Wednesday, November 10, 2004

Exit Poll Findings That I Don't Believe

There's a lot of dodgy data in the NEP numbers. Many people are already disputing their finding of a big jump for Bush among Hispanics, although the L.A. Times poll shows something similar. And some of their state data seem weird -- a 17-point jump for Bush among Massachusetts Catholics -- when no town in the whole state saw any movement to Bush like that -- a 42 percent white electorate in Hawaii -- when the Census shows only 24 percent of state residents being white.

But nothing makes less sense than its finding that Bush won 39 percent -- a gain of 13 points -- in large cities (over 500,000). The 2000 census shows 29 cities with more than 500,000. None of them saw a shift to Bush of anything like that:

City 2000 Bush% 2004 Bush%

New York 18 24
Los Angeles* 32 36
Chicago 17 18
Houston* 54 55
Philadelphia 18 19
Phoenix* 53 57
San Diego* 50 52
Dallas* 53 50
San Antonio* 52 55
San Jose* 34 35
Indianapolis 49 49
San Francisco 16 15
Jacksonville 57 58
Columbus* 48 46
Austin* 47 42
Baltimore 14 17
Memphis* 42 42
Milwaukee 28 27
Boston 20 22
Washington, DC 9 9
El Paso* 40 43
Seattle* 34 34
Denver* 31 29
Nashville 40 45
Charlotte* 51 48
Fort Worth* 61 62
Portland, OR* 28 27
Oklahoma City* 62 64

* -- county data

No city appears to have seen a double-digit Bush gain. The only cities where Bush improved by even five points were Nashville (home of his 2000 opponent) and New York City, site of the GOP convention and of 9/11. In NYC, Bush appears to have gained among Catholics -- especially on Staten Island, hit so cruelly on 9/11 -- and especially among Orthodox Jews, who voted about 60% for Gore-Lieberman in 2000 and about 70% for Bush this year. His NYC numbers are not that spectactular -- Bush won the same percentage that his father got in a three-way race in 1992.

Unless there was some huge shift in turnout to Sun Belt cities, the NEP numbers don't make sense.


Anonymous said...

In terms of the data you supply, don't you think use of county data makes it problematic? I mean, places like Los Angeles County, San Diego County, Harris County (where Houston is), Maricopa County (where Phoenix is - this one is especially problematic) are enormous and full of suburbs. So this could easily mask any changes in the city itself. I suspect your basic conclusion is correct, but I think a tighter analysis is necessary to prove it.

Anonymous said...

I was also skeptical that Kerry really lost a lot of votes in cities as the exit poll analysts claim he did. So I went to the trouble of putting a spread sheet together with the data from 2000 and 2004 compared.

I ran into the same problem as you did that I could only find county results for some of the cities. However, since the exit polls claim that suburban voters ALSO went more strongly for Bush than in 2000, this should hardly skew the data in Kerry's direction (unless the exit poll is complete junk).

I compared the data for the 31 cities with a population of more than 500,000 in 2004. Two of these cities (Tucson and Las Vegas) had just under 500,000 people in 2000. In both cities Gore and Kerry won by a slightly smaller margin than on average. Including these two cities in the comparison with 2000 should not influence the results significantly.

The main result is this: the 2004 exit poll is basically correct, but the comparison data from 2000 used in the newspaper articles is simply wrong. Gore did not get 70 % of the big city vote in 2000. He only got about 59 %. In 2004, Kerry received 60.3 % of the big city vote, a small INCREASE, not a decrease.

Bush's numbers went up from 38.3 % in 2000 to 38.7 % in 2004, also a small increase.

One way to interpret the small changes is that the votes received by Nader and others were split about 3:1 between Kerry and Bush. (In addition, more people turned out to vote, but about proportionally for both of them.)

I will try to alert other blogs to this as well. It would also be nice to get newspapers to run corrections.

One possible explanation for the discrepancy: the data from 2000 might indeed be based purely on city, not county data, whereas the 2004 exit poll data is based on county results. In LA and Chicago, for example, the county results are roughly 10 % worse for Kerry than the city results.

If you are interested in the spread sheet I put together, you can email me at gw0144 at