Trainability assessments and work samples : a field study and a meta-analysis : a thesis presented in partial fulfillment of the requirements for the degree of Master of Arts in Psychology at Massey University

Abstract

This study compared a work sample test with a trainability test for the prediction of typing students grades. A meta-analysis of the work sample literature was also carried out. Participants in the work sample trainability test comparison were 89 female first year Polytechnic typing students. Students were randomly assigned to either the work sample group or the trainability test group. Tutors then administered the relevant predictor and data was collected. Scores on the predictors were later correlated with the students grade in their second terms test. All the obtained correlations were found to be highly significant although the results unexpectedly revealed that the error score on the work sample was the best predictor overall. It was

suggested that the tutors inexperience in

administering trainability tests, their greater familarity with work samples

and certain

deficiencies in the criterion may have contributed to the unexpected trend in the data. Meta-analysis was used to cumulate and average results from many different studies which examined work samples. Studies which utilised training criteria were analysed seperately from those which employed job proficiency criteria. Results from the analysis showed substantial remaining variance following correction for statistical artifacts. The studies were then grouped according to Robertson and Kandola's (1982) classification of work samples in order to identify potential moderator effects. Meta-analysis of subgroups revealed that for all categories, with the exception of group discussion/decision making, considerable variance still remained following correction for statistical artifacts. It is suggested in the

discussion that further research on work samples is required, particularly the development of a classificatory system which can accurately and reliably distinguish between types of work samples. Possibilities for future research on trainability tests are also explored.