Kids' test scores began declining way before COVID. These schools are making gains
The pandemic-era backslide in math and reading scores for students across the U.S. was not a sudden catastrophe but the continuation of a brutal, decade-long "learning recession" that began years before COVID-19's arrival. That's according to the latest Education Scorecard, an annual deep-dive into student data from The Educational Opportunity Project at Stanford University and Harvard University's Center for Education Policy Research.
The new Scorecard, released Wednesday and in its fourth year, offers several revelations for families, educators and policymakers looking for clarity — and hope — at a time when public education has been blamed and battered for those persistent declines in student performance.
Among the report's takeaways: Most states are finally making gains in math; federal relief dollars likely helped the lowest-income districts mount a hearty comeback; and, while most states have yet to make gains in reading, those that have all made legislative changes to how it's taught in their schools.
Before we dive in, one caveat: The annual Education Scorecard includes data from the vast majority of states and Washington D.C. drawn from their own state tests — as opposed to the Nation's Report Card. But some states were excluded for various reasons, including if their state assessments had changed recently (Illinois, Kansas), if test opt-out rates were too high (New York, Colorado) or if a state didn't publish district-level data with enough detail.
'The learning recession'
For nearly a quarter-century, from 1990 to 2013, math achievement among fourth- and eighth-graders "rose steadily," according to the Scorecard's analysis. So steadily that "the average fourth grader in 2013 could perform the same math skills as the average sixth grader could in 1990. That's enormous progress," says Stanford University's Sean Reardon, one of the Scorecard's authors.
Reading gains weren't quite as eye-popping, but they were gains nonetheless.
These sustained gains "may be one of the most important social policy successes of the last half-century that nobody knows about," says Harvard's Thomas Kane, one of the Scorecard's authors. "Racial gaps were narrowing too. We just need to get back on that track."
In short, much was right with America's schools, which makes the decline that began around 2013 "appear more striking and anomalous," the report says.
"Particularly in reading, test scores were going down for four to six years before the pandemic," says Reardon. "In fact, you wouldn't really know there was a pandemic effect if you just looked at the last 10 or 12 years of test scores. There's been just a steady kind of decline regardless of the pandemic."
What might have triggered that decline?
The Scorecard's trigger theories
Scorecard researchers offer two possible explanations for the beginning of schools' learning recession:
1. The fade-out of test-based accountability: Remember the much-maligned federal education law, No Child Left Behind (NCLB), that took a tough-love approach with schools to improve student performance? The law, implemented in 2003, threatened a host of sanctions, including school closure, if student test scores didn't rise, but its standards were seen by many to be not just unrealistic but unattainable. By 2013, the Obama administration began issuing waivers to free states from the law's consequences. According to the Scorecard, 38 states were granted relief in the 2012-13 school year. Eventually, Congress replaced NCLB with a new federal law that de-emphasized test-based accountability.
Around 2013, Kane says, "school districts learned that nobody was looking over their shoulders in terms of student achievement."
While the Scorecard researchers don't draw a direct, causal connection between the declines of test-based accountability and student scores, it's clear that the nation's learning recession began at roughly the same time states and schools stepped back from the punishing consequences of NCLB.
2. Students' social media use: It turns out, 2013 also marks a period of explosive growth in teenagers use of social media. A Pew Research study found that in 2014-15, roughly 1 in 4 teens said they used the internet "almost constantly." By 2022, it was nearly half of teens.
The researchers also point to international testing data that shows that lower-achieving students are the heaviest users of social media. Students who spend more time (7+ hours per day) on social media score below students who spend less (1-3 hours). And this gap, between the highest and lowest performers, began growing before the pandemic, not just in the U.S. but in many other countries too.
The end of the learning recession?
The Scorecard devotes considerable analysis to what's been happening in schools since the end of the pandemic, from 2022 through the spring of 2025. There are signs that the nation's learning recession may be turning around, albeit slowly.
In that span of time, most of the states covered by this year's Scorecard showed students making meaningful improvement in math, with Washington D.C. coming in as the clear winner there. Only five states failed to make gains in math: Georgia, Idaho, Wyoming, Nebraska and Iowa.
Reading, though, remains a cause for concern. While D.C., Louisiana, Maryland and five other states did experience meaningful improvement between 2022 and 2025, most states continued to stagnate or, as in Florida, Arizona and Nebraska, further declined.
It's also worth noting, while schools are once again, on average, regaining ground in math and slowly turning the corner in reading, the declines that began around 2013 have been so steep and lasting that only one state, Louisiana, has returned to 2019 performance levels in both subjects.
No state has returned to 2013 levels, according to Reardon.
"It's easy to be sort of doom and gloom," he adds, "but when you look at the period from the '90s through 2013, we made enormous gains. And we actually narrowed achievement gaps between racial groups. That says we can actually improve our schools in ways that also improve equality of opportunity. We just haven't been doing it for the last decade. But we could do it again."
The U-shaped recovery
The Scorecard reveals a fascinating phenomenon in schools from 2022 to 2025: a U-shaped recovery. Meaning, schools with the least amount of poverty, alongside schools with the most poverty, saw similar gains in math and similarly small losses in reading achievement. That's while the schools in the middle of the income spectrum, at the bottom of this U, improved the least in both subjects.
Why? One theory is that the highest-poverty districts got the most help from Congress in the form of federal COVID relief dollars — money they could spend on interventions such as tutoring and summer school. Districts with the lowest poverty rates got little help from the federal government but were already well-positioned financially. It was the middle-income districts that needed more help but didn't qualify for full federal support.
"If it hadn't been for the federal pandemic relief," says Kane, "we estimate there would have been no recovery on average for the highest-poverty districts."
