Northwestern Model Tracks 35 Senate Races Every 10 Minutes

A new Northwestern University election model now updates Senate race forecasts every ten minutes.

The system, named NowCasting, tracks all 35 contested U.S. Senate races across the country. It calculates the probability of a Republican victory in each state and the likelihood that the party will retain control of the chamber.

Every ten minutes, the algorithm runs one million hypothetical midterm elections. This process simulates 35 million individual Senate race outcomes to generate current probability estimates.

The model places specific weight on market activity from the preceding hour. This design allows the system to detect immediate responses to singular news events. Users can view state-by-state predictions and overall trends on an interactive, color-coded map.

Thomas Miller, a faculty director in Northwestern’s School of Professional Studies, developed the platform. He stated that the system combines the expectations of thousands of traders with statistical modeling. Miller explained, “We’re not just looking at individual races. We’re connecting the dots across the country to understand the bigger picture in the battle for Senate control.”

NowCasting draws on data from two major prediction platforms, Kalshi and Polymarket. Previous models by Miller relied solely on PredictIt. The new system also incorporates statistical methods used in finance and economics to identify relationships among the 35 races.

Miller has built election forecasts for several recent cycles. In 2020, his Virtual Tout system predicted 294 Electoral College votes for Joe Biden, a figure twelve votes below the final total of 306. In 2021, his surveys correctly predicted wins for Democratic Senators Jon Ossoff and Raphael Warnock in Georgia. In 2024, his model favored Donald Trump for president, aligning with the final result despite other major forecasts favoring Kamala Harris.

The key distinction in NowCasting is its treatment of market data. Miller noted that prediction markets provide event contract prices rather than direct probabilities. The system uses statistical calibration to convert those prices into its own estimates of winning likelihoods.