Are People Getting "Dumber"? New U.S. Data Point to a Reverse Flynn Effect

Just over a century ago, the Binet IQ test was developed in France and soon after brought to the United States, where it was administered to schoolchildren, physicians and immigrants arriving at Ellis Island. Throughout the 20th century, average measured intelligence rose steadily — a phenomenon known as the "Flynn Effect." Researchers estimated roughly a three-point gain per decade across many populations and testing instruments.

Recent research, however, suggests that trend may be reversing in some domains. Social scientist Elizabeth Dworak and colleagues analyzed responses from 394,378 American adults (ages 18–90) collected via the SAPA Project from 2006 to 2018 using items from the open-source International Cognitive Ability Resource (ICAR).

What They Measured

Rather than relying on a single composite score, the team examined several cognitive domains separately: matrix reasoning (abstract pattern recognition and problem solving), letter-and-number series (sequential logical reasoning), verbal reasoning (vocabulary and general knowledge), and three-dimensional rotation (mental spatial visualization). They also calculated composite scores across tasks.

Key Findings

The analysis found population-level declines in composite cognitive performance consistent with a "reverse Flynn effect." Composite scores fell across age groups, sexes and education levels. The largest drops occurred among adults aged 18–22 and among participants with lower formal education.

Declines were not uniform across domains. Matrix reasoning and letter-and-number series scores generally declined along with the composite, indicating weaker performance on abstract and logical-reasoning tasks. Verbal reasoning changed little and did not meet the study's threshold for a meaningful decline. By contrast, three-dimensional rotation scores — a measure of spatial ability — improved over the study period, a pattern some commentators have tentatively linked to greater exposure to video games and spatially demanding digital media.

"I spent weeks going back through all the code. I thought I'd messed something up and would have to delay submitting. But then I showed my adviser, and he said, 'Nope, your math is right.'" — Elizabeth Dworak, as quoted in New York Magazine

Context And Caveats

Other indicators line up with these findings: ACT and SAT average scores have declined in recent years. Importantly, the observed downturn in the ICAR data began before widespread access to generative AI tools and predates the COVID-19 pandemic, so neither factor alone explains the trend. The late 2000s — when the decline becomes visible — also coincide with the rapid rise of smartphones and a major shift in reading and media habits. Some observers, like Times columnist James Marriott, argue that a move away from dense print reading toward short-form digital content may encourage different modes of thinking.

Crucially, Dworak's study is based on repeated cross-sectional data, not longitudinal tracking of the same individuals. That design documents population-level shifts but cannot identify causes or rule out sample or measurement effects. The authors and other experts emphasize that the pattern is complex and domain-specific: some cognitive skills may be weakening while others improve, which could reflect cultural adaptation rather than a simple loss of intelligence.

Implications

The findings open important questions for education, workforce planning and public health. If abstract and logical-reasoning skills are declining while spatial abilities rise, educators may need to reconsider curricula and training priorities. The study was published in the journal Intelligence, and the authors call for more research using diverse, representative samples and longitudinal methods to understand causes and consequences.

Bottom line: Measured cognitive performance in the U.S. sample declined in some domains between 2006 and 2018, rose in others, and the causes remain uncertain — pointing to a nuanced, domain-specific shift rather than an all-or-nothing change in human intelligence.