Monday, October 24, 2016

So much for the opinion polls


I'm sure that by now, most of my readers have learned about the incriminating e-mail sent by the Clinton campaign as long ago as 2008, and just revealed by Wikileaks.  In case you missed it, here's the salient excerpt.

I also want to get your Atlas folks to recommend oversamples for our polling before we start in February. By market, regions, etc. I want to get this all compiled into one set of recommendations so we can maximize what we get out of our media polling.

There's more at the link.

Zero Hedge points out:

The email even includes a handy, 37-page guide with the following poll-rigging recommendations.  In Arizona, over sampling of Hispanics and Native Americans is highly recommended: 
Research, microtargeting & polling projects
  • Over-sample Hispanics
  • Use Spanish language interviewing (Monolingual Spanish-speaking voters are among the lowest turnout Democratic targets)
  • Over-sample the Native American population
For Florida, the report recommends "consistently monitoring" samples to makes sure they're "not too old" and "has enough African American and Hispanic voters."  Meanwhile, "independent" voters in Tampa and Orlando are apparently more dem friendly so the report suggests filling up independent quotas in those cities first.
  • Consistently monitor the sample to ensure it is not too old, and that it has enough African American and Hispanic voters to reflect the state.
  • On Independents: Tampa and Orlando are better persuasion targets than north or south Florida (check your polls before concluding this). If there are budget questions or oversamples, make sure that Tampa and Orlando are included first.
Meanwhile, it's suggested that national polls over sample "key districts / regions" and "ethnic" groups "as needed."
  • General election benchmark, 800 sample, with potential over samples in key districts/regions
  • Benchmark polling in targeted races, with ethnic over samples as needed
  • Targeting tracking polls in key races, with ethnic over samples as needed

Again, more at the link.

This absolutely confirms the recent revelation that the Clinton campaign was up to shady tricks (to put it mildly) in major media polling of potential voters.  They've been doing it for years - don't forget that the e-mail quoted above dates back to 2008!

It also explains recent triumphalist claims by the Clinton Campaign, for example:  'Hillary Clinton is so far ahead of Donald Trump in the race for the presidency that she no longer even feels the need to pay attention to the Republican nominee.'  As is now clear, she's mainly ahead in polls that have been deliberately skewed in this way, so as to portray her as so far ahead that the election is effectively a 'done deal'.  I suppose that's to try to persuade potential Trump and Republican voters not to bother to cast their vote, as there won't be any point.  Instead, they should stay home on election day and let events take their presumably inevitable course.

Thing is, of course, they're not inevitable.  Other polls (for example, this one) portray the race as much, much closer.  All of us have a voice, and every voice (and every vote) counts.  It's up to us to use them.

Peter

14 comments:

SiGraybeard said...

Nobody should be surprised that the point of the polls is to make one chosen candidate look good. The bad stuff is reports coming out of South Florida of pollsters going to Hispanic houses, conducting the a poll entirely in Spanish, and then giving the respondent a piece of paper entirely in English asking them to sign - that turns out to be an ideal identity theft vehicle that could be used as a way to get an absentee ballot.

Until about next week at some time, the point of polls is to create opinions, not to measure them. There's a strange symbiosis between the polls and the news. To the chattering media, from the hard left NY Times and CNN to the slightly left Fox News, all they do is report the polls as given gospel.

At some point, they have to try to be accurate so they can sell their services as being "the one who got it right". It's really not worth paying attention to polls until just about election day.

Anonymous said...

As many people have pointed out already, this is not what oversample means. Oversample is when you make your sample size larger to reduce the margin of error among demographics. E.g. The margin of error in a hypothetical statewide poll is +/- 4% with a sample size of 1000. If you were to try to extrapolate the numbers to determine the hispanic demographic of that vote it would be say 100. So with a lower sample the margin of error is much higher. To solve this you would sample 1000 Hispanics. This does not change the result of the larger poll.

Google it, lots of people have explained it better than I can.

Anonymous said...

One other use of polls is to support the results of the election after voter fraud has been committed. if the fraud results in election results similar to the polls then who can complain. And if Trump wins the the polls will be cited to prove voter fraud on the republican side.

Rolf said...

Two things:
What Anon above said.
Psy-ops; a continuous drumbeat of polls showing Shrillary well ahead are wearing, and might depress the anti-Clinton vote by depressing them to the point of giving up, staying home, and (the Dem party hopes) blowing their brains out. Yes, some of the left really do hope that the right (however you define it) would just hurry up and die already so they can loot our stuff with fewer messy objections.

