Positive autobiographical memory recall does not influence temporal discounting: an internal meta-analysis of experimental studies

Abstract

People tend to discount the value of future rewards as the delay to receiving them increases. This phenomenon, known as temporal discounting, may underlie many impulsive behaviors, such as drug abuse and overeating. Given the potential role of temporal discounting in maladaptive behaviors, many efforts have been made to find experimental manipulations that reduce temporal discounting. One class of manipulations that has held some promise involves recalling positive autobiographical memories prior to making intertemporal choices. Just as imagining positive future events has been shown to reduce temporal discounting, a few studies have shown that recalling positive past events reduces temporal discounting, especially if memory retrieval evokes positive affective states, such as gratitude and nostalgia. However, we failed to replicate these findings. Here we present an internal meta-analysis combining data from 14 studies (n = 758) that involved within-subjects positive memory recall-based manipulations. In each study, temporal discounting was assessed using a monetary intertemporal choice task. The average effect size was not significantly different from zero. This finding helps elucidate the neurocognitive mechanisms of temporal discounting; whereas engaging the episodic memory system to imagine future events might promote more patience, engaging the episodic memory system to imagine past events does not.

1. Introduction

People often make tradeoffs between smaller short-term benefits and larger long-term gains. For example, they may have to resist spending money on luxury goods now in order to save more for retirement. In these intertemporal choices, people tend to display temporal discounting. That is, the subjective value of potential rewards decreases as the delay to receiving them increases. People vary in the extent to which they discount future rewards, and higher discount rates measured in the lab are associated with real-world detrimental behaviors, including substance abuse and other addictive behaviors, physical inactivity, and overeating (Amlung et al., 2017; Appelhans et al., 2019; MacKillop et al., 2011; Story et al., 2014). There are also reports of elevated temporal discounting across several psychiatric disorders (e.g., addiction, attention deficit hyperactivity disorder, schizophrenia), suggesting that it may be a critical trans-disease process in psychopathology (Amlung et al., 2019; Lempert et al., 2019). Given the links between temporal discounting and maladaptive behaviors, there have been many efforts to manipulate temporal discounting in the lab, with a view to developing interventions that might promote more future-oriented decision-making. One category of manipulations that has been reported to reduce temporal discounting involves asking participants to recall positive memories prior to making intertemporal choices (DeSteno et al., 2014; Lempert et al., 2017). It is unclear, however, how robust and generalizable those effects are. Here we present data from a set of mostly previously unpublished within-subjects studies in our lab that attempted to reduce temporal discounting by having participants recall autobiographical memories that evoke positive affect.

Many scientists have attempted to manipulate temporal discounting in the lab, in order to better understand the situational factors that impact intertemporal choices (Lempert & Phelps, 2016). Some of these manipulations have focused on explicitly re-framing choices; for example, if the delayed reward is described as the “default” option, then participants tend to make more future-oriented choices (Loewenstein, 1988). Other manipulations have focused instead on changing an individual’s mindset or affective state prior to intertemporal choice (these are sometimes referred to as “priming” manipulations; Rung & Madden, 2018). These priming manipulations have had mixed success, but the one that most reliably reduces temporal discounting involves having people engage in episodic future thinking (Rösch et al., 2022; Rung & Madden, 2018). Episodic future thinking refers to the mental simulation of events that may take place in one’s personal future. It has been proposed that the capacity for episodic future thinking serves to make future outcomes more concrete in order to guide intertemporal decision-making (Boyer, 2008; Bulley et al., 2016). Indeed, in many previous experiments, asking participants to imagine specific prospective episodes before they make an intertemporal choice results in their being more likely to select larger, later rewards, compared to choices in a control condition. In a 2022 meta-analysis, Rösch and colleagues examined 174 studies drawn from 48 articles, and they concluded that there is a medium-sized effect of episodic future thinking on temporal discounting. Notably, in almost all episodic future thinking studies, the imagined episodes have a positive valence. Findings are mixed about whether imagining negative or neutral events also increases the subjective value of future rewards (Bulley et al., 2019; Calluso et al., 2019; Liu et al., 2013; Zhang et al., 2018).

Positive episodic future thinking manipulations are promising and potentially scalable, but the mechanism by which they alter choice remains unclear. On the one hand, the future-oriented content of episodic future thinking might make delayed rewards seem more concrete and/or proximal, and therefore, more valuable. On the other hand, it could be that the process of engaging in mental simulation prior to choice is sufficient to reduce discounting. Mental simulation, whether it be about the future or the past or about counterfactual realities, involves changing one’s perspective. It is possible that priming perspective-taking could facilitate decisions that are valuable from the perspective of the “future self.” One way to test this idea is to see if imagining positive events from the past also reduces discounting. Indeed, one published study (co-authored by one of us, Lempert et al., 2017; see Exps. 1, 2, and 3 below) showed that recalling positive memories prior to intertemporal choice led to more future-oriented choice. This effect was replicated by another group (Ciaramelli et al., 2019). Other studies that have tested the effects of specific emotions on temporal discounting have used similar manipulations and yielded similar findings. When participants recalled events that made them feel grateful, they showed reduced temporal discounting (DeSteno et al., 2014), and when they recalled episodes that evoked nostalgia, they exhibited more financial patience (Huang et al., 2016). Although this literature on positive memory retrieval is small, these findings challenge the idea that episodic future thinking reduces temporal discounting by increasing future orientation. If reminiscing about the past is equally as effective as imagining the future is, then an individual does not need to be explicitly oriented toward future outcomes in order to make decisions that maximize their long-term gain.

We ran a series of experiments to test the robustness of the effect of positive memory retrieval on temporal discounting. In addition to an exact replication of the first study in Lempert et al., 2017 (Exp. 4), these experiments included: using personal photos as cues for episodic memory retrieval (Exp. 5), having participants take a field or observer perspective in their memory (Exp. 6), and conducting the study online (Exp. 7). We also attempted to replicate the gratitude (Exps. 8 - 12 below) and nostalgia (Exp. 13 and 14) effects in our lab.

Foreshadowing the results, we were unable to replicate the effects of positive memory recall, gratitude, or nostalgia on temporal discounting in our follow-up experiments. Instead of publishing each experiment separately, we combined these experiments with the experiments from the original study on positive memory recall and temporal discounting (Lempert et al., 2017) and conducted an internal meta-analysis, which we present here. We believe that publishing these null results is important for several reasons. First, the finding that imagining positive past events is not as effective as imagining positive future events suggests that future orientation, and not just mental simulation or perspective-shifting, is critical for increasing patience. Second, positive memory retrieval is often used as a control condition in experiments that test episodic future thinking manipulations (Daniel et al., 2016; Dassen et al., 2016; Stein et al., 2016), a practice that would be inadvisable if positive memory retrieval were an effective manipulation itself. This paper thus confirms that episodic thinking about the past is a good benchmark comparison condition for episodic future thinking.

