<![CDATA[Stories by Evan Hughes on Medium]]> https://medium.com/@evanhughes?source=rss-e715b2179d83------2 https://cdn-images-1.medium.com/fit/c/150/150/0*_DAfwJklqH-NvQRV.jpeg Stories by Evan Hughes on Medium https://medium.com/@evanhughes?source=rss-e715b2179d83------2 Medium Sun, 31 May 2026 20:42:13 GMT <![CDATA[On polling and percentages]]> https://medium.com/@evanhughes/on-polling-and-percentages-edfe23eaad9a?source=rss-e715b2179d83------2 https://medium.com/p/edfe23eaad9a Mon, 07 Nov 2016 22:32:30 GMT 2016-11-07T22:35:32.395Z With zero expertise in statistics or probability, I’ve been trying to articulate why it feels off to describe a candidate’s chances of winning in percentage terms. Some tentative thoughts: It seems like we mean different things when we say “the Cubs have a 33% chance of winning” vs. “Trump has a 33% chance of winning.” The Cubs are trying to get hits but accidentally make some outs. It all turns on human performance, and we accept that there’s an element of randomness to that. But voting is not about human performance; it’s just human decision-making. You don’t swing and miss while trying to vote Clinton. (Butterfly ballot joke goes here.) In the final days, it feels like the main thing that’s being quantified by 538/Upshot/etc. is the chances that the poll averages are wrong, which is about human performance. But we don’t know much about measuring that because the era of poll aggregation is short, and elections happen infrequently. Hence the variance between 538 and the other models, I guess—it’s a young field, we’re still figuring a lot out. When Nate Silver pegged Obama at 91% to beat Romney in 2012, did he get it right? We don’t know. You’d have to hold the election 100 times.

And yet the models keep putting precise percentages out there, and my scientific conclusion is: I dunno, it feels off.

P.S. Here is a funny tweet on this theme:

Steve Hely on Twitter

Y'all are gonna owe Nate Silver an apology when Trump wins in 35 of 100 multiverses

]]>