| home | authors | calendar colophon | links | newsgroups | newsfeed | new | printer version |  
volume 5
april 2002

Television violence is fully harmless


  What Johnson and his fellow researchers really found (but did not report)
by Ger Tillekens
  On Eastern 2002 a load of papers and news stations worldwide broke the news: now finally there is definite proof of the relationship between viewing television and later aggressive behavior among adolescents. Al least such was the outcome of a long-term research project, published by five researchers from New York in the renown scientific magazine Science. A closer look at the article itself, however, reveals that the analysis is not very solid. The article was filed under "Sociology". A trained sociologist himself, Ger Tillekens reconstructs the data analysis presented in the Science article and argues that the results are based on a full-blown statistical fraud. Moreover, the reconstructed data seem to prove exactly the opposite of what the researchers claim: for adolescents watching television is really fully harmless.
1 Compromising on a linear relationship. "More TV for youths now may mean more aggression later." This provocative line headed the article by Susan FitzGerald in the Philadelphia Inquirer of Friday, March 29, 2002. Below the caption FitzGerald discussed the results of a research project assessing the relationship for adolescents between watching television and aggressive behavior over a seventeen-year interval in a community sample of 707 individuals, that was published that very same day in the renown Science magazine. The article was cowritten by no less than five authors: Jeffrey Johnson, Patricia Cohen, Elizabeth Smailes and Stephanie Kasen working at the Columbia University and the New York State Psychiatric Institute and Judith Brook from the Mount Sinai Medical Center — all institutes located in New York City (Johnson et al., 2002). The Philadelphia Inquirer was not the only paper breaking the news. Spread by the press agencies of Reuters and Associated Press, the news item received a lot of attention from major papers and news stations all over the world. Let's be clear from the start: if the conclusion of the New York research group is really right, it deserves to be heard. Indeed, it is not a minor charge the authors are making against television as they claim to have found long lasting serious effects years afterwards. These negative effects, moreover, do not concern little children who may have problems discerning the representation of symbolic violence on the television screen from concrete violence in real life, but adolescents of fourteen years and older. And, these effects are not restricted to just an emotional response, for instance a heightening of excitement, but come into the open in real aggressive behavior.
  That is a serious allegation and not everybody seemed to be persuaded by this scientific statement. There were clearly some doubts seeping through FitzGerald's review, as she phrased the relationship with the words "may mean". Some papers also quoted other scientists expressing their misgivings about the project (BBC News Services, 2002; CBC News 2002). The abstract, accompanying the original article in the last March edition of the Science magazine, however, was not hindered by such thoughts and more firmly stated: "There was a significant association between the amount of time spent watching television during adolescence and early adulthood and the likelihood of subsequent aggressive acts against others." In the press this idea of an incremental relationship between daily hours of watching television and later aggressive behavior was strengthened by some impressive figures. Increasing the daily exposure to television from less than one hour to three hours to over three hours at age fourteen would result in an growing chance of active involvement in aggressive behavior at age sixteen and even later at age twenty-two. The percentage of aggressive youths would rise from 5.7% over 18.4% to 25.3% of the groups consuming these daily amounts of television time. Reading summaries like this in the press, I expected the researchers had found a clear linear relationship between the daily amount of time spent on viewing television and later aggressive behavior — implying that if young people watch more television, they will be more inclined to show aggressive behavior later on.
Some papers also referred to an odds ratio — an incremental risk index — for falling into aggressive behavior as computed by the researchers for each next step in daily television time. [1] In many press publications, however, Johnson was also quoted summarizing the outcomes in more conditional terms (MSNBC News Services, 2002), :
  "We saw the jump was between less than one hour and more than one hour a day. There was a four-fold increase."
  Only less than one hour? Whatever happened to adolescents watching over one hour? Were they all risking to behave aggressively in the same way? In short, what had happened to the presumed linear relationship? Johnson's remarks, compromising on having found a linear relationship, made me rather suspicious. So I took a critical look at the figures that were presented in the Science magazine itself, and doing so my suspicions grew. The authors presented part of their figures about aggressive acts at "mean age sixteen or twenty-two". What does that mean? And, why did they use measures of association like the Chi² instead of statistical techniques fit for analyzing linear relationships? The caption of the Philadelphia Inquirer was one of the few showing more cautiousness, and rightly so. The data analysis is rather sloppy and, as we shall see, only by squeezing their data and capitalizing on the outcomes, could the researchers maintain that there is a relationship between viewing television and later aggressive behavior. As we all know, you can lie with statistics and most of the times that is rather innocent as most people can see through it. However, you can also cheat with statistics and that is what happens here. As we shall see, one even may call this case a conscious statistical fraud.
  What is wrong with the research, presented by Johnson, Cohen, Smalles, Kasen and Brook? Of course, one could comment on their approach from a theoretical perspective by looking at their ideological assumptions. Here, however, I will just focus on the methodological aspects. As I will show, it seems the researchers had to massage their data to find at least some, be it very weak, indications of the relationship they were after. This weak association, moreover, proves to be an artefact of the way they analyzed their data. We will find some minor flaws along some major frauds. Among the flaws are the presentation of data collection as a longitudinal design, the rather rough definition of the dependent variable aggressive behavior, and the merging of categories of the independent variable daily television time. The more serious frauds regard the treatment of the covariate of low family income, the conscious neglect of the rising risk for adolescents from low income families of falling into aggressive behavior over the years, and the ascription of what really are effects of low family income to the influence of watching television. On these points I will try to reconstruct the data analysis into detail. In between I will explain what controlling for covariates implies and I will also say something about those covariates that did not prove important as predictors for the daily consumption of television time or for the later acts of aggression among adolescents. At the end I will argue that, despite all their efforts, the New York researchers found no association at all. So if their project does prove anything, it is that a relationship between the amount of watching television and later aggressive acts actually does not exist.
2 Talking about a longitudinal design. The data were collected as part of a long-term project, the Children in the Community Study. The knack of research projects like this, no doubt, is their longitudinal design. In this case, if you depend on the usual survey, you would have to ask your respondents at the age of sixteen if they had committed any act of aggression since age fourteen as well as, in retrospect, how many daily hours they used to spend watching television at age fourteen. It is clear that such a design risks a considerable memory bias in respect to the reports about television consumption. Longitudinal projects avoid this risk by observing or interviewing their respondents at the actual relevant moment. Of course, this will take more time and energy as you will have to approach the same respondents several times over a longer period. In this respect the New York project seems impressive. It now spans a full twenty-five years, as the project started in 1975 and the children under study were about six years of age. The study on television and aggression regards the period between age fourteen and age thirty. Data were collected in 1982 — at mean age fourteen — in 1985-1986 — at mean age sixteen — and in the period between 1991 and 1993 — at mean age thirty. This seventeen year period was mentioned frequently by the press as to strengthen the power of the research's argument. The article itself, however, tells quite another story.
  Describing their sample, the authors write: "... for whom data were available through 1991-1993 regarding television viewing" (Johnson, 2002: 2469). So it seems that the data concerning television viewing, or at least some part of it, were not collected in 1982 at age fourteen nor even in 1985 or 1986 at age sixteen, but during the next interviewing period, in the years between 1991 and 1993, when the respondents were actually about twenty-two years old. If that is a correct interpretation of what the authors say in their article, it means that there exists a considerable gap of eight years between the actual situation of viewing televison and the self-reports of the youths about their viewing habits. The data concerning aggressive behavior appear to have been collected even later still, in 2000, at mean age thirty. Or, as the authors write themselves: "The youths in the study, randomly selected from age-eligible offspring, were administered questionnaires that assessed a wide range of aggressive acts in 2000" (Johnson et al., 2002: 2469). The study, so it seems, could easily have been a usual survey as it foregoes all advantages of a real longitudinal design.
  For the remainder the sample seems to be alright. As the researchers show, the sample is rather homogenous in its composition, which is rather good given the small sample size of 707 respondents. The respondents are mainly white (91%), over half of them are Roman Catholic (54%) and all are living in two counties in the northern New York State area. It is reasonable to say that the results can generalized to populations of modern urbanized societies that don't differ much from the Northern American situation. One question remains. In longitudinal research projects like this, many respondents tend to disappear over time as people may die or migrate. Therefore it is always a great concern for researchers using this kind of design, to keep as many respondents as possible in their samples. Moreover, it is good practice to tell something about the survival rate of the cohort, just in case there is some bias that can be attributed to the loss of respondents. The span of the New York project and the peculiar number (707) of respondents suggests that there has been some amount of loss. Chances are some compensation did take place by adding new respondents afterwards.
  In this case, for instance, there is a potential problem of families migrating from so-called bad neighborhoods in search for better perspectives. As aggressive behavior among adolescents is quite common in bad neighborhoods, we can assume that those families will be the ones with an aversion to it. It may be that the New York research group did lose a relatively excessive amount of these families and replaced them with some other families living in bad neighborhoods. It is clear that such an operation could possibly result in an overrepresentation of reports of aggressive behavior from these quarters. As the researchers do not tell us anything about any loss of respondents, we just will have to take their word for it that their sample is representative.
3 Counting aggressive behavior. Of course, to start with the analysis of the data, the researchers have to show that there really is some kind of relationship between aggressive behavior and the daily amount of consumed television hours. And, indeed Johnson, Cohen, Smalles, Kasen and Brook furnish some figures demonstrating such a relationship. In their article they present two tables giving the percentages of "aggressive adolescents" — one at "mean age sixteen or twenty-two", the other at mean age thirty. So the first table gives a provisory answer to the question if watching more daily hours of television in 1983 at a mean age of fourteen corresponds with a higher incidence of aggressive behavior later on. Below we reproduce part of it, concerning the category of aggressive behavior the New York research group calls "assaults or physical fights resulting in injury," as the other category of more serious, criminal acts — "robbery, threats to injure someone, or weapon used to commit a crime" — proved not significant for boys and girls separately. [2] As this is not very convincing, we will skip this finding for the moment and concentrate our attention to the analysis of "assaults or physical fights resulting in injury" and the total scores for boys and girls for both categories of aggressive behavior. The next table shows the results for the relatively light category of "assaults or physical fights resulting in injury":
aggressive   daily hours television   total
behavior   less than 1   1 to 3   3 or more    

