The Relationship Between Subjective Memory Experience and Objective Memory Performance Remains Stable Across the Lifespan

This research was preregistered at Open Science Framework (OSF): https://osf.io/5vcpt. Deidentified participant data and analytical code are available at OSF: https://osf.io/ceq86. The analytical code was based on the Cooper and Ritchey (2022) analysis code available at GitHub: https://github.com/memobc/paper-vividness-features, and extended to account for a between-subjects factor of age group and an additional dependent variable of confidence rating. The materials used can be made available upon request. Procedures were approved by the University of Sussex Sciences & Technology C-REC. A shortened example version of the paradigm can be accessed at https://h02.eventmemory.org/example/instructions.html.

One-hundred young adults (47 females, 6 non-binary, 47 males) and 100 older adults (50 females, 50 males) participated. Seven participants were excluded under our preregistered exclusion criteria: 4 young adults and 3 older adults performed below a threshold of 30% correct for conceptual gist memory (chance = 25%). This resulted in a final sample size of 96 young adults aged 20-25 (M = 21.78, SD = 1.67, 46 females, 6 non-binary, 44 males), and 97 older adults aged 68-75 (M = 71.59, SD = 2.06, 50 females, 47 males). Demographic measures for both age groups are available in Supplementary Materials in Table S1. There was a greater variety in the achieved educational level in the older adult group. For this reason, we estimated linear mixed effect models to verify that vividness and confidence ratings, as well as overall memory performance, are predicted by the main variables of interest while controlling for participants’ educational level, which was not a significant predictor. The results of this analysis can be found in Supplementary Materials in Tables S2-S4 available at https://osf.io/ceq86. All participants were United Kingdom residents.

All participants completed the study in one online session, lasting approximately 45 minutes at a self-guided pace. See Figure 1 for details of the experimental paradigm, which was based on Experiment 2 from Cooper and Ritchey (2022), coded using PsychoPy (Peirce et al., 2019) and stored on a privately hosted web-server. Instructions were provided in written and audio versions simultaneously, and the first phase of the experiment started immediately after the instructions. The study consisted of four parts.

In the first part (encoding phase), participants were shown 24 stimuli sets, each comprising four elements, which constituted a single event. The elements were randomly assigned to one of the 24 events, but the combination of elements into events was fixed across participants. Each set contained a labelled photo of (1) a famous person, (2) an object, (3) a place, and (4) a theme word (e.g. “Party”), presented without a photo. Each element was uniquely associated with a single event. Famous person images were selected based on a pilot study, with participants in the same young and older adult age range as the current study, all of whom were United Kingdom residents. Participants were asked to generate a meaningful, distinct story to create an association between the elements in each stimuli set and then press the spacebar. The time limit was 20 seconds.

In the second part, the test phase instructions followed immediately after completion of the encoding phase. Confidence and vividness were defined using an example during test phase instructions. Participants were told that the ability to vividly visualise an event does not necessarily correspond to how confident one can be in the accuracy of their memory for that event. They were given an example in which a friend tells them that the last time they saw each other, they met in a café. One may be able to vividly visualise sitting inside that café. However, it may not correspond to how confident a person may feel that it really was the place they met their friend most recently. Conversely, it is possible to feel confident about the accuracy of one’s memory without vividly re-experiencing the event. For example, a person may be sure that they last saw their friend in the café, without being able to vividly visualise it.

Memory for all 24 stimuli sets was tested and all test responses were self-paced. First, conceptual gist memory was tested, using a four-alternative forced-choice recognition test, separately for people, objects and places. A theme word was presented as a retrieval cue, together with four word labels of previously studied people, places or objects. One of the labels was associated with the event indexed by the theme word; the other three labels were from different events. Following this, participants were asked to rate how confident they were about their memory and how vivid their memory was using continuous scales, anchored between one and six. Afterwards, participants were tested on their perceptual detail memory using a two-alternative forced-choice recognition test. Each test trial comprised two conceptually identical and perceptually similar pictures of the same person, object or place; one of the pictures was identical to the studied image. The pictures shown in the perceptual detail memory test were always correct in regard to the theme word they were studied with, regardless of whether participants responded correctly to the conceptual gist questions. The test order for the three elements of content manipulation (person, place or object) was counterbalanced across trials but fixed for conceptual gist and perceptual detail questions within the same trial.