2. Materials and Methods

2.1. Study inclusion criteria

The studies included here fulfilled the following criteria:

1. The study either had no age restrictions, or it was a study of young adults (ages 18-35). Thus, we excluded a published study that did not find effects of positive memory recall on temporal discounting in older adults (Lempert et al., 2020a). Given the mixed findings on the effectiveness of episodic future thinking manipulations in older adults (Mok et al., 2020; Sasse et al., 2017), it was not clear a priori whether a memory-based intervention would be effective in this group. In the studies that we included, we did not exclude any participants on the basis of age.

2. Participants did a temporal discounting task that involved a series of choices between smaller, immediate and larger, delayed monetary rewards. In all included studies, participants were either paid according to one of their choices (N = 8 studies) or there was a chance that they would be paid according to one of their choices (i.e., for studies done online through Amazon’s mechanical Turk, participants’ data were submitted to a lottery in which one participant per day of data collection was selected to get one of their choices as a bonus: N = 6 studies). Thus, participants had a monetary incentive to always choose what they preferred on each trial.

3. The manipulation involved participants recalling episodic memories from their personal past that had a positive valence. Note that manipulations still varied according to what kind of positive emotion was induced (e.g., generic positive emotion, gratitude, or nostalgia) and how much the participant mentally elaborated on those memories.

4. The manipulation was within-subjects.

5. The hypothesis in the study was that the positive memory manipulation condition would reduce temporal discounting rates compared to the control condition. Thus, we excluded one study in which participants were asked to recall fifteen events that made them feel grateful, since the purpose of that manipulation was to increase temporal discounting by making retrieval of positive memories difficult (Schwarz et al., 1991). Note, however, that that study found no effect of recalling fifteen “gratitude events” on temporal discounting (t₆₆ = −1.36; p = 0.892; Cohen’s d_z = 0.017).

These criteria resulted in the inclusion of 14 studies performed by the authors over approximately five years, involving 758 unique participants. One of the 14 studies (Exp. 6) tested two positive memory-based manipulations within the same subjects in the same session, however. Specifically, Experiment 6 had three conditions: Control, Positive memory (observer perspective), and Positive memory (field perspective). We expected that both of these positive memory manipulations would reduce temporal discounting, and we did not want to count the same participants twice (or exclude either of these conditions). Therefore, we averaged discount rates between the field and observer perspective conditions, and compared that averaged discount rate to the control condition in order to obtain the effect size for that study.

All participants in all studies provided informed consent, and all studies were approved by the Institutional Review Board of New York University (Exps. 1 - 3) or the University of Pennsylvania (Exps. 4 - 14).

2.2. Study descriptions

2.2.1. Experiment 1. Positive memory recall in-lab (Exp. 1 from Lempert et al., 2017, SCAN)

Participants.

Forty-seven participants completed this experiment (33 F, 14 M; mean age = 21.19; SD = 4.20).

Procedure.

This study involved two in-lab sessions. On the first day, participants wrote about memories prompted by each of thirty life event cues (e.g., family vacation). The cues were compiled from prior studies (Sharot, Riccardi, Raio, & Phelps, 2007; Speer et al., 2014) and were designed to elicit neutral or positive memories. For each cue, participants selected a memory in which they had been personally involved and that had occurred at a specific place and time. For each memory, participants reported a brief description, location, and date. They also gave subjective ratings for valence (1 = neutral; 2 = positive), emotional intensity (1-4: 1 = not intense, 4 = very intense), and feeling (i.e., how they felt when recalling the memory; 1-4: 1 = neutral, 4 = very good). Participants were instructed to select memories that were positive (e.g., visiting Disneyland) or neutral (e.g., packing for a trip), but not negative (e.g., lost luggage).

In preparation for the second session, ten of each participant’s positive memories were selected. These ten had been rated as positive (i.e., valence = 2) and had the highest combined intensity and feeling ratings. They were summarized in subject-specific event cues that the participants reviewed at the beginning of the second session to ensure that they could identify the memory associated with each cue.

Participants returned for the second session three days later to perform an intertemporal choice task (Fig. 1). On each trial, they were presented with a screen showing two options: “$10 today” and a monetary reward of larger magnitude available after a delay (e.g., “$20 in 30 days”; amounts varied from $11 to $40; delays from 4 days to 180 days. All delayed reward amounts were paired with all delays). They pressed a button to indicate which option they preferred. The order of the trials was randomized, and the immediate and delayed reward options switched sides of the screen randomly. After participants responded, they were shown the option they had just chosen for 1 second. After a 2 second inter-trial interval, the next choice screen appeared. There were 60 distinct trial types, repeated twice for a total of 120 trials.

Fig. 1.

Basic task layout for Experiments 1 through 6. Participants were asked to describe positive memories a few days prior to doing this intertemporal choice task. In the Memory condition of the intertemporal choice task, they were asked to imagine those memories (prompted by subject-specific cues) for 14 s. Following this imagination phase, they either made ratings about the memory (Experiments 1 through 5), or spoke about the memory from one of two perspectives (Observer or Field perspective; Experiment 6). Finally, they made a series of intertemporal choices (ranging from 5-7 choices, depending on the study) before moving on to the next memory “mini-block.” In the Control mini-blocks, they were prompted to relax for 14 s. They then either made ratings (Exps. 1-5) or spoke about how they currently felt (Exp. 6). The structure of Experiment 5 was similar, but the subject-specific cues were replaced with photos depicting positive memories that the participant had shared beforehand. In the Control condition of Exp. 5, participants viewed a neutral nature scene and were asked to evaluate it for 14 s.

Participants made these choices in blocks (“Memory” and “Control” blocks). In Memory blocks, participants re-accessed the ten positive memories triggered by cues from their questionnaire on Day 1 before making choices. At the beginning of each memory trial, a fixation point appeared for 3 seconds. Then, a memory cue was displayed for 14 seconds. Participants were asked to recall the memory associated with this cue and to elaborate on it for as long as they could or until 14 s were up. After a 3 s inter-stimulus interval, participants rated the memory on valence, emotional intensity, and feeling (allotted 4 sec for each). Following this, participants made 6 intertemporal choices before the next memory cue appeared. One memory block consisted of 5 memories and 30 intertemporal choices.

In Control blocks, participants first saw the word “Relax” on the screen for 14 s. They were instructed to rest during this time. Then, they rated how tired they were (1-4; 1 = very awake; 4 = very tired), how bored they were (1-4; 1 = not bored; 4 = very bored), and how good they felt (1-4; 1 = neither good nor bad; 4 = very good; 4 s for each rating). Following this, they made 6 intertemporal choices before the next “relax” screen appeared. Each Control block consisted of 5 “relax” screens and 30 intertemporal choices. There were two control blocks and two memory blocks, and the order was counterbalanced across subjects. The same choices were presented in both conditions.

Participants were told at the outset that one of the trials would be randomly selected and they would receive the amount they chose on that trial, at the delay specified. If they chose the immediate reward on that trial, they would receive the money in cash that day. If they chose the delayed reward, they would receive the money in their personal checking account via Paypal (www.paypal.com) after the delay had elapsed. This task was programmed using E-Prime 2.0 Stimulus Presentation Software (Psychology Software Tools).

2.2.2. Experiment 2. Positive memory recall in-lab exact replication 1 (Exp. 1Rep in Lempert et al., 2017, SCAN)

Participants.

