Apologies for the political post, but as a US citizen and voter, you might as well spill a few pixels. Yesterday was, well, unexpected. And that leaves us with an important lesson, that...

The bell tolls for soft sciences

I think I took some screengrabs somewhere of so many websites predicting Clinton's win likelihood percentage. CNN, The New York Times, Nate Flyboy Silver's FiveThirtyEight -- they all had Clinton at a 70+% chance of winning. Even though, in poorly circular reasoning, I imagine, I mostly believed that'd be the outcome, I wondered what stupid alchemical dances some were doing to get percentages to the tenth of a percent. Come on. Show me a predictive poll, and I'll show you the LSU/Alabama game.

Do polls have some predictive value? Sure. Polls are best at predicting what the numbers will be for the next poll taken with similar conditions. They're not as good for predicting something complex that only happens once. If you can't repeat it, is it really a science? We might be better than tea leaf readers now, but we're still just guessing.

I got tired of hearing about the "silent" majority or "quiet" majority -- or whatever the term of art would eventually be chosen by each network's cobbled-together crow eating festival last night as I watched, transfixed. (Kudos to Chuck Todd on NBC for being the first by about a half hour to start chewing, and to immediately accurately describe how large the plate of crow was.) If you'd polled the right people, you'd have gotten answers from the "silents". That is, good polls have no silent demographics. You know what that means: Either pollers are complete idiots and can't sample to save their lives (partially true, I bet), or people aren't always honest when they answer calls from polls. Welcome to the 21st century, "scientists".

We're all in the media now

I wonder how much it's like trying to get an unbiased jury pool. People don't just read the news now, they're saturated with their news. Everyone's tainted. Everyone has a nuanced relationship with not just the news, but the news media. Heck, look how many of us are media producers, are broadcasters. We all have blogs, Medium accounts, Twitter accounts, Facebook pages. We're all publishers, and, worse, so many of us are republishers. We're all broken Facebook news aggregators.

That is to say, people who self-identify as "likely voters" are already by definition folks who listen to the media, participate in the media, in some senses are the media. They know the "right" answers. Like ESPN talking heads predicting football games in their echo chamber, none brave enough to leave the herd, once one person goes on the limb, the lemmings follow. Worse, the lemmings approach the cliff together, checking in with each other, making sure they all agree to follow the same script, and jump all at once. At least if the drop's fatal, they have company. (And that means job security.)

Talking Head Echo Chamber

Don't believe me? Check out the picks for last week's NFL games from NFL Pick Watch:

Grid of NFL picks from last week. Giants swathes of people picking the same winners -- and losers! -- en masse

Not only does everyone pick the same team, they often pick the same losers to win too. Hey, look, we're all 8-5 or 9-4 together!

And that collective echo chamber is what we see here with pollsters. "My poll matches their poll so I'm guessing we're on to something." Yesterday shows us that's not good enough.

Why polls aren't like betting on sports

There's a reason the sports bookies stay rich, relatively speaking, and don't seem to be as bad at picking games as folks on TV. Bookies know something we don't: How people have made their bets to this point. Then they adjust their lines, and eventually make money. It's not really about the games. Casinos are really just selling lines so to minimize their overall risk, and making money on the juice you pay to play the game. They don't care who wins, or even if they "got it right" the first time.

From complex.com:

When oddsmakers create betting lines, the goal is simple: To get people to wager on both teams equally. They want just as many bets to be placed on Team A as Team B. Now, we know what you're thinking: Doesn't that mean that they just end up breaking even? No, they don't, and it's because of the "vig" (also known as the "juice") that's attached to each and every bet that a sportsbook takes.

Here's how it works: When you place a wager in order to win, say, $100 on a point spread bet, you'll be required to bet slightly more than $100. The amount varies depending on the vig that your sportsbook charges. But in most cases, you'll likely be required to place a $110 bet in order to win $100 back. 

This really isn't unlike what Costco's making their profit from memberships, not goods sold.

From Why Membership Fees Are So Important to Costco Wholesale Corporation -- The Motley Fool:

A club, not a retailer
Membership revenue totaled $2.4 billion last year, up 6% from 2013. Meanwhile, Costco booked $3.2 billion in total operating profit. So, subscriber fees accounted for 75% of earnings, which shows why its more useful to think of Costco as a membership club rather than a retailer.

Does that make sense? Not even sports books can really predict a game. They take their insider knowledge -- how people are betting on their initial guess at who is going to win and by how much -- and then tweak the numbers continually to approximate neutrality.

To have a similar situation in politics, we'd have to only let 5% of the population vote each day, factor in the results, and readjust our predictions. (Actually not a completely horrible idea a priori...)

Can you scientifically predict a unique event?

Polling isn't supposed to be prognostication, and it sure isn't a Vegas casino taking bets. It's supposed to be scientifically predictive. Instead, polling has become a Mobius strip of popular opinion that's found it safest to ignore both the influence of fickle psychology and its hand in creating it from its own previous polls.

The worst part is that there's no second chance. This isn't a seven game series, where Clinton could win the next two nationwide elections and pull back ahead, more accurately testing if she really had a 70+% chance of winning this election. Let's painfully undercut my credibility and substitute this from wikipedia.org, as a proxy for what we've all been taught about the Scientific Method:


If an experiment cannot be repeated to produce the same results, this implies that the original results might have been in error.

Here's your corollary: If an experiment cannot be repeated at all, it's not science. There's a reason why doctors practice. You're always a special case. (I swear. ;^D)

Polls are folks practicing alchemy and divination, and we need to make it clear to everyone to base their actions on their own convictions, not the stupid 10,000' view explanations from the THEC (Talking Head Echo Chamber) saying why the entire nation seems to have answered questions with a three-point swing. Don't weigh your vote using the silly rationale of a talking head explaining away a shift in a poll: the way a candidate smiles or stands near someone or waves or sighs or coughs. Vote because of your belief in their platform, and their track record in getting the planks of their platform done.

What a broken mess we've made of elections, regardless of how your candidates fared.

Speaking of echo chambers, did any network call Trump winning before 2:30am? I think the AP was first to call enough states that he couldn't lose, but who was the first group with any substance to call the race for Trump? Clinton's campaign. The same group that wouldn't address their supporters who'd waited hours on hours at the Javits Center. The networks had less gumption than a group of pretty poor losers. I'm not sure who to congratulate from this motley crew.

Labels: ,