forecastingprinciples.com Reviews of Important Papers on Forecasting,
1985-1995 Reviews
Review of:

H. John Bernardin and D.K. Cooke, 1993, Validity of an honesty test in predicting theft among convenience store employees, Academy of Management Journal, 36, 1097-1108.


Theft is an important problem for food stores. It has been estimated that about 45% of employees steal either goods or money. In 1988, the US Government prohibited the use of polygraph testing by most employers. How then can stores predict which potential employees will steal from them? The answer is simple: you ask them.

Here are some of the questions asked about honesty in the London House Test (London House is a company in Park Ridge, Illinois). They are organized into five areas: (1) Thinking about stealing (e.g. ‘How often in recent years have you simply thought about taking money without actually doing it?"), (2) being more tolerant of those who steal (e.g. ‘A young person caught stealing $50 in cash from an employer. If you were his employer, what would you do?"), (3) believing that theft is common (e.g. "How many executives steal from their companies?"), (4) believing in loyalty among thieves (e.g. "If you were caught stealing, would you tell on the people who helped you?"), and (5) rationalizing theft (e.g. "Would you steal if the conditions were right?").

Paper and pencil tests of honesty are a growth industry. Those who develop tests do studies to determine whether the tests provide good predictions. A meta-analysis of 23 correlations using theft as the criterion found that the correlation between the honesty score and theft was 0.50 (McDaniel and Jones, 1988). Most of those studies used self-reports as the criterion, some used supervisor’s ratings of theft, one used a polygraph, but only two used measures that were closely related to actual theft behavior; one was small intakes by Salvation Army charity collectors and the other was a criminal background check. (Incidentally, my interpretation of the McDaniel and Jones’ findings differed from those presented in Bernardin and Cooke, so interested readers should depend on the original source.)

Already I sense that you are dubious about the results of studies done by those who sell the tests. The independence of the forecaster has been shown to be important (e.g. see Shamir’s, 1986, study of pre-election polls). Bernardin and Cooke get around this problem because they are independent academic researchers. The interesting part of this study is that it uses actual theft as the criterion. Bernardin and Cooke examined data on 111 employees of a major retail convenience store who worked at counters of convenience stores or gas station outlets. The London House Personnel Selection Inventory (PSI) had been used to hire these employees, but no on involved in the stores had been informed about the employees’ scores on the honesty subscale of the PSI. (Their inclusion in the overall PSI scores would, however, be expected to lead to a range restriction with respect to honesty scores, and thus to reduce the observed relationship between the honesty score and theft.) Of the 111 employees, 54 had been formally terminated for theft and none had appealed this reason for termination. There was a 0.28 correlation between the honesty subscale and theft (P < 0.01). This result lends additional support to the meta-analysis findings by McDaniel and Jones.

The results are consistent with the belief that paper and pencil tests provide one of the most effective ways to predict honesty. The results also agreed with previous studies that men are more likely to steal than women, and that younger people are more likely to steal, although these findings were not statistically significant. Also, consistent with prior studies, race was unrelated to theft among employees.

This study provides another case where attitude can successfully predict behavior. Note that most of the items related to thoughts about behavior. Also, in view of prior research, it is interesting that the potential bias on the part of the researchers appeared to have only modest effects. One way to overcome bias in such a situation is for the researcher to provide full disclosure of the methods and their limitations (Weimann, 1990, provides evidence on this using pre-election polls). Those who provide the PSI tests seem to provide good disclosure, but this is speculation on my part.

References

McDaniel, M.A. and J.W. Jones, 1988, Predicting employee theft: A quantitative review of the validity of a standardized measure of dishonesty, Journal of Business and Psychology, 2, 327-345.

Shamir, J., 1986, Pre-election polls in Israel: Structural constraints on accuracy, Public Opinion Quarterly, 50, 62-75.

Weimann, G., 1990, The obsession to forecast: Pre-election polls in the Israeli press, Public Opinion Quarterly, 54, 396-408.