The causal effect of educational tracking on student achievement: Evidence from a difference-in-difference approach with matching