Direct and generative autobiographical memory retrieval: How different are they?

Coversheet This is the accepted manuscript (post-print version) of the article. Contentwise, the accepted manuscript version is identical to the final published version, but there may be differences in typography and layout. How to cite this publication Please cite the final published version: Harris, C. B., & Berntsen, D. (2019). Direct and generative autobiographical memory retrieval: How different are they? Consciousness and Cognition, 74, 102793. https://doi.org/10.1016/j.concog.2019.102793 Publication metadata Title: Author(s): Journal: DOI/Link: Direct and generative autobiographical memory retrieval Harris, Celia B.; Berntsen, Dorthe Consciousness and Cognition https://doi.org/10.1016/j.concog.2019.102793 Document version: Accepted manuscript (post-print) General Rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognize and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. If the document is published under a Creative Commons license, this applies instead of the general rights. This coversheet template is made available by AU Library Version 2.0, December 2017 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Direct and Generative Autobiographical Memory Retrieval: How Different Are They? Celia B. Harris Department of Cognitive Science Macquarie University, Australia Dorthe Berntsen Center on Autobiographical Memory Research Aarhus University, Denmark Corresponding author: Dr. Celia B. Harris Department of Cognitive Science Macquarie University Sydney, NSW 2109 Australia Phone: + 61 2 9850 4066 Email: celia.harris@mq.edu.au 1 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Abstract Theories of autobiographical memory have emphasised effortful generative retrieval, but recent research indicates that subjectively effortless direct retrieval is common. We compared the processes of direct and generative retrieval. Sixty-five participants retrieved 24 autobiographical memories across three cue types: concrete, emotional, and personal. We recorded retrieval latency, and participants judged direct versus generative retrieval and rated memory specificity, vividness, significance, rehearsal, and emotionality. Overall, direct retrieval was common, especially for personal cues. Directly retrieved memories were recalled faster, were less likely to be specific, and were rated more significant, rehearsed, and emotional than generatively retrieved memories. The speed of both direct and generative retrieval varied similarly according to cue type, suggesting they did not involve fundamentally different cognitive processes. These findings challenge theories that assume direct retrieval bypasses constructive processes. Instead we suggest that both direct and generative retrieval involve construction that is similarly affected by cue concreteness and relevance. Keywords: autobiographical memory; retrieval processes; direct retrieval; memory qualities; involuntary autobiographical memories 2 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Direct and Generative Autobiographical Memory Retrieval: How Different Are They? Theoretical accounts of autobiographical memory retrieval often distinguish between “generative” versus “direct” retrieval (Addis, Knapp, Roberts, & Schacter, 2012; Conway & Pleydell-Pearce, 2000; Haque & Conway, 2001; Harris, O’Connor, & Sutton, 2015; Uzer, Lee, & Brown, 2012, Williams et al., 2006). Definitions and criteria for distinguishing between direct and generative retrieval vary. Generative retrieval is typically defined as involving a “top-down”, “goaldirected” (Conway & Pleydell-Pearce, 2000) “controlled and effortful” (Anderson, Dewhurst, & Dean, 2017, p. 163) search process, in which individuals search through increasingly specific autobiographical information in order to bring an event to mind. In contrast, direct retrieval is variously described as “automatic and effortless” (Uzer, Lee, & Brown, 2012, p. 1296), “immediate” (Ros, Latorre, & Serrano, 2009, p. 91), or “spontaneous and unexpected” (Conway & Pleydell-Pearce, 2000, p. 275; Williams et al., 2006, p. 363). Traditionally, theories of autobiographical remembering, while acknowledging the possibility of direct retrieval, emphasise and prioritise generative retrieval as the default process by which memories are retrieved and suggest that direct retrieval is comparatively rare; e.g. “the retrieval of autobiographical memories, although occasionally spontaneous and apparently effortless, is more usually an effortful and protracted process” (Haque & Conway, 2001, p. 529). However, there is recent evidence to suggest that subjectively effortless direct retrieval can be as common as effortful generative retrieval, at least in certain circumstances. Recent research by Uzer et al. (2012) was specifically designed to assess the prevalence of direct versus generative retrieval in laboratory-based intentional memorycuing paradigms. They asked participants to elicit memories in response to concrete nouns and emotional cue words presented on a computer screen, and to report for each memory whether it was retrieved directly or generatively. Uzer and colleagues found surprisingly high rates of direct retrieval: about half of the memories were 3 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. reported to have been retrieved directly. They found that this was the case whether they asked participants to make this distinction in terms of retrieval time (“this memory came immediately to mind”), retrieval effort (“I did not have to actively search for this memory”), or the degree to which they brought additional information to mind (“I did not have to think of additional information from my life to help me recall this memory”). This study indicated that direct retrieval may be more common than previously assumed, and may be occurring in relatively high rates even in laboratory paradigms designed to assess generative retrieval (see also Barzykowski & Staugaard, 2016; Harris et al., 2015; Uzer & Brown, 2017). Although Uzer et al. (2012) developed a new method of relying on participant self-report to distinguish direct and generative retrieval, other research has used different kinds of cues to vary retrieval type. In this