Forty-eight participants completed this replication of Experiment 1 (30 F, 18 M; mean age = 23.21; SD = 3.61). The procedure for the replication experiment was identical to the procedure for Experiment 1.

2.2.3. Experiment 3. Positive memory recall replication with functional magnetic resonance imaging (Exp. 4 in Lempert et al., 2017, SCAN)

Participants.

Forty participants completed this study (28 F, 12 M; mean age = 23.03; SD = 3.46).

Procedure.

The procedure in this functional magnetic resonance imaging (fMRI) experiment was very similar to the procedure in Experiment 1, with a few changes. The Day 1 questionnaire was the same, except that a vividness rating question was added for each memory. In preparation for the second session (Day 2), fourteen of each participant’s most positive memories were selected. Before going into the MRI scanner, participants practiced making intertemporal choices, and they also practiced memory recollection and rating with two of the selected memories, which were not later used in the task (leaving twelve memories for the task). Participants also practiced the Control condition tasks.

The intertemporal choice task was done in the MRI scanner. On each trial, participants chose between $20 today and a larger monetary reward available at a later date (e.g., “$36 in 30 days”; amounts varied from $22 to $60; delays were 4, 7, 30, 60, 100 or 180 days. Each delayed reward amount was paired with each delay). The delayed option appeared on the screen for 5.5 sec, and participants pressed a button to indicate whether they preferred the reward on the screen or the $20 default. After this, feedback appeared for 0.5 sec (blue dot = $20 today; green dot = delayed reward). After a jittered ITI (0.5 – 8 sec; mean = 2.5 sec; drawn from long-tailed distribution; Hagberg, Zito, Patria, & Sanes, 2001), they proceeded to the next choice trial. There were 7 of these choices in a row in each block. There were 84 distinct delayed rewards, shown once in each condition, for a total of 168 choice trials. The order of the trials was randomized.

Participants made these choices in blocks (“Memory” blocks and “Control” blocks). In each of the 12 Memory blocks, participants re-accessed 1 of the 12 positive memories for 14 s, triggered by cues from their Day 1 questionnaire. After a jittered ITI (1-8 sec; mean = 4 sec), participants made valence, positive feeling, emotional intensity, and vividness ratings (4 s each). Finally, they did a set of 7 intertemporal choices as described above. In each of the 12 Control blocks, participants first saw the word “Relax” for 14 s. They rested during this time. After a jittered ITI, they rated their boredom, positive feeling, alertness, and excitement before making 7 intertemporal choices. As in the behavioral experiments, one of the participant’s choices was realized at the end of the study. Trials were divided into 6 functional runs. Each run contained 4 blocks in one of these two orders: Memory/Memory/Control/Control or Control/Control/Memory/Memory. Orders were alternated from run-to-run for each subject, and the first order was counterbalanced across subjects. The button presses assigned to the immediate and delayed reward options were also counterbalanced across subjects.

Just as in Experiments 1 and 2, participants were paid according to what they chose on a randomly selected trial in the intertemporal choice task, either in cash (immediate reward) or via Paypal (delayed reward).

2.2.4. Experiment 4. Positive memory recall in-lab exact replication 2 (unpublished)

Participants.

Thirty-eight participants completed this second replication of Experiment 1 (26 F, 12 M; mean age = 23.00; SD = 3.71). The procedure for this experiment was identical to Experiment 1, with the exception that any delayed rewards were paid out on a Greenphire Clincard debit card rather than via Paypal. Immediate rewards were still paid out in cash. While the first three studies were conducted at New York University, this study and all subsequent studies were done at the University of Pennsylvania.

2.2.5. Experiment 5. Positive memory recall in-lab: photo cues (unpublished)

Rationale.

This experiment examined if using more salient cues of positive events (actual photographs provided by the participants that were taken at the events) would increase the efficacy of the positive memory manipulation, compared to cues that were written and generated by the experimenter.

Participants.

Thirty-six participants completed this experiment (30 F, 6 M; mean age = 21.83; SD = 3.77).

Procedure.

This experiment also consisted of two sessions, but the first session was completed online, while the second was completed in-person. In the first session, participants were asked to upload twenty pictures that evoked positive memories from their personal past. For each photo, participants ranked the vividness, intensity, and feeling of their memory of the event in the photo, and how similar they felt to their “past self” from the memory, on a scale from 1-4. Next, they rated the valence of their memory as either positive or neutral. Participants also provided background information on their pictures, including a description of the event pictured, when and where the picture was taken, who took the picture, and who was in the picture. From these twenty pictures, the experimenter selected ten with the highest “feeling” ratings (the majority were ranked a 3 or 4) for the second session. These pictures were then all standardized to a 400x400 pixel size.

The second session took place between 1 day to 2 weeks after the first. In the second session, participants completed an intertemporal choice task very similar to the ones described above. Instead of being cued to recall memories with phrases describing those memories, participants were cued with the photos they had previously uploaded. After recalling a memory for 14 s, they rated the memory’s vividness, feeling, intensity, and valence before making a series of six intertemporal choices. In the control condition, participants were presented with images of outdoor scenes selected because they were rated neutral in pleasantness in a previous study (Pegors et al., 2015). On control trials, participants assessed the quality of the image. After looking at each image for 14 seconds, they rated its visual complexity (1= not complex, 4 = very complex), how good they thought the photo was (1 = just okay, 4 = very good), how good they felt looking at the photo (1= neutral, 4 = very good), and if they liked the photo (1= no, 2 = yes).

The decision task involved choosing between two options. One option was “$10 today” and another was a larger monetary value to be given at later date. The length of delay varied from 1-180 days, and the delayed amount varied from $11-$35. We chose amounts and delays to capture a range of hyperbolic discount rates (range: 0.00018 – 0.25) with the constraints that the immediate amount always be $10 and the delay not exceed 180 days. Participants pressed a button to indicate their choice. The immediate and delayed options switched their position on the screen randomly across trials. After the participants chose, they were shown their selection for 1 second. There were four blocks of trials, each consisting of 5 photos and 30 choices. Participants completed one of two task orders: memory-control-memory-control or control-memory-control-memory. Thus, there were 60 choice trials in each condition and 120 choices total. Participants saw the exact same set of choices in each condition, so any difference in discounting between conditions could be attributed to the manipulation.

Prior to beginning the experiment, participants were instructed that one trial would be randomly selected, and the amount that they chose for that trial would be given to them on a Clincard debit card at the delay indicated.

2.2.6. Experiment 6: Positive memory recall in-lab: field perspective and observer perspective (unpublished)

Rationale.

This experiment explored whether the mechanism by which positive memory recall reduced discounting was by inducing participants to take the perspective of their past selves. This idea was inspired by research showing that the temporo-parietal junction, a region involved in perspective-taking, is critical for patient choice (Soutschek et al., 2016). Here participants were instructed to recall their memories in two different ways: a “field” perspective, which required them to take the perspective of their past selves, and an “observer” perspective, which required them to observe their past selves from a distance. We expected that the field perspective would be more effective than the observer perspective at reducing temporal discounting, but that both types of recall would reduce discounting compared to a control condition.

Participants.

Forty-two participants completed both study sessions (30 F, 12 M; mean age = 22.18; SD = 3.82).

