For a long time, I've been concerned about the fact that all of our pre-election polls deliberately suppress the undecided vote. Instead of measuring how many people have made a decision, and how many have not, most pollsters today follow the format first used by George Gallup 73 years ago - a forced choice, hypothetical question: Who would you vote for if the election were held today (among a list of names).

He first started that in 1935, Alec Gallup acknowledges, because Gallup knew that if he asked who people would vote for in the 1936 election, so many people would say they didn't know, the newspapers wouldn't have found the results interesting. Unfortunately, pollsters have continued with that same question ever since.

The Gallup daily tracking poll from March 2008 until now shows an average of about 5 percent to 6 percent undecided voters, when we really know that many voters start out undecided and make up their minds during the campaign. It seems to me that polls should reflect that dynamic - with a certain proportion of voters who have made up their minds, along with another proportion (higher in the beginning, and tapering as we near Election Day) who are undecided.

My proposal for the new vote question is the following:

"In the presidential election next November, do you support the Republicans John McCain and Sarah Palin, the Democrats Barack Obama and Joe Biden, someone else, or are haven't you yet made up your mind?" [Rotate the first two choices.]

I think it appropriate to follow up with these two questions:

For those who made a decision, I would ask: "Is that a firm choice, or could you change your mind before Election Day"

For those who said they were undecided, I would ask: "Do you lean toward Obama or McCain?" [OR - we could ask the follow-up question, by using the standard "If the election were held today, who would you vote for?" The key is to have a follow-up question that presses the undecided to see if they are actually leaning toward a candidate.]

For the primary set of numbers to be released and tracked, I would report the results based on the first question, but would use the follow-up questions to provide a more in-depth look at the contest.

What I would like to see tracked - on pollster.com and other sites! - are the results based on the first question, so that we can see the size of the undecided from the beginning, and see how it declines as the campaign progresses. I have no problem if pollster.com (for example) also wanted to track the results based on forcing people to come up with a decision, but I think it important to track the undecided vote.

I make a somewhat extended argument for this approach in The Opinion Makers, so I won't do that now - except to say that I think this new approach gives a more accurate picture of the election. My objection to the standard vote choice question is that it misleads people - journalists, pundits, and even politicians - into thinking that voters have largely made up their minds a long time before the election. That seems particularly true during the primary season, when Rudy Giuliani was touted as the frontrunner, at the same time that a special Gallup poll in November 2007 (based on a one-time departure from the standard vote choice question) found that no Republican candidate had more than 5 percent of the vote - even though all the polls (including Gallup) were routinely reporting Giuliani as the dominant frontrunner.

Poll results have a tremendous impact on the campaigns - all the way from fund-raising, to news media coverage, to getting people to work on the campaigns - that I think the news media polls should provide as accurate a picture of the electorate as possible.

For pollsters who still want to make their final predictions, they can use this new version of the question, with the follow-up questions, to pressure voters into making a choice. But I would ask pollsters to make a distinction between 1) their final poll (based on the new vote choice question) and 2) their prediction (based on pressuring respondents to make a choice, and on any other modeling they might do).

In the New Hampshire Primary, that would have revealed (as did the UNH Survey Center) that 21 percent of Democratic voters were still undecided (in the last pre-election poll), but that when pressed for a vote, respondents at that time favored Obama over Clinton.

Do you see any downside to this proposal for a new vote choice question over the standard vote choice question that is used today? How do you react to my criticism that polls today do not give us an accurate picture of the electorate, and that by suppressing the undecided vote, polls not only mislead, but have a distorting effect on the campaigns?