There's a 50/50 Chance that Nate Silver is Full of It.
An unbiased and informative view of polling, the Electoral College and the coming presidential election. Part 1 of 2.
The election of Donald Trump in 2016 in a lot of ways was a disaster. Trump pulled us out of an important treaty with Iran, appointed Supreme Court judges who have damaged women’s reproductive health (among other things), and fanned the flames of social unrest. But I believe the Democratic Party’s response to Trump’s election was also a disaster, and much of that came from the expectations for the election pushed by the likes of Nate Silver, vox.com, and Hillary Clinton’s own campaign. So seeing Silver still plying his trade infuriates me.
Countdown to Futures Past
There are people who are planning to vote in the upcoming presidential election but their cars will break down, or they’ll get sick, or they’ll learn their grandmother needs a ride to Cleveland to have her toe surgically reconstructed. Or they’ll show up to vote and discover there’s something wrong with their registration. There are people who are not planning to vote but change their minds last minute. There are even people who plan to vote but genuinely haven’t decided yet who they’ll vote for.
But not many. Not this close to election day.
The 2024 election is not like a football game or hurricane. It’s not a distant event that’s hard to predict because it will be impacted my variables that haven’t happened yet. For elections the variables have almost entirely already happened. We are not waiting for America to decide whether it likes Trump or Harris better. Mostly we’re just waiting to find out what Americans have already decided.
And right now even that overstates the case. Polling is problematic (we’ll get to that below) but it’s good enough that we can say these days that slightly more Americans are Democrats or Democrat-leaning independents than Republicans or Republican-leaning independents. So that gives Kamala Harris an advantage. Most pollsters predict that Kamala Harris will probably win the popular vote. Not a daring prediction given that Democrats have won every popular vote contest against Trump. Trump has never been popular outside his base.
But American elections are determined by electoral votes. The popular vote doesn’t matter.1 I’ll take up the electoral college in Part 2 of this series. Subscribe to avoid missing that important addition to your life.
Meanwhile, as I said, America has probably already decided that it likes Harris and the Democrats a bit better, but we’re waiting to find out if Harris’ campaign has managed to convert that popular vote advantage into an electoral college victory. Or if the Trump campaign has managed to win enough swing states to take the presidency despite a popular vote disadvantage.
What we don’t need to wait to find out is if Nate Silver, the prediction outfit he launched (fivethirtyeight.com), and similar outfits are full of it. Because regardless of the outcome of the election, anyone who says candidate X has X% chance of winning is always a fraud.
And that makes me angry and that’s why I’m writing this.
Back in the Day
Pity the pollster today. How they must envy their predecessors. From the 1950s through the 1970s families had landline telephones and someone would dutifully answer those phones at dinnertime. Who are we voting for? What car do we drive? How would we characterize our religious practice? Americans would honestly tell any authoritative-sounding caller more or less the truth. But then came answering machines, and then cellphones, and then—for many younger Americans—the landlines disappeared altogether. Plus the lack of respect for authoritative-anything. Many Americans today won’t even answer a call if their smart phone doesn’t recognize the number. This makes reaching a cross section of the public challenging.
Sampling
Even in the good old days days reaching everyone in the state wasn’t possible. Then as now pollsters must convert their sample (those who pick up the ringing phone and actually answer the questions) into a picture of the larger population. Let’s say we’re pollsters conducting a policy survey about statewide preferences for a new state motto Vix superstites, and a hundred state residents answer our survey. We tabulate how many of our sample were male, female, black, white, old, young, rich, poor, single, married, and so on, and compare this to the state’s census data for male, female, black, white, old, young, rich, poor, single, married, and so on. We basically multiple each answer by differing amounts, giving weight to our sample respondents to reflect the demographics of the state as a whole. This works pretty well.
Registered Voters
But when it’s a presidential election we can’t use straight-up census data since not everyone in the state is registered to vote. (In every state you have to be registered to vote in a presidential election.) Registered voters are not an exact cross-section of the state’s demographics and the state doesn’t break down registered voters into helpful research categories as the census does. So pollsters have to do some more mathematical modeling based on how they believe registered voters compare to the state’s demographics. This information is not as accurate as the census, but pollsters are mostly industrious and clever people who tease out the picture in very smart ways. So pollsters probably have a reasonable picture of the demographic breakdown of who is registered in every state.
Likely Voters
But even that isn’t enough! Not all registered voters actually vote! And which registered voters actually vote isn’t consistent from election to election. Turnout can vary depending on whether a voter in a particular election is male, female, black, white, old, young, rich, poor, single, married, or whatever, and these changing results aren’t well understood. So pollsters try to model this too, but it’s a lot more tricky with more guesswork and possibly biased assumptions.
Swinging with Who Answers the Phone
Finally, as said above, reaching people without landlines that are willing to participate in a survey is hard, and pollsters don’t know in each demographic category whether the person talking on the phone is really representative of that demographic category. Are white, male, 30-year-old rural Republicans from Wyoming who answer the phone similar to white, male, 30-year-old rural Republicans from Wyoming who don’t? No one knows! Pollsters try to use exit polling (interviewing voters leaving the polls on election day) to spot check everything else. Does it work?
