Modelling Performance Improvement and Switching Behaviour in Team Learning

Abstract

Accounting educators and agencies have sought to incorporate team learning activities into conventional learning methods. The readiness-assurance process (RAP) of team learning, in which students take quizzes first as individuals and second as members of student teams, has been shown to be effective in this regard. We analyse the RAP with a fixed-effects regression model to identify the factors that contribute to performance improvement and we use ordered logit regression to estimate, probabilistically, switching behaviour within student teams. A longitudinal study was conducted over the course of a semester in which 101 undergraduate accounting students, comprising 22 teams, completed six quizzes. Within-team knowledge disparity was shown to be a significant indicator of performance improvement, and individuals appeared more likely to switch their answers after the first quiz. There were no significant effects for either performance or switching associated with demographic measures of sex and English fluency. Implications for accounting educators are discussed.

Keywords:

• Team learning
• readiness-assurance process
• student teams
• group-performance measurements
• group-structure effects

Acknowledgment

With the traditional disclaimer, the very helpful comments and suggestions of two anonymous referees and an Associate Editor are gratefully acknowledged.

Notes

¹ In four instances, a student did not attend class on the day of the quiz.

² The fixed-effects approach assumes that the dependent variable in a linear regression, in this case R _gq, is a function of a vector z ^T _q of unobservable factors, which enter the regression through the additive term z ^T _q f , where f is a vector of parameters. Setting z ^T _q f = F _q “specifies an estimable conditional mean” which “takes [F _q ] to be a group-specific constant term in the regression model” (Greene, 2003, p. 285).

³ As described in Greene (2003, p. 736–739) ordered logit regression assumes all of the parameters, α _q and β _m , to be fixed over all levels of k.

⁴ One group of students initially performed above the class average on every quiz. We were reluctant, however, to define a dummy variable with only six observations for possible inclusion in the regression.