The over-sampling of Democrats in today’s polls most likely hides a sharp decline in support for President Obama among White voters. If President Obama’s support level among White voters dips a single percent or two below 40, his road to re-election would be in jeopardy. The national polling results today showing President Obama with support levels among Whites between 40-44% likely over-sample support for President Obama by 4% to 8% among this demographic. Poll re-weighting by race achieves an accurate demographic make-up for the United States in 2012 but almost certainly a wholly unrealistic split between self-identified Democrats and Republicans. Because the accurate re-weighting of polls by race often achieves political splits that are not credible, polling organizations give rise to accusations of bias when in reality better selected sample inputs would most likely achieve more credible end results but also meaningfully worse results for the President.
Problems with polls
The majority of polling critiques this election cycle focus almost exclusively on the amount of Democrats versus Republicans surveyed with the observation invariably there are far too many Democrats in the sample. There is much in dispute around this complaint because most polling organizations do not weight polls by the party identification of respondents. Polling organizations argue the disproportionately high amount of Democrats sampled draws a sharp inference there are more Democrats in the overall electorate, not just in the sample size. While it is possible and even probable there are a few more self-identified Democrats in the American electorate (the average in elections since 1984 is 3% more Democrats), the great dispute is the unusually large disparity of Democrats showing up in today’s polls, often as much as 7 to 12% higher than Republicans among the respondents. There are many reasons to challenge this conclusion which I will discuss later, but if we assume these polls have too many Democrats, an interesting phenomenon appears among which Democrats are oversampled.
Most polling methodologies, including how polls are weighted once responses are collected, mirror the Gallup Organization who has been the standard bearer in the US for over 75 years. According to the organization, “After Gallup collects and processes survey data, each respondent is assigned a weight so that the demographic characteristics of the total weighted sample of respondents match the latest estimates of the demographic characteristics of the adult population available from the U.S. Census Bureau. Gallup weights data to census estimates for gender, race, age, educational attainment, and region.” Based on explanations like this there is little reason to suspect intentional political bias in the disparate party weighting, especially when they do not re-weight polls by party identification. Importantly, though, they do re-weight polls by race. This gives rise to some curious issues regarding support levels for the President today.
Racial demographics and voting preference in the US
In the 2008 election, the racial breakdown of the national voting public was 74% White, 12% Black, 9% Hispanic, 2% Asian, 3% Other. When a survey is conducted polling organization re-weight the respondent answers to ensure the each of these groups has accurate representation in the final results. Most national polls reflect this reality usually within a 1% variation for any group.
If you look how each one of these groups vote, you find outsized rates of support for Obama among the non-White groups: Blacks ~95%, Hispanics ~70, Asians ~65%. These levels of support for Democrats are consistent with most modern elections although President Obama has been able to boost these levels slightly above historic averages. In aggregate, non-White support for Obama is roughly 80% in nearly every survey. At the same time Obama, like Democrat Presidential candidates before him, struggles with the White vote. In these same polls, Obama typically averages 40-44% support among White voters. His 2008 support level was 43% and it is widely believed by the Obama campaign among others that he needs support of at least 40% Whites to win the election.
Low potential for over-sampling non-White support for Obama
If we consider the idea that polling today has large over-samples of Democrats, the consistently high percentage of support for Obama among non-Whites makes it almost impossible to over-sample minority groups. First there is not a lot of room for support increases and second, data on the voting trends in non-White groups is often achieved through demographic specific polling of solely Blacks or Hispanics for example. Hence, any over-samples in the non-White demographic would meaningfully alter the already high levels of support for Obama and reveal itself as inconsistent with independent polling. Additionally, any over-samples in the non-White demographic would almost certainly change the racial make-up of the survey and set off red-flags to anyone scrutinizing polls. Therefore it is highly unlikely over-sampled Democrat polls contain an excess amount of non-White voters.
This leaves only White Democrats as the over-represented respondent in these polls that arguably over-sample Democrats. If the average in election turnout since 1984 is 3% more Democrats and these polls have 7 to 11% more Democrats, that means the polls specifically have 4 to 8% more White Democrats surveyed in their likely voter results. The problem for the Obama campaign is if his support level among White voters (74% of the voting public) is between 40%-44% and that support is based on a sampling that over-states his support 4 to 8%, his real level of support is probably closer to 36% or 37%. This is meaningfully below the campaign’s own magic level of 40% and is a huge danger zone for any Presidential candidate no matter how much anyone may spin the demographic changes in today’s America.
