Vividness of imagery and affective response to episodic memories and episodic future thoughts: a systematic review and meta-analysis

ABSTRACT

Recalling personal past events and imagining personal future events are closely linked, yet also show differences. It has been claimed that episodic future thinking produces stronger intensity of in-the-moment affect than does recalling episodic memories [Schubert, T., Eloo, R., Scharfen, J., & Morina, N. (2020). How imagining personal future scenarios influences affect: Systematic review and meta-analysis. Clinical Psychology Review, 75, 101811. https://doi.org/10.1016/j.cpr.2019.101811]. In contrast, the literature indicates that memories are experienced more vividly than are episodic future thoughts, a quality that would be expected to produce a stronger rather than a weaker affective response. In this systematic review and meta-analysis, we examined (a) the intensity of affect, (b) the vividness and (c) the valence of emotion experienced in response to remembering personal past events compared to imagining personal future events. Sixteen studies with a combined sample of 1735 met criteria for inclusion. Remembered past events were experienced more vividly than imagined future events but there was no difference between the two types of representations on emotional intensity. Imagined future events were associated with more positive emotion than memories. Future research could examine factors responsible for the equivalent strength of emotional response in memories and future-thinking despite their differences in vividness.

KEYWORDS:

• Vividness
• anticipatory affect
• episodic memory
• episodic future-thinking

Affective experience is not confined to reactions to events as they occur in real time. Recall of past events has long been seen as contributing to a person’s current emotional life and, more recently, thoughts about the future have been recognised as a source of how someone feels in the here-and-now, a phenomenon labelled anticipatory affect (see MacLeod, 2017). The current review aimed to assess whether recalling past events and imagining future events differ in how much in-the-moment affect they produce and, therefore, how much they may respectively contribute to a person’s ongoing well-being. A second aim of the review was to see whether any affect-eliciting differences between episodic memories and episodic future thoughts were paralleled by differences in vividness of the respective representations, given the generally held view that vividness of imagery is an important determinant of affective experience (Holmes & Mathews, 2010). An apparent contradiction in the literature is that whereas episodic memories are, unsurprisingly, reported as being experienced more vividly than future episodic thoughts, it has been claimed that future episodic thoughts produce stronger affect (Schubert et al., 2020). The present study will review the evidence for both of those claims.

Growing evidence from neuropsychological, cognitive and behavioural studies has highlighted striking similarities between memory and future thinking, leading some to argue that the ability to imagine future personal events is in fact dependent on episodic memory (e.g., Schacter & Addis, 2020). Indeed, neuroimaging studies have shown that common neural substrates are activated when people remember the past and imagine the future (Thakral et al., 2017). How well people are able to recall specific past events is correlated with their ability to imagine specific future events (Williams et al., 1996). Research has also shown similar reductions in the specificity with which past and future events are described among a range of populations characterised by varying levels of cognitive impairment or affective dysfunction (see Hallford et al., 2018 for a review). These converging lines of evidence indicating symmetry between memory and future thinking have led some investigators to propose that they may rely on the same component processes. For example, the constructive episodic simulation hypothesis (Schacter & Addis, 2007) suggests that the episodic memory system plays an integral role in generating episodic future thoughts by extracting and flexibly recombining pieces of stored information from past episodes into mental stimulations of novel scenarios.

Despite convincing evidence for the overlap between remembering the past and imagining the future, a number of important differences have also emerged. One such difference is that, alongside common patterns of neural activity, neuroimaging studies have revealed distinct neural substrates associated with remembering and thinking about the future. For instance, imagining future events appears to recruit a wider neural network than recollecting past events, with greater neural activity observed in the right hippocampus and frontopolar regions (Addis et al., 2007; Benoit & Schacter, 2015), whereas areas involved in visuospatial processing, such as the parahippocampal cortex and posterior visual cortex, are preferentially engaged during remembering than imagining (Addis et al., 2009; Gilmore et al., 2014). Schacter and Addis (2020) have argued that greater activation during future relative to past events likely reflects the more intensive cognitive processing involved in generating novel events that have not occurred in the past.

Behavioural studies have also shown substantial differences in the self-reported phenomenological characteristics of episodic memories and episodic future thoughts. Memories are consistently rated as being more vivid and higher in sensory-perceptual details than are imagined future episodes (e.g., Berntsen & Bohn, 2010). Studies that have employed the Autobiographical Interview procedure (Levine et al., 2002), which differentiates between internal/episodic details (e.g., event details, place, time, emotion, thoughts) and external details (e.g., general knowledge, repetitions, meta-cognitive statements) in event descriptions, have shown that future events contain fewer episodic details than past events (Addis et al., 2008; De Brigard & Giovanello, 2012). Future events are also typically rated as more difficult to generate than memories (D'Argembeau and Van der Linden, 2004; McDonough & Gallo, 2010). Interestingly, despite being more general and less detailed, imagined future events tend to be rated as more positive, more personally significant and more central to one’s identity and life story than remembered events (Berntsen & Bohn, 2010; Rasmussen & Berntsen, 2013). It has been suggested that this positivity for future events might be explained by important functional differences, such as that imagining the future plays a role in regulating emotions, maintaining a positive self-image and encouraging the pursuit of new goals, whereas remembering the past is limited by the reality of past events and thus facilitates problem-solving and learning from past mistakes (Rasmussen & Berntsen, 2013).

