I’m afraid I don’t really have much interesting commentary to contribute to the discourse around Freakonomics. The first edition was published over 10 years ago, and I strongly suspect that the contents seemed much more “rogue” then, than they do now, in the thick of a Big Data revolution where these type of interdisciplinary data-mining projects to form sociological hypotheses have become altogether common and trendy.
Clearly, Levitt and Dubner have a good sense for interesting topics and an accessible approach to exploring them, as they’ve parlayed the success of the book(s) into a wildly popular podcast. As with anything that has pop-sci tendencies, there are critics accusing the authors of being too broad, or of committing the classic correlation=causation gaffe. While I agree that many of the theories and conclusions proposed in this book can absolutely be subject to more scrutiny before being accepted as fact, I didn’t get the sense that Levitt and Dubner were presenting the research in here as “case closed.” Their process — and inherent recommendation — throughout every investigation seemed to be: question the experts and prevailing wisdom, seek knowledge, consult the data, get creative.
They didn’t say, for instance, “People think crime dropped in the nineties because x, y, z. We show that it was absolutely NOT x, y, z; it was c.” Rather, it was more of, “X, y, z are commonly credited for the reason crime dropped in the nineties. We looked into it and they probably did help, but the effect size of these factors is too small to explain the entire phenomenon. What else might have been involved? Well, c is correlated with the crime decrease, and it makes sense because […]; therefore, seems like as strong of a reason as any based on the available data.” Like, I honestly don’t think Levitt and Dubner want people to merely parrot what they’ve read in this book and take it as gospel; it seems contrary to their whole mission in writing the book. So I don’t really hold with the implied criticism that they’ve just replaced one bad set of data and conclusions with another.
It’s clear that the questions that they’re tackling are often fun head-scratchers precisely because the data around them is unresolved: there are a lot of moving parts that are too tough to get into a good predictive model, or maybe there is just not a lot of reliable data out there in the first place. It’s a good thought exercise to generate the hypotheses that they do and demonstrate that there is support for them in the available data, but outside of the legwork required to hunt down the data, most of what’s in the book is not particularly rigorous, to an academic standard anyway. From my perspective, that’s totally fine. But it also meant I didn’t really take much away from reading this. It goes back to what I started off saying — this populist approach to big data and seemingly unrelated datasets may have been ahead of its time, but now this exact fashion of exploratory research is not completely unheard of. I see that as being a very good thing, FWIW, but the Freakonomics podcast (and FiveThirtyEight blog, etc.) are much more currently relevant than this pioneering book.