AI predicts US presidential election

[AngryCanadian]

Not surprising that a leftist Canadian Company is supposedly predicting a Joe Biden Landslide forecast as of Oct 30 2020.

This is the same company that claimed Hillary would win in landslide as well.
AI predicts US presidential election

An Ottawa company says its artificial intelligence can cut through biases, and gather better data than conventional political polls. Xiaoli Li finds out what Polly the AI pollster thinks might happen in the US presidential election.

"AI Predicts Biden Win"
Here's an ironic twist this AI is basically just redoing the polls and seeing the final the outcome. Not much of a actual prediction based on polls. The Canadian media so far has stayed quiet on the Joe Biden Scandal.

Rather then focusing on such issues they would rather have stories like this. Anything to pop up Joe Biden and Harris in a good spotlight. dont forget this data is not the same data was used in the 2016 story. This AI was based off a Canadian tech company. The Canadian media seems to be focusing on the American elections then the scandals that Canada has.

[Occam's Banana]

Yeah, well, I just wrote a Bash script that says Ron Paul is gonna win ...



Garbage In, Garbage Out

[Occam's Banana]

I, for one, welcome our new artificially intelligent pollster overlords.

[nobody's_hero]

Jonathan Pie predicts it as well: (FOUL LANGUAGE WARNING)

[John-G]

The good thing is that if Trump loses as the ai predicts, he would have nobody to blame but himself. The man spent most of his first term trying to appease the no Trumpers, filling his cabinet with wall street folks and managing the wars he promised in his campaign to end.

[Krugminator2]

Anyone who thinks Trump will win on Tuesday is a dimwit. He won in 2016 by a razor thin margin and was polling much better. All data points to aBiden win. That isn't an opinion. That's reality.

That doesn't mean Trump couldnt win. But everything has to go right. Basically has to recover an onside kick and score a touchdown. He probably has a 15% chance and at best 30%

[69360]

Anyone who thinks Trump will win on Tuesday is a dimwit. He won in 2016 by a razor thin margin and was polling much better. All data points to aBiden win. That isn't an opinion. That's reality.

That doesn't mean Trump couldnt win. But everything has to go right. Basically has to recover an onside kick and score a touchdown. He probably has a 15% chance and at best 30%

Who knows, who cares? I'm not voting for Trump or Biden.

The media is pushing Biden so hard I kind of figure that Trump is going to win again. They said Clinton was 90 something percent going to win too. Polls probably don't really work anymore. Trump has been so ostracized by media that many will not admit they plan to vote for him. Biden is as weak a candidate as Clinton was.

[AngryCanadian]

Anyone who thinks Trump will win on Tuesday is a dimwit. He won in 2016 by a razor thin margin and was polling much better. All data points to aBiden win. That isn't an opinion. That's reality.

That doesn't mean Trump couldnt win. But everything has to go right. Basically has to recover an onside kick and score a touchdown. He probably has a 15% chance and at best 30%

Anyone who thinks Trump will win on Tuesday is a dimwit

All the polls in 2016 were saying Hillary would win to. Just like they are now. What makes you think a CNN would be showing a poll Trump winning a landslide or a close race? even during the bush era they were showing Kerry leading.

that many will not admit they plan to vote for him

And they would someone as weak as Biden who will destroy America and its freedom beyond repair? if Biden was in the poll leading as the polls why did he need Obama's face and help for his campaign again?

[nikcers]

Who knows, who cares? I'm not voting

noooooooooooooooooooooooooo

[Anti Globalist]

I have no idea who is going to win. Last time around I thought Hillary was going to win, because I thought Trump was nothing more than a Clinton plant to help ensure a Clinton victory.

[Krugminator2]

Who knows, who cares? I'm not voting for Trump or Biden.

The media is pushing Biden so hard I kind of figure that Trump is going to win again. They said Clinton was 90 something percent going to win too. Polls probably don't really work anymore. Trump has been so ostracized by media that many will not admit they plan to vote for him. Biden is as weak a candidate as Clinton was.

