Great Expectations: Anticipating a Reminder Influences Prospective Memory Encoding and Unaided Retrieval

Abstract

Research in the retrospective memory domain has shown that individuals encode information less effortfully when expecting a reminder system (i.e., external store) to be available at the test. Critically, this expectation leads to worse memory performance when the reminder is unexpectedly removed. The present study examined whether these findings extend to prospective memory (PM) intentions, which are thought to maintain a privileged status in memory and therefore may be less sensitive to expectancy effects. Participants formed the intention to make a special PM response to target items across four ongoing task blocks. Study duration (Experiments 1 and 3), pupil size (Experiment 2), and self-report (Experiments 1–3) indexed encoding effort while learning these targets. Participants had reminders available across the first three blocks (i.e., targets listed at the top of the screen), but not on the fourth. Critically, only one condition was informed that they would not have a reminder prior to encoding targets in the fourth block. Results showed that expecting a reminder lowered objective (Experiments 1 and 3) and subjective (Experiments 1–3) encoding effort and reduced unaided PM retrieval (Experiments 1–3) in the fourth block, independent of memory load (Experiments 3). Objective (Experiments 1 and 3) and subjective (Experiments 1–3) effort also partially mediated the influence of expectations on unaided PM retrieval. These findings suggest PM and retrospective memory encoding operate similarly and that participants can alter learning to more effectively commit PM targets to memory when reminders are not expected.

Prospective memory (PM) is the process of remembering to complete a planned action at the appropriate time in the future, such as taking medication with breakfast. Successful functioning in the modern world requires managing countless intentions over a given week. For example, one must be able to juggle work deadlines, social plans, doctor’s appointments, and taking medications. While previous research has shown that effortful PM encoding strategies lead to better PM retrieval (e.g., Gollwitzer, 1999), recent work has suggested that low-effort encoding can still be effective (Scullin et al., 2018). PM reminders available at retrieval can effectively improve PM retrieval (e.g., Einstein & McDaniel, 1990; Peper et al., 2023) despite recent work in the retrospective memory domain that found expecting a reminder reduces effort at encoding (e.g., Kelly & Risko, 2019). The primary goal of the present study is to use reminder expectations to examine whether PM retrieval benefits from more strategic and effortful encoding (i.e., encoding effort hypothesis).

Successful PM requires effectively encoding new intentions into memory, maintaining the intention in long-term memory over a delay, and retrieving the intention from memory at the appropriate time (Kliegel et al., 2000). In a typical laboratory event-based PM task, participants complete an ongoing task (e.g., one vs. two syllable judgments) and form an intention to make a special keypress (e.g., spacebar) whenever they see any words from a specific set within the next ongoing task. Participants then encode the target words they should remember for later (e.g., pepper, shoe, horse). The intention must then be maintained over a retention interval (e.g., 60 s math distractor), after which participants complete a PM task with target words embedded in the ongoing task. The intention must then be retrieved upon seeing one of the target words. PM retrieval is measured by the proportion of targets that receive a successful PM response. While competing theories of prospective memory differ in the proposed mechanisms underlying retrieval (Einstein & McDaniel, 2005; Smith, 2003; Strickland et al., 2018), all agree encoding is important for successful retrieval.

Prospective Memory Encoding

Several studies have found effortful encoding strategies improve PM retrieval. For example, an implementation intention is an effortful encoding strategy that improves PM retrieval (e.g., Gollwitzer, 1999) by having participants make a verbal statement about their intention, form a mental image of their intention, or both (Scullin et al., 2017). A verbal statement includes both the event (i.e., target) and the action (i.e., PM response) relevant to the intention (e.g., “When I eat my breakfast, then I will take my medication”). In the context of a laboratory event-based PM task, that verbal statement would be “When I see the word ‘pepper’ during the next syllable judgment task, then I will remember to press the spacebar.” Implementation intentions are believed to improve PM by strengthening the association between the PM targets and intended action. Another effortful encoding strategy that can strengthen the memory trace of targets is semantic encoding. McDaniel et al. (1998) had participants in one condition generate synonyms of the PM targets at encoding and found this improved PM retrieval compared to a condition that did not semantically encode the targets. Critically both implementation intentions and semantic processing recruit greater attention and effort at encoding. Conversely, multiple studies have shown that dividing attention at encoding impairs PM retrieval (Einstein et al., 1997; McDaniel et al., 1998). These studies have clearly shown effortful encoding can improve PM retrieval.

More recent evidence, however, suggests that PM encoding need not always be effortful to improve performance (Scullin et al., 2018). Scullin et al. tested whether PM encoding could also occur with minimal thought or effort by inserting thought probes during encoding of PM intentions across eight studies. They had participants encode a single nonspecific category target (e.g., fruits) and form an intention to respond to exemplars from that category (e.g., apple). After studying the target, participants reported what they were thinking about during encoding. Overall, participants reported mind-wandering 42.9% of the time during encoding, while 22.5% of participants barely thought about the PM task. This suggests encoding of PM intentions can occur with minimal thought or effort at a surprisingly high frequency. In their final experiment, Scullin et al. found that longer study duration (i.e., effortful encoding) did not influence PM retrieval, suggesting encoding effort is not always necessary to elicit successful retrieval.

Offloading Memory

One way to compensate for possible consequences of low-effort encoding is through cognitive offloading, which refers to using the external environment to reduce internal cognitive demands (Risko & Gilbert, 2016). Past studies have shown reminders improve PM retrieval in both the laboratory (e.g., Chen et al., 2017; Gilbert, 2015a; Peper et al., 2023; Vortac et al., 1995) and naturalistic tasks (Ihle et al., 2012; Schnitzspahn et al., 2020). Recently, we showed that reminders improve PM retrieval by facilitating the maintenance and retrieval processes (Peper et al., 2023). In a series of experiments, we gave participants reminders by presenting the PM targets at the top of the screen throughout the duration of the PM task. Replicating across four studies, we found that reminders improve PM retrieval, particularly under high memory (i.e., target) load. However, because the focus of that study was on maintenance and retrieval, it is unclear how offloading influences encoding. The present study will use a similar paradigm to test the effect of reminder expectations on PM encoding effort.

Despite the abundance of evidence that reminders can be used to improve memory, reminders can be lost, destroyed, or otherwise fail. People increasingly rely on technology to offload their intentions by, for example, setting reminders on their phones. But what would happen to PM retrieval performance if that phone died when one needed the reminder? Within the retrospective memory domain, Kelly and Risko (2019) tested how a failed reminder affects memory performance. In their experiment, participants completed four blocks of a free recall task. During the first three blocks, participants were able to offload (i.e., write down memory items on a list) during encoding and access the external store (i.e., reminder list) at retrieval. Critically, participants were not able to access the external store at retrieval during the final block. Half of the participants learned they would not have access and did not expect to have the store (i.e., not-expecting reminder condition), whereas the other half expected the store (i.e., expecting reminder condition). The expecting reminder condition captured a situation when a reminder failed (i.e., when one set and anticipated a reminder but later did not have access to it). Their results indicated that participants in the not-expecting reminder condition had better memory performance compared to the expecting reminder condition.

In a more recent study, Kelly and Risko (2022) compared study durations between the expecting and not-expecting reminder conditions. They found that the expecting reminder condition spent less time encoding compared to the not-expecting reminder condition during the final block. A mediation analysis showed that the change in study duration from Block 3 to Block 4 partially mediated the effect of expecting a reminder on memory performance on the final block. These studies provide strong evidence that expecting a reminder reduces encoding effort and unaided retrieval in retrospective memory, which Kelly and Risko referred to as the study effort hypothesis. However, one limitation of this study was that the variability in study durations was restricted by having participants write down each item in the lists. A critical question remains then as to whether unconstrained study durations fully mediate the effect of expecting a reminder on unaided memory or whether another mechanism contributes to the effect.

While PM and retrospective memory share similar processes, there are several principled reasons why the encoding effort hypothesis may not necessarily apply to PM. One reason is that Kelly and Risko (2022) had participants study 20 words in each block list, while PM paradigms often have participants remember only a few intentions at most. Prior research has shown that retrospective memory benefits from offloading depend on memory load (e.g., Storm & Stone, 2015), meaning that similar declines in PM performance from expecting reminders may not be evidenced with lesser loads. Another reason is that Scullin et al. (2018) showed that PM intentions can be successfully retrieved independently of encoding effort. This observation aligns with the intention-superiority effect, which refers to the finding that prospective memories are maintained at a heightened level of activation and are more easily accessible than retrospective memories (Goschke & Kuhl, 1993). Dupre et al. (2024) found that offloaded PM intentions were maintained at similar levels of activation as nonoffloaded intentions during a delay interval between encoding and retrieval, suggesting that offloaded representations can remain equally accessible. Finally, previous research has shown that the repetition of specific targets from encoding during retrieval can produce a discrepancy signal that results in spontaneous (i.e., automatic) retrieval of PM intentions (Lee & McDaniel, 2013). Such automatic processes may be less sensitive to changes in encoding effort based on reminder expectations. Considering the differences between retrospective memory and PM, and the finding that effortful encoding may not be necessary for PM retrieval, there is a need to examine the encoding effort hypothesis in PM.

