28 September 2012

A funny thing happened in IOWA polling....with respect to NEW HAMPSHIRE

...and I think this can be a great learning moment for us all.

First, here is the entire polling for IOWA for 2012:

Nr.IOWADateSampleMoEObamaRomneyUnd.Mar.

AVERAGEN/AN/AN/A48,3344,337,334,00









14Voter/Consumer Research (R)27.09.12500 RV+/-4.446477-1
13PPP (D)26.09.12754 LV+/-3.6514457
12ARG24.09.12600 LV+/-4.0514457
11NBC / WSJ / Marist20.09.12898 LV+/-3.3504288
10YouGov20.09.12766 RV+/-4.24842106
9Rasmussen20.09.12500 LV+/-4.544479-3
8PPP (D)28.08.121244 LV+/-2.8474582
7Rasmussen10.08.12500 LV+/-4.5444610-2
6PPP (D)17.07.121131 LV+-2.91484395
5WAA (R)20.06.121086 LV+/-3.04544111
4Rasmussen12.06.12500 LV+/-4.546477-1
3Marist / Tables31.05.121106 RV+/-3.04444120
2PPP (D)08.05.121181 LV+/-2.95141810
1Selzer / Register18.02.12800 A+/-3.5444610-2


We have an interesting situation that has developed. Of 14 polls within the entire year for IOWA, Obama has won 8, Romney has won 4 and there was one absolute tie.  Of the four polls that showed Romney ahead, 3 have been from Republican or Republican-leaning pollsters and one was from the "Gold Standard" for this state: Selzer. But the Selzer poll is very "cold coffee"; it was from 02/18/2012. I really hope that Selzer will put out a new poll soon. That would be helpful.

However, post conventions, there have been 6 polls of Iowa in the last 8 days. This makes for a pretty healthy "gene pool" of polling values to work with. 4 of those polls show Obama ahead by between +6 and +8. Those are good, solid leads. Two polls show Romney ahead: the RAS 09/20 has Romney up by +3 (the best showing for Romney in any poll of either 2012 or 2011), the other was commissioned by the IOWA state Republican Party and conducted by a Republican firm out of Texas called "Voter / Consumer Research". I have never before, in eight years, ever seen a presidential head-to-head polling value from this firm, which makes it very, very new on the scene. The TIR poll has Romney up by +1. That makes a polling disparity of 7 points compared to PPP (D)'s poll and a polling disparity of 8 points compared to the Marist poll from 09/20. That is really too wide a disparity for a state where the polling average has often been very, very good. Not even to mention the 11 point disparity between RAS and Marist.

So, I read the TIR article (TIR stands for "The Iowa Republican"), and saw the following:

"The major difference between the polls could be attributed to the political make up of the sample.  Public Policy Polling’s Iowa sample was comprised of 37 percent-identified Democrats, 33 percent of identified Republicans, and 27 percent of identified independent or no-party voters.  The TIR/VCR poll was comprised of 36 percent registered Republicans, 35 percent registered Democrats, and 30 percent no-party voters.

Dr. Jan van Lohuizen, the founder and president the Houston, Texas, based polling firm Voter/Consumer Research, noted the recent debate on the make up of recent poll samples.  “There is an on going public debate over media pollsters not weighting their data by party, thus producing samples that are too Democratic,” van Lohuizen said.   “The TIR/VCR poll is weighted by party, but Republicans are weighted down by over a couple of points.”"

The problem with all three boldeds is that they do not match up with the facts at all.

Let's go through this one step at a time. 

First, according to the PPP Poll, the breakdown according to the answers the respondents gave was:

D 37 / R 33 / I 30, not I 27 as the TIR article states. You can find this information at question 13 of of the PDF. This still makes the partisan breakdown D+4, which is similar to as it was at the GE of 2008, but not how it is now.

Here are the EXACT Iowa Voter Registration statistics from September 2012, compared to September 2008, the GE of 2008 and ALSO to the GE of 2004:


State VRDEMGOPINDNO AFFTotalDEM %GOP %IND %NO AFF %MarginMar %State VRGrowth rate
IA 2012-09-0046751666691232242773662212019331,84%31,56%0,11%36,49%60430,29%IA 2012-09-0041,16%
IA 2008-09-002718083619069365758314209583134,26%29,54%0,02%36,18%990144,72%IA 2008-09-002
Diff:-429175005418771534824362-2,42%2,02%0,09%0,31%-92971-4,44%Diff:














IA 2012-09-0046751666691232242773662212019331,84%31,56%0,11%36,49%60430,29%IA 2012-09-004-1,09%
IA GE-2008736244624830974781617214366534,35%29,15%0,05%36,46%1114145,20%IA GE-2008
Diff:-61078442931268-7955-23472-2,50%2,41%0,06%0,03%-105371-4,91%Diff:














IA 2012-09-0046751666691232242773662212019331,84%31,56%0,11%36,49%60430,29%IA 2012-09-004-0,43%
IA GE-200462187362978581877662212940129,20%29,58%0,00%41,22%79120,37%IA GE-2004
Diff:53293393382161-104000-92082,64%1,98%0,10%-4,73%-18690,66%Diff:


We can clearly see that in 2008, shortly before the GE, the Democratic Party had a registration edge of +5.20% over the Republican party, so actually, PPP is weighting is one point less than the VR was in 2008. This is one way to make a comparison. But in 2012, the two parties are practically tied in voter registration, with the Democratic Party currently holding a statistically insignificant +0.29% edge over the Republicans.  This type of VR looks eerily similar to how it looked right before the GE of 2004, where the Republican Party had a statistically insignificant +0.37% VR edge over the Democratic Party. 