methodology, high-imageability, concrete, or personalised cues are used to index direct retrieval and abstract, emotional, or more generic cues are used to index generative retrieval instead of participant self-report (e.g. Addis et al., 2012; Anderson et al., 2017; Williams et al., 2006). This cue-based approach has some consistency with self-report: both Uzer et al. (2012) and Harris et al. (2015) found that concrete nouns led to higher rates of self-reported direct retrieval than emotion cues. However, neither cue type nor retrieval latency perfectly overlapped with selfreported retrieval type. Concrete nouns, assumed to index direct retrieval, resulted in self-reported direct retrieval for about half of trials, and emotion words, assumed to index generative retrieval, resulted in self-reported direct retrieval for about a third of trials (Uzer et al., 2012; Harris et al, 2015). Uzer and Brown (2017) found particularly high rates of self-reported direct retrieval – around 80% – with personally relevant cues that participants generated for themselves. Overall, while cue type does influence rates of direct retrieval, it far from accurately predicts self-reported retrieval type. So far, relatively few studies have been conducted to compare self-reported direct versus 4 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. generative retrieval, and the nature of the distinction – including what information participants rely on to judge the different retrieval types – remains poorly understood. As noted above, different operational definitions of direct retrieval have appeared in both theory and research, sometimes based on effort, sometimes based on retrieval time, and sometimes based on search. Thus, the distinction between direct and generative retrieval is not clearly defined, with debates about the relationship between multiple processes involved in memory retrieval such as cue generation, hierarchical memory search, and memory reconstruction in autobiographical memory retrieval (Harris et al., 2015; Uzer et al., 2012). Uzer et al. (2012) argued that the prevalence of direct retrieval supports a “dual strategies” (p. 1297) interpretation of memory retrieval, according to which direct and generative retrieval represent distinct cognitive processes. Direct retrieval happens when a memory of a past event is immediately and effortlessly specified by the presence of a memory cue. Generative retrieval, on the other hand, involves a protracted search for, and/or generation of, relevant memory cues. According to Uzer et al. (2012), both processes arrive at the same database of “prestored event units”. Following this view, we should expect direct retrieval to be uniformly fast regardless of the nature of retrieval cue, because direct retrieval only happens when the cue is sufficiently strong to specify a memory with no further search or cuing. In contrast, we should expect the retrieval time of generative retrieval to be influenced by characteristics of the memory cues, for example, whether they are concrete or abstract, as chains of cues are generated to yield a cue that can trigger a memory. According to Uzer et al.’s (2012) account, once the two types of retrieval have arrived at a memory, there should be few, if any, differences regarding the subjective qualities of the memories retrieved, and memory qualities should not be influenced by the nature of the cues. This is because both types of retrieval are proposed to access the same database of prestored event representations and to differ only in the cue generation processes that precede retrieval of a memory. 5 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. In contrast, Haque and Conway (2001) distinguished between direct and generative retrieval by proposing that direct retrieval bypasses the top-down memory reconstruction that is involved in generative retrieval (also see Conway & Pleydell-Pearce, 2000). According to this view, we would expect both the retrieval time and the qualities of the memories to interact with the characteristics of the cues. Directly retrieved memories should show uniformly short retrieval times, and their subjective qualities should be unaffected by cue type, since there is no construction involved in their recall. Conversely, we would expect the retrieval time and the characteristics of the generatively retrieved memories to depend on cue type, since the nature of the cue should influence the topdown search and reconstruction processes involved in generative retrieval. In short, both Uzer et al.’s (2012) account and Haque and Conway’s (2001) accounts of memory retrieval processes predict that direct retrieval is always fast, regardless of the cue, albeit for different reasons. Both accounts imply that the nature of the retrieval cue might influence the frequency of direct retrieval, but in cases where direct retrieval is reported, retrieval time should not depend on cue characteristics. Conversely, both accounts imply that the speed of generative retrieval should be cue dependent, as the cue qualities may determine how much additional information needs to be generated, or how much hierarchical top-down search and generation is required, prior to successful retrieval of a specific event. The accounts differ with regards to the qualities of retrieved memories. Uzer et al.’s (2012) account suggests that direct and generatively retrieved memories should be similar in terms of their qualities. Haque and Conway’s (2001) theory is more consistent with differences in the qualities of direct and generatively retrieved memories, and implies that the qualities of directly retrieved memories should be cue-independent, while the qualities of generatively retrieved memories may be cue-dependent, since only generative retrieval involves top-down constructive processes. To examine these two different positions, we examined the retrieval latency for memories 6 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. that were directly retrieved versus generatively retrieved in response to three kinds of cues. We also examined the qualities of memories produced by these two retrieval types and the extent to which their retrieval latencies and phenomenology were cue-dependent. We adopted Uzer et al.’s (2012) methodology to distinguish between direct and