The Present Study

The present study manipulated reminder expectations to examine whether the encoding effort hypothesis extends into the PM domain. The ongoing task was a syllable judgment task (one vs. two syllables), with a PM intention to make a different keypress to specific learned PM targets. Using a similar procedure as Kelly and Risko (2019, 2022), each experiment contained four PM blocks, each contained a target encoding period, a math distractor, and a PM ongoing task. Three between-subjects conditions completed this overall procedure with minor distinguishing differences. Two reminder conditions had reminders during the first three PM tasks, while a no reminder condition did not. The reminder consisted of all PM targets listed at the top of the screen in yellow font for the duration of the PM task. Critically, none of the conditions had a reminder for the final block. The not-expecting reminder condition was explicitly told prior to fourth block encoding that they would not have a reminder during the final PM task. The expecting reminder condition was explicitly told prior to the fourth block encoding that they would have a reminder during the final PM task (see Figure 1).

Figure 1. General PM Task Procedure.

Note. Squares on the left, middle, and right represent the appearance of the PM task for each block in the no reminder condition, not-expecting reminder condition, and expecting condition, respectively. Note that Experiment 3 did not include a no reminder condition. Critically, in every experiment, all eight targets (or eight target-length-matched masks) encoded for that block were listed at the top of the screen during the entire PM task. Experiment 3 also had conditions in which only two PM targets were studied and listed at the top of the screen. While each PM block consisted of target encoding, a distractor, and the PM task, the dark gray squares for each condition and block only show how the PM task appeared. PM = prospective memory. See the online article for the color version of this figure.

Experiments 1–3 used eight specific PM targets in each block, while Experiment 3 included a two-target condition. We directly measured encoding effort through study duration (Experiments 1 and 3) and pupil size (Experiment 2). We anticipated replicating the findings by Kelly and Risko (2019, 2022) that expecting a reminder reduces encoding effort, which in turn, reduces unaided retrieval (i.e., encoding effort hypothesis). If encoding effort and unaided PM retrieval in the fourth block did not differ between the expecting and not-expecting conditions, however, this would suggest that effortful encoding is not necessary to elicit PM retrieval.

Preregistration Statement

The research in the present study was conducted following ethical guidelines approved by the Institutional Review Board at the University of Texas at Arlington. We preregistered how our sample sizes were determined, all exclusionary criteria, all data transformations, and all manipulations¹. Analyses that were not preregistered are specified as exploratory. Open Science Framework preregistrations are linked in the introductions for each experiment.

Experiment 1

Experiment 1 directly assessed the effect of reminder expectations on PM encoding effort by measuring unconstrained study duration. Participants completed four PM blocks where they encoded a set of eight targets (self-paced) unique to each block before completing the PM task, where four of the targets were embedded. The expecting and not-expecting conditions had reminders for the first three PM tasks, while the no reminder condition did not. In the fourth block, the expecting condition believed they would have a reminder, while the not-expecting reminder and no reminder conditions believed they would not.

Based on previous research on the efficacy of reminders in event-based PM (Peper et al., 2023), we predicted reminders would improve PM retrieval in the first three blocks. According to the encoding effort hypothesis, participants should adjust their encoding effort when they expect to have a reminder. This may be particularly noticeable after they successfully retrieve the intention in Block 1. That is, participants with and without a reminder may encode for similar durations during the first block, but those in the reminder conditions should encode targets for less time than participants in the no reminder condition in Blocks 2 and 3. The encoding effort hypothesis is tested by the critical comparison in Block 4. Those in the not-expecting condition should adjust encoding effort in Block 4, resulting in longer study durations (comparable to the no reminder condition). Those in the expecting reminder condition should continue to encode for less time. Critically, the encoding effort hypothesis would predict the expecting condition to have shorter study durations and worse unaided PM retrieval in Block 4 than the not-expecting condition, which would also extend Kelly and Risko’s (2022) work into the PM domain. These hypotheses were preregistered² (https://osf.io/wnpcf).

Method

Participants

The preregistered power analysis was conducted with G*Power based on the medium effect size, which recommended 159 participants. The goal was to obtain .80 power at .05 α probability for an effect size of f = .25 in a one-way analysis of variance (ANOVA) with three groups. We initially collected 175 participants, with the anticipation that some participants would be excluded based on preregistered criteria (described below). Our final sample consisted of 169 participants after exclusions. Participants (aged 17–32) were undergraduates at the University of Texas at Arlington awarded with class credit for their participation. All participants were randomly assigned to either the expecting reminder condition (n = 57), not-expecting reminder condition (n = 57), or no reminder condition (n = 55).

Design

A 3 (condition: expecting, not-expecting, and no reminder; between) × 4 (Block: 1, 2, 3, and 4; within) mixed factorial design was used (Figure 1). Reminder conditions were manipulated between subjects and differed by the presence of a reminder in the first three blocks and the instructions prior to encoding the fourth block. Block was manipulated within subjects such that all participants completed four PM blocks. During the first three blocks, the expecting and not-expecting reminder conditions had reminders for the targets during the PM task and the no reminder condition had no reminder. Before the fourth block, encoding the expecting condition was told they would have a reminder, the not-expecting condition was told they would not have a reminder, and the no reminder condition received the same instructions they had for previous blocks.

Materials

There were 380 words selected from the English Lexicon Project that served as ongoing task stimuli (Balota et al., 2007). Half of the words had one syllable and half had two syllables, each ranging from five to seven letters in length. Twenty words were used for the first practice and reused for the second practice. For the PM blocks, 360 words made up the ongoing task stimuli. Thirty-two additional words were selected as PM targets and matched the ongoing task stimuli in terms of syllable count and letter length. That is, five targets had five letters and one syllable, five had five letters and two syllables, six had six letters and one syllable, six had six letters and two syllables, five had seven letters and one syllable, and five had seven letters and two syllables. Thirty-two unique words with characteristics matching the PM targets were selected as new words in the recognition memory task. All stimuli were presented in uppercase black font, centered on the screen with a gray background. Reminders for all eight targets were centered around the top of the screen in yellow font. Whenever participants had no reminder, masks (e.g., ######) were centered around the top of the screen in yellow font of the same letter length as the targets for that block. This was done to equate screen luminosity across conditions and blocks, which was necessary in Experiment 2 when pupillometry was measured. The math distractor consisted of multiplication problems with answers ranging from 1 to 500.

Data collection was completed in person. Participants consented before answering demographic questions. The experiment was programmed using Python.

Procedure

Broadly, the experiment consisted of two ongoing task practices, four PM blocks, a postexperimental questionnaire, and a recognition memory task. Participants were instructed to make their syllable judgments on English words as quickly and accurately as they could by pressing the “F” key for one-syllable word (e.g., storm) and the “J” key for two-syllable words (e.g., pepper). For the PM task, participants completed the syllable judgment ongoing task with embedded PM target words. The ongoing task and PM tasks were self-paced. Between trials, there was a 500 ms fixation cross before the next word appeared.

After consenting and completing the demographics form, participants sat down at a computer and the lights turned off in the experiment room. Lights were off to equate the procedures across all three experiments in anticipation of using pupillometry in Experiment 2. The experiment began with ongoing task instructions before participants completed a 20-trial practice with accuracy and response time feedback. We intended to have a minimum required accuracy of 75% before participants could move onto the second practice, but an error in the program allowed participants to continue if their accuracy was below that threshold. Then participants completed a second practice of 20 ongoing task trials without feedback.

The instructions for the PM task were given to the participants after they finished both practices. Participants were instructed that we were also interested in their ability to remember to perform an action in the future. Instructions stated they would learn eight target words and they were to press the spacebar instead of making a syllable judgment if they ever saw one of the eight words in the next ongoing task. They were also told they would complete four of these PM blocks. In the two reminder conditions, participants learned they would have a reminder to help them remember the targets. After the PM instructions, all participants summarized the instructions to an experimenter and then took a quiz on the instructions to verify they understood the PM task. We have found in past online studies that instruction quizzes helped to facilitate understanding (Peper et al., 2023). Accuracy on the quiz needed to be 100% for the participants to move on. Otherwise, participants had to go back and read the PM instructions and complete the quiz again.

