Nov 052008
ResearchBlogging.orgThe latest evidence in the debate over the effects of video game violence has arrived in the November edition of the journal Pediatrics. Japanese and American psychologists, including well-known media violence researchers Craig Anderson and Douglas Gentile, report that violent video games constitute a causal risk factor for physical aggression. Perhaps unsurprisingly, the gaming internets have already expressed their disagreement with these results via angry blog postings based on secondary reporting (calmer coverage can be found at Gamasutra). A more professional critique has also been offered, in the form of a post-publication peer review by Texas A&M International University Professor Christopher Ferguson. The paper tries to sell itself as a significant piece of new proof, which it is not. Anderson et al. have found an interesting, if weak, correlation that they cannot prove to be causal, due to the limitations of the methods employed.

The study has two key advantages that, in principle, make it a unique addition to our knowledge about the effects of video game violence. Firstly, it attempts to correlate physical aggression (PA) in teens and kids with habitual exposure to video game violence (HVGV) 3-6 months earlier. While the use of a timecourse alone cannot prove causation, long-term correlations are thought to suggest a causal relationship more strongly than instantaneous correlations. Secondly, the study involves several age groups from two countries, the United States and Japan. Although more children play video games in Japan than in the US, the rate of violent crime in Japanese society is much lower than in the US. This has occasionally been held out as disproof of an HVGV-PA link, but all it really establishes is that other factors play a significant role. Therefore it would be interesting to determine whether cultural differences between the US and Japan alter the effect of HVGV on PA.

Three sets of children (two in Japan, one in the US) filled out questionnaires querying their gaming habits and physical aggression levels. Some months later, these same children were surveyed again to see whether their physical aggression levels had changed. The authors found that HVGV levels at the first time point had a weak correlation (r= 0.28) with PA at the second time point. This effect varied significantly over the individual datasets and was strongest in the youngest age group. However, the r value did not exceed 0.5 for any of these datasets.

In layman’s terms, one could see these results as evidence that HVGV predicts between 8% and 16% of the level of physical aggression, depending on the age group and nationality. I caution my readers that this interpretation is an oversimplification that depends on certain assumptions about the data to have any validity. Because no statistics of the underlying matrices are provided I cannot substantiate those assumptions, so this should not be taken as a definitive description of the study’s findings. Statistics (even averages) imply a model, and should not be trusted if it cannot be proved that the model is appropriate.

These results are interesting and indicate that, although the magnitude of the effect may differ between societies, there is nonetheless a universal positive correlation between HVGV and subsequent physical aggression. Despite the elaborate discussion of youth violence in the paper, this does not directly indicate a linkage to criminal behavior. Moreover, this correlation is difficult to interpret due to the study’s numerous flaws.

There are good reasons to wonder whether the interpretation of the questionnaires produced a valid measure of HVGV at all, the assignment of violence level by genre being particularly suspect. A more significant problem may be that HVGV and PA were assessed by different means in every single group. Each group used different delays between surveys, and each involved differently-aged children. This doesn’t necessarily mean that conclusions drawn by aggregating the three are wrong, and the authors contend that agreement across the varying methodologies indicates robustness. However, the differences in method and subjects multiply the potential sources of error considerably. Since the derived correlations are so weak, this is a significant concern. In addition, because the populations differ substantially in respects other than nationality, it is impossible to accurately assess the effect of culture on the relationship between HVGV and PA. Doubtless future longitudinal studies will apply more uniform methods.

This brings me to another weakness of the study. Scientists will occasionally joke that the best possible set of correlational data is the one that contains only two points, the reason being that you are assured of being able to draw a perfect, straight line through all your data. In practice, however, we know that having limited numbers of data points makes our interpretations much more likely to be wrong. A “longitudinal” study involving two questionnaires given a couple of months apart hardly provides firm footing for a long-term correlation or a causal relationship. The authors acknowledge that the study is limited in this regard, but argue that the short wait would most likely depress correlations from their true value. Still, a longer timecourse with more measurement points would be highly desirable.

The authors make no attempt to account for any confounding factors other than gender. They do not seem to have taken data on family situation, peer influence, parental involvement, or school performance, although all of these factors are known to correlate to greater or lesser degrees with both PA and video game habits. If we only wish to establish that there is a correlation between HVGV and PA that’s not a huge problem. However, Anderson et al. clearly mean to establish video games as a causal factor for aggression. In the absence of controls for confounding factors, that is impossible.

