Newsroom Archive

Bill Bame, a 52-year-old software developer from Hampstead, Md., was selected as the winner of the U.S. Census Bureau's Census Return Rate Challenge. The challenge, conducted from Aug. 31 to Nov. 1, asked contestants to create the best statistical model predicting 2010 Census mail return rates at block group level geographies based on their demographic characteristics. The contest attracted 244 entries.

“During the 2010 Census, the Census Bureau saved an estimated $85 million in operational costs for every percentage point increase in the nation's participation rate by mail,” said Tom Mesenbourg, acting director. "Utilizing the winning model will help the Census Bureau create robust scoring metrics to identify hard-to-count areas and populations in the 2020 Census and save more money."

Bame won the $14,000 top prize with a model that came within 2.59759 percentage points of predicting the actual 2010 Census mail return rate of 79.3 percent. The return rate is the percentage of people who mailed back their 2010 Census questionnaires. Carter Sibley of Boston earned $7,500 for his second-place model, which came within 2.59872 percentage points. Paul Mineiro of Bellevue, Wash., and Jiri Materna of the Czech Republic teamed up for the third-place entry, so Mineiro earned $2,500. The contest rules state that prize money can go only to teams including U.S. citizens or permanent U.S. residents.

There were two entries that had a higher prediction rate than Bame's, however, the contestants did not qualify for a share of the $25,000 in total prizes.

“Interactive Visualization of Census 2010 Response Rates” was chosen as the data visualization entry that best illustrated the intent of the challenge, winning $1,000. Seth Spielman and David C. Folch, both of Boulder, Colo., and Charles R. Schmidt of Bethlehem, Pa., created this project.

“Crowdsourcing solutions to technical problems are one way we are driving innovation and greater efficiency,” said Roderick Little, the Census Bureau's associate director for research and methodology. “The Census Bureau and other federal agencies, while constantly looking for ways to innovate and be more efficient in their operations, will be using information from this challenge to determine how to allocate future data collection resources.”

The competition was conducted on Kaggle.com under Section 105 of the America COMPETES Reauthorization Act of 2011.

To view results of the modeling competition and the methodologies used by the prize-eligible winners, go to https://www.kaggle.com/c/us-census-challenge/leaderboard and https://www.kaggle.com/c/us-census-challenge/forums.