Support NeoGAF

[brainchild]

So with the news that Bernie has won the Michigan primary, it comes as a shock to all of us... except for one person. His name is Tyler Pedigo.

I've been following Tyler's predictions for a couple of weeks now, and I've noticed that his statistical model has been getting astonishingly more accurate with every primary/caucus day. He predicted every Super Tuesday state with the exception of Minnesota, and he's corrected predicted every state since then, with fairly similar margins to the actual results.

As I understand it, Tyler uses a combination of social media and Google trends data points, and refines his predictions by honing in on what variables are key to showing how likely a state is going to vote for one candidate over the other.

Well, yesterday, he looked at his data, as was surprised to find that it was predicting an upset, despite the poll numbers.

Here's what he had to say about it:

It’s a bit unsettling to go against the grain with this forecast. As far as I know, every outlet is projecting a Clinton win tomorrow in both Michigan and Mississippi.

The Sanders campaign must be doing something remarkable in Michigan right now, because the upswing in Sanders popularity among my data sources is undeniable. I am seeing levels of interest in Bernie Sanders in Michigan similar to that of Colorado, Oklahoma, Kansas, and Nebraska. This, along with Michigan’s relatively normal demographic makeup, leads me to personally believe that he does have a chance. It leads my model to estimate that he will win there. Hillary leads every conventional poll, however, which makes me skeptical of these numbers.

Bernie Sanders will be lucky to get above 20% in Mississippi, but I do believe that if he doesn’t win Michigan, the final results will be very close. Here are the numbers:

My official prediction is that Bernie will win Michigan and Hillary will win Mississippi, but in reality Michigan is too close to call with a mathematical model. Elections culminate in a single number after the movement of hundreds or thousands of variables, and as statisticians we can only select a few of those and hope that we account for as much variance as possible. Given the outcome of all the other elections so far this season, the positions of those variables right now in Michigan seem to indicate that a massive upset will happen tomorrow night.

-Tyler

Link


What's incredible is that this guy doesn't rely on polls; just the internet and some fancy math.

Personally, I think that he should be on national television explaining his model, after seeing tonight's results.

 

[Team Alucard]

Spoiler

jk

 

[peaceiscloser]

itshappening.gif

Spoiler

jk

Wasn't the original form of that GIF bannable last election cycle? Or am I thinking of something else

Also they're sure feeling the Bern now.

 

[JesseZao]

Personally, I think that he should be on national television explaining his model, after seeing tonight's results.

That's what's fun about statistics. Dart boards.

 

[danm999]

Impressive accuracy in prediction.

 

[MThanded]

How accurate has his model been for other primaries and caucuses.

Is there a simple way to see this?

 

[commedieu]

Dare I say coincidence that everyone hates the government of Michigan right now? Literally, hundereds...

I mean it's shocking that possibly MI sees Hillary as representing that same exact bullshit that got them where they are now? So an expression of anger?

Or could be something else. But I do think there are underlying issues there. Mich. Gaf? Speak your posts!

 

[Cartman86]

People said the exact same thing about Silver and the 12 election. The nerds are what won that election!!! Yeah!!!!!!! Honestly felt weird in 2012 to have anyone saying that. The second the mainstream media says it you know you're in for some hurt.

 

[MThanded]

Look at his Minnesota prediction and reality.

62 sanders -38 clinton <- reality
47 sanders -52 clinton <- model

Dart board.

I'll follow the blog. I'm interesting in his approach. I think you shouldn't cherry pick though.

 

[Knuckle Sandwich]

I've see that guy pimping his site in the Predictit comments. He has a pretty good track record if my memory serves me.

 

[Leona Lewis]

My name is Tyler.

I hope to eventually work in the field of Public Policy for the federal government of the United States. If I can somehow blend my passion for science into that, I would be a happy individual.

I think someone's going to be a very happy individual when those job offers come rolling in the next 24 hours.

 

[King Tubby]

I'd like to see someone analyze how accurate his analysis is relative to the other analyses we've seen like 538. Getting Michigan right is impressive. Honestly I really am interested in finding out just why every single poll in Michigan was so wrong.

 

[inner-G]

People want an outsider.

I think if Hillary is the nominee we could be in real trouble

 

[Yoda]

Models aren't always right, sometimes that % error becomes realized.

 

[Ziltoidia 9]

Thing is, going on from here, Sanders has more states in play. He has an easier chance to chip away at the delegate lead as he goes. Washington state, Idaho, the Dekotas, Missouri, Wyoming, Utah, Montana, Oregon, Pennsylvania, even California (which is last).

