You Can’t Trust the Polls
Forecasting elections is more art than science. One of the problems in forecasting elections is the nature of polls. Many polls are misleading and suffer from flawed methodology.
I was flying around the world just before the 2016 presidential election, telling audiences in Europe, Australia and here at home that Trump would win. I got a lot of funny looks and gasps from the anchors, who didn’t take me seriously at all.
The major polls were all giving Hillary something like a 95% chance of victory on the eve of the election.
Now, that doesn’t mean that the polls were 95% to 5% in favor of Hillary. They weren’t. They were more like 52% to 48% or 53% to 47%. But the polls were in so much agreement the experts were saying the chances of them being wrong were less than 5%.
Hillary actually did win the popular vote, as we know. But unfortunately for her, most of those people were concentrated in states like California, New York and Illinois, which are only three states out of 50. You need to win state by state to take the Electoral College.
I don’t say it to brag, but I got the last laugh. I wasn’t trying to be controversial for its own sake or to try and make headlines. For me, predicting a Trump victory was simply the result of a lot of hard analysis.
What was I thinking? What was I basing that prediction on?
I basically just looked at the polls and broke them down piece by piece. Whenever you see a national polling result, you have to ask yourself, who are they polling? Everybody? Just registered voters?
Let’s start there, because if you’re not registered you can’t vote. You want to look for polls with registered voters. But even then it’s not good enough because you further have to break it down to likely voters. That’s because approximately 50% of the people who are registered don’t vote.
So if you simply poll registered voters, you’re going to get a lot people who aren’t going actually to vote. That’s not good polling. But it doesn’t end there.
If the poll is improperly weighted towards either Republicans or Democrats, you are likely to get a poor result. For example, leading up to the 2016 presidential election, the polls were clearly oversampling Democrats.
There are slightly more registered Democrats than Republicans, so it is valid if the poll leans slightly Democratic. But by how much?
A fair sample might be 52% Democrat, 48% Republican, give or take. But that’s not what the polls were doing. They were leaning maybe 57% Democrat versus 43% Republican. In other words, the polls were oversampling the Democrats.
Then, within the Democrat sample they were oversampling African-Americans. African-Americans will tend to vote 85 to 90% for Democrats, which is fine. Everyone is free to vote anyway they choose, and Democrats are widely considered the party more attuned to the concerns of minority groups. I’m not saying it’s right or wrong, and I’m not getting into partisan politics here.
But as an analyst I have to recognize the common perception and factor it into my analysis.
From a polling perspective, the problem is if you put more African-Americans in your sample, you’re not only oversampling Democrats. Within the sample you’re missing white Democrats who might vote Republican. And there were a lot of those folks in 2016.
So all those polls before election day 2016 were showing a decisive Hillary win. But if you really break down the polls you can spot these statistical errors. All you have to do is back them out. This is what I did in 2016, which led me to conclude Trump would win.
I would see a poll that had Hillary ahead by two or three points and say, “Okay, but you’re oversampling Democrats, or you’re oversampling African-Americans, or what have you. Let’s adjust for all those things.”
When I made the adjustments, Trump was ahead. So when I was saying Trump was going to win, I didn’t think I was way out on a limb. I thought I was simply using good science. You could see a Trump victory coming if you knew how to break down the polls.
Additional factors you need to consider, for example, are voting differences between men and women, young and old, college-educated versus non-college educated, et cetera. If you don’t take a careful look at all these factors, you’re not going to get a true picture.
In reality, it’s very, very difficult to get right.
If you go back to the presidential election in 2012, for example, five million evangelicals didn’t vote. They were not going to vote for Obama, but they didn’t vote for Romney either. They stayed home because Romney was a Mormon and many Evangelicals have a problem with Mormons.
Now, this is the kind of thing you won’t hear on national TV. It’s a sensitive subject because it involves religion and people steer away from it. But if you’re an analyst trying to get things right, you need to account for factors like this, no matter how uncomfortable it is or how you feel about it personally.
Getting back to the 2016 election, there was other evidence too I saw going in Trump’s favor, some of it being purely anecdotal. For example, I look at lawn signs. I’m probably one of the last people in America who still takes a bus occasionally. The thing about being on a bus for three hours or four hours is that you see a lot of lawn signs. I didn’t see any for Hillary on some of my trips.
The point being, you cannot take polls at face value. You have to tear them apart, piece by piece. Look for oversampling. Look for things like registered voters versus likely voters. Look at all these factors and make the right adjustments to get to the right result.
And right now I’m analyzing the current election the same way I analyzed the 2016 election.
There are 435 separate House elections. But when you boil that number down, about the great majority are foregone conclusions. The districts are either solidly Democrat or Republican, so there’s no contest.
For example, certain districts in New York City or Chicago or certain California cities are guaranteed to be Democratic. Certain districts in other parts of the country are guaranteed to go Republican. No one worries about those districts or polls of those districts because you know who’s going to win. In some of these races, the other side doesn’t even run an opponent.
There are only about 70 to 80 districts of the 435 that are going to decide the outcome of the House elections. Some are too close to call.
The Republicans can afford to lose seats. They just can’t afford to lose 23. The Democrats need to pick up those 23 seats in order to retake the House tomorrow. That’s a bit more daunting than it might seem at first. The problem is that in 2016, the majority of these seats voted for Trump.
There are eight districts in the East that are going to tell the tale. These are bellwether districts, meaning they could be Republican or Democrat. If the Republicans keep them, that’s good news for the Republicans. Likewise, if the Democrats start to scoop them up, that’s good news for the Democrats.
Two of these districts are in New Jersey, the 3rd and the 7th. Two are in New York, the 19th and the 22nd. In Pennsylvania it’s the 17th district. In Virginia it’s the 10th district. In North Carolina it’s the 9th district. And in Florida, it’s the 26th district.
Again, that’s a total of eight districts. Watch these eight districts. If the Democrats take six or more, you can go to bed early because Democrats are going to take the House. If the Republicans can hold six or more, you can also go to bed early because the Republicans are going to hold the House.
If they’re even, put another pot of coffee on because you’re going to be up late. But those are the bellwether districts. Republicans need to hold them. Democrats need to take them.
It’s going to be close. Right now, here are what my models are showing for the overall elections:
They have Republicans expanding their lead in the Senate, 54-46. They also have Republicans holding onto the House, 221-214.
As I said earlier, forecasting elections is more art than science. We’ll have to see. But I fully expect to be up late tomorrow night.
for TheDaily Reckoning