Colby Cosh on the recently published findings of a p-hacking conspiracy study on how the election of President Donald Trump was reflected in the birth ratio of liberals in Ontario:
Front view of Toronto General Hospital in 2005. The new wing, as shown in the photograph, was completed in 2002.
Photo via Wikimedia Commons.
On Monday there came a surprising piece of science news from BMJ Open, an open-access title affiliated with the British Medical Journal. It seems two researchers from Mount Sinai Hospital in Toronto, an endocrinologist and a statistician, have convinced themselves that the election of Donald Trump to the American presidency in November 2016 had a nerve-shattering effect on Ontario. The province of Ontario, that is, not the Los Angeles suburb.
Trump’s victory, according to the researchers, was so awful that, like a war or a disaster, it briefly altered the sex ratio in live births in the province. This is, I should say, a fairly well-established effect of extreme social traumas. When mothers experience physiological stress, the uterine environment becomes less hospitable, and male fetuses, more vulnerable to such changes, become less likely to survive pregnancy. (This makes sense from a Darwinian standpoint, because girls are more valuable than boys in replacing population after a calamity.)
In 2020 nobody should need me to say that a cute, counterintuitive scientific “result” like this, appearing in the newspapers on literally the day of its publication, should be greeted with extreme skepticism. The sex ratio at birth, always expressed in medical literature as a ratio of boys to girls, tends to hover around 1.06 under natural circumstances. (Even in an advanced civilization, things even out within the age cohort over the next 20 years as the lads explore dirt bikes, rock fights, and roofs.)
The Mount Sinai researchers, Ravi Retnakaran and Chang Ye, had records of the sexes of all children born in Ontario from April 2010 to October 2017. Even in a place as large as Ontario, the ratio naturally bounces around randomly between 1.1 and 1.0, and there are seasonal effects that the duo corrected for.
There is no obvious signature of a Trump effect in a scatterplot of the adjusted data, which serves as a warning that the effect being claimed may be an artifact of analysis. But when you apply “segmented regression” using the same parameters as Retnakaran and Ye, you find that the (unadjusted) ratio dipped to 1.03 in March 2017, the fifth month after Trump’s win, and then climbed to 1.08 in June and July before reverting to the long-term norm.