Procedure.

Again, this experiment involved two sessions. The first session was an online survey, which participants completed at least one day before the second session. In the survey, participants wrote about memories prompted by twenty different life event cues taken from Lempert et al. (2017). The cues probed for positive memories. Just as in Experiments 1-4 above, participants rated these personal memories after describing them. In addition to valence, vividness, and intensity ratings (same as in Experiments 1-4), participants also rated “feeling now” (1-4; 1= neutral; 4 = very good), “feeling then” (1-4; 1= neutral; 4 = very good), “feeling of re-experiencing” (1-4; 1= not at all; 4 = very much), similarity to past self (1-4; 1= very different; 4 = very similar), personal importance (1-4; 1= not important at all; 4 = extremely important), connectedness to others (1-4; 1= very distant from others; 4 = very close to others), and subjective temporal distance (1-4; 1= very long ago; 4 = very recent).

After describing and rating memories, participants completed a 17-question “screener” intertemporal choice task. This 17-item abbreviated monetary choice questionnaire probes a range of hyperbolic discount rates from 0.0001 to 0.2525, uniformly sampled in log-space, with the options in each question being similarly valued at a given discount rate. To incentivize these “screener” questions, 1/10 of all participants received one of the amounts they chose on the delay chosen in the form of an Amazon gift card. The screener intertemporal choice task served two purposes. First, subjects who chose either all immediate or all delayed rewards in the first session were not invited for the second session. Second, the discount rates computed based on the first session choices were used to generate a subject-specific set of choices for the intertemporal choice task in the second session. This subject-specific choice set oversampled choices on which participants would be close to indifferent between the immediate and delayed options. We generated three distinct sets of 35 choices, one for each condition described below, for a total of 105 choices.

Before the second session, the experimenter selected seven memories for each participant and summarized them with subject-specific cues. These memories were positive and had the highest intensity, feeling, and personal importance ratings. These cues were shown to each participant at the beginning of the second session to ensure that they could identify the associated memory. Participants did an intertemporal choice task (details: two options presented simultaneously; randomized order and screen placement; self-paced with 1s feedback; 2 s inter-trial interval) with three conditions: “Field Perspective,” “Observer Perspective” and Control. In both the Field Perspective and Observer Perspective blocks, participants were prompted with the seven positive memory cues. In the Field Perspective block, participants recalled each memory from the “field” or “first-person” perspective for 14 s. Then, for 15 s, participants reported out loud (responses were recorded with a tablet) what they felt and thought during the recalled experience (they were asked: “What were you feeling and thinking in this memory?”). After a 2-s interval, participants made five intertemporal choices before the next memory cue. In the Observer Perspective block, the same memory cues were shown in the same order and participants recalled each memory from an “observer” or “third-person” perspective (they were prompted with the question: “What did you look like in this memory?”). Then, for 15 s, they reported what they looked like during the recalled experience. After a 2-s interval, participants made five intertemporal choices before the next memory cue. “Field” and “observer” perspectives were explained to participants at the beginning of the study, and the question prompts helped to further cement these instructions, since the two different questions require two different perspectives to answer them.

In Control blocks, participants saw the word “Relax” for 14 s and were instructed to not think about anything in particular during that time. Next, for 15 s, participants reported how they were feeling right now (prompt: “How are you feeling right now?”). Then, they made five intertemporal choices before the next “Relax” screen. Each Control block had 7 “Relax” screens and 35 intertemporal choices. The order of the three blocks was counterbalanced across subjects.

At the start of the second session, participants were told that one trial would be randomly selected and they would receive the amount they chose on that trial on a debit card at the delay specified.

2.2.7. Experiment 7: Positive memory recall replication online (unpublished)

Participants.

This study was conducted online via Amazon’s mechanical Turk using Qualtrics (Provo, UT). Sixty-five participants (36 F, 28 M, 1 not reported; mean age = 37.94; SD = 12.99) completed the study.

Procedure.

Participant completed one session divided into 4 parts: a 17-item intertemporal choice task (same as the “screener” task described in Exp. 6 above); then an imagination task; then another, very similar 17-item intertemporal choice task; and finally, another imagination task. Participants were assigned to one of four conditions for the initial imagination task: Positive Past, Positive Future, Negative Past, and Negative Future. Here we only include participants in the Positive Past condition. In this condition, participants were instructed to “Please imagine a positive past event that happened to you.” They were given some examples of positive events (e.g., “a night you enjoyed out with friends,” “when you got a new job”). Then participants spent at least two minutes writing about the event (at least 75 words were required). As a guide, they were told to write in detail about when and where the event took place, the persons and objects present, the actions that occurred, and how they felt. Then they rated the vividness, valence, arousal, subjective distance, objective distance, extent of observing and extent of being involved in the event. Following this imagination task, participants completed another set of intertemporal choices (discount rates were compared post-imagination to pre-imagination). In the last part of the study, participants were selected a neutral event from the past 24 hours and wrote about it for at least two minutes. They then rated the neutral event in the same manner as the past positive event. This last block was included to try to minimize experimenter demand effects.

Participants were told at the outset that one out of every hundred participants would be randomly selected to receive what they chose on one randomly selected trial in the task. Participants were paid via bonus through Amazon’s mechanical Turk.

2.2.8. Experiment 8: Gratitude recall in-lab (unpublished)

Rationale.

This experiment attempted to replicate DeSteno and colleagues (2014), who showed that inducing gratitude with a writing task reduced temporal discounting.

Participants.

Thirty-four individuals from the University of Pennsylvania community who had not previously participated in similar studies participated (demographic information is not available, but recruitment was similar to other in-lab studies).

Procedure.

Participants completed one session, in which they did the following sequence twice: (1) writing task, (2) self-report questionnaire, and (3) intertemporal choice task. The writing tasks were administered in Qualtrics, and the intertemporal choice tasks were administered in E-Prime (Psychology Software Tools, Sharpsburg, PA). The two iterations were identical except for the writing task, which was a Gratitude writing task in one case and a Typical Day writing task in the other (order counterbalanced across subjects). In the Gratitude task, participants wrote for five minutes about a situation that made them feel grateful. In the Typical Day (control) task, participants wrote for five minutes about a typical day. Subjects wrote on paper, which was not collected to encourage participants to write honestly. After each writing task, participants were given one minute to think about the event that they wrote about. Next, they rated how they felt at that moment, specifically the extent to which they felt frustrated, happy, content, thankful, appreciative, grateful, pleasant, confident, sad, bored, annoyed, accomplished, and angry, all on a scale from 1 (not at all) to 5 (very much). The purpose of this questionnaire was to ensure that our manipulation was effective at inducing gratitude. When averaging the responses for the “thankful,” “grateful,” and “appreciative,” questions, we confirmed that participants felt significantly more gratitude after writing about a situation that made them feel grateful compared to after writing about a typical day (t₃₃ = 6.49; p < 0.001).

After this questionnaire, participants completed a 17-item intertemporal choice questionnaire (the same one used in Exps. 6 and 7). Participants saw similar, but not identical, choice sets in both conditions (amounts and delays were jittered so that participants never saw the same question twice). On each trial, participants pressed a button indicating which option they preferred, and the selected option was shown before the next trial appeared.