% of boys:   8.9 (4/45)   27.5 (55/200)   41.7 (48/115)   29.7 (107/360)
% of girls:   2.3 (1/43)   8.6 (16/186)   9.3 (11/118)   8.1 (28/347)

% of total:   5.7 (5/88)   18.4 (71/386)   25.3 (59/233)   19.1 (135/707)
  Table 1: Percentage of youth reporting non-criminal aggressive behavior (assaults or physical fights resulting in injury) at mean age sixteen or twenty-two related to daily television time at mean age fourteen
(source: Johnson et al., 2002: 2469)
  That's it! The simple table above comprises the main argument of Johnson and his team (see Table 1). It sketches a relation between an independent variable — daily hours of televison at age fourteen — and a dependent variable — aggressive behavior in later life at age sixteen or twenty-two. The original indication of aggressive behavior was "aggressive acts reported at mean age sixteen or twenty-two." The first question we encounter here, is the work definition of the dependent variable. Phrased otherwise, what does aggressive behavior — and therefore a rise in aggressive behavior — mean in the context of this research project? The answer can be short: a youth is taken to have been aggressive as he or she has shown aggressive behavior at least once in all those years. To measure this Johnson and his team took both the accounts of mother and child, reporting aggressive between age fourteen and age sixteen or twenty-two. Information derived from police records was added to these self-reports: "In 2000, data were also obtained from New York State and Federal Bureau of Investigation records regarding arrests and charges for adult criminal behavior" (Johnson et al., 2002: 2469). So, the scores of aggressive behavior in this research project do not indicate the amount or weight of aggression for a particular youth, but it just tells us if a boy or girl at least once has shown at least some form of aggression between the age of fourteen and sixteen or twenty-two. So the original heading of the table has to be read as "percentage of youngsters reporting at least one aggressive act at mean age sixteen or twenty-two."
  It seems the researchers had some problems filling the cells of their tables. Now it is always a problem to find facts like aggressive behavior in a standard population. Things like aggression or criminal acts tend to follow a loglinear relationship. Only a minor percentage of the population is responsible for the greater part of all aggressive and criminal acts. Estimates for boys in adolescence are that about seventy-five percent of all criminal and serious aggressive acts are committed by only five percent of the age group. The remaining twenty-five percent of crimes are just incidental acts by boys who slip into it only once or twice. Several facts indicate that the authors had indeed trouble finding enough observations of aggression amongst the adolescents in their sample. For instance they were taking the reports of the mother as well as the children themselves, and "... considering [aggressive acts] present if reported by either informant" (Johnson et al. 2002: 2469). Also the low numbers of added cases at age twenty-one indicates, that the researchers had problems filling the cells of their tables. Though it is not easy to find, the article actually does offer some information on this point. Between the age of sixteen and the age of twenty-two, so we can deduce from some supplementary material on the Internet, there are only an added 39 new cases of aggressive behavior — over a period of six years that is about six on a yearly base and it certainly is a small percentage of the whole sample.
aggressive   daily hours television   total  
behavior   less than 1   1 to 3   3 or more    

A: % of total:   5.7 (5/88)   18.4 (71/386)   25.3 (59/233)   19.1 (135/707)   .022
B: % of total:   3.4 (3/88)   9.6 (37/386)   14.6 (34/233)   10.5 (74/707)   .013
C: % of total:   5.7 (5/88)   22.5 (87/386)   28.8 (67/233)   22.5 (159/707)   .024

