<<

JOSEPH PRICE Brigham Young University

GORDON B. DAHL University of California, San Diego*

Using Natural to Study the Impact of Media on the Family

The randomized trial is the in including television, movies, print and digital scientific research and is used by several fields advertisements, and the Internet (Pergrams & to study the effects of media. Although useful Zaradic, 2006). Some of these media may have for studying the immediate response to media negative effects, whereas other types of media exposure, the experimental approach is not well may have positive effects. Social scientists from suited to studying long-term effects or behavior a variety of disciplines have used different outside the laboratory. The ‘‘natural experi- techniques to model and estimate these impacts. ment’’ approach, a quasi-experimental method In this paper, we discuss what economists commonly used by economists, exploits settings have learned about these questions through in which there is naturally occurring variation the use of ‘‘natural experiments.’’ We view that ‘‘randomly’’ influences the amount and type the economist’s approach as complementary to of media that individuals are exposed to. We dis- those in other disciplines (and also recognize cuss the methodological issues of natural exper- its occasional use by other disciplines). A iments and the ways in which findings based primary aim of this article is to encourage a on this approach can complement the findings more interdisciplinary approach to evaluating from laboratory or other experimental settings. the impact of the media on the family. The examples we discuss highlight the value of The gold standard in scientific research is interdisciplinary work in better understanding the randomized controlled trial. In a randomized media’s impact on family issues such as fertility, controlled trial, subjects are randomly assigned divorce, domestic violence, and child well-being. to a treatment or control group. By construction, a correctly implemented randomized The question of how media affects families is yields a causal estimate of the effect of treatment important from both a scientific and practical on the outcome of interest because it balances all perspective, as it can help inform the decisions of of the potential factors between the both practitioners and parents. Family members two groups. There are, however, limitations to are increasingly exposed to a variety of media, real randomized trials. First, many experiments would be unethical or immoral. Second, many important outcome measures are impractical to measure in a laboratory. Third, laboratory Department of , Brigham Young University, and especially field experiments can be very Provo, UT 84602. ([email protected]). expensive to administer, especially for research *Department of Economics, University of California, San questions that involve large interventions or Diego, La Jolla, CA 92093 ([email protected]). require large samples. Attrition is also a major Key Words: family, media, natural experiments problem in long-term experimental studies. Family Relations 61 (July 2012): 363 – 373 363 DOI:10.1111/j.1741-3729.2012.00711.x 364 Family Relations

When laboratory or field experiments are not for a better understanding of media’s effect on possible, economists have tried to take advantage families. We then provide an example of a of ‘‘natural experiments.’’ At the broadest level, natural experiment that cannot be well studied in a natural experiment is a situation that is thought the laboratory, namely, the effect of unexpected, to approximate a randomized trial even though negative emotional cues on domestic violence. there was no actual involved In the second half of the paper, we turn to the (DiNardo, 2008). Economists have searched for arguably more important question, namely, the settings in which there is naturally occurring long-term effects of repeated media exposure variation in treatment, for example, a sudden on families. We explore how economists’ use of change in public policy (e.g., law changes), natural experiments adds to our understanding of random environmental shocks (e.g., weather), the long-term causal effects of media exposure, or other unexpected events (such as the birth of especially television exposure. The varied twins). The setting must be one where treatment natural experiments we discuss show that long- is arguably independent of all the confounding term exposure has both positive and negative factors that might otherwise bias an estimate. effects on families. In this section, we provide Natural experiments are a quasi-experimental several examples of how economists evaluate the methodology, and as such the key question validity of a natural experiment. Throughout, we for internal validity is whether the identifying point out the benefits and potential challenges to variation (the ‘‘treatment’’) is actually random. the use of natural experiments. Except in rare cases, there are no direct statistical tests to establish internal validity. This is because NATURAL AND LABORATORY EXPERIMENTS it is impossible to test whether treatment is IN THE SHORT RUN correlated with an unobserved variable. Instead, the researcher must provide arguments and Whereas natural experiments have a long history indirect tests in an attempt to convince the in economics (Meyer, 1995), the use of natural reader that the natural experiment can be used to experiments by economists to study the effects estimate a causal effect. To start, the researcher of media is more recent. An early pioneer lays out why the natural experiment occurred and in the use of natural experiments to study why it is likely to introduce random variation the effect of media is the sociologist David in treatment. The