generative retrieval; a self-report technique in which participants were asked to rate whether they had “searched” and used additional information to retrieve the memory. That is, the degree of search was given as the defining feature of direct retrieval. We aimed to compare directly retrieved memories and generatively retrieved memories, and to examine the extent to which prevalence, retrieval speed, and qualities of direct and generative retrieval varied depending on the nature of the retrieval cue. Method Participants Sixty-five (54 females and 11 males, mean age 22.95 years, SD = 3.65, range 20-46 years) psychology undergraduates from Aarhus University, Denmark, participated in this study in return for cinema tickets. All participants were informed that their responses were anonymous, that they should not elicit anything that was uncomfortable for them to disclose, and that they were free to withdraw at any point during the procedure. The design was fully within-subjects. Materials Participants elicited memories in response to 24 cues. Eight of the cues were Danish translations of the concrete noun cues used by Uzer et al. (2012): book, pill, bag, dog, river, bread, pencil, chair, radio. Eight of the cues were Danish translations of the emotion word cues used by Uzer et al. (2012): surprised, bored, sad, afraid, frustrated, happy, amused, daring, satisfied. The final eight cues were “personal cues”, the names of four people and four places that were idiosyncratic to each participant, and provided by participants via email prior to the experimental session as described below. 7 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Procedure Personal cue elicitation. After participants signed up for the study, one week before the scheduled experimental session the experimenter emailed them and asked them to provide some information about themselves. Specifically, they were asked to list the first and last name of four people and four specific places that had impacted their lives regularly in the last three years. They were also given further instructions on how to give specific responses, and told to exclude direct family members, their home, and the university campus. These instructions were based on Uzer and Brown (2017) and were designed to generate unique, specific, distinctive memory cues. Participants emailed their responses to the experimenter. Memory elicitation and rating. Participants were tested in groups of between 2 and 8. They were seated at individual computers. The experimenter told them that they would be presented with a series of cue words on the computer, and for each one, they should report the first memory that came into their mind in response to this cue. Participants were encouraged to think of specific events, to sample widely from their life, and to try not to repeat events if possible. The experimenter told participants that they should press the space bar as soon as they had the event in mind. They would then be prompted to type a description of the event and to answer some questions about it. The experimenter then gave participants detailed instructions about how to report the retrieval type for each event that they recalled. She explained that there are two ways that people can retrieve memories, when asked to recall personal events in response to cues: the first is when the cue directly triggers a memory and no additional information needs to be thought about; the second is when the cue does not directly trigger a memory so additional information from one’s life is thought about in order to arrive at a specific memory. These instructions were a direct Danish translation from Uzer et al.’s (2012) Experiment 3, where they developed instructions for distinguishing direct and generative retrieval without defining them in terms of effort or time for 8 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. participants, to avoid confounding reaction time or retrieval difficulty with self-reported retrieval process. Each item was presented in a random order on the computer screen. At the bottom of the screen, an instruction appeared reminding participants to “Press the spacebar as soon as you have a memory in mind”. If participants pressed the spacebar, they were first prompted to report the retrieval type. To avoid biasing responses in either direction, we used two versions of this question and counterbalanced which version participants saw. Half the participants reported retrieval type by pressing “1” for direct and “2” for generative retrieval; the other half reported retrieval type by pressing “1” for generative and “2” for direct retrieval. Participants then typed in a description of the memory. They also reported whether it referred to a general or specific event, and again we used two versions of this question and counterbalanced versions across participants to avoid response bias. Half the participants reported specificity by pressing “1” for specific and “2” for general event; the other half reported specificity by pressing “1” for general and “2” for specific event. Finally, participants reported how old they were in years at the time of the event, and rated on 7-point Likert scales the memory's clarity (1 = not at all clear, 4 = somewhat clear, 7 = extremely clear), personal significance (1 = not at all significant, 4 = somewhat significant, 7 = extremely significant), rehearsal (1 = not rehearsed at all, 4 = somewhat rehearsed, 7 = rehearsed a great deal), and valence (1 = highly negative, 4 = neutral, 7 = highly positive). If participants did not press the spacebar, the item timed out after 90 seconds, and the computer automatically presented the next item. The procedure was repeated until participants had completed all 24 items. This procedure took approximately 1 hour and 20 minutes, although it varied between participants. Memory coding. Two independent raters (native Danish speakers) coded two aspects of the elicited autobiographical memories. First, they coded the specificity of the memory according to a 5-category system, the “SCEPT Decision Tree” (Raes, Hermans, Williams, & Eelen, 2007). Using 9 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. this system, they scored whether each memory was: (1) a specific event; (2) specific extended i.e. lasting more