Maybe the polls for next Tuesday are right, but if it turns out next Tuesday is pretty split (no matter how much the % wins) then I think this is going until June. I also think that as it goes, with Trump cooling off in terms of the heat on his campaign, voters might deflect to Sanders.

 

[Afrodium]

Dare I say coincidence that everyone hates the government of Michigan right now? Literally, hundereds...

I mean it's shocking that possibly MI sees Hillary as representing that same exact bullshit that got them where they are now? So an expression of anger?

Or could be something else. But I do think there are underlying issues there. Mich. Gaf? Speak your posts!

I don't think debates actually have too much of an effect on polls, but I will say that Sander's answers to questions in Sunday's Michigan debate regarding the Flint water crisis were way better than Clinton's. I plan on voting for Clinton, but seeing her answer a Flint woman's question about winning back her trust in the government with "I'll double and triple check everything" really highlighted why so many people don't find her attractive at all.

 

[brainchild]

How accurate has his model been for other primaries and caucuses.

Is there a simple way to see this?

Look at his Minnesota prediction and reality.

62 sanders -38 clinton <- reality
47 sanders -52 clinton <- model

Dart board.

Clearly you don't understand how this works.

His model builds upon itself with each data set that comes in, looking at the patterns of shifting values between variables. This is not a dart board.

I'd liken his model to a game of Battleship, if anything.

 

[ShironRedshift]

Dare I say coincidence that everyone hates the government of Michigan right now? Literally, hundereds...

I mean it's shocking that possibly MI sees Hillary as representing that same exact bullshit that got them where they are now? So an expression of anger?

Or could be something else. But I do think there are underlying issues there. Mich. Gaf? Speak your posts!

As a Clinton supporter in Michigan, I can only imagine that the bullish polls for Clinton and Michigan being an open primary state is at least part of the picture. Because of those factors, I nearly opted for a Republican ballot myself to vote against Trump but decided against it at the last moment because I didn't see the point of doing that and just couldn't bring myself to vote for any of the Republican candidates anyway. But others may have saw things differently and just gotten too confident and decided that Clinton didn't need their help and opted to vote for Republican, I guess.

But I can't really speak for anyone other than myself really since I'm a college kid and Central Michigan, like most college campuses, leaned heavily towards Bernie both among my friends and the polling that was done by the school newspaper and such and it's thus no surprise to me that he dominated in Isabella County (and the same with the other college towns).

But even being overconfident in Clinton and people opting vote strategically with a Republican ballot instead probably can only account for a part of that difference. Something more is probably going on, but I personally am not sure what that would be exactly. Like, I don't want to attribute much to the debate since that was really too recent to have much of an impact and even so I definitely personally wouldn't have called in a blowout for Sanders or anything regardless (Sanders trying to connect what happened to Flint to Wall Street in any way and using it as a segue for his stump speech, even if not completely inaccurate, came off as tone-deaf to me and that combined with like doubling down on his tweet connecting the state of Detroit and how it has areas of decrepit and abandoned buildings to free trade policies when the reality is so much more complex than that though I was disappointed that Clinton didn't more effectively nail him on that one, but she did rather effectivley pin him as a one-issue candidate and him falling right into that trap by going on a "But let's talk about Wall Street" segment anyway which had even the audience laughing), so it would have to be something more than just that as well I'd imagine.

But in any case it's definitely going to be interesting to learn more as people start to try to dig into what happened exactly.

 

[timetokill]

Maybe it's not what went wrong, but what went right?

 

[MThanded]

Clearly you don't understand how this works.

His model builds upon itself with each data set that comes in, looking at the patterns of shifting values between variables. This is not a dart board.

I'd liken his model to a game of Battleship, if anything.

I'm saying your indictment of 538 sort of works here too. Especially since his prediction outcome set is so small.

Good to have more data. However, it will be interesting to see how the model holds up. I'll keep an eye open.

I'll say his inbox will definitely be blowing up though.

 

[Damaniel]

It comes down to underestimating the effect of independent turnout in an open primary. I wouldn't have expected such high turnout from indies, but they came out in droves and broke heavily for Bernie.

I don't think poll results will be quite so skewed in closed primary states, but then again I didn't think Bernie would win MI either. Regardless, I think I'll follow this guy's blog - if anything, I like to see alternative approaches to determining primary winners and how they hold up going forward.