Right now, even before I go on, it should be readily apparent to every reader that therefore, the Independent vote in IA will be the vote that makes the difference.

So, if you compare PPP's current weighting of it's results to current VR stats, then the DEMS are overrepresented by 3.7 points.

However, if you take the TIR sample, which claims D 35 / R 36 / I 30, it is also off, for it is overrepresenting the R edge by almost 1 point. In no way is the R component of this TIR poll "weighted down by a couple of points", as the article claims. That is just a plain old lie that can be refuted by VR statistics immediately.

So, were we to have a poll based only on current VR stats, the breakdown SHOULD look like this:

D 32 / R 32 / I 36

Therefore, measured only against VR stats, both polls have a mathematical bias: PPP's toward the DEMS, TIR's toward the GOP. That still does not completely explain a 7 point margin variance in a state where the two major parties are currently absolutely neck and neck in voter registration.

So, let's look at the exit polls from both 2008 and 2004. This is incredibly handy, as we can compare them to the end VR statistics and see how things did or did not match-up.

IOWA, GE 2008, EXIT POLLS:

(screenshot)



So, according to the breakdown, the actual turnout (if the exit polls were correct) in IA was:

D 34 / R 33 / I 33, margin: D +1. If that is the case, then there is a 4 point difference between VR in Iowa from the end of October 2008 and the actual turnout, and yet, Obama won IA by a near landslide +9.53%. This happened because Obama had a massive lead in the Independent vote, of +15.

Go back to 2004:

IOWA, GE 2004, EXIT POLLS:



So, according to the breakdown, the actual turnout (if the exit polls were correct) in IA in 2004 was:

D 34 / R 36 / I 30, margin = R +2. If that is the case, then with a +2 point edge for the Republicans in 2004, but Bush only winning the state by +0.67%, assuming the same voting intensity for each party along party lines, then it must mean that the Independent vote in 2004 was practically a draw between Kerry and Bush.

This is important, because TIR is assuming an "R" component that identical to the 2004 GE exit polls. But in 2004, with these stats, it was a very close race in the independent vote.

This can therfore only mean one thing: TIR is seeing a much, much larger margin for Romney under the Independent voters than either PPP, ARG, Marist, YouGov, and even WAA (a very Republican firm) is seeing for Obama.

---------------------------------------------------------------

This disparity PROVES a point I have been making all along: that all polls have a certain amount of mathematical bias within them, and for this very reason I accept ALL polls, even ones that I would consider to be outliers, and trust that the average is much closer to the truth than the individual polls themselves.

Without the TIR poll in the mix, the current margin average in IOWA would be: Obama +5.00. With the TIR poll in the average, it is, as you can see: Obama +4.00. At this point in the game, +4 looks every bit as reasonable to me as +5. And both values are outside the standard MoE. Obama is winning Iowa, plain and simple.

Consider that voter registration in IA has reversed for the Democrats by -4.91% since the GE of 2008.

Let's take Obama's margin from 2008 and subtract that difference right now:

9.51 - 4.91 = 4.60, right between +4 and +5.
IOWA vs. NEW HAMPSHIRE, GE performance since 1996:

Let's look at another factor: 

in 2008, the two states where Obama won by circa +9.5% were:

IOWA: Obama +9.53%
NEW HAMPSHIRE: Obama +9.61%
Disparity, IA to NH 2008: 0.08%

These two states also had similar voting records in 2004:

IOWA: Bush +0.67%
NEW HAMPSHIRE: Kerry +1.37%
Disparity, IA to NH 2004: 2.04%

And also in 2000:

IOWA: Gore +0.31%
NEW HAMPSHIRE: Bush +1.27%
Disparity, IA to NH 2000: 1.58%

And also in 1996:

IOWA: Clinton +10.34%
NEW HAMPSHIRE: Clinton +9.95%
Disparity, IA to NH 1996: 0.39%


Take a look at the current polling for New Hampshire on the BATTLEGROUND REPORT for 09/27

Obama +4.00, exactly the same average as for Iowa. In other words, the states with the most similar margins to each other in 2008 also have identical margin averages right now in 2012. That is what we call statistical stability.

Let's take a quick look at what has happened with NH's VR statistics in the last 4 years:

State VRDEMGOPINDNO AFFTotalDEM %GOP %IND %NO AFF %MarginMar %State VRGrowth rate
NH 2012-01-010226720259968030456179124928,65%32,86%0,00%38,49%332484,20%NH 2012-01-010-10,86%
NH 2008-01-008258776271220035768888768429,15%30,55%0,00%40,29%124441,40%NH 2008-01-008
Diff:-32056-112520-53127-96435-0,50%2,30%0,00%-1,80%-208042,80%Diff:














NH 2012-01-010226720259968030456179124928,65%32,86%0,00%38,49%332484,20%NH 2012-01-010-17,45%
NH GE-2008282421280507039560095852829,46%29,26%0,00%41,27%19140,20%NH GE-2008
Diff:-55701-205390-91039-167279-0,81%3,59%0,00%-2,78%-351624,00%Diff:


Oh, look, the Democratic Party has also had slippage in the VR stats for NH, very similar to Iowa. A major difference here is that at the current time, overall registration in NH has shrunk by 11% over two years (sorry, these are the most recent VR stats for NH, they are stingy...)

FACIT: I can live with accepting all poll values, even if they appear to be outliers. And remember, other polls of Iowa are sure to follow, and they will either confirm or negative the poll put out by the Republican Party of Iowa.

And the next time people want to poo-poo all over polls, just remember this thread....

No comments:

Post a Comment

Constructive comments and critique are always welcome. Please keep it polite and respectful.