Lies and damn lies
In his opinion piece for the News (“D’AMBROSIO: Race, class and statistics,” April 21, 2017), Justin D’Ambrosio writes: “Statistics presuppose a domain of which they are supposed to be representative, and the statistic in question may not be representative of every subdomain.”
The meaning might escape you, as it did me. My best guess is that D’Ambrosio means to say that a general statistic — say, the radioactivity of Chernobyl — does not betray the assumption that each acre (or “subdomain”) scores the same on the Geiger radioactivity counter. And while that’s true, I’d still ask for a gas mask on my visit.
In vague terms, D’Ambrosio chastises Yale students in a Yale bubble for not being comfortable enough to explore New Haven and for allowing a misplaced trust in statistics to obscure our experience of the city. His diatribe carries an unmistakable tone of a guilt trip, if only because “race” and “class” are mentioned in the title of his piece but play no role in its content.
It’s true that Yale students likely don’t spend their time in a thorough analysis of “block-by-block” New Haven crime. Nonetheless, this information is readily available through sites like neighborhoodscout.com. But even if we didn’t know of the shootings and robberies just beyond our “walls and gates,” I would still advocate for every Yale student to exercise caution in the city. After all, in the absence of more detailed information, isn’t it better to be safe than sorry?
Still, in calling statistical inference “nonmonotonic,” D’Ambrosio implies our inability for anything but disparate information processing and binary uptake functions. D’Ambrosio would benefit from a study of Bayesian statistics and specifically posterior probabilities — those conditional on past events. Just because D’Ambrosio perceives our existent information about New Haven as not sufficiently granular, it does not follow that “learning new information can invalidate what we thought was a good inference.”
A basic exercise: Neighborhood X has an 80 percent chance of violence; we avoid it — rightly so. Assume we also avoid a specific block Y, within neighborhood X. A report comes out: Y only has a 20 percent risk level within X as a whole. We are best served by combining the information, and most of us do just that. A logical Yalie would update her prior knowledge to say that although Y doesn’t mean guaranteed death, she probably still wouldn’t want to be there.
New Haven is not that dangerous. But D’Ambrosio’s vague progressiveness should not be enough for us to self-flagellate and blame our “bad inference.” Perhaps, instead of blinding ourselves to actuality, feigning interest and blaming each other for being sheltered, we should look critically at New Haven’s injustices. Or we could run some regressions.
Daniel Kipnis ’19