 

[SecretMoblin]

He predicted every Super Tuesday state with the exception of Minnesota

We're very powerful yes

 

[Makai]

According to Nate Silver, it's the biggest upset ever.

 

[brainchild]

I'm saying your indictment of 538 sort of works here too. Especially since his prediction outcome set is so small.

Good to have more data. However, it will be interesting to see how the model holds up. I'll keep an eye open.

I'll say his inbox will definitely be blowing up though.

Actually, I suggested that Nate work on getting more data outside of polls as well. For instance, if Nate had a model similar to Tyler's, I don't think he would have predicted a 99% win for Clinton.

Tyler said it was very clear that SOMETHING was happening that the polls weren't accounting for. There would be no reason for him to go against every pollster unless he felt that he had real reason to doubt their accuracy.

Of course, Tyler may be wrong in the future, but when he was wrong about Minnesota, he changed the sources of his data to account for that. Nate is not likely to change his sources to data outside of polling/weighting pollsters, which could significantly inhibit his forecasting abilities in the future.

tl;dr version

Tyler's model is more flexible than 538's, and that allows him to be more accurate with more data.

 

[injurai]

My major takeaway in this election, is how undemocratic superdelegates are. I'm disgusted by them.

 

[TheMan]

big picture, does this really help him?

 

[Darksol]

big picture, does this really help him?

Nah.

 

[egruntz]

Actually, I suggested that Nate work on getting more data outside of polls as well. For instance, if Nate had a model similar to Tyler's, I don't think he would have predicted a 99% win for Clinton.

Tyler said it was very clear that SOMETHING was happening that the polls weren't accounting for. There would be no reason for him to go against every pollster unless he felt that he had real reason to doubt their accuracy.

Of course, Tyler may be wrong in the future, but when we was wrong about Minnesota, he changed the sources of his data to account for that. Nate is not likely to change his sources to data outside of polling/weighting pollsters, which could significantly inhibit his forecasting abilities in the future.

tl;dr version

Tyler's model is more flexible than 538's, and that allows him to be more accurate with more data.

Thanks for sharing this. He has some really interesting blog posts in general. I'll keep an eye out.

Interesting that as a result of Minnesota's numbers being offed, he figured the outlier was Iowa instead and that affected his methodology. So he revisited before he ran numbers on Michigan? I'll be very curious to see what his numbers say in 1 week....

 

[Anondreas]

My major takeaway in this election, is how undemocratic superdelegates are. I'm disgusted by them.

I couldn't agree more.

 

[Makai]

big picture, does this really help him?

Not in delegates, but potential momentum going into other states.

 

[smokeandmirrors]

came outta left field

 

[TheLostBigBoss]

My major takeaway in this election, is how undemocratic superdelegates are. I'm disgusted by them.

would you call it undemocratic for a party to select it's own candidate?

There are plenty of nations that have parties select their own PM internally, superdeligates are simply a way for a party to have more control of their candidates and make sure something like Donald Trump can never happen.

I don't see and issue with that.

 

[brainchild]

Thanks for sharing this. He has some really interesting blog posts in general. I'll keep an eye out.

Interesting that as a result of Minnesota's numbers being offed, he figured the outlier was Iowa instead and that affected his methodology. So he revisited before he ran numbers on Michigan? I'll be very curious to see what his numbers say in 1 week....

His model is updated every time his predictions turn out to be right or wrong. In fact, his model relies on the confirmations of his predictions in order to become more and more accurate over time.

I'm very interested in next week's predictions as well

 

[evanmisha]

Re: Super delegates

Didn't Hillary lock up a ton of them early on in the 2008 race, but then they changed their loyalty to Obama when the polls started going in his direction?

 

[casinoman921]

Could end up being a fluke. Other than maybe Iowa, the polls have been pretty good this year (for Democratic side), and I don't see any major shifts next week. Ohio could be a lot closer, but Hillary should still be favored nicely in Illinois, North Carolina, Florida, and Missouri.

 

[Kickz]

Re: Super delegates

Didn't Hillary lock up a ton of them early on in the 2008 race, but then they changed their loyalty to Obama when the polls started going in his direction?

Yup, the fact as they are used as propaganda to inflate delegate counts is unfortunate since they aren't set in stone.

 

[hodayathink]

Re: Super delegates

Didn't Hillary lock up a ton of them early on in the 2008 race, but then they changed their loyalty to Obama when the polls started going in his direction?