PM blocks consisted of an encoding period, a math distractor, and a PM task. The 32 PM targets were randomly assigned eight to a block for each participant. During encoding, participants studied eight target words one at a time for as long as they wished (i.e., completely self-paced study). After studying the targets, participants were reminded of the PM task instructions before beginning a math distractor. The math distractor had participants complete multiplication problems for 60s. The purpose of this was to ensure memory for the intention was moved to long-term memory instead of being maintained in working memory (Graf & Uttl, 2001). After finishing the math distractor, participants began the 84 trial PM task with four of the eight targets appearing on trials 20, 40, 60, and 80. The reminder conditions had the eight targets listed at the top of the screen throughout the entire PM task while the no reminder control had to rely on internal memory alone. Upon completing the PM task, participants were told they finished that block and did not need to remember the eight targets from that block any longer.

The first three PM blocks occurred exactly as they were described above, but the critical manipulation happened during the instruction period immediately before fourth block encoding. In the expecting reminder condition, participants were told they would have a reminder for the target words (i.e., same instructions as the previous blocks). Participants in the not-expecting reminder condition were told they would not have a reminder to help them remember the target words. That information was highlighted in a different color to draw attention to this fact. The no reminder condition received the same instructions they had previously with no mention of a reminder. Importantly, after the math distractor, all participants completed the fourth PM task without a target reminder.

After the four PM blocks, participants completed a postexperimental questionnaire. They answered questions assessing their retrospective memory for the PM task, and the reminder conditions answered questions about the reminders, including a manipulation check that assessed whether they believed they would have a reminder in the fourth block. Two multiple-choice questions asked participants whether they remembered they were supposed to do a secondary task in addition to the ongoing task and what they were supposed to do for the secondary task. An incorrect response for that question and having a zero for PM retrieval across all four blocks signified a retrospective memory error (see exclusionary criteria). The reminder conditions were also asked whether they expected a reminder on the final block (i.e., manipulation check). All participants were asked to rate from 1 (not at all effortfully) to 7 (extremely effortfully) how effortfully they studied the targets in Block 4. Finally, participants completed a 64-trial recognition task for the PM targets.

Preregistered Exclusionary Criteria

Participants were excluded from each condition for the following:

1. Both failing to detect any PM targets and forgetting the prospective memory task (n = 0).

2. Getting below 60% accuracy on the PM block ongoing task (n = 4).

   • expecting = 2, not-expecting = 1, no reminder = 1
3. Having PM block ongoing task response times greater than 3 or less than 3 SDs from the group mean (n = 2).

   • expecting = 0, not-expecting = 0, no reminder = 2

4. Making false alarms (i.e., PM response on nontarget trials) on over 15% of trials (n = 0).

Results

Our preregistration lists our analyses as a series of 3 (Condition: not-expecting, expecting, and no reminder) × 4 (Block: 1, 2, 3, and 4) mixed ANOVAs for PM retrieval, study duration, uncontaminated recognition, and ongoing task performance (accuracy and response times). The preregistration was written for a dissertation, and the current article has been narrowed in scope and the results are limited to the critical measures to support the encoding effort hypothesis, which includes PM retrieval and study duration. We also include exploratory analyses on subjective effort. Analyses for all the preregistered dependent variables can be found in the Supplemental Materials.

While the main effect results for both condition and block are confounded by the manipulation in Block 4, a significant 3 × 4 interaction justifies probing further. Upon a significant interaction, one-way ANOVAs tested whether the two reminder conditions differed in the first three blocks during which they experienced an identical procedure. When no differences were observed, participants in the expecting and not-expecting conditions were then combined into a single reminder condition for additional power and ease of interpretation. That is, a series of 2 (reminder: reminder vs. no reminder) × 3 (Block: 1, 2, and 3) mixed ANOVAs were conducted whenever there was (a) a significant 3 × 4 interaction and (b) no differences between reminder conditions in the first three blocks. Post hoc analyses for block effects used Bonferroni-corrected p values of .017. For the second set of analyses probing a significant 3 × 4 interaction, we compared performance separately for the three conditions only in Block 4. We interpreted the Greenhouse–Geisser corrected values in instances in which Mauchly’s test indicated that the assumption of sphericity was violated (in which, the F-test is subscripted using “F_GG”) and used Bonferroni-corrected p values of .017.

To verify that our manipulation was working correctly, we examined how many participants believed the specific reminder instructions prior to Block 4 (proportion who did not believe instructions: not-expecting: n = 9 [16%]; expecting: n = 17 [30%]). Participants in the expecting reminder condition failed the manipulation check if they answered “no” to the question of whether they expected to have a reminder during the fourth PM task. Participants in the not-expecting reminder condition failed the manipulation check if they answered “yes” to the question of whether they expected to have a reminder during the fourth PM task. Results were the same regardless of whether the participants who failed the manipulation check were excluded. We therefore report analyses that include all participants.

PM Retrieval

PM retrieval was calculated as the proportion of target trials (out of four total) on which participants pressed the “spacebar” rather than making an ongoing task response in each block. Figure 2 presents the PM retrieval results separately for each condition and block.

Figure 2. PM Retrieval Across All Four Blocks Separated by Condition in Experiment 1.

Note. The circles represent mean performance, and the error bars reflect plus or minus standard error. PM = prospective memory. See the online article for the color version of this figure.

All Blocks.

For PM retrieval, there was a significant interaction between condition and block, F_GG(5.79, 480.47) = 9.09, p < .001, ${\eta }_p^2=.099$ that justified the following probing analyses. A series of one-way ANOVAs tested for differences between the two reminder conditions in the first three blocks when they were identical. Importantly, there were no differences in PM retrieval between reminder conditions (F’s < 1.12, p’s > .293).

Blocks 1–3.

In Blocks 1–3, the two reminder conditions were collapsed into a single reminder condition. Reminders improved PM retrieval overall, F(1, 167) = 18.21, p < .001, ${\eta }_p^2=.098$. There was no effect of block (F < 1). There was a significant reminder by block interaction, F(2, 334) = 3.73, p = .025, ${\eta }_p^2=.022$. Post hoc comparisons to explore the interaction used a Bonferroni-adjusted significance level of p < .017. While there was no PM retrieval difference between reminder and no reminder conditions in Block 1, F(1, 167) = 4.25, p = .041, ${\eta }_p^2=.025$, reminders improved PM retrieval in Block 2, F(1, 167) = 6.11, p = .014, ${\eta }_p^2=.035$, and Block 3, F(1, 167) = 26.60, p < .001, ${\eta }_p^2=.137$.

Block 4.

In Block 4, there was a difference in PM retrieval between conditions, F(2, 166) = 7.18, p = .001, ${\eta }_p^2=.08$. Block 4 analyses supported the encoding effort hypothesis by showing PM retrieval was better in not-expecting condition compared to the expecting condition, F(1, 113) = 13.38, p < .001, ${\eta }_p^2=.106$. However, PM retrieval in Block 4 was no different between the not-expecting condition and the no reminder condition (F < 1).

Study Duration

Study durations were calculated by averaging the length of time each target was studied. Figure 3 presents the results separately for each condition and block.

Figure 3. Study Duration Across All Four Blocks Separated by Condition in Experiment 1.

Note. Study duration refers to the average time in milliseconds spent studying each target. The circles represent mean study duration, and the error bars reflect plus or minus standard error. See the online article for the color version of this figure.

All Blocks.

For study durations, there was a significant interaction between condition and block, F_GG(4.32, 358.83) = 9.75, p < .001, ${\eta }_p^2=.105$ that justified the following probing analyses. A series of one-way ANOVAs tested for differences between the two reminder conditions in the first three blocks when they were identical. Importantly, there were no differences in study duration between reminder conditions, F’s < 2.03, p’s > .157.

Blocks 1–3.

In Blocks 1–3, the two reminder conditions were collapsed into a single reminder condition. Reminders reduced study durations, F(1, 167) = 26.60, p < .001, ${\eta }_p^2=.137$. Study durations also decreased across block, F_GG(1.39, 232.86) = 14.29, p < .001, ${\eta }_p^2=.079$. Critically, there was an interaction between reminder and block, F_GG(1.39, 232.86) = 6.50, p = .005, ${\eta }_p^2=.037$. Post hoc comparisons to explore the interaction (Bonferroni p < .017) showed no differences between reminder and no reminder conditions in Block 1 (F < 1). However, reminders reduced study durations in Block 2, F(1, 168) = 7.99, p = .005, ${\eta }_p^2=.045$ and Block 3, F(1, 168) = 5.96, p = .016, ${\eta }_p^2=.034$. This is consistent with the encoding effort hypothesis that expecting a reminder reduces encoding effort after the first experience of retrieval with a reminder.

Block 4.

In Block 4, there was a difference in study duration between conditions, F(2, 166) = 3.69, p = .037, ${\eta }_p^2=.039$. Block 4 analyses supported the encoding effort hypothesis by showing the not-expecting condition had longer study durations than the expecting condition, F(1, 112) = 11.32, p = .001, ${\eta }_p^2=.092$. However, study durations in Block 4 were no different between the not-expecting condition and the no reminder condition (F < 1).

Exploratory Analysis of Subjective Effort

Block 4.