One area where the evidence is less clear from the existing literature is the emotional impact of thinking about the future relative to remembering the past. Studies that have assessed participants’ subjective emotional response to recalling memories and simulating future events have found mixed results. Some studies indicate that memories were experienced as more emotionally intense than imagined future events (e.g., Rasmussen & Berntsen, 2014); some demonstrated a stronger affective response to simulating future episodes (e.g., De Brigard & Giovanello, 2012), while others have found equivalency (e.g., Özbek et al., 2017). The fact that memories do not clearly produce a stronger affective response is somewhat surprising given that remembered past events are consistently rated as being more vivid and detailed and there is evidence to suggest that vivid mental representations evoke stronger emotional responses (D’Argembeau & Van der Linden, 2006; Holmes & Mathews, 2010). For example, in the mental imagery literature, it has been shown that visual imagery elicits a stronger affective reaction compared to verbal processing of the same material and this association has been found for both positively and negatively valenced material (Mathews et al., 2013). Moreover, in the clinical sphere, it is widely recognised that vivid and intrusive mental images are prevalent across a range of psychological disorders, including post-traumatic stress disorder, social phobia and obsessive-compulsive disorder (Hackmann & Holmes, 2004; Holmes & Mathews, 2010), thus indicating that vivid mental images may play an important role in emotional distress. In light of this evidence, one might expect to observe a stronger affective response to recalling memories compared to imagining future events.

A recent meta-analysis by Schubert et al. (2020) sought to clarify these mixed findings by comparing the affective response to future imagination versus memory, concluding that “imagining the future evokes stronger affect than remembering the past” (Schubert et al., 2020, p. 1). However, this conclusion is undermined by the fact that the review extracted and utilised measures of valence (how strongly positive or negative the feelings experienced were when thinking about the event) from included studies, rather than measures of intensity (the strength of feeling, irrespective of valence) (N. Morina, personal communication, 29th August, 2022). Using a valence rating scale as a simple linear scale to indicate strength of feeling means that a score of −3 (extremely negative) would be taken as indicating less strength of feeling experienced than would a score of −1 (mildly negative). Of course, this is the reverse of how it should be if strength of affect is the variable of interest. The review by Schubert et al. (2020) appears to be more about how positively or negatively someone reacts emotionally to thinking about a memory or future event, rather than simply the strength of their feeling per se.

It remains unclear whether people experience more affect when they imagine the future compared to remembering the past. The data on vividness of representations seems clearer. To compare vividness and affective impact of remembering versus imagining the future, we selected studies that measured both vividness and intensity (strength) of affective response for episodic memories and episodic future thoughts. We only included studies where the measure of intensity was unambiguously about the feelings elicited in the moment. For example, studies where the intensity measure could be interpreted as asking for an appraisal of the event (e.g., The recalled/imagined event is emotionally intense) rather than describing the in-the-moment experience (e.g., the emotions I have when I imagine/recall the event are intense) were excluded, as were studies that did not fully describe the intensity measure in enough detail to rule out ambiguity. Finally, in light of the findings of Schubert et al. (2020) that appear more relevant to valence, we also had an additional, secondary aim of comparing past and future representations on how positively or negatively participants rated their emotion in response to recalling/simulating events.

Method

The review was conducted in accordance with the PRISMA statement guidelines for the reporting of systematic reviews and meta-analyses (Liberati et al., 2009) and the protocol prospectively registered with the PROSPERO database (PROSPERO ID: CRD42021231054).

Search strategy

The following search terms were applied to title, abstract and keywords in three electronic databases (PsycINFO, PubMed and Web of Science): (“episodic future thinking” OR “future thinking” OR “future-directed thinking” OR prospection OR “imagining the future” OR “episodic simulation” OR “mental time travel”) AND (“episodic memor*” OR “autobiographical memor*” OR “remembering the past” OR retrospection). No restrictions for language and year of publication were applied.

After duplicates were removed, the titles and abstracts of all studies (N = 1775) were screened by the second author based on the inclusion and exclusion criteria, and articles considered to be relevant were retrieved in full-text and assessed for eligibility. A second reviewer screened 296 studies (18%) of the titles and abstracts retrieved from the initial search, along with full texts of potentially eligible articles. Where there was a disagreement (4 studies), the two reviewers discussed the eligibility of the article until an agreement was reached. An additional hand-search was independently conducted by the second author, who checked the reference lists of all eligible studies. See Figure 1 for the results of the search strategy.

Figure 1. PRISMA flow diagram of study selection process.

Inclusion and exclusion criteria

Studies were included in the review if they met the following criteria:

• Adult participants (≥ 18 years of age);

• Participants were explicitly instructed to recall personal past events and to imagine plausible events in their personal futures;

• At least one measure of vividness¹ and one measure of emotional intensity² was employed. A secondary outcome of interest was emotional valence³, although it was not necessary for studies to include a measure of valence to be included in the review;

• Included statistical analyses comparing past and future conditions on the outcomes of interest;

• Published in a peer-reviewed journal before 11th September, 2022, with full text available in English.

