The flaw in the Bloomberg “Obama is going to win” poll

It always pays to go back to the description of a poll.   Pretty much everyone agrees this poll is shocking, showing President Obama with a 53% to 40% lead over Mitt Romney.

http://media.bloomberg.com/bb/avfile/rQyA68BW5P20

First thing to notice is the small sample size of a national poll

1,002 U.S. adults ages 18 and over, June 15-18, 2012
Including 734 likely voters in the Nov. 2012 election

but more importantly

“Responses were weighted by age and race to reflect the general population based on recent census data.”

so they took the “Likely Voters” answered and then amplified the numbers based on the “generation population”, NOT the characteristics of the population who vote.   It’s widely know that the turnout of young people is far less than senior citizens.

Let’s say when they did the poll, they took the group of people aged 20 to 24 – which according to the US Census is

http://www.census.gov/compendia/statab/2012/tables/12s0007.pdf

21,586,000 out of an 18+ population of 234,564,000  which is 9.2% of the voting age population.    (Note this is the voting age population, not people who are registered to vote or eligible to vote or haven’t lost the right to vote for a felony conviction)

So they had  734 likely voters – you would expect that 67 of the people in the poll would be 20-24 years old.   But let’s guess they only got 34 people who passed the “are you likely to vote” screen…  for those 34 people, their votes would count double, effectively completely undoing the “likely voter” screen.   But they took the pie and sliced it one cut thinner – they also weighted it by race.   It’s not a big secret that certain ethnic groups (especially “Hispanics”) don’t show up to vote in proportion to their numbers in the Census (maybe because they are here illegally?)

If you need another marker of just how skewed this survey and its weighting is – they asked people are they going to vote for a Republican or a Democrat for the House Race in 2012.   The House is made up of districts of roughly equal population.   Bloomberg’s “poll” found that  people were going to vote (or leaning)  48% Democrat, 41% Republican.

In 2010, Republicans won 51.4% of the vote and Democrats 44.8% of the House vote and Repulicans won 242 of 435 seats (55.6%)   [total popular vote is a pretty meaningless statistic at a national level]

I see no evidence yet to make me think voters are going to abandon Republicans in the House by 10%.

Who is Selzer & Company?   Do they have any prior experience in political polling?

Selzer is Ann Selzer – she’s done polling for the Iowa Register since 1997.

Here is a pretty complete biography:
http://businessrecord.com/main.asp?Search=1&ArticleID=14471&SectionID=16&SubSectionID=56&S=1

Here is some video from just before the 2010 election where she appears on PBS