Interactions Are Critical

Abstract

Although we agree with Theobold and Freeman (2014) that linear models are the most appropriate way in which to analyze assessment data, we show the importance of testing for interactions between covariates and factors.

Laboratory Courses with Guided-Inquiry Modules Improve Scientific Reasoning and Experimental Design Skills for the Least-Prepared Undergraduate Students

“…For studies with control and intervention groups, ANCOVA can be used to control for regression to the mean (Barnett et al. , 2005), and a significant interaction between pretest score and treatment group on posttest score would indicate differential effects of an intervention based on student preparation (Beck and Bliwise, 2014).…”

“…, 2014; Theobald and Freeman, 2014; Theobald, 2018). To test for homogeneity of slopes (i.e., that the slope of the relationship between absolute gain and pretest score was the same for each course level), we included the interaction between course level and pretest score in our initial model (Beck and Bliwise, 2014). Because the interaction was not significant ( F (2, 455.7) = 0.38, p = 0.68), we removed the interaction from the final model.…”

Modeling Course-Based Undergraduate Research Experiences: An Agenda for Future Research and Evaluation

“…Third, certain CUREs may be subject to the same recruitment or selection bias as research internships, in that they only involve a small group of high-achieving or research-interested students ( Brownell et al ., 2013 ). Moving forward, it will be important to study the impact of CURE instruction using approaches that control for student-level variables and nonrandom assignment ( Theobald and Freeman, 2014 ; Beck and Bliwise, 2014 ), such as propensity score matching (e.g., Schultz et al , 2011 ) or regression discontinuity design (e.g., DesJardins et al , 2010 , 2014 ).…”

The Relationship between Perceptions of Instructional Practices and Student Self-Efficacy In Guided-Inquiry Laboratory Courses

“…To determine the effect of course level on postsemester self-efficacy and whether course level should be included as a covariate in subsequent analyses, we compared a model with just presemester self-efficacy as an explanatory variable against a model with presemester self-efficacy and course level as explanatory variables. The second model was originally fit with an interaction effect between presemester self-efficacy and course level ( Beck and Bliwise, 2014 ), but the interaction was removed from the final model because it was not significant (X 2 = 2.11, df = 2, p = 0.35). The simpler model with just presemester self-efficacy as a covariate was a better model than a model that included course level, based on minimizing Akaike information criterion (AIC; Table 2 ), so course level was excluded from subsequent models.…”