  Table 2: Percentages of youth reporting categories of aggressive behavior: (A) non-criminal aggressive behavior; (B) criminal aggressive behavior; (C) total of A and B; all at mean age sixteen or twenty-two related to daily television time at mean age fourteen (source: Johnson et al., 2002: 2469)
  The more serious category of criminal aggressive acts — including robbery, threats to injure someone, or the use of a weapon to commit a crime — does not seem to add much to the number of adolescents showing aggressive behavior either. Table 2 above shows the outcomes of the research project for both categories of criminal acts. Because of the overlap the summing up of the cases of the lighter (A) and the more serious category (B) will only add 24 new cases to the files: sixteen cases spending between one and three hours and eight cases spending three hours or more on television a day. So, all in all, the research project offers us a very rough and potentially exaggerated indication of aggressive behavior.
4 Imploding television time. Now, what of the independent variable in the equation: the daily amount of consumed television time? First, of course, we have to mention that the New York research group did not measure the amount of violence on television directly. The researchers just take it for granted that watching television equals being exposed to images of violence and that watching more hours of television will be more of the same. To ground their case they only give some circumstantial evidence. At the start of their article they refer intensively to the long-term Annenberg research project of Gerbner and his colleagues for the years between 1967 and 1989 (Gerbner and Signorielli, 1990; Gerbner, Morgan and Signorielli, 1994). For that period the Annenberg research tells us that US prime-time television programming on average showed three to five violent acts per hour and that children's Saturday morning programs on average showed about twenty to twenty-five violent acts per hour. We can take that to be an indication of the amount of violence on television in 1982 — the year when the New York respondents were fourteen years of age. However, it will not tell us if the adolescents under study really were all watching programs of a violent nature. Moreover, it does not inform us about the cumulative effect of watching more hours of television a day. Is being exposed to about four acts of violence on television — we take this to be the prime-time mean for people watching less than hour — less problematic than being exposed to about ten acts of violence — the prime-time mean for watching between one and three hours? And how much does the impact of television violence increase when the amount of televison time is stretched out to three hours or more? As the New York research group only offers information about the sheer amount of consumed television time, we have again only a very rough indication.
  A second and related problem arises, where the researchers implode the scores between one and three hours of daily television time into one category. The units in which television is aired, however, are much smaller than that. An episode of a cartoon series such as "Dragon Ball Z" — by the way a serious source for concern for many educationalists — takes half an hour. The same time is taken by most episodes of comedy series like "Dharma and Greg". Soap series like "The Bold and the Beautiful" and most adventure series, ranging from the rather peaceful "Star Trek" over the humorous "Martial Law" to the suspense-filled "La Femme Nikita", all will take an hour. Even movies like the fearful "Rambo" or the more innocent classic Disney animation "Snow White and the Seven Dwarfs" will just take about one hour and forty-five minutes — all including commercials. So merging all television time between one hour and three hours into one category will not discriminate between an adolescent looking at just one series episode and a more heavy youthful television consumer watching one complete movie and yet one series episode. There really is no sound theoretical argument to take such large steps in counting television time. Surely, the New York research team must have had more specific information about the daily amount of consumed televison time in its data set. One can only assume they had special reasons to merge all television time between one and three hours into one category. As a social scientist and researcher myself, I can think of only one reason: heightening the association between television time and aggressive behavior to attain a sufficient significant relationship. Table 3 below shows what this means with a fictive example.
  daily hours television R df Chi²
  less than 1 1 to 2 2 to 3 3 or more        

A: 8.9 (4/45) 27.5 (55/200) 41.7 (48/115) .221 .049 2 17.774
B: 8.9 (4/45) 27.8 (25/90) 27.3 (30/110) 41.7 (48/115) .205 .042 3 17.780
C: 8.9 (4/45) 32.7 (103/315) .172 .030 1 10.686
D: 21.5 (29/135) 34.7 (78/225) .140 .020 1 7.002

