National polls have been relatively convergent for a month, but state polls are all over the place. There have been a couple of very nice articles about this, and I want to add my two cents on this issue.
When looked at by polling agent and result there are only two poll results. One group shows a Romney lead of 2-6 points, the other an Obama lead of about the same magnitude. The average of these two spikes shows a tight race, but in fact the polls show two completely divergent results.
The reason is the sample mix used. The polls favoring Obama are using a voter mix very much like 2008. This mix has an electorate with Democrats outnumbering Republicans by 10 points, and Independents leaning Democratic. This sample assumes a democratic turnout as high or even higher than 2008 (itself a record), and an increased participation by Hispanics, the young and Blacks. The polls favoring Romney assume an electorate closer to 2004 than 2008. This sample has a democratic edge of 3-4 points and independents split closer to Republicans than to Democrats.
The problem witha veraging these two very different samples is that they both can't be correct. It is entirely wrong to combine the two halves of a bimodally dsitributed population and then claim to understand 'the average'. One of these polling samples is going to be wrong, so we must look at what evidence there is to support either sample.
This is generally bad news for Obama. Gallup's survery of voter preferences found a distribution that was actually +3 Republican. Early voting totals indicate the GOP picking up 300-500,000 votes off early voting alone. The internal results of polls are similar. When independent voters are asked their preference Romney has a lead of at least 11 points.
The 'big leads' for Obama have been obtained by poll samples that are as much as +13 Democratic, an advantage never seen yet in presidential elections. Quality evaluation of the electorate suggests it will be much more like 2004 than 2008. And Obama is barely tied or trails even with these favorable polling samples. One of the hallmarks of folly is to ignore data and accept as real only that which you wish to see. The democrats are doing that right now.
In 1870 France was the acknowledged military power of Europe. France's armies had defeated one colonial group after another quickly and brutally. This was achieved despite manifest problems with supply, transport, command structure and conscription. Prussia was a military irrelevancy. Its infantry weapon was old, Prussia was small, had no military tradition in 100 years and had only fought a 'bruderkrieg' with Austria. In going to war everyone thought France would crush Prussia, and would do so rapidly.
But while no one was paying any attention Prussia had created a net of railroads to bring their armies rapidly to the front, a supply system of similar efficiency, and a rifled steel artillery that was the best in the world, and against which the French had outranged Bronze smoothbore artillery.
The result was a stunning, crushing win for Prussia. Six weeks after the war started the entire French profesisonal army was surrounded or had surrendered. French troops lay hungry in open fields under the rain while their tents and food piled up at stations because the French thought of the trains but not the wagons on the other end. Napoleon III did not give overall command to anyone but himself, and his corps commanders did not feel empowered to take independent action in the face of the flexible arrangements of the Prussian army. French soldiers were brave but hopelessly outclassed.
I believe this is what will happen to Obama on the 6th. Everyone will be surprised by a 6 point Romney win for 300 electoral votes. Silver's model is irrelevant, because it assumes the top line of the polls is right, and this year it is not. The University of Colorado model is closer. In only 5 days we will wake up and know. Until then, pay close attention to the polling sample before taking any poll seriously.
Ted
When looked at by polling agent and result there are only two poll results. One group shows a Romney lead of 2-6 points, the other an Obama lead of about the same magnitude. The average of these two spikes shows a tight race, but in fact the polls show two completely divergent results.
The reason is the sample mix used. The polls favoring Obama are using a voter mix very much like 2008. This mix has an electorate with Democrats outnumbering Republicans by 10 points, and Independents leaning Democratic. This sample assumes a democratic turnout as high or even higher than 2008 (itself a record), and an increased participation by Hispanics, the young and Blacks. The polls favoring Romney assume an electorate closer to 2004 than 2008. This sample has a democratic edge of 3-4 points and independents split closer to Republicans than to Democrats.
The problem witha veraging these two very different samples is that they both can't be correct. It is entirely wrong to combine the two halves of a bimodally dsitributed population and then claim to understand 'the average'. One of these polling samples is going to be wrong, so we must look at what evidence there is to support either sample.
This is generally bad news for Obama. Gallup's survery of voter preferences found a distribution that was actually +3 Republican. Early voting totals indicate the GOP picking up 300-500,000 votes off early voting alone. The internal results of polls are similar. When independent voters are asked their preference Romney has a lead of at least 11 points.
The 'big leads' for Obama have been obtained by poll samples that are as much as +13 Democratic, an advantage never seen yet in presidential elections. Quality evaluation of the electorate suggests it will be much more like 2004 than 2008. And Obama is barely tied or trails even with these favorable polling samples. One of the hallmarks of folly is to ignore data and accept as real only that which you wish to see. The democrats are doing that right now.
In 1870 France was the acknowledged military power of Europe. France's armies had defeated one colonial group after another quickly and brutally. This was achieved despite manifest problems with supply, transport, command structure and conscription. Prussia was a military irrelevancy. Its infantry weapon was old, Prussia was small, had no military tradition in 100 years and had only fought a 'bruderkrieg' with Austria. In going to war everyone thought France would crush Prussia, and would do so rapidly.
But while no one was paying any attention Prussia had created a net of railroads to bring their armies rapidly to the front, a supply system of similar efficiency, and a rifled steel artillery that was the best in the world, and against which the French had outranged Bronze smoothbore artillery.
The result was a stunning, crushing win for Prussia. Six weeks after the war started the entire French profesisonal army was surrounded or had surrendered. French troops lay hungry in open fields under the rain while their tents and food piled up at stations because the French thought of the trains but not the wagons on the other end. Napoleon III did not give overall command to anyone but himself, and his corps commanders did not feel empowered to take independent action in the face of the flexible arrangements of the Prussian army. French soldiers were brave but hopelessly outclassed.
I believe this is what will happen to Obama on the 6th. Everyone will be surprised by a 6 point Romney win for 300 electoral votes. Silver's model is irrelevant, because it assumes the top line of the polls is right, and this year it is not. The University of Colorado model is closer. In only 5 days we will wake up and know. Until then, pay close attention to the polling sample before taking any poll seriously.
Ted
And yet...Obama did win, getting 50.5% of the popular vote to Romney's 47.9%, 61 million votes over Romney's 58 million, and 303 electoral votes to Romney's 206. In the number of states carried, it was indeed tighter, though still ending in Obama's favor, with 25 (plus District of Columbia) to 24 for Romney (with only Florida still being indeterminate as of 3 days post-election).
ReplyDeleteSo the pre-election polls showing Obama in the lead turned out to be better guessers (since, it seems to me, this is what small political polls of a large population comes down to: guessing).
David