Confounding Carelessness? Exploring Causal Relationships Between Carelessness, Affect, Behavior, and Learning in Cognitive Tutor Algebra Using Graphical Causal Models

Studies have found positive correlations between affective states (e.g., confusion, boredom) and learning outcomes in educational technologies like ASSISTments and Carnegie Learning's Cognitive Tutor. The adage that "correlation does not imply causation" is especially apt in light of these observations; it seems counterintuitive that increasing student boredom or confusion (e.g., designing systems that bore or confuse students) will benefit learning. One hypothesis to explain positive correlations between boredom and learning suggests that carelessness is a “confounding” common cause of boredom and another construct linked to learning. We consider a Cognitive Tutor Algebra dataset in which boredom and confusion are positively correlated with learning. Prior causal modeling of this data suggests that various behavioral and affective features (e.g., boredom and gaming the system) share unmeasured common causes. We provide a correlational analysis and causal models of this data that situate carelessness among behaviors and affective states to determine whether (and how) carelessness plays a confounding role.