researcher then continues by Phillips. He initially called his settings ‘‘found empirically ruling out alternative explanations experiments,’’ although the idea is the same with a variety of indirect and placebo tests as a natural experiment. In a series of papers, as falsification exercises. Although the indirect he documented that suicide rates increase after tests will vary from application to application, mass media reporting of suicides, including some of the more common ones include an increase in teenage suicides (Phillips & showing (a) the treatment occurs before the Carstensen, 1986). In similar papers, Phillips outcome, (b) the treatment cannot be predicted found that motor vehicle accident fatalities by lagged outcomes, (c) there are no differential also increase after newspaper stories about preexisting trends in the outcome, and (d) the suicide (Phillips, 1979). Phillips has also used observed covariates are balanced in the treatment a similar methodology to study the impact of and control samples. Often, supporting evidence mass media violence on homicides, using both comes from analysis of auxiliary . fictional television stories and news reporting as In this paper, we develop the idea of natural natural experiments (Phillips, 1983; Phillips & experiments as we discuss the contributions of Hensley, 1984). In each case, the idea behind economists to the literature on media’s causal Phillips’s natural experiments is that, once effect on families. We begin by comparing and seasonal patterns have been controlled for, media contrasting natural experiments to those done reports of suicides, murder, or other violent in the laboratory, which, by design, estimate behavior should be random events. The spikes short-term effects. In this first section, we focus in homicides and suicides following such events on two examples. We start with the effect of can be argued to be the causal result of media exposure to violent movies in the laboratory exposure. He argued that these effects operate versus the field. We outline why the two through imitation or suggestion. By their design, approaches identify different effects and why these experiments are well suited to estimate these complementary approaches are important short-run effects but not long-run impacts. Using Natural Experiments 365

One reason to conduct natural experiments the audience for blockbuster movies is high. is that they serve as useful complements to Broken down by time of day, in the evening laboratory experiments. Over the past several hours when people are in the theaters, there is a decades, laboratory experiments have greatly significant negative effect on crime for strongly added to our understanding of the causal effects and mildly violent movies; violent crime falls of media in the short run. The experimental lit- by around 1.2% per million violent movie erature using the laboratory, largely conducted goers. In contrast, weekends with nonviolent by psychologists, has focused heavily on study- blockbuster movies have small and insignificant ing the effect of media violence on aggressive effects. The authors interpreted this first effect behavior, both for violent video clips and violent as ‘‘voluntary incapacitation.’’ On evenings video games (see Anderson & Bushman, 2001; with high attendance at violent movies, many Anderson et al. 2003; Bushman & Anderson, potential violent criminals choose to be in 2002). This experimental literature has con- a movie theater and are incapacitated from vincingly demonstrated a causal link between committing a crime during the movie. violent media on aggressiveness immediately In the night-time hours after a violent movie after exposure, including a sharp increase in is over, crime falls by an even larger percent, aggression for children and young adults. These whereas nonviolent movies again have no findings have led many scientists, parents, and discernable impact. This delayed effect initially politicians to worry about the risk of media seems at odds with the psychology experiments, violence and its negative impacts. which find large increases in aggressiveness due To focus on how the use of a natural to arousal immediately following exposure. It experiment can add to what can be learned is, however, important to recognize that the from the laboratory, we make a comparison two methodologies capture different effects. to laboratory experiments that expose subjects The laboratory experiments estimate the impact (typically children or college students) to short, of exposure to a violent movie compared to violent film clips. That literature provides causal a nonviolent movie. The natural experiment evidence on the short-run effect of media allows individuals to choose between a movie violence on aggressiveness but not whether this and an array of alternative activities. Dahl and translates into higher levels of violent crime in DellaVigna’s (2009) paper provides evidence the field (through arousal or imitation). Dahl that violent movie attendance is displacing and DellaVigna (2009) addressed the question more violent-prone alternative activities (such of whether violent movies increase violent crime as drinking at a bar). using a natural experiment. They exploited the The natural experiment emphasizes that time natural experiment induced by variation in when use plays an important role in how violent violent movies are released. media affects crime in the short run. Violent The idea behind this natural experiment is media keeps violent people busy during the that in some years there will be strongly violent movie and changes an evening’s activities even movies on certain weekends during the year after the movie is over, so that violent movie wheres in other years those same weekends