I'm going to start rolling out some state Democratic presidential primary polls. My approach to polling is different from most in that I rely on past voter history from the vote file (RBS) rather than calling randomly generated numbers (RDD) and asking some screening questions. RDD works well in high profile elections, but can miss regional and demographic patterns that influence lower turnout elections.

For example, I polled Texas at approximately the same time as ARG a couple of months back. Their results showed Obama only two points behind Clinton, while I had him in third behind Edwards. I was able to reconcile the two polls by reweighting my regional numbers to match the general population. Obama was very strong in the big cities, but very weak in South Texas. Problem is, the big cities have very low turnout in primaries. South Texas' turnout percentage is three times that of Dallas. By sampling based on actual voter history, these proportions are already baked into the data.

Last week, Datamar released a surprising poll that showed Edwards in the lead. I think they are using RBS, as they use

a comprehensive predictive model of "likely" voters, based on election cycles and other factors of voters in Florida. The survey was conducted using an automated telephone dialer and the voice of a professional announcer. A Datamar proprietary algorithm was used to generate random samples of voters for calling.

They sample 'voters' based on 'election cycles,' which is a description that could apply to my methodology. Maybe I'm reading too much into that description, but either way, the demographic profile is very close to a sample I had FloridaVoterFile.com pull based on my criteria. I was surprised at the low Hispanic percentage, but Florida Hispanics are 1) registered at a much lower rate than the general population, 2) in younger, low turnout age groups compared to the genpop, 3) more likely to be Republican than Hispanics nationwide, meaning fewer Hispanic Florida Primary-voting Democrats than you might expect.

Datamar uses different age groupings than FloridaVoterFile, but the percentages look like they could match up to the General Election profile among registered Democrats. Edwards is strong in the older age groups, and with 42% of Datamar's sample over 65, there should be no surprise that he polled well. However, Democratic Primary elections in Florida have a significantly higher senior percentage than General Elections. Edwards may actually have a bigger lead than Datamar suggests.