Nate Silver is the most accurate political oddsmaker who publishes his findings. For example Nate Silver had Trump as 29% to win election day 2016, which turned out to be pretty good. He has him at 10% right now. https://fivethirtyeight.com/features...mpression=true

One positive for Trump is a few polls are breaking his way. I have seen about 400 political ads in the last two days. Wouldn't be surprised if those slightly benefit Trump more than Biden.

[AngryCanadian]

Nate Silver is the most accurate political oddsmaker who publishes his findings. For example Nate Silver had Trump as 29% to win election day 2020, which turned out to be pretty good. He has him at 10% right now. https://fivethirtyeight.com/features...mpression=true

One positive for Trump is a few polls are breaking his way. I have seen about 400 political ads in the last two days. Wouldn't be surprised if those slightly benefit Trump more than Biden.

You mean that same site that famously predicted 2016? 71% Hillary for sure she was to win.

[Krugminator2]

You mean that same site that famously predicted 2016? 71% Hillary for sure she was to win.

Yes. That's the one. And the fact that you (and most of the rest of the world) doesn't understand how right he got that election is a failure of government schools. A basic class in probabilities and statistics should be mandatory for every high schooler.

""If you don't get this elementary, but mildly unnatural, mathematics of elementary probability into your repertoire, then you go through a long life like a one-legged man in an ass-kicking contest."

[Occam's Banana]

In terms of percentages:

Trump's "chance" of winning can reasonably be expressed in only one of two ways: 0% or 100% (i.e., 0-in-1 or 1-in-1).

Harris's "chance" of winning can reasonably be expressed in only one of two ways: 0% or 100% (i.e., 0-in-1 or 1-in-1).

The "chance" of anyone else (Jorgensen, et al.) winning can reasonably be expressed in only one way: 0% (i.e., 0-in-1).
(It is conceivable that one of Jorgensen, et al. might win, in much the same way that it is conceivable that pink unicorns exist.)

Any other statements about the percentage "chance" of anyone winning are literally meaningless. [1] They are math-abusing expressions of ignorance and uncertainty, dressed up in (more or less elaborate) statistical gibberish and parading around in a "pretense of knowledge."



[1] To say that there is an X% "chance" of a thing occurring (where 0 < X < 100) means that a thing is expected to occur at an average rate of X times per 100 opportunities. But there is only going to be one election (i.e., one "opportunity"), the winner will win only once (not "X times"), and there will be no "average rate" (because there will be no rates to average).

Nate Silver had Trump as 29% to win election day 2016, which turned out to be pretty good.

What does this even mean? I cannot make any sense of this claim.

By what standard is Silver's incorrect prediction (namely, that Trump would lose) to be assessed as having been "pretty good?"

What difference would have been observed if Silver had put Trump's "chance" at 19% or 39% instead of 29% (apart from the utterly trivial observation of the mere relative ordinal proximities of 19%, 29% and 39% to 100% - which is what Trump's "chance" of winning actually turned out to be)?

In what sense can Silver's prediction be said to have been "pretty good" when flipping a coin would have given you better odds of correctly guessing the outcome?

[Weston White]

Ummm...isn't IA supposed to be about making logical determinations programmatically, while "learning" from relevant variables and data--and is not about making predictions? The latter is not "IA" but "predictive software", which is to date utter crap. (E.g., police departments no longer seem interested in applying software such as PredPol.)

[Occam's Banana]

Yes. That's the one. And the fact that you (and most of the rest of the world) doesn't understand how right he got that election is a failure of government schools. A basic class in probabilities and statistics should be mandatory for every high schooler.

"If you don't get this elementary, but mildly unnatural, mathematics of elementary probability into your repertoire, then you go through a long life like a one-legged man in an ass-kicking contest."

Anyone familiar with the mathematics of elementary probability knows that it is by definition meaningless speak of the statistical probability that an unrepeatable one-off event (such as a US general presidential election) will produce this, that or the other particular outcome.

