Read on Twitter
"Ok Dario, I see Biden is up +10 points in National polling. But the polls were WRONG in 2016, how can you trust them now?"
Here's a useful thread to debunk this conspiracy theory once and for all.
Warning: Mucho Texto
Here's a useful thread to debunk this conspiracy theory once and for all.
Warning: Mucho Texto
"The polls were WRONG in 2016":
This shows a misunderstanding of how polling works. In 2016, the latest polls up before the election, showed Clinton was up 3 points NATIONALLY. Most polls have about a 3 point +/- error margin, with a 95% confidence interval.
This shows a misunderstanding of how polling works. In 2016, the latest polls up before the election, showed Clinton was up 3 points NATIONALLY. Most polls have about a 3 point +/- error margin, with a 95% confidence interval.
This (contrary to popular belief) does not mean a 95% chance the result is correct, but that 95% of the possible outcomes, fall within this 3 point range. The most likely outcomes being closer to the proposed result, and less likely outcomes branching out from there.
Now this 3 point margin of error, exists for BOTH projections. I.e.:
If Trump is slated to have 42% and Clinton 45% - that means that within this error margin, Trumps actual number could be 45% AND Clintons could be 42%. This is within the range of possiblities (however slim).
If Trump is slated to have 42% and Clinton 45% - that means that within this error margin, Trumps actual number could be 45% AND Clintons could be 42%. This is within the range of possiblities (however slim).
This is in fact, a 6 point total SWING that is possible (but unlikely). However still predicted under the statistical uncertainty of the polling.
Now, obviously, the job of the pollsters, is to get as close as possible to the actual result every time, and to lower uncertainty.
Now, obviously, the job of the pollsters, is to get as close as possible to the actual result every time, and to lower uncertainty.
Now, the NATIONAL popular vote polls of 2016, were actually even closer than you think, since Hilary did win the popular vote - which is what they try to predict.
However the issue, was in some state polling. A main problem in hindsight, was the lack of weighing by education.
However the issue, was in some state polling. A main problem in hindsight, was the lack of weighing by education.
In 2016, many pollsters did not weigh(adjust) for education levels in the samples, and that seemed to be a major factor in peoples preference in Trump vs Hilary.
This has been analyzed and indentified since then. https://www.nytimes.com/2017/05/31/upshot/a-2016-review-why-key-state-polls-were-wrong-about-trump.html
This has been analyzed and indentified since then. https://www.nytimes.com/2017/05/31/upshot/a-2016-review-why-key-state-polls-were-wrong-about-trump.html
Since then however, the pollsters are constantly improving (as is their job), and have implemented these measures, to make sure that polling this time around, is even more accurate. https://fivethirtyeight.com/features/what-pollsters-have-changed-since-2016-and-what-still-worries-them-about-2020/
Again, the polls in 2016 weren't _wrong_, they were mostly within margin of error, and comfortably so, however, polling can _always_ be improved to better represent the results, and generally as a profession, does pretty well at political predictions.
"Ok but what if Biden falls within these Margins?":
Generally when we try to get an accurate picture of how the election is going, we want to collect what is called an "aggregate of polls".
Generally when we try to get an accurate picture of how the election is going, we want to collect what is called an "aggregate of polls".
An aggregate, is a combination of all credible polls (that fall within certain criteria) because generally the more credible data you have, the lower your error margins are.
However, it's important to adjust for big outliers (on both sides) and a pollsters general record.
However, it's important to adjust for big outliers (on both sides) and a pollsters general record.
There are many sites that do this. The most famours ones are fivethirtyeight and realclearpolitics. They will collect all credible registered pollsters, and do an aggregate for you. This makes it super easy to get a good view on both national and state polling!
Now when we adjust for aggregates. We see that Clinton up until the election, had about a 3 point lead.
H: 45%
T: 42%
As mentioned previously, with a 3 point error, that leaves Trump with a decent chance to still take it.
H: 45%
T: 42%
As mentioned previously, with a 3 point error, that leaves Trump with a decent chance to still take it.
Now for the 2020 election. This same number looks like this:
B: 52
T: 42
With a 3 point error margin, with the best possible result for Trump, this is still WAY outside what a victory possibly could look like.
B: 52
T: 42
With a 3 point error margin, with the best possible result for Trump, this is still WAY outside what a victory possibly could look like.
Now, obviously popular vote alone, doesn't give you a victory. You could still win the electoral college, without winning the popular vote. However, state polling does not look much better for Trump right now.
In order to win, Trump would have to win the following states, that Biden has a lead in (from slimmest lead to broadest lead): (+ for Biden atm)
Georgia (+.1)
ME2(+.4)
NC(+1.5)
AZ(+3)
FL(+3.5)
NE2(+4)
PA(+7)
Georgia (+.1)
ME2(+.4)
NC(+1.5)
AZ(+3)
FL(+3.5)
NE2(+4)
PA(+7)
Suppose Trump makes a magical upset for each and every state, and takes it within margin of error. He would still have to win Pennsylvania, which is currently even OUTSIDE margin of error (3 points). https://projects.fivethirtyeight.com/polls/president-general/pennsylvania/
Not to mention, that for every time Trump beats the odds in one state, it is increasingly unlikely that he would beat them in every state, and even less unlikely (i cannot stress this enough, INSANELY unlikely) that he would beat them outside the margin of error.
"So are Trump vs Hilary and Trump vs Biden even comparable races?"
No. Not in a statistical sense. Hilary and Trump, in the month leading up until the election, if you looked at the actual polling data, were always way closer than people gave them credit for.
No. Not in a statistical sense. Hilary and Trump, in the month leading up until the election, if you looked at the actual polling data, were always way closer than people gave them credit for.
The polls were not "wrong" in 2016. They made their best prediction, that prediction was within margin of error. Most pollsters gave Trump about a 30% chance of winning, and he did.
This time around, looking at the actual data, today, Trump has less than a 5% chance of winning.
This time around, looking at the actual data, today, Trump has less than a 5% chance of winning.
Thanks for coming to my TED talk. Obviously I could write and talk about this in more detail, but this is meant as a quick debunk on this specific claim that seems to crop up increasingly now.
But what does all this actually mean?
But what does all this actually mean?