A one-way ANOVA was conducted to test whether self-reported Block 4 encoding effort differed between conditions when assessed during the postexperimental questionnaire. For subjective effort, there was a significant difference between conditions, F(2, 166) = 15.80, p < .001, ${\eta }_p^2=.160$. To probe the effect, we compared the not-expecting and expecting conditions, as well as the not-expecting and no reminder conditions. The not-expecting condition (M = 5.11, SE = 0.15) reported more subjective effort than the expecting condition (M = 3.81, SE = 0.21) during Block 4 encoding, F(1, 112) = 24.65, p < .001, ${\eta }_p^2=.180$, but there was no difference between the no reminder condition (M = 4.95, SE = 0.16) and the not-expecting condition (F < 1).

Exploratory Mediation of Study Duration on PM Retrieval

Block 4.

An exploratory mediation analysis was conducted using the PROCESS Macro 4.2 Model 4 for SPSS (Hayes, 2017) to test whether the effect of expecting a reminder on PM retrieval in Block 4 was mediated by encoding effort. The reminder condition variable (x) only included the expecting and not-expecting conditions to capture the effect of expecting a reminder. For this analysis, study durations were converted to seconds from milliseconds for interpretation of b coefficients. The path from reminder condition to PM retrieval (c) was significant, b = −0.19, t(112) = −3.53, p < .001, and the association between encoding effort to PM retrieval was significant, b = 0.03, t(111) = 4.39, p < .001. The link between reminder condition to encoding effort (a) was also significant, b = −2.71, t(112) = −3.37, p = .001. For the c’ path, the effect of expecting a reminder on PM retrieval was still significant after accounting for encoding effort as a mediator, b = −0.12, t(111) = −2.30, p = .023. A bootstrap approach with 5,000 samples revealed an indirect effect of expecting a reminder on PM retrieval, b = −0.07, SE = 0.02, 95% CI [−0.12, −0.03], using confidence intervals (i.e., not overlapping with zero) as an index of significance. These results suggest that the consequences of expecting a reminder on unaided PM retrieval are mediated, in part, by encoding effort.

Exploratory Mediation of Subjective Effort on PM Retrieval

Block 4.

An exploratory mediation analysis was conducted to test whether the effect of expecting a reminder on PM retrieval in Block 4 was mediated by subjective effort during Block 4 encoding. The reminder condition variable (x) only included the expecting and not-expecting conditions to capture the effect of expecting a reminder. The path from reminder condition to PM retrieval (c) was significant, b = −0.19, t(112) = −3.53, p < .001, and the association between subjective encoding effort to PM retrieval was significant, b = 0.06, t(111) = 3.14, p = .002. The link between reminder condition to encoding effort (a) was also significant, b = −1.30, t(112) = −4.97, p < .001. For the c’ path, the effect of expecting a reminder on PM retrieval was still significant after accounting for encoding effort as a mediator, b = −0.12, t(111) = −1.99, p = .049. A bootstrap approach with 5,000 samples revealed an indirect effect of expecting a reminder on PM retrieval, b = −0.08, SE = 0.03, 95% CI [−0.15, −0.02], using confidence intervals (i.e., not overlapping with zero) as an index of significance. These results suggest that the consequences of expecting a reminder on unaided PM retrieval are mediated, in part, by subjective encoding effort.

Exploratory Strategy Usage

Participants were asked at the end of the experiment if they would change strategies (yes/no) upon studying a new list if they knew a reminder would not be available. A chi-square test found that participants in the expecting reminder (M = 0.83) condition were more likely to indicate they would switch strategies than those in the not-expecting (M = 0.57) condition, χ² (1, N = 117) = 9.54, p = .002.

Discussion

Experiment 1 manipulated reminder expectations and examined the role of encoding effort on PM retrieval. Replicating previous work using a similar task (Peper et al., 2023), reminders improved PM retrieval. Critically, after a single experience with reminders, participants expecting a reminder spent less effort encoding the next set of targets, as evidenced by shorter study durations in the reminder conditions selectively during the second and third blocks. This is consistent with the encoding effort hypothesis that participants noticed the success of PM retrieval in the first block and then encoded with less effort in the second and third blocks. Critically, during Block 4, participants expecting a reminder studied targets for less time and reported less subjective effort at encoding compared to participants who did not expect a reminder. While the reduction in effort did not reduce PM retrieval when the reminders were present, lower encoding effort worsened PM retrieval when the reminders were unexpectedly unavailable for the expecting condition in Block 4. Exploratory mediations found that the effect of expecting a reminder on unaided PM retrieval was partially mediated by encoding effort (study durations and subjective effort), providing evidence of a causal effect. It is important to draw attention to the partial mediation that replicated Kelly and Risko’s (2022) work despite study durations that were not constrained by writing down each item. Overall this pattern of result is consistent with the encoding effort hypothesis, replicating in PM the pattern observed in retrospective memory (e.g., Kelly & Risko, 2022) and suggesting shared mechanisms underlie the effect of expecting reminders in both retrospective memory and PM despite the intention-superiority effect (Goschke & Kuhl, 1993) and fewer to-be-remembered items. That is, expecting a reminder reduces encoding effort but their presence at retrieval offsets memory deficits that may otherwise occur in both domains.

Experiment 2

Experiment 2 was designed as a conceptual replication of Experiment 1 using a physiological measurement of encoding effort that provided an assessment of how attentional effort at encoding influences PM retrieval. Pupillometry is a methodological technique that can be used to assess attentional effort (Unsworth & Miller, 2021). A task-evoked pupillary response (TEPR) is the degree of pupil dilation in response to a stimulus and is a reliable measure of attentional effort and task engagement (Beatty & Lucero-Wagoner, 2000; Just & Carpenter, 1993). Extant research has observed that attentional effort at encoding is positively related to recall in a paired associates task (Miller & Unsworth, 2021), a free recall task (Ariel & Castel, 2014), and a recognition task (Papesh et al., 2012). For example, Miller and Unsworth had participants encode word pairs (e.g., pepper–table) and at test presented the first word (“pepper-?”) to prompt retrieval of the associated second word (“table”). They found that participants with higher rates of successful retrieval showed larger TEPRs at encoding. While these findings clearly demonstrate a relationship between attentional effort at encoding and memory retrieval, these findings have not been extended into the PM domain. During encoding in each block of Experiment 2, every target was studied for 5s. Reminder condition was manipulated between subjects, with all participants randomly assigned to either the no reminder condition, not-expecting condition, or expecting condition. According to the encoding effort hypothesis, encoding TEPRs should not differ in the first block, but the reminder conditions should have smaller encoding TEPRs in Blocks 2 and 3. Then in Block 4, those in the expecting condition would encode targets less effortfully (i.e., smaller TEPRs) than the not-expecting condition in Block 4 and this would lead to worse PM retrieval. These hypotheses were preregistered³ (https://osf.io/t8xwz).

Method

Participants

The preregistered power analysis was conducted with G*Power based on the large effect size (d = .80) found in Ariel and Castel (2014) between high-value and medium/low-value words at encoding, which recommended 123 participants. The goal was to obtain .90 power at .025 α probability for an effect size of d = .80 between the expecting and not-expecting conditions and the not-expecting and no reminder conditions in Block 4 TEPRs. Our rationale was that a power analysis for the post hoc comparisons between conditions after accounting for the Bonferroni correction would be more conservative. We initially collected data from 140 participants with the anticipation that some would be excluded based on preregistered criteria. Our final sample was 133 participants after exclusions. Participants (aged 17–37) were undergraduates at the University of Texas at Arlington awarded with class credit for their participation. All participants were randomly assigned to either the expecting reminder condition (n = 45), the not-expecting reminder condition (n = 43), or the no reminder condition (n = 45).

Design and Materials

The overall design and all materials were identical to those in Experiment 1.

Procedure

While most of the procedure was identical to Experiment 1, there were a few differences in Experiment 2 that primarily addressed issues of timing and luminosity that are necessary to consider when using pupillometry. During encoding, the target study duration was 5s for each target. A mask fixation of seven characters, the maximum letter length for each target (Miller & Unsworth, 2021), was placed between each target during encoding. Each ongoing and PM task trial lasted 3,000 ms, and a mask (i.e., #####) appeared after a response was made with the same number of characters as the word stimulus on that trial. For example, if a participant pressed the F key for the word broad after 1,000 ms, then a five-character mask appeared for 2,000 ms before the 500 ms fixation appeared that demarcated the next trial.

Eye-Tracking and Pupillometry

Pupillometry data were collected by a GazePoint GP3 eye tracker. Participants sat approximately 70 cm away from the eye tracker with their chin resting in a chin rest to prevent movement. Before starting the task, participants completed a 5-point calibration. The eye tracker sampled pupil dilation and gaze binocularly at 60 Hz. Experiment rooms were windowless, and the lights were turned off so that pupil dilation was more sensitive to variations in attentional effort (Miller & Unsworth, 2021). Pupillometry data were calculated separately for encoding trials (5,000 ms) and PM block trials (3,000 ms).