than a day; (3) a lifetime period (e.g. “when I was in high school”); (4) a categoric or repeated event (e.g. “going to a restaurant”); or (5) a semantic associate (e.g. “my dog is friendly”). Both coders scored all of the elicited memories, and their agreement was 84.38%. Disagreements were resolved by discussion between the two coders. Results Analysis Strategy and Manipulation Checks. Initial tests indicated that we could collapse across participants in the different counterbalancing conditions. For each participant, we scored the proportion of elicited memories that they rated as directly retrieved, and the proportion of memories they rated as specific. An independent samples t-test comparing rates of direct retrieval across counterbalancing conditions showed no significant difference, t(63) = 0.63, p = .532. Similarly, an independent samples t-test comparing rates of specific recall across counterbalancing conditions indicated no significant difference, t(63) = 1.22, p = .229. That is, the way in which we asked participants to report the retrieval type and specificity for each memory did not appear to bias their overall responding rate in either direction. Participants successfully recalled memories within the time limit to 96.41% of the memory cues, leading to 1500 elicited memories from 1560 presented cues. A 3-level (cue type) within subjects ANOVA on proportion of memories successfully recalled yielded a significant main effect, F(2,63) = 15.15, p < .001, ηp2 = .19. Follow up pairwise comparisons (reported ps throughout include a Bonferroni adjustment for multiple comparisons where appropriate) indicated that personal cues (M = 99.23%, SD = 0.03) and concrete cues (M = 97.69%, SD = 5.80) both resulted in higher rates of retrieval success than emotion cues (M = 91.54%, SD = 13.28), all ps < .002, with no significant difference between personal and concrete cues, p = .176. Despite these differences in 10 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. success rates between cues, participants successfully recalled memories in response to most of the cues. Prevalence and Speed of Direct and Generative Retrieval For the memory qualities (reported by the participant or scored by the coders), we used IBM SPSS Statistics 24 to apply multi-level models that analysed memories as nested within the random factor of “participant” who recalled them, to take into account that each participant contributed multiple memories to the dataset (see Wright, 1998). We also included the cue order (i.e. a number between 1 and 24 to indicate when it was presented) and cue content (i.e. a number between 1 and 24 to indicate which specific item the cue was) as random factors in the model. We used the maximum likelihood method and we specified a random intercept. This approach was adopted due to the variability in the rates of direct retrieval across participants. Reported direct retrieval ranged from 21% to 79% of elicited memories for each participant, such that creating mean scores at the participant level would bias the results and give participants who reported fewer “direct” memories more weight in the dataset (for a discussion and comparison of both approaches to analysis, see Harris et al., 2015; Wright, 1998). Overall, participants reported direct retrieval for 713 (47.5%) elicited memories. To test the relationship between cue type and retrieval type, we conducted an ANOVA for the effects of cue type (concrete vs. emotion vs. personal) on retrieval type. Direct retrieval was scored as “1” and generative retrieval was scored as “0” so that the means represent proportion of memories directly retrieved. We specified cue type as a fixed factor, and participant, cue order, and cue content as random factors. This analysis yielded a main effect of cue type, F(2,760.63) = 50.79, p < .001. The model parameters for the difference between concrete cues and personal cues were β = .211, SE = .03, t(341.32) = 7.13, p < .001, 95% CI of the difference = .15-.27; the model parameters for the difference between emotion cues and personal cues were β = .290, SE = .03, t(1308.18) = 9.71, p < 11 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. .001, 95% CI of the difference = .23-.35. Follow up pairwise comparisons indicated that regardless of retrieval type, personal cues resulted in the highest rates of direct retrieval (M = 0.64, SD = 0.48), followed by concrete cues (M = 0.43, SD = 0.49), and emotion cues resulted in the lowest rates (M = 0.35, SD = 0.48), all ps < .026. Thus, consistent with previous research, direct retrieval was common, and the nature of the cue influenced the rates of direct retrieval (Uzer & Brown, 2017; Uzer et al., 2012). To test the relationship between reported retrieval type and retrieval speed, we conducted a 2 (retrieval type: direct vs. generative) × 3 (cue type: concrete vs. emotion vs. personal) ANOVA on time taken to recall a memory (reported in seconds throughout). We specified retrieval type and cue type as fixed factors, and participant, cue order, and cue content as random factors. This analysis yielded a main effect of retrieval type, F(1,1495.10) = 174.02, p < .001, and a main effect of cue type, F(2,1441.30) = 40.06, p < .001. Overall, regardless of cue type, reported direct retrieval was significantly faster than reported generative retrieval, as expected (see Table 1). Follow up pairwise comparisons indicated that regardless of retrieval type, personal cues resulted in the fastest retrieval (M = 6.56, SD = 8.19), followed by concrete cues (M = 10.16, SD = 12.58), and emotion cues resulted in the slowest retrieval (M = 15.48, SD = 16.20), all ps < .001. However, these main effects were moderated by a significant interaction between cue type and retrieval type, F(2,1478.48) = 4.34, p = .013. Separate one-way ANOVAs tested the effect of cue type for directly retrieved memories and generatively retrieved memories. For both multi-level ANOVAs, we specified cue type as a fixed factor and we specified participant as a random factors (we removed cue order and cue item from these analyses as the model would not converge when they were included). For both directly and generatively retrieved memories, there were significant main effects of cue, F(2,677.65) = 24.26, p < .001 and