Reward does not modulate forgetting in free recall tests

Abstract

Reward is thought to attenuate forgetting through the automatic effect of dopamine on hippocampal memory traces. Here we report a conceptual replication of previous results where we did not observe this effect of reward. Participants encoded eight lists of pictures and recalled picture content immediately or the next day. They were informed that they could gain monetary reward for recalling the pictures, with the level of reward indicated through the frame surrounding the picture. Reward was manipulated both within and across lists. Bayesian statistics found moderate evidence for the null hypothesis that reward does not modulate forgetting in human free recall.

When people encounter items that they believe are useful to them (for example, those that will help them gain reward), they later remember such items particularly well. While there is plenty of evidence that people actively use strategies to maximize the chances of remembering reward-predicting experiences, in addition to the preferential attention such experiences attract obligatorily (Pearson et al. 2015; Hennessee et al. 2019), it is not yet entirely clear how the impact of competition for memory resources unfolds through time—from encoding, through maintenance, to eventual retrieval.

Many previous publications have reported that reward increases delayed memory more than immediate memory or that its memory-enhancing effects are only significant after a prolonged retention interval (Wittmann et al. 2005; Murayama and Kuhbandner 2011; Montagrin et al. 2013; Murayama and Kitagami 2014; Spaniol et al. 2014). For example, in a study that used a between-group manipulation, where some participants were promised a monetary reward for answering trivia questions but others were not (Murayama and Kuhbandner 2011), reward increased delayed, but not immediate, memory. The dominant interpretative framework for such results is that they are due to catecholamine-dependent, and specifically dopaminergic, modulation of memory consolidation of hippocampal memory traces (McGaugh et al. 2000; Adcock et al. 2006; Lisman et al. 2011; Yamasaki and Takeuchi 2017).

In keeping with this framework, even though previous studies used recognition memory tests, we predicted that reward will attenuate forgetting in free recall as well. Delayed free recall tests of memory are relatively rare because in the second, delayed, session, it is difficult to instruct the participant to recall from a specific list they encoded during the first session. This practical difficulty means that delayed free recall tests are typically limited to resource-intensive experiments in which participants encode and recall a single list or to “final free recall” test of all studied lists together. We have overcome this limitation in our previous study by presenting participants with categorized lists (e.g., landscapes and household objects) and cuing them with category names prior to the free recall test (Talmi et al. 2021). Note that we refer to the memory test as “cued free recall” to accurately describe that a list-wise but not item-wise cue was present. Surprisingly, we found that items that predicted high reward were not preferentially retained, suggesting no detectable role of reward prediction on consolidation. In that experiment, we promised participants £1 for remembering some pictures, but only £0.10 for remembering others. Their ability to describe the pictures was tested after 1 min and after 24 h. Memory at the immediate test showed that the local context was key to the effect of reward on memory. Reward improved recall only when items that predicted high and low reward were studied and recalled in the same list. Reward did not increase memory when items that predicted high and low reward were presented and recalled separately, in blocked lists. Crucially, reward did not modulate forgetting rates, computed as proportions of items retained, and even increased forgetting when it was computed simply through the number of items recalled. A reanalysis of forgetting rates found that evidence for the null effect ranged from anecdotal for the pure lists (BF01 = 2.89) to moderate for the mixed lists (BF01 = 6.56).

To the best of our knowledge, only one other study investigated the effects of reward and delay on memory recall (Horwath et al. 2022). They, too, found no interaction between reward and delay, supporting the claim that reward does not modulate forgetting when memory is measured with free recall tests. Interestingly, one of the experiments Horwath et al. (2022) conducted was similar to the traditional “mixed list” design, where high- and low-reward items were interleaved, but in the other experiment, participants encoded high- and low-reward items in blocks, a design that resembles a traditional “pure list” condition. Crucially, in both experiments, participants were required to recall all items together. While Talmi et al. (2021) were unable to pinpoint whether the greater effect of reward in the mixed list than in the pure list condition was due to encoding or retrieval effects, Horwath et al. (2022) observed that reward enhanced recall regardless of study conditions, suggesting that competition during retrieval, rather than during encoding, underlies the list composition effect reported by Talmi et al. (2021).

