Catenaa, Thursday, November 13, 2025-Researchers led by Lehigh University statistician Taeho Kim have unveiled the Maximum Agreement Linear Predictor (MALP), a method designed to produce predictions that better align with actual outcomes.
Unlike traditional approaches that minimize average error, MALP maximizes the Concordance Correlation Coefficient (CCC), which measures how closely predicted and observed data align along a 45-degree line.
The team tested MALP using both simulations and real-world datasets, including ophthalmology scans and body fat measurements.
In an eye study comparing older Stratus OCT devices with newer Cirrus OCT systems, MALP more accurately predicted Stratus readings from Cirrus data than the standard least-squares method, demonstrating improved agreement despite slightly higher average error.
Similarly, MALP provided better alignment with actual body fat percentages compared to conventional predictive methods.
Kim explained that MALP is particularly useful in contexts where the goal is not just closeness but agreement between predicted and actual values.
The approach combines precision, how tightly predictions cluster, with accuracy, how closely they follow the ideal reference line. By focusing on alignment rather than correlation alone, MALP offers a more robust framework for applications in health, biology, social sciences, and other fields relying on predictive modeling.
While traditional least-squares methods remain valuable for error reduction, MALP offers a complementary tool when agreement is paramount. The researchers plan to extend MALP beyond linear predictors to create a broader Maximum Agreement Predictor applicable across more complex datasets.