F(2,758.92) = 22.32, p < .001, respectively. Fixed effects comparisons for the different 12 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. levels of cue type indicated that, for directly retrieved memories, concrete and personal cues resulted in significantly faster retrieval than emotion cues, model parameters β = 4.38, SE = .65, t(683.29) = 6.71, p < .001, 95% CI of the difference = 3.10-5.66 seconds. However concrete and personal cues did not differ from each other, β = .44, SE = .60, t(674.68) = 0.75, p = .456, 95% CI of the difference = -0.72-1.61 (see Table 1 for means). For generatively retrieved memories, concrete and personal cues resulted in significantly faster retrieval than emotion cues, model parameters β = 8.72, SE = 1.36, t(768.10) = 6.42, p < .001, 95% CI of the difference = 6.05-11.38 seconds. Personal cues resulted in significantly faster retrieval than concrete cues, model parameters β = 3.54, SE = 1.37, t(763.83) = 2.59, p = .010, 95% CI of the difference = 0.86-6.27 seconds. Overall, these results suggest that regardless of whether direct or generative retrieval was reported, retrieval latency was cue-dependent, with retrieval to abstract emotion cues being particularly slow (see means in Table 1). Despite the average differences between direct and generative retrieval in terms of time taken to recall, it is important to note that there was a large range in elicitation latencies, and latency was not necessarily a proxy for retrieval type (see Figure 1 for the distribution). There was a considerable amount of overlap in elicitation latencies, rather than a clearly bimodal distribution indicating distinct “fast” and “slow” retrieval trials (see Figure 1). We return to this point in the Discussion. Retrieval type and Memory Specificity For each elicited memory, participants rated whether it referred to a specific event or not. We calculated the proportion of events elicited to each cue that were specific. Overall, rates of specific retrieval were very high (76.53% of the memories), consistent with experimental instructions to recall specific events. To test the relationship between retrieval type and specificity, we conducted a multi-level ANOVA for the effects of retrieval type (direct vs. generative) on participant-rated 13 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. specificity. Specific event retrieval was scored as “1” and non-specific retrieval was scored as “0” so that the means represent proportion of memories that were specific events. We specified retrieval type as a fixed factor, and participant as a random factor (we removed cue order and cue content as the model could not achieve convergence with these random factors included). This analysis yielded a main effect of retrieval type, F(1,1480.80) = 4.83, p = .028 (see Table 2 for model parameters). Generative retrieval resulted in higher rates of specific memories (M = 0.80, SD = .40) than direct retrieval did (M = .73, SD = .44). These results were confirmed by more detailed coding conducted by trained, independent coders. These coders assigned participants’ event descriptions as belonging to one of 5 categories: specific event, repeated event, categoric event, lifetime period, or semantic information. There were 1492 memory descriptions, with 8 missing (from participants pressing “enter” to continue without typing a memory description). We calculated the proportion of events elicited to each cue that were specific vs. not specific (collapsed across the other 4 categories), and conducted the same multilevel ANOVA reported above. The results of this analysis mirrored participants’ self ratings, yielding a main effect of retrieval type, F(1,1468.80) = 6.63, p = .010 (see Table 2 for model parameters). As previously, this analysis indicated that generative retrieval resulted in higher rates of specific memories (M = 0.83, SD = .38) than direct retrieval did (M = .76, SD = .43). For participant-rated and coded specificity, we were interested in whether memory specificity was cue dependent in similar or different ways for direct vs. generative retrieval. Means split by cue type are presented in Table 1. We conducted two multi-level ANOVAs for retrieval type (direct vs. generative) × cue type (concrete vs. emotion vs. personal) on specificity ratings. We specified retrieval type and cue type as fixed factors, and participant, cue order, cue content, and intercept as random factors. For both specificity measures, this analysis yielded a significant main effect of retrieval type, as reported above, and but no main or interaction effects of cue type, all Fs 14 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. < 2.49, all ps > .084. Therefore, memory specificity varied with retrieval type but not with cue type. Overall, generative retrieval yielded more specific memories than direct retrieval. Retrieval type and Memory Qualities Across reported age, and ratings of clarity, rehearsal, significance, emotional valence, and emotional intensity, we conducted a series of multi-level ANOVAs for retrieval type (direct vs. generative) on participant responses, to examine whether there were differences between directly and generatively retrieved memories. As above, we specified retrieval type as a fixed factor, and participant, cue order, cue content, and intercept as random factors. Table 1 presents means and Table 2 presents model parameters across quality variables. For reported age of the memory, there was no effect of retrieval type, F(2,1456.70) = 1.05, p = .306. Similarly, for the emotional valence of the memory, there was also no effect of retrieval type, F(2,1467.55) = 1.89, p = .170. However, for the remaining measures, retrieval type did influence memory quality. For reported clarity, the main effect of retrieval type, F(1,1468.61) = 12.90, p < .001, indicated that directly retrieved memories were rated as more clear and vivid than generatively retrieved memories. For reported personal significance, the main effect of retrieval type, F(1,1460.77) = 27.68, p < .001, indicated that directly retrieved memories were rated as more personally significant than generatively retrieved memories. For reported rehearsal, the main effect of retrieval type, F(1,1472.38) = 17.96, p < .001, indicated that directly retrieved