Can You Give an Accurate Explanation for Why the Polling in 2016 Failed to Predict Trump's Victory?

[Sunstone]

Can you give a scientifically accurate explanation for why the polling in the 2016 presidential election failed to predict Trump's victory in the electoral college vote but did in fact predict his loss in the popular vote?

BONUS QUESTION: On a scale of one to five, how knowledgeable of statistics do you consider yourself? The greater the number, the more knowledgeable.

 

[Estro Felino]

Because , I think, media only focus on California and NY, almost ignoring the will of the flyover people.
And this happens in polls too.

 

[HonestJoe]

Can you give a scientifically accurate explanation for why the polling in the 2016 presidential election failed to predict Trump's victory in the electoral college vote but did in fact predict his loss in the popular vote?

I thought this had been explained and fairly well understood. The popular vote is easier to predict because it’s a simple count and would only turn on millions of votes. The actual electoral college is harder because it is more complex and has the potential to turn on a handful of districts, potentially a handful of votes in specific places. There is basically much more scope for error and a much tighter margin for that error to be significant there than with the popular vote.

BONUS QUESTION: On a scale of one to five, how knowledgeable of statistics do you consider yourself? The greater the number, the more knowledgeable.

Maybe a 2.5 on statistics in general and 1.5 on electoral statistics specifically (i.e. enough to know there is a difference ). Most of my knowledge is based on common sense and literacy (which is sadly enough to be above average in a lot of circumstances).

 

[Enoch07]

Can you give a scientifically accurate explanation for why the polling in the 2016 presidential election failed to predict Trump's victory in the electoral college vote but did in fact predict his loss in the popular vote?

BONUS QUESTION: On a scale of one to five, how knowledgeable of statistics do you consider yourself? The greater the number, the more knowledgeable.

Hillary didn't campaign in certain states because she thought they would vote Democrat (as per usual). But Trump changed their minds. Also conservatives usually poll low, something about having a job, family and more important things on their mind then answering polls.

 

[Augustus]

Can you give a scientifically accurate explanation for why the polling in the 2016 presidential election failed to predict Trump's victory in the electoral college vote but did in fact predict his loss in the popular vote?

Both races should have been given as 'too close to call', but people like constructing narratives out of noise and then constructing more narratives to explain why their narrative built on noise were wrong.

 

[PureX]

People lied about their intent to vote, and for whom.

 

[Phaedrus]

Hilary had the popular vote. Unfortunately, an outdated and unnecessary Electorate College decided to go against the wishes of the people and elect Trump instead. It has happened before, and always the Republican was chosen when the people wanted a Democrat. Clearly, there is a bias within the EC.

 

[Stevicus]

Can you give a scientifically accurate explanation for why the polling in the 2016 presidential election failed to predict Trump's victory in the electoral college vote but did in fact predict his loss in the popular vote?

BONUS QUESTION: On a scale of one to five, how knowledgeable of statistics do you consider yourself? The greater the number, the more knowledgeable.

I don't think anyone can give a "scientifically accurate explanation" for it. We're talking about human choices based on whimsy and frivolity.

The use of polls itself could also be a factor. In addition to all the other falderal which accompanies every election, the voters are inundated with endless polls and predictions which are ostensibly designed to influence voters more than anything else.

 

[bobhikes]

Can you give a scientifically accurate explanation for why the polling in the 2016 presidential election failed to predict Trump's victory in the electoral college vote but did in fact predict his loss in the popular vote?

BONUS QUESTION: On a scale of one to five, how knowledgeable of statistics do you consider yourself? The greater the number, the more knowledgeable.

Everything I read says it didn't fail. All polling gave Hillary a slim advantage over Trump. She did win the popular vote and only lost the electoral vote. A poll would have to be really huge to be able to decide electoral over popular and the pollsters did not do anything like that.

Closest poll to the election,

Poll
Bloomberg/Selzer
Nov 4 – Nov 6, 2016
799 Likely Voters

Trump 43
Clinton 46
Other 5
Undecided 6
Spread Clinton +3

 

[Brickjectivity]

I think its a good thing that the polls can't accurately predict the outcome. It works both ways. Gotta realize, too, that its a first past the post system, which means that through the electoral college the minority can still win.

The purpose of the electoral college is to allow the minority to have more influence, specifically it helps the less populated states. That is one of its functions.

Senator Clinton had a fair chance of winning last election cycle. She could have won, but small variables effected a different outcome.

 

[metis]

I don't claim"accuracy", but when I was out of the country my oldest daughter called and told me that Comey had just announced that he was reopening the investigation on Hillary just days before the election, and I told her that this would likely give Trump the election, and I believe it did.