Exclusion criteria were as follows:

• Participants characterised by primary physical health problems (e.g., cancer), neurological disorders, developmental disorders, brain injuries and substance misuse/addictions;

• Involuntary episodic thinking tasks (e.g., experience sampling methods and mind-wandering);

• Episodic thinking tasks about implausible events or events that do not involve oneself (e.g., events relating to friends, family or fictional characters);

• Episodic thinking task embedded within another task or preceded by an experimental manipulation.

Data extraction and quality assessment

The following study characteristics were extracted: authors, year of publication, country, study design, population, age, sample size, the proportion of females, type of intervention, number of cues, cue valence, temporal distance of cues, response format, response time given and outcomes. Means and standard deviations for the outcomes of interest were also extracted for each study to calculate effect sizes.

The Downs and Black Checklist for Measuring Quality (Downs & Black, 1998) was used to assess methodological quality of included studies. Each of the 18 items is rated on a two-point scale (0 = no or unable to determine, 1 = yes), with the exception of one item which is rated on a three-point scale (0 = no or unable to determine, 1 = partially, 2 = yes). The maximum score that can be awarded is 19. No studies were excluded from the review based on quality assessment due to the risk of over-exclusion, which could limit the validity of results (Meline, 2006).

Data analysis

All analyses were performed in Meta-Essentials (Suurmond et al., 2017), specifically using the Excel workbook designed for estimating differences between two dependent groups with a continuous outcome, since all of the included studies had a within-subjects or mixed design. Due to expected heterogeneity between studies, random-effects models were used to estimate all effects, as this assumes that the true effect will vary between studies (Borenstein et al., 2009). Standardised mean differences were calculated using Hedges’ g, which provides a corrected estimate to account for the small upward bias that occurs when sample sizes are small (Hedges, 1981). Hedge’s g and 95% confidence intervals (CI) were estimated for each study from reported descriptive statistics, including means (M), standard deviations (SD) and sample size. Because correlations between memory and episodic future thinking scores were not reported any of the included studies, a correlation coefficient of r = 0.50 was assumed in order to account for dependencies between memory and future conditions due to within-subjects designs. The results were subsequently re-examined using a sensitivity analysis, by imputing a higher (r = 0.70) and a lower (r = 0.30) correlation; however, this had little impact on the effect sizes and therefore an assumed correlation of r = 0.50 is reported for all analyses. Studies with multiple conditions (N = 6), for example, where future thinking and memory were measured using cue words and important event cues, were collapsed into one condition using Borenstein et al.’s (2009) recommended method for estimating a summary effect across outcomes. Studies with more than one group (N = 8) had the groups combined into a single group using an online tool (StatsToDo, 2021). However, for two studies that had two conditions (Gehrt et al., 2020 – everyday events and anxiety-related events; Ikier et al., 2022 – real-self and counterpart-self conditions), the results are only reported for the everyday events condition and the real-self condition because these conditions were most comparable with the other studies. Effect sizes were interpreted as small (0.2), medium (0.5) and large (0.8) (Cohen, 1992). For all studies, effect sizes were coded in the direction of future thinking, so that positive effect sizes indicate outcomes in favour of future thinking and negative effect sizes indicate outcomes in favour of memory.

Heterogeneity between studies was assessed using Cochran’s Q and the I² statistic. Cochran’s Q tests whether variation between study effect sizes is larger than would be expected to occur by chance, with a p value of < .05 indicating that heterogeneity is present. I² shows the extent of variability among effect sizes, with I² values of 25%, 50% and 75% considered to indicate low, moderate and high levels of heterogeneity (Higgins et al., 2003). To explore potential sources of heterogeneity, subgroup analyses were planned for: (1) age groups; (2) response format; (3) cue type; (4) cue valence; (5) temporal distance and (6) study quality. This was assessed using random-effects models with pooled effect sizes for each subgroup and a between-group Q test (Q_between), with a statistically significant Q_between statistic indicating a significant difference between groups. Heterogeneity within each subgroup was also quantified using the I² statistic.

Publication bias was assessed by visually inspecting funnel plots, whereby an asymmetrical distribution of points in the scatter plot is indicative of publication bias. Furthermore, Rosenthal’s (1979) failsafe N was calculated for each meta-analysis to account for the file drawer problem (i.e., the tendency for significant positive results to be published over null results). A failsafe N is an estimate of the number of additional non-significant unpublished studies required to reduce the overall effect size to non-significance. The trim-and-fill method (Duval & Tweedie, 2000) can be used to estimate an adjusted effect size that accounts for funnel plot asymmetry and hypothetically missing studies; however, this procedure was not used in the present study due to the large amount of between-study heterogeneity found in the models, which can undermine the validity of this method (Peters et al., 2007). Therefore, sensitivity analyses were conducted to examine the effect of manually removing outlier studies from the models.

Results

Study characteristics

Figure 1 illustrates the study selection process. A summary of the study characteristics of the sixteen eligible studies (1735 participants, 66.8% female, mean age 30.05 years) is presented in Table 1. All studies were published between 2010 and 2022.