In the third part of the study, participants indicated whether they recognized each of the famous people. Finally, the fourth step consisted of two questionnaires: Vividness of Visual Imagery Questionnaire (VVIQ, Marks, 1973) and the Survey of Autobiographical Memory (SAM, Palombo et al., 2013).

Following Cooper and Ritchey (2022), we carried out the original analysis as applied in their paper. Additionally, we analysed confidence ratings. Critically, we also investigated age as a between-subjects factor. The main analyses of interest focused on the trial-wise relationship between subjective ratings of confidence and vividness and objective memory attributes. This was done in two ways: using within-participant correlations between subjective memory ratings and each of the memory attributes, and predictive linear models with repeated cross validation exploring which memory attributes predict subjective memory ratings.

Additionally, we explored individual differences in the relationship between the subjective ratings and memory attributes. We calculated correlations between mean subjective memory ratings and two questionnaires - SAM and VVIQ. We also ran K-Means clustering analysis to group participants based on the similarity, defined as Euclidean distance, of the correlations between all memory attributes and the subjective memory ratings.

Furthermore, we conducted an exploratory generalised linear mixed-effects model which contained data from both subjective ratings and age groups within a single model. The results of this analysis can be found in Supplementary Materials at https://osf.io/ceq86.

We first tested for differences in memory performance and subjective memory ratings between the two age groups. Older adults performed significantly worse than young adults on both conceptual gist and perceptual detail memory. This was measured using a Mann-Whitney U-test separately for conceptual gist and perceptual detail. Young adults remembered significantly more items at conceptual gist memory level (M_young = 2.31, SD_young = 0.96) than older adults (M_older = 2.04, SD_older = 1.05), U = 6085.0, p \< .001, r = -0.31, as well as at perceptual detail memory level (M_young = 2.18, SD_young = 0.82, M_older = 2.07, SD_older = 0.84), U = 5606.5, p \< .05, r = -0.20.

Older adults reported significantly lower memory vividness ratings (M_young = 3.28, SD_young = 0.91, M_older = 2.48, SD_older = 0.94), U = 2457.0, p \< .001, r = 0.47, and confidence ratings (M_young = 3.75, SD_young = 0.94, M_older = 2.68, SD_older = 1.05), U = 2084.0, p \< .001, r = 0.55, in contrast to young adults. Furthermore, there was a strong significant trial-wise correlation between confidence and vividness ratings within both young adults’ group (r(2207) = 0.78, p \< .001) and older adults’ group (r(2230) = 0.90, p \< .001).

For subsequent analyses, and following Cooper and Ritchey (2022), perceptual detail memory was set to incorrect if the answer to the corresponding conceptual gist element was incorrect within the same event. This is because detailed perceptual information, always tested on correct event elements, could not have contributed to the subjective memory experience if the gist of the feature was not first successfully recalled. To verify that this decision does not affect the overall pattern of results, we have also run all analyses on data where detail memory performance was not set to be conditional upon gist memory performance. Out of 34 analyses, only one result has changed, with a correlation between detail memory for object and vividness ratings becoming significant in older adults when detail performance is not restricted to trials where gist memory is correct. The results of the verification analyses can be found at https://osf.io/ceq86.

We tested whether conceptual gist and perceptual detail memory for place, person and object are significantly correlated with vividness and confidence ratings, using one sample t-tests with Bonferroni correction (these correlations are shown in Figure 2A). These t-tests were performed separately for vividness and confidence correlations, and separately for young and older adults. Replicating Cooper and Ritchey (2022), conceptual gist memory for all three types of content (place, object, person) was significantly related to vividness ratings amongst young adults (mean z >= 0.31, ts(95) >= 10.39, ps \< .001, d >= 1.06). We found the same pattern for conceptual gist memory in older adults (mean z >= 0.35, ts(96) >=11.43, ps \< .001, d >= 1.16). As in Cooper and Ritchey (2022), only perceptually detailed memory for the place was significantly correlated with memory vividness amongst young adults (z = 0.08, t(95) = 2.85, p \< .05, d = 0.29), with person and object memory scores being non-significant (z \<= 0.07, ts(95) \<= 2.31, p >= 0.14, d \<= 0.24). However, we did not find this relationship in older adults, with none of the perceptual detail memory attributes being significantly correlated with memory vividness (z \<= 0.06, ts(96) \<= 2.13, p >= 0.21, d \<= 0.22). Nevertheless, there was no significant difference between the older and young adult groups’ correlations of memory vividness and detail place memory, t(180.52) = 1.63, p = 0.10, d = 0.24.

