What to make of the new Field Poll on Prop. 30? Looking back at the state’s political history, most propositions in roughly similar situations pass, but tax-increase measures fare especially poorly in the final week, with undecided voters breaking sharply toward opposition, and sometimes even with former supporters migrating to the no column.

Here are the numbers:

Using Field Polls from the past 15 years, and restricting the list to cases in which the survey’s final day of interviewing occurred less than 10 days before the election, we get a set of 82 surveys on ballot measures. (For a similar analysis of the previous Field Poll on Prop. 30, go here.)

In the broadest terms, the final Field Poll is a pretty good indicator of what will happen on Election Day. Of the 43 measures that had a lead in the poll, all but six passed. Of the 38 measures that trailed, all but two failed. (One measure was tied at the final poll.)

The overall pattern is that undecided voters break more toward opposition than support, but not enough to defeat a measure that has a lead. Comparing the final poll to the outcome on Election Day, opposition increased an average of 10.4 percentage points, while support increased 5.3 points.

But those numbers include measures that were either far ahead or far behind in the polling. What about past measures that were leading with a week to go, but with support under 50 percent, as Prop. 30 is now?

There are 17 such measures, though four of them were a matched set of Indian gaming propositions on the 2008 ballot. Treat those as a single case, and we have 14 relevant examples. In the final poll, those 14 measures averaged 45.6 yes, 36.3 no, 18.1 undecided, with an average lead of 9.3 points. That makes them roughly analogous to Prop. 30’s current numbers: 48-38-14, for a 10-point lead.

So what happened to those 14 measures at the polls? Nine of them passed and five failed. On average, opposition increased 10.7 points from poll to Election Day, and support increased 7.4 points, meaning that the undecided broke about 59-41 towards opposition. If you apply that formula to Prop. 30’s current numbers, it would end up passing 53.7 to 46.3.

But here’s a critical caveat: None of those 14 measures asked people to raise their own taxes. In the past 15 years, there have only been five statewide propositions to increase taxes, so it’s difficult to generalize.

Still, the tax measures typically fared more poorly in the final week than other propositions. In two of the five cases, the level of support actually dropped, meaning that not only did the undecided voters move towards a “no,” but some of the previous supporters shifted camps as well. If that happens this year, Prop. 30 would obviously be dead.

On the other three tax-increase measures, support increased, but slightly. The undecided voters broke sharply toward the opposition, by almost a three-to-one ratio. If you apply that ratio to Prop. 30’s 14 percent of undecided voters, it would pass, but very narrowly, 51.5 to 48.5.

In terms of specific numbers among the tax measures, Prop. 30 seems closest to the two tobacco taxes of the past 15 years. In 1998 Prop. 10 led 50-41-9 with a week to go and hung on to pass 50.5 to 49.5. In 2006 Prop. 86 was in a dead heat 45-45-10 and died 51.7 to 48.3.

Depending on your position about Prop. 30, you can probably read these numbers to suit your own ideological preconceptions. My own view, for what it is worth, is that the historical record is actually more positive for the governor than I would have thought. I expected to find that measures polling under 50 percent with a week to go almost invariably lose, and that is not really the case. Still, the measure’s passage is, obviously, far from a sure thing. We’ll find out next week.