Table 1. Study characteristics.

Author (year), country	Population	N	Mean age (SD)	Female %	Type of intervention	Number of cues	Cue valence	Temporal distance of cues	Response format and time given	Outcomes
Akdere and Ikier (2021)	Turkish younger (Y; n = 30), middle-aged (MA; n = 30) and older adults (OA; n = 30)	90	Y: 29.27 (4.57) MA: 48.39 (6.04) OA: 68.20 (6.21)	Y: 60.0 MA: 66.67 OA: 60.0	Describe an important past event, an important future event (plus a counterfactual event and an additional future event)	4	Not specified	Not specified	Written, no time limit	Vividness, emotional intensity, emotional valence
Anderson and Evans (2015), UK	University students: dysphoric group (n = 31) vs. non-dysphoric group (n = 28)	59	Dysphoric:22.48 (3.89) Non-dysphoric:21.61 (3.69)	Dysphoric: 70.97 Non-dysphoric: 60.71	Describe 4 past events and 4 future events	8	Not specified	2 past month, 2 past year, 2 next month and 2 next year	Written, no time limit	Vividness, emotional intensity, emotional valence
Berntsen and Bohn (2010), Denmark	University students	122	26.53, range 21 - 47	84.43	Describe 5 past and 5 future events in response to cue words and 5 past and 5 future important events	20	Not specified	Not specified	Written, no time limit	Vividness, emotional intensity, emotional valence
D’Argembeau et al. (2010), Belgium	Young adults – native French speakers	100	24.30, range: 18.0–30.0	49.0	Describe the last/next time you meet with a friend and something you did/will do during your last/next vacation	4	Not specified	Not specified	Verbal, no time limit	Visual details, feeling of emotion
De Brigard et al. (2016), USA	Young adults (YA; n = 30) and older adults (OA; n = 30)	60	YA: 22.35 (3.27) OA: 69.32 (4.85)	YA: 63.33 OA: 60.0	Describe 2 memories, 2 unlikely future and 2 likely future events (plus 4 counterfactual)	10	5 positive, 5 negative	Not specified	Verbal, up to 3 min	Vividness, emotional intensity, emotional valence
Ernst and D’Argembeau (2017), Belgium	Native French speakers, mostly university students	92	28.54 (10.39)	48.91	Describe 10 past events and 10 future events in response to cue words	20	Not specified	Not specified	Verbal, no time limit	Sensory details, emotional intensity, emotional valence
Gehrt et al. (2020), Denmark	Patients with health anxiety (HA; n = 32), patients with obsessive compulsive disorder (OCD; n = 33) and controls (n = 32)	97	HA: 40.03 (9.58) OCD: 30.82 (10.10) Controls: 40.0 (11.44)	HA: 62.50 OCD: 84.85 Controls: 62.50	Describe an everyday past event and an everday future event and a past event and future event in which you felt/would feel anxious or scared.	4	2 negative, 2 not specified	2 last year, 2 next year	Written, no time limit	Vividness, emotional intensity, emotional valence
Ikier et al. (2022)	University students	108	22.57 (1.42)	53.70	Describe an important past event, an important future event (plus an important counterfactual event)	3	Not specified	Not specified	Verbal, no time limit	Visual imagery, emotional intensity
O'Toole and Berntsen (2020)	Pregnant women (n = 59) and control women (n = 59)	118	Pregnant: 28.3 (4.8) Controls: 24.3 (2.2)	100	Describe 4 past and 4 future events in response to cue words at Time 1 (3 months before due date) and at Time 2 (3 months after expected birth)	8	Not specified	Not specified	Written, no time limit	Vividness, emotional intensity, emotional valence
Ottsen and Berntsen (2015), Denmark	Middle Eastern (ME) sample from 2 universities in Qatar (n = 124) and Scandinavian (SCA) sample from a university in Denmark (n = 128)	252	ME: 22.7 (6.2) SCA: 22.9 (3.7)	ME: 50.0 SCA: 57.03	Describe 3 past events that have been “most important in the story of your life” and 3 important events that will occur in your personal future.	6	Not specified	Not specified	Written, no time limit (4 lines provided for written description)	Vividness, emotional intensity, emotional valence
Özbek et al. (2017), Turkey	University students	69	21.09 (2.20)	57.97	Describe 3 memories of important past events and 3 important imagined future events (plus 3 counterfactual events)	9	Not specified	2 past month, 2 past year, 2 past 5 or more years, 1 next month, 1 next year and 1 next 5 or more years	Written, no time limit	Vividness, emotional Intensity, emotional valence
Özbek et al. (2020), Turkey (Study 1) USA (Study 2)	Study 1: Turkish young adults (YA; n = 72) and older adults (OA; n = 41) Study 2: American young adults (YA; n = 32) and older adults (OA; n = 34)	Study 1: 113 Study 2: 66	Study 1: YA: 20.15 (1.94) OA: 65.10 (5.62) Study 2: YA: 26.50 (3.14) OA: 63.88 (4.44)	Study 1: YA: 65.28 OA:51.22 Study 2: YA:53.13 OA: 58.82	Describe 3 memories of important past events and 3 important imagined future events (plus 3 counterfactual events)	9	Not specified	2 past month, 2 past year, 2 past 5 or more years, 1 next month, 1 next year and 1 next 5 or more years	Written, no time limit	Sensory imagery^a, emotional intensity, emotional valence
Rasmussen and Berntsen (2013), Denmark	University students	158	24.44 (4.13)	83.54	Describe 2 past events and 2 future events	4	2 positive, 2 negative	Not specified	Written, no time limit	Sensory imagery, emotional intensity, emotional valence
Rasmussen and Berntsen (2014), Denmark	University students	151	23.96 (3.87)	84.77	Describe 4 memories and 4 future events in response to high imageable and low imageable cue words	8	Not specified	Not specified	Written, 3 min	Vividness, emotional intensity, emotional valence
Rubin (2014), USA – Experiment 1	University students	76	19.02 (1.06)	60.53	Describe 6 past events that “impact you a lot” and 6 future events that “would impact you a lot” (plus 3 neutral past events to anchor participants’ responses)	15	3 neutral, 6 positive, 6 negative	6 past year, 6 next year	Verbal, no time limit	Visual details, emotional intensity, emotional valence

