, as well as to improve efficiency. It includes indicator variables for both parents’ incomes and educational attainments, measured in the offspring’s birth year.5
Let j be a subscript for family. The family fixed effects (FFE) model can then be described as follows:
(2)
where TV is actual exposure, and bi is a vector of birth order fixed effects. The latter are included to avoid birth order effects to contaminate our estimate of β, as it will almost always be the case that the younger sibling has been exposed to at least as much TV at age 18 as the older sibling. Since we analyze boys and girls separately, the FFE model can only use data from families with at least two sons or two daughters, respectively.
IV. Main Results
We present the main results from the baseline instrumental variables (IV) and family fixed effects (FFE) models in Table 3. A first point to note is that the results are remarkably similar for the IV and FFE models, despite that they rely on different sources of identification and also involve different data sets (the number of observations is 60 percent lower in the FFE models than in the IV models). With the standardized cognitive ability score as the outcome, both models give an estimated coefficient on our television exposure measure of −0.012, that is, a negative effect of exposure of 1.2 percent of a standard deviation. Since the IQ scale has a standard deviation of 15, this estimated effect of one additional year of full television coverage would roughly correspond to a reduction in IQ of 0.18 points. By comparison, Brinch and Galloway (2012) estimate the effect of one year of schooling to be around 3.7 IQ points using an education reform in Norway in the 1960s. Likewise Carlsson et al. (2015) use a quasi-experimental setting in Sweden to exploit variation in test-taking date for young Swedish males preparing for military service. Their results imply that one year of schooling raises crystallized (synonyms and technical comprehension) test scores by around 20 percent of a standard deviation, corresponding to around 3 IQ points, but they found no effect on fluid (spatial and logic) intelligence tests. Taken at face value, our estimate thus indicates that, for example, 10 years of full cable TV coverage has a negative impact on cognitive ability (−1.8 IQ) comparable to around half a year’s schooling.
Main Results. Instrumental Variables and Family Fixed Effects Estimates of the Effect of One Additional Year with Cable Television Coverage
Moving on to high school completion, our results indicate that one extra year of cable TV coverage reduces the probability of high school completion for men by age 21 by around 0.4 percentage points. To again put the results in perspective, we note that the estimated effect on high school completion of one year of cable TV exposure constitute 1/12 of the corresponding statistical association between one additional year of parental education and high school completion, as reported by Bratsberg, Raaum, and Røed (2012, Figure 5). For women, we find no effect at all on high school completion, regardless of model specification. This is consistent with the research showing greater importance of the home and school inputs for boys than for girls. Boys’ higher impatience may also make them more susceptible to the temptation of easy entertainment at the expense of more cognitively challenging activities.6
V. Robustness
In this section, we evaluate the robustness of our findings with respect to the sources of identification—in terms of the way we allow for differentiated time trends in the two outcomes—and with respect to the functional form of the relationship between TV exposure and outcomes.
A. Differentiated Time Trends
Although we are confident that the expansion of cable television was supply side–driven, and thus not causally affected by factors determining local trends in cognitive ability, we have emphasized that spurious correlations cannot be ruled out. We now examine our findings’ robustness with respect to the inclusion of additional sets of control variables that are designed to capture differential time trends in the two outcome variables. In the IV model, these variables serve the purpose of controlling for any selective migration unaccounted for by parental characteristics (pi), as well as for developments of local learning environments that are spuriously correlated with cable TV expansion, whereas in the FFE model they only serve the latter purpose. We add three types of differentiated trend controls. The first is a set of time-varying municipality characteristics included to account for differential local trends in peer environments that potentially vary across municipalities; that is,
(3)
The vector of municipality characteristics includes municipality-level average education, male and female income, and male and female employment rates (five variables). These are all measured in the year of each cohort’s birth; hence, they are not absorbed by the municipality fixed effects (which are common for all cohorts).
The second type of trend control is a vector of county \times birth year fixed effects, entered in the form of 247 additional dummy variables:
(4)
Obviously we cannot use municipality–year fixed effects in our models because that would totally absorb the effects of interest and induce a multicollinearity problem. However, by including regional birth-year fixed effects at a somewhat higher geographical level, we can control for time variations that are common within regions. There are 19 counties in Norway, and each county covers approximately 23 municipalities on average. A somewhat unfortunate consequence of using county–year fixed effects is that we effectively eliminate the influence of observations from Oslo, the capital city, which is by far the largest municipality and forms its own county.