Here in WA it's entirely vote-by-mail. No precincts to go into a booth. So I have ZERO faith that there are not boatloads of ballots waiting in the wings to make sure that things have the "right" result. Zero. I'm voiceless. But that's white male privilege for you.

Total banana republic.

William Bianco said...

Gentlemen,

Please google the term "oversample" - it's a standard tactic for increasing the accuracy of a poll that increases the precision of estimates for subgroups. No reputable pollster reports raw data from an overweighted sample. In fact, no one reports unweighted data, as there are various corrections that increase (have been found to increase) the precision of poll-based predictions. Polling is something of an art, people of good will can disagree about how best to construct a sample and interpret poll data, but to jump to pollsters being in the tank for Clinton based on oversampling is madness.

I say this all with the greatest possible sympathy to your point of view. Zero Hedge says some smart things. This isn't one of them.

WB

William Bianco said...

And FWIW, I base my argument on 30 years of interpreting and doing surveys as a professor and as a consultant to various individuals and groups. There is a simple explanation for why virtually all polls show Trump behind -- he's losing.

Will said...

Professor Bianco,

the fact that you are/were a professor brings doubt to your conclusion that Trump is losing, as the inherent political ideology of virtually all staff members of the schools will bring a strong negative bias to bear on the subject.
One of the factors that the left cannot explain in relation to the poling numbers is the overflowing crowds that show up for every Trump gathering, even in the face of direct violence by Clinton supporters, and the virtual dearth of supporters that show up for Clinton.

In addition, to maintain accuracy in surveys, you have to understand how the data can be mismanaged, both through incompetence and deliberate fraud. I'm sure that observations by you on this subject would be considered carefully.

bmq215 said...

Peter, you're probably not especially familiar with standard statistical terminology but oversampling is a common technique intended to increase precision, rather than to bias any overall results. These emails simply highlight those groups for which the dems want the highest quality information. It is therefore entirely unsurprising that the oversampled groups are those that seem likely to have dem leanings.

bmq215 said...
This comment has been removed by the author.
William Bianco said...

Will, FWIW, Trump's crowds aren't so big - Romney was attracting larger audiences at this stage of the campaign, or so his staff say (sorry - don't have a reference). That's a fact, not an interpretation.

McChuck said...

You flaw is in assuming that those who own the polling companies care about their reputation for fairness and accuracy. Most of them don't. They care about supporting their friends, and ensuring that the polling supports the position/side that paid for the poll.

Pollsters are human, and most of the major ones collude with the Democrats and the media (but I repeat myself). It has been obvious for a long time. The pattern is always the same.

The media and the polls also influence the primary races to try to ensure that the selected candidates are elected. The Democrats pick their favorite Republcans to run against, then those people get the favorable news and opinion pieces, and favorable polling numbers. After the primary, the media starts turned by against them. The Republican polls well early, then starts dropping until in October the Democrat shows a steady lead.

You do know that there are published polls and unpublished polls, right? The published polls are primarily propaganda. The unpublished polls show the truth, but privately. These are the ones that the pollsters try their best to be correct with.

It's like the difference between writing in your private diary versus writing a newspaper editorial.

McChuck said...

And I based my statementsabove on a former career in intelligence/counterintelligence. I know information operations when I see them.

BTW, where do you think most third party candidates get their funding and support? What would Trump's polling numbers look like without the unprecedented support for Gary Johnson, the Libertarian candidate?

McChuck said...

Technical quibble versus layman's words. Accurate observation, but wrong conclusion.

If a poll samples 1000 registered Democrats, and only 10 registered republicans, then shows the Democrat candidates with insurmountable leads, is the poll still accurate?

bmq215 said...

Minecraft Chuck, depending on the methods they're using the answer to that could very well be yes.

I know, I know, that seems ridiculous given the comically lopsided example that you listed but bear with me. Statistical analysis is rarely as straightforward as the examples you see in high school textbooks.

For example, perhaps we wish to predict the overall result of the election but desire a much smaller confidence interval (i.e. more precise estimates) for how registered democrats will vote than registered republicans. We can poll many more dems than republicans and then weight each group by the party affiliation proportions for the area of interest. This could yield a result that is relatively accurate despite the lopsided sampling.

Of course the actual methods used by pollsters are much more complicated than the ones I'm describing here, but it should be clear that unequal polling can still provide accurate results. In fact, in many cases it may actually produce estimates that are more accurate.