LOS ANGELES (AP) -- Two pillars of Democratic Gov. Gray Davis' past support -- union members and Hispanics -- failed him Tuesday as those voters drifted toward Republican Arnold Schwarzenegger, a survey of voters found.

Amid extraordinary voter discontent with the state's economy and Davis' job performance, nearly half of Hispanics and a similar proportion of union members voted to recall the governor, according to an exit poll conducted for The Associated Press and other news organizations.

Less than a year ago, nearly two-thirds of Hispanic voters and voters from union households supported Davis' re-election, and 70 percent of both groups voted for Davis in his first run for governor in 1988, exit polls found.

This year's statewide survey of 4,172 voters, including Election Day interviews and a telephone survey of absentee voters during the past week, was conducted by Edison Media Research of Somerville, N.J., and Mitofsky International of New York City. Interviews were conducted in English and Spanish. The sampling error margin was plus or minus 1.5 percentage points for the full sample, larger for subgroups.

Since the special election was announced in July, labor unions had spent millions to push their members to vote no on the recall, yes on Lt. Gov. Cruz Bustamante and no on Proposition 54, the racial privacy initiative.

About four in 10 voters from union households supported Schwarzenegger in the gubernatorial replacement vote even though during the campaign he called unions "special interests" and vowed to renegotiate union contracts with state workers. About a third of the electorate were union members or had union members in their household.

Proposition 54 was defeated, but voters approved the recall of Davis. In an election sparked by the state's fiscal woes, voters were split on whether California can resolve its budget deficit without raising taxes, the poll found.

Eight in 10 voters said the state's economy was not so good or poor, and seven in 10 disapproved of how Davis was handling his job. Nearly half of all voters strongly disapproved, and among them, nine in 10 voted for the recall and seven in 10 voted for Schwarzenegger, the exit poll found.

Voters were divided on whether Tuesday's election to recall Davis, which cost about $66 million, was worth the money.

Among Election Day voters, about seven in 10 said they made up their minds more than a month ago on how to vote on the recall. The survey offered little evidence that late-breaking allegations that Schwarzenegger grouped women had much effect on how people voted.

The poll found that six in 10 of all voters valued their candidate's positions on the issues above the candidate's leadership or personal qualities.

About two-thirds of voters -- including a quarter of Schwarzenegger's own supporters -- said the movie star, who participated in only one pre-election debate, didn't address the issues in enough detail.

More than half of California voters expressed positive feelings about the state's new domestic partners law, one of the measures Davis signed in the run-up to the recall. Two in 10 called themselves enthusiastic about the law, which gives same-sex partners many of the same rights and responsibilities as married couples, but at least one in 10 said they are angry about it.

Davis signed the domestic partnership law in September, granting same-sex couples in California nearly all the same rights and responsibilities as married spouses. The governor also reversed position on a bill allowing illegal immigrants to get a California drivers license and signed that into law. Only a quarter of voters agreed with that position, the exit poll found.

Fewer than one in 10 Election Day voters reported having problems with voting equipment or the length of the 135-candidate ballot.

Edison/Mitofsky conducted the Election Day survey at 60 randomly selected precincts around California, and the telephone poll of absentee voters from Sept. 29 through Oct. 5. Results from the telephone poll were weighted so that those responses represented 26 percent of the sample, the estimated size of the absentee vote.

(Copyright 2003 by The Associated Press. All Rights Reserved.)