Theft at Work

Measurement of Theft

Perhaps the most critical feature of employee theft is that it is difficult to detect. It is clearly undertaken by employees with the intent of going undetected and thus stands in contrast with most other organizational phenomena of interest to the industrial/organizational psychologist. This problem of detection has widespread implications for research and practice. One issue is that it makes it difficult to even document the extent of the problem. The proportion of employees caught stealing is generally very small. For example, a common strategy for test validation is to test applicants, put them on the job, measure the behavioral outcome of interest, and then examine the relationship between test scores and outcomes. When this is done with theft as the outcome of interest, rates of detected theft over the first year of employment among typical populations [Page 801](e.g., entry-level retail workers) are in the 1% to 3% range. Although there is general agreement that some theft goes undetected, there is no agreement as to the proportion. Published estimates of the extent and cost of the employee theft problem reflect untested assumptions about the rate of undetected theft.

A second implication of the difficulty-of-detection problem is that it makes research on employee theft hard to interpret. For example, in trying to document psychological characteristics of employee thieves, one faces the question of whether detected thieves constitute a random sample of all thieves, or whether those caught are different in important ways from those who steal and are not caught. Organizations using selection systems, for example, hope to screen out individuals prone to theft, not merely those prone to get caught while stealing. Another research implication is that the statistical tools used to examine the relationship between psychological variables and employee theft (e.g., the correlation coefficient) cannot be interpreted in the normal manner when a variable under study is highly skewed. The maximum value of a correlation drops as the proportion caught/not caught stealing departs from 50%. At a 98%-to-2% split, the maximum possible correlation is .39, rather than the expected 1.0, and thus correlations with theft need to be interpreted relative to this maximum value.

...