[https://twitter.com/NateSilver538/st...38586520662018]

LMGDAO. Nate Silver: "But, hey, look! The excellence of our incorrect prediction is demonstrated by this unfalsifiable counterfactual about the crap we might have gotten if our prediction hadn't been wrong!"

[Weston White]

dupe post.

[Krugminator2]

Trump's "chance" of winning can reasonably be expressed in only one of two ways: 0% or 100% (i.e., 0-in-1 or 1-in-1).

No. I don't even know what to say to that.

In what sense can Silver's prediction be said to have been "pretty good" when flipping a coin would have given you better odds of correctly guessing the outcome?

No. It wouldn't give you better odds of correctly guessing.

Anyone familiar with the mathematics of elementary probability knows that it is by definition meaningless speak of the statistical probability that an unrepeatable one-off event (such as a US general presidential election) will produce this, that or the other particular outcome.

Elections happen every four years. Hundreds of years of data.


Simple question that you should answer to yourself. How do you think professional sports bettors make money?

[nobody's_hero]

No. I don't even know what to say to that.

I'm not a weatherman, but I can accurately tell you there's a 50% chance of rain every day, regardless of the weather.

Meaning, it will either rain, or it won't, and you know what? I will be absolutely right, every single day.

That's what I think OB was criticizing. A "chance of winning" is statistical magic.

[Krugminator2]

I can accurately tell you there's a 50% chance of rain every day, regardless of the weather.

Unless you live in a place that rains 182 days a year on average using many years as data, what you just said is false.

And even if rained 182 days a year on average, weather patterns would matter when forecasting the probability or rain on a given day.

[nobody's_hero]

Unless you live in a place that rains 182 days a year on average using many years as data, what you just said is false.

And even if rained 182 days a year on average, weather patterns would matter when forecasting the probability or rain on a given day.

Read what I said again.

It will either rain, or it won't. It will either rain, or it won't. It will either rain, or it won't. 50% chance of rain.

If it rains 365 days a year, I'm still right.

What Nate Silver or any other of these 'pollsters' claim to be doing when they make predictions about a 'chance to win' is not statistics, it is magic. It's not slight of hands, exactly, but more like, slight of words, or slight of numbers.

[A Son of Liberty]

Simple question that you should answer to yourself. How do you think professional sports bettors make money?

I'm not a stats guy by any stretch, but from this layman's elementary observation it seems that there are considerably fewer variables in predicting the outcome of a sporting event than there is in predicting the outcome of a national election.

[nobody's_hero]

I'm not a stats guy by any stretch, but from this layman's elementary observation it seems that there are considerably fewer variables in predicting the outcome of a sporting event than there is in predicting the outcome of a national election.

In sports, horse racing, or slots, for probability statistics to have any meaning they have to happen x number of times per y number of outcomes. A team that has had a record of 20-0 W/L is reasonably going to be predicted to win against a team that has a 0-20 W/L record. In those cases, the stats of a "chance to win" might actually mean something.

Trump will never run against Hillary again. It happened once, and no further data can be gleaned from it.

Trump has never won or lost an election against Biden. It will happen once on Nov. 3, and unless Trump comes back in 4 years and Biden hasn't been kicked out in favor of Harris, no further data could be gleaned from it.

[A Son of Liberty]

In sports, horse racing, or slots, for probability statistics to have any meaning they have to happen x number of times per y number of outcomes. A team that has had a record of 20-0 W/L is reasonably going to be predicted to win against a team that has a 0-20 W/L record. In those cases, the stats of a "chance to win" might actually mean something.

Trump will never run against Hillary again. It happened once, and no further data can be gleaned from it.

Trump has never won or lost an election against Biden. It will happen once on Nov. 3, and unless Trump comes back in 4 years and Biden hasn't been kicked out in favor of Harris, no further data could be gleaned from it.

In essence, the prediction is not "who will win"; it's "who are people more likely to vote for".

[nobody's_hero]

In essence, the prediction is not "who will win"; it's "who are people more likely to vote for".

