The Problem With the Pew Poll

[McCamy Taylor]

Abstract: This is a long thread, so I post my conclusion right here. The recent Pew presidential poll that everyone is so excited about predicts that both Hispanics and Blacks are going to stay home this election, while white men will follow the old adage “Vote early, vote often.”

Team Romney must be jumping for joy. The latest Pew Poll shows him surging ahead of Obama among likely voters, 49 to 45% in the wake of last week’s debate. Sure, Romney lied and lied and again. But, we are told, America secretly wants a braggart and a bully to be its leader----

Excuse me, make that white America. Male white America. But America is no longer overwhelming white. Minorities are set to become the new “majority” sometime in the near future. And as their numbers rise, their political clout rises, too.

I decided to do a little investigating, so I turned to a reliable polling firm---Pew--- to see how the rising percentage of minority (nonwhite) voters has affected presidential races in the recent past. Here is a tabulation of votes cast by race from 1988 to 2008. Note that the total percentage was 82% white in 1988. By 2008 that number had fallen to 73.4 percent. Minority participation went up, with 24% of the votes cast in 2008 coming from Blacks, Hispanics and Asians. Black women, in particular, saw their participation rise.

The levels of participation by black, Hispanic and Asian eligible voters all increased from 2004 to 2008, reducing the voter participation gap between themselves and white eligible voters. This was particularly true for black eligible voters. Their voter turnout rate increased 4.9 percentage points, from 60.3% in 2004 to 65.3% in 2008, nearly matching the voter turnout rate of white eligible voters (66.1%). For Hispanics, participation levels also increased, with the voter turnout rate rising 2.7 percentage points, from 47.2% in 2004 to 49.9% in 2008. Among Asians, voter participation rates increased from 44.6% in 2004 to 47.0% in 2008. Meanwhile, among white eligible voters, the voter turnout rate fell slightly, from 67.2% in 2004 to 66.1% in 2008.

http://pewresearch.org/pubs/1209/racial-ethnic-voters-presidential-election

Now, let’s do what most of us never do. Let’s look at the makeup of the so called “likely” voters that Pew polled last week. To do so, we have to turn to page two (something we really ought to do before we start talking about poll numbers). According to Pew, this year minority voters are going to become apathetic, while white voters will develop a re-surging interest in presidential politics.
http://www.people-press.org/2012/10/08/romneys-strong-debate-performance-erases-obamas-lead/2/

Pew says that Latino voter turnout is going to be bad. In 2008, Hispanics cast 7.4% of the vote. This year, despite their growing numbers, they will count for only 7% of the vote----even though their percentage participation has been growing steadily for the last decade. (Go, look at the numbers). The curve suggests that with steady population growth and participation, Latinos should make up 9% of the votes cast.

Blacks are also planning to stay home, according to Pew, with their 12% in 2008 dropping to 11% this year. Why? Are they really so disgusted by the abolition of Don’t Ask Don’t Tell, that they plan to send Obama a strong message by allowing a Mormon to become president? I think we all know about the Church of Jesus Christ of Latter Day Saints’ problems with Black folks. Whom do African-Americans fear more? Gays in the military or a member of a Church who was brought up to believe that they are a cursed race?

“Other Non-Hispanic”---a new group for Pew which replaces Asian in 2008 jumps from 2.5 to 5%. Who are the “Other Non-Hispanics”? Asians? Native Americans? People of mixed race (Latino plus Black, Black plus white)? Wish I knew. If we just lump them in with minority voters, then Pew is projecting that 23% of the votes cast will be from non-European-Americans—a slight increase from 2008. If the increase is made up of people who just wouldn’t say, or maybe whites who are proud that their great grandmother was Cherokee, then Pew is guessing that minorities are going to stay home---or else have their votes challenged in such great number by poll workers that their population gains will mean nothing.

Now, if you pay attention to population demographics, you are probably wondering “What about that surge in Latinos that we keep reading about?” According to the U.S. Census, the number of Hispanics in the U.S grew from 35 million to 50 million between 2000 and 2010. In 2008, those voters went 2 out of 3 for Obama.

http://www.huffingtonpost.com/2011/05/31/hispanic-population-rising-faster-than-anticipated_n_869209.html

Polls this year suggest that Latinos still favor Democrats by a margin of 2 to 1. Too bad for Obama that Pew thinks they are sitting this one out.

Two point five percent here and two percent there and one percent way over there really adds up when you are talking about a five point lead. Maybe Pew needs to repeat their poll and this time make sure that the ethnic demographics take into count the trend towards rising minority voter participation. Unless they are convinced that the GOP’s suppress the vote campaign is going to be massively successful.