^a Ratings of vividness, (p)re-living, visual imagery and auditory imagery averaged into sensory-imagery factor.

Study design

Eight studies employed cross-sectional within-subjects designs, with the majority of studies using student samples (N = 7) and one study employed a sample of young adults. The remaining eight studies used cross-sectional mixed designs, of which four studies compared younger adult and older adult populations (Akdere & Ikier, 2021; De Brigard et al., 2016; Özbek et al., 2020 – Studies 1 & 2). One study compared dysphoric and non-dysphoric participants from a non-clinical student population (Anderson & Evans, 2015), one study compared Middle Eastern and Scandinavian student populations (Ottsen & Berntsen, 2015), one study compared pregnant and non-pregnant women (O'Toole & Berntsen, 2020) and another study included a 3-way comparison of patients with severe health anxiety, patients with obsessive-compulsive disorder (OCD) and a control group (Gehrt et al., 2020).

Sample characteristics

All studies appeared to use non-probability sampling methods, with three studies reporting that participants were recruited as part of an undergraduate teaching course (Berntsen & Bohn, 2010; Rasmussen & Berntsen, 2013, 2014) and two studies recruited though student participant pools or sign-up sheets at universities (Özbek et al., 2017, 2020 – Study 1). One study recruited through TurkPrime Panels using Amazon’s Mechanical Turk workers (Özbek et al., 2020 – Study 2); one study recruited through clinics at a university hospital (Gehrt et al., 2020) and another study recruited from two sites providing childbirth classes, a midwife consultation clinic and a database of volunteers (O'Toole & Berntsen, 2020). The remaining eight studies did not clearly state how participants were identified and recruited.

Memory and future thinking tasks

There was considerable variation across studies in the assessment of episodic memory and episodic future thinking. Four studies presented a series of cue words and elicited descriptions of past events and future events associated with each word (Berntsen & Bohn, 2010; Ernst & D’Argembeau, 2017; O'Toole & Berntsen, 2020; Rasmussen & Berntsen, 2014). Seven studies asked participants to recall (imagine) personally important past (future) events, three of which also included cues of varying temporal distances (Özbek et al., 2017, 2020 – Studies 1 & 2). Two studies employed valence-based cues (Rasmussen & Berntsen, 2013; Rubin, 2014), and Gehrt et al. (2020) instructed participants to identify anxiety-eliciting everyday events. D’Argembeau et al. (2010) cued specific common scenarios and Anderson and Evans (2015) used a minimal cueing strategy of temporal distance only. Finally, De Brigard et al. (2016) employed cues that varied in terms of both valence and perceived likelihood that had previously been established. The studies also differed with respect to response format (11 written versus 5 verbal) and whether a time limit was imposed (two studies).

Measurement of outcomes of interest

The measurement of current emotional intensity was relatively uniform across studies. All of the studies used 7-point Likert scales to measure emotional intensity (e.g., “The feelings I experience as I recall/imagine the event are intense: 1 = not at all, 7 = to a very high degree”). Of the fourteen studies that measured emotional valence, all studies employed 7-point scales, with the majority of studies (N = 13) assessing valence on a scale from −3 (“extremely/very negative”) to +3 (“extremely/very positive”) and one study (De Brigard et al., 2016) used a scale from 1 (“negative”) to 7 (“positive”).

There was greater variation in the measurement of vividness across studies. All studies measured vividness on a 7-point scale. Ten studies specifically used the term “vivid(ness)” in the item descriptor or response options (e.g., “This memory/ imagined event is vivid: 1 = not at all, 7 = to a very high degree”). Ernst and D’Argembeau (2017) measured sensory details (“I can see, hear or perceive what happened/will happen: 1 = not at all, 7 = a lot”), while Rasmussen and Berntsen (2013) measured sensory imagery (“While remembering/imagining this event, I can see and hear it in my mind: 1 = not at all, 7 = to a very high degree”) and three studies (D’Argembeau et al., 2010; Ikier et al., 2022; Rubin, 2014) measured visual details/imagery (e.g., “While remembering/imagining the event, I can see it in my mind: 1 = not at all, 7 = as if it were happening now”).

