Calculate the weighted polling average for each state where polls are available.

• All polls were gathered from RealClearPolitics, Silver Bulletin, and Race to the WH.

• Only polls from the last three weeks are included in the polling averages. If the only polls included are not from the last three weeks, then all polls from the most recent month available are considered instead.

• Polls are weighted based on the following factors:

  • Recency: Based on how many days the poll was released before election day.

  • Sample Size: Based on the number of people that answered the poll.

  • Poll Quality: Uses a score system of 1-10 based on the letter grade given to the pollster by Silver Bulletin. If the pollster is not included in Silver Bulletin's list, then a default score of "5" is given.

  • Sample Quality: Uses a tiered score system based on whether the poll was conducted amongst likely voters (Score = 1.0), registered voters (Score = 0.8), or another voter base (Score = 0.6).

• Additional Note: Official best-case scenarios for the incumbent and challenger party candidates are tabulated based on the weighted average margin of error and "absolute" best-case scenarios are tabulated based on additionally considering undecided voters.

Create a hypothetical margin of victory for each state from regression-selected fundamentals factors.

• Several factors were tested as potential predictors using the most recent race for each of the 100 Senate seat, of which the following were selected as the strongest using LASSO:

  • State Partisan Lean: The state difference from the national popular vote for the last two presidential elections, with the most recent race being given more weight.

  • Fundraising: The difference in individual contributions between the incumbent and challenger party candidates.

  • Incumbency: A numerical categorical variable based on whether the incumbent Senator is running for re-election (2 = Elected incumbent is running for re-election, 1 = Appointed/successor incumbent running for a full term, 0 = No incumbent is running).

  • Candidate Quality: The average difference between either the incumbent or challenger party nominees' performance in their last two statewide campaigns to the most recent presidential election result in that state. Polling numbers are used alternatively if either nominee did not compete in a previous statewide election.

  • Positive Party Turnout Difference: The percentage difference in the number of people who voted in the incumbent party's primary from the number of people who voted in the challenger party's primary.

  • Competitive Primary for Challenger Party: Based on whether the primary for the challenger party was competitive or not, which is determined based on whether that party's nominee either failed to secure at least two-thirds of the primary vote OR failed to secure at least a 15% margin of victory against their closest opponent.

  • Political Experience: A numerical categorical variable based on the amount of political experience both the incumbent and challenger party candidates have (0 = No elected experience, 1 = State legislators and local officials, 2 = U.S. House members and statewide officers excluding Governors, 3 = Current or former Governors, U.S. Senators, or national executive officers.)

  • National Environment Implication from Special Elections: The projected national environment based on the average difference between all congressional and state legislative district special elections and their partisan lean from the last two years.

• These factors were then placed into an ordinary least squares (OLS) regression formula capable of calculating hypothetical margins of victory for historical and future races.

Determine the influence percentage polls and fundamentals will have in the official forecast for each state, then apply to create a prediction.

• The percentage of influence polls will have in the forecast is tabulated using an average of "recommended influence" percentages, based on the following five polling factors:

  • # of Polls: The number of polls released on the state within the last three weeks. At least three polls are required to get at least 50% weight.

  • Poll Consensus: A binary variable based on whether all polls used for the state agreed on a predicted winner (Yes = 100% weight, No = 50% weight).

  • % of Undecided Voters: The weighted average percentage of undecided voters. 3% or fewer respondents must be undecided for polls to get at least 50% weight.

  • Lopsided Victory: A binary variable based on whether the margin of victory for predicted winner in the weighted polling average is outside the margin of error (Yes = 50% weight, No = 100% weight). This weighting is applied due to the error difference between the predicted and actual margin of victories being greater in safe states than swing states.

  • # of Days Prior to Election Day: The number of days before Election Day the most recent poll for the state was released. The most recent poll must be released either 21 days or sooner before Election Day to get at least 50% weight.

• The percentage of influence fundamentals will have is based on the remainder not given to polls. If no polls were released in the state, then fundamentals gets 100% weight in determining the forecast.

• Calculate the weighted polling average of the polling and fundamentals margins of victory to generate an official projected margin of victory.