One possibility for the aberrant observation in memory recall, compared with previous recognition studies, is the nature of the test. However, a less interesting reason for the null effect of reward on forgetting in our previous study is floor effects, given that delayed memory in the most decisive condition, where memory for each list was only tested once, either immediately or after a 24-h interval, was poor, averaging 20%. In Horwath et al. (2022), the encoding stage was repeated three times to prevent floor effects, but perhaps as a consequence, no significant forgetting occurred in their first experiment. The question of how motivation influences forgetting is important, and we therefore sought to replicate our previous work conceptually while making some changes to improve average performance in the delayed test. We have preregistered the experiment reported here (https://osf.io/fep6d) with a focus on illuminating the effect of reward on forgetting.

Participants were recruited from the Cambridge Sona volunteer pool. Inclusion criteria were fluent English and normal or corrected to normal hearing and vision, and exclusion criteria were history of psychiatric or neurological disorders and taking psychoactive medicine. Forty participants between the ages of 18 and 40 yr completed two study sessions. We have preregistered a sample size of N = 34 “clean” data sets, which allows detection of a medium effect size with a power of 0.8. In our previous study, admission of rehearsal was an exclusion criterion, and we used the same criterion here. While most participants (N = 31) expected a memory test in the second session, only three participants actively rehearsed between sessions. The final sample included N = 37, of whom 24 identified as female and one identified as nonbinary. We have preregistered excluding participants who have not recalled any pictures in a particular reward condition. One participant did not recall any pictures in two of the delayed conditions and was thus excluded from the respective analyses.

The materials—eight sets of colored photographs that depicted eight different topics (e.g., household objects and landscapes)—were identical to those used in Talmi et al. (2021). Regardless of whether the list was allocated to the pure or the mixed list condition, all the pictures in each list belonged to one category. The order of pictures in each list and its assignment to the eight list composition × reward × delay conditions were randomized for each participant.

Participants were presented with eight lists of 16 pictures. Pictures were presented for 2 sec each, with a jittered stimulus interval with an average of 4 sec. Participants were told that if they see the instruction “Recall,” they should write down a short description of the content of the pictures they saw. The name of that category was announced before the first picture and accompanied the instruction to recall. Half the lists were tested in the first session, with 4 min for each recall period (i.e., one list for each list composition [mixed/pure] × reward [high/low] condition). Participants were further told that each picture would be framed in either blue or brown and that the color indicates how much money they may get if they recall that picture, with one color assigned to high reward (£0.5) and one assigned to low reward (£0.05). The assignment of color to reward condition was counterbalanced across participants, and participants were explicitly informed what color indicated high reward and then tested to ensure they remembered the color. At the end of session one, participants were thanked and reminded of their video call appointment the next morning. The same experimenter met them in the video call session and asked the debriefing questions listed below. Like in the immediate test, the participants were given the category name and had 4 min to retrieve each list. All eight lists from the encoding were tested in the delayed test (one list from each list composition [mixed/pure] × reward [high/low] × delay condition [only immediate test/both immediate and delayed test]). Half of the lists had already been tested in the immediate test, and the other half were tested for the first time in the delayed test (one list for each list composition × reward condition).

