Modelling police officers' judgments of the veracity of suicide notes.

• Author: Snook, Brent
• Position: Canada

Within the realm of judgement and decision-making (JDM) studies, there is an ongoing debate, known as the rationality debate, on the subject of human decision-making complexity and level of information processing (see Gigerenzer, Todd, and the ABC Research Group 1999; Gilovich, Griffin, and Kahneman 2002; Kahneman, Slovic, and Tversky 1982). Along this dimension of complexity, there is a gradient that ranges from extremely simple and uncomplicated decisionmaking strategies, known as heuristics, to complex and computationally expensive strategies, referred to as fully rational (Payne, Bettman, and Johnson 1993). Much recent bounded-rationality research has established that people use simple heuristics to make a range of decisions that lead to adaptive and successful outcomes (see Gigerenzer 2008). Interestingly, this simple heuristic research has begun to shift away from modelling somewhat benign tasks (e.g., predicting the largest of two cities; see Gigerenzer and Goldstein 1996) toward consequential, real-world decisions in domains ranging from medicine to law (e.g., Dhami 2003; Dhami and Ayton 2001; Dhami and Harries 2001; Smith and Gilhooly 2006). The purpose of the current study is to continue this progression by examining the possibility of modelling judgements made by police officers regarding the veracity of suicide notes.

According to Goldstein and Hogarth (1997), debate regarding rationality initially favoured fully rational strategies as the ideal standard of human judgement and this provided the ground against which bounded-rationality research emerged. More recently, Gigerenzer et al. (1999) developed a number of fast and frugal strategies that could be empirically tested and compared to more complex strategies in real-world environments. These simple heuristics are each composed of, and differentiated by, three main components: (1) guide for information search, (2) rules for stopping search, and (3) rules for decision making based on results from the first two steps. By differentially combining these components, heuristics such as Take the Best were conceptualized (see Gigerenzer and Goldstein 1996). In a number of studies comparing the predictive accuracy of these heuristics against that associated with more complex models, it was found that the simple heuristics performed as well or better on a variety of decision tasks. For example, Czerlinski, Gigerenzer, and Goldstein (1999) found that the Take the Best heuristic performed as well as more complex statistical formulae on 20 different prediction tasks (e.g., predicting the number of car accidents along a stretch of highway and predicting high school drop-out rates).

Similarly, Dhami and Ayton (2001) developed the Matching Heuristic (MH), which has been the focus of a significant amount of empirical testing (Dhami 2003; Dhami and Harries 2001; Kee, Jenkins, McIlwaine, Patterson, Harper, and Shields 2003; Smith and Gilhooly 2006). The MH can presumably be used to model any binary decision in which decision makers have multiple pieces of information (cues) available to them. The MH conforms to the three components of any simple heuristic discussed earlier. First, the MH searches through information according to the descending utilization validity of each cue, which represents the frequency with which a particular cue value coincides with the non-default of the two decision outcomes. Secondly, the search for information is stopped when a cue is found that allows a decision to be made. (Returning to the example, a police officer may be able to render a decision after using only two pieces of information, and therefore stops searching for more information). Thirdly, the MH offers a model of decision making that contains only the cue that led to a stop in the search for information, together with the cues, if any, that were searched before that cue was reached (see procedure for a more detailed description of the MH).

As mentioned, a number of studies have been conducted that compared the ability of the MH to model human decision making with that of more complex strategies. The consistent and robust finding from this body of research is that the MH, at a minimum, equals the performance levels of a variety of more complex strategies and does so by ignoring much of the available information. For example, Dhami and Harries (2001) compared the ability of the MH and that of a regression model to capture how physicians made medical decisions. In this study, physicians were asked to decide whether to prescribe lipid-lowering drugs to a hypothetical cardiac patient, based on a description containing information on 12 symptoms. Although the MH models used fewer cues (symptoms) than the regression models, both types of models fit the physicians' judgements equally well. These authors concluded that the MH was, indeed, the more enticing model because it offers a simpler, and therefore more psychologically plausible, cognitive strategy (for similar findings, see Smith and Gilhooly 2006).

Another study by Dhami (2003) sought to establish the validity of the MH in a legal context by comparing it to Franklin's Rule (FR)--a compensatory strategy. According to Dhami (2003), FR requires cues to be differentially weighted according to their relative importance to the decision task. After each available cue is assigned a subjective relative weight, all are considered in a compensatory fashion, whereby combinations of cues for and against a particular decision option can cancel each other out according to their respective weights (see also Payne et al. 1993). This study focused on UK magistrates' decisions to either grant or deny bail to actual persons appearing before them in court. Considering the judicial practice of due process that is supposed to ensure a comprehensive and fair judgement, one could assume that the rational strategy would model the judges' bail decisions more accurately than a simple heuristic. However, Dhami (2003) found that the MH more accurately described the judges' decision making and was able to more accurately predict judges' future bail decisions (i.e., cross-validation).

This review of the relevant studies suggests that the MH is an accurate and psychologically plausible model of professional decision making. In the current study, the MH and FR were each used to model police officers' decision-making processes as they attempted to judge the veracity of suicide notes. The MH was used as the representative simple heuristic and FR was used as an example of a rational strategy. Of note, FR was chosen as a comparative benchmark for the MH, but this does not imply that police officers' decision making should (or can) adhere to this strategy.

Demonstrating the applicability of heuristics in police decision making, specifically suicide investigation, would bolster simple heuristic research by extending the range of domains in which heuristics have been investigated, while also confirming their psychological plausibility. Moreover, the results of such an investigation are valuable because judging the veracity of suicide notes is an important decision that often has serious consequences in equivocal death analysis (EDA). When presented with an equivocal death, accurately determining whether it is a homicide, suicide, or accidental death is essential to ensuring justice is done as well as to preventing police organizations from wasting valuable resources on unnecessary criminal investigations (Ault, Hazelwood, and Reboussin 1994). As far as we know, there is...