Early literacy skills are critical to the educational outcomes of young students. Accordingly, 19 states have early literacy policies that require grade retention for underperforming readers at the end of third grade. However, there is mixed evidence about retention's effectiveness and concerns that retention may disproportionately impact traditionally disadvantaged student groups. Using regressions and a regression discontinuity design, we examine retention outcomes under Michigan's early literacy law, the Read by Grade Three Law. We find that Black and economically disadvantaged students are more frequently eligible for retention and retained than their peers. While controlling for students' test performance, particularly their math scores, eliminates this disparity for Black students, it persists for economically disadvantaged students. We show that differences in average math performance, exemption characteristics, district characteristics, and eligibility-induced student mobility across districts do not explain the disparities in the implementation of retention by economic disadvantage status.