Poll Explorer

Surveys only capture a sample of the population, so we know that the result probably won’t exactly match the “true” result that we would get if we surveyed everyone in the population or that we would expect to see on election day.

Without adjustment, surveys tend to over represent people who are easier to reach and underrepresent those types of people who are harder to reach. In order to make the results more representative we weight the data so that it matches the population – based on a number of demographic measures. Weighting is a crucial step for avoiding biased results, but it also has the effect of making the margin of error larger. Using US Census data and voting history, we can get a rough sense of the proportions of gender, race, age, and political/voting history we would expect to see in our sample.

We can either weight data based off of the most recent census (the default) or exit polling.

To create a view of what will happen on election day, it is important to only sample those who vote. However, many people ineligible to vote and eligible voters who end up deciding not to vote still show up in our survey sample. To resolve this, we create a probabilistic likely voter model and weigh our results accordingly.

Our likely voter model is based on the Perry-Gallup index outlined in Pew 2016.

We give respondents "points" that estimate how likely we think they are to vote. More points means we think they're more likely to vote.

How many points to give respondents who...