will attendees are less prone to commit violent crime. have nonviolent blockbuster movies. Which year The psychology experiments emphasize content happens to have a violent blockbuster movie alone, which explains the difference in findings. should otherwise be uncorrelated with violent To be clear, Dahl and DellaVigna (2009) were crime, after flexibly controlling for seasonality. not arguing for a catharsis effect from violent As an example, in the second week of February movies; they actually found some evidence for 2001, the strongly violent movie Hannibal was an arousal effect, but argued that this effect released, whereas in other years, the second (which increases violent crime) is more than week in February was usually dominated by offset by the much larger time use effect (which blockbuster romantic comedies. This provides a lowers violent crime relative to the alternative natural experiment, where the treatment is the activity). availability of a violent movie compared to the It is also important to point out that control of the availability of a nonviolent movie the laboratory experiments are capturing the on similar weekends. reaction to media violence in a representative Dahl and DellaVigna (2009) found that sample of the population, whereas the natural violent crime is actually lower on days when experiment results are largely driven by which 366 Family Relations types of individuals choose to sort into violent are broadcast via the media? To help frame this movie attendance. As such, the laboratory question, consider the theory of gain-loss utility setting is not representative of the effect of developed by economists and psychologists, exposure in most field settings, where consumers which says that individuals asymmetrically choose what media to consume. Rather it is frame gains and losses around a rationally representative of an instance of unexpected expected reference point (Koszegi & Rabin, exposure, such as a violent trailer placed within 2006). In simple terms, this theory says two family television programming. things: (a) an individual’s happiness depends We believe that the causal estimates from not on actual outcomes but on outcomes judged both the laboratory and the natural experiment relative to expectations, and (b) unexpected are interesting and informative for answering disappointments have a larger impact on different questions. For example, the natural individuals compared to pleasant surprises. experiment findings are important for evaluating To test such a theory in the context of policies that would restrict access to violent domestic violence, one would need an exper- movies, because these policies will lead to a iment where family violence outcomes could substitution toward alternative activities that be measured after both unexpected positive and may well be more violent in the short run. We negative surprises. Card and Dahl (2011) use also believe, however, that differences between the natural experiment created by NFL football the laboratory and field can be narrowed by game broadcasts. The setting is motivated by changes in the design of laboratory experiments. three reasons. First, a large number of fans are For example, laboratory experiments could strongly attached to local teams, with regular- incorporate sorting into a violent movie to season Sunday games averaging a television replicate selection in the field or can change audience equal to 25% of all local households. exposure to a full-length movie (Lazear, Second, there are detailed game that Malmendier, & Weber, 2012). make it easy to identify more or less salient Although Dahl and DellaVigna (2009) found games. Perhaps most importantly, the existence no impact of movie violence in the weeks fol- of a well-organized betting market allows one to lowing exposure, their design, like the laboratory infer the expected outcome of the game, which experiments, cannot address the more important serves as the reference point for gain-loss utility. question about the long-run effect of violent Each of these characteristics is crucial to their media consumption. In the next section, we turn ability to use this setting as a natural experiment. to natural experiments that can get at the effects Does this natural experiment provide ‘‘ran- of long-term exposure to television (although dom’’ positive and negative upset shocks? If not necessary violent media). Before doing so, betting markets are unbiased predictors of the we provide one more example of a natural exper- game outcome, the answer is yes. With unbi- iment that addresses short-run effects. ased betting markets, after conditioning on the The previous example illustrates how labora- pregame point spread any differential effect of a tory and natural experiments complement each win versus a loss is the causal effect of the game other by answering different questions. Many outcome by construction. So a sufficient condi- times, however, laboratory or field experiments tion for the validity of this natural experiment for important family outcomes cannot be done is easy to state and verify. Indeed, empirical due to ethical or practical considerations. One evidence suggests one cannot make money by such example is whether unexpected emotional betting against the market (i.e., there is strong cues (or ‘‘visceral factors’’) play an important evidence the market is unbiased). role in precipitating family violence. Media can Card and Dahl (2011) found that police have a powerful potential to affect people’s emo- reports of male-on-female intimate partner tions, as has been demonstrated in the laboratory. violence rise 10% in areas where the local But linking this evidence to an effect on domes- NFL team lost a game they were favored to tic violence is difficult to