Regarding confidence ratings, the results followed a nearly identical pattern to vividness. Conceptual gist memory for all three types of content was significantly related to confidence ratings in both young adults (mean z >= 0.34, ts(95) >= 11.64, ps \< .001, d >= 1.19), and older adults (mean z >= 0.36, ts(96) >= 12.41, ps \< .001, d >= 1.26). Perceptual detail memory for the place and object was significantly correlated with confidence ratings in young adults (mean z >= 0.10, ts(95) >= 3.10, ps \< .05, d >= 0.32), but not person detail memory (mean z = 0.03, t(95) = 0.87, p = 1.00, d = 0.09). None of the perceptual detail memory attributes were correlated with memory confidence in older adults (mean z \<= 0.07, ts(96) \<= 2.22, ps >= 0.17, d \<= 0.23).

To test whether older adults rely on conceptual gist to a greater extent than young adults to assess their subjective memory experience while controlling for the effect of memory content, we also ran age (2 levels, between subjects: young adults, older adults) x content (3 levels, within subjects: place, person, object) x specificity (2 levels, within subjects: gist, detail) mixed effect ANOVAs, separately for confidence and vividness ratings. The dependent variable was the correlation between the subjective rating and the memory attributes. Across both subjective measures, there was a significant main effect of specificity for vividness correlations, F(1, 191) = 193.48, p \< 0.001, η²_G = 0.195, and confidence correlations, F(1, 191) = 198.99, p \< 0.001, η²_G = 0.207. There was no significant main effect of age or content, and no significant two-way interactions between specificity and age on vividness or confidence correlations.

Following Cooper and Ritchey (2022), we used a predictive linear regression procedure to predict subjective ratings using performance on the six memory attributes (these results are shown in Figure 2B). We replicated the pattern of results for vividness ratings of young adults. We defined the best model as the one with the highest out-of-sample R-squared value. We identified the best model to predict vividness ratings included four memory attribute predictors: conceptual gist memory of object, place and person, as well as perceptual detail memory for place (highest out-of-sample R-squared = 0.24; SD = .04). The best model to predict young adults’ confidence ratings included all three conceptual gist elements, as well as perceptual detail memory for object and place (highest out-of-sample R-squared = 0.43; SD = .04).

For older adults, the best model containing all six memory attributes as predictors explained the most variance in both vividness ratings (highest out-of-sample R-squared = 0.25; SD = 0.04) and confidence ratings (highest out-of-sample R-squared = 0.31; SD = .04).

Given the substantial individual variability in the findings of Cooper and Ritchey (2022), we also explored individual differences in the relationship between the subjective ratings of confidence and vividness and memory attributes. First, we conducted correlations between mean subjective memory ratings (separately for confidence and vividness) and each of the memory attributes, as well as scores on the two questionnaires - the episodic component of the Survey of Autobiographical Memory (SAM-episodic; Palombo et al., 2013) and the Vividness of Visual Imagery Questionnaire (VVIQ; Marks, 1973). These correlations were evaluated at a one-tailed Bonferroni-corrected threshold of r = 0.233, expecting a positive correlation between the scores on the questionnaires and the mean subjective memory ratings.

Cooper and Ritchey (2022) reported significant but weak correlations between mean memory vividness ratings and both the SAM score, as well as the VVIQ score. Furthermore, in their study, mean vividness scores were correlated with all six memory attributes, although to a greater extent with the conceptual gist memory attributes than with the perceptual detail memory attributes. The current study partially replicated Cooper and Ritchey (2022), as shown in Figure 3. In the young adult group, there was a weak but significant correlation between mean memory vividness ratings and the SAM-episodic questionnaire scores, but not the VVIQ scores. Neither questionnaire was correlated with mean memory confidence ratings in young adults. In the older adult group, mean vividness and confidence ratings were significantly correlated with SAM-episodic scores (rs >= 0.33). Furthermore, mean memory vividness but not mean memory confidence was correlated with the VVIQ scores (r = 0.30). Amongst older adults, there was a weak but significant correlation between the SAM-episodic score and gist person memory performance (r = 0.30), but none of the other five memory attributes. VVIQ scores were not significantly correlated with any of the memory attributes amongst older adults, similarly to the results obtained in the young adults group.