Participants were paid on a debit card according to what they chose on one trial in one of the intertemporal choice tasks (a coin flip determined whether the post-gratitude or post-typical day task was selected).

2.2.9. Experiment 9: Gratitude recall in-lab replication with new choice set (unpublished)

Rationale.

This experiment examined if using “round” delayed reward amounts and delays would increase the efficacy of the gratitude manipulation. We define “round” amounts as multiples of $5 (e.g., $10 or $25). We define “round” delays as ones that can be converted into a discrete number of weeks or months (e.g., 7 days or 30 days). Previous research has shown that people are more likely to choose delayed rewards when the amounts in the task include decimal points, possibly because this reduces positive affect associated with the immediate rewards or changes the mindset with which people approach the task (Fassbender et al., 2014). Therefore, we reasoned that “round” amounts and delays would be more intuitive and make the decision-making process easier, and possibly more amenable to change. In addition, this experiment also held the immediate reward constant. We reasoned that these two changes from Experiment 9 might enhance the efficacy of our manipulation by (1) focusing participants’ attention on evaluation of the delayed, rather than the immediate, reward, the evaluation of which may be more likely to be affected by episodic processes, and (2) making it easier for participants to evaluate the delayed reward in general.

Participants.

Forty participants from the University of Pennsylvania community who had not done a similar study previously completed this study (24 F, 16 M; mean age = 21.55; SD = 3.03).

Procedure.

The procedure was similar to Experiment 8, with a few changes. First, the tasks were all administered in Qualtrics, instead of a combination of Qualtrics and E-Prime. Second, the intertemporal choices were altered. Participants completed two different 14-item intertemporal choice tasks, one following the Gratitude writing task and one following the Typical Day writing task. Each choice was between $10 today and a larger amount of money available after 7, 30, 60, or 180 days. The possible delayed reward amounts were $15, $20, $25, $30, $35 and $40. The precise combinations of delays and delayed reward amounts were different for the two tasks, so that no trial was repeated. The discount rates that could be computed from the two 14-item choice sets were comparable, and which choice set was paired with which condition (Typical Day or Gratitude) was counterbalanced across subjects. After both imagination and intertemporal choice tasks, participants rated the extent to which within the last 24 hours they felt each of the emotions that they had rated previously (since general levels of gratitude may be correlated with discount rate: Dickens & DeSteno, 2016). As in Experiment 8, participants received what they chose on one randomly-selected trial on a debit card.

Once again, we confirmed that participants felt significantly more grateful after the Gratitude writing task than after the Typical Day writing task (t₃₉ = 8.74; p < 0.001).

2.2.10. Experiments 10 and 11: Gratitude recall replication online (unpublished)

Participants.

Seventy-four participants completed Experiment 10. Three were excluded for not following instructions (did not write about their typical day and/or gratitude), leaving n = 71 (28 F, 43 M; mean age = 33.28; SD = 9.91) in final analyses. Seventy-seven participants completed Experiment 11. One was excluded for not following instructions, leaving n = 76 (47 F, 28 M, 1 not reported; mean age = 37.39; SD = 12.85) in final analyses.

Procedure.

Experiment 10 was identical to Experiment 9, except that data were collected online using Amazon’s mechanical Turk and both writing tasks were shortened from 5 to 3 minutes. Experiment 11 was identical to Experiment 10, except that the intertemporal choice tasks contained delayed reward amounts that were not rounded to the nearest $5 increment. In Experiment 11, each choice was between $10 today and a larger amount of money available after 7, 30, 60, or 180 days. The possible delayed reward amounts in one task were $16, $21, $26, $29, $34 and $39, and in the other were $14, $19, $24, $31, $36, and $41 (counterbalanced between gratitude and typical day conditions). This way, no combination of delay and delayed reward amount was ever repeated.

For these and all other studies described below, one participant was randomly selected from all participants who completed the task on any given day, and the selected participant received what they chose on one randomly selected trial as a bonus. Participants did not know how many participants there were that day, so they could not know their chance of being selected. Paying one participant on each day that data were collected ensured that an immediate reward payment was possible.

According to participant ratings, feelings of gratitude were significantly induced in both Experiment 10 (difference between Gratitude and Typical Day conditions: t₇₀ = 6.28; p < 0.001) and Experiment 11 (difference between Gratitude and Typical Day conditions: t₇₅ = 8.21; p < 0.001).

2.2.11. Experiment 12: Gratitude event list online (unpublished)

Rationale.

This experiment tested a novel manipulation intended to increase feelings of gratitude through ease of retrieval. We hypothesized that listing five experiences that they were grateful for would feel easy to participants, and that participants would therefore feel they had a great deal to be grateful for. This hypothesis was based on previous research (Schwarz et al., 1991) showing that generating six examples of one’s assertive behavior increased participants’ ratings of their own assertiveness, while being asked to generate 12 examples decreased ratings of assertiveness, presumably since generating 12 examples was more difficult and therefore participants felt they were not as assertive. We also included a “difficult retrieval” condition in this study, in which participants were asked to list 15 experiences they were grateful for. Data from the difficult retrieval condition are not included in the current meta-analysis, however, because we expected that retrieval difficulty would decrease feelings of gratitude, and thus potentially increase temporal discounting (i.e., that condition did not fulfill our inclusion criteria).

Participants.

Ninety-four participants completed this study. Ten were excluded for not following instructions, leaving n = 84 (44 F, 39 M, 1 not reported; mean age = 34.18; SD = 8.74) in final analyses.

Procedure.

Participants from Amazon mechanical Turk first completed a 17-item intertemporal choice task; this was used to establish a pre-intervention discount rate. Then, participants completed a writing task in which they were instructed, “Now please list 5 personal experiences that make you feel grateful when you remember them. You do not need to write more than one sentence per blank, but all of the experiences must be distinct from each other.” Each blank was preceded by the prompt, “I feel grateful when I remember the time that…” Participants had to type a minimum of 10 characters for each blank and could not proceed until at least one minute had elapsed. After this writing task, participants made the same emotion ratings as in Experiments 8-11. Then, they completed a second 17-item intertemporal choice task, with similar (but distinct) choices and a future self-continuity task (Ersner-Hershfield et al., 2009). In the future self-continuity task, they saw a series of circles representing different degrees of overlap between their current self and their future self and selected which degree of overlap best described how similar and connected they felt to their future self; then they rated on a scale from 1 to 7 how much they liked and cared for their future self. The order of the future self-continuity task and the second intertemporal choice task was counterbalanced between subjects. Finally, participants rated how difficult it was to list 5 experiences that they were grateful for (scale of 1 to 10), and the extent to which they felt each of the emotions that they had rated previously within the last 24 hours.

Since this experiment did not include a control, “typical day,” writing task, we could not conduct the manipulation check in the same way here as in Experiments 8-11. Instead, we compared the gratitude ratings directly following the event listing task to the feelings of gratitude that participants reported having in the last 24 hours. We found that feelings of gratitude were significantly elevated following the gratitude event listing task (t₈₄ = 4.01; p < 0.001). Note that this is a conservative analysis, since the ratings from the last 24 hours may be influenced by the event list task.