Hiding the decline
The issue with the suspect polling internals and media embrace of the figures is the consistent lead for Obama would be immediately challenged if his support levels dropped dramatically among the outlined racial groups. Support levels of 60% among Hispanics (9% of the voting public) or 80% among Blacks (12% of the public) would jump off the page to poll watchers. The same holds true for support levels of 36/37% among Whites (74% of the voting public). It would be near impossible for Obama to win the Presidency with support levels like the ones I just outlined. Unfortunately support for President Obama among White voters has declined from 43% in 2008 to apparently as low as 36%-37% in today’s polls absent unrealistically high levels of self-identified Democrats. With White voters making up 73-74% of the electorate and support levels in the upper 30s, it is inconceivable President Obama has the advantage these polls lead readers to believe. But the results largely go unchallenged in the media despite the impractical internal party identification make-up.
Polling bias and Party identification
When we reflect on accusations of bias in polling based on party identification, it seems hard to justify when most organizations do not adjust their polls based on this metric. These organizations do, however, run the risk of confirmation bias where the media and polling firms have a predilection towards one candidate and upon achieving results they agree with fail to challenge outlier data like unrealistic Democrat turnout levels in 2012. Inconvenient poll compositions like the fantastical party identification of respondents shake the credibility of desired outcomes but no explanation is given for such oddities. This leaves more fair-minded poll watchers uneasy with the factual reporting on data with obvious internal issues while partisans react more strongly with bias accusations not substantiated based on the available data. The over-sampling of Democrats may not be showing the bias of polling organizations but it is likely hiding the decline of dwindling White support for Obama.
This only raises the question of where the polling firms are getting their samples from — possibly heavy Democrat districts — because the end results are party identification breakdowns unrealistic in today’s electorate. In 2008 seven percent more Democrats than Republicans identified themselves as such on election day, well above the historic average of 3%. This was a big change from 2004 when party identification was evenly split between the Democrats and Republicans. But there were many reasons for the strong Democrat turnout that do not exist today. The top of the ticket was a historic candidate (first Black President), America had war and Bush fatigue, the financial meltdown created an anti-Republican wave, and his opponent wasn’t the strongest (good biography, bad and underfunded candidate). These factors led to a strong Democrat self-identification advantage at the voting booth in 2008. But in the 2012 election, none of the advantages outlined above are there for Obama and many of those factors are now largely working against the President: 8%+ unemployment for three years, sub-2% GDP, 23 million unemployed, Arab Spring blowing up and casting the historic vote in 2008 is yesterday’s news. Additionally the Romney campaign ground game has exceeded the McCain campaign across many metrics as much as 10- to 15-fold.
Despite the stark changes in each of these factors, polling outfits thus far have consistently sampled an election turnout often greater than candidate Obama’s 2008 best-in-a-generation advantage.
That means something else is going on. But the polling organizations shrug their shoulders and have been found to say the losers in the results are just crying sour grapes. This is even though their sample outcomes have party identification splits unrealistic beyond any stretch of reason. Sadly no credible defense is given for the unusual party split in these results which gives rise to charges of bias whether intentional or accidental. If the polling firms believe today’s electorate will exceed the incredible 2008 advantage Obama achieved they should make the argument to justify results that contain suspect internal data. But they would also have to explain why the 2008 election gave Democrats massive majorities in the House of Representatives yet today’s electorate will likely return massive majorities in the House to Republicans. It defies all logic. But very likely due to “confirmation bias” the media and polling organizations report favorable results for President Obama without challenge.
There are many explanations for odd internal data in polls as well as the built in accuracy issues that come with the very nature of polling. As Michael Barone writes, “it’s getting much harder for pollsters to get people to respond to interviews. The Pew Research Center reports that it’s getting only 9 percent of the people it contacts to respond to its questions. That’s compared with 36 percent in 1997.” But consistently unrealistic sample outputs give rise to greater scrutiny from the polling outfits and media organizations who report the results uncritically for whatever their reasons may be.