Preregistered Exclusionary Criteria

Participants were excluded from each condition for the following:

1. Both failing to detect any PM targets and forgetting the prospective memory task (n = 0).

2. Getting below 60% accuracy on the PM block ongoing task (n = 6).

   • expecting = 2, not-expecting = 3, no reminder = 1
3. Having PM block ongoing task response times greater than 3 or less than 3 SDs from the group mean (n = 1).

   • expecting = 1, not-expecting = 0, no reminder = 0

4. Making false alarms (i.e., PM response on nontarget trials) on over 15% of trials (n = 0).

Results

Behavioral Analyses

For the behavioral data, the analytic approach was identical to Experiment 1, except there were no study durations. PM retrieval was submitted first to a 3 (Condition: not-expecting, expecting, and no reminder) × 4 (Block: 1, 2, 3, and 4) mixed ANOVA. When there was a significant interaction, one-way ANOVAs tested for differences between the two reminder conditions in the first three blocks when the conditions were identical. The not-expecting and expecting conditions were then combined into a single reminder condition if no differences between them were found to compare performance to the no reminder condition across the first three blocks. Subsequently, Block 4 performance was compared across all three conditions. Results for recognition memory and ongoing task performance can be found in the Supplemental Materials.

To verify that our manipulation was working correctly, we examined how many participants believed the specific reminder instructions prior to Block 4 (proportion who did not believe instructions: not-expecting: n = 4 [9%]; expecting: n = 13 [29%]). Results were the same regardless of whether the participants who failed the manipulation check were excluded. We therefore report analyses that include all participants.

PM Retrieval

All Blocks.

For PM retrieval, there was a significant interaction between condition and block, F(6, 390) = 10.56, p < .001, ${\eta }_p^2=.140$ that justified the following probing analyses (Figure 4). A series of one-way ANOVAs tested for differences between the two reminder conditions in the first three blocks when they were identical. Importantly, there were no differences in PM retrieval between reminder conditions (F’s < 1.63, p’s > .205).

Figure 4. PM Retrieval Across All Four Blocks Separated by Condition in Experiment 2.

Note. The circles represent mean performance, and the error bars reflect plus or minus standard error. PM = prospective memory. See the online article for the color version of this figure.

Blocks 1–3.

In Blocks 1–3, the two reminder conditions were collapsed into a single reminder condition. Reminders improved PM retrieval overall, F(1, 131) = 19.52, p < .001, ${\eta }_p^2=.130$. There was no effect of block, F(2, 262) = 1.63, p = .198, ${\eta }_p^2=.012$ and no reminder by block interaction (F < 1).

Block 4.

In Block 4, there was a difference in PM retrieval between conditions, F(2, 130) = 12.93, p < .001, ${\eta }_p^2=.166$. Block 4 analyses supported the encoding effort hypothesis by showing PM retrieval was better in not-expecting condition compared to the expecting condition, F(1, 86) = 19.49, p < .001, ${\eta }_p^2=.185$. PM retrieval was no different between the not-expecting condition and the no reminder condition (F < 1).

Eye-Tracking Encoding Analyses

Observations were removed for falling outside of the plausible range for a human pupil diameter (<2 mm or >8 mm; Mathôt, 2018). Participants were excluded from pupil analyses (but not behavioral analyses) if they had missing data (e.g., missing values during blinks) on 50% or more of the trials. This resulted in 11 participant exclusions for encoding TEPRs. We handled missing data for included participants with linear interpolation that filled in observations with values forming the best fitting line between valid observations within a trial (Gross & Dobbins, 2021). A low-pass Butterworth filter was applied to the data to preserve the overall structure while rounding any erratic observations across 10 Hz to minimize potentially erroneous variability between observations (Bowling et al., 2019). After interpolation and smoothing, pupil dilations for each trial were averaged into 200 ms bins and baseline corrected to calculate the TEPRs. The baseline correction involved subtracting pupil diameter from the final 200 ms of the intertrial interval (fixation) from the bins in each trial. The Supplemental Materials also presents details for TEPRs on target trials, nontarget trials, and recognition memory task trials, as well as pretrial pupil diameter during the PM task.

The average encoding change in TEPR across the entire encoding interval, collapsed across condition and block, is plotted in the Supplemental Materials. TEPRs generally declined until about 1,200 ms, which appears to reflect a pupillary light reflex (Binda et al., 2013). We therefore calculated encoding TEPRs using the 1,200 ms bin as the baseline to correct for the pupillary light reflex, similar to other researchers who examined TEPRs at encoding (Miller & Unsworth, 2021). The results were generally similar for encoding trial TEPRs whether we corrected for the pupillary light reflex or not, so we only reported results for TEPRs corrected for the pupillary light reflex.

Encoding TEPRs were submitted to a 3 (Condition: not-expecting, expecting, and no reminder) × 4 (Block: 1, 2, 3, and 4) mixed ANOVA with the same analysis plan as PM retrieval.

Encoding TEPRs

All Blocks.

For encoding TEPRs, there was a no interaction between condition and block (F_GG < 1), no main effect of condition (F < 1), and no main effect of block, F_GG (2.82, 335.41) = 1.15, p = .328, ${\eta }_p^2=.010$ (Figure 5).

Figure 5. Mean TEPRs on Encoding Trials Across All Four Blocks Separated by Condition.

Note. Mean encoding TEPR refers to the average baseline-corrected pupil size in millimeters across the length of each encoding trial after correcting for the pupillary light reflex (i.e., average pupil size from 1,200 ms to 5,000 ms). The circles represent mean pupil size, and the error bars reflect plus or minus standard error. TEPR = task-evoked pupillary response. See the online article for the color version of this figure.

Exploratory Analysis of Subjective Effort

A one-way ANOVA was conducted to test whether self-reported Block 4 encoding effort differed between conditions when assessed during the postexperimental questionnaire. For subjective effort, there was a significant difference between conditions, F(2, 130) = 31.00, p < .001, ${\eta }_p^2=.323$. To probe the effect, we compared the not-expecting and expecting conditions, as well as the not-expecting and no reminder conditions. The not-expecting condition (M = 5.09, SE = 0.21) reported more subjective effort than the expecting condition (M = 3.00, SE = 0.23) during Block 4 encoding, F(1, 86) = 45.11, p < .001, ${\eta }_p^2=.344$, but there was no difference between the no reminder condition (M = 5.00, SE = 0.20) and the not-expecting condition (F < 1).

Exploratory Mediation of Subjective Effort on PM Retrieval

Block 4.

An exploratory mediation analysis similar to Experiment 1 tested whether the effect of expecting a reminder on PM retrieval in Block 4 was mediated by subjective effort during Block 4 encoding. The path from reminder condition to PM retrieval (c) was significant, b = −0.24, t(86) = −4.42, p < .001, and the association between subjective encoding effort to PM retrieval was significant, b = 0.05, t(85) = 2.57, p = .012. The link between reminder condition to encoding effort (a) was also significant, b = −2.09, t(86) = −6.72, p < .001. For the c’ path, the effect of expecting a reminder on PM retrieval was still significant after accounting for encoding effort as a mediator, b = −0.14, t(85) = −2.18, p = .032. A bootstrap approach with 5,000 samples revealed an indirect effect of expecting a reminder on PM retrieval, b = −0.10, SE = 0.04, 95% CI [−0.18, −0.04], using confidence intervals (i.e., not overlapping with zero) as an index of significance. These results suggest that the consequences of expecting a reminder on unaided PM retrieval are mediated, in part, by subjective encoding effort.

Exploratory Strategy Usage

Participants were asked at the end of the experiment if they would change strategies (yes/no) upon studying a new list if they knew a reminder would not be available. A chi-square test found that participants in the expecting reminder (M = 0.88) condition were more likely to indicate they would switch strategies than those in the not-expecting (M = 0.70) condition, χ² (1, N = 94) = 4.51, p = .034.

Discussion

Experiment 2 was a conceptual replication of Experiment 1 to test the effect of reminders and reminder expectations on PM retrieval and encoding effort using pupillometry. The behavioral results in Experiment 2 replicated the pattern observed in Experiment 1. Available reminders improved PM retrieval in Blocks 1–3 and expecting a reminder reduced subjective effort and unaided PM in Block 4. The exploratory mediation analysis revealed subjective effort during Block 4 encoding partially mediated the effect of reminder expectations on unaided PM retrieval. However, using encoding TEPRs as an index of effort at encoding, reminder experience and expectations did not influence encoding effort, unlike Experiment 1. It is possible that reminder expectations did not influence encoding effort, but this is inconsistent with the behavioral results.

