Improving Time Estimation in Witness Memory

The present study sought to determine whether witness memory for duration could be improved. In three studies, we examined the effects of unpacking (breaking an event into its component parts), anchoring (supplying participants with a reference duration), and summation (summing component estimates). Participants watched a video-recorded mock crime and provided duration estimates for components of the crime (e.g., casing the car, unlocking the door, etc.) and for the total crime. Results indicate that bias in estimated duration was less for the sum of the parts than it was for the overall duration estimate. Further, the sum of the individual parts did not equal the total estimate, even though all estimates were given in sequence. Summing the component parts could be a more successful intervention than anchoring or unpacking and is easy to employ with witnesses.


Confidence
Participants' confidence in their estimates was compared using an 8 (item being estimated) x 2 (Unpacking) x 2 (Anchoring) mixed-model ANOVA. There was a significant effect of item, with confidence varying dependent on what was being estimated, F(7,434) = 14.34, p < .001, η 2 = .19. As can be seen in the mean values reported in Table 1, post hoc analyses (LSD) indicated that confidence ratings could be placed into 3 somewhat equivalent groups (estimates in those groups were not significantly different from each other, but significantly different from the estimates in the other groups). Receiving the highest confidence ratings was a group that included duration estimation for unlocking the door, removing the item and the total duration for the crime. Participants were somewhat less confident in their ratings of duration for walking up to the car, casing the car and for the height of the perpetrator. The least amount of confidence in ratings was reserved for the weight of the perpetrator and the number of visible cars. It is notable that participants were least confident for ratings that tended to have the smallest average bias. Participants were most biased in their estimations of duration, but also most confident. Confidence was not influenced by either of the manipulations, anchoring and unpacking, ps > .25, η 2 s < .02. The average accuracy of the time estimates was collapsed across segment and participants and plotted as a function of confidence and anchor. As can be seen in Figure 1, accuracy increased with confidence when an anchor was given, whereas there was no relationship between confidence and accuracy when no anchor was given. However, the error bars are overlapping, indicating no significant differences. Participants' height and weight were recorded to examine whether or not these factors influenced judgments of the perpetrator's height and weight (anchors can influence judgments such as height, Mussweiler &Strack, 2000, andweight, Frederick &Mochon, 2012).

Other Measures
We examined bias ( There were no gender differences in bias for the variables measured here (ps > .17).
Though height of the participant was not related to their estimation of perpetrator's height, r(150) = .037, p = .652, there was a small correlation between participant's weight and the estimation of the perpetrator's weight, r(150) = .210, p = .010. It is possible that participants used their own weight as an anchor for their judgment of perpetrator's weight.

Confidence
Confidence for the estimates was compared using a 9(item being estimated) x 2 (Unpacking) x 2 (Anchoring) mixed-model ANOVA. There was a significant effect of item, with confidence varying dependent on what was being estimated, F(8,1136) = 46.90, p < .001, η 2 = .25 (see Table 2). However, ratings of confidence for duration estimates were influenced by the  The average accuracy of the time estimates was collapsed across segment and participants and plotted as a function of confidence and unpacking condition. As can be seen in Figure 2, accuracy increased with confidence when total was given last (in the unpacking condition) compared to when the total was given first (in the no unpacking condition). However, the error bars are overlapping suggesting no difference in confidence accuracy calibration across conditions.

Figure 2.
Average accuracy (+2 SE) of time estimates as a function of confidence level for the unpacking (total given last) and no unpacking (total given first) conditions.

Study 3
There were no gender differences in estimations of duration (ps > .11).

Confidence
Confidence in the estimates of duration was compared using a 5 (event being estimated) x 2 (Unpacking) mixed-model ANOVA. There was a significant effect of event, with confidence varying dependent on what was being estimated, F(4,340) = 8.50, p < .001, η 2 = .09. Post hoc analyses indicated that confidence was lowest for duration of the last segment of the video, examining the tablet and leaving, but somewhat equivalently higher for all other aspects (see Table 3). Confidence was not influenced by the unpacking manipulation, ps > .27, η 2 s < .02. The average accuracy of the time estimates was collapsed across segment and participants and plotted as a function of confidence and unpacking condition. As can be seen in Figure 3, accuracy increased with confidence in both unpacking conditions. There was no significant difference in confidence-accuracy calibration as evident by the overlapping error bars. Unpacking in Study 2 and anchoring in Study 1 tended to increase confidence-accuracy calibration, although the differences were not statistically significant. In Study 3, accuracy increased with confidence, regardless of unpacking condition.