Hot on the heels of a recent report that postulated a lower-than-commonly-believed economic benefit of formal schooling comes a brand new analysis that offers more proof of high marginal returns to education. Who is winning the battle of the impact estimates?
The earlier report (which we will call Clark, the name of its lead author) was a meta-analysis of 79 previous studies that the authors concluded showed evidence of two types of publication bias. These combined to suppress findings of small or null impacts of an additional year of education on individuals’ future earnings, leading to an overestimation of positive benefits. If those biases were not in place, Clark says, typical income boosts would be in the range of 0 to 3 percent. At least two of the papers cited in Clark were co-authored by Harry Anthony Patrinos and George Psacharopoulos, researchers from the University of Arkansas and the London School of Economics, respectively. Now Patrinos and Psacharopoulos have published a brand-new report that once again supports the more-commonly-found positive impacts in the 8 to 10 percent range…and more besides.
Their new paper is also a meta-analysis, comprising 191 impact estimates from 145 studies across 54 countries, including 45 of the studies also analyzed in the Clark paper. Cognizant of the findings in Clark, Patrinos and Psacharopoulos look to overcome the publication bias questions by compiling the “most comprehensive” collection of studies that report both ordinary least squares (OLS) regressions and other credibly causal estimates of economic impact to allow for comparison—especially estimates using instrumental variables (or IV) regressions. They also document and discuss the changing geographic and temporal composition of the evidence base (that is, increasing evidence from low-income and low-educational-attainment contexts) and why that matters to outcomes. Their paper, they write, “does not provide new causal estimates of the return to schooling; instead, it analyzes how different identification strategies shape the magnitude and interpretation of reported returns, documenting systematic patterns across credible designs and clarifying their economic meaning.”
By comparing the impact findings in the two types of regressions, they observe three widely-documented empirical findings in their meta-analysis. First, estimated returns to schooling are consistently positive under both OLS and causal identification strategies. Second, IV-based estimates of impact exceed OLS estimates in the majority (78 percent) of cases reviewed. Third, the magnitude of the IV–OLS gap shows systematic variation driven by income levels. Specifically, the gap is substantially larger in lower- and middle-income economies than in high-income contexts. This pattern is not driven by a small number of outliers but reflects a broad regularity across studies, instruments, and study locations. Patrinos and Psacharopoulos interpret these findings as reflecting high marginal returns in low-educational-attainment environments and the fact that IV analysis identifies local average treatment effects of specific policy changes.
If true, these findings would serve to refute the idea that publication bias is driving up impact estimates, as the Clark paper argues. Instead, they indicate that causal estimates are specifically identifying very high returns to education for first-generation secondary or post-primary completion, which is more prevalent in lower- and middle-income economies. And while other returns may be lower—some even near zero despite being in similar low-income contexts, as Clark argues—the average income gains for an additional year of formal education would still settle out in the typical 8 to 10 percent range due to the very high impacts in those specific contexts.
Patrinos and Psacharopoulos conclude that their findings not only refute Clark, but actually make an even stronger case for the value of education for disadvantaged and lower-income populations. “Well-designed education policies,” they conclude, “that expand access to high-quality schooling remain a central instrument for promoting individual prosperity and supporting long-run economic development.” They recommend future research to quantify the relative contributions of specific contextual factors such as national income level and local test score variations and expanded investigations in using credible causal designs in low-income country settings where data are currently scarce.
SOURCE: Harry Anthony Patrinos, and George Psacharopoulos, “Causal returns to education,” International Journal of Educational Development (March 2026).
