Social inequality in student achievement is a core topic in the educational sciences. Various developmental mechanisms have been proposed to underlie its emergence. First, the Matthew effect assumes a path-dependent process of cumulative advantage, in which achievement gains are a function of prior achievement. This mechanism predicts increasing inequality as initial differences amplify over time. Second, mechanisms of cumulative advantage due to exposure effects have been proposed, in which the additive effects of social class lead to higher levels of inequality. Third, it has been proposed that schooling has inequality-reducing effects that compensate for differences in developmental relevant home resources. This paper discusses the possibilities of modelling these mechanisms when only a limited number of repeated assessments are available. It illustrates the use of multilevel formulations of growth curve and simplex models, drawing on annual assessments of reading and mathematics achievement (grades 4 to 6; N=3169). The results indicated decreasing individual differences, thus refuting a key aspect of the Matthew hypothesis. Results from growth curve models were in line with the compensation hypothesis as long as initial achievement was not controlled for. Simplex models reflecting path-dependent developmental processes suggested that social background is generally unrelated to achievement growth.