Curiously, the authors also do not seem to have measured HVGV at the later time point. One objection to existing research linking HVGV to real-world violence has been that the observed correlations exist because people predisposed to violence choose to enjoy violent media. Testing the hypothesis that PA at the initial timepoint predicts HVGV at the second timepoint seems like an obvious thing to do, if only to squelch this objection. This seems to me particularly worthwhile, because the predictive power of HVGV for later aggression appears to be less than the instantaneous correlation between HVGV and aggression, significantly so for the older group. In light of these facts, the choice not to assess HVGV at the later time seems extremely odd.

Despite these flaws, this research is a step in the right direction. We need longitudinal studies, carefully controlled for confounding factors, over a range of ages and nationalities to parse out the true effects of video games on aggressive behavior in teens and adults. I do not find the present study terribly convincing, and I particularly dislike the more sensationalistic high points of its discussion section. Nonetheless, I hope that the authors will take criticisms like those of Dr. Ferguson into account as they design studies that will more rigorously investigate the causal relationship between HVGV and PA.

Only a particularly obstinate person would deny that there is a correlation between the intake of violent media, including video games, and aggressive behavior. They may inspire aggressive behavior, or serve as an outlet for existing aggression; either way, the correlation ought not be ignored. However, video games are just one, and doubtless not the most important one, of a constellation of potential factors affecting child behavior. Without a genuine analysis of the complicated causal relationships among these it is impossible to provide good advice to parents, doctors, and psychologists. The present study does not fill that gap in our understanding; it is doomed by its single-minded focus on video games and failure to account for confounding factors. While it is of value to know that the correlation between violent video games and aggression transcends national and cultural boundaries, it would be of greater value to know whether excessive playing of violent video games is a cause of aggressive behavior, a result of pre-existing aggression, or both. That is a question this research does not adequately, much less conclusively, address.

C. A. Anderson, A. Sakamoto, D. A. Gentile, N. Ihori, A. Shibuya, S. Yukawa, M. Naito, K. Kobayashi (2008). Longitudinal Effects of Violent Video Games on Aggression in Japan and the United States PEDIATRICS, 122 (5) DOI: 10.1542/peds.2008-1425

Aug 142008
ResearchBlogging.orgHow does the human brain react to the communication of emotion? Does the observation or imagination of emotions have anything in common with the personal experience of them? It is possible that the brain uses a setup in which seeing a person experience an emotion, imagining that emotion, and feeling that same emotion all use completely independent circuitry. Yet since all of these experiences make references to the same emotional state, it is also reasonable to think that some of the pathways are shared. In a recent article from PLoS ONE, a team of researchers uses functional Magnetic Resonance Imaging (fMRI) to determine similarities and differences in the patterns of brain activation following various means of communicating disgust. PLoS ONE is open access, so go ahead and open the article up in another window.

First, a word about fMRI, for those unfamiliar with it. As the name would suggest, fMRI is an elaboration of the standard MRI techniques used image the interior of your body without the use of potentially harmful radioactivity. Neuronal activity in the brain causes a local depletion of oxygen from the blood, followed by a localized increase in blood flow. Because the magnetic properties of iron in the blood change with its oxygenation state, it is possible to detect these hemodynamics using magnetic resonance imaging. Thus, fMRI is able to indirectly detect neural activity, although the fMRI signal lags behind activity by a few seconds. A given fMRI signal also encompasses a large number of individual neurons and therefore can only serve as a rough map to where things are happening in the brain. These temporal and spatial limitations limit the conclusions that can be drawn reliably from fMRI, but the observed correlations can provide valuable insights.

Jabbi et al. used fMRI to map the neural response of subjects to various encounters with disgust. Previous research had shown that a particular region of the brain (the IFO) showed increased activity when subjects either tasted something disgusting, or viewed a short clip of someone else tasting something disgusting. For this study, Jabbi et al. had participants read short scripts (samples can be found in the supplementary materials) intended to make the reader imagine being disgusted, pleased, or not feeling anything. They found that reading disgusting passages induced a neural response in this region of interest, just as it had for the cases of tasting or observing disgust.

While this may seem completely unsurprising, it bears some consideration. The experience of personal disgust differs significantly from the experience of observing disgust in others. Similarly, imagining or reading about disgust creates a very different subjective experience than, say, drinking quinine. Given that these are all quite different feelings, it is somewhat surprising that a single area is activated by all three.

