Skip to main content

Table 3 Signal detection theory

From: Risk assessment and decision making about in-labour transfer from rural maternity care: a social judgment and signal detection analysis

Theory Design
SDT uses vignettes in which a forced dichotomous choice task is used to test the participant’s ability to detect a “signal” against a background of “noise”. For each vignette the participant decides whether the signal is present or absent. The signal can be any event or state that the person has to judge and the noise is the additional information which is presented. When used in decision making research SDT is based on the notion that a decision maker must have the ability to detect the need to take action i.e. to discriminate between high and low levels of risk in a case, and have a personal decision threshold which determines the level of risk they will accept before deciding to take action. SDT assumes that on average, skilled people are more likely to take action where there are higher levels of risk than in low level risk cases. At either end of the risk spectrum (high risk to low risk) the majority of skilled decision makers would agree, but there is a “grey area” where cases in which there is a need to take action, and those where no action is required overlap. The point at which the decision to act is made indicates the individual decision threshold. Information about the level of risk comes from the case assessment and is case specific. The personal decision threshold is based on belief about the likelihood and utility for possible outcomes and is relatively fixed across cases. For example, a clinician who believes that failure to progress in labour is likely, or that it will result in very negative consequences will require a lower level of risk before taking action than the clinician who believes it is unlikely to happen or have only minor consequences.
SDT uses vignettes which are developed as for SJT described in Table 2. However, to allow for the SDT analysis vignettes are specifically selected for inclusion in the task so that 50% have an average of a high level of overall risk across the factors (e.g. 60 out of 100), and are designated to be signal or “should take action” cases. 50% are selected to have a low average amount (e.g. 40 out of 100), of risk across the case factors and are designated as no signal or “should not take action” cases.
Administration Analysis
Participants are asked to decide for each case whether they would take action or no action Using this method, for each vignette a decision to act could be a true positive or false positive. The decision making performance of participants is captured by their true positive and false positive rates. These scores are turned into two indices of performance, ability to discriminate “should act” cases from “should not act” cases and the decision threshold (willingness to act) which is determined by the level of risk required in the case, before the decision to act was made. These analytic methods yield standard errors for the relative weights and thresholds and this allows comparisons between individual midwives and obstetricians using Z-tests. Ability has a minimum of zero when the participant has no ability. Willingness has a negative value when the participant has a greater willingness to act.