For many states it seems to. In many states general opinion polls fit with registered voter polls which fit with likely voter polls which fit with past histories and exit polls. So pollsters can be reasonably sure of what’s going on. So they are sure for instance that Wyoming and Texas are going to vote Republican, and Connecticut and California are going to vote Democrat. In fact more than 40 states will reliably vote Democratic or Republican in the presidential election.
The states that the pollsters aren’t sure about are the swing states. In the 2024 election Arizona, Nevada, Wisconsin, Michigan, Pennsylvania, North Carolina, and Georgia are in this category. They are not “swing states” because there are lots of voters in these states who don’t know how they’re going to vote or who might change their minds. They are only “swing states” because the pollsters don’t know which way they will go. Arizona, Nevada, Wisconsin, Michigan, Pennsylvania, North Carolina, and Georgia have already determined the outcome of the election, but our pollsters just don’t have the ability to figure out what they’ve determined.
Enter the Hucksters
There’s nothing wrong with not knowing. There’s nothing wrong with mystery!
But where there is mystery in come the hucksters! Always a buck to be made and fame to be had in telling the movers and shakers of society that polls when looked at the right way, with lots of statistics and percentages, can tell us what they can’t really tell us:
FiveThirtyEight, the prognosticator Silver founded in 2008 and left earlier this year, gives Harris a 56% chance of winning, while DecisionDeskHQ gave her a 54% chance. The Hill has Trump winning with a 52% chance.
(Nate Silver recently wrote an op-ed for the New York Times saying the candidates have a 50/50 chance of winning.)
That’s a ridiculous misuse of percentages.
Percentages are great when used properly. If a pollster says 36% of Virginians prefer vanilla ice cream and 47% chocolate, that can be useful information. Also it can be checked. Another pollster can come along and do another poll and see if the numbers match. Results aren’t always tested. Claims by recognized authorities often are blindly accepted and spread about faster than theater gossip. But the polls themselves, even bad polls, could be tested. There’s nothing fraudulent about such polls.2
But saying a candidate has a certain percentage chance of winning is nonsense because there is no way for anyone else to falsify or check the percentage. How can it be proven wrong?
There is no way to falsify the claim that a certain candidate has a 65% or 95% or 35% chance of winning. If the candidate who allegedly had a “80% chance to win” loses the election the huckster claims, Hey, 80% chance was still not 100%! We didn’t say the chance was 100%! If the candidate who allegedly had a “80% chance of winning” wins, the huckster claims it’s proof the prediction process works!
Assuming an asteroid doesn’t hit the earth or some other disaster doesn’t prevent the election from happening, there is a 100% chance that either Kamala Harris will win or Donald Trump will win or it will be a tie. Quote me on that! There is also a 100% chance that we don’t have any way of knowing now which of those outcomes will occur.
Can Campaigns Change Minds?
At the beginning of a campaign season things are different, especially during the primaries. The outcome isn’t just unknown, it truly is as yet undetermined. Most voters truly haven’t decided. Candidates can promise programs and policies that can change the world, and convince voters to trust them to implement those promises. As favorite candidates drop out, as scandals pop up, voters change their minds and preferences.
The swift Democratic convergence around Kamala Harris, her convention speech, and her successful debate against Trump really did raise her stature and win over voters. Maybe she could have kept that going if she had had a series of promises and programs ready to roll out, but her nomination came too suddenly and oddly—considering she was vice president for someone as old as Biden—she doesn’t seem to have been ready with a program. The only messages I get from her are she’ll do her best if elected, she needs money, and Trump is bad. The only messages I get from Trump are he’ll do great if elected, he needs money, and the Harris, Biden, and the Democrats are bad. But people have made up their minds and at this point we’re waiting to find out what they’ve decided.
So why can’t the Nate Silvers of the world just say this? Why invent these silly percentages? Why pretend we’re in a football game waiting to see if one of the teams will surge ahead?
What they’re doing is converting a lack of knowledge about what has been happening in the world, into a mystery about what will happen in the world. They’re converting ignorance of the present into fake expertise about the future. They’re telling the types of people who read the New York Times that no one knows what’s about to happen because it’s an unfolding story, it’s the future. This hides the truth that maybe they could have known what voters wanted for years, and crafted policies, programs, and promises to convert the Democratic advantage into a more certain electoral victory. These elections don’t have to be always fingernail-holding, cliff-hanging dramas. I’ll take this up down the road.3
Meanwhile, I don’t want to go back to years with the streets, narcissistic public vendettas, and gossip and preening. I remember that all too well. Constant chaos and conflict. So as for me I’m not going to vote for Trump just because it would be fun to stick it to the Democrats, or blow up a corrupt system. That’s my choice. But I’m not going to try to convince any of you. The rest of you have already made your decision. I’m looking forward to finding out what you’ve decided.
Except in discourse. It’s odd to hear claims that Hillary Clinton was unpopular when she beat Trump by a sizeable majority in the popular vote. The truth is that Clinton was plenty popular; she just ran a bad campaign that failed to convert her popularity into an electoral college victory.
There are polls that are actually fraudulent. Push-polls for example are polls by a political group used to try to change the mind of respondents by phrasing the question in a certain way or having a series of questions in a particular order. Also in recent years political campaigns seem to share bad polls in the effort to create a narrative of their candidate winning (on the assumption that undecideds will want to join the winning team) or losing (in order to raise money). Campaigns game polls the way colleges game rankings.
Big mistake writing on this subject....