Finally, we add in a vector of prereform (1980) birth-municipality characteristics interacted with birth-year dummy variables; that is,
(5)
The prereform municipality characteristics include variables capturing the average education, income, and employment rates in the adult population, as well as population density and average ability level among men; see Table 1 for details. This gives us 91 additional covariates, which are included in the manner of Duflo (2001) to control for unobserved local trends in outcomes that vary systematically with factors that potentially influenced (or correlated spuriously with) network expansion. Trends that differ according to initial population density are included here for the reason that population density was a particularly important factor behind the cable network expansion.7 And trends that differ according to the initial average ability level are included to account for possible influences of mean-reversion tendencies in the data.
By including all these additional trend terms in our model, we run the risk of “overcontrolling” and thus erroneously attribute some of the true impacts of cable TV to other factors, as some of the trends we control for may have been endogenously caused by differential exposure to commercial TV. For example, this could be the case if the true effects of exposure are heterogeneous or nonlinear, such that they are not appropriately captured by our single TV exposure variable, but instead soaked up by the heterogeneous trend terms. We nevertheless think of this as a useful exercise in order to assess robustness with respect to alternative identification sources. We present the results in Table 4. The point estimates vary somewhat from model to model, with a pattern of declining effects as we allow for more differentiated local trends. Although this suggests that we should be a bit careful about interpreting the point estimates, we view the main results as rather robust. None of the qualitative conclusions from the previous section need to be modified, although the statistical significance for the FFE model do evaporate in the model with all trend terms included simultaneously. It worth noting that while point estimates differ across the different model specifications, the estimated standard errors are almost the same. The reason for this is that although the added controls soak up variation in the TV coverage variable—leading to larger standard errors—they also contribute to a drop in the overall residual variance—leading to smaller standard errors. In our case, these two forces largely cancel out.
Robustness with Respect to Differentiated Time Trends. Instrumental Variables and Family Fixed Effects Estimates of the Effect of One Additional Year with Cable Television Coverage
B. Functional Form
We now turn to the issue of functional form specification. So far we have assumed that the effects of cable TV exposure operate in a linear fashion, such that the marginal effect of an additional exposure year is the same regardless of initial exposure. Although this appears to be a rather restrictive—and poorly justified—assumption, we will argue that it is defendable on the grounds that we need to exploit the admittedly small independent variation that we have in TV exposure as efficiently as possible. However, it accentuates the need for checking whether this somewhat arbitrary functional form restriction is critical for our findings. Hence, we now relax the linearity assumption, and instead include cable TV exposure in our models in the form of a series of dummy variables, representing [0–2), [2–4), [4–6), [6–8), [8–10), [10–12), [12–14), [14–16), and [16–18] years of exposure, respectively. Otherwise, the estimated models are exactly as in the baseline models (Equations 1 and 2) described in Section III.
We present the results from this exercise graphically; see Figure 6. While there are some differences between the IV and FFE models, the main message coming out of these results is that the identified negative impacts for boys are indeed monotonously increasing in the exposure time, and the linearity assumption is not critical for our findings. For girls, the FFE model now actually indicates a positive impact on high school completion of TV exposure. Such an effect may be rationalized directly by the negative impact on boys, as poorer school performance among boys in areas with high coverage may have contributed to a lowering of the overall high school passing standards and thus reduced the degree of competition for girls. However, given that these results are not confirmed in the IV estimation, and also that only one of the FFE coefficients is statistically significant, we should probably not put too much emphasis on this result.
Instrumental Variables and Family Fixed Effects Estimates of the Effect of Years with Cable Television Coverage (with 95 percent confidence intervals)
Notes: All models include birth-year fixed effects. The IV models also include municipality fixed effects and parental characteristics. The FFE models include family fixed and birth-order fixed effects. The confidence intervals are computed based on standard errors clustered on municipality.
VI. Mechanism
Why does access to commercial cable TV reduce the cognitive ability and educational performance of young boys? And why do we not see a similar negative effect on educational performance for young girls? A simple answer to the first question is that large doses of the kind of light entertainment that commercial TV offers are not particularly cognitively stimulating. However, this does not explain why girls are not negatively affected.
In this section, we take a closer look at what auxiliary data sources on youths’ actual time use can tell us about the relationship between cable TV access, TV viewing, and the alternative activity of reading. In doing this, we also examine the role of socioeconomic background, as we suspect that cable TV access may affect offspring in different types of families differently, both in terms of its impact on the volume of TV consumption and in terms of the kinds of activities that it substitutes. Finally, we examine the extent to which there are “critical ages” at which access to cable TV has a particularly large (or small) influence on cognitive developments.