Yup. The superdelegates are going to vote with whomever the popular consensus is. Right now, that's still Hillary, and if it become Bernie before the convention, then the superdelegates will switch this time, too.

 

[Neo C.]

would you call it undemocratic for a party to select it's own candidate?

There are plenty of nations that have parties select their own PM internally, superdeligates are simply a way for a party to have more control of their candidates and make sure something like Donald Trump can never happen.

I don't see and issue with that.

When you have a two party system, the least you want is to see the party having even more control of their candidates.

We accept internal election (most of the time), because we have 4 or 5 parties to choose, discounting the below 5% parties.

 

[injurai]

would you call it undemocratic for a party to select it's own candidate?

There are plenty of nations that have parties select their own PM internally, superdeligates are simply a way for a party to have more control of their candidates and make sure something like Donald Trump can never happen.

I don't see and issue with that.

A two party system exists out of necessity, to balance the left and the right of whatever the political spectrum of a nation might be. Parties don't exist for the few that agree with establishment politics to rule the middle. They exist to find the middle.

All constituents of a nation should have an equal hand in balancing where the middle lies. Right now superdelegates disenfranchise the voter. The Democratic party would rather act to lean into to the right to prove they work across party lines, rather than strengthen the party's solidarity with the left. Then the "party proper" wonders why it flounders during the midterms. The right has cracked, yet things don't move to the left. Status quo politics are disgusting.

 

[brainchild]

Could end up being a fluke. Other than maybe Iowa, the polls have been pretty good this year (for Democratic side), and I don't see any major shifts next week. Ohio could be a lot closer, but Hillary should still be favored nicely in Illinois, North Carolina, Florida, and Missouri.

Considering that his model has been getting MORE accurate with each result (not less), I highly doubt that.

He literally said that there was blatant information/data that the polls weren't accounting for, and he was right. He didn't just throw out a guess, the data in his model was in direct contradiction to the polls.

 

[casinoman921]

Considering that his model has been getting MORE accurate with each result (not less), I highly doubt that.

He literally said that there was blatant information/data that the polls weren't accounting for, and he was right. He didn't just throw out a guess, the data in his model was in direct contradiction to the polls.

Oh sorry, I wasn't referring to his model. Just all the polls that had Clinton up 20-plus points being hilariously off. You don't see that too much.

 

[shadowsdarknes]

It's truly impressive to see Bernie's supporters turn out in such numbers.

There is a voice for an outsider. Even if Bernie is not the guy this time, someone may fill the void one day.

 

[Ziltoidia 9]

Thing about it is, after next Tuesday, Sanders could win 12 states in a row. And if the trend of him winning big in the smaller stakes by 30 points continue, he will rack up delegates fast.

 

[eBay Huckster]

god damn it this is turning out more like Mondale-Hart than I thought it would

 

[Abounder]

Interesting. All I can add is obvious stuff like how polls have been failing globally - people are avoiding phone calls, and the people are angry vs the elites. Pound per pound Bernie has definitely been dominating social media and populism for the Democrats. Meanwhile Hillary only stands to lose ground as the race goes on with incidents and trends like #WhichHillary

 

[royalan]

My major takeaway in this election, is how undemocratic superdelegates are. I'm disgusted by them.

How is that your takeaway when superdelegates haven't come into play yet?

Definitely following this guy's blog.This has been an interesting night.

 

[billeh]

I've been getting GOP calls, just telling them I'm voting for Trump and strongly dislike everything about Rubio. That's what I think of them calling FIVE TIMES when I get home from work.

I'm voting for Bernie. :0)

 

[brainchild]

Oh sorry, I wasn't referring to his model. Just all the polls that had Clinton up 20-plus points being hilariously off. You don't see that too much.

If those polls are off by that much again by next week, somebody dun fucked up and it's time for an methodological overhaul.

 

[K.Sabot]

How is that your takeaway when superdelegates haven't come into play yet?

Definitely following this guy's blog.This has been an interesting night.

They come into play when pledged delegates inflate a Clinton lead to look much more dire for Sanders than it really is, disenfranchising uninformed or undecideds which plays directly into Clinton's favor.

But it's true that they technically are not in play until the end.

 

[eBay Huckster]

If those polls are off by that much again by next week, somebody dun fucked up and it's time for an methodological overhaul.

And if anyone's under-weighting independents to the extent that the Mitchell pollster was in particular, they need to self-immolate.