Masked identity priming reflects an encoding advantage in developing readers.

The masked priming technique is widely used to explore the early moments of letter and word identification. Although this technique is increasingly used in experiments with young readers, the mechanism in play during masked priming with early readers has not yet been fully explored. We investigated the masked priming effects from a modeling perspective; we instantiated competing theories as data models (using Bayes factors) and as a computational model (diffusion model). We carried out a masked priming experiment using identity primes with second- and fourth-grade participants, and we analyzed the data through an evidence accumulation model lens. The priming effect manifests as a shift in the response time distribution, which in evidence accumulation models is accounted for by changes in the encoding process. We describe such changes as savings that have three features of theoretical importance. First, they are numerically very close to the stimulus onset asynchrony between primes and targets. Second, they remain relatively constant from second grade to fourth grade. Third, they seem to operate at the level of abstract orthographic representation because the priming effect occurs in both case-matched and case-mismatched pairs. These findings also have consequences for the practice of data transformation in developmental research; some patterns of data, when transformed, would produce spurious effects.