Figure 1
A computational approach to the study of uncertain threat anticipation. a) Depiction of threat-anticipation studies to compare certain-vs-uncertain conditions. b) P(Threat) (top) and hazard rate (bottom) reveal that threat-context and hazard rate are confounded, and that the cumulative hazard rate is greater during uncertain trials. c) We addressed this confound by comparing two uncertain conditions. d) Uncertain conditions differed in the timing of P(Threat) (top), but were matched through most of the trial. Cumulatively and during periods of matched P(Threat), the hazard rate (bottom) was higher in the early-threat condition. e) Participants were exposed to each condition 50 times in a pseudo-randomized order. All trials ended in shock. f) During the testing phase, trials continued and participants earned money until avoidance or shock.
Figure 2
Higher hazard rates cause more avoidance behavior and fear/anxiety. a) Survival analysis revealed a significant difference between avoidance behavior at all timepoints (log-rank test, N = 42, 𝑥2 = 259.3, p < .005). b) In epochs during which P(Threat) values were matched (see Figure 1d), participants were significantly more likely to avoid the early-threat condition, where the hazard rate was higher (independent-samples t-tests, N = 42: [6–10s: t = 8.09, p < .001], [11–15s: t = 7.80, p < .001], [16–20s: t = 4.71, p < .001], [21–25s: t = 8.30, p < .001]). c) In a final, 30-second, unavoidable-shock trial for each condition (counterbalanced), participants reported significantly more fear/anxiety in the early-threat condition (Wilcoxon rank-sum test, N = 21, U = 328.5, p < .005). This analysis held after statistically accounting for the cumulative number of shocks received in each condition, as can be seen in the early-late fear/anxiety ratings residualized for the number of shocks (c, right). d) In a forced-choice test, 19 of 21 participants identified the early-threat condition as more anxiogenic (binomial test, N = 21, p < .001). Error bars represent the 68% CI, corresponding to the standard error of the mean. ** = p < .005, *** = p < .001.