do, and laboratory and win by four or more points. Consistent with field experiments attempting to measure actual reference point behavior, a loss when the game domestic violence would be both impractical was expected to be close or when the local team and unethical. was expected to lose does not have a significant Is family violence linked to unexpected effect on family violence. Also consistent with disappointments (negative emotional cues) that the theory, they found little protective effect of Using Natural Experiments 367 an unexpected win, pointing to an important natural experiments that rely on timing, it is asymmetry in the reaction to unexpected gains important to control for seasonality, as in the versus unexpected losses. Dahl and DellaVigna (2009) example. This is Timing is an important element of their also true for natural experiments where there natural experiment, and Card and Dahl (2011) is a conditioning variable that captures all the match the hourly local police reports of violence available information up to a point in time, such to the time frame of the football games. The as a natural experiment involving stock market increases in violence after an upset loss are prices or betting odds (see Card & Dahl, 2011). concentrated in a narrow time frame at the end of the game, as one might expect if the spike in violence is due to a transitory emotional LONG-RUN EFFECTS OF MEDIA shock. These patterns are more pronounced for The short-run effects of media are primarily upset losses in games that are more emotionally of interest because of the belief that the charged, such as games against a traditional accumulation of short-run effects is likely to rival, when the local team was still in playoff result in long-run changes in behavior. The contention, or after a particularly frustrating ability to influence long-run media exposure in performance (an excessive number of sacks, an experimental setting, however, is extremely turnovers, or penalties). Their paper suggests limited. As a result, most of the research on long- that a better awareness and management of run effects of media is based on correlations expectations could help to reduce violence between media exposure and the outcome of within families. interest (controlling for as many confounding One drawback of many natural experiments is factors as possible). There is often emphasis that it can be difficult to identify the mechanisms, put on collecting longitudinal data so that even if a causal effect can be documented. the temporal ordering of media exposure and For example, in the Card and Dahl (2011) measured behavior can be established. paper, although their analysis points to a causal Even with the best longitudinal data, the spike in domestic violence after an unexpected primary concern about nonexperimental data loss, the mechanism is difficult to unravel remains: Individuals get to choose the amount without further information. For example, the and type of media exposure. As such, there may effect could operate through increased alcohol be unobserved characteristics about individuals consumption in response to the unexpected that influence both their media choices and their loss, but their design and data can only hint outcomes (even if these outcomes are measured at whether this is the mechanism. Another at a later period). Many of these omitted vari- inherent challenge with natural experiments is ables can be included as controls in the analysis, that they are not always easy to replicate and may but there is always the possibility that important not have external validity. For example, other characteristics are not observed or might even types of unexpected and upsetting events may be unmeasureable. If these unobserved charac- affect different segments of the population or teristics affect both media consumption and the operate in different ways, and therefore have a outcome, the resulting estimate of media expo- dramatically different impact on violent crime. sure will be biased. This problem of unobserved It is important to discuss the role of control heterogeneity affects many important areas of variables in natural experiments versus both research that involve individual choices, such as laboratory experiments and field experiments. church attendance, marriage, and paternal and In laboratory and field experiments, the only maternal employment, all of which are likely reason to include control variables is to to influence important family outcomes but for improve the efficiency of the estimates (i.e., to which the selection problem is difficult to deal decrease the standard errors). This is because with. with a laboratory or field experiment, the Although observational data in these settings treatment is asymptotically uncorrelated with can provide important descriptive relationships, the covariates as a result of randomization. economists’ focus on causal estimates has forced In many natural experiments, however, it is them to look for natural experiments, a quasi- important to include a set of control variables, experimental method that tries to approximate because often the treatment is only a random the gold standard of a randomized trial. The event conditional on those controls. In the key to this approach is to look for a situation 368 Family Relations in which there are two groups of individuals occurs. In fact, one of the primary criticisms of with similar characteristics but who experience the natural experiment approach is that it limits different levels of exposure to media (due to researchers to looking at those settings in which reasons that are unrelated to their own individual a natural experiment has occurred. For better or attributes and choices). worse, this feature of natural experiments has led