Research has found that deception detection accuracy in the context of suspect interrogation hovers around chance levels. Geiselman (2012) adapted the cognitive interview (typically used for witnesses) for use with suspects (CIS) and found that judgments of deception were more accurate than previous interrogation techniques. The current study attempted to use the CIS to improve deception detection with Reality Monitoring (RM: Vrij et al., 2008), which has already been validated in the context of witness statements. One hundred sixty-six undergraduate students were randomly assigned to 2 conditions. In the Truthful condition, participants played a game with a confederate, whereas in the Deceptive condition, participants rehearsed (but did not experience) a synopsis of the game scenario. Participants in the Deceptive condition were also instructed to steal $10 from a confederate's wallet. In both conditions, $10 was purported to be missing and a researcher blind to condition conducted a CIS. Statement veracity was coded using 6 of the RM criteria advanced by Vrij et al. (frequency of visual, auditory, spatial, temporal, cognitive, and affective details). According to results from a MANOVA, truthful and deceptive statements differed significantly on all RM criteria, with the exception of affective details, validating the importance for evaluation of statement veracity (p ≤ .01). Further, a binary logistic regression found that combining the RM criteria together correctly classified 86.6% of statements, χ(²)(6) = 114.4, p < .001, with excellent sensitivity and specificity (.899 and .833, respectively). As well, Visual, Auditory, and Cognitive details uniquely predicted condition. Findings support using RM criteria to detect deception in interviews conducted with the CIS.

No abstract

This study explores twitter data about insurance and natural disasters to gain business insights using text analytics. The program R was used to obtain tweets that included the word ‘insurance' in combination with other natural disaster words (e.g., snow, ice, flood, etc.). Tweets related to six top Canadian insurance companies as well as the top five insurance companies from the rest of the world, including the new entrant Google Insurance, was collected for this study. A total of 11,495 natural disaster tweets and 19,318 insurance company tweets were analyzed using association rule mining. The authors' analysis identified several strong rules that have implications for insurance products and services. These findings show the potential text mining applications offer for insurance companies in designing their products and services.