  Table 3: A fictive data example of data compressing
(R and Chi² are significant at the 0.01 level for all rows)
Table 3 above shows what happens if we merge categories of television time. As a result the variance in the scores of aggressive behavior as explained by the amount of television time will vary according to the resulting smoothness of the patterns and the sharpness at the extremes. A fuzzy four-category system, discriminating between one hour, two hours, three hours and over three hours of television time, will explain 4.2% of variance in the number of aggressive acts — see row B. Remember, I just made up the data of rows B and D. [3] Row A represents the real data of the New York research project concerning boys and less serious aggressive acts. The pattern here is sharper, giving more weight to the extremes. Because of that, we here find an explained variance that is slightly higher: 4.9%.
  In social science merging initial scores of a variable is not considered to be good practice, especially as it results in very small versus very great categories. That amounts to proving that small groups deviate from the mean represented by the far more greater group. For that reason alone rows C and D represent bad examples. In row C those watching less than one hour of television are distanced from all other ones. This combination of categories of television time will only explain 3% of variance in aggressive behavior. Maybe this option would have been a good choice for the New York researchers, though, as in the end this proves to be the significant cutting point of their data. Of course, the result would not have looked very neat with such different group sizes. And as we will see later on, the New York had yet another very good reason to keep adolescents watching three hours or more of televison a day apart as a separate category. In row D, were the cutting point is drawn between all youths watching less than two hours and all youths watching over two hours of television time, the differences in television time will only explain 2% of variance in aggressive behavior. If the figures of row B resemble the original data, the New York researchers have chosen the optimal solution by merging television time between one and three hours into one category. The profit of 0.7% explained variance at first sight may seem only marginal, but — as we shall see — the relations between watching television and aggressive behavior really lie in this order of magnitude.
  Their work definitions of "aggressive behavior" and "television time" offer the New York researchers some significant relationships, but — as we have seen — these are not very strong. For the category of more serious aggressive behavior the differences in hours of daily television consumption not even are significant for boys and girls separately. For the same reason we can eliminate girls as a risk group in the category of less serious aggressive behavior. Only for the boys in this category of aggressive behavior there remains a significant relationship between acts of aggression and the daily amount of hours spent before the television set. Out of 360 males almost one third (107 out of 360) had been involved in some form of light aggression at age sixteen or twenty-two. Moreover, for boys there seems to be a severe rise in relation to the daily amount of television time. The numbers rise from 8.9% (4 out of 45) for those watching television for less than an hour each day, to 27.5% (55 out of 200) for those watching television for one to three hours a day, and to 41.7% (48 out of 115) for those watching television for over three daily hours. At first sight these numbers may seem rather impressive, and there really seems to be a linear relationship. The correlation coefficient, however, comes out rather low (r = .221; p < .001). Though significant, this correlation amounts to an explained variance of only (.221² =) 4.9%. NBC News (2002) headed its report of this news item with the line: "Study of teens shows boys especially at risk." As the figures show, they better should have written: "If there is any risk involved, it regards only boys."
  Clearly all aggressive behavior amongst youths can not be attributed exclusively to the amount of couch-potatoship alone. Taking boys and girls together and amassing all categories of aggressive behavior, the variance explained by television time is only 2.4%. Except for watching television, there must be other variables at work, causing or attributing to aggressive behavior. Moreover, these yet unknown variables may also contribute to the amount of watching television. If so the relationship between daily televison time and aggressive behavior may be confounded by these so-called third variables. If boys, for instance, watch more television and also are more inclined to aggressive behavior than girls, gender differences may explain both the amount of television time and aggressive behavior. The relationship between television time and aggressive behavior then may even be completely spurious. Luckily there are some statistical techniques that will help researchers in controlling for the influence of this kind of variables.
5 Controlling for third variables. "The association," the article's abstract tells us, "remained significant after previous aggressive behavior, childhood neglect, family income, neighborhood violence, parental education, and psychiatric disorders were controlled statistically." How to explain away so-called third variables? The answer is: by treating them as covariates. For an example, let's say we found a small but significant correlation between television time and aggressive behavior of .155; seemingly implying that (.155² =) 2.4% of variance in aggression can be explained by the observed differences in hours of television time. By the way, that is exactly the amount of explained variance the New York research group found for all aggressive acts (see Table 2). Now take "low family income", or better said "poverty", as a third variable. We know that poverty and related circumstances as living in bad neighborhoods will heighten the change of aggressive behavior. It is also known that poorer people and their children will spend more time watching television on a daily base than people with more economic and cultural capital at their disposal. For the economic and intellectual rich there are many alternatives for television such as theaters and art exhibitions. So we can assume poverty to be related to aggressive behavior as well as to television time. Just for the example, let we say that poverty correlates significantly with aggressive behavior (r = .157) and with television time (r = .112). Given these figures and helped by a statistical computer program we can easily compute that the total variance within aggressive behavior is explained by poverty and televison time for about 4.1%. Introducing poverty as a third variable has heightened the total amount of explained variance. Now, what part of this can be attributed to the influence of both variables separately?
  The trick of controlling for a third variable lies in checking the amount of variance in aggressive behavior that can be explained by poverty alone. Helped by a simple, but sophisticated statistical formula, we subtract the variance in aggression caused by television time from the total variance in aggressive behavior. As a result we get what is called a residual, indicating the amount of variance in aggressive behavior that is not explained by television time. Next we do the same for poverty by the subtracting the variance in television time from that in poverty. We now can compute the correlation between both residuals. In our example this correlation — called part correlation — between the residuals of poverty and aggressive behavior comes out at .140. We now know that poverty alone explains (.140² = ) 1.96% of the variance in aggressive behavior. Turning the tables we can do the same for the relationship between televison time and aggressive behavior. In this case the result comes out at about the same percentage. Separately television time will explain 1.9% percent of the variance in aggressive behavior. Taken together that will take care of 3.86% of the explained variance.
  What happened to the remaining 0.24% of variance in aggression? Well, this has to be attributed to the interaction of poverty and television time. As poverty is related to television time and television time in turn to aggressive behavior, some of the found aggressive behavior must be caused by poverty indirectly by way of its effect on watching television. As we have said, in our example the total explained variance amounts to 4.1%. So from the original amount of the variance, explained by television time only a small percentage (0.24 / 2.4 = 10%) is attributable to the interaction between poverty and television time. Based on that figure we now can recompute the correlation between poverty and television time as the square root of 10%, which is .316. Roughly said, this is the arithmetic operation we perform when we treat poverty as a covariate. In scientific articles this usually will be reported in figures like the one we see below:
  Figure 1: A causal model for the relationship between televison and aggressive behavior with poverty as a third variable
  Figure 1 presents the outcomes of our example and everything we wrote down now can be inferred from this figure. As we see, the initial correlation between poverty and reported aggressive behavior has diminished. Instead of an r = .157 we now find r = .140, implying that poverty by itself explains (.140² =) 1.96% of variance in aggressive behavior. The correlation of poverty with television time has risen from r = .112 to r = .316. By its relation to television time, poverty explains, we can deduce, an added (.316² * .155² =) 0.24% of the variance in aggressive behavior. Summing things up, poverty now directly and indirectly is responsible for 2.2% of the explained variance in aggressive behavior, leaving (4.1% - 2.2% =) 1.9% as the influence that can attributed to television time alone. One can easily see that it is mainly a theoretical decision how to frame the model. We can, for instance, switch poverty and watching television. Though one can think of some reasons why a lot of television watching will make you poor, on theoretical grounds this is not very plausible. The present computations, which by the way come close to the real data of the New York research project, may seem rather unintelligible. Rest assured, you can forget them for the moment. The main thing to keep is mind, is that we now can maintain that television time, controlled for the effects of poverty, still explains about 1.9% of variance in aggressive behavior.
  Remember, the daily amount of time spent on television initially explained 2.4% of variance in all aggressive acts. Stepping back from 2.4% to 1.9% is not a considerable reduction. However, in the present model aggressive behavior as well as the daily consumption of television time are not yet fully explained. There can be other factors attributing to both of them, which — stepwise introduced into the equations — may reduce the explained variance somewhat more. These factors must have a relation with both television time as well as aggressive behavior. It is good practice in social research to leave factors not significantly relating to each other out of the modeling. For that reason, as we now shall see, the New York research team had to dismiss a lot of potential covariates from their analysis.
6 Dismissing some covariates. The New York research project acknowledges the importance of third variables by controlling for no less than eleven of them, which on a conceptual level we can group into three large categories. The first one we can label as "severe educational experiences". This category includes: (1) "aggressive peers during childhood", (2) "frequent school violence during childhood", (3) "harsh punishment during childhood or adolescence", and (4) "childhood neglect". In their supplementary material on the Internet the New York researchers provide some information, proving that all of these variables are related to aggressive behavior. However, except for the last one, they show no significant relationship with the daily amount of television time. As I have said, the golden rule is to keep these variables out of the analysis and that is exactly what the researchers do. Of course the lacking significance can be blamed to a wrong way of measuring these traits or to other flaws of the instruments. For the time being, however, we may conclude that, as far as the figures go and except for "childhood neglect", there is no relationship between severe educational experiences at the one hand and television time at the other. Otherwise said, there is no evidence that children and youths are inclined to spent more hours watching television a day because of these conditions and thus become more aggressive.
A second category includes personal traits. From these (5) "male sex" — according to your liking due to male hormones or to gender specific life styles — shows a very strong relationship with aggressive behavior. A result that is not very surprising. However, because gender is not related to the amount of consumed television time, it also had to be removed from the analysis. The same goes for (6) "low verbal intelligence", which in turn is related to television viewing but not to committing any aggressive acts. The remaining two variables, though, prove to be relevant. About (7) "earlier aggressive behavior" the authors do not offer us any factual information, but as they include it in their analysis we can assume it had at least some relationship with later aggressive acts as well as television time. The last variable in this category concerns (8) "psychiatric disorders", which exerts influence on both the amount of television viewing and aggression. As we can deduce from the information in the supplementary material no less than 303 out of the 707 respondents (or 42.9%) can be included into this category. [4]
  Up to now there are only three variables left as covariates: "childhood neglect", "earlier aggressive behavior" and, whatever that may be, "psychiatric disorders". The third category of covariates adds another three relevant variables. It comprises variables related to the socio-economic background of the respondents: (9) "low parental education" defined as "less than a high school education for either parent", (10) "growing up in an unsafe neighborhood", and (11) "low family income" which better can be labeled "poverty" as it was defined as "mean income below the US Poverty Level" (Johnson et al. 2002: 2469). Though the authors refrain from telling, we can trust those factors are correlated. Moreover, all these variables are related both to aggressive acts and daily hours of television consumption and thus seem fit for introduction as covariates into the relationship between television time and aggression. But wait, there is something strange going on with the last one.
low family
youths for whom aggressive acts
were reported at mean age
    +     -        
  n   % n   %   n %

