Avram,
what I understand in this paper that it is not just a notion of correlation by year – but correlation by county, correlating to the number of cell towers there. They also looked at a couple of competing factors and found them lacking. This seems like a reasonable correlation, not yet causation, but enough to take it seriously. I agree that the drug relationship does not seem to have data behind it.
Maybe, it is so clear to them that murders relate to drugs – something that is not obvious for, B’H, outsiders like you and me.
I understand that they are focusing on murders and not drug-related arrests, because the latter number is less reliable and more prone to be correlated to some hidden factors (police presence, politics). This makes sense to me.
I personally would suggest a more general explanation – teens are sitting in front of the screen (possibly doing computer crimes or aveiros) rather than with other human beings. We know recommendation to make a potential murderer a shochet. Not sure, whether a conversion of a potential drug dealer into a cyber-criminal is a good deal, maybe it is.