27 October 2012

DISCONNECTS - THREE OF THEM




As a guy who has been watching and studying the 2012 polling numbers for the last two years (yes, there was also considerable polling for 2012 in 2011), it is becoming to apparent to me that there are a number of DISCONNECTS going on that need some attention.


I. National Polling vs. State Polling

First, and probably most importantly, the incredible disparity in national polling numbers, ranging from Romney +7 at one point (Gallup) to Obama +6 on the same day (Reuters) should tell any reasonable person that a 13 point disparity in national polling cannot exist in the same universe at the same time. That is simply too wide a disparity. This means quite clearly that among the national trackers, if this trend continues to election day, some trackers are going to have a lot of egg on their face after the election.

To be honest, it is hard to imagine any polling firm wanting to do national numbers like this. Imagine: our Union currently numbers around 311,000,000 people, a good 210,000,000 of which must be registered voters. How can I write this with certainty? Well, in 2008, 131,000,000 ballots were cast for president and the voter turnout was 61%, which means that there were 214,000,000 registered voters in 2008. Our Union has grown since then, so it is logical to assume that there are probably more than 210,000,000 RV, but let's just stick with 210,000,000 to be safe.

A standard national poll uses between 1,000 and 1,500 likely voters in its model for calculation. Gallup uses 2,700, but spread over 7 days. PPP uses 1,200 per day. Let's take the highest possible side and say that every national poll were to have, say, 2,000 LV in the group, double the norm.

2,000 / 210,000,000 = 0.000952%, or just shy of 1/10,000th of a percent! This means that each likely voter, assuming a massively large pool of 2,000 LV, would stand for around 100,000 voters. There is absolutely no way to accurately gauge the nation based on national polls. Now, in elections that aren't even close, usually the polls show a rough sketch that is pretty close to reality.  But in a close election, national polls can go all over the place.

Let's take 2008 as an example. Here is how the final national polls for 2008 looked:


Pollster
Date
Obama
McCain
Margin
Gallup FINAL
11/04
53
42
+11
11/04
52
46
+6
11/04
51
46
+5
11/04
54
43
+11
GWU/Battleground
11/04
49
44
+5
Marist
11/04
52
43
+9
Harris
11/04
52
44
+8
ARG
11/04
53
45
+8





11/03
50
45
+5
FOX
11/03
50
43
+7
CBS
11/03
51
42
+9
IPSOS
11/03
53
46
+7
IBD/TIPP
11/03
48
43
+5
Pew
11/03
52
46
+6
NBC/WSJ
11/03
51
43
+8
ABC/WAPO
11/03
53
44
+9
DEMCORPS
11/03
51
44
+7





Average:

52.00
44.13
+7.87
Average 2:

51.47
44.06
+7.41
Average 3:

51.47
44.13
+7.34
MEAN:

51.65
44.11
+7.54
-GE 2004



+2.46
ADJ. MEAN



+10.00


Pretty instructive, eh? Looks nothing nothing like 2012. On the last two days of the campaign, the highest poll value was Obama +11 (Gallup and Zogby), the lowest poll value was Obama +5 (KOS, GWU Battleground, IBD/Tipp). That is a six point disparity, not a 13 point disparity as we have been seeing in the national polling from 2012.

But about half of this national polling was based on RV calculations, not just LV calculations.

Inn 2008, I took three averages from the National Polls:

"Average 1 = only the polls from today
Average 2 = All polls (without repeaters) from the last 3 days (maximum 10 polls allowed)
Average 3 = average without the highest and the lowest individual margin.
Mean = the average of these three averages."

Notice that my mean value from these three averages was Obama +7.54, which was my prediction on election day. Obama won by +7.26, so the prediction was pretty accurate. Notice what average three does: it throws out the highest and the lowest "outlier". Notice how well these three averages worked. Notice how average 3 alone came the very closest to Obama's actual winning margin in 2008. 

But that was not necessarily a close election, so this stuff is easier to be unified about. I may start taking the average of averages pretty soon, maybe not.


II. The Electoral College Rat-Race

Second, it is not about a national race, anyway, but rather, 51 individual presidential elections designed to designate Electors to the Electoral College, which then meets on January 6th, 2013 (well, probably January 7th this time around, as the 6th falls on a Sunday). So, the composite averages of the state polling give us a much clearer picture in a close election. And the composite polling still shows Obama closer to 270 than Mitt Romney. In fact, more of the time, Obama has been clearly over 270.

