FarmerJohn
Has No Life - Lives on TB
By Harry Enten FiveThirtyEight
AUG 9, 2016
We’ve reached that stage of the campaign. The back-to-school commercials are on the air, and the “unskewing” of polls has begun — the quadrennial exercise in which partisans simply adjust the polls to get results more to their liking, usually with a thin sheen of math-y words to make it all sound like rigorous analysis instead of magical thinking.
If any of this sounds familiar — and if I sound a little exasperated — it’s probably because we went through this four years ago. Remember UnSkewedPolls.com? (The website is defunct, but you can view an archived picture of it here.) The main contention of that site and others like it was that the polls had too many Democratic respondents in their samples. Dean Chambers, who ran the site, regularly wrote that the polls were vastly undercounting independents and should have used a higher proportion of Republicans in their samples. But in the end, the polls underestimated President Obama’s margin.
Now the unskewers are back, again insisting that pollsters are “using” more Democrats than they should, and that the percentage of Democrats and Republicans should be equal, or that there should be more Republicans. They point to surveys like the recent one from ABC News and The Washington Post, in which 33 percent of registered voters identified as Democrats compared to 27 percent as Republicans. That poll found Hillary Clinton ahead by 8 percentage points.
But let’s say this plainly: The polls are not “skewed.” They weren’t in 2012, and they aren’t now.
The basic premise of the unskewers is wrong. Most pollsters don’t weight their results by party self-identification, which polls get by asking a question like “generally speaking, do you usually think of yourself as a….” Party identification is an attitude, not a demographic. There isn’t some national number from the government that tells us how many Democrats and Republicans there are in the country. Some states collect party registration data, but many states do not. Moreover, party registration is not the same thing as party identification. In a state like Kentucky, for example, there are a lot more registered Democrats than registered Republicans, but more voters identified as Republican in the 2014 election exit polls.
A person’s party identification can shift, and therefore the overall balance between parties does too. Democrats have typically had an advantage in self-identification — a 4 percentage point edge in 2000, a 7-point advantage in 2008 and a 6-point edge in 2012, according to exit polls — but they had no advantage in the 2004 election. Since 1952, however, almost every presidential election has featured a Democratic advantage in party identification.
Here’s the margin that Democrats have had in self-identification since 1952, according to the American National Elections Studies and, starting in 1972, exit polls.
And it’s not crazy to think Democrats will have an advantage in party identification in 2016. With a controversial nominee, many Republicans might not want to identify with the GOP, and may be calling themselves independents.
You should also be skeptical of other attempts to reweight pollsters’ data. One website, LongRoom, claims to “unbias” the polls using “actual state voter registration data from the Secretary of State or Election Division of each state.” The website contends that almost every public poll is biased in favor of Clinton.
Think about what that means: The website is saying that a large number of professional pollsters who make their living trying to provide accurate information — and have a good record of doing so — are all deliberately biasing the polls and aren’t correcting for it. Like many conspiracy theories, that seems implausible. [emphasis added]
I’d also point out that election offices from different states collect different data. Some states don’t have party registration; other states don’t collect data on a person’s race; some states collect data on neither. There are some companies that try to fill in missing data for each state, though it costs a lot to get that data. Isn’t it more plausible the people who get paid to know what they are doing are right, while some anonymous website on the internet with unclear methodology is wrong?
Of course, unskewing is simply one of many ways of pretending Clinton hasn’t jumped out to a large post-convention lead against Donald Trump. You could also ask us to imagine a world without polls. You could allege, without any evidence, that outright election fraud will take place. Or you point to Trump’s rally sizes, though George McGovern in 1972, Walter Mondale in 1984 and Mitt Romney in 2012 all had large crowd sizes and lost.
People, though, should stick to reality. Right now, Clinton is leading in almost every single national poll. She leads in both our polls-plus and polls-only forecasts. That doesn’t mean she will win. The polls have been off before, but no one knows by how much beforehand, or in which direction they’ll miss. For all their imperfection, the polls are a far better indicator than the conspiracy theories made up to convince people that Trump is ahead.
http://fivethirtyeight.com/features/the-polls-arent-skewed-trump-really-is-losing-badly/
AUG 9, 2016
We’ve reached that stage of the campaign. The back-to-school commercials are on the air, and the “unskewing” of polls has begun — the quadrennial exercise in which partisans simply adjust the polls to get results more to their liking, usually with a thin sheen of math-y words to make it all sound like rigorous analysis instead of magical thinking.
