This week, you probably saw some headlines that said things like, "Poll: Clinton clobbers potential GOP foes." Which sounds pretty definitive. But! You may have also noticed that the Republican National Committee did not publish a press release that read, "Piss it, we're conceding the race and regrouping for 2020." Why is that?
Well, there's a quirk in the science of polling, which holds that leading up to any presidential election, there will be months in which the head-to-head polling of the race is very accurate and other months in which it's not very inaccurate. I've prepared a little guide here, ranking the 10 best months for polling accuracy for the next presidential election, in order from least to most accurate:
10. March 2016
9. April 2016
8. June 2016
7. May 2016
6. July 2016
5. August 2016
4. September 2016
3. October 2016
2. November 2016
1. December 2016
As you can see, if you're a reporter and you want to obtain the most accurate possible snapshot of who is going to win a presidential election from a pollster, the best time to call him up is between twenty and fifty days after the election. You will ask, "Who is going to be the next president?" and he will say, "The guy who won the election last month." You can't go wrong.
You may have also noticed a trend, in which the nearer you are to Election Day, the easier it gets to predict an outcome. And, indeed, past experience bears this out. A week before the 2012 election, most pollsters were uncannily predicting that the winner was going to either be Barack Obama or Mitt Romney.
But the further back you go, the murkier it gets. And look, here's some math from political scientists Christopher Wlezien and Robert Erikson, helpfully provided, with zen-like patience, by Brendan Nyhan of the Columbia Journalism Review. Think of this as my ranked list in chart form (you'll note I've accounted for the odd quirk that seems to hold that May polls are slightly more accurate than June polls):
The bottom line, as Nyhan notes, is that "polls conducted even 300 days before an election have virtually no predictive power."
From there, we can extrapolate. How accurate are 2016 head-to-head polls in November 2015? They are zero accurate. What about July 2015? They equal "not accurate." April of 2015? They are wholly antipodal to accuracy. And thus, in January 2015, these polls will be the null set of accuracy.
I bring this up because "political science Twitter" -- one of the few Twitter subcultures that do not essentially promote deleting your Twitter account as the path to a better life -- is hard at work calming people down about the polls that generated all these hot, hot, headlines. Listen to the nice political scientists, you guys! I promise that unlike virtually everyone else who writes about politics (including on occasion myself), they mean you no harm.
At this point, you may be wondering, "Well, if polling is so inaccurate until you get very close to an election, why do they continue to do it?" The short answer is that pollsters ask many questions that are more interesting than "Who would you vote for in this head-to-head matchup?" The answers just don't make for banner headlines.
We will know more in 20 months than we do now. Feel free to relax.
[Would you like to follow me on Twitter? Because why not?]