Ah, but that's not what Nate et. al. are saying. Otherwise they would go strictly based on polling, and after 2016, there's not a polling agency in the USA that wants to stake their organization's credibility on polls. They need to measure something else. They can't just say that calling around asking opinions isn't exactly reliable anymore, because their careers depend on it.

"Chance to win" is a relatively new metric in elections. And frankly, it's a cop-out. You can never be wrong (unless I guess you go for a flat 0%, or the other way, 100%), and in those terms, you get to somehow maintain your credibility. I'm almost willing to bet that if other polling agencies raised their "chance to win" for Trump, Nate would instantly "out-bid" them by a few percentage points, but not so much as to make much difference. The only motive would be to be "more right" (or less wrong) than the other pollsters if it should happen that Trump wins. Ultimately Nate gets to claim that he was 'the most accurate' so everyone around him will say what a genius he was with numbers! , but scientifically, it's a pile of garbage.

Maybe it's just me, but I don't remember growing up ever hearing about a candidate's 'chance to win' or having anyone brag about such predictions. It's New Age analytics, or as I prefer to call it, magic.

[Occam's Banana]

Trump's "chance" of winning can reasonably be expressed in only one of two ways: 0% or 100% (i.e., 0-in-1 or 1-in-1).

No. I don't even know what to say to that.

There's no need to say anything, because the statement I made is simply and indisputably correct.

To say that an event has a percent "chance" P of occurring literally means that the event is expected to occur, on average, P times for each 100 opportunities, proportional to the ratio E / O_E (where E is the total expected number of occurrences of the event, and O_E is the total number of opportunities for the event to occur).

In other words: P / 100 = E / O_E, or P = E / O_E * 100. This is what "percent" means.

There will be one and only one opportunity for Donald Trump to win the 2020 election (O_E = 1).

Either Trump will lose that election (E = 0) or Trump will win that election (E = 1).

There are no other possible outcomes. (That is to say, there are no other possible values for E.)

Therefore, Trump's percent "chance" of winning can only be either 0 (= 0 / 1 * 100) or 100 (= 1 / 1 * 100).

Any assertion of some other percent "chance" for a Trump win is literally gibberish. Given the clear and unambiguous meaning of "percent," a claim that Trump has a 30% "chance" of winning, for example, can only be interpreted in one of two ways: that (1) Trump is expected to win, on average, 3 times out of every 10, or (2) Trump will achieve only three-tenths of a win. But neither (1) nor (2) makes any kind of sense. In the case of (1), the election isn't going to happen 10 times (or 100 times, or 1000, or even just twice), and in the case of (2), it is not at all clear what "three-tenths of a win" could possibly even mean.

The only way to even begin trying to make any kind of sense out of a claim that Trump has a 30% "chance" of winning is to do something like pretending that we have a magical "reset" button that somehow allows us to "rerun" the election over and over again. We could then suppose that saying Trump has a 30% chance to win means that we would expect Trump to win, on average, 3000 times out of every 10,000 "reset" trials that we run.

But this doesn't make any sense, either - because it would require us to discard a little thing called "cause and effect" (and without cause and effect, any talk of "predictions" or "chances" becomes completely incoherent and senseless). Assuming that nothing relevant changes from any one of our trial runs to any other, all the same causes will necessarily produce all the same effects - and thus, Trump would either win all of our runs, or he would lose all of them. So this would bring us right back to the fact that he only has either a 0% "chance" of winning or a 100% "chance" of winning, and that saying he has any other "chance" of winning (such as 30%) does not actually mean anything sensible.

The only way around this would be to drop our assumption that nothing relevant changes from one run to another - but this would take us straight into a fantasy Never-Never Land of "alternate dimensions" or "parallel universes" or some other kind of counterfactual (and thus, unfalsifiable) silliness ... (And why? Just so we can continue pretending the claim that someone has, say, a 29% "chance" of winning the election [1] actually means something sensible ... ?)



