October 20, 2016
This week, we’re releasing our latest set of battleground polls. We polled likely voters in Georgia, Florida, North Carolina, and Virginia as part of our new series of state surveys in this election season. Along with our state polls, our daily Presidential tracker showcases the power of Lucid’s platform and our marketplace for online sample. While we’re excited by how new tools like ours can help transform the polling world as traditional telephone samples become increasingly difficult and expensive to model, we’re also humbled by the challenge that pollsters go through every day to produce and publish reliable results.
We spent a lot of time crunching our latest numbers this week and found and corrected some anomalies in our calculations. As a result, we decided to go back and re-check our previous results to ensure these issues were not missed for earlier publications. In fact, we did find two errors impacting the results we published last week for Iowa, Indiana, Pennsylvania, and Ohio.
Neither of the issues we detected had an impact on who we had in the lead for either the Presidential race or each state’s respective Senate campaigns. In the first error, we did not properly exclude a small number of incomplete surveys. This had a minimal impact in the Presidential results as you’ll see in the accompanying table.
The second issue involved our calculation of the final Senate results. Instead of adding undecided voters who leaned toward one candidate or the other as we had intended, we misclassified leaners as pure undecided voters. While this mistake also did not impact the direction of any of the campaigns, it caused us to overstate the uncertainty of voters in these elections.
The corrected results are as follows:
While the error was unfortunate, we’re proud that our results stack up favorably to other published polls. With so much changing about the polling industry, we’re excited to take part in the dialog and grow Lucid’s impact as a solutions provider for political researchers.