Errors in Detecting Financial Deception: A Cognitive Modeling Approach

Organizational agents often need to determine whether business information provided under conflict of interest is intentionally misleading or false. Detecting strategically manipulated information is in general difficult and prone to failure. This research explores the errors made by 18 loan officers when examining misleading financial information. The process traces of the loan officers are compared with the behavior of a cognitive model of detection success derived from Social Contract Theory and the Theory of the Detection of Deception. The results show that one of the keys to successful detection is the ability to ‘coming to think’ about deception, i.e., the ability to begin interpreting perceived anomalies in the received information as generated by the sender’s malicious manipulations. The findings on the distribution of different error types are consistent with the theory, statistically significant, and pragmatically meaningful. Modeling the Detection of Deception The increasing diffusion of electronic commerce practices has brought an increased sensitivity to the security and quality of financial information communicated among business parties under conflict of interest. A great deal of work is done to understand, design and build secure information architectures that are able to insure the identity of those involved in the communication, as well as to protect the communication contents from third-parties tampering. By contrast, relatively less attention (March, 1990) has been paid to the problem of evaluating whether the received information has been maliciously manipulated by the sender. Such manipulations are designed to foster misrepresentations of crucial aspects of business transactions and in this way influence the behaviors of the information recipients (e.g., Hansen et al., 1996; Lucas, 1993). Business instances of this ‘deception detection’ problem include the detection of manipulated insurance claims, applications for loans, and financial reports. In general, detecting strategically manipulated information is difficult (Ekman, 1995). Failure occurs even when the detectors are trained professionals using sophisticated information technologies and motivated by high stakes (Albrecht et al., 1995). The research presented here is a part of a larger investigation of the determinants of performance at detecting deception in financial settings (Grazioli, 1997). Here we focus on comparing the behavioral trace of a sample of organizational agents with the behavior of a cognitive model of successful detection. The model is based on the integration of Cosmides’ Social Contract Theory (1985; 1992), with Johnson’s (et al., 1992, 1997) Theory of Fraud Detection. It implements a psychologically plausible strategy for detecting misleading financial information. The strategy consists of interpreting perceived anomalies in the received information in the light of the goals and possible actions (manipulations) that can be ascribed to the information provider. Processing is composed of four steps: (1) the detection process is triggered by the identification of anomalies, defined as observations that violate their expectations; (2) these anomalies prompt a search for hypotheses that explain them, including the hypothesis that the anomalies result from malicious manipulations; (3) the selected candidate hypotheses are then evaluated and (4) combined to form an overall assessment. The theory has argued that a key step for success is the generation of the hypothesis that information has been manipulated by its provider (as opposed to say identifying anomalies in the provided information, or evaluating candidate hypotheses). Detection failure is explained as the selection of alternative and incorrect hypotheses. These alternative hypotheses explain manipulation-related anomalies as the results of circumstances independent from the information provider (e.g., a low market demand), or as the result of the deceiver’s unintentional action (e.g., an accounting mistake). A Field Experiment on Deception Detection The banking industry offers an excellent setting to investigate how organizational information recipients (commercial loan officers) detect intentionally incorrect or misleading information communicated under conflict of interest (applications for a business loan). An increasing number of banks accept electronic filings of loan applications and feed the provided financial information into decision support systems used by loan officers to make a decision about the loan. Eighteen loan officers from a large national bank volunteered in a field experiment designed to study the effect of various levels of knowledge of the goals and possible actions of the information provider on detection effectiveness (Grazioli, 1997). The subjects averaged 15 years of experience in the financial industry, of which 9.5 as a loan officer (std.dev. 5.4 and 7.3,