“We don’t just have egg on our face. We have an omelette.”
That’s what Tom Brokaw said in 2000 when NBC called Florida for Gore a tad too early. NBC made the call at 7:50 p.m. on election night, and it was based partially on exit polling data. Several other media outlets followed their lead and made the same call.
Nate Silver drew up a 2016 electoral map to show us all “What a Clinton Landslide Would Look Like.”
A few hours later, media mea culpas for calling Florida too early were all the rage.
Yet, here we are, nearly 16 years later, and the mainstream news narrative basically has them projecting that Hillary Clinton has locked up enough electoral votes to win the election. With 0 percent of precincts reporting nationwide and zero votes cast, Joe Scarborough was projecting Hillary Clinton will win all the blue states on the NBC election map this morning.
538’s Nate Silver drew up a 2016 electoral map to show us all “What a Clinton Landslide Would Look Like.”
Nate Cohn at The New York Times’ The Upshot says it’s “too strong” to call the race over, but Hillary Clinton’s probability of winning is about the same as “hitting a field goal from the 20 yard line.”
The Upshot also ran a map of what a Hillary Clinton landslide would look like. I’m not sure if it was the same map 538 ran because a pop-up said I couldn’t look at it unless I paid 99 cents for a 4-week subscription. It wasn’t worth it. Why burn a dollar on a hypothetical Clinton landslide map when I can grab a blue crayon and make my own?
Why has the media basically upgraded Hillary from Democratic nominee to presumptive POTUS?
“Big Data” has us awash in a deluge of polling numbers, and the mainstream media— grateful for the Trumpkins-stay-home-because-Hillary-already-won narrative— has lost sight of what polls are and what they represent. They’re not a scoreboard; they’re hypotheticals that rely on a set of assumptions and predictions being true. In a year where the Trump train has already rolled over so many deeply held electoral assumptions, and the electorate’s perception of the candidates is monumentally different than ever before, the media should have more skepticism and less glee when it comes to the polls.
Surely Nate Silver doesn’t want to have to pen a sequel to his May 18 article, “How I Acted Like a Pundit and Screwed Up on Donald Trump.”
Here’s why polling data isn’t worthy of the media flash-mobbing it has inspired:
1. Trump has a big bump coming, I guarantee it.
There are registered voter (RV) polls and there are likely voter (LV) polls. Registered voter polls make no attempt to predict what the electorate will look like. Basically, unless you have a morning show on MSNBC, they’re useless. Will the electorate look like a representative sample of registered voters weighted to reflect census data? No. Using RV polls to decide who will win an election is like trying to predict who will win a football game without knowing which players will show up. RV polls have about the same predictive powers as a Magic 8 ball.
To quote Nate Silver, “polls of all registered voters or all adults usually overstate the performance of Democratic candidates.”
Likely voter (LV) polls are much more valuable than RV polls because — if the group that conducted the LV poll does a good job predicting who will vote — then it predicts much better than an RV poll what will happen on Election Day.
If you take all the polls listed on RealClearPolitics that were conducted entirely or partially in August, and break them down into which ones are LV, and which ones are RV. Clinton leads Trump by about 9 points in the RV average, but in the average of the LV polls, Clinton leads Trump by only about 4 points.
Soon, all the polls will be LV and Trump will get a bump simply from the change in methodology.
2. What’s more important? That we produce quality data or that the talking heads have stuff to say?
There are two kinds of polls — probability polls and nonprobability polls. Probability polls are based on random sampling. Nobody can choose to be included in the population being sampled from because it would bias the results.
Nonprobability polls allow people to opt-in to the polling population, undoubtedly skewing the results.
So why would anyone use a nonprobability poll? Because the jabbering heads need data to talk about all day long, and nonprobability polls are a way cheaper way to get it.
This is the first election I’ve ever seen nonprobability polls treated like sound scientific polling data. The two I’ve seen popping up over and over are Survey Monkey and Yougov.
Here’s three biases that exist in a Yougov survey.
- People have to be members of Yougov to be sampled. Today, I said to myself, these Yougov samples are too biased against Trump. I’m going to go sign up for Yougov and even it out. I’ve now “opted in,” which taints the population they’re sampling from — and so can you.
- You have to surf the interwebs to participate, so this is almost certainly undersampling older less tech-savvy people.
- It’s definitely oversampling people who give a crap about polls and undersampling the group of people who don’t want to sit around on their computer or phones and take surveys — I’m guessing that’s a sizable group.
Yesterday, my friend texted me results from the Zip App saying that 89 percent of respondents would rather be “punched in the face” than vote from Hillary Clinton. Those numbers carry as much weight with me as a Survey Monkey or Yougov poll — zero.
3. Not even the pipe-smokingest, tweed-wearingest Ivy league professor — whose election model has predicted every winner since the beginning of the Cenozoic era — knows who will actually show up and vote.
It’s always hard to determine who will show up to the polls. Gallup gave up and isn’t even trying this time around. It’s always hard to predict who will show up, but in this unprecedented political climate, it’s going to be impossible.
Will the Democrat voters who turned out so enthusiastically for Obama show up for Hillary too? Will the electorate look like it did in 2012 and 2010? Will it look like it did in 2004? Or will it be an electorate we’ve never seen before? Once you predict who they’ll be, then you have to predict what they will they care about. And — cherry on top — you get to predict it all without knowing what will happen between your poll and Election Day.
I’m not saying Trump is winning — I sincerely doubt that’s the case. I’m saying the media should never treat the polls like a scoreboard and certainly not in August.