2.2.12. Experiments 13 and 14: Nostalgia recall online (unpublished)

Rationale.

Experiments 13 and 14 attempted to replicate a previous study (Huang et al., 2016) that found that recalling nostalgic events made participants more patient in intertemporal choice. Nostalgia is frequently elicited by recall of positive past events, including events that induce gratitude. Therefore, we wanted to test whether this emotion moderated the effect of positive memory recall on temporal discounting. It is worth noting that nostalgia itself is not a strictly positive emotion; a nostalgic reminiscence is often accompanied by a sense of loss for the past (Holak & Havlena, 1998). Nevertheless, this emotion was induced by recalling autobiographical memories that were typically positive. We also cannot rule out the possibility that the positive memories retrieved in other studies did not also induce mixed emotions as well. To test whether that the effect of nostalgia may be different from the effects of gratitude or of positive memory recall generally, we included nostalgia as a potential moderator in the analyses below, though we note these results should be interpreted with caution, as only two of our fourteen experiments were designed to induce nostalgia.

Participants.

Seventy-four participants completed Experiment 13. Six were excluded for not following instructions, leaving n = 68 (39 F, 29 M; mean age = 35.57; SD = 11.79) in final analyses. Seventy-six participants completed Experiment 14. Seven were excluded for not following instructions, leaving n = 69 (37 F, 32 M; mean age = 35.45; SD = 10.55) for final analyses.

Procedure.

Experiments 13 and 14 were similar to Experiments 10 and 11 (gratitude recall online). The control condition was a task that involved writing about a typical day for 3 minutes. The intertemporal choice tasks included 14 items each; in Experiment 13 they featured “round” delayed reward amounts and in Experiment 14 they featured delayed reward amounts that were not in $5 increments. The only difference from Experiments 10 and 11 was that, in the experimental condition, rather than writing about a past event that made them feel grateful, participants wrote about a past event that made them feel nostalgic. After this 3-minute writing task, participants thought about what they wrote for 1 minute, just as in the gratitude studies. Participants also made the same emotion ratings as in the gratitude studies, though they additionally rated the extent to which they “felt nostalgic” and were “having nostalgic feelings” (scales from 1 to 9) after both writing tasks. We averaged the answers to these two questions and used this average as our measure of nostalgic feeling (as in Huang et al., 2016). We found that in both experiments, people reported feeling more nostalgic following writing about a nostalgic experience, compared to following writing about a typical day (n = 2 missing nostalgia ratings in Exp. 13, n = 1 missing nostalgia ratings in Exp. 14; Experiment 13: t₆₅ = 7.34; p < 0.001; Experiment 14: t₆₇ = 8.72; p < 0.001).

2.3. Statistical analysis

In all studies, the dependent variable was the same – hyperbolic discount rate. Discount rates were calculated separately for the experimental and control conditions. Participants’ intertemporal choice data were fit with the following logistic function using maximum likelihood estimation:

$$P_1=\frac{1}{1+e^{-\beta (SV1-SV2)}},P_2=1-P_1$$

Here, ${\mathrm{P}}_1$ refers to the probability of choosing the delayed option, and ${\text{SV}}_1$ and ${\text{SV}}_2$ are the subjective values of the delayed and immediate options, respectively. The subjective value of the options was assumed to follow a hyperbolic discounting function (Kable & Glimcher, 2007; Mazur, 1987):

$$SV=\frac{A}{1+kD}$$

Where $\text{SV}$ is the subjective value, A is the amount, $\mathrm{D}$ is the delay to receiving the reward, and $k$ is the subject-specific discount rate parameter (higher $k$ values correspond to more impatience). The hyperbolic function fits temporal discounting data well (Kable & Glimcher, 2007, 2010). Since discount rates are not normally distributed, these parameters were log-transformed before statistical analyses were performed.

To compute an effect size Cohen’s $d_z$ for each study, we performed a paired t-test comparing the log-transformed discount rate in the experimental condition to the control condition. We computed our meta-analysis and follow-up meta regressions in STATA, using a random-effects meta-analytic model. We evaluated the heterogeneity across studies by calculating the $I^2$ value, and determined significance using the $Q$ statistic.

2.4. Potential moderators

The following were tested as potential moderators in a series of random-effects meta-regressions.

Publication status.

Effect sizes are likely to be larger in published studies, since null results are often unpublished. Therefore, many meta-analyses test for publication bias, to examine the likelihood that significant results are driven by the fact that mostly positive results are available to be found and included in the meta-analysis. The current paper includes mostly unpublished studies, making publication bias less of a concern. Nevertheless, for completeness, we chose to include the three positive memory recall experiments from Lempert et al. (2017). We felt this was important because several of our unpublished studies were exact or conceptual replications of experiments from that paper. To test for any effects of publication status, we compared the effect sizes in the Lempert et al. (2017) experiments (Exps. 1, 2, and 3) to the effect sizes from the unpublished experiments.

Gratitude.

One previous study, with a between-subjects design, found that having participants recall memories that made them feel grateful reduced temporal discounting (DeSteno et al., 2014). We attempted to replicate this effect within-subjects in Experiments 8-12. Note that the manipulation in Exp. 12 was different from the one used in the DeSteno et al. (2014) study, however, and that the precise question sets and samples (online or in-lab) also varied. We tested whether positive memory manipulations that specifically induced gratitude were more or less effective than other manipulations.

Nostalgia.

One paper, reporting the effects of several between-subjects experiments conducted online, found that inducing nostalgia (usually through a writing task) led to more patient consumer decisions (Huang et al., 2016). Unlike our studies here, which involve fitting a temporal discounting rate to a series of choices, most of the studies in Huang et al. (2016) asked participants to make a single choice between a smaller, sooner reward and a larger, later reward (e.g., $20 today vs. $30 in one month) or to rate the extent to which they would be patient in different scenarios (e.g., waiting for a webpage to load). Relatedly, there is also a line of research showing that nostalgia leads to more future orientation and future self-continuity (Cheung et al., 2013; Sedikides et al., 2016; Sedikides & Wildschut, 2016; Zhou et al., 2012). Thus, we attempted to replicate this effect in a within-subjects design with a temporal discounting task. We tested whether positive memory manipulations that specifically induced nostalgia (Exps. 13 and 14) were more or less effective than other manipulations.

Experiment setting (online or in-lab).

Although episodic future thinking has been shown to reduce temporal discounting in online experiments (Sofis et al., 2020; Sze et al., 2017), the effect sizes of experimental manipulations done online may be smaller than those that are done in the lab. One potential reason is that, when participants are doing experiments online, they are not monitored to ensure that they are not doing other tasks at the same time. The risk that participants are not fully engaged is especially high for our experiments here, which involved writing and thinking about previous memories. We attempted to reduce this risk by requiring a character limit for the writing tasks and excluding participants who we judged not to be following instructions. Nevertheless, we expected that in-lab experiments would show larger effects than online ones, so we tested experiment setting as a moderator.

Immediate reward stability.