+ 32   49.2 33   50.8   65 9.0
- 166   25.9 476   74.1   642 91.0

  198   28.0 509   72.0   707 100.0
  Table 4: Low family income at age fourteen and aggressive acts at age twenty-two (source: Johnson et al. 2002, Supplementary material)
  Ever wondered yet, what Johnson and his team mean when they speak about the amount of aggressive acts "at mean age sixteen or twenty-two"? Table 4, reconstructed from information hidden in the corners of their supplementary material on the Internet, provides the answer. As we can see, suddenly the total amount of aggressive acts here has risen from the original 159 to 198 cases. We may assume that this last figure refers to the number of aggressive cases at age twenty-two. A curious fact is that this table is the only one of which the total deviates from the original one. So we may also surmise that the last part of the cryptical indication "at mean age sixteen or twenty-two" refers to the use of just this covariate in the analysis. Now, why did the New York researchers — for this variable and for this variable alone — use the figures at age twenty-two instead of those at age sixteen? Up until now the data manipulations of the New York researchers may be called flaws, but it is here that the reworking of their data becomes fraudulent. Let's try to reconstruct the facts and figures.
7 Reconstructing some data. At age twenty-two almost half of the boys and girls from low income families, Table 4 tells us, had shown some aggressive act at age twenty-two — to be more precise: 32 out of 65. How many of them had committed at least one aggressive act at age sixteen? Let's do some statistical detective work in reconstructing the data for age sixteen, as used by Johnson and his team. For the totals and subtotals the supplementary material offers the necessary clues. Table 5 below shows the results of our reconstruction:
poverty aggressive   daily hours television   total total
  behavior   less than 1   1 to 3   3 or more   at 16 at 22

+ +   1     10     17     28   32  
+ -   7     11     19     37   33  
  subtotal:     8     21     36     65   65

- +   4     77     50     131   166  
- -   76     288     147     511   476  
  subtotal:     80     365     197     642   642

  total:     88     386     233     707   707
  Table 5: Reconstructed data describing the relation between low family income (poverty), daily hours of television at age fourteen and aggressive acts at age sixteen as used by Johnson et al. for their analysis
The red colored figures in Table 5 are derived directly and indirectly from the tables and parameters in the Science article. Of the total of 198 aggressive cases at age twenty-two there are 32 coming from low income families. For age sixteen this number, it seems, was proportionally lowered to 28 out of 159. [5] So at age sixteen 37 out of 65 youths from low income families were qualified as not (yet) aggressive. Given a total number of 159 aggressive cases at age sixteen, 131 aggressive cases must have come from better financial quarters, leaving 511 non-aggressive ones from a similar economic background. The subtotals for the categories of consumed television time can be reconstructed as 8, 21 and 36 for the low income group and as 80, 365 and 197 for their better-off peers. From the Science article we also know that the total of young people under study who had shown some form of aggressive behavior at age sixteen adds up to 5 for those watching television for under one hour, to 87 for those watching between one and two hours, and to 67 for those watching three hours or more — totaling to 159. So we only have to guess two cells in the table and we can almost automatically fill in all the other ones. Here we have set the number of the low income group having shown some aggressive behavior and watching under one hour to 1, and the number of the same category watching over three hours to 17 (for the supposed figures at age twenty-two see Appendix 1).
  How to interpret this result? At first sight the number of 28 "aggressive" cases among youths from lower income families at age sixteen has grown to 32 at age twenty-two. So at age twenty-two an added (4/65 =) 6% of the young people with a poor background had shown some aggressive behavior, at least so it seems. The number of youth with better financial resources shows a comparable increase: here we find an added 35 cases of aggressive behavior, or (35/642 =) 5.5%, again so it seems. Computing backwards from age twenty-two, however, does not mean that at age sixteen these proportions really were the same. This curious operation of imputing the data of age twenty-two into those of age sixteen makes things very complex. It amounts to saying that some youths who possibly did not show any aggressive behavior at age sixteen, should and would have done so looking backwards in time from what they had done at age twenty-two. And, consequently that other youths who really did show aggressive behavior at age sixteen, did not because they did not fall back into it in the intervening years between age sixteen and age twenty-two. This of course is really crazy statistics. We can assume the authors had some reason for doing this. The only reason I can think of, is that the real figures at age sixteen did not show any relevant associations between poverty and aggression. So we can assume the real figures must rather have resembled those shown in table 6 below:
poverty aggressive   daily hours television   total total
  behavior   less than 1   1 to 3   3 or more   at 16 at 22

+ +   1     5     10     16   32  
+ -   7     16     26     49   33  
  subtotal:     8     21     36     65   65

- +   4     82     57     143   166  
- -   76     283     140     499   476  
  subtotal:     80     365     197     642   642

