Does More Accurate Confidence in Memory Change Reminder Choices?

Leaving an object beside the door transfers part of a future action to the environment. Its position can remind you what to do when the time comes. Using such support and deciding whether to use it are different problems, however. A reminder may improve success while choosing it at every opportunity may not maximize a particular objective. Before asking whether reminders work, it helps to specify what the choice is supposed to achieve.

An experiment published by Nicole C. Engeler and Sam J. Gilbert in 2020 separated judgments about memory from decisions to use external support. The researchers tried to make people's performance predictions more accurate by providing feedback on those predictions. They then tested whether improved self-evaluation changed reminder preferences. The outcome suggests that evaluating one's ability and selecting a strategy are not controlled by a single switch. [1]

Writing a future action into the environment

Participants dragged numbered circles out of a box in ascending order. A new number appeared when a circle left. Some new circles initially appeared in another color, then faded to yellow after two seconds. When their turn arrived, they had to be moved to a designated alternative edge. Participants therefore followed the numerical sequence while retaining intentions for actions not yet due. [1]

An external reminder could be created by moving a special circle near its assigned edge in advance. This did not complete the action early. Its position later supplied a cue when the number's turn arrived. The experiment thus compared retaining an intention internally with attaching it to a visible arrangement. The support actually tested was spatial positioning on a screen, not a smartphone alarm application.

Each trial contained twenty-five circles, including ten special targets. Holding several intentions at once made support useful. Accuracy in forced reminder trials was approximately 96 percent in both groups, compared with averages of approximately 50 to 56 percent without reminders. A bias toward reminder choice therefore did not mean reminders were ineffective. They substantially improved success on the target task. [1]

More correct responses need not mean more points

The researchers introduced a point system to evaluate choices. Correct targets were worth ten points when participants relied on internal memory. With reminders, each correct target was worth between one and nine points. Participants were instructed to select the strategy they expected to yield the most points. Accuracy and reward were thereby separated: external support could be more reliable while earning less for every correct response. [1]

Consider an illustration that is not experimental data. Suppose unaided accuracy is one-half and reminder-supported accuracy is perfect. Expected unaided reward is then five points per target. If a supported correct response is worth five points, the options have equal average value. Five is not the optimal number of reminders; it is the reward value of one correct supported target. The study's indifference threshold is based on this kind of comparison.

Performance before preference

Participants first completed trials with the strategy imposed: five using internal memory and four using external reminders. These forced trials allowed the researchers to calculate a point-based threshold from each participant's accuracy with the two strategies. Nine choice trials followed, offering the different reminder values. The values at which participants actually selected support were compared with the performance-based threshold. [1]

Under this definition, reminder bias means selecting the external strategy at lower reminder rewards than expected for maximizing points. Optimal consequently has a restricted meaning: increasing average score. Preferences concerning mental effort, discomfort about errors, or reduced variability are not assigned separate values in that benchmark. Departing from it should not automatically be described as irrationality throughout everyday life.

Comparing predictions with outcomes

During forced trials, the training group predicted the percentage of targets they would complete correctly. Afterward, participants saw both their prediction and their actual performance, with feedback indicating underestimation, overestimation, or accurate estimation. The control group did not receive this prediction-and-metacognitive-feedback package. That does not mean controls saw no task information: correctness signals and point information were also part of the procedure. [1]

The training group's average performance judgments became better aligned with actual performance. Controls underestimated their abilities, particularly unaided memory. But an accurate group average does not show that every individual predicted perfectly. The outcome concerned global estimates of success in this task, not knowing whether every particular intention would be remembered or demonstrating increased memory capacity.

Choices did not show the predicted change

The anticipated second step did not emerge: better-calibrated self-evaluation did not eliminate reminder bias relative to the point benchmark. Both groups showed the bias. Its numerical value was even larger in the training group, but that reverse difference cannot be presented as evidence that training worsened choices. The planned one-tailed test addressed the predicted reduction, and the authors did not establish a reverse-direction effect. [1]

This does not make beliefs about memory irrelevant. The experiment shows that one intervention changed average self-evaluation without demonstrating the expected improvement in choice. Avoiding mental effort, preferring reliable outcomes, and disliking errors are possible explanations. They were not individually manipulated here to establish their causal roles. Saying that several factors may contribute is different from identifying one hidden explanation as proven.

The distinction also prevents a misleading interpretation of training success. If an intervention improves one measure but not another, the first improvement remains meaningful. What fails is the assumption that changing the first outcome necessarily changes the second. In this case, confidence calibration and support selection needed separate evaluation rather than one being treated as a substitute for the other.

Where the interpretation must stop

The study tested 133 people and retained 116 after applying its specified exclusion criteria. Those criteria included task performance and patterns of strategy choice. Participants were online workers reporting a United States location, and payment did not depend on points earned. Instructions to maximize points should therefore not be equated with actual financial loss or the consequences of forgetting an important everyday obligation. [1]

The forced-performance phase also preceded the choice phase. If further practice or fatigue changed accuracy, the threshold derived earlier might not fully represent later performance. Training combined making predictions with feedback, so the contribution of each component was not isolated. Nor was this a long-term follow-up showing durable changes in daily reminder habits. The observations concern one session and the particular support arrangement used within it.

The findings are not a recommendation to switch reminders off. The stronger conclusion is methodological and psychological: actual memory performance, beliefs about that performance, and decisions to use support need to be measured separately. A useful tool can remain attractive even after people estimate their unsupported ability more accurately. Understanding that attraction requires specifying both the benefits the tool supplies and the objective against which its use is judged. [1]

Sources

[1] Nicole C. Engeler and Sam J. Gilbert (2020), The effect of metacognitive training on confidence and strategic reminder setting, PLOS ONE 15(10): e0240858. Open-access primary research.

Source access and factual review: 15 October 2026. The main methods, results, and discussion were directly read; the authors' reported preregistration, raw data, and analysis code were not independently audited. Original explanation based on the Creative Commons Attribution-licensed research, attributed to Engeler and Gilbert. The numerical reward illustration is explanatory, not the researchers' data.