Methodological quality

Total scores on methodological quality ranged from 10 to 14, with a median score of 11. Areas of strength across studies included clear description of aims and hypotheses, primary outcomes, sample characteristics and main findings, along with appropriate statistical analyses, reporting of estimates of random variability for main outcomes and results not based on data dredging. Common weaknesses were a lack of information about the representativeness of study samples, lack of consideration of possible confounders, not reporting withdrawals and drop-outs and not reporting a priori power analyses.

Meta-analysis findings

Vividness

The meta-analysis revealed a medium effect size (g = −0.61, 95% CI −0.81 to −0.41, p < .001; see Figure 2): participants experienced episodic memories as significantly more vivid than episodic future simulations. The heterogeneity statistics indicated a large amount of heterogeneity between studies (Q = 177.60, p < .001, I² = 91.55%). Visual inspection of the funnel plot highlighted some asymmetry, with three studies to the left of the plot (Anderson & Evans, 2015; Ernst & D’Argembeau, 2017; Rasmussen & Berntsen, 2013) and three to the right of the plot (Akdere & Ikier, 2021; D’Argembeau et al., 2010; Rasmussen & Berntsen, 2014). However, Rosenthal’s failsafe N indicated that a minimum of 248 studies with null or weak effects would need to exist before the combined effect size became statistically non-significant, thus suggesting that this model is likely to be robust to the effect of publication bias. Nevertheless, a sensitivity analysis was conducted to test the effect of removing these six studies from the main model, which only marginally reduced the effect size to g = −0.55 (95% CI −0.62 to −0.49, p < .001) while significantly decreasing the level of heterogeneity (Q = 6.58, p = .681, I² = 0%).

Figure 2. Forest plot for differences between memory and future thinking. Note. Effect sizes in positive direction favour future thinking and effect sizes in negative direction favour memory.

Emotional intensity

The random-effects meta-analysis found a small effect size in favour of memory, though the overall effect was non-significant (g = −0.10, 95% CI −0.25–0.05, p = .156; see Figure 3). The heterogeneity between studies was high (Q = 102.34, p < .001, I² = 85.34%). Inspection of the funnel plot identified three influential studies (Anderson & Evans, 2015; Ikier et al., 2022; Rubin, 2014) to the left of the plot with relatively large effect sizes and confidence intervals. However, the effect sizes were still significantly heterogeneous after removing these three studies from the model (Q = 46.82, p < .001, I² = 74.37%), while the overall effect of the reduced model remained non-significant (g = −0.05, 95% CI −0.17–0.08, p = .435). The forest plot for the full model demonstrates that a consistent pattern was not observed across studies. Five studies (Anderson & Evans, 2015; Berntsen & Bohn, 2010; D’Argembeau et al., 2010; Ikier et al., 2022; Rubin, 2014) showed a significant effect in favour of memory, two studies (Akdere & Ikier, 2021; Rasmussen & Berntsen, 2014) showed a significant effect in favour of future thinking, while the remaining nine studies had non-significant effects. Furthermore, Rosenthal’s failsafe N for the overall effect size was 0, indicating that publication bias is a concern for this model.

Figure 3. Forest plot for differences between memory and future thinking on emotional intensity. Note. Effect sizes in positive direction favour future thinking and effect sizes in negative direction favour memory.

Emotional valence

Fourteen studies were included in the meta-analysis and the combined studies had a significant medium effect size of g = 0.56 (95% CI 0.27–0.86, p < .001; see Figure 4). This indicated that emotional valence was significantly higher (i.e., more positive) for future thoughts relative to memories. The heterogeneity statistics indicated that there was significant variation around the estimated mean (Q = 232.06, p = <.001) and that a very high proportion of the observed variance was due to differences between studies (I² = 94.40%). Inspection of the funnel plot highlighted one outlier to the right of the plot (Özbek et al., 2017). The removal of this study from the model made some difference to the effect size (g = 0.48, 95% CI 0.23–0.74, p < .001), although heterogeneity was only trivially different (I² = 93.66%). Rosenthal’s failsafe N for the full model was 101, thus estimating that 101 unpublished or future studies with null effects would need to be included to make the results of this meta-analysis non-significant. Hence, it appears that the mean effect size is not heavily influenced by publication bias.

Figure 4. Forest plot for differences between memory and future thinking on emotional valence. Note. Effect sizes in positive direction favour future thinking and effect sizes in negative direction favour memory.

Subgroup analyses

Planned subgroup analyses were conducted to assess for differences in effect sizes based on study characteristics that may have had a moderating effect. The studies were grouped based on age groups, response format, cue type, temporal distance of events and study quality.

With respect to vividness, the results indicate that the effect sizes did not significantly vary according to any of the subgroup variables specified above. Greater homogeneity between effect sizes was observed among studies that employed important event cues (I²= 39.64%); however, moderate to high levels of heterogeneity were evident in the remaining subgroups.