This procedure resembled the one used in our previous study with the following preregistered exceptions. First, the name of the category from which pictures were drawn was announced before the study list was presented and announced before the cued free recall test in order to decrease measurement noise and equate the immediate and delayed conditions (in our previous study, it was only announced before the test in the delayed recall condition). Second, here we framed both high- and low-reward stimuli using equiluminant brown and blue frames, while the previous experiment had only framed high-reward pictures. Unfortunately, the original preregistered experiment had to be abandoned in March 2020 when the COVID pandemic hit. We relaunched it in June 2021 once the laboratory reopened and introduced the following nonregistered changes to meet COVID-related restrictions. First, the experimenter could not be in the same room as the participant for more than a few minutes. Because they could not monitor participants during the completion of the distractor task and there was a chance that the participant would begin recall during that period, we decided to eliminate the distractor task that separated the encoding and the cued free recall test in our previous experiment. Second, to decrease in-person contact with participants, we conducted the second session online using a video call. This meant that participants wrote their answers longhand in the immediate test (as in the previous study) but typed it into the chat in the delayed test. Third, we formalized the debriefing so that we could more clearly capture which participants actively rehearsed between the two sessions. Before the delayed recall was introduced, the experimenter asked the following questions and noted down participants’ responses. (1) After you left the laboratory yesterday, did you think back about your experience? (2) Did you talk about your experience with friends or family? (3) Did you write about your experience? (4) Can you tell me a little what you thought, talked about, or wrote about? (5) Did you think about this session today? (6) Can you tell me a little what you thought about? (7) Did you have any specific expectations about what we would do today? Participants 8 and 10 were deemed to have rehearsed on the basis of their responses to questions 1–7. Participants 11 onward were asked two additional questions to facilitate this decision: (8) Did you expect a memory test? (9) Did you rehearse the material? Participant 20 was excluded based on their response to question 9. Fourth, to increase delayed memory performance, especially given the online nature of the delayed test, we decreased the retention interval from 24 h (M = 18.29 h, SD = 1.18 h). The first session always took place in the afternoon, and the second session always took place the next morning. Finally, because the promise of £1 had a robust effect in the previous data set, and because we expected increased levels of recall in this experiment, we decided to only offer £0.50 in the high-reward condition to save resources.

We have preregistered two main analyses. The first focused on list composition effects using a three-way repeated measures ANOVA of proportion recalled, with the factors list composition (mixed list vs. pure list), reward (high reward vs. low reward), and retention interval (immediate test vs. delayed test), and a follow-up focusing only on the delayed test—first using a 2 × 2 repeated measures ANOVA for data in each session, and then planned t-tests to examine the effect of reward in each list composition condition. The analysis was conducted in SPSS 28.0. The second analysis focused on forgetting rates. We preregistered using the d2 measure of retention (Ennaceur and Delacour 1988), computed as follows for each “list by reward” condition:

$$\mathrm{Fogetting}\mathrm{scores}={\displaystyle \frac{\left(\mathrm{average}\mathrm{immediate}\mathrm{recall}-\mathrm{average}\mathrm{delayed}\mathrm{recall}\right)}{\left(\mathrm{average}\mathrm{immediate}\mathrm{recall}+\mathrm{average}\mathrm{delayed}\mathrm{recall}\right)}.}$$

The preregistration does not specify which analysis would be applied to forgetting rates, and we have decided to use Bayesian t-tests focusing on the comparison between high and low reward within each list composition condition. We used the “Related Samples Inference: Normal” in SPSS with diffuse priors around the mean and variance.

All results below only refer to the more interesting condition from lists that were only recalled once. In our previous study, we also reported data from lists that were recalled twice—both immediately and after a delay—because those results were less vulnerable to floor effects. The current focus on forgetting rates renders twice-recalled data less relevant, because the delayed data are contaminated by the first recall test and therefore are not reported here.

Written responses were classified as “correct” or as “intrusions.” A subset of coded data was recoded blindly by one of the investigators for quality assurance purposes. Figure 1 shows that the study was successful in its aim to increase average delayed memory performance. While in our previous study delayed memory was vulnerable to floor effects, this was no longer the case here. Figure 1 also suggests that the 18-h retention interval was sufficient to induce forgetting and that reward influenced the proportion of items recalled. The analysis of proportion recalled yielded three significant effects: a main effect of time interval (F_(1,36) = 283.85, P < 0.001, partial η² = 0.89), a two-way interaction between list composition and reward (F_(1,36) = 10.39, P = 0.003, partial η² = 0.22), and a three-way interaction (F_(1,36) = 4.24, P = 0.047, partial η² = 0.1; all other effects had P > 0.10). In the two-way analysis of immediate memory, the interaction was significant (F_(1,36) = 11.57, P = 0.002, partial η² = 0.24). Reward increased memory in mixed lists (t₍₃₆₎ = 3.0, P = 0.005) but not in pure lists, where the number of items recalled from low-reward pure lists was significantly higher than the number of items recalled from high-reward pure lists (t₍₃₆₎ = −2.56, P = 0.015). The main effects were not significant (P > 0.10). In the two-way analysis of delayed memory, neither the interaction of list composition and reward (F_(1,36) = 3.89, P = 0.06, partial η² = 0.1) nor the main effects were significant. Planned paired t-tests showed that reward increased delayed memory significantly in mixed lists (t₍₃₆₎ = 2.23, P = 0.032) but not in pure lists (t₍₃₆₎ = −0.80, P = 0.43 < 1).