[1] Speaking of which, I asked the following question earlier regarding Nate Silver's claim that Trump had a 29% "chance" to win the 2016 election, but I did not get an answer, so I will take this opportunity to ask it again: Apart from the trivial fact of the relative ordinal proximities of 19%, 29% and 39% to 100% (or to 0%, for that matter), what observable (or measurable, or identifiable) difference would there have been if Silver had put Trump's "chance" of winning the 2016 election at 19% or 39%, rather than at 29%? IOW: What was that 29% value actually supposed to mean, in any falsifiable sense, as distinct from any other number that might have been plucked from out of thin air?

In what sense can Silver's prediction be said to have been "pretty good" when flipping a coin would have given you better odds of correctly guessing the outcome?

No. It wouldn't give you better odds of correctly guessing.

Silver gave a 29% "chance" for a Trump win.

Flipping a coin would have given a 50% "chance" for a Trump win.

Trump won.

Do the math. (Hint: 50 > 29).

Anyone familiar with the mathematics of elementary probability knows that it is by definition meaningless speak of the statistical probability that an unrepeatable one-off event (such as a US general presidential election) will produce this, that or the other particular outcome.

Elections happen every four years. Hundreds of years of data.

And Tuesdays happen every week. So what? Are you seriously suggesting that Nate Silver, et al. have incorporated "data" from 19th-century US presidential elections into their horoscopes "models?" If not, then what is the point of this remark? If so, then given that "Nate Silver [...] publishes his findings" (as you said earlier in this thread), can you provide a link to (or citation of) the "data" he used from, say, the Hancock vs. Garfield election of 1880 - or even just to his analysis of that "data" and its relevance to his "model?"

TIA

Simple question that you should answer to yourself. How do you think professional sports bettors make money?

Besides just "to myself," I'll even answer it "out loud" ...

Professional sports bettors make money by placing successful bets on sports events. The amounts and placements of those bets are determined by making educated guesses about the outcomes of those events. Successful bettors are able to make such guesses more or less consistently because the outcomes of sports events - unlike the outcomes of elections - are determined by the application of a limited number of well-understood constraints and highly regularized rules, in relation to which the bettor can take into account a finite range of clearly and objectively well-defined statistical factors (such as "batting averages," "passing yards," "free throw percentages," etc., etc.).

The same outline, BTW, applies to the question of how meteorologists are able to forecast the weather. Their forecasts - unlike forecasts of election outcomes - are made solely and strictly within the ambit of a finite number of rigorously defined and objectively measurable parameters (such as temperature, humidity, barometric pressure, etc.). This is what makes it possible to perform meaningful statistical analyses of historical meteorological data, and to usefully apply the results of such analyses to the forecasting of weather phenomena (as is clearly not the case in attempts to forecast electoral phenomena).

[Krugminator2]

There's no need to say anything, because the statement I made is simply and indisputably correct.

Might be the dumbest thing I have ever read.

Any assertion of some other percent "chance" for a Trump win is literally gibberish.

Bookmakers handicap presidential elections.

The only way to even begin trying to make any kind of sense out of a claim that Trump has a 30% "chance" of winning is to do something like pretending that we have a magical "reset" button that somehow allows us to "rerun" the election over and over again.

You can model elections off of polling data.

what observable (or measurable, or identifiable) difference would there have been if Silver had put Trump's "chance" of winning the 2016 election at 19% or 39%, rather than at 29%? IOW: What was that 29% value actually supposed to mean, in any falsifiable sense, as distinct from any other number that might have been plucked from out of thin air?

If bookmakers have Trump as a 4-1 underdog, then there is no bet to be made if your model is 19%. If the model says 39% and you make enough of those bets you will end up with all the money in the world.

Flipping a coin would have given a 50% "chance" for a Trump win.

See if you make millions with that strategy

Professional sports bettors make money by placing successful bets on sports events.

No difference between political handicapping and sports handicapping. Same thing.

[pcosmar]

No difference between political handicapping and sports handicapping. Same thing.

And either could be completely wrong. and often are.. and YOU LOSE.

I would not bet on Creepy Joe..

[spudea]

All data points to a Biden win.

Maybe you are looking at the wrong data. Trump is going to win.

[KEEF]

I predict we all will lose, and that is with a p-value way under 0.05.