  total:     88     386     233     707   707
  Table 6: Reconstructed data describing the relation between low family income (poverty), daily hours of television at age fourteen and aggressive acts at age sixteen — showing no significant correlations between poverty and aggression
It is only an educated guess and the real data could have looked somewhat different, but not by far (see Appendix 2). If our researchers really had figures resembling those in Table 6 they would not have found any significant relationship between poverty and aggression. They would have had to drop this covariate from their equations. It may be clear that this makes their analysis not very trustworthy, to say the least. But, there even is a more serious problem, which makes the whole operation not only inproper but also fraudulent. If we compare the figures of Table 6 with those in Table 5 we see, that the rise in aggressive behavior between age sixteen and age twenty-two is mainly due to youths from low income families. Among those youths there now are six new cases of aggressive behavior in the category of one to three hours of television time and ten new cases in the category watching television over three hours. That is a rise in aggressive behavior of (16/65 =) 25% among youths from low income families against only (12/642 =) 2% among youths from higher income families. [6] If our reconstruction is correct, this implies that a higher incidence rate between age sixteen and age twenty-one of youths from low income families, due to their circumstances, will almost fully explain the differences between both tables. If so, silently imputing data from a later age into the analysis of an earlier one clearly comes down to a forgery of the data set, in short a fraud.
8 An artefact of method. If, as we think, the original data about "low family income" at age sixteen did not prove significantly related to "aggressive behavior", why didn't Johnson and his cowriters just simply do away with it? Why did they need this variable so urgently as to go to such complex data manipulations? One should expect the researchers would have been glad to remove "low family income" from their equations, as each new significant covariate will diminish the amount of variance to be explained by "television time". If so, of course, they would not have been able to say they had controlled the relationship for income, which would have diminished the power of their argument considerably. Moreover, due to some special characteristics of the data, the reconstructed "low income" variable has some nice advantages, at least if you want to to prove the influence of television time on aggressive behavior. In the imputed scores "aggressive behavior" rises considerably in the second and, more pronounced, in the third category of television time. [7] Because, as we shall see, the imputed scores on "low family income" are inversely skewed with those on "aggressive behavior" they will isolate some variance in "aggressive behavior" uniquely to be explained by "television time".
  Something more is going on with the "low family income" variable — and by the way with most other variables related to this one as well. They are rather skewed in relation to "television time". Of these variables "low family income" is the one showing the most conspicious pattern. The main reason is that the "low income" variable only differentiates between two groups whereas the "television time" variable comprises three categories. Moreover, because of the small percentage of poor people and the relationship between poverty and television time, adolescents from "low income" families threaten to be overrepresented in the third and highest category of "television time". As a consequence youths from "higher income" classes will be overrepresented in the first two categories of "television time". This is not to say that the New York sample is not representative. Admittedly the percentage of adolescents from low income families in the sample (9%) corresponds roughly with that of the population. Based on the numbers of 1997, the US Census Bureau (2002) estimated that 15.6% of the adult population of New York City — compared to 13.3% for the US — and 24.7% of the New York City children — compared to 19.9% for the US — have family incomes below the US Poverty Level in 2002. No, the real problem is, that in relation to "television time" the scores on the "low family income" variable come out as inversely skewed with those on "aggressive behavior". The plotted lines in Figure 2 below visually show what this implies.
  Figure 2: Graphical representation of the relations between "daily hours television" and "aggressive behavior" and "low family income" based on the figures from table 5 (standardized scores).
  Because of the inverted relationship between "low family income" at one hand and "aggressive behavior" at the other, the introduction of the variable "low family income" as a covariate will accentuate some left-over variance in the residual of ""aggressive behavior" to be explained by the residual of "television time". This will be equal to the relative differences between the percentage of "low family income" youths (blue line) and the percentage of aggressive cases (green line) in each category of "television time". As these differences are related to "television time" (red line), this variable will explain most of this variance after "low family income" is introduced into the equations. Now the variance at the top of television time, ranging over one hour and more, will be reduced by the introduction of the covariate. The differences between the second and third category no longer are significant. At the start however, the differences between the first and the second category, will be left to be explained by television time. Here, this comes down to the 1.9% of variance left to "television time", which we found above (see Figure 1). That is the basis on which Johnson and his team can maintain that television time at age fourteen explains aggressive behavior at age sixteen — or twenty-two — even after the data are controlled for "low family income".
  Now what does "televison time" really explain after the introduction of "low family income" as a covariate? Most poignant here, as we can see in Figure 2, is the difference in the first category of this variable where we find relatively less youths showing "aggressive behavior" than there are youths from "low income families". So, simply said, if "television time" does explains anything all, it has something to do with the relative low level of "aggressive behavior" in respect to the relatively high percentage of "low level income" of youths watching less than an hour. In this category both youths from "low income" families and youths from "higher income" families show less aggressive behavior than those in the other categories of "television time". However, does this really imply that this difference can be attributed to the fact that they watch less television? We can think of many good reasons, why this small group deviates in "aggressive behavior" from the more common pattern, represented by youths watching television for over one hour a day.
9 Reversing the conclusion. If this reconstruction of their data analysis is correct, we can reverse the conclusions of the New York research team. Instead of a causal relationship between watching television and aggressive behavior among adolescents, this is what Johnson and his fellow researchers really found:
  1. Except for the effects of the manipulated "low family income" variable, at this point we may assume that, as far as the data of the Children in the Community Study go, any relation between televison time and later aggressive acts is spurious. So, until proven otherwise, we can maintain that there is no real relationship between the "daily amount of television" or even "television violence" at age fourteen and later "aggressive behavior".
  2. There is a small group of boys and girls watching television for less that an hour each day who also show few aggressive behavior, at least — we must add, because of the imputation of the figures at age twenty-two — they will not show much additional aggressive behavior between the age of sixteen and twenty-two. This group recruits its members not only from better-off families, but also from families with incomes below the US Poverty Level.
  The first of these conclusions may be clear by now. The second one, however, deserves some more thought. As we have seen, instead of indicating that watching television will increase the risk of aggressive behavior, the figures show that if you belong to the slight percentage of adolescents — 88 out of 707, or 11% of the sample — watching almost no television, you will probably show no aggression at all. Which is really something else. Discussing this, we will have to keep in mind that the low incidence rate of these boys and girls in "aggressive acts" only covers 1.9% of the variance in all "aggressive" behavior in the population. We also have to consider that the low score on aggressive behavior possibly is related to the fact that the youths from "low income" families in this category are less accident-prone to "aggressive behavior" between age sixteen and age twenty-two — and probably already at an earlier age — compared to their peers spending more time watching television. There seems to be something special about this small group of boys and girls, both from low and high income families.
  For youths in this category living in "high income" families the explanation may be rather simple, as they may be recruited from really high income classes. The "low income" variable of the New York research project does not take these classes into account, and this fact is mainly responsible for the skewedness of the variable. Looking back at Figure 2 it is easy to see that a more differentiated income variable, with an added middle class and higher class category, probably would have resulted in a less skewed blue line, leaving less to be explained in "aggressive behavior" by "television time". High income classes, comprising professionals and intellectuals, moreover cultivate quite another lifestyle with the consumption of high culture competing successfully with television and also with less feel and need for physical aggression. If we reckon only half — about 40 — of this group to live in really high-income families, the total amount of explained variance will rise from 4.1% to 4.9%, but the direct impact of television on television, though still significant, will dwindle to almost nothing — even without controlling for other covariates. This is what the British media expert Guy Cumberbatch, from the Communications Research Group based in Birmingham, said about the Children in the Community Study (BBC News Services, 2002):