Relatively few studies have considered how the particular choice sets used to assess temporal discounting affect behavior (cf. Lempert et al., 2015). However, when one of the two options is held stable, this can change how participants allocate attention. For example, in studies that hold the immediate reward constant, only the delayed reward has to be re-evaluated on each new trial. Thus, holding the immediate reward constant might decrease any effects of positive memory recall manipulations on evaluations of the immediate reward. Therefore, we tested if holding the immediate reward constant led to a larger effect size.

3. Results

Looking across all experiments, we found that the average effect of positive memory recall on temporal discounting was small (Cohen’s d_z = 0.079; 95% CI: [−0.008, 0.166]) and not significantly different from zero (z = 1.79; p = 0.074; Fig. 2). The heterogeneity between studies was modest (I² = 30.01%) and not significant (Q (13) = 18.54; p = 0.138). Next, we conducted a series of meta regressions to examine the effects of potential moderators.

Fig. 2.

The effects of positive memory retrieval on temporal discounting. Forest plot showing within-subjects effect sizes (Cohen’s d_z) for each experiment in the meta-analysis, along with overall effect size. The dependent variable in each case is the difference in hyperbolic discount rates. Although the effect was positive (θ = 0.079), it was not significantly different from zero (z = 1.79; p = 0.074). Therefore, the effects of positive memory-based manipulations on temporal discounting are modest at best. Note that the first three studies are drawn from a published paper (Lempert et al., 2017), but since we used different inclusion criteria, the effect sizes reported here differ from those previously published.

Publication status.

Experiments 1-3 were previously published in Lempert et al. (2017), while Experiments 4-14 have not been published previously. There was a significant effect of publication status on effect size, with Experiments 1-3 showing larger effect sizes (β = 0.281, 95% CI: [0.091, 0.471], z = 2.90; p = 0.004).

Gratitude.

Experiments 8-12 attempted to induce feelings of gratitude with positive memories. Positive memories that induced gratitude were not more effective at reducing temporal discounting than those that did not induce gratitude (β = −0.121, 95% CI: [ −0.286, 0.043], z = −1.44; p = 0.149). When controlling for the effects of publication status (a significant moderator), gratitude was still not a significant moderator of effect size (β = −0.053, 95% CI: [−0.211, 0.105], z = −0.66; p = 0.511).

Nostalgia.

Two experiments (Experiments 13 and 14) asked participants to recall memories that made them feel nostalgic. Nostalgia was not a significant moderator of the effect of positive memories on temporal discounting (β = −0.008, 95% CI: [−0.250, 0.235], z = −0.06; p = 0.951) even when controlling for the effect of publication status (β = 0.067, 95% CI: [−0.124, 0.257], z = 0.69; p = 0.491).

Experiment setting.

Eight of the 14 experiments took place in the lab, whereas six were conducted online via Amazon’s mechanical Turk platform. Experiment setting was a significant moderator (β = 0.174, 95% CI: [0.026, 0.323], z = 2.31; p = 0.021), such that experiments that were done in the lab yielded larger effect sizes. However, experiment setting was no longer a significant moderator after controlling for publication status (β = 0.084, 95% CI: [−0.087, 0.256], z = 0.96; p = 0.336), suggesting that this effect may be driven by the fact that the three previously published studies were all done in person, rather than online.

Immediate reward stability.

In ten of the 14 studies, the immediate reward was the same on each trial (e.g., always $20 today). Whether the immediate reward was held constant had no significant moderating effect (β = 0.131, 95% CI: [−0.051, 0.312], z = 1.41; p = 0.158), even when controlling for publication status (β = 0.070, 95% CI: [−0.095, 0.234], z = 0.83; p = 0.406).

4. Discussion

In this meta-analysis of studies from our lab, most of them previously unpublished, we found that positive memory recall-based manipulations were not effective in reducing temporal discounting within-subjects. Of the moderators we tested, only publication status had a significant effect, with the three experiments published in Lempert et al. (2017) showing larger effect sizes than the previously unpublished studies. After controlling for publication status, inducing specific emotions such as gratitude or nostalgia, conducting the study in-person rather than online, or keeping immediate rewards constant did not have a significant moderating effect.

This meta-analysis included only studies that were conducted by the first or last author (Lempert and Kable). A major strength of this approach is that the studies included are fairly homogeneous. The study designs were very comparable, and the data were analyzed in the same way in all studies. This homogeneity was confirmed statistically; the I² measure of heterogeneity was ~30%, which is low and not statistically significant. Thus, there are unlikely to be systematic differences between studies beyond what would be expected from sampling error alone. Doing an internal meta-analysis also increased our power to detect effects because we did not have to control for factors that might vary across institutions and labs. Our approach also addresses one of the major limitations of meta-analysis: the problem of comparing “apples to oranges” (Sharpe, 1997). The biggest threat to the validity of internal meta-analysis is selective reporting (Vosgerau et al., 2019), but here we have included all of the available data that fulfilled inclusion criteria. The main limitation of our approach is that our conclusions may not generalize to other designs. For example, since our manipulations were all within-subjects, we cannot rule out the possibility that the effects of positive memory recall on temporal discounting may be larger when assessed between-subjects (Ciaramelli et al., 2019; DeSteno et al., 2014; Huang et al., 2016).

All of the manipulations we examined here were within-subjects, while the majority of temporal discounting manipulations have been attempted between-subjects (Lempert & Phelps, 2016; Rung & Madden, 2018b). A meta-analysis of the effects of episodic future thinking manipulations on temporal discounting (Rösch et al., 2022) found that effect sizes were smaller in within-subjects than between-subjects designs. This result is somewhat surprising because (1) the potential for experimenter demand effects is higher in within-subjects designs, and experimenter demand would inflate effect sizes, (2) within-subjects designs have more power to detect effects, and (3) there is substantial between-subject variability in temporal discounting that might reduce effect sizes in between-subjects designs. One possible explanation for the smaller effect sizes in within-subjects designs is that there are “carryover” effects; perhaps participants are unable to “turn off” episodic future thinking when switching from experimental blocks to control blocks, and therefore, control discount rates are lower than they would be otherwise (Rösch et al., 2022). This explanation is unlikely to explain our null results, however, since only some of our studies alternated experimental and control blocks. In some of our experiments, discount rates after autobiographical memory recall were compared to discount rates before the intervention, so it would not be possible for episodic memory recall to exert an effect on choices in the control condition. Instead, we interpret our results as confirming that temporal discounting rates are quite resistant to change, especially when discounting measurements are done around the same time. This resistance to change is consistent with the view that temporal discounting is a reliable individual difference metric (Hardardottir, 2017) that can potentially be used as an endophenotype or vulnerability factor for some disorders (Bickel, 2015; Lempert et al., 2019).