memories were rated as more frequently rehearsed than generatively retrieved memories. For emotional intensity, we calculated scores by obtaining the absolute distance from the “neutral” midpoint of 4 on the valence scale for each memory. This score gave an indication of rated emotion regardless of valence, with a lower score indicating a rating closer to the midpoint. Scores could range from 0 to 3. For reported emotional intensity, the main effect of retrieval type, F(1,1483.08) = 24.60, p < .001, indicated that directly retrieved memories were rated as more emotionally intense than generatively 15 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. retrieved memories. For the characteristics that varied with retrieval type, we were interested in whether memory qualities were cue dependent in similar or different ways for direct vs. generative retrieval. Means split by cue type are presented in Table 1. We conducted a series of multi-level ANOVAs for retrieval type (direct vs. generative) × cue type (concrete vs. emotion vs. personal) on clarity, rehearsal, significance, and emotional intensity ratings. We specified retrieval type and cue type as fixed factors, and participant, cue order, cue content, and intercept as random factors. For all characteristics, this analysis yielded a significant main effect of retrieval type, as reported above, and a significant main effect of cue type, all Fs > 40.34, all ps < .001. Importantly, the interaction between retrieval type and cue type was not significant for any of these comparisons, all Fs < 1.39, all ps > .251. These findings suggest that, although directly retrieved memories were on average rated higher across these phenomenological characteristics, both directly and generatively retrieved memories showed higher phenomenology ratings for memories retrieved to emotion and personal cues than for those memories elicited to concrete cues (see Table 1). That is, for both direct and generatively retrieved memories, associated phenomenology was cue-dependent in similar ways. We return to this point in the Discussion. Discussion We replicated recent findings (Barzykowski & Staugaard, 2016; Harris et al., 2015; Mace, Clevinger, Delaney, Mendez, & Simpson, 2017; Uzer et al., 2012) indicating a high prevalence of self-reported direct memory retrieval. Direct retrieval was reported on approximately half of the trials. Direct retrieval was more common in response to concrete, highly imageable cues than to abstract, low imageable emotion cues (see also Anderson et a., 2017; Williams et al., 2006), but most often was reported in response to personally-relevant cues (see also Uzer & Brown, 2017). The findings for personally-relevant cues suggest that cues that were personally meaningful and 16 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. relevant were more likely to directly activate a memory. Meaningfulness is likely to be idiosyncratic: some generic cues triggered direct retrieval, but the cues that resulted in direct retrieval for some participants did not necessarily result in direct retrieval for others. Overall, the nature of the cue was not a good proxy of self-reported retrieval type, with direct retrieval reported on less than half the trials in response to concrete nouns. This finding suggests that paradigms that use concrete and abstract cues as the method for distinguishing direct and generative retrieval (e.g. Anderson et al., 2017; Williams et al., 2006) do not neatly separate retrieval trials into the same categories as those derived from self-report. Directly retrieved memories were elicited faster than generatively retrieved memories on average, consistent with previous findings (Uzer & Brown, 2017; Uzer et al., 2012). Interestingly, at the level of individual responses, retrieval latency did not necessarily mirror retrieval type. In fact, the slowest 30% of the self-reported directly retrieved memories took more than 5 seconds to recall, and 10% took more than 11 seconds to recall, while 30% of the self-reported generative retrieval trials took less than 6 seconds. One possible interpretation of the substantial range and overlap of retrieval latencies associated with both direct and generative retrieval is that participants simply are not accurate when self-reporting retrieval type. Reliance on self-report assumes that participants can accurately reflect on their retrieval processes. Alternatively, these surprising latency results may yield insights into the nuances of the distinction between direct and generative retrieval. We defined direct retrieval for participants in terms of the amount of information they had needed to search for in order to recall the target event, consistent with Uzer et al., (2012). Some previous research has used speed of retrieval as a proxy for indexing whether retrieval was direct or generative (e.g. Anderson et al., 2017; Haque & Conway, 2001; Williams et al., 2006) , and some definitions of direct and generative retrieval refer to retrieval speed: direct retrieval is “immediate” (e.g. Ros et al., 2009, p. 91) while generative retrieval is “protracted” (e.g. Haque & Conway, 2001, p. 529). 17 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Our findings suggest it is possible for direct retrieval to be slow, and for generative retrieval to be fast. Directly retrieved memories had distinct phenomenological characteristics compared to generatively retrieved memories. First, directly retrieved memories were less likely to refer to specific events than were generatively retrieved memories. Differences for specificity were small, but we had explicitly instructed participants to report specific events and the majority of memories were specific regardless of retrieval type. We also found that directly retrieved memories on average were rated as clearer, more personally significant, more frequently rehearsed, and more emotionally intense than generatively retrieved memories. These differences in phenomenology of memories retrieved via direct versus generative retrieval are consistent with prior research (Barzykowski & Staugaard, 2016; Harris et al., 2015). Overall these findings suggest that direct retrieval accesses the most available and personally relevant memories. Most importantly, we examined two possibilities