<< Why no trial for bin Laden? | Main | Stay Denied to Mississippi Quadruple Murderer >>


Unconditional Differences, Video Games, and Violence

| 0 Comments
Among the worst sources of disinformation in public policy debates is the study that merely finds a correlation between two variables but is claimed to show a causal relation.  The study shows A is correlated with B.  Therefore, it is claimed, A causes B and we have to crack down on A to reduce B.  There is a host of other possibilities, one of which is that A and B have no direct relation to each other but both have some relation to C.

In the current issue of the Journal of Law and Economic, Michael Ward of UT Arlington has an article titled Video Games and Adolescent Fighting.  Here is the abstract (emphasis added):

Psychologists have found positive correlations between playing violent video games and violent and antisocial attitudes. However, these studies typically do not control for other covariates, particularly sex, that are known to be associated with both video game play and aggression. This study exploits the Youth Risk Behavior Survey, which includes questions on video game play and fighting as well as basic demographic information. With both parametric and nonparametric estimators, as there is accounting for more demographic covariates, the video game effects become progressively weaker. The overall link between video games and fighting is modest and not statistically significant. The remaining positive association appears only for individuals who play 4 or more hours per day.

Boys, on average, play more video games than girls and also, on average, get into more fights.  Well, duh!  So if you just do a correlation of gaming and fights without controlling for sex, you get a positive correlation.  But it's not real, or at least not strong enough to say for sure it is real.

We see stuff like this all the time, especially with racial claims.  We need to get rid of the notion that simply showing a correlation can establish anything or even shift a burden of proof to the other party.  A correlation by itself is such weak evidence of causation that it should not even be admissible, much less permitted to raise a presumption, at least until the most plausible alternative reasons for the correlation have been ruled out.

Leave a comment

Monthly Archives