Except in rare cases, there are generally not many researchers to use historical data to look any direct statistical tests that can be used at interesting stages in the diffusion of different to establish validity, because the question is forms of media (Gentzkow & Shapiro, 2008; whether treatment is correlated with an unob- Stromberg,¨ 2004) and then provide thoughtful served variable. Rather, it is a combination analysis of whether the results at a certain of understanding why the natural experiment place and time have external validity and can occurred, indirect and placebo tests as falsifi- be applied to other settings. cation exercises, and supplementary empirical The ideal natural experiment is often unavail- evidence that together can provide a convincing able. As a result, researchers will examine situ- case. Many of the common and convincing indi- ations in which the variation in media exposure rect tests involve the timing of the treatment and is not as random as the topography of Indonesia, outcomes. Such arguments and indirect tests aim but might still unfold in ways that are unrelated to to convince the reader that the source of varia- the characteristics of the individuals being influ- tion in media exposure is not being influenced by enced by the media. R. Jensen and Oster (2009) the decisions and characteristics of the individ- used a 3-year panel of individual-level data from ual (particularly those related to the outcome of 180 villages in India. During this time period, interest). Because each application is unique and 21 of these villages received cable television for has its own set of plausible alternatives that must the first time. Their empirical approach com- be ruled out, we explore arguments and indirect pared outcomes ‘‘between survey rounds across tests for quasi- in the context villages based on whether (and when) they added of the several examples that follow. cable television’’ (p. 1059). They found that the Recent work by Olken (2009) provided an introduction of television decreased acceptance excellent illustration of the natural experiment of domestic violence, increased the amount of approach. Indonesia’s mountainous terrain pro- autonomy experienced by women, and led to vides a situation in which two very similar and lower rates of fertility. closely located villages can experience very dif- One of the concerns about this setting is ferent levels of television signal strength based that trends in other variables (such as income on which side of the mountain they happen to be or ‘‘modernity’’) may cause certain villages to on. By combining models of signal strength with both adopt cable television and also experience detailed geographic information system data on changes in the outcomes of interest. R. Jensen the topography of Indonesia, Olken constructed and Oster (2009) used many of the traditional a measure of the signal strength in each village methods in this vein of research to provide sup- and separated the variation in the strength of port for the validity of their natural experiment. signal across villages into the part that is due First, they showed that there were no preexisting to predictable factors (distance to major city, differential trends in these outcomes between the elevation, etc.) and the part due to the ‘‘hap- villages that did or did not adopt cable television. penstance of topography.’’ He found that the Second, they showed that the outcomes of inter- topography-induced differences in exposure to est are not correlated with future cable access. television and radio cause individuals in the more That is, behavior this year cannot be predicted by media-exposed villages to become less involved using information on whether you adopt cable in community organizations. next year, which would often be the case if the Although not the focus of his research, the relationship being estimated is driven by under- approach Olken (2009) used could be applied to lying but unobserved differences between the outcomes that are more specifically related to the treatment and control groups. Third, they col- family, such as fertility, divorce, or child well- lected additional information about how cable being. The ability to extend this type of research operators made the decision about which vil- hinges crucially on the availability of the relevant lages to operate in. They found that this decision data for the outcome of interest in locations is largely driven by issues related to the cost and time periods where the natural experiment of supplying the cable rather than village-level Using Natural Experiments 369 demand issues (in particular those that might of the shows (Chong & La Ferrera, 2009; La influence the outcomes). Ferrara, Chong, & Duryea, 2008). Rather than A piece of supporting evidence in the natural looking at the diffusion of a media technology, experiment approach is to provide a theoretical they examined the diffusion of a particular explanation for the relationship between the type of show on television, the soap opera. In change in media and the outcome of interest. R. Brazil, the television network Rede Globo had Jensen and Oster (2009) discussed some of the a virtual monopoly on the production of soap ways in which television may have brought about operas, a set of shows that depict a lifestyle the changes they document (some of which they that involves smaller family sizes and higher were able to provide supporting evidence for levels of divorce and infidelity. These papers using additional data they collected). Increased exploited differences in the timing of when the television access may alter fertility choices by Globo network entered different areas of Brazil providing more information on family planning and found that the arrival