A. TV Consumption and the Role of Socioeconomic Status
Access to commercial television channels can affect the nature of television consumption in two ways: by affecting the total time spent watching TV and by affecting the type of programs being watched. The Norwegian Broadcasting Corporation (NRK), which held a legal monopoly until 1981, has had and continues to have a broad public service mandate. In their articles of association, it is stated that “The purpose of the NRK’s overall public media services is to meet democratic, social, and cultural needs in society” (NRK 1996). Further, “[t]he NRK should promote public debate,” “offer services which can be a source of inspiration, reflection, experience and knowledge through programs of high quality,” and “contribute to public education and learning.” The new channels that arrived with cable television had no public service mandate and no regulation of content other than pornography and violence (Regulations Relating to Broadcasting 1997). Their program profiles were indeed markedly different. According to a comparative analysis undertaken in 1993, the contents of the main Scandinavian cable channels in Norway (TV Norge and TV3) were 75 percent entertainment and 10 percent advertisements, while that of the NRK was almost 40 percent news, documentaries, science, nature, or similar (NRK 1993), with no advertisements. It is also clear that the new channels were watched—in 1992, the NRK’s share of total viewing time was down to 64 percent (MMI 1992).
In this subsection, we use auxiliary data from media surveys to shed light on how the introduction of cable TV influenced media habits among offspring from families with different socioeconomic status (SES). Since 1991 statistics Norway has undertaken surveys of the population’s media habits. An effort is made to obtain responses from kids as young as nine years old; thus, for the age group 9–18 we can get a quite good picture of how having a cable connection correlates with actual television watching in the period from 1991 onwards. When it comes to television, respondents are asked both about what type of connection they have and about how much they watched yesterday. They are also asked what type of programs they watched yesterday, and each program is coded as belonging to one of several categories. This makes it possible to distinguish educational programs (news, debates, information and documentaries, science, nature, and quiz) from other more entertainment-oriented shows (sports, kids/youth, religious, theatre and ballet, classical music, films, TV series, pop music, and other entertainment).
In Table 5, we present a summary of these data, illustrating time use and television content for boys and girls by SES, as well as a series of regression coefficients capturing the estimated influence of access to cable TV. We have defined three SES groups based on the education of the head of household: less than high school, high school, or college/university. For each of these groups, we have estimated ordinary least square regressions, using minutes spent yesterday on TV and reading, respectively, and the number of educational and other shows watched, as the dependent variables, and a dummy indicating access to cable TV as the key explanatory variable. In these data, we expect a considerable attenuation bias, as they are collected in a period with relatively high coverage of commercial TV channels through parabolic antennas in the no-cable group (approximately 30 percent). To come as close as possible to the approach in our main analysis, we have included controls for area type (cities with population >100,000, 20,000–100,000, <20,000, and “rural” areas), age, and year.
The Estimated Impact of Cable Connection on Television Consumption and Reading for Adolescents Aged 9–18 by the Head of Household’s Highest, Completed Education
Although the conclusions that can be drawn from the results presented in Table 5 are limited, given the small sample sizes and the resultant large statistical uncertainty, there are some noticeable patterns that emerge. First, for both boys and girls, there is a social gradient in TV watching, possibly related to variations in parents’ patience and effort in keeping their offspring focused on more cognitively stimulating activities. Time spent watching TV is decreasing with parental education. Second, particularly for boys, there is also a social gradient in reading. Time spent reading is increasing with parental education. Moving on to the estimated impacts of cable TV on media use, we note that access to cable TV implies much more television watching, as well as a marked shift toward noneducational shows. And, interestingly, it appears that the estimated impacts on time use are generally larger for offspring with higher SES. This is particularly the case when we look at the impacts on reading, where we find the strongest negative effects of cable TV among boys and girls with a college/university educated parent. Hence, there is some suggestive evidence here indicating that the higher TV consumption resulting from cable TV access substituted for more cognitively stimulating activities in families with higher SES.
Given these differences in impacts on actual time use, we now return to the impacts of cable TV on cognitive ability and high school completion, and we reestimate the baseline model described in Section III separately for offspring with different SES. We present the results in Table 6. Although the differences are relatively small and not statistically significant, the point estimates show indications of larger effects for families with higher SES. For the highest SES group, we now even obtain a slightly negative impact on high school completion for girls. In light of the examination of time use patterns, it is natural to interpret these findings as a result of the larger impacts that cable TV access has on time use in families with higher SES. It appears that TV viewing substitutes for more valuable and cognitively stimulating alternative activities—such as reading—in families with more educated parents. This also suggests that it may not be the TV viewing itself that primarily gives rise to the adverse impacts identified in this paper, but rather the reduction in the alternative activities that it substitutes.
