Hur sannolikt är ”sannolikt”? - Ett kvasi-experiment om kognitiva bias i underrättelsetjänsten

Sammanfattning: Studies on cognitive psychology have shown that estimates of probability are influenced by the way in which semantic and numeric information is presented to us – so called framing bias. Further, studies on intelligence populations have indicated that bias is present in the analytic work. The aim of this study was to explore how framing bias can manifest in estimates of probability in intelligence analysis -using three probability frames: unlikely, plausible and likely, as well as high versus low consequence frames. Further the study explored whether there are differences between a Swedish intelligence population and a control group in such framing effects. The participants were tasked with reading 6 fictive intelligence reports and answer questions about the probability of the described event occurring. The participants were also asked to translate semantic frames of probability into numeric values, independent of contextual information. Results found a framing effect for both the control- and experimental group based on whether expected consequences were framed as high or low. However, when testing the different intelligence departments individually, the effect was only significant for one. Further, the results showed that there was no significant difference between the three intelligence-departments translations of semantic probability frames. The study suggests that more research is needed on the area of how to mitigate bias in intelligence analysis.