Barack Obama is ahead of Hillary Clinton in North Carolina by 12 points. No, it's 5 points. Wait, now the lead is 14 points.

That's all in the same week, as reported in an N&O story Thursday that referred to various polls that are being conducted on the Democratic presidential primary. The 12-point lead was reported by Raleigh-based Public Policy Polling.

Only a week earlier, PPP had said Obama was ahead by 25 points. Had the race tightened that much -- the poll was taken in the midst of the latest Jeremiah Wright controversy -- or is something wrong with the numbers? Meanwhile, SurveyUSA was showing Obama's lead at only 5 points and the Rasmussen Reports poll had him up 14 points.

The polls are an interesting, and fun, tip sheet for the political horse races, but not necessarily a take-it-to-the-bank guide. Last month, PPP showed Democratic gubernatorial candidates Beverly Perdue and Richard Moore virtually tied one week, then Perdue ahead by 10 points two weeks later (the polling method changed.) Another poll at about the same time by the Civitas Institute had Perdue and Moore tied.

The day before the April 22 Pennsylvania primary, Public Policy Polling showed Obama ahead of Clinton 49-46. Clinton ended up winning by nearly 10 percentage points.

With such wandering numbers, what's a voter to believe? And why should a newspaper publish them?

Bill Krueger, The N&O's political editor, says the various polls, even when they conflict, are a useful running sample of what the voters are thinking at a given moment in time. The N&O reports the latest findings as they are released by the pollsters on a weekly or monthly basis, but, Krueger says, not as definitive information on the races. "We're factually reporting the results of a poll," he said. "I don't see that we're putting some kind of seal of approval on it." He noted that poll results don't run as news stories but as items in Under the Dome, the political gossip column that runs on Page 5B.

The N&O itself was criticized when it published its own poll on the elections in early April. The poll, conducted by Braun Research for The N&O and The Charlotte Observer, showed Moore leading Perdue 25-19, but with a huge undecided vote of 53 percent. Even in the Obama-Clinton contest, the undecided vote was 39 percent, compared with less than 20 percent in other polls.

"I certainly hope the McClatchy papers will not start reporting these numbers like they're the gospel truth," Tom Jensen, communications director of Public Policy Polling, wrote in a blog. "Because the simple reality is that they're wrong." The N&O did rely on the numbers for an editorial about how candidates can't put much stock in polls with so many undecideds.

The undecided vote in that poll was so large because Braun surveyed registered voters who said they intended to vote in the primary. Wouldn't we all say that? Most other polls narrow the sample to people who had voted in the last election.

The broader sampling was deliberate, said Joe Dennery, the Charlotte Observer editor who coordinated the survey with Braun, because it focused more on the issues that were on the minds of North Carolina voters, rather than on the horse-race numbers. "If I had it to do over, I would say we shouldn't have used the term 'likely voters,' " in the story about the poll, he said.

In retrospect, the survey probably shouldn't have done a horse-race reading at all, since the numbers were so misleading.

The lesson from all this is that readers need to regard the polls with skepticism and to check the accompanying information about how the poll was conducted -- sample size, margin of error, wording of questions, how people were surveyed and other key factors. And check the surveying organization. Public Policy makes its money working for Democratic candidates (though none in the races it polls), Civitas is a conservative think tank and Braun Research doesn't specialize in political polling.

With all those caveats, why should we pay attention to polls at all? Because, says UNC polling expert Phil Meyer, they stimulate interest in the elections. Studies show that people who read polls are more likely to vote and that they're better informed about the issues in the election.

"It kindles interest in the election," Meyer said. "How interested would you be in a basketball game if you couldn't see the scoreboard?"

Other politics

I've been asked whether The N&O intends to endorse in the presidential primary. Not unless something changes after I've written this column.

Steve Ford, editorial page editor, said the paper hasn't endorsed in primaries in at least a dozen years. Part of the reason, he says, is that the paper could end up arguing against itself.

If The N&O endorsed Clinton or Obama in the primary, then it could find itself in the fall election choosing between the Democrat it had already endorsed and, presumably, John McCain. So if The N&O endorsed McCain, it would be opposing the Democrat it endorsed in the primary. (OK, you don't need to tell me how many times The N&O has endorsed a Republican for president; we're talking hypothetical here.)

The same reasoning applies for all other races on the ballot -- governor, U.S. Senate, General Assembly, county commissioner, etc. The N&O did endorse last week in the Durham school board race, but that is a nonpartisan, nonprimary election that decides who will take office.

Too bad. The N&O could be a king-, or queen-, maker. Before the Iowa caucuses, editors at the Des Moines Register were wined and dined, literally, by Hillary and Bill Clinton and Barack Obama and were wooed by Madeleine Albright, Walter Mondale, Robert Kennedy Jr. and Gen. Wesley Clark. (Ford says the N&O editorial page staff is not amenable to being wined, dined or wooed.)

The Des Moines Register endorsed Clinton; Obama won.