That being said, is there a correspondence between the closeness of a race in the NPV and the electoral college spread? Not really. Here are the 13 closest elections in our history since the inclusion of the GOP in the electoral college in 1856, by ascending order of margin:



Year
DEM %
GOP %
Other %
Margin %
DEM EC / %
GOP EC / %
EC Margin / Margin %
NPV winner
EC winner
1880
48.22%
48.31%
3.47%
GOP +0.10%
155 / 42.0%
214 / 58.0%
GOP +59 / GOP +16.0%
Garfield
Garfield
1960
49.72%
49.55%
0.74%
DEM +0.16%
303 / 56.4%
219 / 40.8%
DEM + 84 / DEM +15.6%
Kennedy
Kennedy
2000
48.38%
47.87%
4.25%
DEM +0.52%
266 / 49.4%
271 / 50.4%
GOP +5 / GOP +1.0%
Gore
Bush, Jr.
1884
48.84%
48.25%
2.87%
DEM +0.57%
219 / 54.6%
182 / 45.4%
DEM + 37 / DEM +9.2%
Cleveland
Cleveland
1968
42.72%
43.42%
13.86%
GOP +0.70%
191 / 35.5%
301 / 55.9%
GOP +59 / GOP +20.4%
Nixon
Nixon
1888
48.63%
47.80%
3.57%
DEM +0.83%
168 / 41.9%
233 / 58.1%
GOP +5 / GOP +16.2%
Cleveland
Harrison
1976
50.08%
48.02%
1.90%
DEM +2.06%
297 / 55.2%
240 / 44.6%
DEM + 57 / DEM +10.6%
Carter
Carter
2004
48.27%
50.73%
1.00%
GOP +2.46%
252 / 46.7%
286 / 53.2%
GOP +34 / GOP +6.5%
Bush, Jr.
Bush, Jr.
1876
50.92%
47.92%
1.16%
DEM +3.00%
184 / 49.9%
185 / 50.1%
GOP +1 / GOP +0.2%
Tilden
Hayes
1892
46.02%
43.01%
10.97%
DEM +3.01%
277 / 62.4%
145 / 34.7%
DEM + 132/ DEM +27.7%
Cleveland
Cleveland
1916
49.24%
46.12%
4.64%
DEM +3.12%
277 / 52.2%
254 / 47.8%
DEM + 23 / DEM +27.7%
Wilson
Wilson
1896
46.71%
51.02%
2.27%
GOP +4.31%
176 / 39.4%
271 / 60.6%
GOP +95 / GOP +21.4%
McKinley
McKinley
1948
49.55%
45.07%
5.38%
DEM +4.48%
303 / 57.1%
189 / 35.6%
DEM + 114 / DEM +21.5%
Truman
Truman


Notice that by the two closest elections of all time (1880, 1960) by popular vote margin, the EV spread was actually pretty wide. Notice also that we have had an electoral "backfire" even with one candidate winning by a +3.00% margin, and yet, losing in the EC (1876).  

So, it is entirely possible for either Obama or Romney to win the popular vote by maybe as much as 2-3%, but the "other guy" could win in the Electoral College, or worse yet, there could be an absolute tie in the EC, which could become a real mess.


III. The Sleeping Bear. Awake, or not?

Third, I still contend that there is one demographic that is neither being polled in correct numbers, nor it is being calculated correctly: the LATINO vote.

Latino Decisions, CNN and Telemundo have all put out independent-from-each-other pollling of Latinos in the USA and come up with pretty much identical margins: Obama is winning the Latino Vote by a crushing +50% over Romney, 75%-25%.

Just the most recent SUSA poll of New York, from 10/26, shows Obama taking 80% of the Latino Vote in that state, which is in line with the national Latino Vote polling.

But Gallup is predicting just a +10 in the Latino Vote for Obama. Those two stats cannot jive with each other. And I would bet good money that Gallup is way off on this one.

In the next days, I am going to start to publish all Latino Vote details from every national poll I can find up till election day.

But it is becoming apparent to me that pollsters are making the very same mistake they made in 2008 and 2010 - namely, they are vastly undercalculating the Latino Vote. This is why Obama vastly outperformed his end polling in a number of states in 2008, i.e. CO, NM and NV. And this explains why most of the pollsters really missed the three critical marquee Senatorial races in 2010: CO, NV, WA. Hmmmm, you would think that polling organizations would learn from their mistakes, but apparently not.

This topic is going to be covered in a number of very detailed threads starting on Sunday and going till election day.

No comments:

Post a Comment

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