2022

Using AI to solve the Land Bank Algorithm

Country/area: United States

Organisation: Eye on Ohio, the Ohio Center for Journalism

Organisation size: Small

Publication date: 27/12/2021

Credit: Emily Crebs, Jim Crowley, Ron Calhoun, Sara Stoudt, Lucia Walinchus, and Rich Weiss

Biography:

ucia Walinchus is an award-winning journalist, attorney and ice hockey addict. She is currently the Executive Director at Eye on Ohio, the Ohio Center for Journalism. Walinchus has written more than 500 articles for various publications throughout her career and was named a 2016 Fulbright Berlin Capital Program Scholar. She has been featured as a guest speaker on CNN and is a contracted freelancer for the New York Times. By investigating police practices throughout Oklahoma, she was able to write an exposé detailing how infrequently Oklahoma police fingerprint evidence, especially in rural areas, even if possession is an issue. For another story, she analyzed thousands of records to determine that Oklahoma City Landlords win 95 percent of contested cases. She recently teamed up with the Pulitzer Center to show how a tax loophole raises property tax rates for small business owners, and spearheaded a major investigation with the Cincinnati Enquirer that showed African American neighborhoods have far more stops than white ones. Walinchus has a degree in Journalism from American University and a Juris Doctorate from California Western School of Law.

Emily Crebs is a junior studying journalism through the Honors Tutorial College at Ohio University. At Ohio U, Emily is the editor-in-chief of the online student-run publication The New Political. She has also worked as a staff writer and news editor at The New Political where she has written in-depth features and investigative stories. After college, Emily hopes to work as an investigative reporter.

Sara A. Stoudt, Ph.D joined the Bucknell University Mathematics Department as a tenure-track Assistant Professor in the fall of 2021. She received her PhD in statistics from the University of California, Berkeley where she was also a Berkeley Institute for Data Science Fellow. Her research focus is on ecological applications of statistics and statistics communication. At Berkeley she was advised by Will Fithian and Perry de Valpine and taught writing for statistics with Deb Nolan. Previously, she received a B.A. in Mathematics from Smith College with an emphasis on Statistics. Check out Deborah Nolan and my new book, “Communicating with Data: The Art of Writing for Data Science”

 

Project description:

Using machine learning methods, Eye on Ohio looked at property remediation in several counties to look deeper at a process that has transformed the rust belt over several years. 

Certain factors such as proximity to valuable properties or the race of the majority of students in a school district made a property more likely to be picked for the land bank in some counties.

And in certain areas, officials responsible for economic recovery are the same people in charge of that remediation. 

 

Impact reached:

According to the National Land Bank Network at the Center for Community Progress, there are over 200 land banks nationwide. Eighty-two of those are Ohio county land banks, and several Ohio cities have land banks as well.

Over the years, the legislature has been gradually expanding powers to land banks. Under an Ohio Revised Code Section 294’s “expedited administrative tax foreclosure,” Board of Revision officials theoretically only hear foreclosure cases where the properties are abandoned and there are no other legal issues. But we found  in most small counties, BOR members bring the cases in the first place, and later decide who gets the properties. That’s because of overlap between city officials, the BOR, and the Land Bank.

Unlike a traditional levy, it’s the land bank board, not voters, who determine how much money goes to the land bank.

Is diverting $500K+ from schools worth it for ~115 properties? Until this project, most people didn’t even know the details of how this worked.

Each land bank has a policy that essentially says “we do the best for the community with what we have.” What does that mean, mathematically? This is the first project that has shown the inherent conflicts of interest built into the system have affected those decisions.

This just published so it’s too early to tell the full impact, but this project has been making waves among our readers, listeners, and viewers.

Land banks are supposed to “clean up communities” and for the most part they do that. But we found they could also be away for public officials to deny applicants in choice spots or to deny political rivals.

Techniques/technologies used:

This is the culmination of months of work, over 5,000 lines of code, hundreds of public records requests, and several trips to counties all over the state to see how this works in practice.

This project is an example of some of the really cool things R does: machine learning! HTML maps! Geolocating stuff!

The st_distance() function for Cuyahoga county alone produces a matrix that’s 11 GIGS!

We used Github, Google Sheets, and Google maps for version control and to flag suspect problems. We used RStudio to create Rmd scripts, and within RStudio we used many built-in packages to pull in data, read databases, clean data, combine data, map parcels, etc.

For a detailed explanation of each step, please see our github repository.

What was the hardest part of this project?

The hardest part, conceptually, was to think about how location affected decisions. What does it even mean to be in a good location? The school district? What if the county has open enrollment? Is it location to amenities? Does a 5 or ten minute drive matter or overall distance?

By picking out spots that our maps showed might be interesting, and by talking to people near those properties, we were able to get a sense that it was distance to several key locations that mattered.

The hardest part, logstically, was putting data from six different counties in the same format- some data was missing or wrong, too. That was the most frustrating and time-consuming.

What can others learn from this project?

They can learn a lot- this project is open source!

We started with property because there’s a lot of great parcel data out there and it’s all public records.

This is a technology that the government and thousands of companies are using to save money. What if we looked beyond that? You could also apply it to lots of decisions that companies and governments make about reader’s lives. There are endless possibilites to take this project and apply it to other public records: parole maybe? Medicare/Medicaid?

Equally important to study and write about: How is AI already making decisions about our lives and are those decisions fair?  It’s important to have a firm grasp on this process so you can spot flaws.

Project links:

eyeonohio.com/how-do-public-officials-make-land-bank-decisions-artificial-intelligence-may-seek-patterns/

eyeonohio.com/plaintiff-defendant-and-judge-how-some-ohio-counties-entrust-the-same-officials-to-collect-taxes-and-wipe-tax-liens/

eyeonohio.com/sidebar-what-is-artificial-intelligence-and-why-use-it-to-look-at-public-records/