Rise In Violence Linked To Poor Statistics. Or Climate Change
http://www.thegwpf.org/rise-violence-linked-poor-statistics-climate-change/
The most charitable way to describe the result of this Science paper is … complete and utter nonsense. I do not want to exaggerate, but this paper is not even a mess. But at least the BBC believed them.
The computer revolution has been as bloody, wrenching, and disruptive as any other conflict. But unlike in territorial wars where there are actual body counts, the wounds caused by the proliferation of easy and cheap computing are chiefly psychic. (The most common injury is morbid over-certainty.)
Problem with computers is that they are irredeemably stupid. The computer doesn’t know anything: it can only do what it is told. And if it is told to take this set of numbers and that set of numbers and to mix them in a certain way, it will do it, creating pretty pictures of the result. And as computers get better, it can do these blind operations faster and produce prettier pictures.
That’s where the trouble starts. People start believing in the pictures because they’re so pretty and quantitative. The result is—and I’m sorry for this joke—not a pretty picture.
It is here Yours Truly must accept some of the blame. I and my brother (and sister!) statisticians gleefully create these algorithms the computers chew upon. We do this for fun, for the sheer sake of climbing the mountain because it is there. This wouldn’t be so bad except that we release these algorithms into the wild, where we naively hope they will do no harm.
Forgive us! Because instead they have wreaked havoc on the minds of men (and women!).
Perhaps no clearer example of the devastation can be had than by the Science paper “Quantifying the Influence of Climate on Human Conflict” by Solomon Hsiang, Marshall Burke, and Edward Miguel, all of whom appear to be economists, the Lord bless them. Here is their abstract (those who are squeamish are advised to look away):
…Drawing from archaeology, criminology, economics, geography, history, political science, and psychology, we assemble and analyze the 60 most rigorous quantitative studies…We find strong causal evidence linking climatic events to human conflict across a range of spatial and temporal scales and across all major regions of the world. The magnitude of climate’s influence is substantial: for each 1 standard deviation (1σ) change in climate toward warmer temperatures or more extreme rainfall, median estimates indicate that the frequency of interpersonal violence rises 4% and the frequency of intergroup conflict rises 14%. Because locations throughout the inhabited world are expected to warm 2 to 4σ by 2050, amplified rates of human conflict could represent a large and critical impact of anthropogenic climate change.
Well, you can see it for yourself. Poor men. It’s almost as if they believed what they were writing!
I particularly admire the sciencey shading.
They mixed data from sources as disparate as the MSNBC and Fox News, they compared apples to roller blading, they contrasted black with semiotics. Data from last Tuesday was said to be equal in veracity to that culled from 8000 BC. They dumped into a computer a bunch of numbers lots of people found from all over the place, measuring God knows what, and produced lots of sharp graphics and one big quantitative result that hot, rainy weather is bad for you.
Which is exactly why everybody is moving from Michigan to South Carolina (at least 2σ difference in mean temperature): to get to where the action is and join a gang.
The authors mixed violent personal crime in the USA in one model, with inter-group riots in India in a second model, with redistributive inter-group conflict in Brazil in a third, to global civil conflict onset in a fourth, to civil war incidence in Africa in number five, to rape in the USA in number six, to political leader exit (oh! the violence!) worldwide in seven, to domestic violence in Australia in eight, and then added these to models nine, ten, etc., etc., etc. Sixty—60—in total!
Sometimes temperature at the local level was the “independent” variable, or causative agent, and sometimes it was “jurisdictional” temperature. Still other instances: daily max county (not city) temps, stadium temps (yes: football areas), state rainfall losses, pixel (yes, pixel) temps, municipality rainfall by divisions, pixel standardized precipitation loss, and, naturally, pacific ocean temperature. Oh, and ENSO and the North Atlantic Oscillation, and floods and storm numbers, and (who could leave this out?) the Nile river overflowing.
Whatever weather or climate variable “worked” was the one they used to predict whatever kind of “violence” could be quantified. The whole thing was then duct-taped together and lathered with colorful Bondo to conceal the gaping holes in commonsense. But at least they produced wee p-values (def: less than the magic number, which is what makes it science).
The most charitable way to describe the result is complete and utter nonsense. I do not want to exaggerate, but this paper is not even a mess. My God! the work, the hours the authors must have toiled. The supplementary material alone is thirty-eight pages!
At least journalists, the dears, believed them.
Full comment
Sent by gReader Pro