Addendum: I don't want to give the impression that Pew "fixed" their poll. However, I wonder if they were so excited at finding any likely voters when they sampled cell phone users that they got sloppy on their other demographics. It is so much easier to find people on landlines.

[Ghost of Tom Joad]

1. thanks for the summary

I was beginning to wonder if all the complaints from the right about previous polls were starting to have an effect on pollsters.

[Dawson Leery]

2. This poll should not be taken seriously based upon the exclusion of Latinos alone.

[DemocratSinceBirth]

3. According To 2008 Exit Polls , Latinos Comprised 9% Of The Electorate

http://edition.cnn.com/ELECTION/2008/results/polls/#USP00p1


I think out of the 1,116 respondents in Pew's likely voter model only 57 were Latino.

If people of color come out in the same numbers as they did in 08 Romney will likely lose. The Obama campaign thinks it can win with 38% of the white voter. They received 43% of the white vote in 2008. Recent Democratic presidential losers like Kerry, Gore, and Dukakis all received greater than forty percent of the white vote.

[cthulu2016]

6. That 57 is an unweighted number

It is not desirable to have a tiny sample that is weighted up to the proper demographic level because it increases the potential for error by magnifying the effect of individual responses.

But if the Pew Latino sample was 57 (5% of the total) that doesn't mean those responses counted for only 5% of the poll. They are weighted up to match the demographic model.

The unweighted Pew poll sample is 56% female, but nobody mentions that as a flaw. (?) But the female opinion didn't count for 56% of the reported results. It is weighted down to 53%. (Or whatever number Pew uses)

It is possible for a pollster to have a bad demographic model. If Pew had a notion that the electorate is 60% female it would cause nonsensical weighted results.

But the reported unweighted sample sizes do not indicate how much the groups counted in the poll.

[cthulu2016]

4. "Pew thinks..." No, Pew doesn't "think" that

Pew reports that in their random interviews with black and hispanic voters the black and hispanic voters are less likely to vote this year.

The premise of the OP is false.

[DemocratSinceBirth]

5. How Likely Is It That Latinos Will Real Comprise 45% Less Of Their Share Of This Electorate

Than they did in 08?

[cthulu2016]

8. That requires substituting my assumptions for data

I can look at a poll result and say, "I do not think that will happen." Perhaps latino voters get more excited about elections at the last minute.

I am not saying that the result is certain to match voting behavior a month from now.

I am, however, saying that if more Latinos, for whatever reason, are telling pollsters they are less than 100% certain to vote then the pollster cannot arbitrarily count them as likely voters.

Polls are measurements, not predictions.

Imagine if latinos this year were telling pollsters they are going to vote 100%.

Could the pollster reduce the latino weighting in likely voters because the pollster didn't buy the result because that was a lot higher than they actually voted in 2008?

[McCamy Taylor]

7. Dude, part of the art of conducting surveys is knowing how to do a statistical sample.

If Pew decides that a sample is representative of likely voters with only 7% Latinos, then someone at Pew decided that Latinos either will not turn out to vote the way they have before or they will not have their votes counted.

Reputable pollsters (and researchers) look at their data to make sure that it makes sense---unless they have an agenda or a deadline. I predict that Pew's credibility is going to suffer on this one.

[cthulu2016]

9. Dude, read replies #3 and #5

You are not mistaking how polling works, on a couple of levels.

We cannot deduce weighting from unweighted sample sizes. (which is all you or I know about the Pew poll.)

Likely voters are based on what the person being polled says, not adjusted to a house model of the likely electorate. Likeliness is not weighted. Age, sex, region, gender, race, cell phones and other broad demographics are weighted in the initial sample.

Things that are changeable are not weighted. Party ID, ideology, likelihood to vote.

Since the raw sample size (unweighted) is reported for both registered and likely voters the initial demographic weighting persists.

Say the initial sample of black voters was 112. (It happened to be 116) 112 is 10% of the total sample of 1118. So the answers given by self-identified black voters are counted at 130% to equal the demographic model of 13%.

Some of those 112 people make it to "likely voter." Say 82 of the 112 are liklies. So the unweighted sample size for black likely voters is reported as 82. But that 82 people are given the weight of 106.6 people.

These little grey numbers are unweighted sample, not how much the groups were counted in the total. Notice that women are 56% of the sample. It should be 53%, so why isn't anyone freaking about that obvious Obama bias? Does Pew (or anyone) think the female electorate will be 56% this year? No. That is an unweighted number.