Development and initial psychometric calibration of a screening instrument for receptive-analytical skills in critical thinking: TKM-Chem for Kazakhstani pre-service chemistry teachers

Teacher education in Kazakhstan lacks validated domain-specific instruments for assessing critical thinking. This study reports the development and initial psychometric calibration of TKM-Chem, a screening measure targeting basic receptive-analytical critical thinking skills in chemistry education contexts. The instrument was administered in parallel Kazakh and Russian forms to a non-probability sample of 253 bachelor's students enrolled in initial chemistry teacher-education programs. The finalized 26-item form was analyzed using classical test theory, confirmatory factor analysis, exploratory structural equation modeling, bifactor modeling, two-parameter logistic (2PL) item response modeling, local independence diagnostics, and differential item functioning (DIF) analysis. A general factor dominated the internal latent structure (explained common variance = 0.870; hierarchical omega = 0.922–0.947), whereas specific subscale reliabilities net of the general factor approached zero (0.044–0.144), leaving the total score as the sole defensible unit of interpretation. The observed total-score distribution was bimodal (bimodality coefficient = 0.700; Hartigan's dip D = 0.087, p < 0.0001): in this sample, 28.9% scored within the range expected under chance-level responding, while 17.8% obtained the maximum score. Empirical decomposition of the floor stratum showed that it is not homogeneous: accuracy exceeded chance on a cluster of six items while accuracy on the remaining twenty items was at or below chance, and a mixture model assigned three quarters of the stratum to a chance-responding class. High internal consistency (alpha = 0.94) was partly dispersion-driven and inflated by local item dependencies. No severe cross-language DIF was flagged among retained items, though the Russian-language subsample ( n = 61) affords limited statistical power for equivalence claims. The instrument provides targeted precision within the lower-to-middle range of the latent trait (θ from −2.00 to +1.05), functioning adequately for aggregated baseline screening, whereas individual-level high-stakes decision-making and broad population generalizations remain psychometrically unwarranted.