An alternative possibility is that eight encoding trials in each block did not provide enough observations to reliably assess encoding effort. Although previous research has shown that TEPRs are a robust measure of attentional effort (Beatty & Lucero-Wagoner, 2000; Just & Carpenter, 1993; Kahneman & Beatty, 1966) and larger TEPRs at encoding are related to better memory retrieval (Ariel & Castel, 2014; Miller & Unsworth, 2021; Papesh et al., 2012), a primary methodological difference between those experiments and the present study is in the number of encoding trials used to calculate the TEPRs. We presented eight target items in each encoding block. Papesh et al. (2012) had participants encode 80 items and Miller and Unsworth (2021) had 50 items and 90 items in their Experiment 1 and 90 items in their Experiment 2. The large standard errors in the present study suggest eight encoding trials may have been too few to observe any stable differences between reminder conditions or across blocks (see Supplemental Materials). Considering the limitation of measuring pupillary responses over so few trials, encoding TEPRs should be interpreted with caution. Given the mixed findings between Experiment 1 (study duration) and Experiment 2 (TEPRs), the subsequent experiment sought to replicate and extend the findings from Experiment 1 using study duration as the primary measure of encoding effort.

Experiment 3

One criticism of the previous experiments is that studying eight PM targets places relatively high demands on retrospective, rather than prospective, memory processes. PM consists of a prospective component associated with remembering that something needs to be done at the appropriate moment and a retrospective component associated with remembering what needs to be done (Einstein & McDaniel, 1990). The prospective component is unique to PM tasks, whereas the retrospective component is common to both PM and retrospective memory tasks. The fact that expecting a reminder in the fourth block reduces target detection could simply reflect deficits in retrospective memory processing, similar to previous work in the retrospective memory domain (e.g., Kelly & Risko, 2019). Scullin et al. (2018) also used a single PM target when they showed no association between study duration and PM retrieval, albeit with nonspecific category targets (e.g., fruit) rather than specific targets (e.g., apple) like the current experiment.

Experiment 3 sought to replicate Experiment 1 using only two targets, which places negligible demands on the retrospective component. Performance was compared between high (eight targets) and low (two targets) retrospective memory load for those expecting or not-expecting a reminder in Block 4. Similar to Experiment 1, we allowed unlimited study duration to index encoding effort. We hypothesized that a similar pattern of results would be found under low load and high load, such that in both cases expecting a reminder reduces encoding effort and subsequent memory. This would be evidenced by a main effect of expectation and no interaction between expectation and load. However, we also thought that the expectation effect may be more pronounced under high load, in which case an interaction may be present. Finding an expectation effect under low load would indicate that expectations can uniquely affect the prospective component of PM. Alternatively, if encoding effort only operates on the retrospective memory component, then an interaction effect should be seen such that expectations influence performance only under high load. These hypotheses were preregistered (https://osf.io/syt34).

Method

Participants

The preregistered power analysis recommended a minimum of 180 participants (45 per condition), following exclusions, to detect a medium-sized effect (f = .3) with a power of .80 and Pα of .05. We initially collected online data (using E-prime Go) from 208 participants with the anticipation that some would be excluded based on preregistered criteria. The final sample, after exclusions, included 190 participants (aged 17–29) that were randomly assigned to the expecting low (n = 52), expecting high (n = 49), not-expecting low (46), and not-expecting high (n = 43). Thus, the final sample size in the not-expecting high is slightly under the recommended 45 participants per condition. This study was completed online using E-Prime Go.

Design, Materials, and Procedure

A 2 (reminder expectations: expecting vs. not-expecting; between) × 2 (target load: low [2] vs. high [8]; between) × 4 (Block: 1, 2, 3, and 4; within) design was used. The materials and procedure were nearly identical to Experiment 1. Participants studied two (low load) or eight (high load) targets in each block, but only two targets were presented during the ongoing task. In the low load condition, both studied targets in each block were presented during the ongoing task. In the high load condition, two (of eight) targets were randomly selected for each participant to be presented in the ongoing task. Targets appeared on trials 40 and 80 of the 84-trial ongoing task block. The recognition memory consisted of the eight studied targets and eight new items in the low load condition, and 32 studied targets and 32 new items in the high load condition.

Dependent Variables

The dependent variables were the same as in Experiment 1. However, because there were no “uncontaminated” items in the low load condition (i.e., all targets that were studied also appeared in the ongoing task), recognition memory is on overall performance.

Exclusionary Criteria

Participants in each condition were excluded for the following preregistered reasons:

1. Both failing to detect any PM targets and forgetting the prospective memory task (n = 8).

   • expecting low = 2, expecting high = 3, not-expecting low = 1, not-expecting high = 2

2. Getting below 55% accuracy on the PM block ongoing task (n = 0).

3. Making false alarms (i.e., PM response on nontarget trials) on over 15% of trials (n = 0).

4. Having PM block ongoing task response times greater than 3 or less than 3 SDs from the group mean (n = 2).

   • expecting low = 1, expecting high = 0, not-expecting low = 1, not-expecting high = 0
5. Failing an attention check during the postexperimental questionnaire (n = 1).

   • expecting low = 0, expecting high = 3, not-expecting low = 0, not-expecting high = 1
6. Participants with average study durations of greater than 3 or less than 3 SDs from the group mean (n = 7). This exclusion was added to the potential high variability from studying only two target words in the low load condition.

   • expecting low = 2, expecting high = 2, not-expecting low = 1, not-expecting high = 2

One additional participant was excluded from the expecting low condition due to having an average study duration in one block that was greater than 9 SDs greater than the group mean. This participant had an average study duration of approximately 5 s, 5 s, and 1 s across the first three blocks, respectively, followed by a study duration of approximately 68 s in Block 4. No other participant had an average study duration in Block 4 of over 30 s, and the highest average study duration in any block by any participant was 37 s (all others were under 30 s). While this exclusion was not preregistered, we deemed it justified based on its extremity.

Results

Preregistered analyses included a 2 (Expectation: expecting vs. not-expecting) × 2 (Load: low vs. high) × 4 (Block: 1, 2, 3, and 4) ANOVA for PM retrieval and study duration. Planned analyses were also conducted between conditions across the first three blocks and separately for the fourth block. Results for recognition memory and ongoing task performance can be found in the Supplemental Materials. Greenhouse–Geisser corrected values are provided in instances in which Mauchly’s test indicated that the assumption of sphericity was violated (in which, the F-test is subscripted using F_GG), and post hoc analyses for block effects used Bonferroni-corrected p values of .017.

To verify that our manipulation was working correctly, we examined how many participants believed the specific reminder instructions prior to Block 4 (proportion who did not believe instructions: not-expecting low: n = 1 [2%]; not-expecting high: n = 1 [2%]; expecting low: n = 7 [13%]; expecting high: n = 8 [17%]). Results were the same regardless of whether the participants who failed the manipulation check were excluded. We therefore report analyses that include said participants.

PM Retrieval

All Blocks.

For PM retrieval, there was an effect of block, block: F(3, 558) = 24.16, p < .001, ${\eta }_p^2=.115$ and load, load: F(1, 186) = 32.09, p < .001, ${\eta }_p^2=.147$, but not expectation, expectation: F(1, 186) = 1.46, p = .228, ${\eta }_p^2=.008$ (Figure 6). There was a significant interaction between block and expectation, Block × Expectation: F(3, 558) = 14.10, p < .001, ${\eta }_p^2=.070$ and between block and load, Block × Load: F(3, 558) = 6.42, p < .001, ${\eta }_p^2=.033$. There were no other significant interactions (F’s < 1.87, p’s > .163).

Figure 6. PM Retrieval Across All Four Blocks Separated by Condition in Experiment 3.

Note. The circles represent mean performance, and the error bars reflect plus or minus standard error. PM = prospective memory. See the online article for the color version of this figure.

Blocks 1–3.

Across Blocks 1–3, performance was higher under low load, load: F(1, 186) = 14.76, p < .001, ${\eta }_p^2=.074$. There were no other significant effects (F’s < 2.76, p’s > .098).

Block 4.

Block 4 analyses supported the encoding effort hypothesis by showing PM retrieval was better in not-expecting condition compared to the expecting condition, expectation: F(1, 186) = 29.80, p < .001, ${\eta }_p^2=.138$. Performance was also higher under low load, load: F(1, 186) = 48.91, p < .001, ${\eta }_p^2=.208$. Critically, there was no interaction between the two (Expectation × Load: F < 1).

Study Duration

All Blocks.

For study duration, there was an effect of block, block: F_GG(2.84, 527.75) = 13.11, p < .001, ${\eta }_p^2=.066$ and load, load: F(1, 186) = 3.40, p < .001, ${\eta }_p^2=.115$, but not expectation, expectation: F(1, 186) = 2.94, p = .088, ${\eta }_p^2=.016$ (Figure 7). There was a significant interaction between block and expectation, Block × Expectation: F_GG(2.84, 527.75) = 4.98, p = .003, ${\eta }_p^2=.026$. There were no other significant interactions (F’s < 3.41, p’s > .066).

Figure 7. Study Duration Across All Four Blocks Separated by Condition in Experiment 3.