Figure 1.

Proportion recall as a function of list composition and reward in the immediate (top) and delayed (middle) test and as a function of all three factors (bottom). Error bars refer to the standard error.

Figure 2 suggests that the forgetting rates were similar in each one of the four list composition × reward conditions. Bayesian t-tests provided moderate evidence for the null hypothesis that reward did not influence forgetting from mixed lists (BF01 = 7.56) or pure lists (BF01 = 7.59).

Figure 2.

Forgetting rates as a function of list composition and reward. Error bars refer to the standard error.

In this preregistered study, we replicated a null effect of reward on forgetting in cued free recall. Our results provide evidence that this is not due to a null effect of reward on memory, because both here and in our previous experiment reward increased memory recall robustly. Replicating much previous work, when high- and low-reward items were studied and tested together, reward increased recall. We observed this effect both in the immediate test and in the delayed test. Replicating our own and other previous studies (Nilsson 1987; Ngaosuvan and Mantila 2005; experiment 1 in Stefanidi et al. 2018; Talmi et al. 2021), reward did not increase recall in pure lists. We observed that in the immediate test of pure lists, reward actually decreased recall here, a finding we have not observed before and that would require replication. The average performance in the delayed recall test here was better than in our previous study, providing reassurance that our previous finding of a null effect of reward on forgetting in cued free recall was not due to floor effects. We have corroborated the frequentist analysis with a Bayesian analysis, which provided moderate evidence for the null.

Classical work in experimental psychology has concluded that cognitive manipulations that increase immediate memory do not modulate the forgetting rate (Loftus 1985; Slamecka 1985), and our results align with this literature. In contrast, there are many reports that emotional arousal and reward, which often increase immediate memory, also attenuate forgetting in healthy volunteers, an effect mediated through neuromodulation by norepinephrine, cortisol, or dopamine secreted during encoding that stabilize the consolidation of hippocampal memory traces (McGaugh et al. 2000; Lisman et al. 2011; Yamasaki and Takeuchi 2017). One possibility worth exploring is that the time delays used in animal testing where modulated consolidation has been observed as a result of emotion and motivation are not directly homologous to those used in human testing. For example, norepinephrine modulates consolidation after 24 h but not after 1.5 h in rodents (Bianchin et al. 1999), even though both of these retention intervals are considered tests of long-term memory in humans, which rely on an intact hippocampus. Indeed, although the effect of reward on memory tested 24 h after encoding has been shown to be mediated by the interaction of the reward system and the hippocampus (Adcock et al. 2006; Bowen et al. 2020), similar findings have been reported even when the retention interval was <1.5 h (Gruber et al. 2016). Nevertheless, human findings showing that reward enhances memory after a delay but not immediately point to effects on consolidation that were not observed here. Future research may help elucidate why only some memory-enhancing manipulations influence forgetting.

Alternatively, the neural mechanisms that support memory across varying retention intervals interact with the nature of the material tested and the type of test used. For example, the renowned psychologist Andrew Mayes (Isaac and Mayes 1999), who has recently departed, has shown that forgetting rates in organic amnesia vary as a function of the degree to which materials were organized and whether memory was tested using free recall, cued recall, or recognition. In another example, (a cued recall task of paired associates, where participants were cued with one member of the pair to recall the other), reward enhanced cued recall only after a delay and only when the stimulus was deemed “boring” (Murayama and Kuhbandner 2011).

Speculatively, our results suggest that it may be important to examine whether reward interacts with encoding processes that may be influenced by the stimulus set and the retrieval demands to influence human forgetting. Neuromodulation through automatic effects of dopamine may have more marked effects on tests that are less strongly affected by strategic processes compared with the rather unconstrained recall of organized sets of complex, novel, natural picture scenes. Future research can test this conjecture directly by comparing delayed recall of different materials or comparing performance on two or more memory tests.