Comparison of Polls to Election Results

[DuJay]

Comparison of polls immediately before Election Day to election results shows that the national popular vote and a Clinton win was accurately forecast. The state-by-state popular vote was also fairly accurate for Clinton, but not for Trump. Polls did not forecast surges of vote for Trump in states with close races like the key swing states of MI, WI, and PA, which were forecasted for Clinton, and also NC, SC, MS, OH, IA, and AK. Trump won all of these states due to the Trump surges which had not been forecasted.
 
This picture does not seem to add up. How is it that the polls could be accurate for national popular vote and for state popular vote for one candidate, but not for the other?

Since votes for Clinton did not drop in the swing states, the Trump surges must have been all of the undecided polled voters. Especially in MI, WI, and PA, where polls forecast a Clinton win, why didn’t Clinton get even close to a proportionate share of the undecided vote?

Is it just coincidence that the Trump surges happened to occur in all the key swing states Trump needed to win the electoral college, but not in the neighboring states?

Doesn’t this picture look suspicious, as if some manipulation of the votes may have occurred?

[Tender Branson]

That is not entirely true ...

Polls in MI and PA definitely narrowed in the final days: I remember that the polls had Hillary up by 5-10 in PA and MI 2 weeks before election day and the final polls only by 2-3%, so the trend towards Trump was there.

Some polls, I guess Trafalgar even had Trump slightly winning there.

So, the polls were not terribly wrong using the MoE.

WI polls were definitely off though.

[JustinTimeCuber]

Polls just give a rough estimate of what will happen in an election. When people vote, there are certain factors- mainly, which candidate did a better job at getting their supporters to turn out?

The difference between the inaccuracy of national forecasts and state forecasts was less extreme than some people may think. Using 538's final election forecast and comparing that to the national results, we see:

Actual	Forecast	Difference (Actual vs Forecast)
D+2.0	D+3.6	R+1.6

In a few swing states:

State	Actual	Forecast	Difference	Difference Anomaly*
MI	R+0.2	D+4.2	R+4.4	R+2.8
WI	R+0.7	D+5.3	R+6.0	R+4.4
PA	R+0.7	D+3.7	R+4.4	R+2.8
FL	R+1.2	D+0.6	R+1.8	R+0.2
NC	R+3.6	D+0.7	R+4.3	R+2.7
NV	D+2.4	D+1.2	D+1.2	D+2.8
NH	D+0.3	D+3.6	R+3.3	R+1.7

*Difference Anomaly is defined such that if the difference is R+1.6, the difference anomaly is EVEN.

You can see that Trump outperformed more than he did nationally in most of the important swing states. Hillary outperformed more than she did nationally (since she underperformed nationally, that number is negative) in some solid blue states. Hillary won by 30.1 in California but was only projected to win by 23. Factoring in the Trump national swing, she would have won by just 21.4. That's a big difference of 8.7 points.

[DuJay]

I appreciate  the detailed analysis giving specific values for the differences between polls and election results.

However, the analysis still begs the question why the anomalous Trump surges happened to occur in all the key swing states Trump needed to win the electoral college vote.

[Tender Branson]

I appreciate  the detailed analysis giving specific values for the differences between polls and election results.

However, the analysis still begs the question why the anomalous Trump surges happened to occur in all the key swing states Trump needed to win the electoral college vote.

That's not true: States such as IN, MO, etc. also had polls that were much tighter than the election day result.

Probably because white-working class voters refused to tell pollsters their real intention and then unleashing their wrath against Hillary and Dems on election day.

The same to a smaller extent with Latinos in states such as AZ, TX etc. - just in favour of Hillary.

But none of this is any indicator that the election was rigged, like you suggested above ...

[DuJay]

If I am interpreting Justintimecuber's analysis correctly, the only anomalous states are those listed. The list includes all the key swing states Trump needed to win the electoral college vote.

[vote for pedro]

Polling isn't an exact science.  Plenty of people won't answer; or won't answer truthfully, when every media outlet proclaims their candidate cannot win.

[vote for pedro]

If I am interpreting Justintimecuber's analysis correctly, the only anomalous states are those listed. The list includes all the key swing states Trump needed to win the electoral college vote.

You are interpreting his analysis incorrectly.    Only the key swing states you asked about are listed.  And those states have relatively small anomalies.  You could look at the anomalies for non-swing states but no one cares!  If the Democrat is winning D.C. by 70 points or 87 points, they are still winning D.C.

In a few non swing states:

State	Actual	Forecast	Difference	Difference Anomaly*
DC	D+86.8	D+70.5	D+16.3	D+17.9
WV	R+41.7	R+26.5	R+15.2	R+13.6
SD	R+29.8	R+15.5	R+14.3	R+12.7
HI	D+32.2	D+23.7	D+8.5	D+10.1

*Difference Anomaly is defined such that if the difference is R+1.6, the difference anomaly is EVEN.

The farther a state is from becoming a swing state, the less effort is going to be wasted on polling there. 


I'll leave it to you to look at the results vs. polls for Minnesota and Maine.  The same so called "Trump surges" you are talking about happened in those places.

[BoAtlantis]

I appreciate  the detailed analysis giving specific values for the differences between polls and election results.

However, the analysis still begs the question why the anomalous Trump surges happened to occur in all the key swing states Trump needed to win the electoral college vote.

It could be that those states happened to have disproportionate % of Obama voters who wanted to think it out before going to Trump. People who cross party lines to vote obviously tend to take longer to ponder before making the decision.

[DuJay]

Is it not anomalous that all the key swing states needed for an electoral college win were all  anomalies, to the degree they were, in the comparison of polls to election results?

[zorkpolitics]

There is a simple answer for why Trump won.  He campaigned in swing states.

Compared to 2012 Trump improved on Romney by 5.4% in the 13 swing states he actively campaigned in, ran ads, had rallies.  In the other 37 states and DC, he actually performed 0.1% worse than Romney.
So where local people heard his campaign they voted for him, where they didn't and only were fed a diet of media driven negative reporting, he did not do better than Romney.

So in swing states which flipped from Obama to Trump the poll models were based on the 2012 electorate, but that electorate did not vote the same in 2016 in the swing states, and the polls over predicted the vote for Clinton.  In the rest of the country, absent local campaigning by Trump, the 2012 electorate was a good model for 2016 and the polls more closely predicted the result.

[RI]

Is it not anomalous that all the key swing states needed for an electoral college win were all  anomalies, to the degree they were, in the comparison of polls to election results?

JTC's "Difference Anomaly" by state:

The polling error in WI, MI, and PA was much smaller than in neighboring Midwestern states IN, IA, MO, OH, etc. In general, the polling in swing states was pretty good compared to other states.

[SingingAnalyst]

The same thing happened in 1992. Gallup's final poll showed 14% for Perot; he won 19%.

Of course, this didn't affect the final result.

Trump voters, like Perot voters, were more likely to refuse to answer polls.