Basics of Election Prediction

Methods for prediction

We will visit some information sources you may consider when designing your own election prediction strategy. Each source can be used as a starting point, and then expanded upon or combined with another approach, to obtain the best possible prediction.

Current Polling Data

One strategy for predicting the winner in an election would be to use the latest aggregate polling data from a reputable source, e.g. FiveThirtyEight or The New York Times.

  • Polls may use land lines (robocalls), cell phones, or web surveys.
  • Some polls are of likely voters, while others may not restrict to this group
  • Polls vary in quality (you can find some quality ratings online)
  • Polls may be subject to bias (e.g., nonresponse bias, assumptions involved in determining “likely voters”)
  • Polls associated with margin of error
  • More polls focus on national sentiments than on state-specific sentiments

Sources of election uncertainty

  • Sample sizes of polls
  • Individual changes in turnout
  • Systematic changes in turnout (different turnout patterns from historical records, e.g. more young people vote)
  • Individual variation in support (undecided voters)
  • Unmeasured bias in polls

Historical results

  • Use historical data from past elections, e.g. within a house district, within a state, nationally
  • Useful when a state consistently votes for the same party but less useful for swing states
  • Check out the NC Voter Record

Voter Turnout Data

  • States like NC make available data on who votes in each election
  • Voter turnout data can be used to construct a voting history for certain districts, demographic groups, etc.
  • Locations with low or variable turnout often harder to predict
  • Many better predictions take voter turnout data into account in some way

Ancillary Data

  • Voter results for other elections, e.g. Senate results when predicting the presidental election
  • “It’s the economy, stupid” – James Carville
  • Societal movements (e.g., #BlackLivesMatter)
    • Incumbency
    • Pandemics
    • Wars, hurricanes, etc.
    • These factors may be helpful when predicting results if they are correlated with the election results. (Remember to use caution with this type of approach because correlation does not imply causation!)