Hindsight bias in metamemory: outcome knowledge influences the recollection of judgments of learning

ABSTRACT

Hindsight bias describes people’s tendency to overestimate how accurately they have predicted an event’s outcome after obtaining knowledge about it. Outcome knowledge has been shown to influence various forms of judgments, but it is unclear whether outcome knowledge also produces a hindsight bias on Judgments of Learning (JOLs). Three experiments tested whether people overestimated the accuracy of their memory predictions after obtaining knowledge about their actual memory performance. In all experiments, participants studied 60 cue-target word pairs, made a JOL for each word pair, and tried to recall the targets in a cued-recall test. In Experiments 1a and 1b, people recollected their original JOLs after attempting to recall each target, that is, after they obtained outcome knowledge for all items. In Experiments 2 and 3, people recollected their original JOLs in a separate phase after attempting to recall half the targets so that they had outcome knowledge for some but not all items. In all experiments, recollected JOLs were closer to actual memory performance than original JOLs for items with outcome knowledge only. Thus, outcome knowledge produced a hindsight bias on JOLs. Our results demonstrate that people overestimate the accuracy of their memory predictions in hindsight.

KEYWORDS:

• Metamemory
• metacognition
• judgments of learning
• hindsight bias
• outcome knowledge

Disclosure statement

No potential conflict of interest was reported by the author(s).

The data for all experiments are available at https://osf.io/6jpcy and Experiments 2 and 3 were preregistered (https://osf.io/nehf2; https://osf.io/mrkzq).

Notes

1 In many studies on the hindsight bias, correctly recollected judgments (ROJ = OJ) are excluded when computing ΔHB (see Pohl, 2007; Pohl & Erdfelder, 2017). In the experiments reported here, all results were unchanged when excluding correctly recollected judgments.

2 Notably, the three relevant effect sizes for the hindsight bias found in Experiment 1a were ${\eta }_{\mathrm{p}}^2$ = .19 for the interaction between JOL type and recall success, d = 0.63 for the difference in resolution, and d = 0.44 for ΔHB. To ensure sufficient power, we assumed a medium effect to be able to detect effects in all three measures.

3 We did not specify predictions for ΔHB in the pre-registrations for Experiments 2 and 3. For reasons of consistency to Experiments 1a and 1b, we included ΔHB as a standard measure of hindsight bias magnitude (Pohl, 1992; Pohl & Erdfelder, 2017).

4 In the pre-registrations for Experiments 2 and 3, we did not specify that we planned to exclude participants with zero or perfect recall performance. For reasons of consistency, we applied the same exclusion criteria used in Experiments 1a and 1b. Importantly, including participants with zero or perfect recall performance from Experiments 2 and 3 did not change the reported results.

5 In the pre-registration, we specified that RJOLs would be a compromise between the anchor level and actual memory performance. More specifically, we expected lower RJOLs than OJOLs in the high-anchor group and higher RJOLs than OJOLs in the low-anchor group of items without outcome knowledge, resulting in an interaction between JOL type and recall success, that is qualitatively different from the hindsight bias interaction between JOL type and recall. Although this prediction was not supported, the absence of the interaction here underlines the interpretation that the hindsight bias on JOLs is limited to items with outcome knowledge.

Additional information

Funding

This work was supported by the University of Mannheim’s Graduate School of Economic and Social Sciences and by a grant from the Deutsche Forschungsgemeinschaft (DFG) to Monika Undorf (UN 345/1-3).