Note. Study duration refers to the average time in milliseconds spent studying each target. The circles represent mean study duration, and the error bars reflect plus or minus standard error. See the online article for the color version of this figure.

Blocks 1–3.

Across Blocks 1–3, study duration changed across blocks, block: F_GG (1.93, 20.31) = 20.31, p < .001, ${\eta }_p^2=.098$. There was a significant reduction in study duration from Block 1 to 2, F(1, 189) = 32.11, p < .001, ${\eta }_p^2=.145$. The numerical reduction from Block 2 to 3 was not significant (F < 1). Study duration was also longer under low load, load: F(1, 186) = 25.94, p < .001, ${\eta }_p^2=.122$. There were no other significant effects (F’s <3.36, p’s > .068).

Block 4.

Block 4 analyses supported the encoding effort hypothesis by showing that study duration was longer in not-expecting condition compared to the expecting condition, expectation: F(1, 186) = 12.87, p < .001, ${\eta }_p^2=.065$. Study duration was also longer under low load, load: F(1, 186) = 22.95, p < .001, ${\eta }_p^2=.110$. Critically, there was no interaction between the two, Expectation × Load: F(1, 186) = 1.52, p = .219, ${\eta }_p^2=.008$.

Exploratory Analysis of Subjective Effort

Block 4.

A 2 × 2 ANOVA was conducted to test whether self-reported Block 4 encoding effort differed between conditions when assessed during the postexperimental questionnaire (subjective effort from 1 to 7: not-expecting low: M = 5.39, SE = 0.19; not-expecting high: M = 4.63, SE = 0.23; expecting low: M = 4.02, SE = 0.22; expecting high: M = 3.67, SE = 0.22). This analysis supported the encoding effort hypothesis by showing that subjective effort was higher in the not-expecting condition compared to the expecting condition, expectation: F(1, 186) = 29.10, p < .001, ${\eta }_p^2=.138$. Subjective effort was also higher under low load compared to high load, load: F(1, 186) = 6.61, p = .011, ${\eta }_p^2=.208$. Critically, there was no interaction between the two (Expectation × Load: F < 1).

Exploratory Mediation of Study Duration on PM Retrieval

Block 4 Objective Effort.

The Lavaan package from R was used to specify an exploratory mediation model in which objective encoding effort (study duration in milliseconds) simultaneously mediated the relation between expectation (Subscript 1) and load (Subscript 2) and PM retrieval in Block 4.

For reminder expectation effects: The path from expectation condition to encoding effort (a₁) was significant (b = −1.20, z = 3.60, p < .001), the path from encoding effort to PM retrieval (b) was significant (b = 0.02, z = 3.18, p = .001), and the path from expectation condition to PM retrieval (c₁) was significant (b = −0.14, z = 5.41, p < .001). The effect of expecting a reminder on PM retrieval after accounting for encoding effort as a mediator (c₁’) was also significant (b = −0.12, z = 4.65, p < .001). Critically, there was an indirect effect of expecting a reminder on PM retrieval, b = −0.02, SE = 0.01, 95% CI [−0.041, −0.008], using confidence intervals (i.e., not overlapping with zero) as an index of significance. These results suggest that the effect of expecting a reminder on unaided PM retrieval is partially mediated by encoding effort, such that reduced study duration in the expecting condition is associated with worse memory.

For load effects: The path from expectation condition to encoding effort (a₂) was significant (b = 1.55, z = 4.82, p < .001), the path from encoding effort to PM retrieval (b) was significant (b = 0.02, z = 3.18, p = .001), and the path from expectation condition to PM retrieval (c₂) was significant (b = 0.18, z = 6.88, p < .001). The effect of expecting a reminder on PM retrieval after accounting for encoding effort as a mediator (c₂’) was also significant (b = 0.15, z =5.60, p < .001). Critically, there was an indirect effect of expecting a reminder on PM retrieval, b = 0.03, SE = 0.01, 95% CI [0.029,0.070], using confidence intervals (i.e., not overlapping with zero) as an index of significance. These results suggest that the effect of load on unaided PM retrieval is partially mediated by encoding effort, such that reduced study duration in the high load condition is associated with worse memory.

Exploratory Mediation of Subjective Effort on PM Retrieval

Block 4 Subjective Effort.

The Lavaan package from R was used to specify an exploratory mediation model in which subject encoding effort simultaneously mediated the relation between expectation (Subscript 1) and load (Subscript 2) and PM retrieval in Block 4.

For reminder expectation effects: The path from expectation condition to encoding effort (a₂) was significant (b = −0.59, z = 5.49, p < .001), the path from encoding effort to PM retrieval (b) was significant (b = 0.06, z = 3.39, p = .001), and the path from expectation condition to PM retrieval (c₂) was significant (b = −0.14, z = 5.91, p < .001). The effect of expecting a reminder on PM retrieval after accounting for encoding effort as a mediator (c₂’) was also significant (b = −0.10, z = 4.07, p < .001). Critically, there was an indirect effect of expecting a reminder on PM retrieval, b = −0.03, SE = 0.01, 95% CI [−0.058, −0.012], using confidence intervals (i.e., not overlapping with zero) as an index of significance. These results suggest that the effect of expecting a reminder on unaided PM retrieval is partially mediated by encoding effort, such that reduced subjective effort in the expecting condition is associated with worse memory.

For load effects: The path from expectation condition to encoding effort (a₂) was significant (b = 0.27, z = 2.50, p = .012), the path from encoding effort to PM retrieval (b) was significant (b = 0.06, z = 3.39, p = .001), and the path from expectation condition to PM retrieval (c₂) was significant (b = 0.18, z = 6.98, p < .001). The effect of expecting a reminder on PM retrieval after accounting for encoding effort as a mediator (c₂’) was also significant (b = 0.16, z =6.32, p < .001). Critically, there was an indirect effect of expecting a reminder on PM retrieval, b = 0.02, SE = 0.01, 95% CI [0.002, 0.034], using confidence intervals (i.e., not overlapping with zero) as an index of significance. These results suggest that the effect of load on unaided PM retrieval is partially mediated by encoding effort, such that reduced subjective effort in the high load condition is associated with worse memory.

Exploratory Strategy Usage

Participants were asked at the end of the experiment if they would change strategies (yes/no) upon studying a new list if they knew a reminder would not be available (proportion “yes”: not-expecting low: M = 0.44; not-expecting high: M = 0.60; expecting low: M =0.69; expecting high: M = 0.88). A logistic generalized linear model analysis indicated that participants in the expecting reminder condition were more likely to indicate they would switch strategies than those in the not-expecting condition, expectation: χ² (1, N = 190) = 6.45, p = .011. There was no difference between low and high load, χ² (1, N = 190) = 3.74, p = .053, and no interaction between the two (Expectation × Load: χ² < 1).

Discussion

Experiment 3 examined whether the high demands on the retrospective memory component of PM may explain the expectation effects in Experiments 1 and 2. The results suggest that is clearly not the case, as there was no interaction of expectation and load in any analysis. Critically, even with only two targets the results entirely replicated those of Experiment 1—expecting a reminder in Block 4 reduced encoding effort and reduced PM retrieval, and encoding effort partially mediated the effect of expectations on PM retrieval. This runs contrary to Scullin et al. (2018) findings showing no association between study duration and PM retrieval with a single nonspecific target and suggests that encoding effort and PM retrieval may only be related when a target is specific.

Because our interpretation of the results largely depends on null interaction effects with load, we also conducted exploratory analyses separately for the high- and low-load conditions (see Supplemental Materials for full analyses). In the high load condition, we replicated all effects reported in the full analysis that align with the encoding effort hypothesis. Participants in the not-expecting condition in Block 4 showed higher PM retrieval rates, longer study durations, and greater subjective effort reports compared to those in the expecting condition. Objective and subjective effort both partially mediated the effect of expectations on PM retrieval. Finally, participants in the expecting condition were more likely to indicate that they would switch strategies if they knew no reminders would be available. In the low load condition, we replicated all the same effects consistent with the encoding effort hypothesis, except that study duration did not partially mediate the effect of expectations on PM retrieval (note, however, that subjective effort did). Given the null interactions, along with the largely consistent results when examining the effects separately for each load condition, this suggests that retrospective memory demands are not a primary factor contributing to our findings.

One unexpected finding was that participants studied targets for longer and reported higher subjective effort under low load compared to high load. This could reflect that two, but not eight, targets were within all participants’ regions of proximal learning (Metcalfe & Kornell, 2005). Participants may try hard encoding both items under low load, while under high load participants realized they could only remember so many items and therefore selectively prioritized effortfully encoding a subset of items or distributed attention across all eight items but studied each individual item with less effort. However, it is important to note that load was manipulated between subjects, so participants were not actually making effort assessments relative to lists with more or fewer words. The critical finding is that reminder expectations within a single load condition (e.g., low load), which are more directly comparable, produced differences in encoding effort.