Instrumental Variables and Family Fixed Effects Estimates of the Effect of One Additional Year with Cable Television Coverage. By Parents’ Education
Our finding that the adverse impacts of cable TV are larger for offspring with educated parents may at first sight seem surprising, given the previous evidence reported by Gentzkow and Shapiro (2008) and Kearney and Levine (2015) that TV has the largest impacts on offspring with low SES. However, it should be kept in mind that Gentzkow and Shapiro (2008) evaluated the introduction of TV as such, whereas Kearney and Levine (2015) evaluated the impacts of a particularly educating show (Sesame Street), and both reported positive impacts on the cognitive developments of preschoolers. In contrast, we examine the impacts of the introduction of commercial entertainment channels into an already existing TV environment and find negative impacts. Since we also provide suggestive evidence indicating that it is neither the TV viewing itself nor its content shift after deregulation, but rather the activities it crowds out that underlies the adverse effects, it is perhaps not surprising that the effects may be larger for offspring with higher SES.
B. Exposure at Different Ages
In our baseline model, we have assumed that commercial television exposure has the same impacts on cognitive ability and schooling outcomes, regardless of its timing within the 0–18 year age span. This is a questionable assumption. The literature on the importance of relatively early environments (Heckman 2006; Conti, Heckman, and Pinto 2016; Chetty, Hendren, and Katz 2016) and “critical periods” for cognitive development (Bleakley and Chin 2004; Van den Berg et al. 2014) suggests strongest effects in the preschool and elementary school periods.
Before we examine the impacts of cable TV exposure at different ages, Figure 7 shows what the media surveys can tell about the age profiles for actual television watching “yesterday” among those with and without access to cable TV. Up to age 14, there is a steady increase in the number of minutes spent on television watching for both groups. And those with cable connection watch considerably more television than those without, with a possible exception for the very youngest respondents.
Television Watching by Age (Smoothed), from the Media Use Survey, 1991–2004
To examine exposure impacts by age, we have split the television exposure variable into three separate periods of six years each, corresponding to the preschool, elementary school, and middle and high school periods, respectively, and include all three variables simultaneously in the baseline models (Equations 1 and 2) instead of the single TV exposure variable. We present the results in Table 7. Since the three coverage variables are highly correlated, and since exposure in early childhood almost always implies exposure at higher ages also, it is difficult to obtain a sharp identification of age-specific exposure effects, and few of the coefficients in Table 7 are statistically significant. However, taken at face value, they indicate that the adverse impacts on cognitive ability are highest for exposure in middle and high school periods, whereas the impacts on high school completion are largest for exposure during elementary school. Hence, our analysis does not lend support to the hypothesis that the preschool age is the most important period in this context. This is consistent with the suggestive evidence presented above that the adverse effects of TV viewing to some extent result from its displacement of reading, which is not a very common activity for preschool children. It is also consistent with the age profiles for actual TV watching shown in Figure 7, which suggests that access to cable TV has a smaller influence on time spent watching for the smallest kids.
Instrumental Variables and Family Fixed Effects Estimates of the Effect of One Additional Year with Cable Television Coverage, by Age of Exposure
VII. Conclusion
In this paper, we have used the geographically staggered expansion of cable TV in Norway during the 1980s and 1990s to examine the impacts of access to commercial TV channels on children and youths. Our outcome measures include ability test scores (measured at age 18) for men and high school completion at age 21 for both men and women. Our findings indicate that commercial TV affects ability test scores and high school completion rates negatively for men, but has no significant effect on high school completion for women. The estimated effects on male outcomes are moderately sized, but far from negligible. For example, one year of full cable television coverage lowers intelligence test scores by approximately 1.2 percent of a standard deviation, corresponding to 0.18 IQ points. This is roughly 4.5 percent of the previously estimated impact of one year extra schooling (Brinch and Galloway 2012). The same increase in cable TV reduces the high school completion rate of men at age 21 by around 0.4 percentage points.
We started out this paper by pointing out that many countries have seen a recent decline in intelligence test scores and questioned whether this could be explained by the increased access to commercial television channels. During the data period covered by our analysis, the average IQ score among Norwegian male conscripts declined by around 1.9 points (Figure 4), and average cable television exposure increased by three years. Based on our baseline model and data for aggregate cable television coverage, we estimate that the expansion of cable TV can account for a 0.5 point decline, that is, around 26 percent of the overall decline. Hence, although we do find negative effects of commercial television access on cognitive ability, the rollout of cable television can only explain a modest fraction of the apparent overall decline in IQ scores.
