Don't buy a computer, but if you do

Over at India Uncut, Amit Varma points to study by Todd Kendall which claims to demonstrate that the spread of the internet causes a decline in the incidence of rape. Kendall's argument is that the Internet makes pornography more widely available, especially to men in the 15-19 age group, and serves as a substitute for sexual violence.

It's an interesting paper, but reading it I had several concerns.

My biggest problem with Kendall's empirical results is that he runs a regression with both % of households accessing the Internet and % of households with computers as independent variables, and while the sign for Internet access is significant and negative, the sign for households with computers is significant and positive (it's the most significant variable in his regression) - a fact that Kendall conveniently neglects to mention in his paper, let alone provide an explanation for. Kendall's justification for including % of households is to seperate the effect of the Internet from other technological influences, but that would imply a non-significant effect of owning computers but a significant effect of Internet access. As it is, we have two variables in the right hand side of the equation that we would expect to be highly correlated (Kendall does not bother to provide us with a correlation table, but computer ownership and internet access pretty much have to be positively correlated) and they enter the regression with signs that are opposite and significant. Personally, I'd love to see what happens to the coefficient on Internet access if Kendall runs his regression without % of households with computers in there. I'm unconvinced that it would continue to be significant.

Think about it this way. Kendall tells us that, on an average, a 10% increase in Internet penetration causes a 7.3% decline in incidence of rape [1]. But if you believe his results in Table 4, a 10% increase in the % of households owning a computer causes a 6.4% increase in the incidence of rape. So the net effect of buying a computer and using it to access the Internet on the incidence of rape is a mere 0.9% (yes, yes, I know it doesn't work that way - which is exactly my point). But who are all these households who are buying a computer but not using it to connect to the Net? And what, according to Kendall, is the reason that they're more likely to commit rape? Frustration about poor connectivity? Isn't it more likely that what we're seeing is just multicollinearity unrealistically inflating the regression estimates?

My second concern with Kendall's study is that he assumes implicitly that the spread of the Internet has no effect on the reporting of rape, so that changes in the number of rapes reported is a valid measure for number of rapes actually taking place. Kendall acknowledges that there is a measurement problem here, but sees no reason to believe that the spread of the Internet may be causing a systematic bias in his measurement. He even makes some arguments for why his measures may be underestimating the effect.

Yet the study itself suggests one potential reason why the results might be biased. Kendall tells us that rapes that don't get reported tend to be those committed by people known to the victim - date rapes, for instance. Kendall also cites previous literature that tells us that the Internet facilitates more dating and other face-to-face interactions and that this may increase the opportunities for rape. Put those two together and it suggests that the spread of the Internet increases the opportunities for the kind of rapes that tend to go unreported. Is it possible, therefore, that the effect Kendall is capturing is really a reflection of the fact that the Internet is shifting the incidence of rape from assaults on strangers (which have high reporting rates) to date rape (where reporting rates are low), causing reporting rates to go down? I'm not saying this is necessarily happening - I'm simply saying that it's an interpretation of Kendall's results (such as they are) that would be consistent with the literature that Kendall himself cites, and that he doesn't consider.

Finally, it's interesting that though Kendall has a panel data set, he doesn't actually account for lagged effects in his model. So what we're seeing is the absolute level of rape in the state (and the absolute level of internet usage) not the change in rape incidence. Personally, I would love to see a regression where Kendall includes the previous year's rape incidence for the state on the right hand side or, even better, takes first differences for his variables of interest. That would tell us whether changes in the spread of the Internet were really driving changes in the incidence of rape.

None of this is to say that what Kendall is saying is necessarily wrong, though personally I'm sceptical about the argument that access to porn is a substitute for rape (in Kendall's terms, I'm firmly in the camp of those who believe that rape is about power rather than about lust). It's simply to suggest that Kendall's results and the interpretation he puts on them are extremely questionable, and we should be careful before drawing any real conclusions from them.

[1] I'm assuming that Kendall's reported coefficients are adjusted for the fact that his dependent variable is in logs.