Sampling for Florida's Democrats

I have an idea for meeting Florida’s need to have a do-over in the Democratic primary. As I understand it, the biggest sticking point in this on-again, off-again proposition is the cost of a statewide primary election, a problem my proposal addresses. Instead of spending tens of millions on a new statewide primary, why don’t the Democrats conduct a cost-effective census-like sampling of registered Democrats in the state to ascertain their presidential preference? It would cost a fraction of what a statewide primary election would — and produce an indisputable picture of what Florida Democrats want their delegates to do.

Note that I use the word “sampling” to describe what cynics and critics might call a “poll.” If proponents of my idea ever want to move the concept ahead, never use the word “poll” to describe what I am suggesting.

First, it’s simply not an appealing term to describe anything with such serious intent. Too often the words “rigged,” or “slanted,” or just “bad” appear in conjunction with the word “poll.” But it’s also an inaccurate term, because my plan, outlined below, is vastly different from political polling.

So “sampling” is the key word here. Some will remember that Democrats have championed sampling in the conduct of the decennial census. In 1999 this issue went all the way to the Supreme Court. The Bush administration and Republican-appointed justices of the Supreme Court quashed sampling. So Florida Democrats should warm to the concept naturally.

The general outline of the concept would be simple. Let Sens. Barack Obama (Ill.) and Hillary Rodham Clinton (N.Y.) campaign in the state for a week or two. During this campaign period, a coalition of public polling organizations (I’d nominate Gallup, Pew and CBS/New York Times) would take the list of Florida voters registered as Democrats. Researchers would choose by random means a statewide sample of 10,000 voters. These 10,000 voters should be sent an overnight letter notifying them of their inclusion in the sample that will select Florida’s delegates for Obama or Clinton. Fourteen attempts should be made to telephone each voter over the seven days of the census, two calls per day at different times. Messages should be left with a toll-free number for the convenience of those who would prefer to call back for their interview.

This telephone number would also be included in the pre-notification letter. The call-back option would address issues like cell phones-only voters, no-phone voters, or unlisted numbers. Voters calling in would have to provide some information to confirm their identities.

As I stated above, this is not “just a phone poll.” Today’s political polls would almost never use pre-notification, 14 call-backs over seven days and a toll-free inbound number. But more importantly, I would propose that results of this survey NOT be weighted to the population that was sampled. Like a walk-in election, there is no assumption that those voting are a truly cross-sectional sub-sample. If more senior citizens in the sample cooperate and complete their interview at a higher rate, then they constitute a disproportionately higher percentage of the votes counted.

If you are interested, the 10,000 sample would produce a margin of error of ±1.2 percent at the 99 percent confidence level in a close race like we’d expect. Normally, political polls suffice with 95 percent confidence.

Surely Democrats could devise a way to allocate votes if the results are even closer to dead-even.

This project should cost less than $1 million and set a seminal example for the public use of sampling in the electoral process.

This approach wouldn’t work as well in Michigan because voters there don’t register by party. And the fact that Michigan allocates voters by congressional district would also raise the sample size by a factor of 15, eliminating the savings possible in Florida.

Hill is director of Hill Research Consultants, a Texas-based firm that has polled for GOP candidates and causes since 1988.