# I need a nonsense detector on my model

So I'm trying to check internet availability's effect on terrorism. I was just going to run a regression of

Number_of_Attacks ~ %access_of_internet + region(categorical)

To code %access_of_internet, I would code each attack with the proportional access of internet in that country in that year.

From this, I would interpret the coefficient to say that for every change in access to internet, the number of attacks will change by Y. Theoretically, I would make some argument about the strategic value of the internet. If the coefficient is positive, I could say something about how the internet broadens networks of terrorists and facilitates planning. If it is negative, I will say something about how the internet, contrary to popular belief, just makes it easier to catch terrorists due to a data trail.

Is this sound? If not, how would you suggest I fix it to answer the question: "does the internet facilitate terrorism?"

• Sounds like nonsense to me. Internet access is one of many possible source of radicalization, and access to the internet won't cause radicalization in the absence of other additional causes (you have internet access but are unlikely a terrorist). But it's certainly not the only cause, nor is it even necessary (pre-Internet terrorism being the obvious example). So your model is incorrectly specified, which means you lose most of the desirable qualities of the coefficient estimates. – Sycorax Apr 21 '16 at 13:19
• Lots of incorrectly-specified models have significant and robust coefficients. Setting aside any numerical questions, just think through what you're saying: people who can access the internet carry out terrorist attacks. That's insane. Can you think of at least one other reason people carry out terrorist attacks, other than having access to the internet? Cf. omitted variable bias. – Sycorax Apr 21 '16 at 13:25
• What about internet access meaning access to more information, which could be an eye-opener for potential terrorists. What if this information make them less determined that bombing people is a good idea? What if internet access dilutes the influence of other sources of information, where those other sources are much more likely to support terrorism (e.g. meetings of local terrorists as an information source)? – Richard Hardy Apr 21 '16 at 13:39
• I don't understand what you're asking. If the question is how do I ensure that my model makes sense, that is science and substance, not statistics. Otherwise put, you are, or should be, one nonsense detector and the same applies to your readers. @C11H17N2O2SNa is making more or less the same point by challenging your model; I agree with his main argument (I'd prefer not to use the word "insane" in discussion). – Nick Cox Apr 21 '16 at 13:41
• The relationship between general access to the internet and an individual's access is tenuous. People can rarely or never use the internet even in places with high potential access, and vice versa. To do a clean analysis, you'd need at least to measure each individual's internet consumption. This may be helpful. – Stephan Kolassa Apr 21 '16 at 13:43