Nevertheless, some manipulations do consistently lead to within-person changes in discounting (see Rung & Madden, 2018 for a meta-analysis). As mentioned above, episodic future thinking systematically reduces discounting (Rösch et al., 2022), but other types of “priming” manipulations have weak and inconsistent effects (Rung & Madden, 2018). Priming, in this context, typically involves giving participants a preliminary task that is meant to change their affective state or the content of their thoughts; these tasks are often framed as being unrelated to the intertemporal choice task (Rung & Madden, 2018). The manipulations that we assessed here fall into this priming category. In contrast, manipulations that involve the explicit re-framing of choice outcomes, such as making the delayed reward the default option (Loewenstein, 1988) or explicitly signaling the opportunity costs of choosing immediate rewards (Radu et al., 2011) lead to large, replicable reductions in discounting (Lempert & Phelps, 2016; Rung & Madden, 2018). This is consistent with the finding that framing manipulations that do not require effort on the part of the individual tend to be more effective at promoting behavior change (Beshears & Kosowsky, 2020). It is not always feasible to re-frame choices, however, thereby limiting the real-world scope for such manipulations.

While our null finding might be a setback in the search for manipulations that change discounting, it is a step forward for understanding the cognitive mechanisms of discounting. Knowing that the content of episodic simulation has functional consequences at the time of choice adds to our understanding of the mechanisms underlying intertemporal choice. Our results also help contextualize other findings in the literature. For example, individuals with hippocampal damage have deficits in both episodic future thinking and episodic memory retrieval, but their discount rates are similar to healthy controls (Craver et al., 2014; Kwan et al., 2012). In contrast, individuals with semantic dementia, who have a more significant deficit in episodic future thinking but relatively spared episodic memory retrieval, have steeper discount rates (Chiong et al., 2016). Thus, mechanisms involved in future thinking specifically, rather than episodic simulation more generally, appear to drive patient choice. Our results also have practical relevance for experimental design. Studies that test the effects of episodic future thinking on discounting can use autobiographical memory retrieval as a well-matched control condition.

In conclusion, an internal meta-analysis, which pooled data from 758 unique participants, shows that the effects of positive memory recall on temporal discounting are modest at best. Publishing our largely unsuccessful attempts to manipulate temporal discounting may help shift the focus of research in this area to more fruitful avenues. Overall, we hope that “clearing out our file drawer” not only contributes to the literature on behavior change, but also inspires others to do the same.

Table 1.

Study information, including number, genders, and average age of participants, setting, specific emotion elicited, nature of control condition, and details of choice sets used. Note that the first three studies are drawn from a published paper (Lempert et al., 2017), but since we used different inclusion criteria, the participant information reported here differs from that which was previously published. fMRI = functional magnetic resonance imaging. “Round” amounts are monetary amounts that are multiples of 5 (e.g., $9 is “unround,” while $10 is “round”). “Round” delays are delays that correspond to a discrete number of months or weeks (e.g., 30 days is “round,” since it is equivalent to 1 month, but 29 days is “unround”).

Study	N	Sex	Age	Setting	Emotion elicited	Control condition	Choice characteristics
1. Positive memory recall (Lempert et al., 2017)	47	33 F, 14 M	M = 21.19 SD = 4.20	In-lab (NYU)	Positive (generic)	“Relax” screen, blocks interspersed	Immediate amount: $10 Delayed amounts: varied, “unround” Delays: 4, 7, 30, 60, 100, 180 d
2. Positive memory recall rep1 (Lempert et al., 2017)	48	30 F, 18 M	M = 23.21 SD = 3.61	In-lab (NYU)	Positive (generic)	“Relax” screen, blocks interspersed	Immediate amount: $10 Delayed amounts: varied, “unround” Delays: 4, 7, 30, 60, 100, 180 d
3. Positive memory recall fMRI (Lempert et al., 2017)	40	28 F, 12 M	M = 23.03 SD = 3.46	In-lab (MRI; NYU)	Positive (generic)	“Relax” screen, blocks interspersed	Immediate amount: $20 Delayed amounts: varied, “unround” Delays: 4, 7, 30, 60, 100, 180 d
4. Positive memory recall rep2 (unpublished)	38	26 F, 12 M	M = 23.00 SD = 3.71	In-lab (Penn)	Positive (generic)	“Relax” screen, blocks interspersed	Immediate amount: $10 Delayed amounts: varied, “unround” Delays: 4, 7, 30, 60, 100, 180 d
5. Positive memory recall: photo cues (unpublished)	36	30 F, 6 M	M = 21.83 SD = 3.77	In-lab (Penn)	Positive (generic)	Nature photos, blocks interspersed	Immediate amount: $10 Delayed amounts: varied, “unround” Delays: varied, “unround”
6. Positive memory recall: field & observer perspective (unpublished)	42	30 F, 12 M	M = 22.18 SD = 3.82	In-lab (Penn)	Positive (generic)	“Relax” screen, blocks interspersed	Immediate amount: varied, “unround” Delayed amounts: varied, “unround” Delays: varied, “unround” Choices around indifference point
7. Positive memory recall online (unpublished)	65	36 F, 28 M, 1 not stated	M = 37.94 SD = 12.99	Amazon mTurk	Positive (generic)	Pre-intervention	Immediate amount: varied, “unround” Delayed amounts: $25, $30, $35 Delays: varied, “unround” Same choices in both conditions
8. Gratitude in-lab (unpublished)	34	n/a	n/a	In-lab (Penn)	Gratitude	Writing about typical day, block order counterbalanced	Immediate amount: varied, “unround” Delayed amounts: $25, $30, $35 Delays: varied, “unround”
9. Gratitude in-lab – “round” number amounts (unpublished)	40	24 F, 16 M	M = 21.55 SD = 3.03	In-lab (Penn)	Gratitude	Writing about typical day, block order counterbalanced	Immediate amount: $10 now Delayed amounts: varied in $5 increments Delays: 7, 30, 60, 180 d
10. Gratitude online – “round” number amounts (unpublished)	71	28 F, 43 M	M = 33.28 SD = 9.91	Amazon mTurk	Gratitude	Writing about typical day, block order counterbalanced	Immediate amount: $10 now Delayed amounts: varied in $5 increments Delays: 7, 30, 60, 180 d
11. Gratitude online – not “round” number amounts (unpublished)	76	47 F, 28 M, 1 not stated	M = 37.39 SD = 12.85	Amazon mTurk	Gratitude	Writing about typical day, block order counterbalanced	Immediate amount: $10 now Delayed amounts: varied, “unround” Delays: 7, 30, 60, 180 d
12. Gratitude online: list of 5 events (unpublished)	84	44 F, 39 M, 1 not stated	M = 34.18 SD = 8.74	Amazon mTurk	Gratitude	Pre-intervention	Immediate amount: varied, “unround” Delayed amounts: $25, $30, $35 Delays: varied, “unround” Same choices in both conditions
13. Nostalgia online - “round” number amounts (unpublished)	68	39 F, 29 M	M = 35.57 SD = 11.79	Amazon mTurk	Nostalgia	Writing about typical day, block order counterbalanced	Immediate amount: $10 now Delayed amounts: varied in $5 increments Delays: 7, 30, 60, 180 d
14. Nostalgia online – not “round” number amounts (unpublished)	69	37 F, 32 M	M = 35.45 SD = 10.55	Amazon mTurk	Nostalgia	Writing about typical day, block order counterbalanced	Immediate amount: $10 now Delayed amounts: varied, “unround” Delays: 7, 30, 60, 180 d

Data availability

Participant-level data from all studies are available online in an Open Science Framework repository (https://osf.io/ngk43/).