Replicating the original study, young adults’ mean vividness scores were significantly correlated with all conceptual gist memory attributes (rs >= 0.37) and perceptual detail memory attributes (r >= 0.27). Mean confidence ratings were also significantly correlated with all conceptual gist memory attributes (rs >= 0.67) and perceptual detail memory attributes (rs >= 0.36). We found no correlation between young adults’ mean memory performance on any of the six memory attributes and the two questionnaires.

Both mean memory vividness and confidence were significantly correlated with all three gist memory attributes in older adults (rs >= 0.45). However, we found no significant correlations between mean vividness or confidence ratings and detail memory attributes.

Second, we performed K-Means clustering to group participants based on the similarity (Euclidean distance) of the correlation between all memory attributes and subjective memory measures. To obtain the optimal number of clusters (k, up to a maximum of 10), we identified the elbow of the curve for inertia values, defined as the sum of squared distances of each participant to their closest cluster centre. For each k iteration, K-Means was run 500 times, returning the cluster assignments with the lowest inertia. K-Means clustering was run separately for confidence and vividness, and separately for older and young adults.

Cooper and Ritchey (2022) reported that even though young adults’ subjective sense of vividness is more closely tied to gist over detail memory at a group level, there are individual differences in that pattern. Using K-Means clustering analysis, they found that participants were clustered into 3 groups. One group showed an equally strong relationship between vividness ratings and all conceptual gist and perceptual detail memory attributes combined, a second group showed a stronger relationship between vividness ratings and conceptual gist memory attributes, and the final group showed a stronger relationship between vividness ratings and perceptual detail memory attributes. We found a similar pattern of results, though with a varying number of clusters throughout the analyses.

The following graphs shown in Figure 4 represent participant clusters from K-Means analysis, which grouped the participants based on their pattern of correlations between vividness memory ratings and the six memory attributes. The results of analyses ran on confidence ratings can be found in Supplementary Materials in Table S5 and in Figures S2-S5. All correlations were z-scored. Note that outliers were identified and removed before the analysis was run for each group and each dependent variable. Outliers were identified as participants with any z-scored memory score with an absolute value greater than 2.5. For the number of participants included in each analysis please see Supplementary Table S5.

Older adults showed a similar pattern of individual differences to the young adult sample in Cooper and Ritchey (2022), with the optimal number of clusters identified being three. The first cluster (N = 32) showed a strong relationship between vividness ratings and all six memory attributes; the second cluster (N = 25) showed a stronger relationship between vividness ratings and the place, object, and person elements of perceptual detail memory; and the third cluster (N = 29) revealed a stronger relationship between vividness ratings and the place, object and person elements of conceptual gist memory as well as detail place memory. The clusters did not differ in their relationship between vividness ratings and detail place memory.

In the young adult sample from the current study, we identified four clusters. The first cluster (N = 25) showed a stronger relationship between vividness ratings and conceptual gist memory of place, object and person elements; the second cluster (N = 29) between vividness ratings and perceptual detail memory of object elements; the third cluster (N = 17) had a very limited relationship with memory attributes overall; and the fourth cluster (N = 18) showed a strong relationship between vividness ratings and all six memory attributes.

This study aimed to establish whether the relationship between objective and subjective measures of memory is consistent across the lifespan, both when considering memory at the level of conceptual gist, as well as perceptual detail. We found that older adults performed at a lower level then young adults, yet the correspondences between objective measures and vividness and confidence ratings (our subjective measures), were very similar across both age groups. Both young and older adults rely on gist to a greater extent than detail to judge their subjective memory experience, which replicates and extends the results of Cooper and Ritchey (2022) who demonstrated this effect in young adults. Contrary to some previous studies, we neither found that older adults were spared in their gist-based memory, nor that vividness ratings were higher in the older adults. Across both age groups, the greater reliance on conceptual gist follows a similar pattern across vividness and confidence ratings, with the two subjective measures being strongly correlated. Our results suggest that throughout the lifespan, the amount of conceptual information retrieved about an event relates to the ability to reexperience it vividly, and to have confidence in one’s memory.

