Connect with us

Hi, what are you looking for?

Policy

A State Uses AI to Determine School Funding. Is This the Future or a Cautionary Tale?

Renee Fairless runs three K-8 public charter schools in Nevada that enroll a total of nearly 4,000 students. Almost all of the students are English learners, economically disadvantaged, or both.

Until recently, the state each year provided the Mater Academy of Nevada schools with a base amount of roughly $7,000 per student, as well as an extra dose of funding for roughly 1,700 students the state categorized as “at risk.”

This school year, Fairless’ funding haul looked a little different: Only 45 students qualified as at risk and, therefore, were eligible for the additional aid.

Now Fairless is pondering cuts to services like subsidized college classes for high schoolers and after-school tutoring for struggling students. “We don’t know where we’re going to get the money,” she said.

Fairless’ experience isn’t unique. The number of students across Nevada who are eligible for at-risk aid dropped from 288,000 for the 2022-23 school year to 63,000 for the current school year.

That means nearly 75 percent of students deemed at risk in 2022 lost that distinction the following year, meaning their districts lost the targeted aid that accompanied it. The share of Nevada’s 485,000 K-12 students for whom schools were receiving at-risk aid dropped from nearly 60 percent to just 13 percent.

The drop is even more acute in some individual districts and schools. One charter leader said last week during a meeting of the state Commission on School Funding that only three of his school’s 800 students qualified for at-risk aid this year—and all three have already graduated, because the formula lags a year behind current data.

Tens of thousands of children in Nevada didn’t become dramatically less needy—and, as a result, less expensive to educate—overnight.

Rather, as part of a last-minute revision to a school funding bill signed by the governor in 2023, the state launched an effort to change the formula that determines the number of at-risk students in each of the state’s 750 traditional public and public charter schools.

State officials say the previous calculations dramatically overcounted the number of at-risk students who would most benefit from additional funding. They say they aren’t surprised the new number is much lower, and they’re happy they can target more available dollars per pupil to the students who stand to benefit the most.

But education advocates and some school leaders in the state are raising alarms about the effects of the state’s attempt at a more precise approach to targeting funding to at-risk students. In recent months, they’ve sought more transparency from the state education department and a third-party contractor about how the formula functions, how it was developed and what role machine-learning technology played, and what it means.

“We really need to make certain that we right-size this funding so we can adequately provide services to students without taking resources away from students who don’t need the same services,” said Paul Johnson, chief financial officer for the 1,200-student White Pine district in the rural northeastern part of the state. “It’s quite the conundrum.”

These challenges are specific to Nevada for now, but they have nationwide implications. Nevada is offering the nation a cautionary tale about the perils, both perennial and contemporary, of crafting a funding formula that effectively supports students with varying needs—and makes sense to the people it affects.

Reworking a school funding formula is naturally a fraught undertaking, without relying on a new technology.

But artificial intelligence may be the latest tool states use to assist with school funding formulas that opens up more complications than the problems it solves, said David Knight, assistant professor of education finance and policy at the University of Washington College of Education.

“A major strength of any good school finance formula is to have something that’s transparent and clear,” Knight said. “We need the public to understand the funding model because we need them to support it.”

Nevada uses AI to predict the likelihood a student will or won’t graduate

The new source of data for determining the number of at-risk students represents a significant departure from the typical methods states use to assess funding needs. Many of those methods have been in place for decades with few adjustments.

Nevada developed the new approach to calculating aid for at-risk students using a digital tool from Infinite Campus, a technology company that sells student information systems and other data repository tools to hundreds of school districts nationwide.

Nevada has contracted with the company since 2016. It’s one of six states where every district uses the Infinite Campus platform to keep track of students’ attendance, behavior, and grades, among other details. The other states are Delaware, Kentucky, North Carolina, South Dakota, and Hawaii, which has only one school district.

The company’s “early-warning system,” comparable to others that schools have been using for years, employs a machine-learning algorithm to assess the likelihood that each student whose data enters the system will or will not graduate.

“It seemed to address a lot of the areas that we thought were important to a student’s makeup of who they are, what it is to actually succeed,” Megan Peterson, deputy superintendent for the Nevada state education department’s student investment division, told Education Week in an interview this month. “We’re trying to ensure every student has the skills to meet what their vision of success is.”

The algorithm makes its predictions based on an assessment of 150 million student years’ worth of data on students’ academic performance, attendance, behavior, demographics, and family characteristics, said Charlie Kratsch, the founder and CEO of Minnesota-based Infinite Campus.

