The tough tone of the Pennsylvania Democratic campaign tarnished both candidates -- more so Hillary Clinton, with 67 percent of voters saying she attacked Barack Obama unfairly.
Yet it appears to have worked: Late deciders favored Clinton by a wide margin, boosting her to an essential victory in the state.
Clinton beat Obama by 59-41 percent among voters who decided in the last few days, countering his equally large edge among those who picked their candidate between a week and a month ago. It was the sixth primary in a row in which she won late-choosing voters.
Other support groups were crucial for Clinton: Women accounted for 58 percent of voters, a record for Pennsylvania primaries (though about what it's been elsewhere this year), and she won them by 59-41 percent; she won white women even more broadly, by 68-32 percent.
Clinton again won seniors, this time by 63-37 percent. And she won voters who haven't been through college by a bigger-than-usual 58-42 percent margin.
Obama countered with broad support in Philadelphia (66-34 percent over Clinton) and a 4-point better showing among white men, a swing group in these primaries, than he managed in Ohio, which Clinton won March 4.
But Obama still lost white men, by 14 points. (He split men overall with Clinton, 51-49 percent.)
Obama also won 90 percent of African-Americans. But at 15 percent, their turnout was lower than the Pennsylvania primary record, 17 percent in 1988 and 1984.
In a difference from supposedly similar Ohio, 47 percent of Pennsylvania voters were college graduates -- usually a better group for Obama -- compared with just 38 percent in Ohio. But college graduates divided by a close 51-49 percent between Obama and Clinton, compared to his 9-point advantage in this group across all primaries to date.
Another difference strongly favored Clinton: White Catholics accounted for an unusually large share of voters -- 32 percent, vs. 16 percent in previous primaries this year -- and they favored her by a very wide 72-28 percent, a gap that's appeared in some previous contests (e.g., New Jersey and Rhode Island) but not most.
One reason is that white Catholics in Pennsylvania were less apt to hold college degrees -- again reflecting Clinton's better showing among less-educated voters.
While two-thirds of voters said Clinton attacked Obama unfairly, 50 percent also said Obama unfairly attacked Clinton. Both numbers were higher than in previous primaries overall -- by 15 points for Clinton and 12 for Obama -- reflecting the negative tone of the campaign's closing days.
However, voters who said Obama attacked unfairly were more apt to punish him for it: Clinton won those voters by 67-33 percent; of those who said Clinton attacked unfairly, Obama won by a narrower 54-46 percent.
For many voters, moreover, it didn't matter; more than usual decided early. Sixty-one percent said they picked their candidate more than a month ago, compared with 45 percent in previous primaries this year.
Yet as noted, those who did decide late went for Clinton, reversing Obama's edge among those who decided in the previous week to a month.
Despite Clinton's victory in the state, overall expectations were on Obama's side. Fifty-five percent said they expected him, not Clinton, to be the party's eventual nominee.
And Obama supporters were more confident: just 5 percent of his supporters thought Clinton will win the nomination; by contrast, 23 percent of Clinton supporters said they think Obama will be the nominee.
If Obama does win the nomination, a quarter of Pennsylvania Democrats say they'd either support John McCain or sit out the contest entirely; if Clinton's the nominee, one in six said they'd either vote for McCain or sit it out.
That, however, may indicate the heat of the Democratic contest more than it predicts the future; the question's akin to asking a quarreling couple what they'll be doing for Valentine's Day. It might end ugly -- but they also may kiss and make up.
At the same time, there is a racial element to these views. Among white voters who said the race of the candidates was important in their vote -- albeit a small group, 12 percent of all voters -- just 55 percent said they'd support Obama vs. McCain.
Of the rest, 27 percent said they'd back McCain, and 15 percent said they wouldn't vote. Whites who discounted race as an issue were 17 points more apt to favor Obama against the Republican.
Fifty-eight percent saw Clinton as honest and trustworthy; in a hold-their-nose result, 23 percent of those who didn't see her as honest voted for her anyway.
More overall -- 67 percent -- saw Obama as honest. And despite the sniping that flew during the campaign, Obama and Clinton were rated equally on being "in touch with people like you" -- two-thirds for both.
Fifty-five percent in Pennsylvania rated the economy as the top issue in the election, similar to what it's been in all Democratic primaries to date. Nearly nine in 10 said the nation's economy is in a recession; 41 percent said it's a "serious" recession.
On attributes, as elsewhere, the top quality cited by far was the candidate who "can bring about needed change"; that again worked for Obama, with "change" voters favoring him by a vast 69-31 percent.
Clinton came back among voters focused on experience -- winning 94 percent of their votes -- as well as the relatively few who cared most about electability.
Clinton won small town and rural voters, 63-37 percent; weekly churchgoers, 59-41 percent; and gun owners, 63-37 percent -- all groups of interest given the controversy over Obama's comment about small-town voters clinging to guns and religion.
In Ohio, whose primary preceded those comments, Obama did better among weekly churchgoers (losing them by 4 points) -- but worse among small-town and rural voters, who favored Clinton by 70-26 percent. Gun ownership wasn't asked there.
Among other differences with Ohio, Obama did better in Pennsylvania with seniors -- he lost them by 63-37 percent, better than the 72-26 percent tally in Ohio. But there were more of them in Pennsylvania, mitigating the change.
Results reported in this analysis have been updated with a final weighting of exit poll data.