based on opposing theoretical accounts of direct and generative retrieval. According to Uzer et al. (2012), both direct and generative retrieval access “pre-stored event representations” (p. 1306), and differ only in the amount of cue elaboration that precedes retrieval. This account appears to predict that the retrieval latency of generative retrieval – but not of direct retrieval – should be cue dependent, and that the qualities of directly and generatively retrieved memories should be the same as each other. According to Haque and Conway (2001), generative retrieval involves a top-down memory construction process which direct retrieval bypasses. Similarly, Barzykowski and Staugaard (2016) suggested that direct retrieval may come about because less memory reconstruction is required for frequently rehearsed events, and argued that generative retrieval involves more memory reconstruction. This latter account therefore also predicts that the retrieval latency of generative retrieval – but not of direct retrieval – should be cue dependent, but that the qualities of directly and generatively retrieved memories may differ. 18 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Counter to both these theories, we found that the speed of direct and generative retrieval were cue dependent in broadly similar ways, with emotion cues resulting in the slowest direct retrieval as well as the slowest generative retrieval. Uzer et al. (2012) reported no interaction between retrieval type and cue, but they interpreted that finding as indicating that retrieval latencies for both retrieval types were similarly cue-independent. Instead, our findings as illustrated in Table 2 show that the retrieval latencies for both retrieval types are similarly cue-dependent. Moreover, phenomenology of direct and generative retrieval were also similarly cue dependent, with emotion cues and personal cues leading to richer phenomenology than concrete cues, regardless of retrieval type. Thus, our results do not support the view that autobiographical memory retrieval accesses prestored event representations, nor that direct retrieval bypasses memory construction, suggesting instead that direct and generative retrieval do not involve fundamentally different processes. We propose a third theoretical position: that both direct and generative retrieval involve construction of autobiographical memories, differing only in the degree of subjective effort required for retrieval. Further work is needed to understand what pre-retrieval processes participants are reflecting on when they report a lack of search or effort in recall. There are multiple effortful preretrieval processes including cue elaboration, semantic association, memory construction, control processes, and the inhibition of irrelevant information, and each of these may be reduced in direct retrieval. Further research is needed to understand how these multiple pre-retrieval processes can vary – independently or together – and how these variations influence subjective experiences of retrieval as well as the resulting memory phenomenology (see also Harris et al., 2015). Importantly for definitions of direct retrieval which characterize them as “spontaneous” or “unexpected”, the profile of directly retrieved memories is not similar to the profile of involuntary autobiographical memories, as observed in diary studies. First, involuntary autobiographical memories, by definition, occur with no preceding instruction or attempt to retrieve a memory 19 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. (Berntsen, 1996). In contrast, directly retrieved memories, when measured in experiments, are always preceded by cues and an explicit instruction to retrieve a memory in response to each cue. Second, the characteristics and phenomenology of the two classes of memories also differ. For instance, in diary studies, involuntary memories are more likely to be specific than voluntary memories (Berntsen, 1998; 2010; Mace, 2006; Schlagman & Kvavilashvili, 2008), although diary studies do not directly instruct participants to report only specific events in the same way that laboratory methods do. Involuntary memories are associatively cued by features in the ongoing situation rather than an intention to retrieve (Berntsen, 1996; Mace, 2006; Berntsen & Jacobsen, 2008; Rasmussen & Berntsen, 2011). In diary studies, involuntary retrieval results in memories with somewhat idiosyncratic or ‘unusual’ content (Berntsen & Hall, 2006). Several studies have found them to be less relevant to the person’s life story (e.g. Cole, Staugaard, & Berntsen, 2016; Johannessen & Berntsen, 2010; Rubin et al., 2008) and less rehearsed than voluntary memories (Berntsen, 1998; Johannessen & Berntsen, 2010; Rubin, Boals & Berntsen, 2008; Rubin, Dennis, & Beckham, 2011). Overall, our results regarding the specificity of directly retrieved memories, and their phenomenological characteristics, suggest that direct retrieval involves recalling memories that are already highly accessible, relevant, and frequently rehearsed. Extensive executive search for these memories may not be required because they are readily accessible when in “retrieval mode” with the intention to recall (Tulving, 1983, 2002). Overall, directly retrieved memories do not fit the profile of involuntary memories and the two phenomena cannot be conceptually equated. In terms of specificity and phenomenology, the profile of directly retrieved memories (compared to generatively retrieved) is directly opposite to the profile of involuntary memories (compared to voluntary) for several important variables. Based on our findings, it appears that a conscious initiation of the retrieval process, with the general intention to recall a memory, matters for the characteristics of the memories that are recalled, such that direct retrieval and involuntary retrieval 20 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. are different despite their shared lack of effort. Intentional recall – even when relatively effortless i.e. direct retrieval – appears to facilitate retrieval of memories that are most personally-relevant and frequently rehearsed. Overall, these findings contribute to efforts to understand the nature of the distinction between direct and generative