Lawson Comp Sci building ccambron

The Lawson Computer Science building houses the office of Dan Goldwasser.

Two Purdue professors are creating a website for the public that will use live data to predict the outcome of legislation at the state-government level.

Eric Waltenburg, a Purdue professor of political science, and Dan Goldwasser, a professor in the department of computer science, are bringing their disciplines together to create a new predictive model, according to the Purdue News Service.

The model, which uses available data from the state legislature, began in Indiana, but now the project is being tested in Oregon and Wisconsin, said Goldwasser.

The idea of the model is to use live data, like that of social media, to predict what will happen with new bills.

“We are also collecting data, which is not official, and this includes coverage of political activities, prominent people (and) social media accounts,” Goldwasser said.

Quality of data is another question entirely. Rosalee Clawson, a political science professor, said that the ability to predict outcomes lies in the information that can be obtained by the system.

“To develop a forecasting model, it’s critical to have high-quality data on the variables that matter for predicting the electoral outcome,” Clawson said.

The professors are taking steps to move from a fixed model to one that can take in data as it appears.

“One of the things that we want to do is to open it up and say, ‘We predict that these are trends that you’re going to see in the next year,’” Goldwasser said of the website. “Or, ‘These are the types of legislation which we think (are going to) get passed in the next year.’”

The process will utilize machine learning, which is on the frontier of modern research capability.

“(Machine learning is) a new way to instruct computers to do what you want them to do,” Goldwasser said. “Without involved specific programming for any task.

“At least (in the) idealized view of machine learning, the same algorithm can generate many different systems for many different tasks by just taking data as an input.”

The perception of the model’s results is an important point, said Clawson.

“If a pollster says, ‘Hillary Clinton has a 70 percent chance of winning the (election),’ the media often report it as ‘Hillary is predicted to win,’” Clawson said. “But that’s way too simplistic. If Hillary has a 70 percent chance of winning, that also means that Donald Trump has a 30 percent chance of winning.”

Goldwasser clarified that the amount of reliability they hope to achieve with the model is not necessarily a measurable number.

“We can think about that number as a way to try and understand if this is an easy problem or a hard problem,” Goldwasser said. “Meaning that if we don’t try hard and we get 100 percent (accuracy), maybe it’s not a hard problem.”

He chuckled as he said that 100 percent accuracy is not likely, but that it’s a future hope for the model.

Recommended for you