In election season, we regularly see references in the USA to "The popular vote" as well as nationwide polls comparing presidential candidates. These are self-destructive, and ideally should be curtailed.

There is no popular vote in US election law. People talk about one because newspapers add up the 51 different races that are done and publish a number. People make several significant errors about this number. They talk about a candidate "winning" or "losing" this popular vote, which of course is not correct -- you can't win or lose a race you're not even trying to compete in, and no candidate does so.

Instead what people are probably interested in is a different number, "What the popular vote might be, if the USA had an actual popular vote race which determined a winner." Many people would prefer a popular vote to the electoral college, for fairly obvious reasons. Problem is that the newspaper published number is not that number and may not even be close to it. It might be possible to try to calculate the theoretical number people would like to use to compare, but at present nobody does that.

Many countries don't have a popular vote but sometimes report it. In Canada, in the last election the Liberal party got 32.6% of the unofficial "popular vote," and while there are people who lament this, there's actually much less consternation over this than what it seen in the USA when the electoral college has an opposite sense than either the fake popular vote or even a real one. The Liberals rule with a minority but came close to the full power of a majority. In the UK, Labour got 33.7% of the "popular vote" and did win complete control via a majority.

Why the newspaper "popular vote" isn't real

The published popular vote has no scientific validity, unfortunately. That's because US elections consist of about 5-8 "real" elections in swing states, and 40+ partially pretend elections in safe states. The candidates campaign and spend money almost entirely in the swing states. In safe states, they don't see the thousands of rallies and billions of ad pieces and other aspects of campaigning. Only those who watch the news get information, though today more also get it online. The voters in the swing states are told 50 times every day how important their vote is, and they go to the polls expecting to be able to make a difference. In the safe states they are told, and they know, their won't change the Presidential result one whit. They go just on principle, or because they are civic minded, and also for downballot races where they can make a difference. But we know, based on the difference in turnout in off-years, that a lot of them don't care that much about those downballot races. Turnout is also lower in safe states, for reasons you might expect.

In our quest to figure out what the popular vote would be if it existed, we have a problem. The safe state voters are what we would call "self-selected." Whether they voted or not is based on how keen they are in appearing in the result, not how keen they are on the race, which they can't change. While it is true that in low-turnout elections like the USA, even the swing state voters have an element of self-selection, but it's a very different one, and the rules of the election define the count as being the sum of the preferences of the people who take the effort to go to the polls. As such, they measure that precisely.

Self-selection introduces a bias into a result. It no longer matches what we are hoping to measure -- for example what a real popular vote would be like. In statistics class the teaching is clear. You can't add results with different biases in them. It's Apples and Oranges. If you try to add them, you might be getting a very wrong number, and you don't know how wrong it is.

When newspapers add the "popular vote" they could make one total for the swing states and one for the safe. The swing total would be pretty valid (there are still some biases, due to election manipulation and interference) and the safe state result would still be invalid, but at least the numbers would be more correctly grouped. Nobody does this.

Could we fix this?

We could get a better approximation. We still face the problem that we can't say "Hilary Clinton won the popular vote" because she wasn't trying to win a vote that doesn't exist. Until people believe their safe state votes actually meant something, it's hard to say what they would have done.

There could be effort into trying to understand the difference in voting patterns between safe and swing states. In particular, you would love to compare the same state. Some states switch between being safe and swing, but it's at least 4 years apart. You might be able to look at results when there is an important real race, like for Governor, or Senator or a ballot proposition, that pulls out as many voters as the Presidential race. Unfortunately there are still going to be many factors putting bias into your totals. As far as I know, nobody has attempted to try to control for this bias, if they even can, and if they have, they do not publish it.

In addition to the actual election result, it's surprising how many Presidential polls are done nationally. While polls are also done in swing states, a surprising effort goes into the national polls, presumably because they are better newspaper fodder, being interesting to everybody, not just those interested in a particular state. (Most companies doing national polls do make the statewide breakdowns available to paying customers but they don't show in popular coverage.) Sites like five-thirty-eight try to use those national polls (as it's all they have sometimes) to measure what changes might be going on in individual states, but I am not sure they should. Data from the actual state is always better. Polls have so much bias in them already -- did you know that in order to get once answer a typical pollster has to make 15 phone calls? -- and I have joked that a good poll result would be "100% of voters have landlines, answer calls from strangers, and are home when pollsters call." Pollsters try to compensate for this bias but it's insurmountably large, which is why there is so much error.

So when you look at this page of polls try to ignore the national polls. You can request it show you only a particular state -- focus on Pennsylvania.

Having a real popular vote

Many people hate the electoral college, and want to replace it with a popular vote. That's not an unreasonable thing, though it turns out the quirks of the college were deliberate at the time they were made. But if it is desired, it's a tall order. Since 2000, the college has helped the Republicans over the Democrats, and that means the Republicans aren't going to support its elimination, even though in the past it's gone the other way.

There's an effort that has been underway called the National Popular Vote Interstate Compact which seeks to solve it by getting states with 270 votes agree to cast their electoral votes with the popular vote. If the compact existed, there suddenly would be a real popular vote because this would make the races in all states equally real. Problem is, 17 states and DC have joined this compact, but they are all "blue" states. A few swing states have it in Committee, and Nevada needs referendum approval. But as an all blue-state effort, it's very hard for a red state to join.

A much simpler plan might be easier to make work. You don't need 270 votes. Instead, if you had 50 safe blue votes and 50 safe red votes, as you would if California, Texas and one other large red state like Georgia -- that's just three states! -- it would work. That's because while they only swing things 100 votes, any election that is so close that the electoral college and popular vote are opposite won't be a college landslide, and 100 votes difference in the college will flip it as well. It's less clear what happens if there are 3 real candidates.

But in reality, the college only makes very minor differences. Differences of 0.5 to 2%. In most of the world the differences are much larger. The reality is that parties adapt to flaws like this. They should not have to, but they find ways. They adjust their platforms to compensate for anything not flat about the playing field, and the contest continues. There is not a natural 50-50 division in American politics or any other system. Rather, that's an adaptation to the rules.