Probing gaps in educational outcomes within the U.S.: A dual moderation multiple mediator latent growth model.

Racial/ethnic disparities in math achievement are especially troubling because math proficiency predicts long-term educational outcomes, but the mechanisms underlying these disparities remain unclear. Previous research has demonstrated that across diverse samples, both within and outside the United States, the relation between students' academic aspirations and later postsecondary attainment is mediated by initial levels of math ability and by growth in that ability across time. The key issue examined in this investigation is the extent to which students' underestimation or overestimation of their math ability (i.e., calibration bias) moderates those mediated effects and whether this moderation varies as a function of race/ethnicity. Using data from two longitudinal national surveys (i.e., NELS:88 and HSLS:09), these hypotheses were tested in samples of East Asian American, Mexican American, and Non-Hispanic White American high school students. In both studies and in all groups, the model explained large portions of the variance in postsecondary attainment. In East Asian Americans and non-Hispanic White Americans, calibration bias moderated the effect mediated by 9th grade math achievement. The strength of this effect was greatest at high levels of underconfidence and steadily weakened as self-confidence grew, suggesting that some degree of underconfidence may be achievement-promoting. Indeed, in the East Asian American sample, this effect became negative at high levels of overconfidence (i.e., academic aspirations actually predicted the lowest postsecondary attainment levels). Educational implications of these findings are discussed and possible reasons for the failure to find moderation effects in the Mexican American sample are explored.