InnovateUS 
Who wins Texas in the U.S. midterm election? Midterm Night Desk gives you different answers depending on what evidence you ask it to consider.
Combine October surveys from the Civic Health and Institutions Project (CHIP50) with public polls, and Democrat James Talarico has a 63% chance of beating Republican Ken Paxton.
Remove the public polls, and his chances rise slightly to 66%. Switch to “Fundamentals only”—electoral history, incumbency and the national environment—and they fall to 32%.
One setting changes and the favorite flips.
He wanted to see how quickly he could turn his team’s data into something people could use. “The answer was: remarkably fast.”
Northeastern’s David Lazer built the experimental dashboard with Claude in a couple of hours. He wanted to see how quickly he could turn his team’s data into something people could use. “The answer was: remarkably fast.”
The Texas figures are an invitation to investigate. What makes the tool interesting is the ability to pull apart its answer.
How much does the forecast depend on what voters say now? What happens when you rely instead on a state’s political history?
The speed of the build says little about the strength of the model. Even this version “required expertise,” Lazer says, and a solid one “would have taken significantly more time and more expertise.”
But a polished interface can make an unfinished experiment look authoritative. As Lazer puts it:
“It’s easy to quickly build something that has the appearance of credibility (LLMs are really good at that). It’s another thing to be truly trustworthy.”
CHIP50 supplies something an afternoon of coding cannot: years of survey research. An outgrowth of the COVID States Project, CHIP50 brings together Northeastern, Harvard, Rutgers and the University of Rochester.
Its surveys reach all 50 states and Washington, D.C., giving researchers evidence about places that rarely get battleground attention. Scholars can also apply to add questions and, if accepted, collect data free of charge.
Lazer’s concern is whether that work can keep going: “the sustainability of the system that creates the information that a model like this needs.”
Lazer built a new way to explore the data in hours. Collecting the next wave still takes people, money and voters willing to answer questions.