SMART: Student modeling approach for responsive tutoring

This paper describes a new student modeling paradigm called SMART. The premise is that a single, principled approach to student modeling, involving both theoretical and empirical methods, can render automated instruction more efficacious across a broad array of instructional domains. After defining key terms and discussing limitations to previous student modeling paradigms, I describe the SMART approach, as embedded within a statistics tutor called Stat Lady (Shute and Gluck, 1994). SMART works in conjunction with a tutor design where low-level knowledge and skills (i.e., curricular elements) are identified and separated into three main outcome types. Throughout the tutor, curricular elements with values below a pre-set mastery criterion are instructed, evaluated, and remediated, if necessary. The diagnostic part of the student model is driven by a series of regression equations based on the level of assistance the computer gives each person, per curriculum element. Remediation on a given element occurs when a subject fails to achieve mastery during assessment, which follows instruction. Remediation is precise because each element knows its location within the tutor where it is instructed and assessed. I end with a summary of results from two controlled evaluations of SMART examining the following research issues: (a) diagnostic validity, (b) individual differences in learning from Stat Lady, (c) affective perceptions of the tutorial experience, and (d) contributions of mastery and remediation to learning outcome and efficiency. Comments about related and future research with this paradigm are offered.