|
November 5, 2013 General Election Recap
The 2013 St. Petersburg City election is finally over, and now that we have the full voter data from the Pinellas County Supervisor of Elections office, we can do a follow-up on how our polling stacked up against the final election results. First, a few words on election's eve polling specifically. This was the third and final time that we performed polling on the night before an election. I say the final time because I have come to believe that it is just not as accurate. I can't offer up any concrete explanations for this, but for both the primary and the general election, results we were about 3% off of our week-before polling results, and that was 3% in the wrong direction. Earlier this year we also conducted a private poll for a client the night before an election, and again it was more inaccurate than the poll we conducted of that race two weeks before. My guess is that a segment of people are tired of polls by the night before, or they want to purposely respond with the opposite of who they voted for, but whatever the reason three strikes and it's out for us. Now, on to comparing the polling and election results. The final election results showed Rick Kriseman winning over Bill Foster by a margin of 11.8%. Excluding our election's eve polling, the last two polls we conducted(October 14th and 24th) showed about a 6% lead for Rick Kriseman over Bill Foster. After removing the undecided responses, Kriseman's lead rises to about 7%. These two polls used samples and scientific weighting based upon the August Primary voter demographics. The August Primary Election had about 2,500 less voters than the November General Election ended up having, and it also had some different demographics. The gap between Democrats and Republicans for the primary was 11.56%, and for the general it was 14.73%. That amounts to almost 2,000 more democrats voting in the general election than voted in the primary. This is greater than a 3% gap difference compared to how we were pulling our sample of voters and calculating our scientific results. Taking this into account, we can adjust our poll results to be more than a 10% lead by Kriseman. This means, after adjusting for final voter turnout(demographics), our poll responses were less than 2% off of the final election results. Another factor in our polling accuracy is that since October 16th, we are not able to call cellphones in any way, due to FCC regulatory changes. This did shrink our available sample, but after comparing results from the last four city polls we conducted with and without cellphones, we were never more than 1% off of the results without cellphones. So, this was not a significant factor related to the accuracy of our poll results here in St. Petersburg. Something else I want to mention is that six weeks before the general election, we conducted a poll with a sample and scientific calculations based on registered voter demographics, which has a higher ratio of Democrats to Republicans than the primary voter demographics do. This poll ended up with a 9.7% lead by Kriseman. After removing the undecided responses, that lead goes up to 10.7%, which was only 1.1% off of the final election totals. Even though this poll ended up being closer to the final election results than any other poll released during the general election cycle, several people reguarded it as an outlier. The Bill Foster campaign even dedicated one of its weekly Foster Feed emails to criticizing this specific poll as scientifically inaccurate "junk science". We also polled the four City Council races that were on the same ballot. We did not poll these races as often because none of them were as close as the race for Mayor was polling, and there was much less interest in these races, as evidenced by the fact that the final election had over 10% of the people who voted not voting in any of the City Council races. Even with our October 14th polling showing undecided responses accounting for 26% to 37% of the responses in these races, we ended up being only 1% to 5% off of the final vote percentages for each candidate in all four city council races after removing the undecided reponses. It's important to remember that polling is a science, and as such it needs to be adjusted from time to time reflecting new information. Looking back on our polling for the general election, we should have better analyzed the absentee ballot returns, and adjusted our random sample demographics to account for the higher numbers of Democrats that were voting early. As always, we will integrate this new information into our methodology to give us better polling results in the future. Matt Florell President - Fextel, Inc. |