of this network led to alternatives. It may also change the value of increased rates of divorce (Chong & La Ferrera, women’s time or increase the relative value of 2009) and lower levels of fertility (La Ferrara leisure for men, leading men to grant greater et al., 2008). liberty for women to do more outside of the The primary concern about this natural home. Also, television exposes rural villages experiment is whether the Globo network to urban culture, a culture that exhibits fewer decided to locate in those areas that were already cases of acceptable wife beatings, less interest starting to experience the demographic changes in having their next child be male, and greater that are depicted in the soap operas. The authors autonomy for women. used many of the common approaches in this There have also been studies outside of literature to establish the validity of the natural economics that use the natural experiment experiment. First, they examined changes in approach to study this issue. Centerwall (1989) fertility during the years before and after the used differences across countries in the timing entry of Globo and found no decrease in fertility of television adoption to look at the effect prior to entry but a sharp decline immediately it had on homicide rates. Joy, Kimball, and after the arrival of Globo. Second, they showed Zabrack (1986) examined a town in Canada that which areas received the Globo network that did not receive television until 1973 and are not the areas that were already experiencing compared the change in physical aggression larger changes in fertility or divorce. Third, among children between this town and two they documented that the presence of Globo control towns. Both of these studies found that in neighboring areas did not influence local the introduction of television led to an increase in fertility (which would be the case if the results violence. Subsequent criticisms of Centerwall’s were driven by unobserved cultural or economic research highlight some of the issues that we factors that might be common to adjoining discuss in this paper, such as the importance areas). of examining preexisting trends (G. Jensen, La Ferrara et al. (2008) also used within- 2001) and controlling for confounding factors show variation to provide further evidence that (Felson, 1996). the effects are operating through the content In both studies, the small number of relevant being depicted on these shows. They found geographic units (three countries or three that children born in areas receiving Globo towns) makes precise inference a challenge. In are much more likely to be named after the contrast, the more recent research in economics main characters of the soap operas aired that used multiple geographic units within a single year. Also, the fertility effects are strongest for country. This makes it possible to control for the women who are closest in age to those of country-level changes over time and provides the main female characters in that particular enough distinct geographic units to provide year. These two examples illustrate the type of meaningful after properly additional evidence that can be used to support clustering the standard errors (Bertrand, Duflo, the main findings of a natural experiment and & Mullainathan, 2004). help provide insight into the specific mechanism A pair of recent papers using data from Brazil through which the effect is operating. provides more specific evidence that the effect Other examples of recent research, although of television can operate through the content less focused on outcomes related to the family, 370 Family Relations highlight the type of natural experiments that statistically significant). For these households, can be used to examine the more general the additional media time was likely replacing effects of media on society. Each paper follows other activities that would have provided better the same general approach: (a) documenting cognitive development. They found, however, the differences across similar individuals in the much larger positive effects for children from amount or type of media they were exposed households where English is not the primary to (the strength of the natural experiment), language, for children whose mothers have less (b) providing evidence that the difference in than a high school education, and for non- exposure was not determined by other factors White children. These homes provide much that could induce a spurious relationship less language-rich environments (Dickinson & (the validity of the natural experiment), and Snow, 1987), so that the alternative use of time (c) documenting the changes in the outcome of would have provided less cognitive development interest that accompanied the change in media than television. In addition, television may have exposure. pulled children away from harmful influences DellaVigna and Kaplan (2007) exploited outside of the home or diffused situations in the variation in the content that households receive home that lead to abuse. This last point illustrates on television, focusing on the effects of the the importance of considering both the exposure introduction of Fox News (a conservative cable effect and the time-use effect of media (a point news channel) that had reached about 20% also highlighted by the Dahl & DellaVigna, of towns in the United States by November 2009, paper discussed in the previous section). 