While there are many approaches to integrated STEM instruction (iSTEM), the integration of engineering design is the most widely-studied and practiced pedagogical approach to iSTEM in K-12 classrooms. Research has shown that the inclusion of engineering-design improves students' attitudes, as well as interest and engagement in pursuing STEM-related careers. Furthermore, studies have shown enhanced 21st century skills for students engaged in iSTEM learning contexts. However, more research is needed to understand how iSTEM and its critical features are operationalized to promote positive student outcomes. To address this need, this study examined the relationship between student cognitive engagement in iSTEM and its hypothesized predictors: curricular opportunities for STEM content integration, engagement in multiple solution development, agency in STEM practices, evidence-based reasoning, data practices, and collaboration. The study is guided by Roehrig et al.'s (2021) Detailed Conceptual Framework of Integrated STEM and Moore et al.'s (2014) framework for Quality K-12 Engineering Education. We utilized multinomial logistic regression (MLR) analysis due to the polytomous categorical distribution of the outcome variable. This study used classroom video data from previous work that examined the presence of critical features of K-12 iSTEM. Scores using a novel and validated iSTEM observation protocol (Dare et al., 2021) from 2,007 iSTEM lessons were used. Through preliminary analyses, we determined that the assumptions for MLR have been sufficiently met. Three categories of the outcome variable, student cognitive engagement, reported on were lessons that provide opportunities for students to (1) analyze/evaluate STEM concepts, (2) use/apply STEM concepts, and (3) know/understand STEM concepts (which was set as the baseline or reference category). All predictor variables except for curricular opportunities for collaboration and data practices were statistically significant in the model. The final MLR model has a total of 12 predictor categories. The deviance goodness-of-fit test indicated that the model was a good fit to the observed data, χ2(234) = 207.605, p = .892, with 137 (36.2%) cells having zero frequencies. The final model statistically significantly predicted the outcome variable over and above the intercept-only model, p < .001. Furthermore, it has a pseudo R-squared value of.643 (Nagelkerke R2) and correctly classified 72.8% of cases. Among other findings, we found that the odds of multidisciplinary lessons providing opportunities for students to analyze and/or evaluate STEM concepts was 2.401 times higher than that for monodisciplinary lessons, χ2(1) = 24.963, p < .001. In addition, lessons with opportunities for students to redesign a solution to the engineering task are more likely to provide opportunities for students to analyze/evaluate STEM concepts (exp(B) = 126.038) compared to lessons without such curricular opportunity, χ2(1) = 22.033, p < .001. In conclusion, engineering-centric iSTEM instruction that engage students in higher levels of cognition are marked by the presence of multidisciplinary content, engagement in designing solutions to an engineering problem, agency in STEM practices, and evidence-based reasoning.