Of course, there is a fine line to consider here — the passages meant to make the subjects imagine disgust may have actually disgusted them. The paragraphs that the authors make available in the supplementary materials are written in second person and involve things like accidentally ingesting animal waste. Because the subjects are reading passages that ask them to imagine themselves being disgusted, and the passages are themselves disgusting, the act of imagination may be contaminated by an immediate personal experience of disgust. In a more elaborate experiment it might be of value to use passages written in the third person. Additionally, it might be useful to employ passages in which the characters, because of particular phobias or personal experiences, are disgusted by items or actions the reader is likely to find innocuous.

Whether the readers where themselves disgusted or not, the overall response in the brain differed for each of the stimuli, as shown by a map of correlated activity (Figure 2). While the area outside the IFO activated by observation was relatively small, both the disgusting taste and the disgusting scripts produced widespread activity relative to a neutral taste or script. In general there was not much overlap between the networks, except for a small region shared by the imagination and experience groups. The authors propose that the similarities of imagining, observing, and experiencing emotion are due to the common activation of the IFO, while the differences between these are due to the largely distinct networks of correlated activity. Different modes of exposure to disgust may therefore act in complementary, rather than independent, ways.

Additionally, this result appears to be consistent with the view that our recognition of observed disgust and our imagination of disgust rely on an internal simulation of our own feelings of disgust. However, these experiments cannot establish exactly what a particular region of the brain is doing, so this remains an open question.

While this research does not indicate whether these results can be generalized to other emotional states, this finding may interest developers of media that make use of multiple modes of communication, specifically video games. Games often rely on video cutscenes to convey story and emotion, but this approach may be wasting a significant amount of potential. The participatory nature of games makes it possible to approach emotional communication not only through the observational route, but also the experiential route.

Consider the case of Agro’s fall in Shadow of the Colossus. Observing the cutscene, and hearing the voice of Wander, the player can understand that Wander feels grief at this event, in much the same way that anyone watching a movie could understand it. Additionally, the emptiness of the game’s landscape and the forced collaboration between the player and the Agro AI has helped to create a relationship between the player and the horse. Thus, in observing Agro’s fall, the player may feel his own sense of grief at the event, increasing the emotional resonance of the moment.

This suggests a possible, if lengthy, experiment. It would be interesting to compare the fMRI profile of a subjects observing Agro’s fall under two conditions: one in which they have actually played the game up to that point, and another in which they have watched the game as a movie, with exploration and battles recorded previously from an expert player’s run. Would the first group have activity in both the observational and experiential networks, or would each group activate a different network? What implications might these outcomes have for the development of emotionally fulfilling games?

Of course fMRI studies are not some holy grail that makes everything clear. The work of Jabbi et al. has given us a rough map to where things are happening, but understanding exactly what is happening and how it is happening will require additional experiments and possibly new investigative techniques. Nonetheless, this is an interesting piece of the puzzle, and perhaps some food for thought.

Mbemba Jabbi, Jojanneke Bastiaansen, Christian Keysers (2008). A Common Anterior Insula Representation of Disgust Observation, Experience and Imagination Shows Divergent Functional Connectivity Pathways PLoS ONE, 3 (8) DOI: 10.1371/journal.pone.0002939

Feb 272008
ResearchBlogging.orgCourtesy of Kotaku I learned of an article on the response of players to violent (and non-violent) video games out of a Finnish lab (1). The study has garnered attention because its results seem to contradict popular rhetoric that games encourage violence and desensitize their players to it. What Ravaja et al. found was that players of a first-person shooter had negative emotional reactions to killing or wounding opponents, and a positive emotional reaction to “dying” in game. These results seem odd, but I feel they’re reasonable and hint at interesting avenues for future research.

First, the methodology. The authors of this paper used several physiological measurements in combination with subjects’ reports of emotional experience and a personality test. In the case of the physiological measurements, skin conductance level (SCL) was used to assess general arousal, and EGM recordings of muscle activity determined positive (zygomaticus, orbicularis oculi) and negative (left corrugator) emotional valence. The researchers studied 36 students (70% male) aged 20-30 playing three (one “training”, two “play”) 5-minute sessions of James Bond 007: Nightfire or Super Monkey Ball 2. If I’m reading the “Event Scoring” section right some of these players had good reason to be frustrated: it looks like someone managed to die 8 times (although it’s not clear whether this refers to a single 5-minute play session or a sum over both).

