Methodology

We conducts polls using Google Consumer Surveys.

We run “horse-race polling” attempting to predict the results of state-wide and national elections. We use a single-question format, to minimize both cost and response drop-off. We include the major candidates in the election, along with an “I don’t plan to vote” choice, and an open-ended text field for people to volunteer that they are undecided, refuse to answer, or prefer another candidate. For non-Presidential races, we also provide a choice for people mistakenly geo-targeted into the survey.

From this data, we then re-calculate the percentages with non-voters excluded, while still including undecided voters. Google Consumer Surveys can, in some cases, weight results based on age and gender. We generally will not use these when reporting top-line results, though we may comment on if they cause a significant change to results.

We avoid much of the “special sauce” that regular polls use to try to get more predictable results. We do not weight (or even ask about) race, ethnicity, political affiliation, or past voting history. We believe that the reach of the Google advertising network gives us a statistically representative sample of the American public, and our past polling results have supported that belief.

For further questions, please email Alexander Power at alex@pow3.com .