2000. They first showed that the availability of this channel in different towns appears to be largely idiosyncratic after conditioning on a set DISCUSSION of basic controls. They found, however, that the The purpose of this article is to describe the towns that received Fox News between 1996 economist’s approach to examining the effect and 2000 experienced a 0.4 to 0.7 percentage of media on outcomes that are important to the point increase in the fraction of votes going family. The primary approach of economists to the Republican candidates. They noted that has been the use of natural experiments that this change in behavior might represent a provide quasi-random variation in the amount, ‘‘temporary learning effect for rational voters, type, or content of the media that individuals or a permanent effect for non-rational voters have access to. Examples of these natural subject to persuasion’’ (p. 1187). experiments include changes in the content of Gentzkow and Shapiro (2008) used informa- movies released in a particular week, differences tion on the timing of the arrival of broadcast across areas in the strength of television signals television in different counties in the United they can receive, and differences in the timing of States to examine the pressing concern voiced when different technologies or specific content by parents about whether ‘‘television rots the providers arrive in different parts of a country. brain.’’ They used a large nationally representa- Each of these create settings where otherwise tive data set on test scores during the 1940s and similar individuals experience different types of 1950s (when television diffusion occurred in the access to media, and these differences can be United States) and found that each additional used to estimate the effect of media on families. year of childhood exposure to television actu- Table 1 provides a summary of each of the ally increased test scores during adolescence by natural experiments that we describe in this 0.02% of a . paper. The information in the publication year They also considered whether different column highlights the fact that economists are groups of children were affected differently recent arrivals to studying the effects of media, by exposure to television. This part of their an area of research that has been examined by analysis highlighted the fact that, for many other fields for a much longer period of time. As children, the effects of exposure to media are such, it is easy for these new arrivals to ignore the determined largely by the type of activities that important research done in other fields and the are being displaced by television. For children accumulation of knowledge that has occurred in the types of homes with the highest levels of over the decades in which economists played parental involvement, the effect of television almost no role in this research area. It is also was negative (although this effect was not easy for fields that have a more established Using Natural Experiments 371 was introduced had a 0.4 to Fox News number of children by 1% andgiving the birth probability in of that year by 6% of separation or divorce by 7% 0.7 percentage point higher Republican votethe share 2000 in presidential elections compared to 1996 preference by 70%, the autonomy indexnumber by of 41%, acceptable the beating situations bythe 46%, probability and of being pregnant atby time 53% of survey average test scores by 0.02% of a standard deviation 7% fewer social groups, and adultspercent attending fewer 11% group meetings violent crime by 1.3% during thein hours the the theater movie and is by 1.9%after in exposure the hours immediately violence; a loss when the team washave expected a to close lose game or has small and insignificant effects The availability of Rede Globo increases the probability Towns where The introduction of cable television decreases son Each additional year of access to television increased An additional channel of television reception leads to and voter turnout separated or divorced violence, son preference, women’s autonomy, and fertility mathematics, spatial reasoning, verbal, and reading community organizations and measures of trust Fertility rates The availability of Rede Globo decreases the average Republican vote share Share of women who are Attitudes about domestic Test scores in Violent crimeIntimate partner violence Each million people watching a Upset violent losses movie lead decreases to a 10% increase in intimate partner Participation in Fox became available in different the Rede Globo network, the major provider of soap operas in Brazil News parts of the United States the Rede Globo network, the major provider of soap operas in Brazil television in rural India television was introduced to different local markets in the United States during the 1940s and 1950s of blockbuster movies provided by comparing final NFL game outcomes to pregame point spreads strength due to the topography of villages in Indonesia Variation in the timing of availability of Variation in the timing of when Variation in the timing of availability of Variation in the timing of when Temporal variation in the violence level Unexpected wins and unexpected losses . Summary of Natural Experiments by Economists Related to the Effects of Media Table 1 on television television television preschooler theaters cues from local NFL games Fox News 2007 2008 Soap operas on 2008 Television as a Kaplan & Duryea Shapiro DellaVigna & La Ferrara, Chong, Chong & La Ferrera 2009 Soap operas on R. Jensen & Oster 2009 Cable television Variation in the timing of cable Gentzkow & Card & Dahl 2011 UnexpectedOlken emotional 2009 Television and radio Differences in television and radio signal PaperDahl & DellaVigna 2009 Violent movies in Year Media Exposure Natural Experiment Outcomes Main Finding 372 Family Relations history of research in this area to discount the general aggression model. Personality and Social results of economists who are new arrivals to Psychology Bulletin, 28, 1679 – 1686. this research area and in the nascent stages of Card, D., & Dahl, G. B. (2011). Family