PHOENIX (Jon Kamman, Arizona Republic) August 20, 2005 - Arizona's turnout in last year's presidential election was an incredible 90.1 percent of registered voters, the Census Bureau now reports.

Incredible is right. Figures released today are far from reality.

That's because they are based on a survey, rather than an actual count.

Official turnout, as counted by Arizona elections officials, was 77.1 percent, or 13 points below the survey figure.

Polling and election experts say the wide difference underscores the danger of attaching too much validity to surveys, which can have weaknesses ranging from too small a sampling to outright lying by participants.

At the same time, numbers crunchers say the census survey provides, at least on a national level, the best data available for tracking demographic trends and voting behavior from election to election.

"It's so incredibly difficult to find numbers you can rely on in this field because voting is secret," said Dan Seligson, spokesman for electionline.org, a nonpartisan organization in Washington, D.C., that researches and analyzes how elections are conducted across the country.

Nationally, the survey showed a 4.3 percentage point increase from 2000 in balloting among citizens who were eligible to register. Men, women, Whites, Blacks, Asians and Hispanics all posted gains, bringing the turnout to just under 64 percent of eligible adults, according to the report.

Asians had the lowest turnout, at 44 percent, followed closely by Hispanics, at 47 percent. The other racial groups all scored at least 60 percent.

Viewed in terms of how many registered voters actually cast ballots, the survey showed a nationwide turnout of 88.5 percent, an increase of 3 points from 2000.

On the state level, however, the results proved unreliable.

The survey underestimated the number of registered voters by nearly 160,000, or more than 6 percent. It overestimated the number of people who actually cast ballots by more than 200,000, or about 10 percent.

With errors of that magnitude, figures purporting to show turnout among such smaller groups as Hispanics, veterans or college graduates are subject to even higher margins of error.

Although a survey by its nature is "never accurate," the voting study is worth doing, said Curtis Gans, director of the Washington, D.C.-based Committee for the Study of the American Electorate.

"It's the only good source for trend lines," said Gans, whose analyses of elections over nearly 30 years have made him one of the nation's top authorities on voting.

Census Bureau analyst Kelly Holder said a variety of factors can introduce flaws into the survey, which attempts to reach more than 100,000 people.

State registration figures often are higher than those calculated in the survey because the states are not immediately removing the names of people who have died or moved away, she said.

The only nose-by-nose count in the whole process is the actual census, conducted five years ago. After that, the bureau releases annual estimates of population and its racial and ethnic makeup, creating one margin for error.

That inaccuracy can be compounded when the bureau then estimates the number of people 18 and older.

Moreover, respondents are not always accurate, even if they mean to be truthful, in answering the survey, she said.

Seligson said his group's reliance on the survey data never has been challenged.\