Comparing regression task performance of students under three experimental conditions

Abstract

This experimental study attempts to investigate how and to what extent peer collaboration affects students to evaluate regression model fitting and diagnose the problems associated with an unfitted model. Forty-nine students who voluntarily participated in the study were randomly assigned to one of these three groups: A, B, or C and there were four stages in the study. Students in groups A and B were allowed to have peer discussions about the workflow of a regression modelling in Stages I and III respectively, whereas students in Group C worked on their own in both stages. In Stage II or IV, all students in these three groups attempted three questions individually. It was found that students with pre-task discussion (A) demonstrated higher ability of accomplishing straightforward tasks, e.g., differentiating the functional roles of two variables, x and y; reading statistical graphs; defending successful data transformation, etc. than students with post-task or no discussion. Students with post-discussion (B) were more proficient in modelling and graphing regression data and validating the assumption of data normality than those with pre-task discussion or no discussion at all. On the other hand, students who had no discussion (C) performed better in checking the assumptions of homoscedasticity, independence, and linearity of data; hypothesizing data transformation; making a prediction with a transformed regression model and so on than those students with pre-task or post-task discussion. Hence, the effect of peer collaboration on students’ regression diagnostics performance is dependent on the types of regression tasks.