Investigating Microadaptation in One-to-One Human Tutoring

Abstract

The authors investigated whether some advantages of tutoring over other instructional methods are due to microadaptation, or, tutors basing their actions on assessments of tutees they develop during tutoring. In a 2 × 2 between-subjects experiment, independent variables were shared experience (tutors either worked with the same or a different tutee in each of four segment of the tutoring session) and communication context (face-to-face or computer-mediated). Although there were no overall learning differences across experience conditions, tutees who worked with the same tutor demonstrated better learning of concepts initially discussed during the final tutoring segment. Shared experience led to accurate competence assessments only in the computer-mediated context, suggesting that cognitive load influences assessment development. However, there was no evidence of micro-adaptation.

Keywords:

• human tutoring
• microadaptation
• learning
• assessment of tutee

Notes

In this article, knowledge is defined as the ability to produce facts or explain concepts; learning is operationally defined as the shift from unsuccessful to successful production and application of facts or concepts in an assessment scenario.

Because we are investigating the relation between tutors’ competence assessments and their tutoring actions, we need to know whether tutors’ assessments of tutee competence are related to tutees’ actual competence. That is, we are interested in whether tutors develop accurate assessments of their tutees’ competence relative to other tutees. Accuracy of competence assessments is related to the strength of correlations between tutors’ assessments and tutees’ actual general competence.

In piecemeal and overlay assessments, tutors predict tutee's responses to specific questions. However, for piecemeal assessments, tutors predict how many questions the tutee will get correct, whereas for overlay assessments, tutors predict which questions the tutee will get correct. For example, if the tutee's pattern of correct answers is 10110, then the tutee has gotten three of five answers correct. If the tutor predicts that the tutee will get three of five answers correct, then the absolute accuracy of the tutors’ piecemeal assessment is 100%. However, if the tutor is also asked which questions the tutee will answer correctly, and the tutor predicts that the tutee's pattern of correct answers is 01011, then the tutor is only 20% accurate on an overlay model assessment.

Although the effect of shared experience increased with tutors’ physics knowledge, there was no significant effect of experience condition among tutees of higher-knowledge tutors (p =.15).

FIGURE 1 Relationship between tutor pretest and tutee total posttest score by condition.

Relevant concepts assessed were (a) the kinematics equation without time, (b) the final velocity of the cannonball at the top of the arc of its flight is zero, (c) its (vertical) acceleration is g (about 10 m/s²), and (d) its (vertical) displacement is 5 m (given in the problem statement).

In the Different condition, there was a context by tutor pretest by tutee midtest interaction, F(1, 27) = 5.34, p =.03, η_p² =.17. In the FTF Different condition, there was a marginally significant negative correlation between tutor pretest and tutor Likert rankings (r = –.50, p =.06). There were no significant effects in the CM Different condition.

In the Different conditions, there was again a significant negative correlation between tutor pretest and tutors’ predictions (r = –.40, p =.01). However, this correlation was significant only in the FTF Different condition (r = –.47, p =.04).

Furthermore, there were no interactions between prior tutoring experience or tutor pretest and either type of competence assessment. Thus, higher-knowledge tutors or tutors with more prior experience were not more likely to adapt question difficulty.

One possibility is that tutors in a CM context were able to maintain a running tab of tutees’ knowledge deficits in relation to their own knowledge level (there was also a negative correlation between lower-knowledge CM tutors’ initial knowledge and their piecemeal assessments). Perhaps once that difference becomes too large (more likely for higher-knowledge tutors), tutors are no longer sensitive to further differences. However, this issue needs further investigation.

Additional information

Notes on contributors

Stephanie Ann Siler

Stephanie Ann Siler (Ph.D., University of Pittsburgh) is a Research Psychologist in the Department of Psychology at Carnegie Mellon University. Her main interests are science learning and instruction—particularly in interactive contexts.

Kurt VanLehn

Kurt VanLehn is The Diane and Gary Tooker Chair for Effective Education in Science, Technology, Engineering and Math School of Computing, Informatics and Decision Science Engineering at Arizona State University. His research interests include intelligent tutoring systems, educational data mining, cognitive science, and learning science.