Our key new finding is that the relationship between objective and subjective measures of memory was the same in both the younger and the older age groups. Similarly to younger adults, the older adults predominantly based both ratings of memory vividness and confidence on the retrieval of conceptual gist-level information about the events. Interestingly, while objective levels of performance were lower in the older adults, this was mirrored by lower ratings of vividness and confidence. This suggests that older and younger adults base their phenomenological experience of remembering on the retrieval of qualitatively similar information. Furthermore, this reliance on conceptual gist-level information supports the view that a coherent memory trace involves the binding of primarily gist-level information about distinct features of an event. In contrast, accessing the perceptually detailed information about each feature appears to be a separate, independent process (Cooper & Ritchey, 2019; see also Andermane et al., 2021).

We found little difference between subjective ratings of vividness and confidence, with both being predicted by conceptual gist memory to a greater extent than perceptual detail memory. During the test phase instructions, confidence was defined in terms of belief in how accurate the memory was, whereas vividness was defined in terms of the ability to visualise an event in our mind’s eye. Nevertheless, both within-subject and between-subject ratings of confidence and vividness were highly correlated (Figure 2 and Figure 3; see also Robinson et al., 2000, and Sharot et al., 2007, for similar findings when rating autobiographical memories). It remains an open question whether there are situations when the subjective experiences of memory confidence and memory vividness dissociate (for further discussion, see Folville et al., 2021).

In our study, the older adults performed more poorly on the episodic memory tasks, both at the level of conceptual gist and perceptual detail. This is somewhat at odds with previous studies reporting that gist memory remains intact amongst older adults (for reviews, see Devitt & Schacter, 2016; Grilli & Sheldon, 2022). For example, older adults often retain information at a superordinate level or they retain information about prototypical details, but are poorer at remembering specific details (Flores et al., 2017). In addition, when testing memory for complex, structured material, such as stories or TV shows, the main ideas of the storyline were well-retained in older adults, but they were poorer at remembering literal or perceptual details (Adams et al., 1997; Delarazan et al., 2023). It has been argued that older adults use their schematic knowledge to interpret and remember key information about their experiences, but remember fewer specific details (see also Amer et al., 2022; Grilli & Sheldon, 2022; Umanath & Marsh, 2014). In our study, participants had to create a novel association between a famous person, object and a place, thus not allowing older adults to maintain their performance via the use of pre-existing schemas. Moreover, “gist-level” in our study still referred to specific elements (individuals, objects, and locations) rather than more categorical or superordinate classes of elements (e.g. actors, tools, and interiors).

Older adults often report equally high or even higher experiences of vividness when rating their memories compared to younger adults, even if their objective performance on memory tests is poorer (Folville, Bahri, et al., 2020; Folville et al., 2021). We did not find this; older adults’ subjective ratings were significantly lower than those of younger adults, and this was in line with their poorer performance on the task. Interestingly, studies that reported the overestimation of vividness amongst older adults used free recall (Folville, D’Argembeau, et al., 2020; Robin & Moscovitch, 2017). By contrast, our forced-choice recognition procedure may have influenced our participants’ subjective ratings of their memories. Older adults may have monitored their subjective memory experience by knowing that one of the choices they were presented with was correct and yet they did not recollect it. Nevertheless, the existing literature on the impact of ageing on memory monitoring (“metamemory”) is not clear-cut. A number of previous studies assessing judgements-of-learning and monitoring of forgetting found either negligible or non-existent differences between young and older adults (Halamish et al., 2011; Hertzog et al., 2010; Hines et al., 2009). In contrast, age-related declines have been reported using a different assessment of metamemory (episodic memory “feeling-of-knowing”; Souchay et al., 2000).