The assessment weighs 75 factors including test scores, household size, number of documented behavior infractions, and even the number of times students and parents log into their school’s platform for grades.

Each student in the system has a “grad score” the algorithm assigns—a single number “like a credit score” between 50 and 150 that fluctuates as often as every day, Kratsch said.

Districts and schools in several states use grad scores to identify individual students who would benefit from special attention—those on the lower end of the spectrum.

Nevada uses them for a different purpose: Each Oct. 1, all students that the system slots into the “high risk” and “medium risk” categories qualify for additional state aid for their districts—35 cents extra for every dollar they get through the state’s base per-pupil funding.

The Infinite Campus system considers students who fall below the fifth percentile of all the scores in a sample to be at high risk for failing to graduate, Kratsch told Education Week. Students between the fifth and 20th percentiles are considered medium-risk, Kratsch said.

States and districts are welcome to make different judgments about which scores constitute different risk levels, he said. Infinite Campus doesn’t get involved in what a state or district chooses to do with the grad scores at their disposal.

“School finance is a sticky wicket. We are not consultants in that area. We’re not qualified to go in and recommend where they would cut that line,” Kratsch said. “We build the tool, we create the number, and we make it as accurate as we possibly can.”

But Nevada ended up sticking with the percentiles laid out by Infinite Campus.

The 63,000 students who fell below the 20th percentile had grad scores of 72 or lower, Peterson told Education Week. All of those students qualify for weighted at-risk funding from the state—$198.7 million in total, or $3,137 per student, beyond the base state aid of $7,073 per pupil.

Nevada excludes students with disabilities and English learners from its count of at-risk students. Those students instead receive additional funding at a different weight than at-risk students, even if they would otherwise qualify for at-risk aid.

Knight believes a better approach to using the grad score as a basis for funding would have been to gradually taper funding weights as students’ scores rise, rather than setting a threshold at which districts receive funding for an at-risk student and above which they don’t.

“The whole problem with FRL that we hate is that it’s binary,” Knight said of using students’ eligibility for free and reduced-price lunch as a factor in funding formulas.

The new formula is no different in that regard.

With new technology comes new challenges and risks

The new formula in Nevada relies on technology that even its developers describe as a work in constant progress. It could foreshadow a broader shift toward using artificial intelligence tools to shape funding decisions.

Nevada is the first state to use the Infinite Campus early-warning system as the basis for allocating state resources, though other states have expressed interest, Kratsch said.

“I would say that when a state successfully rolls out something like this and can show the value of it, history shows that many other states tend to follow suit,” he said.

Ryan Watkins, a professor of educational technology at George Washington University, reviewed for Education Week a 2019 Infinite Campus white paper that breaks down how the early-warning system works. He said the approach to collecting and synthesizing data largely appears sound and in keeping with best practices.

“If I was a data scientist, I would probably make many of the decisions that they made in trying to set something like this up,” Watkins said.

Even so, the approach is not without detractors.

An early-warning system in Wisconsin that developed a score meant to assess a student’s likelihood of completing high school on time drew criticism from researchers and scrutiny from reporters. A 2023 investigation by The Markup, a nonprofit technology news organization, found that the system was wrong about students of color far more often than it was wrong about white students.

More recently, Nevada school funding commission members said they’ve heard from school leaders who believe the Infinite Campus algorithm gives lower scores to girls than boys with otherwise comparable data.

The Infinite Campus team has shown Nevada figures indicating that its grad score predictions are accurate roughly 95 percent of the time.

That said, the quality of the grad score’s prediction varies slightly depending on several factors, Kratsch said.

It’s more likely to accurately predict the trajectory of a 9th grader who’s been in the same state since kindergarten than that of a 3rd grader with less data, Kratsch said. And it’s more likely to struggle to predict the graduation prospects for an 11th grader who just arrived from another state that doesn’t put data into the Infinite Campus platform, Kratsch said.

“The wider you use the tool, the more that’s factored out—any individual student is going to have less impact over the larger population,” he said.

Infinite Campus data scientists retrain and update the grad score algorithm roughly once a year.

For the next revision cycle, the company is pondering removing students’ race and gender from the list of 75 factors that play a role in determining a student’s grad score, Kratsch said.

Nevada is still figuring out how to target a limited pot of state aid

Even the most fervent critics of the new approach to calculating at-risk funding in Nevada agree that previous approaches were far from perfect.