If any of this sounds familiar — and if I sound a little exasperated — it’s probably because we went through this four years ago. Remember UnSkewedPolls.com? (The website is defunct, but you can view an archived picture of it here.) The main contention of that site and others like it was that the polls had too many Democratic respondents in their samples. Dean Chambers, who ran the site, regularly wrote that the polls were vastly undercounting independents and should have used a higher proportion of Republicans in their samples. But in the end, the polls underestimated President Obama’s margin.
Now the unskewers are back, again insisting that pollsters are “using” more Democrats than they should, and that the percentage of Democrats and Republicans should be equal, or that there should be more Republicans. They point to surveys like the recent one from ABC News and The Washington Post, in which 33 percent of registered voters identified as Democrats compared to 27 percent as Republicans. That poll found Hillary Clinton ahead by 8 percentage points.
But let’s say this plainly: The polls are not “skewed.” They weren’t in 2012, and they aren’t now.
The basic premise of the unskewers is wrong. Most pollsters don’t weight their results by party self-identification, which polls get by asking a question like “generally speaking, do you usually think of yourself as a….” Party identification is an attitude, not a demographic. There isn’t some national number from the government that tells us how many Democrats and Republicans there are in the country. Some states collect party registration data, but many states do not. Moreover, party registration is not the same thing as party identification. In a state like Kentucky, for example, there are a lot more registered Democrats than registered Republicans, but more voters identified as Republican in the 2014 election exit polls.
A person’s party identification can shift, and therefore the overall balance between parties does too. Democrats have typically had an advantage in self-identification — a 4 percentage point edge in 2000, a 7-point advantage in 2008 and a 6-point edge in 2012, according to exit polls — but they had no advantage in the 2004 election. Since 1952, however, almost every presidential election has featured a Democratic advantage in party identification.
Here’s the margin that Democrats have had in self-identification since 1952, according to the American National Elections Studies and, starting in 1972, exit polls.
And it’s not crazy to think Democrats will have an advantage in party identification in 2016. With a controversial nominee, many Republicans might not want to identify with the GOP, and may be calling themselves independents.
You should also be skeptical of other attempts to reweight pollsters’ data. One website, LongRoom, claims to “unbias” the polls using “actual state voter registration data from the Secretary of State or Election Division of each state.” The website contends that almost every public poll is biased in favor of Clinton.
Think about what that means: The website is saying that a large number of professional pollsters who make their living trying to provide accurate information — and have a good record of doing so — are all deliberately biasing the polls and aren’t correcting for it. Like many conspiracy theories, that seems implausible. [emphasis added]
I’d also point out that election offices from different states collect different data. Some states don’t have party registration; other states don’t collect data on a person’s race; some states collect data on neither. There are some companies that try to fill in missing data for each state, though it costs a lot to get that data. Isn’t it more plausible the people who get paid to know what they are doing are right, while some anonymous website on the internet with unclear methodology is wrong?
Of course, unskewing is simply one of many ways of pretending Clinton hasn’t jumped out to a large post-convention lead against Donald Trump. You could also ask us to imagine a world without polls. You could allege, without any evidence, that outright election fraud will take place. Or you point to Trump’s rally sizes, though George McGovern in 1972, Walter Mondale in 1984 and Mitt Romney in 2012 all had large crowd sizes and lost.
People, though, should stick to reality. Right now, Clinton is leading in almost every single national poll. She leads in both our polls-plus and polls-only forecasts. That doesn’t mean she will win. The polls have been off before, but no one knows by how much beforehand, or in which direction they’ll miss. For all their imperfection, the polls are a far better indicator than the conspiracy theories made up to convince people that Trump is ahead.
http://fivethirtyeight.com/features/the-polls-arent-skewed-trump-really-is-losing-badly/
