LatestPollResults.com has presented an unbiased probability analysis of the Electoral College for Presidential elections since the Bush/Gore contest in 2000. Our poll-based analysis determines probabilities for each candidate by analyzing the latest poll results in each state. In 2010, we extended our poll-based analysis to the Senate. In 2016, in addition to our poll-based analysis, we began our market-based analysis by tracking the pricing of futures contracts traded at the PredictIt.org marketplace. For 2018, we're adding our analyst-based analysis of the major political analysts who provide ratings for every seat in both the Senate and House.
Our model has been consistent through the years. For the Presidential election, we determine the probability for the Democrat and Republican candidate in each state, the District of Columbia, the two Maine districts and the three Nebraska districts which also have Electoral Votes. Thus, there are 56 individual probabilities for each Presidential candidate. For the Senate, we determine the probability for each party's candidate for each Senate seat. For 2018, there are 35 Senate seats up for election. For the House, we determine the probability for each party's candidate in all 435 House seats.
Mathematically, it presents a challenge to calculate every possible outcome in a Presidential, Senate or House election from the individual state/seat probabilities. Most political websites instead opt to run thousands (or even hundreds of thousands) of election "simulations" in an attempt to estimate the overall probability for each party to win the Presidency or control the House or Senate. But instead of estimating the probability with simulations, LatestPollResults calculates each party's exact probability of winning.
Senate Election - Similarly, in analyzing the Senate, using the probabilities for the Senate races, we compute the probability of every possible seat distribution to obtain each party's overall odds of controlling the Senate. In 2018, 35 Senate seats will be decided resulting in 2^35 or over 34 billion outcomes.
House Election - For the House, using the probabilities for the House races in each district, we compute the probability of every possible seat distribution to obtain each party's odds of controlling the House. With 2^435 possible outcomes, there are over 88 duoquadragintillion outcomes. This number is 131 digits long and dwarfs the number of atoms in the universe ... by 50 orders of magnitude!
To determine the individual state/seat race probabilities, we use one of three distinct methods:
Our poll-based method is simple. We use the latest poll from each state/seat race to find the probability of winning for the Democrat and Republican. The probability can be derived from the poll percentages for each candidate and the sample size of the poll. Once we have the individual probabilities for each state/seat race, we analyze all possible outcomes. Through and including 2016, our poll-based method has a perfect record projecting the winner for the Presidency and the party to control the Senate:
In 2016, we began tracking and analyzing an alternate set of probabilities for the Presidential and Senate races. PredictIt.org has emerged as the leading prediction marketplace for real-money politically-based contracts. The U.S Commodity Future's Trading Commission (CFTC) has issued a no-action letter for PredictIt.org to operate its marketplace and U.S. traders can legally buy and sell politically-based contracts based on an event occurring ... or not occurring.
Each contract trades in cent intervals between 1 cent and 99 cents and has a specific expiration date. Traders can buy or sell at the market price, any time until the contract expires. At expiration, the value of the contract will be settled at $1.00 if the event occurs and $0.00 if it doesn't. If the transaction is profitable, PredictIt deducts a 10% commission from your winnings. From the contract pricing, we derive the implied market-based probability for each candidate. For example, assume that a trader purchases a contract for the Democrat to win in Florida and pays 60 cents for the contract. If the Democrat loses Florida, the value of the contract will be $0.00 and the trader will have lost 60 cents. If the Democrat wins Florida, the contract will settle at $1.00 and the trader will win 40 cents per contract, less 4 cents for the 10% commission. The net winnings will be 36 cents. The implied market-based probability of risking 60 cents to win 36 cents is 62.5%.
Based on the bid/ask prices for futures contracts trading on the PredictIt.org marketplace, we calculate the implied probability for the Democratic or Republican candidate in each state/seat race. Since trading occurs 24/7, these market-based probabilities offer a live snapshot of each race. Note, there is a contract for who will win the Presidency, which party will win control of the Senate, and which party will win control of the House. Several websites report the market odds for these contracts. However, we compute the overall probability of winning from every possible outcome using the underlying individual state/seat race probabilities.
In 2016, the market-based model for both the President and Senate strongly pointed to a Clinton victory and a Democratic Senate. Entering election day, the market-based odds indicated Clinton's odds of winning were 87.6% while the odds that the Democrats would control the Senate were 65.8%. These market-based projections conflicted with our poll-based projections and proved to be inaccurate.
For 2018, we're analyzing the major political analysts who provide ratings for every seat in both the Senate and House. The Cook Political Report, Inside Elections, and Sabato's Crystal Ball are well known, widely respected, and have been publishing results for many election cycles. Seats are rated either Safe, Likely, Lean, or Tilt (Inside Elections only) for the Democrat or for the Republican or as a Toss Up.
To use our probability model, we need to assign a probability to each seat rating. We are using the accuracy of their past performance as a proxy for the probability of each qualitative rating. From these individual seat probabilities, for each analyst, we analyze all possible outcomes and calculate the overall probability of winning for each party based on historical accuracy.
LatestPollResults is operated and co-authored by Steve and Mark Delano.