City Research Online

Graduate recruitment at professional entry level: clinical judgements and empirically derived methods of selection

Harvey-Cook, J.E. (1995). Graduate recruitment at professional entry level: clinical judgements and empirically derived methods of selection. (Unpublished Doctoral thesis, City University London)


This research provides evidence to support the argument that selection procedures dependent upon clinical judgements, being used in the chartered accountancy profession, may well provide results not significantly different from those obtained by chance. Research has suggested that personality type, choice of vocation and performance are predictable from personal histories (Holland, 1976; Owens and Schoenfeldt, 1979; Eberhardt and Muchinsky, 1982a; Super, 1980; Wernimont and Campbell, 1968) and using a predictive model approach to scoring biographical data (biodata) is explored here as a means of improving the selection function. Part I of this study develops predictive models for scoring the biodata of applicants to the profession. An original contribution is made by carefully comparing two empirical model-building methodologies: the generally accepted, non-parametric, Weighted Application Blank technique and the parametric, logistic regression technique. The validity of both are explicitly tested using information from a sample of 23 training offices from 22 medium size chartered accountancy firms. The sample trainees were all non-accounting graduates entrants entering between 1985 and 1987 (N=665). Evidence is provided of the superiority of the results of the parametric models, in terms of true predictive validity. Relevant theory and the important implications of the results for related biodata studies generally are discussed. The result of applying the models to applicants, rather than recruits, is examined in a pilot study. An original approach to scoring applications is presented. Specifically developed software is provided to minimise both processing time and error margins. The biodata logit scores of the applicants and their likely success as trainees as indicated by that score, are compared with the firm's decision whether to accept or reject. Severe problems inherent in the judgemental approach to selection are revealed and the superior performance of the model-based approach demonstrated. Part II addresses the crucial issue of long term validation of biodata models by scoring a sample of recruits from 3 representative firms' 1988-90 entrants (N=323). The evidence does not support criticism of long term validity, as the logit models demonstrate effective performance, measured interms of the probability of correct classification, successfully predicting the criteria on those entering the profession up to 5 years after subjects used in model development. It is suggested that poor methodology may be responsible for excessive loss of validity over time in other studies and their lack of use of hard data. In addition, original evidence is provided to support the hypothesis of the generalizability of such models (i) across organizations and (ii) across samples significantly different from the development sample. This evidence suggests that, not only may the models be used to score applicants accounting firms of different sizes (and are therefore not organization-specific) but they may be used to score accounting graduates, who differ considerably from the original development sample (indicating that they, are not sample specific). The appropriateness of using these models in a manner similar to psychometric tests is considered. An assessment of approximate net profit associated with successful, failing or partially successful trainees is made. Accounting graduate trainees are more financially viable than non-accounting graduates

Publication Type: Thesis (Doctoral)
Subjects: H Social Sciences > HB Economic Theory
Departments: Bayes Business School
Doctoral Theses
Bayes Business School > Bayes Business School Doctoral Theses
PDF - Accepted Version
Download (12MB) | Preview



Downloads per month over past year

View more statistics

Actions (login required)

Admin Login Admin Login