For emotional intensity, the significant Q_between statistic for type of cue suggests that the true mean effect varies depending on the type of cueing method used to elicit memories and future thoughts (Q_between(3) = 14.54, p = .002). The results indicate that studies which employed simply cue words had a significant small overall effect size favouring future thinking (g = 0.17, 95% CI 0.13–0.20) with high homogeneity between studies (I² = 0%). On the other hand, studies that required participants to produce important events (g = −0.08, 95% CI −0.32–0.16), employed valence-based cues (g = −0.24, 95% CI −0.53–0.04) or other cueing methods (g = −0.22, 95% CI −0.49–0.05) found small mean effect sizes in favour of memory, albeit non-significant. Nevertheless, these findings should be interpreted with caution, given that the subgroups were comprised of a small number of studies and they were highly heterogeneous in their effect sizes. There were no significant differences between the remaining subgroups.

The subgroup analyses for emotional valence revealed significant between group differences when studies were grouped according to type of cue (Q_between(3) = 63.15, p < .001) and cue valence (Q_between(1) = 47.19, p < .001). Studies that used important event cues (g = 0.94, 95% CI 0.52–1.36) and cue words (g = 0.64, 95% CI 0.52–0.75) had significant large and moderate effect sizes respectively, indicating that important future events and word-cued future events evoked more positive emotions than memories. Studies which employed “other” cueing strategies similarly found a moderate effect size in favour of future thinking (g = 0.45, 95% CI −0.18–1.09), though it did not reach statistical significance, whereas studies that elicited events using valence-based cues found a small overall effect size in favour of memory (g = −0.04, 95% CI −0.18–0.09), though again this effect was non-significant. Nonetheless, the results of this subgroup analysis should be treated with caution as there was high heterogeneity in two of the groups and the number of studies in each subgroup is likely too small to draw meaningful conclusions.

Furthermore, a significantly larger overall effect size was observed among studies that did not specify cue valence (g = 0.79, 95% CI 0.51–1.04) compared with studies which used positively and negatively valenced cues (g = −0.02, 95% CI −0.13–0.09), indicating that neutral/unspecified cues elicited more positively valenced emotions for future events compared to positively and negatively valenced cues. Thus, this suggests that when cues are more specified, people show little difference on emotional valence; however, when the cueing methodology provides more freedom, they appear to produce future events that are more positive. Heterogeneity in the positive and negative valence cue group was substantially lower (I²= 19.55%) compared to the not specified group (I²= 88.42%), indicating that differentiating studies in terms of cue valence partially explained the large amount of heterogeneity. Despite this, the positive and negative cue valence group only consisted of four studies, which may be too small to draw definite conclusions. The remaining subgroup analyses did not highlight any further significant differences in effect sizes between groups and did not appear to account for the high levels of heterogeneity.

Finally, two of the studies were with clinical samples. To test effects without those samples, the main analyses were re-run excluding those two studies. The effects remained the same - vividness: g = −0.61 (95% CI: −0.81 to −0.41, p < .001); intensity: g = −0.10 (95% CI: −0.25–0.05, p = .237); valence: g = 0.56 (95% CI: 0.27–0.86, p < .001).

Discussion

The current study sought to systematically review and meta-analyse findings from studies that have directly compared the subjective experience of episodic memory to episodic future thinking, specifically in terms of vividness and emotional intensity, with emotional valence as a secondary outcome. It has been cited in the literature that future episodic thoughts produce more affect than memories (e.g., Schubert et al., 2020), which seems inconsistent with claims that memories are more vivid and higher in sensory detail (e.g., Berntsen & Bohn, 2010; Ernst & D’Argembeau, 2017), given the link between vividness and affective response (e.g., Holmes & Mathews, 2010). A systematic search identified 16 studies with a total of 1735 participants that met criteria for inclusion for the primary aims of the study, 14 of which also provided valence data.

Memories of past events were experienced as more vivid than imagined future events. There was significant variation between study effect sizes, accounted for by a minority of studies; the removal of which had little impact on the effect size. Furthermore, there was little indication of publication bias affecting the overall effect size. Subgroup analyses did not highlight any significant between-group differences in effect sizes based on sample age and methodological differences. Nevertheless, most subgroups in these analyses were comprised of a small number of studies, which may preclude meaningful interpretations due to limited statistical power to detect effects.

In contrast, the second meta-analysis found no difference between episodic memory and episodic future-thinking on emotional intensity. However, again this was not a consistent result, as five studies indicated that memories were experienced as significantly more emotionally intense than future events, two studies found a significant effect in favour of future imagination, and the remaining nine studies had non-significant effects. The removal of three outlier studies made little difference to the overall effect size or the level of heterogeneity. Additionally, there was some reason for concern about publication bias. Subgroup analyses indicated that the type of cueing methodology used to elicit memories and future events may impact the findings. Specifically, the results suggested that future events generated in response to cue words were experienced as more emotionally intense than memories, whereas the remaining cueing methodologies had non-significant effects. It is not entirely clear why this should be the case. Future episodes generally provide more positive affect than memories and perhaps single cue words provide more scope for an enhanced idealised future than do more constrained prompts such as a personally important event. However, this explanation is speculative and the findings should be considered tentative, as heterogeneity within the subgroups was large and the small number of studies in the group comparisons limits our ability to derive definite conclusions.