Another motivation was the educational gender gap. In the same manner as with intelligence test scores, we can calculate cable television’s contribution to the increasing gender gap in high school graduation. Taking the estimates of a zero effect for women and a negative effect of 0.4 percentage points per year of exposure for men, we get that cable television explains a 1.2 percentage point decline in high school graduation for men in this period, or about 20 percent of the increase in the gender gap. One may speculate whether the asymmetric effect across gender, which has a priori support in previous research, also applies to other types of media that provide light entertainment.
Our findings suggest that offspring from families with the highest socioeconomic status are not at all immune from the harmful effects of commercial television. To the contrary, our point estimates indicate that the negative impacts of TV exposure are largest for boys with highly educated parents. We have provided suggestive evidence that this apparently “inverted” social gradient arises because the activities crowded out by TV watching are more cognitively stimulating in families with highly educated parents. In particular, based on time use data, we have shown indications that the negative effect of cable TV access on time spent reading is particularly large for boys with educated parents. Based on these findings, we hypothesize that the negative effects identified in our paper do not primarily stem from increases in TV consumption per se, but rather from the resultant reduction in reading activities.
If it indeed is the case that the adverse effect of commercial TV arises because nonproductive television viewing crowds out more cognitively stimulating reading, we might also expect adverse effects on other outcomes normally assumed to be positively affected by reading. There is a large literature showing that there appear to be (small) positive impacts of leisure reading on noncognitive skills such as empathy and the capacity to identify and understand other people’s states of mind; see Mumper and Gerrig (2017) for a recent survey and meta-analysis and Kidd and Castano (2013) for experimental evidence. There is also a literature indicating that reading may be a fruitful long-term strategy for keeping the brain active and for preventing the risk of Alzheimer’s disease and other forms of dementia; see Stern and Munn (2010). However, how a more broad-based transition from leisure reading to consumption of modern multimedia entertainment will affect social and cognitive skills among men and women over the longer term is still basically unknown—and a fascinating area for future research.
Appendix
Standardized Income at Age 40 by Ability Test Score
Note: Data based on income of 40-year-old men in 2013. Income is first averaged by ability score, then standardized.
Footnotes
↵1. Recent negative trends in IQ scores have been reported for Norway (Sundet, Barlaug, and Torjussen 2004), Australia (Cotton et al. 2005), Denmark (Teasdale and Owen 2008), Britain (Shayer, Ginsburg, and Coe 2007), Sweden (Rönnlund et al. 2013), the Netherlands (Woodley and Meisenberg 2013), Finland (Dutton and Lynn 2013), and France (Dutton and Lynn 2015).
↵2. Time spent on sports and outdoor activities also declined somewhat during this period, from around 52–41 minutes for boys and 39–28 minutes for girls.
↵3. We treat the test score distribution as an interval scale, even though in principle it is ordinal. To test this assumption, we have rescaled the test scores according to the relationship between mean income levels by test scores for 40-year-olds in 2013 in order to capture information about actual intervals between scores. Figure A.1 in the Appendix shows the highly linear relationship between the original ability scores and the rescaled, standardized measure. The rescaling makes essentially no difference for the results.
↵4. Former head of the union of commercial cable-TV operators in Norway, Knut Børmer, personal communication. Terje Frøsland, of the state owned Norwegian Telecommunications (Televerket, from 2005 Telenor), personal communication.
↵5. Parental characteristics include four dummy variables for father’s education (less than high school, high school, bachelor, master), four dummy variables for mother’s education, ten dummy variables for father’s earnings, and ten dummy variables for mother’s earnings (with nonemployment as separate categories).
↵6. By pooling the data for for men and women and then incorporating gender interactions on all variables (including the fixed effects), we have also tested the statistical significance of the gender difference in the reported effects on high school graduation. These tests show that the difference between the −0.0046 (for men) and the −0.0015 (for women) coefficients in the IV model is not statistically significant at conventional levels (p-value = 0.19), whereas the difference between the −0.0042 and 0.0040 coefficients in the FFE model is statistically significant at the 5 percent level (p-value = 0.03).
↵7. We have also estimated the model using a categorization of population density in deciles in this interaction instead of using it directly as a scalar. This makes essentially no difference for the results.
- Received March 2016.
- Accepted August 2017.
