An exploratory investigation of a methodology for constructing homogeneous keys for a biographical inventory

The present research was an exploratory investigation of a methodology for constructing homogeneous keys for a biographical inventory. The methodology investigated was a computerized adaptation of a technique used previously by DuBois, Loevinger and Gleser (1952). The primary objective of the study was to determine whether relatively independent, homogeneous keys could be constructed with this technique. Of secondary concern was the relationship of the resulting keys with various criterion information. Computer programs were written to (1) compute a matrix of inter-item covariances, (2) select nuclei of three items having high covariances inter se, (3) add items to each nucleus so as to maximize the saturation of that key, defined as the ratio of the inter-item covariance to the key variance, and (4) compute the between*-key and item-key correlations. The biographical data source was a 55 item, 229 alternative background questionnaire which was administered with a selection battery of four cognitive tests. The applicant population consisted of 2590 men who had applied for operating jobs at a west-coast oil refinery. The keys were constructed on a random sample of 1000 non-hires drawn from the pool of 2590 applicants. Thirteen keys were generated which had an average saturation of .54 and an average intercorrelation of .10. The keys were labeled Academic Ability, Scientific Interests, Mechanical- Electrical Orientation, Athletic Involvement, Paternal Blue Collar Background, Maternal Employment History, Rural Background, Risk Taking Behavior, Social Aggressiveness, Upper-Middle Class Background, Parental Permissiveness, Favorable Self-Perception, and Musical Interests. The keys were scored on all 168 applicants who had been hired and who were still employed after 10 months. The resulting scores were intercorrelated with scores received on the cognitive tests in the selection battery, scores on a key which had been empirically derived from the same questionnaire, and on 2, 6, and 10-month criterion ratings. Although there was a severe restriction in range, various homogeneous keys related moderately well with the criterion measures and compared favorably in these relationships with the empirically derived key. It was concluded that (1) the methodology investigated did provide a practical means of generating relatively independent and homogeneous keys, and (2) the keys related to various criterion measures in meaningful ways. Suggestions for further increasing the utility of the computer programs, and several recommendations for additional research, were made.