  "How many families do you know where children watch this amount or less? These are highly unusual families — the kind who are more likely to be taking their children to art galleries and museums. And there are so few of them compared to the rest of the children studied ... This is a case of torturing the data to make it fit a theory."
  In regard to youths from the high income group Cumberbatch is quite right. The adolescents from "low income" families, though only 1% of the sample, however, are also important for the variance explained by television, because of their relatively low incidence rate in aggressive behavior. For the boys and girls of this category we have to find yet another explanation. We can think of two. For one these girls and boys may be from families of the impoverished rich — for instance one-parent families with divorced or widowed mothers trying to keep to their former life style. For the other they may be from poor but religious families with faith and conviction keeping them away from watching television as well as from aggression. If we reckon two-thirds of them — only 6 cases — to be the exponents of such a traditional life style deviating from the standard pattern, the remaining tiny direct impact of television on aggressive behavior no longer will be significant. For all practical purposes the relationship now will look like the one, shown in Figure 3 below.
  Figure 3: A causal model for the relationship between "daily hours television" and "aggressive behavior" and "low family income" after introducing "life style" as an intermediating variable
  As we can see, there is no direct relationship left between "television time" and "aggressive behavior". This is what Jonathan Freedman, a psychologist working as media researcher at the University of Toronto, said about it (Vedantam, 2002):
  "It has nothing to do with TV — it has to do with lifestyle. People who watch more than three hours of TV are different than those who watch less than an hour."
  Freedman's remark is right to the point. Applying these same methods, the New York research team could have proved the impact on "aggressive behavior" of almost every element of a lifestyle deviating from the special conditions of the small group watching television for less than an hour a day — eating junk food for instance. Special conditions may also explain the one other effect the New York research group found when looking for the effects of television at an even later age. This time they only found an effect for women. For those women watching television for three daily hours or more at mean age twenty-two will result in a higher rate of aggression at mean age thirty. Though the aggressive cases only amount to a small percentage of 4% (15 out of 347) of all women, television now not only attributes significantly to assaults or physical fights but also to more serious aggressive behavior. As we know that violence for women mostly is associated with private life, we may assume that the aggressive acts and the flight towards television both may be caused by the special conditions of an unhappy marriage or even violence at home.
  Let us also not forget what the New York research project does show about aggression. Though their numbers may be exaggerated, the relatively large proportion of American boys — 33.6% if we add up more and less serious acts of aggression — that has been involved in acts of aggression of some sort at twenty-two, is rather high. At least we now can say this percentage can not be attributed to the consumption of television time at an earlier age. The main thing explaining this, the research project inadvertently indicates, are the social conditions connected with male adolescence, with poverty and social exclusion and with individual psychiatric difficulties in living up to the standards of modern life. We must also keep in mind that we are not talking about little children, but about youths making their own choices of what television has to offer and that these choices may be steered by these factors. In the New York research the effects of male gender" and of severe educational experiences at home and at school — "aggressive peers during childhood", "frequent school violence during childhood", "harsh punishment during childhood or adolescence", and "childhood neglect" — all proved to be significantly related to "aggressive behavior". Next to that there are the indications of social exclusion: "low parental education", "growing up in an unsafe neighborhood", and "low family income" — and the reported level of psychiatric disorder. These elements prove to be the main causes of aggressive behavior. Then why does the New York research group insists on blaming television?
10 Thinking things over. The quick acceptance of the conclusions of the New York research by many papers and news stations fits in with the moral panic about television, dating back to 1954-55 when the United States Senate Committee on the Judiciary — more precisely the Subcommittee to Investigate Juvenile Delinquency — held a series of hearings on the impact of television programs on juvenile crime. Since its early beginnings people are eager to blame television for the social problems surrounding them, especially in regard to children and youth (Cumberbatch, 1994). It took some time, but in the 1980s many social scientists jumped the bandwagon. Since the start of the 1980s many scientific organisations took to their professional duty of civil responsability to issue warnings about the effects of television on children and adolescents. In 1982 the National Institute of Mental Health (1982) warned parents: "Violent programs on television lead to aggressive behavior by children and teenagers who watch those programs." In February 1985 an urgent resolution of the American Psychological Association (1985) was published, followed by many other publications. [8] Against this background project leader Johnson of the Children in the Community Study, explaining his results, could maintain (MSNBC News Services, 2002):
  "The evidence has gotten to the point where it's overwhelming."
  Even if his accusations were right, Johnson here clearly is exaggerating the impact of television on actual violence. Others too go to great lengths to prove that they are right in their misgivings about television. Defending the outcomes of his New York colleagues, Bad Bushman — a professor of psychology at Iowa State University at Ames who, with his colleague Craig Anderson wrote a commentary accompanying the New York study in Science magazine — stated for the Washington Post (Anderson and Bushman, 2002; Vedantam, 2002):
  "The correlation between violent media and aggression is larger than the effect that wearing a condom has on decreasing the risk of HIV. It's larger than the correlation between exposure to lead and decreased IQ levels in kids. It's larger than the effects of exposure to asbestos. It's larger than the effect of second-hand smoke on cancer."
  Again the actual situation is grossly exaggerated. Transmission of HIV through unprotected intercourse with HIV-patients is estimated at largest at 3 percent for each contact. Consistent use of condoms has been found to reduce the risk of HIV transmission by 87 to 95 percent. So the remaining risk is about 0.3%, which conforms to a risk reduction of about 2%, mentioned by Bushman and Anderson (2001: 481) elsewhere. Comparing HIV-prevention to television consumption, however, is nonsense. Even we take televison violence to be responsible for at least some real violence among youths, they don't have to watch only once to incur such a risk of falling into aggressive behavior. As the New York project indicates, youngsters can watch years of television for long hours each day with only a small risk of falling into aggressive behavior. Moreover, banishing all violence from all forms of popular culture is not comparable to promoting condoms in an anti-HIV-campaign. It rather compares to issuing an interdict of all sexual intercourse to prevent the disease. The incidence of lung cancer as a cause of death may be high among smokers, but is rather sparse in the whole population — an estimated 7%. Aggressive behavior, on the other hand, at least as measured in the New York research project, seems to be more common. Smoking cigarettes is one of the main causes of lung cancer: over 80% of all cases is due to this habit. Besides television — and again if we take the 1.9% of explained variance of the New York research project as a true measure of televisions impact — there are many other and far more important factors contributing to actual violence. As every epidemiologist knows, one has to cure the most influential factors first.
  One surely can have one's doubts about violence on television, in contrast to the real dangers of urban life it is however far less dangerous. It is this real violence which seems to be at the core of the growing anxieties about television violence, culminating in the US in the debate about the V-chip. From the 1960s on, a growing mass mobility and changing cultural patterns have led to a more intensive and intermingled cultural use of anonymous public places by formerly socially segregated populations, thereby enlarging the incidence of violence and crime. Though it could have been far more, violence and crime in the public arena rose considerably from the 1960s to the 1970s, to stabilize later on (Young, 1999). Because of the anonymity of public space these forms of violence and crime now have become almost unpredictable incidents that can happen to anyone. At the same time people became more sensitized to the potential dangers of public places, as everyone claimed the right to feel her/himself save at any place and at any time. This paradox between a growing need to feel save in all circumstances and the increased risk of unpredictable incidents, so it seems, lies at the bottom of the growing concern about public safety. One can try to reconcile this paradox by removing one of the main causes: social exclusion and severe educational experiences at home and at school. Many people, including Anderson, Bushman and Johnson, however, seem bent on solving the paradox in a more mythical way by creating the image of nostalgic past with no or at least less overt aggression, which has been perverted by the commercial drives of the media. As such, next to a professional attitude of civil responsibility, their stance demonstrates a fear for modernity, motivating their crusade against television.