General Discussion

The present study examined how offloading influences the encoding and retrieval of PM intentions and the consequences of unexpectedly removing access to reminders. We replicated previous research showing that having reminders available at retrieval (Blocks 1–3) improved PM retrieval compared to having to remember the intentions internally (e.g., Gilbert, 2015a, 2015b; Gilbert et al., 2020; Kelly & Risko, 2019, 2022; Peper et al., 2023; Vortac et al., 1995). We also extend previous research from the retrospective memory domain by showing that PM retrieval dropped considerably when reminders were unexpectedly taken away, regardless of memory load. We suggest that these findings are due, at least in part, to reduced encoding effort when reminders are expected to be available later. When participants know reminders will not be available, they change their encoding strategy to more effectively commit PM targets to memory. Below we discuss the theoretical implications of the present findings, limitations of the project, and future directions.

Previous research suggests that one reason participants offload is to minimize the amount of effort to complete the task, as maintaining memory representations internally can be demanding (Sachdeva & Gilbert, 2020). The present study indicates that offloading can also reduce the encoding effort needed to achieve an acceptable level of memory performance. Because the quality of encoding is irrelevant when the exact details are provided as reminders during retrieval, this effectively allows participants to encode intention-related details with minimal effort. Based on previous research demonstrating effortful encoding is not required for event-based PM tasks (Scullin et al., 2018), perhaps in part due to the privileged status of future-oriented intentions (Goschke & Kuhl, 1993), we reasoned that unexpectedly removing these reminders system may not negatively impact PM retrieval. However, this clearly was not the case, suggesting that some form of effortful encoding is required to effectively retrieve intentions later.

The encoding effort hypothesis (Kelly & Risko, 2022), applied to PM tasks, assumes that reduced encoding effort reduces unaided PM retrieval. In support of this hypothesis, participants in the expecting condition, who had the lowest PM retrieval, also had lower study duration (Experiments 1 and 3) and self-reported effort (Experiments 1–3). Furthermore, it was found that shorter study duration (Experiments 1 and 3) and subjective effort (Experiments 1–3) were associated with worse PM retrieval. In contrast, Scullin et al. (2018) found no association between study duration and PM retrieval. One primary methodological difference is that Scullin et al. (2018) used nonspecific category targets (e.g., respond to any animal) whereas the present study used specific targets (e.g., respond to rabbit), which theoretically can be retrieved using different processes. Nonspecific targets place greater demands on preparatory attention (Brewer et al., 2010; Smith, 2003), whereas specific targets can be retrieved spontaneously (Einstein & McDaniel, 2005; Lyon & Hicks, 2023). One means of spontaneous retrieval is when previous exposure (i.e., encoding) heightens the familiarity of the target relative to other nontarget stimuli. The discrepancy between the target and nontarget stimuli can trigger a search in memory for the source of the discrepancy, leading to the automatic retrieval of the intention (discrepancy-plus-search; McDaniel & Einstein, 2007). In this case, studying a specific target longer could increase the familiarity of it and enhance retrieval, but studying a nonspecific target longer would not enhance the familiarity of an exemplar that appears in the PM task. Thus, future work directly comparing the role of encoding effort on specific and nonspecific PM retrieval may better help elucidate the cognitive processes underlying PM retrieval.

While the present study provides some evidence in favor of the encoding effort hypothesis more generally, it does not fully explain the effect of reminder expectations on unaided PM. Study duration (Experiments 1 and 3) and subjective effort (Experiments 1–3) only partially mediated the role of expectancy on unaided PM retrieval, suggesting that there may be additional mechanisms underlying performance. One possibility is that reminder expectations influence retrieval processes independently of encoding. Kelly and Risko (2022) note that having a reminder unexpectedly taken away may result in a surprise or alarm, which in the context of PM, can usurp working memory resources that are needed to maintain the intention in awareness while simultaneously performing a demanding ongoing task. A PM task is well-suited to test this possibility. Anything that redirects working memory resources should be reflected by changes in ongoing task performance, such as slower response times or lower accuracy (Smith, 2003). However, in all experiments, we found that expecting a reminder had no effect on ongoing task performance when the reminder was not present (see Supplemental Materials). Additionally, the retrospective reports of how participants noticed the PM targets indicated there were no differences between reminder conditions in the degree to which participants monitored for or spontaneously retrieved the intention when there was no reminder (see Supplemental Materials). Together the ongoing task performance and subjective retrieval strategy reports suggest that having a reminder unexpectedly taken away does not occupy working memory resources during retrieval.

Alternatively, it may be that while increasing effort at encoding tends to improve retrieval, some encoding strategies are more effective than others, and it can be difficult to tease apart the roles of strategy and effort. In fact, other work has argued that depth of processing at encoding can be independent of encoding effort (Vincent et al., 1993). Vincent et al. gave some participants a high-effort shallow processing task at encoding and other participants a low-effort deep processing task at encoding. The latter group had better memory performance. Kelly and Risko (2022) measured study durations and asked participants at the end of the experiment whether they used encoding strategies. They found that both study duration and, separately, use of strategy, partially mediated the effect of reminder expectations on unaided retrieval in the final block. Although we did not explicitly ask participants the types of strategies they used, we did ask whether they would study information differently during a new list if they knew reminders would not be available. Indeed, those in the expecting condition of each experiment were more likely to indicate that they would switch strategies, suggesting that they had some awareness that whatever strategy they were using was not effective. Considering the past findings in the context of the present study, it is therefore possible that people who expect a reminder can (a) select a less effective encoding strategy while maintaining the same amount of effort, (b) maintain the same strategy while reducing their level of effort, or (c) select a less effective encoding strategy and reduce their level of effort. Theories of metacognition can help to better understand the decisions underlying encoding strategies and effort when reminders are involved.

Two primary processes underlying metacognition play a key role in theories of cognitive offloading (Risko & Gilbert, 2016) and can be applied to understand why reminder expectations can influence encoding effort and strategy. Monitoring refers to one’s awareness of their own cognitive performance, whereas control refers to the decision to update or maintain a cognitive strategy based on the assessment made by the monitoring process (Nelson & Narens, 1990). The knowledge updating framework describes four core processes involved in encoding strategy selection (Dunlosky & Hertzog, 2000). Strategies vary in effectiveness, and people can monitor differences in effectiveness and update their knowledge about how effective that strategy is before they utilize the newly acquired knowledge. One way people do this is by making confidence judgments about their performance during retrieval that are used to update their knowledge and influence which encoding strategy they select next. Previous research on cognitive offloading has shown having access to an external store (e.g., the Internet) increases confidence in retrieval (Dunn et al., 2021; Pieschl, 2021). Experiencing the ease or effectiveness of retrieval with reminders (i.e., monitoring) could cause participants to update their knowledge about their belief in the effectiveness of reminders, thereby changing their subsequent encoding strategy (i.e., utilization) to engage in less effortful processing and earlier termination of study when expecting to have another reminder at retrieval (Kelly & Risko, 2022). Murphy (2023) examined the effect of reminder expectations on free recall performance with word lists that varied in value using a design similar to the present study. An important difference was that participants could only set reminders for some but not all of the words. Results showed words with higher values were offloaded more often and forgotten more when the reminder was unexpectedly unavailable, stressing the importance of control decisions during encoding with reminders.

Another aspect of metacognitive control during encoding strategy selection is referred to as the “selection of kind of processing” (Bjork et al., 2013; Nelson & Narens, 1990; Nelson et al., 1994). The design of the present study did not allow us to determine whether reminder expectations influenced what type of processing participants selected during encoding, but it is possible that the not-expecting condition engaged in deeper processing than the expecting condition when given the freedom to do so in Block 4. A future study that employed the same design as Experiment 1 could discriminate between effort and strategy by allowing self-paced study and asking participants about their strategies immediately after encoding. We would predict that study durations and strategy effectiveness would fully mediate the effect of reminder expectations on unaided PM retrieval.

Conclusions

The present study replicated previous work showing reminders improve PM. Without reminders, more effortful processing at encoding tends to produce better memory retrieval for specific targets. Our results suggest people encode intentions less effortfully—and possibly choose less effective strategies—when they expect to have reminders at retrieval. This points to the importance of metacognition in the study of cognitive offloading. The present study also contributes to the theoretical understanding of how reminders affect PM and provides avenues for future research. For populations with PM deficits, setting reminders with technology has recently been recommended as a strategy for improving PM (e.g., Scullin et al., 2022). The results of the present study show that reminder recommendations should contain nuance to avoid unintended consequences of depending on reminders. There are also similarities here to work in human-automation interactions, as reminders serve to “automate” human memory. An automation failure has unfortunate consequences on cognitive processes in the context of PM and likely beyond. As technology becomes more popular as a tool for offloading and automation, it is clear from the current project that the technology must be reliable or the user could make critical errors if the technology fails.