The science of reading effect
There's been an important wild card in the effort to improve students' reading skills: A movement among states to change their approach to teaching reading to young children by embracing the "science of reading." As of March, the Scorecard says, most states had passed new literacy laws, including doubling down on the importance of teaching phonics.
The Scorecard authors note that all seven of the states (plus D.C.) that saw reading gains between 2022 and 2025 had put comprehensive science of reading reforms into place. Of the states that had not by January 2024, none saw improvement. The connection between these reforms and improved results isn't necessarily causal, they warn, but there's clearly a link.
With most states struggling to make reading gains, one district-level success story highlighted by the Scorecard stands out: Baltimore City Public Schools. In spite of the challenges posed by poverty — most students there qualify for free or reduced-price meals — Baltimore students have been making striking reading gains.
Under CEO Sonja Brookins Santelises, the district reformed its approach to literacy. It embraced the science of reading even before the pandemic and years ahead of the national wave of state-based literacy legislation.
When Brookins Santelises took the lead in Baltimore in 2016, she says she quickly embraced the science of reading districtwide and its emphasis on phonics, as opposed to the whole language approach, which teaches children to guess at words using cues from a text's pictures.
"I remember gathering the [district's] literacy department. And I said, 'If you want to do whole language, there are other districts in Maryland that are doing whole language, and you are free to go there. We are not doing that in Baltimore City. I respect you, but you cannot stay here. I've been ferocious about it ever since."
'Kiss your brains!'
The benefits of these changes appear to have been twofold. During the pandemic, the Scorecard shows Baltimore schools lost far less ground in reading than schools with similar levels of poverty. Then, in 2022, with those practices firmly in place, the city's reading scores began to skyrocket, erasing pandemic-era losses and rising back around 2017 levels.
Baltimore's successful approach to teaching literacy was on full display on a recent May morning, in veteran teacher Kimberly Lowery's kindergarten class at Johnston Square Elementary. Lowery sat at the front of a rainbow-colored reading rug, running through a series of phonics-based games that her kindergarteners seemed to genuinely enjoy.
There was letter-sound bingo, guess-the-sound flashcards and even a visit from a special spelling helper — a toy owl, named Echo, who lives at the end of a yardstick. If the kids' laughter and cheering isn't sign enough that they're learning, district data shows that, by the end of last year, three-quarters of Lowery's students were reading at or above grade level.
Lowery told the children to kiss their brains and asked, "You guys are super-duper what?"
In unison, the children hollered, "Smart!"
"Yes you are," Lowery answered.
Edited by: Nirvi Shah and Steve Drummond
Visual design and development by: LA Johnson
Facts Only
The Education Scorecard, released by Stanford and Harvard researchers, analyzes student performance data from most U.S. states and Washington D.C.
Math and reading scores for U.S. students began declining around 2013, marking the start of a "learning recession."
From 1990 to 2013, math achievement among fourth- and eighth-graders rose steadily, with fourth-graders in 2013 performing at sixth-grade 1990 levels.
Reading scores also improved during this period, though not as dramatically as math.
The decline in scores coincides with the relaxation of test-based accountability under No Child Left Behind (NCLB), with 38 states granted waivers in 2012-13.
Teen social media use surged around 2013, with nearly half of teens reporting "almost constant" internet use by 2022.
From 2022 to 2025, most states showed gains in math, with Washington D.C. leading, while reading scores stagnated or declined in many states.
Federal COVID relief funds helped high-poverty districts recover, while middle-income districts saw less improvement.
States adopting "science of reading" reforms, emphasizing phonics, saw reading gains, with Baltimore City Public Schools as a notable example.
No state has returned to 2013 performance levels in math or reading.
Baltimore's literacy reforms, implemented before the pandemic, included a shift from whole language to phonics-based instruction.
Baltimore's reading scores have rebounded to around 2017 levels, with three-quarters of kindergarteners in one class reading at or above grade level by 2025.
Executive Summary
Full Take
The Education Scorecard presents a nuanced picture of U.S. education, revealing both progress and persistent challenges. The strongest version of this narrative acknowledges the significant gains made from 1990 to 2013, particularly in math, and the subsequent decline that began well before the pandemic. The report rightly highlights the potential role of policy shifts, such as the relaxation of NCLB accountability, and societal changes, like the rise of social media, in explaining the downturn. However, it stops short of establishing direct causality, which is appropriate given the complexity of educational outcomes.
The U-shaped recovery pattern, where high- and low-poverty districts saw similar gains while middle-income districts lagged, suggests that federal relief funds played a crucial role in supporting the most vulnerable schools. This raises questions about the long-term sustainability of such gains without continued targeted funding. The emphasis on the "science of reading" reforms is compelling, particularly with Baltimore's success story, but it's important to note that correlation does not equal causation. The report's cautious language on this point is commendable.
The broader implications of this analysis are significant. If the learning recession is indeed tied to policy and societal shifts, it suggests that educational outcomes are deeply intertwined with broader systemic factors. The report's focus on equity—highlighting both the narrowing of racial gaps during the progress years and the current disparities—underscores the need for policies that address structural inequalities.
Bridge questions to consider: What other factors, beyond accountability and social media, might explain the 2013 decline? How can the successes of high-poverty districts like Baltimore be scaled effectively? What role should federal funding play in sustaining educational recovery?
Counterstrike scan: If this narrative were part of a coordinated campaign, it might emphasize the failures of public education to undermine trust in institutions or push for privatization. However, the report's balanced approach, acknowledging both progress and challenges, does not align with such a pattern. It presents a credible, evidence-based analysis without overt manipulation.
Patterns detected: none
Sentinel — Human
LIKELY_HUMAN (confidence: 0.25)