A wide body of academic research shows that money matters for improving student outcomes in schools. Researchers have also shown that schools need to spend more than the base amount on some disadvantaged students—such as students in poverty, English learners, and students with disabilities— in order for them to achieve outcomes comparable with their peers.

The state used to provide additional aid to schools that had a large share of students in vulnerable populations. Then, when the state transformed its entire school funding formula in 2019, all students eligible for free and reduced-price meals were considered “at risk” by the state.

But that number has become less helpful over time, state officials have said, because of the federal Community Eligibility Provision, which affords the option of free and reduced-price meals to all students in a school where more than 25 percent qualify for the benefit, down recently from the previous threshold of 40 percent.

Preliminary figures the state has reviewed suggest the number of students eligible for free and reduced-price meals next year will be 380,000, or more than 78 percent of the state’s K-12 population.

Knight said most states still use the simple measure of free and reduced-price lunch eligibility, even as it becomes less useful. Even with his reservations about the grad score model, “I would applaud the state for trying to experiment with need-based funding,” he said.

State officials say they’re pleased to see that the number of students deemed at risk is now closer to the annual share of roughly 19 percent of Nevada students who have failed to graduate on time in recent years. With the allocated funding from the legislature, that number maximizes the number of dollars per pupil who most need the investment, Peterson said.

“When we [use the free and reduced-price lunch eligibility count], we’re less than $200 to $300 per pupil in being able to serve children,” Peterson said. “There’s nothing we can provide a student in terms of services with $200 a student.”

Some observers see it differently. Mark Mathers, chief financial officer of the Washoe County schools, said last week that the alignment of those two figures concerns him. “If 20 percent of kids were at risk and 20 percent don’t graduate, then we’ve done nothing to help those kids,” he said.

School leaders have raised other objections as well. Dusty Casey, executive director of Oasis Academy, a K-12 charter school, said the vast majority of his school’s at-risk funding went to his elementary students. “I truly don’t believe all of our at-risk students are in elementary school,” Casey said.

State and company officials have acknowledged school leaders’ concerns and said they plan to continue to adjust their practices in the coming years.

Underlying all of these thorny questions is the state’s persistent underfunding of K-12 education more broadly, Jason Goudie, chief financial officer for the Clark County school district, said during the commission meeting. “If our per-pupil base was $20,000, I bet you we would not be having this discussion.”

Without robust state funding that keeps up with inflation, though, attention shifts to the definition of an at-risk student.

If “at-risk” were redefined in the state to align with the poverty level, the result might be different. Just shy of 16 percent of Nevada’s school-age children are in families living in poverty, which translates to closer to 80,000 children.

At the Mater Academy of Nevada, students routinely clean out baskets of fruit and muffins staff put out for them, and some report not having regular access to a washer and dryer at home, Fairless said.

She believes her school deserves the resources that the at-risk funding stream previously provided—but she’s less confident that “at risk” is the right phrase to use.

“I don’t know that I would want anybody to look at a 4th grader, 5th grader, and already be categorizing them as a student we fear would drop out,” she said.

There’s little empirical evidence to show the amount of additional funding necessary to ensure students in marginalized groups have access to the same educational opportunities as higher-income peers, leaving states flying blind when developing funding formulas, Knight said.

“There is zero research on how much more it costs to educate a kid with a grad score of 72 versus a grad score of whatever else,” he said.

The fundamental question, according to Paul Johnson, the White Pines district CFO, is about the role schools are expected to play. Should they be narrowly focused on providing instruction, or are they responsible for filling gaps left by spotty social services?

“What are the risks that we are supposed to mitigate as an education system? Is it purely a risk of not graduating, or is it deeper than that? Is it social-emotional-academic development? Is it behavioral issues with students?” Johnson said. “In my opinion, it’s a little bit deeper than being at risk of not graduating.”

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

You May Also Like

News

The University of Southern California, reeling after a controversy over its valedictorian selection, announced Friday that its main commencement program would eliminate outside speakers...

News

In recent decades, American universities have expanded their diversity programs to address concerns about the underrepresentation of minority groups on campus. But over the...

News

It’s a nightmare scenario for Democrats: Protesters disrupt their convention this summer; they clash with the police; chaos seems to take hold. It may...

News

As a throwback ski destination, Alta, thinks small, with a one-room public school to match. WHY WE’RE HERE We’re exploring how America defines itself...