An interesting analysis. I believe, however, on a cursory examination
that there are some potentially serious problems with the main result of
the paper, that boys exposed to violent TV are more likely to experience
anti-social behavior problems.
My suspicion is the effect for boys is driven by one or at most 3
observations. The authors state eliminating the outlier (a boy) who
watched 5 hours of 'violent' TV a day only attenuated the result for boys.
What, exactly, would happen if that observation had been switched from a 1
to a 0 on the dependent variable? What would be the effect if the 3 boys
with the highest TV watching were simply eliminated? It is there, after 2
hours a day that the lowess of probability rises substantively. It would
be interesting to know what the BPI scores were for those three boys
scoring highest on violent TV watching.
That is technical, but crucial. Still on a technical level but less
obvious is why the dependent variable was coded as it was. The BPI is a
continuous measure, and I am curious why this dichotomization was used.
Collapsing a continuous measure into a dichotomous one loses information.
It could be, for example, that boys (or girls) spending less time watching
violent TV could score higher on this sub-scale for the BPI than others
spending more time doing so, but when it's dichotomized they all look the
same.
Given that the authors did it this way though, I am curious as to why
the 88th percentile was chosen instead of the 90th, when the latter was
the stated criterion. Of course, they could be the same thing, but if so,
why state the 88th was chosen? If they are not the same, by choosing the
lower criteria, a few extra kids are classified as having a 1 on the
dependent variable than would be if the higher one was chosen.
Conflict of Interest:
None declared
eLetters: ![[Read eLetters]](/icons/misc/sectionDOWN.gif){kind=icon}
