How is it that the (Scientific) Polls, Once Again Got it Wrong!!

[Steve R.]

Seems that the entire polling industry needs to find another line of work. Many of the pollster were predicting a "blue wave" where the populous would rise-up and totally crush those evil disgusting Republicans. That never occurred. Biden was supposed to annihilate Trump, but that never happened. Instead we have a Presidential election too close to call. Republicans, in Congress were supposed to be swept in a tidal wave from office, instead they were able to hold onto their share of Congress. The supposedly scientifically based polls once again got it absolutely wrong!!!

Trump trauma: How the media blew it once again
Pollsters got it wrong again
'Biggest losers': Pollsters flunk election forecasts in 2016 sequel
Hope in the 'Great Blue Wave' of 2020

As an aside, those conducting polls are supposed to be trained in the scientific process, yet they once again were wrong. So this raises a parallel question concerning the validity of what we have been hearing from "experts" concerning Covid-19. What the "experts" say, in light of the pollster being wrong, is that the Covid-19 recommendations should viewed with a degree of skepticism. Yes, one can easily say that the science behind polling is significantly different from epidemiology. Nevertheless the public is being jerked around by fear mongering pundits who have used their supposed scientific knowledge to manipulate the public. Such as the now defunct "blue wave" or that the current response by the Trump administration to Covid-19 is somehow "ineffectual".

 

[The_Doc_Man]

The polls go by whoever responds. Usually that means the guys & gals disgruntled about something. Which this year was MORE likely to be the Democrats.

 

[Isaac]

They're really terrible.

Here's how I view it.

- Many of these people are professional data scientists (the guys who do the real work behind the scenes, not the bosses of course)
- They live & die by their career like everyone else
- They don't want to appear stupid
- Everyone knew about the shy trump voter effect, and a hundred other reasons visible to everyone why the polls were off
- Made the mistakes several times before

And one more:

- They know about the "give up" influence if polls are TOO low in a given area, for that candidate

And finally:

Their results were intentional & had a purpose.

 

[Vassago]

They are just polls. People also lie a lot in polls. I don't ever trust them.

 

[Steve R.]

Their results were intentional & had a purpose.

The polls go by whoever responds. Usually that means the guys & gals disgruntled about something. Which this year was MORE likely to be the Democrats.

They are just polls. People also lie a lot in polls. I don't ever trust them.

In summary, the pollster's are a bunch of charlatans.

 

[Isaac]

In summary, the pollster's are a bunch of charlatans.

It's only logical to assume that they wield their influence with a purpose, just like everyone else and especially media-related segments of industry.
Especially since this being drastically wrong, and only in the last few cycles, and only going one direction, hasn't always been like this

 

[AccessBlaster]

One pollster called it right the last two times by factoring in the shy vote. The Trafalgar group

 

[Isaac]

One pollster called it right the last two times by factoring in the shy vote. The Trafalgar group

Yep - like I said, it was quite possible the whole time. And these people are not stupid.. Not that many other conclusions to reach..

 

[Jon]

Polling requires both subjective and objective uses of data. Garbage in, garbage out. Accounting, you would think, requires just objective decision making. Yet I read that there are lots of subjective decisions that go on. Strange, considering it is just about numbers. But maybe classifying data comes into it and that may be more subjective.

And so it is with polling. How do you factor in the shy Conservative effect? What modifier do you use when you don't know if they are shy or not? That becomes more subjective and therefore leads to error and bias.

 

[Isaac]

How do you factor in the shy Conservative effect? What modifier do you use when you don't know if they are shy or not? That becomes more subjective and therefore leads to error and bias.

You'd have to ask the pollsters that got it right on how they did it.
According to Trafalgar, the way they successfully did it is

Two other proprietary digital methods they don’t share publicly.

My one idea is to use a DIFFERENT medium than standard polling, (say, focus groups in a totally different and perhaps more trusted atmosphere), and figure out the proportion of Trump voters that would actually lie to a phone pollster.

 

[AccessBlaster]

Also Trafalgar adds another matrix that he doesn't disclose. Pollsters sign non-disclosure forms so as not to reveal the routine.

 

[Pat Hartman]

Polls and statistics are the same. You can make them say whatever you want by manipulating the sample. I get why people do it with statistics but I don't get why they do it with polls. I would thing that you would want the best, most representative sample set you could get unless you're trying to sell something such as the "best" restaurant in town based on an exit poll of diners.

 

[Isaac]

Then there's a whole 'nother aspect I haven't seen mentioned yet.

It's the error and bias that is introduced by the person deciding which polls to include in their average, summary, recommendation, or newscast.
And that is where there is no doubt at all of the manipulative purpose behind that particular step.

 

[The_Doc_Man]

@Isaac - you are right, of course. It is possible to make the polls come out any way you want just by judicious choice of your polling population. So many polls have a built-in bias simply based on what group was chosen to be canvased.

If you had a truly unbiased polling group that somehow COULD represent the USA in microcosm, pollsters could easily anticipate the outcomes. But these days with the divisiveness permeating the country, finding that microcosm is incredibly difficult.

 

[Isaac]

@Isaac - you are right, of course. It is possible to make the polls come out any way you want just by judicious choice of your polling population. So many polls have a built-in bias simply based on what group was chosen to be canvased.

If you had a truly unbiased polling group that somehow COULD represent the USA in microcosm, pollsters could easily anticipate the outcomes. But these days with the divisiveness permeating the country, finding that microcosm is incredibly difficult.

Yeah - and I think it was you who recently mentioned about the size of the sample. Even I - not a math person - learned in Political Science "Surveys and Research" class the basic guideline of, the bigger the better! Some of these pollsters, their sample size really shocked me. I'm sure they believe they are compensating for it by some uber-sophisticated selection or randomization process, and by publishing a margin of error, but I mean they REALLY push the envelope. I saw polls with only 300 people. ??

 

[The_Doc_Man]

Yes, @Isaac - I have to find my book on the subject, but basically there are square roots involved in the computation of the error. Therefore, the error involved in small samples is fairly high. In small-sample cases that accidentally happen to be consistent, it would not be rare to find that the estimated error of the sample is greater than the actual variance of the population as a whole. In elections as close as the current presidential race, an error of 3% is SO large that you would have to say it came from a statistically useless sample.

 

[Jon]

Does it come down to, "You can never trust ths experts?" See masks thread. [*Runs for cover.*]

 

[Isaac]

Never fully trust any industry who changes their minds every 2 weeks or 2 months or 2 years