Polls reflect what people believed in days before the day that the polls are announced publicly, so that lag in time can't reflect last-minute changes.

 

[Revoltingest]

If polls don't take into account geographical distribution,
then the electoral college can throw a monkey wrench
into a simple Dem vs Pub percentage.

 

[9-10ths_Penguin]

Can you give a scientifically accurate explanation for why the polling in the 2016 presidential election failed to predict Trump's victory in the electoral college vote but did in fact predict his loss in the popular vote?

The all-or-nothing assignment of electoral votes for most states, combined with very small winning margins in several cases were both big factors. Whole states swung one way or the other based on variation within the polls' confidence intervals.

Also, though I'm not sure how much it was a factor, my impression is that "likelihood to vote" measures aren't terribly accurate. That may have also played a role.

BONUS QUESTION: On a scale of one to five, how knowledgeable of statistics do you consider yourself? The greater the number, the more knowledgeable.

Four, I guess? I don't deal with multivariate stats very often, but I analyze stats for a living.

 

[idav]

Can you give a scientifically accurate explanation for why the polling in the 2016 presidential election failed to predict Trump's victory in the electoral college vote but did in fact predict his loss in the popular vote?

BONUS QUESTION: On a scale of one to five, how knowledgeable of statistics do you consider yourself? The greater the number, the more knowledgeable.

Every stat has a two percent margin for error and many of the states were 51% to 49% very narrow victories. Be better off flipping a coin. Florida is always a toss up along with other swinger states. I can predict with high accuracy that the president election in 2020 will be a two percent difference on who wins.

A question I am curious about is the percentage of states that were correctly predicted. Swinger states are likely harder to predict.

 

[Thief]

Can you give a scientifically accurate explanation for why the polling in the 2016 presidential election failed to predict Trump's victory in the electoral college vote but did in fact predict his loss in the popular vote?

BONUS QUESTION: On a scale of one to five, how knowledgeable of statistics do you consider yourself? The greater the number, the more knowledgeable.

I have seen a report intended to answer that very question
found it on YouTube....as I recall

and the documentary goes into detail of how people manipulate the media
in covert attempts to influence public opinion and the vote

If I can find it again ….I will drop a link

 

[9-10ths_Penguin]

Every stat has a two percent margin for error and many of the states were 51% to 49% very narrow victories. Be better off flipping a coin. Florida is always a toss up along with other swinger states. I can predict with high accuracy that the president election in 2020 will be a two percent difference on who wins.

The confidence interval varies from poll to poll, depending mainly on the sample size. For instance, this random poll from Michigan just before the election gave a 95th percentile confidence interval of +/- 3%:

Michigan (November 5, 2016) v1 (1) | Hillary Clinton | United States Government

And within that confidence interval, they were bang on.

Now... something I see in Canadian election coverage but - at least from what I've seen - not so much in American coverage: the term "statistical tie." If two parties/leaders/whatever come out so close in the poll results that the two results are within each others' confidence intervals, the news will just describe it as a tie for all practical purposes.

... but I guess when every poll in almost every state results in a statistical tie, reporting it that way doesn't create much fodder for panel discussions with analysts. A +/-3% confidence interval is massive when the winning margin in the state is 0.3%.

Oh - the other thing to keep in mind with confidence intervals: they're almost always given for the 95th percentile. IOW, 1 out of 20 times, on average, the real result will be outside the confidence interval. So if you, say, slice the electorate into 50 state-sized pieces, odds are that in most elections, 2 or 3 states will be way off what the polls projected, even after taking into account all the limitations of polling data.

Edit: I'm being serious about the difference in how polls are reported. Here, it's very common for a reporter to say something like "the polls show the Liberals 0.5% ahead of the Conservatives, but the poll margin of error was plus or minus 2.5%, so that lead isn't meaningful. The two parties are in a statistical tie."

 

[Augustus]

Every stat has a two percent margin for error and many of the states were 51% to 49% very narrow victories. Be better off flipping a coin. Florida is always a toss up along with other swinger states. I can predict with high accuracy that the president election in 2020 will be a two percent difference on who wins.

The real MoE is actually a fair bit larger than the stated MoE as it assumes accuracy of sample (well often they don't assume this and just fudge the number a bit based on what they assume the sample imbalance to be).

If we are very generous and give them a 5% real MoE then a 54-46 is really a 49-59 v 41-51 i.e too close to call. How many places fall into a MoE such as this?

Also you get loads of BS narratives forming 'Clinton's lead increases to 2%' is simply noise and should be 'race still too close to call'.

 

[Brickjectivity]

Four, I guess? I don't deal with multivariate stats very often, but I analyze stats for a living.

That's at least a 4.5. Maybe you aren't a CDC researcher, but that is a 4.5.