Finally, emotional valence was significantly more positive for future, compared to past, events with a medium effect size. Substantial heterogeneity was not reduced by the removal of an outlier study. Subgroup analyses indicated that the greater positivity for future events was reduced in studies that employed valence-based cues that prompted for positive and negative events, compared to studies that used other cueing methods and did not specify the valence of events. Hence, this suggests that when the valence of cues is more specified, people show little difference in the valence of emotional responding; yet, when they are given more freedom, people tend to produce future events that are more emotionally positive.

With respect to vividness and emotional valence, the results of our meta-analyses are in line with conclusions drawn in the existing literature that memories of past events are recalled with higher levels of vividness and sensory detail than are imagined future episodes and that future, as compared to past, events are experienced more positively (e.g., Berntsen & Bohn, 2010; D'Argembeau and Van der Linden, 2004; Rasmussen & Berntsen, 2013). However, the results of our subgroup analyses indicating differences in assessment methods may influence the strength of the effect suggest that particular cues might provide scaffolding for producing more vivid and positive events. Given the substantial variability in assessment methods employed in the current literature, further examination of the impact of cueing methods on the phenomenological qualities of memories and future simulations could be a key avenue for a future meta-analytic review.

The lack of difference in emotional intensity runs counter to the claim of Schubert et al. (2020), but as outlined earlier these authors used valence ratings as opposed to intensity ratings for many of their included studies. As such, their conclusions are indicative of how positively or negatively people feel when they recall/imagine events, rather than the strength of feeling evoked. In fact, the magnitude of the effect they report in favour of episodic future thought for “intensity” (g = .45) was not dissimilar to the effect we report here for valence (g = .56) and bears no relation to the effect we found when using studies’ intensity ratings (g = -.10), further reinforcing that their conclusions relate to valence not intensity.

Nevertheless, there is still an interesting dissociation between vividness and intensity, in that future episodic thoughts are experienced less vividly than memories yet produce equivalent levels of affect. This pattern of findings suggests that other variables may be at work. One candidate variable is that future events are seen as more positive and idyllic, while another candidate may be the personal significance of events, given that a number of studies have shown that future events tend to be viewed as more positive and personally important than past events (e.g., Berntsen & Bohn, 2010; Rasmussen & Berntsen, 2013). Indeed, several recent studies have found that personal importance of events was positively associated with anticipatory pleasure (i.e., pleasure experienced in the here-and-now in response to the prospect of future events) (Barsics et al., 2016; Hallford et al., 2020). Thus, it may be that the greater personal significance and/or the greater positivity of future events cancel out the lower vividness and thus, there could be two routes to affect with one being higher in memories and the other being higher in future thinking. One way to test this would be to manipulate the valence and personal significance of event cues and to assess the impact on affective responding. Further work to improve our understanding of characteristics of episodic memory and future thinking that uniquely predict affective response is needed. Identifying factors that may be amenable to intervention could have important clinical implications for enhancing positive affect and/or reducing negative affect in clinical populations that are characterised by deficits in memory and/or future thinking, such as major depression and schizophrenia (Dere et al., 2010; Hallford et al., 2018).

The findings demonstrate that episodic memories and future thoughts have equivalent affective power but that for future thinking this is linked with more positive affect. On the face of it this would seem to suggest that future-thinking is more important than memory for well-being. However, as noted by Rasmussen and Berntsen (2013) the two types of thinking may have different functions, where the positivity of thoughts about the future serve to motivate people to approach future goals, with memories providing more of a corrective reality and problem-solving role. As such, both would be important for well-being because the sustainability of a person’s well-being is likely to depend on being able to move realistically towards future positive outcomes as well as simply being able to imagine them.

To our knowledge, this is the first meta-analysis that summarises in one place the evidence for the phenomenological qualities of vividness, emotional intensity and emotional valence for past versus future episodic thinking. Our findings demonstrate that memories are experienced as significantly more vivid than imagined future episodes, while future events are judged as significantly more emotionally positive than past events. However, a consistent effect was not observed across studies for emotional intensity, pointing to factors other than vividness being important in eliciting in-the-moment affect. These findings have important theoretical and clinical implications, and indicate a need for future research to examine the emotional experience of memory versus future thinking using more detailed measures of affect and to explore the mechanisms through which emotional reactivity occurs.

The authors report there are no conflicts of interests.

*References marked with an asterisk are included in the meta-analysis.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.

Notes

1 For the purpose of this review, vividness is defined as “the quality of producing very clear pictures in the mind” (Oxford Learner’s Dictionaries, 2021). Therefore, to be included in the review, studies needed to either include the word “vivid(ness)” in the item descriptor or clearly assess the quality of producing pictures in the mind.

2 Emotional intensity is defined as intensity of emotion experienced clearly in the act of recalling or imagining the event.

3 Emotional valence is defined as the judged valence (i.e., positive vs. negative) of emotion experienced in the act of recalling or imagining the event.