Mathematics Partial Associations with Subject-Specific Science Confidence: Evidence from 11 Education Systems in TIMSS 2023

Source abstract

Dimensional comparison theory holds that achievement in one school subject is positively associated with self-concept in that subject and negatively associated with self-concept in dissimilar subjects. Applications within the science domain have characterized biology, chemistry, physics, and earth science relative to one another while omitting mathematics from the estimated model. The present study estimated the partial association of mathematics achievement with each of four subject-specific science confidence scales, conditional on achievement in all four science content domains, across the 11 TIMSS 2023 Grade 8 education systems administering subject-specific questionnaires. Because the achievement indicators are strongly intercorrelated, each coefficient is interpreted as an association with the residual variation in mathematics achievement that survives the conditioning, which retains 30% to 52% of its variance by system, and not as a measure of mathematics-specific competence. Zero-order associations were positive for all four subjects; conditioning on science achievement reversed the sign for biology under every parameterization examined. The conditional association was negative for biology confidence (β = −0.158, 95% CI [−0.224, −0.093], Holm-adjusted p = 0.001) and positive for physics confidence (β = 0.093, 95% CI [0.041, 0.146], Holm-adjusted p = 0.009). The chemistry association was positive but specification-dependent (β = 0.066, p = 0.035) and is reported as inconclusive, and the earth science association was not distinguishable from zero. Forming the change in each within-science cross-path inside every replicate and plausible value, both prespecified physics–chemistry paths were attenuated by the mathematics anchor (Δ = −0.027, unadjusted p = 0.012, Holm-adjusted p = 0.024; Δ = −0.014, unadjusted p = 0.040, Holm-adjusted p = 0.040), while other cross-paths were amplified or reversed; a covariance-matched placebo anchor established, algebraically and numerically, that the expected mechanical attenuation is zero. Forward and reverse associations did not differ, though the design excludes only asymmetries larger than about 0.08 to 0.13. Between-system heterogeneity was the most consistent finding (I2 = 56–87%): prediction intervals included zero for all four associations, and while biology was the most negative subject in 11 of 11 systems and the biology–physics contrast held in direction in every system, the complete four-subject ordering recurred in 7 of 11 and had a probability above 0.50 in only 4. Subject-specific science confidence relates heterogeneously to mathematics achievement, and aggregation across science subjects can obscure associations of opposing sign.