The rematch between Gov. Chris Gregoire and Dino Rossi is either as close as it was four years ago — or not. There have been 15 polls published on this race since the first of July, including three Elway Polls. Among the last 12 polls conducted by national pollsters watching this race, the average is 49 percent to 48 percent, in favor of Gregoire. The Elway Poll's three-month average is 51-39, which includes the most recent version, of Oct. 19. What's up?

One explanation is our finding of 10 percent still undecided. The other polls average 3 percent undecided. The Elway Poll numbers for Gregoire (average 51) are close to those of the national pollsters (49), well within the margin of error. The difference has been in our consistently lower numbers for Rossi and our higher undecided numbers. Why would that be?

The reasons for different survey estimates for the candidates have to do with the mechanics of the surveys. There are differences between The Elway Poll and the national pollsters [PDF] as to who we sample, who asks the questions, and how the questions are asked. The chief suspect in the mystery of the undecideds is the difference in the way the question is worded. Not all of the national pollsters reveal their question wording, but most ask a variation of this: "If the 2008 election for governor of Washington were held today, would you vote for — Republican Dino Rossi or Democrat Christine Gregoire?" We ask, "As things stand today in the race for governor, for whom are you inclined to vote between Christine Gregoire, who prefers the Democratic Party, and Dino Rossi, who prefers the GOP Party."

Note two subtle but not unimportant differences. First, we do not ask the respondent to pretend that the election is today. We ask who they are "inclined to vote for." Also, we do not push undecideds to indicate which way they "lean." Our method typically results in a larger undecided total in our surveys, which, in turn, results in a wider gap between the candidates. Second, while others identify the candidates as Republican or Democrat, the Elway Poll identifies the candidates as they appear on the ballot. Thus, Rossi was listed as "prefers the GOP party."

We found in our June survey that 25 percent of the registered voters did not know what "GOP" (Grand Old Party) stands for. This confusion may contribute to the higher number of undecided voters in our poll. It also may depress the Rossi total. In last week's poll, 12 percent of respondents did not know that Rossi is a Republican. Of those, only 15 percent planned to vote for him. On the other hand, of those who did know his affiliation, three times as many, 45 percent, planned to vote for him.

The outcome of the race will depend on these undecided voters and new voters.

Allocating undecided respondents is always a challenge for pollsters. One way is to assume that late-deciding voters will break for the challenger by as much as three to one. That would be the most generous to Rossi, but it is not an unwarranted assumption. So give Rossi 75 percent of the undecideds in our last poll and Gregoire leads by 53-47.

Second, the new voters. New voter registration here and across the country is breaking all records. The profile of the electorate is changing faster than the polls can measure. It is not just the demogrpahics of the new voters, but finding them in the first place. Younger voters are far more likely to be cell-phone-only users, which makes them more difficult to contact by conventional survey research. Pollsters everywhere are losing sleep over what these new voters might do to their models this year.

The common assumption is that the new voters are overwhelmingly likely to vote Democratic. Notably, our October poll had the highest proportion of Democrats in the 16-year history of the Elway Poll: 46 percent said they would register as a Democrat if they had to register by party in order to vote. The number of Democrats has been rising in Washington, as it has nationally, so it is possible that we just detected the surge. But 46 percent may be too high, so to temper that single poll result, we can average the proportions over the last three months. That results in 43 percent Democrats, 29 percent Republicans, and 28 percent independents in Washington.

But now look at the comparable three months leading up to the 2004 election, when the averages were 40 percent Democrats, 32 percent Republicans, and 29 percent independents. If we give further benefit of doubt to the idea that there may be too many Democrats in our sample, we can statistically adjust party identification back to 2004 levels and apply those adjustments to our findings. Doing that shrinks Gregoire's lead to 47-43, with 10 percent undecided. Now give Rossi 75 percent of the undecided and he has a tiny 50.5-49.5 edge. Do those numbers look familiar?

Of course, if the 2008 Democrat surge is real, and their proportion is higher than the 2004 average of 40 percent, then Gregoire's numbers go up.

So here's the bottom line. If Republican-Democrat party identification is still at 2004 levels, and if Rossi gets 75 percent of the undecideds, then he was leading by 50.5-49.5 with two weeks to go. If the Democratic surge materializes, depending on its magnitude, then Gregoire was leading by as much as 54-46, even giving Rossi three of every four undecided voters.

One last caveat. Often lost in the heat of an election are the realities that surveys do not predict the future and they deal in probability, not certainty. They are estimates or snapshots of the electorate at a specific moment. The poll we are looking at here was taken two weeks before Election Day. At that time the range of estimates was from a narrow Rossi lead to a substantial lead for Gregoire. Statistical probability and political events both favored a Gregoire lead.