To ensure accurate results, we use a set number of shuffled phone numbers of registered voters. Only one survey is allowed per phone number per run, and we keep logs to ensure accountability. For most polls, we do allow people to call in as well within 24 hours of the end of the polling if they were unable to take the survey when we called them. But again their phone numbers are validated against our calling list, so no poll-stuffing is possible.|
The list of people to poll comes from the publicly available state-registered voters list for the state of Florida. It is not a complete list of registered voters, it is only those that added their phone numbers when they registered to vote(which is optional) in addition to white pages phone book listings for registered voters.
We are not affiliated with any political party, and we always strive to keep any appearance of a bias on an issue out of how we do our polling.
As of the March 2012 St Petersburg General Survey, the statistics for the General St Petersburg surveys are scientifically weighted by political party, race, gender and age. This is done by taking the percentages(down to the tenth decimal point) of each of those demographics for the entire voting population of St. Petersburg, and using all four categories to weight the results that we get to adjust our sample to fit the actual demographics of the city. Because we use such a large sample size, the weighting has not changed the results that much, no more than 0.4% at most.
Starting in the Spring of 2012, we were working to create a framework to allow us to scientifically weight our polling results by political party, race, gender and age. After we finished, we adjusted the results from our previous 3 St Pete Polls(June 2012, May 2012, March 2012) using our new scientifically-based algorithms to weight each response to each poll question to ensure a proportional and more accurate total percentage. Due to the large sample size we use, about 8 times a standard opinion poll sample size, and the fact that the sample we use is already pretty demographically representative of the voters in the City of St Petersburg, the differences between the raw percentages and the scientifically weighted percentages were very small, usually below 0.3%. We also did extensive testing using different sample sizes, and as expected, we saw a larger difference when we lowered the sample size, but the algorithm test results were always closer to our full sample scientific results, so we know our process is working as it should.
What this means for us going forward is that we will be able to present poll results that are at least within a 4% margin of error and a 95% confidence level, depending on the sample size. We will still do some non-scientific polls, depending on where we are getting the data from, but most of our polls going forward will be labled as "scientifically weighted" if they use our new analysis methods.
During the 2012 General Election, our polling results were very close to the official election results. For a summary of how we did, click here.