Judging from the abundant and expanding literature on educational inequalities, the apparent consensus

is that divergent educational outcomes of individuals can be explained by two main mechanisms: classspecific

differences in children’s skills (primary effects) and educational choices, net of skills

(secondary effects). Contrary to the widespread agreement that primary effects stem from differences in

parental tangible and cultural resources and that secondary effects result from decisions made based on

class-specific constraints, we contend that parents across social classes invest in their children

differently for equivalent levels of children’s perceived skills—thus, primary effects and secondary

effects work in similar ways. Formulating a rational-choice model of primary effects, we hypothesize

that parental investments in children’s skills during the early stages of their life course are stratified

according to the combined effects of relative risk aversion and cumulative advantage. Using data from

the German Socio-Economic Panel (SOEP) and controlling for parental resources, we tested this

hypothesis by asking whether mothers spend more or less time engaged in cognitively stimulating

activities with their preschool children at age 5 to 6 depending on the child’s observed past language

skills and the parents’ social class. Ordinary least squares estimates suggest that mothers’ activities with

their children not only are stratified across social classes but also reflect a class-specific reaction to

children’s observed past ability consistent with our assumed mechanisms. Lower-class mothers invest

in cognitively stimulating activities in a selective way: the higher the perceived ability of the child, the

more effort is invested in these activities; in contrast, higher-class mothers do not differentiate their

investment according to their child’s observed ability. Level of maternal education and extent of child

care support in the household have an additional positive impact on levels of activity. Implications of

the results and possible extensions of the model are discussed.