violence contributing to this research question. and football: The effect of unexpected emotional The interdisciplinary nature of the conference cues on violent behavior. Quarterly Journal of Economics, 126, 103 – 143. that led to this special issue illustrates the degree Centerwall, B. (1989). Exposure to television as a to which important advances in understanding risk factor for violence. American Journal of can occur when considering the approaches , 129, 643 – 652. of multiple fields. Ultimately this type of Chong, A., & La Ferrara, E. (2009). Television interdisciplinary can lead to a more and divorce: Evidence from Brazilian novelas, comprehensive view of the effects of media Journal of the European Economic Association, 7, on important family outcomes. This article 458 – 468. highlights some of the ways in which the Dahl, G. B., & DellaVigna, S. (2009). Does movie economist’s approach can complement the violence increase violent crime? Quarterly Journal important research findings of other fields. First, of Economics, 124, 677 – 734. our discussion of the effects of movie violence DellaVigna, S., & Kaplan, E. (2007). The Fox News effect: Media bias and voting. Quarterly Journal illustrates the importance of considering both the of Economics, 122, 1187 – 1234. exposure effect and the time-use effect of media. Dickinson, D. K., & Snow, C. E. (1987). Interrela- The total effect will depend on both the content tionships among prereading and oral language of the media and the activities that are displaced skills in kindergartners from two social classes. during the time spent using the media. Second, Early Childhood and Research Quarterly, 2, the natural experiments that we describe provide 1 – 25. one way to study the long-run effects of media on DiNardo, J. (2008). Natural experiments and quasi- important outcomes (such as divorce, fertility, natural experiments. In S. Durlauf & L. Blume and domestic violence) for which traditional (Eds.), The new Palgrave dictionary of economics experiments would be unethical or immoral and (2nd ed.). Basingstroke, UK: Palgrave Macmillan. Felson, R. (1996). Mass media effect on violent handles the challenging issue of unobserved behavior. Annual Review of Sociology, 22, heterogeneity. 103 – 128. Gentzkow, M., & Shapiro, J. (2008). Preschool television viewing and adolescent test scores: NOTE Historical evidence from the Coleman study. This research was supported by a Mentoring Environment Quarterly Journal of Economics, 123, 279 – 323. Grant from Brigham Young University. We are grateful Jensen, G. (2001). The invention of television as a for the research assistance provided by Nathan Nazzise and cause of homicide: The reification of a spurious Nolan Pope and for the comments provide by participants at relationship. Homicide Studies, 5, 114 – 130. the Media and Family Conference at BYU. Jensen, R., & Oster, E. (2009). The power of TV: Cable television and women’s status in India. Quarterly Journal of Economics, 124, 1057 – 1094. REFERENCES Joy, L., Kimball, M., & Zabrack, M. (1986). Tele- Anderson, C., Berkowitz, L., Donnerstein, E., Hues- vision and children’s aggressive behavior. In man, L. R., Johnson, J., Linz, D., Malamuth, N., T. Williams (Ed.), The impact of television: & Wartella, E. (2003). The influence of media A natural experiment in three communities violence on youth. Psychological Science in the (pp. 303 – 360). Orlando, FL: Academic Press. Public Interest, 4, 81 – 110. Koszegi, B., & Rabin, M. (2006). A model of Anderson, C., & Bushman, B. (2001). Effects of vio- reference-dependent preferences. Quarterly Jour- lent video games on aggressive behavior, aggres- nal of Economics, 121, 1133 – 1165. sive cognition, aggressive affect, physiological La Ferrara, E., Chong, A., & Duryea, S. (2008). Soap arousal, and prosocial behavior: A meta-analytic operas and fertility: Evidence from Brazil.CEPR review of the scientific literature. Psychological Discussion Paper #6785. Science, 12, 353 – 359. Lazear, E. P., Malmendier, U., & Weber, R. A. Bertrand, M., Duflo, E., & Mullainathan, S. (2004). (2012). Sorting in experiments with application to How much should we trust differences-in- social preferences. American Economic Journal: differences estimates? Quarterly Journal of Eco- Applied Economics, 4, 809 – 811. nomics, 119, 249 – 275. Meyer, B. D. (1995). Natural and quasi-experiments Bushman, B. J., & Anderson, C. A. (2002). Violent in economics. Journal of Business & Economic video games and hostile expectations: A test of the Statistics, 13, 151 – 161. Using Natural Experiments 373

Olken, B. (2009). Do television and radio destroy Phillips, D. P. (1983). The impact of mass media social capital? Evidence from Indonesian villages. violence on US homicides. American Sociological American Economic Journal: Applied Economics, Review, 48, 560 – 568. 1(4), 1 – 33. Phillips, D. P., & Carstensen, L. L. (1986). Clustering Pergrams, O. R. W., & Zaradic, P. A. (2006). Is of teenage suicides after television news stories love of nature in the US becoming love of about suicide. New England Journal of Medicine, electronic media? 16-year downtrend in national 315, 685 – 689. park visits explained by watching movies, Phillips, D. P., & Hensley, J. E. (1984). When vio- playing video games, Internet use, and oil lence is rewarded or punished: The impact of mass prices. Journal of Environmental Management, 80, media stories on homicide. Journal of Communi- 387 – 393. cation, 34, 101 – 116. Phillips, D. P. (1979). Suicide, motor vehicle fatali- Stromberg,¨ D. (2004). Radio impact on public ties, and the mass media: Evidence toward a theory spending. Quarterly Journal of Economics, 119, of suggestion. American Journal of Sociology, 84, 189 – 221. 1150 – 1174.