Subjective experience of memory and imagination varies substantially even among young people (D’Argembeau & Var der Linden, 2005). Moreover, ageing amplifies individual differences in cognitive abilities (Rabbitt, 2019). Therefore, ageing research should consider not only group-level differences between young and older adults, but also within-group individual differences. We explored the relationship between memory performance and subjective memory experience across the lifespan using clustering analyses. We identified three clusters in the older adult group: one relied more heavily on conceptual information to inform their subjective vividness judgement, one more on perceptual detail information, and the third relied on both conceptual and perceptual detail information. This is very similar to the findings of Cooper and Ritchey (2022) in young adults. In our young adult group, we identified four clusters: one cluster showed a stronger relationship between vividness ratings and conceptual gist memory, the second cluster with perceptual detail of object elements, the third had a very limited relationship with memory attributes overall, and the fourth cluster showed a strong relationship between vividness ratings and all six memory attributes. An interesting avenue for future research would be to investigate whether these individual differences in behaviour can be identified at a neural level. Self-reported individual differences have been related to differences in medial temporal lobe (MTL) connectivity (Sheldon et al., 2016). The posterior brain regions underlying visual-perceptual processing have been associated with the scores on the SAM episodic subscale, and the inferior and middle prefrontal cortical regions with the SAM semantic subscale (Sheldon et al., 2016). Given our findings, individual variability in the relationship between episodic memory performance and its phenomenology should be taken into consideration when investigating the neural underpinnings of memory vividness.

In our procedure, the conceptual gist-level questions were always presented first, followed by subjective ratings and then the perceptual detail questions. It is possible that both age groups anchored their subjective ratings on their performance on the gist memory question, and this may explain the correspondence between the subjective measures and gist memory performance. However, this explanation is not supported by the findings from the first experiment conducted by Cooper and Ritchey (2022), where participants were asked to rate their memory’s vividness first, before being tested on conceptual gist or perceptual detail memory. In this experiment, we found the same relationship between conceptual gist memory performance and vividness ratings. Nevertheless, collecting vividness ratings before testing memory resulted in generally low vividness ratings, whereas if conceptual gist memory was tested first, there was a larger range in vividness ratings and they were also generally higher.

Another aspect of our design was the collection of both gist and detail memory measurements for every trial, which permitted within-subjects analyses of how these measures are mutually related to vividness. The order of the memory tests was fixed, with gist-level questions always preceding detail-level questions. This was necessary because perceptual detail questions were always tested on the correct conceptual-level item. For example, if “Brad Pitt” was the person element within an event, then the perceptual detail question always showed two similar photos of Brad Pitt, irrespective of whether the participant correctly identified Brad Pitt as being the person present in the event. Consequently, it would not be possible to test gist memory after revealing the correct answer in the perceptual detail memory test. Future research could incorporate testing the conceptual gist memory and perceptual detail memory on separate events, or test gist memory across all trials before testing perceptual detail memory afterwards, to overcome this limitation. Overall, the current study does not provide evidence for the overestimation of vividness and confidence amongst older adults, and demonstrates that healthy older adults are capable of monitoring their subjective memory experience effectively when tested on a recognition memory task.

In conclusion, we show that both older and younger adults show a close correspondence between their ability to recognise gist-level conceptual information about events and their subjective ratings of how well they can remember these events. By contrast, the ability to recognise more fine-grained perceptual features of events is largely unrelated to the subjective experience of remembering, across both age groups. Unlike several previous studies using different memoranda and procedures, we neither observed increased memory vividness ratings in the older adults, nor a relative preservation of gist-level memory performance. Overall, this study provides evidence suggesting that the relationship between memory performance and subjective memory experience remains mostly stable across the lifespan. Nevertheless, the links between objective measures of memory and the phenomenological experience of remembering are only beginning to be understood and provide an exciting new direction for understanding the multifaceted ability that recollecting the past is.

Contributed to conception and design: M.R., K.M.M., C.M.B., F.D.L.

Contributed to acquisition of data: K.M.M.

Contributed to analysis and interpretation of data: K.M.M., S.C.B., C.M.B., M.R.

Drafted and/or revised the article: K.M.M., C.M.B., F.D.L.

Approved the submitted version for publication: K.M.M., C.M.B., F.D.L, S.C.B., M.R.

This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme, grant number: 819526 to C. M. B., and from a Sussex Neuroscience PhD scholarship to K.M.M. Additional support was provided by National Science Foundation Grant BCS-2047415 awarded to M.R.

No competing interests exist for all authors of this submission.

The hypotheses, design and the analysis plan of this study were preregistered at https://osf.io/5vcpt. The deidentified data and analyses scripts can be found at Open Science Framework (OSF): https://osf.io/ceq86. The stimuli used can be made available upon request.