  This specific combination of a fear for modernity and a professional attitude of civil responsibility may have driven the New York researchers to manipulate their data as to provide the outcomes they wished for. Good intentions, however, are never an excuse for adjusting your data to fit your expectations. The tinkering with the "low family income" variable may be due to a combination of zealotry and incompetence. If not, one may even call it a conscious fraud. Anyway, to the standards of social science this article should never have passed the screening procedure of, nor have been published in a high-profile magazine like Science.
  At the end of this lengthy review we again quote the Canadian psychologist Freedom. This is what he said about the New York research project to CBC News (2002):
  "It doesn't tell us anything we didn't know before. Television violence and television is not harmful. I'm not a great lover of most of what's on television, but it doesn't make you aggressive; it doesn't make you violent."
  As we have seen, a critical look at the Children in the Community Study, fully confirms Freedom's view. The New York researchers, we must admit, have done their utmost best to find an association between televison and aggressive behavior among youths. But even after counting every aggressive act of adolescents over a period of eight years, after massaging their data by combining categories of television time, projecting some of their figures back from age twenty-two to age sixteen, and capitalizing on skewed covariates, they only find a minor relationship, that can easily be explained away as an artefact of their analysis. Though television may not be that bad, studies like these certainly are. It really is bad science and because of the failed screening procedure of the magazine that published it, so we may add, it certainly is bad Science too.
1. Though not having found a linear relationship, the authors offer us an odds ratio based on incremental steps in the daily consumption of television time. The odds ratio here indicates how much more likely it is to find the trait of aggressive behavior among adolescents in each next category of daily television time. For instance for boys and less serious aggressive behavior the article gives an odds ratio of 1.95. This means that if the initial percentage for boys is 8.9%, watching television for between one and three hours a day will raise the percentage of aggressive behavior on a later age to (8.9 * 1.95 =) 17.4%. Again we can expect almost a double incidence rate of (17.4 * 1.95 =) 33.8% for aggressive behavior among adolescents watching for three hours or more on a daily base. Return to text
2. I must admit it is rather unclear to me why threats were incorporated into the second, most serious category of aggressive acts, as words do not harm much. I take it to be serious threats, for instance with a weapon at hand. Return to text
3. Some figures in Tables 3, 5 and 6 are just educated guesses. Of course I dare the research team of the Columbia University to give the ones they used for their original analyses. Return to text
4. Most research in the Netherlands estimates the percentage of psychiatric disorders among youths at about 5% of the population (Rispens et al., 1996). It makes one wonder what is wrong with American youth or — as both parties could report this characteristic — with their parents. Return to text
5. The supplementary material offers a subtotal of 32 youths from low income families and a subtotal of 166 youths from high income families having shown aggressive behavior at age twenty-two. So for age sixteen one would rather have expected reconstructed subtotals like (32/198 * 159 =) 26, (65 - 26 =) 39 and (166/198 * 159 =) 133, (642 - 133 =) 509 (Chi² = 12.59; df = 1). However, this does not conform to the Chi² of 17.41 (df = 1), supplied by the New York research team for the association between low family income and aggressive acts. This Chi², by the way, is not the correct one for the data in Table 4 at age twenty-two, which comes out at 15.99 (df = 1), so it has to refer to the reconstructed data at age sixteen. The nearest subtotals fitting in with the Chi² presented by the New York research group are 28, 37 and 131, 511. For the association between low family income and daily hours of watching television at age fourteen the Chi² is 17.44 (df = 2), conforming to a table with rows 8, 21, 36 and 80, 365, 197. Return to text
6. Part of the inferred increase of aggressive cases among youth from low income families between age sixteen and age twenty-two may also be due to a compensation of loss in the sample (see section 2, paragraph 4). Return to text
7. Next to the one mentioned in section 4, paragraph 5, this is the other reason the New York team had to preserve the third category of "television time" in its data set. Return to text
8. The moral panic in the US about television and violence is marked by a whole series of scientific reports. Next to the congressional hearings starting in the 1950s and continuing till the 1990s, public warnings have been given by the National Commission on the Causes and Prevention of Violence (Lange, Baker and Ball, 1969); the Surgeon General's Scientific Advisory Committee on Television and Social Behavior (1972); the Group for the Advancement of Psychiatry Staff (1982); the National Institute of Mental Health (1982; Pearl, Bouthilet and Lazar, 1982); and the National Research Council (Reiss and Roth, 1993). Reports from the American Psychological Association on this topic include the "Task Force on Television and Society" (Huston, et al., 1992) and the "Commission on Violence and Youth" (American Psychological Association, 1993; Donnerstein, Slaby and Eron, 1994). All of these reports take the stance that watching television for children and adolescents more or less equates to being exposed to violence and that this can led to forms of negative arousal on an emotional and even behavioral level. Return to text
  • Anderson, Craig A., and Brad J. Bushman (2002), "The effects of media violence on society." In: Science, Volume 295, 5564 (29 March 2002), 2377-2379.
  • American Psychological Association (1985), "APA resolution on television violence and children." Retrieved from: American Psychological Association Online Public Affairs, 2002.
  • American Psychological Association (2002), "Violence on television. What do children learn? What can parents do?" Retrieved from: American Psychological Association Online Public Affairs, 2002.
  • American Psychological Association Commission on Violence and Youth (1993), Violence and youth. Psychology's response. Summary report (PDF-format). Washington, DC: American Psychological Association.
  • BBC News Services (2002), "TV - violence link disputed." In: BBC News, Health, Friday 29 March, 2002.
  • Bushman, Brad J. and Craig A. Anderson (2001), "Media violence and the American public. Scientific facts versus media misinformation." In: American Psychologist (PDF-format), 2001, 56, 6/7, 477-489.
  • CBC News (2002), "Researchers split over TV - violence ties." In: CBC News, Friday, 29 March 2002.
  • Cumberbatch, Guy (1994), "Legislating mythology. Video violence and children." In: Journal of Mental Health, 1994, 3, 485-494.
  • Donnerstein, Ed, Ron G. Slaby and Leonard D. Eron (1994), "The mass media and youth aggression," In: L.D. Eron, J. Gentry and P. Schlegel (eds.), Reason to hope. A psychosocial perspective on violence and youth. Washington DC: American Psychological Association, 1994.
  • FitzGerald, Susan (2002), "More TV for youths now may mean more aggression later." In: The Philadelphia Inquirer, Friday, March 29, 2002.
  • Gerbner, George, and Nancy Signorielli (1990), Enduring patterns. Television violence profile 1967 through 1988-89.. Pennsylvania: Annenberg School for Communications, 1990.
  • Gerbner, George, Michael Morgan and Nancy Signorielli (1994), The turning point. Television violence profile no. 16: om research to action . Pennsylvania: Annenberg School for Communication, 1994.
  • Group for the Advancement of Psychiatry Staff (1982), Child and television drama. The psychosocial impact of cumulative viewing. New York: Mental Health Materials Center, 1982.
  • Huston, Aletha C. , Ed Donnerstein, Halford Fairchild, Norma D. Feshbach, Phyllis A. Katz, John P. Murray, Eli A. Rubinstein, Brian L. Wilcox, and Diana Zuckerman (1992), Big world, small screen. The role of television in American society. Lincoln, NE: University of Nebraska Press, 1992.
  • Johnson, Jeffrey G., Patricia Cohen, Elizabeth M. Smailes, Stephanie Kasen and Judith S. Brook (2002), "Television viewing and aggressive behavior during adolescence and adulthood." In: Science, volume 295, 5564 (29 March 2002), 2468-2471. The article — abstract; full article; reprint (PDF-format); and supplementary material — can be read and downloaded at the web site of the Science Magazine.
  • Lange, David L., Robert K. Baker and Sandra J. Ball (1969), Mass media and violence. A report to the National Commission on the Causes and Prevention of Violence. Washington : United States Government, 1969.
  • MSNBC News Services (2002), "Just one hour of TV linked to violence. Study of teens shows boys especially at risk." In: msnbc.com, March 28, 2002.
  • National Institute of Mental Health (1982), Television and behavior. Ten years of scientific progress and implications for the eighties. Volume 1. Rockville, MD: US Department of Health and Human Services.
  • Pearl, David, Lorraine Bouthilet and Joyce Lazar (1982), Television and behavior. Ten years of scientific progress and implications for the eighties. Rockville, Md.: US Department of Health and Human Services, National Institute of Mental Health, 1982.
  • Reiss, Albert J., and Jeffrey A. Roth (eds.) (1993). Understanding and preventing violence. Report of the National Research Council Panel on the Understanding and Control of Violent Behavior. Washington DC: National Academy Press, 1993.
  • Rispens, Jan, Jos Hermanns and Wim Meeus (1996), Opvoeden in Nederlands [Parenting in the Netherlands]. Assen: Van Gorcum, 1996.
  • Surgeon General's Scientific Advisory Committee on Television and Social Behavior (1972), Television and growing up. The impact of televised violence. Report to the Surgeon General, United States Public Health Service from the Surgeon General's Scientific Advisory Committee on Television and Social Behavior. Rockville, MD: National Institute of Mental Health, 1972.
  • US Census Bureau (2002), "State and county quick facts: New York." In: US Census Bureau, last revised: Thursday, 7 February 2002.
  • Vedantam, Shankar (2002), "Study ties television viewing to aggression. Adults affected as well as children." In: Washington Post Online, Friday, March 29, 2002.
  • Young, Jock (1999), The exclusive society. Social exclusion, crime and difference in late modernity. London: Sage, 1999.
  2002 © Soundscapes