So, what did the Ravaja et al. observe? SCL appeared to increase after every kind of game event, though not uniformly so. This suggests basically that in-game events are stimulating, which is probably good news for the developers. Then the authors try to use the EGM data to parse exactly what kind of emotional response the player is having to various in-game events. The researchers find that zygomaticus and orbicularis oculi activity decreases in response to killing an opponent and increases in response to the death of the player. As one might expect, corrugator activity increases in response to opponent deaths and decreases in response to player death. The negative emotional valence was attenuated in players who had high Psychoticism scores on the personality test. Player self-reporting of emotional experience suggested that the primary emotion felt during play was fear, with lesser numbers reporting pleasant relaxation, anger, or depression (probably that guy who died 8 times). Players also reported strong feelings of joy. However, it is not clear whether these reported emotions come exclusively from the Nightfire test or include the Monkey Ball results. As gamers are well aware, different games can produce vastly different emotions.

What are we to make of all of this? Well, the first thing is to remember not to take this too seriously. The sampling here was very limited and included only a single day of play. Moreover, it did not involve anyone under the age of 20. So, we cannot gain any information here about long-term desensitization to violence, developmental effects, etc. It would be foolish to draw very general conclusions here. The games used do not sample all of the genre space and do not include extreme visuals. However, we can take away a few ideas that can be explored in future research.

First, it would seem that a player’s reaction to in-game violence depends on his or her own personality. Negative responses to opponent deaths were slightly attenuated in individuals who had higher than average psychoticism scores (this does not indicate that they were psychotic). This suggests a common-sense idea, namely that whatever negative emotional effects games produce can be reduced by positive context. Similarly, negative contexts will exacerbate negative effects. Emotionally healthy people who play games will probably have emotionally healthy responses to them. However, again, this is something that needs to be examined more thoroughly in children and adults.

Second, these results may point to part of the nature of the entertainment experience. Remember that one of the primary emotions reported by players was fear. First-person shooters and other twitch-style games rely on a sort of catharsis for their emotional impact. Anxiety is developed and sustained throughout a level as danger is repeatedly encountered, often building to a fever pitch with a very difficult boss fight of some kind. The end of the level is accompanied by a quick easing of that anxiety, a release of the held breath, that may produce a positive emotional response. The authors suggest that a form of that relief may also be occurring when the player “dies”, mitigating or overwhelming the negative emotions associated with failure. In this regard it is somewhat regrettable that the study limited the play time. It would have been interesting to compare the emotional response to successful completion of a level with the emotional response to “death”. The authors also suggest that failure in a ‘fake’ context might be associated with a positive challenge. One caveat with the “player dies” result is that this was the least-sampled condition.

And what are we to make of the negative emotional response to the deaths of enemies? Well, it suggests that one of the central assumptions of the “murder simulator” argument, that games associate positive emotions with violent acts, is not necessarily true. One might also conclude that even digital people seem like people, and that some gamers therefore feel they have transgressed even when they kill an AI character. However, player attitudes towards virtual opponents are likely to be strongly determined by context. It would be interesting to see if these results would be recapitulated if the researchers were to, say, set the difficulty to a very high level and force the player to play until the level was completed. It would also be interesting to compare player responses between “killing” a standard enemy and a difficult boss. Also, it’s important to remember that these techniques measure facial movements, not emotions themselves. Perhaps the players are just wincing as a reflex, without any higher emotional thought about it.

Ultimately, this study is too small and too limited to be ultimately probative as regards even the short-term emotional effects of playing video games (remember, the total play time for each individual game was only 15 minutes). The unexpected emotional responses to player and opponent deaths, however, suggest at least that there is a great diversity of emotional responses to video games. From a scientific standpoint it will be interesting to see how further research develops these findings into a more coherent view of response.

From a gaming standpoint, it will be interesting to see if developers try to calibrate games to play off these responses. In a certain sense this has already happened: Shadow of the Colossus is an example of a game that (occasionally) worked to inspire a feeling of guilt about killing your enemies. Will this research inspire designers to take that a step further, to actively try to make players feel unhappy about killing their enemies? Will players respond to such an approach?

1. Ravaja, N., Turpeinen, M., Saari, T., Puttonen, S., Keltikangas-Järvinen, L. (2008). The psychophysiology of James Bond: Phasic emotional responses to violent video game events.. Emotion, 8(1), 114-120. DOI: 10.1037/1528-3542.8.1.114

 Posted by at 4:00 AM