retrieval, using the relatively recent self-report methodology to distinguish between these two types of retrieval. There are distinct phenomenological characteristics associated with memories recalled via direct versus generative retrieval, consistent with previous research, suggesting that participant self-report of retrieval processes reflects genuine differences in the nature of memories recalled. However, our findings that the qualities of directly retrieved and generatively retrieved memories were similar in terms of their cue-dependence suggests that these retrieval types are not fundamentally different in terms of the cognitive processes involved. Our findings were counter to the predictions derived from of two existing models of direct versus generative autobiographical memory retrieval. Counter to these models, the present findings suggest that both direct and generative retrieval are the result of similar constructive memory processes, even though there are differences in the content of memories recalled. Direct retrieval is experienced when memories that are already highly accessible are retrieved, but direct retrieval does not appear to reflect a fundamental difference in the nature of the pre-retrieval processes involved in bringing a memory to mind. 21 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Author note This work was supported by the Danish National Research Foundation (DNRF89), the Danish Council for Independent Research: Humanities (FKK) and the MindLab UNIK initiative at Aarhus University, which is funded by the Danish Ministry of Science and Technology and Innovation. Celia Harris was supported by an Australian Research Council Discovery Early Career Research Award (DE150100396). We are grateful for this support. 22 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. References Addis, D. R., Knapp, K., Roberts, R. P., & Schacter, D. L. (2012). Routes to the past: Neural substrates of direct and generative autobiographical memory retrieval. NeuroImage, 59, 2908–22. doi: 10.1016/j.neuroimage.2011.09.066 Anderson, R. J., Dewhurst, S. A., & Dean, G. M. (2017). Direct and generative retrieval of autobiographical memories : The roles of visual imagery and executive processes. Consciousness and Cognition, 49, 163–171. doi:10.1016/j.concog.2017.02.010 Barzykowski, K., & Staugaard, S. R. (2016). Does retrieval intentionality really matter? Similarities and differences between involuntary memories and directly and generatively retrieved voluntary memories. British Journal of Psychology, 107, 519-36. doi: 10.1111/bjop.12160 Barzykowski, K., & Staugaard, S. R. (2018). How intention and monitoring your thoughts influence characteristics of autobiographical memories. British Journal of Psychology, 109, 321–340. doi: 10.1111/bjop.12259 Berntsen, D. (1996). Involuntary autobiographical memories. Applied Cognitive Psychology, 10, 435-454. doi: 10.1002/(SICI)1099-0720(199610)10:5<435::AID-ACP408>3.0.CO;2-L Berntsen, D. (1998). Voluntary and involuntary access to autobiographical memory. Memory, 6, 113–141. doi: 10.1080/741942071 Berntsen, D. (2009). Involuntary autobiographical memories: An introduction to the unbidden past. Cambridge, UK: Cambridge University Press. Berntsen, D. (2010). The unbidden past: Involuntary autobiographical memories as a basic mode of remembering. Current Directions in Psychological Science, 19, 138–142. doi: 10.1177/0963721410370301 Berntsen, D., & Hall, N. M. (2004). The episodic nature of involuntary autobiographical memories. Memory & Cognition, 32, 789–803. doi: 10.3758/BF03195869 23 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Berntsen, D., & Jacobsen, A. S. (2008). Involuntary (spontaneous) mental time travel into the past and future. Consciousness and Cognition, 17, 1093–104. doi: 10.1016/j.concog.2008.03.001 Cole, S. N., Staugaard, S. R., & Berntsen, D. (2016). Inducing involuntary and voluntary mental time travel using a laboratory paradigm. Memory and Cognition, 44, 376–389. doi: 10.3758/s13421-015-0564-9 Conway, M. A., & Pleydell-Pearce, C. W. (2000). The construction of autobiographical memories in the self-memory system. Psychological Review, 107, 261–288. doi: 10.1037/0033295X.107.2.261 Haque, S., & Conway, M. A. (2001). Sampling the process of autobiographical memory construction. European Journal of Cognitive Psychology, 13, 529–547. doi: 10.1080/09541440125757 Harris, C. B., O ’Connor, A. R., & Sutton, J. (2015). Cue generation and memory construction in direct and generative autobiographical memory retrieval. Consciousness and Cognition, 33, 204–216. doi: 10.1016/j.concog.2014.12.012 Johannessen, K. B., & Berntsen, D. (2010). Current concerns in involuntary and voluntary autobiographical memories. Consciousness and Cognition, 19, 847–60. doi: 10.1016/j.concog.2010.01.009 Mace, J. H. (2006). Episodic remembering creates access to involuntary conscious memory: Demonstrating involuntary recall on a voluntary recall task. Memory, 14, 917–924. doi: 10.1080/09658210600759766 Mace, J. H., Clevinger, A. M., Delaney, D. M., Mendez, A. S., & Simpson, S. H. (2017). Voluntary remembering : Elucidating the mental strategies used to recall the past, Applied Cognitive Psychology, 163, 156–163. doi:10.1002/acp.3313 24 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Raes, F., Hermans, D., Williams, J. M. G., & Eelen, P. (2007). A sentence completion procedure as an alternative to the Autobiographical Memory Test for assessing overgeneral memory in non-clinical populations. Memory, 15, 495–507. doi: 10.1080/09658210701390982 Rasmussen, A. S., & Berntsen, D. (2011). The unpredictable past: Spontaneous autobiographical memories outnumber autobiographical memories retrieved strategically. Consciousness and Cognition, 20, 1842–1846. doi: 10.1016/j.concog.2011.07.010 Ros, L., Latorre, J. M., & Serrano, J. P. (2009). Working memory capacity and overgeneral autobiographical memory in young and older adults. Aging, Neuropsychology, and Cognition, 17, 89–107. doi: 10.1080/13825580903042650 Rubin, D. C., Boals, A., & Berntsen, D. (2008). Memory in posttraumatic stress disorder: Properties of voluntary and involuntary, traumatic and nontraumatic autobiographical memories in people with and without posttraumatic stress disorder symptoms. Journal of Experimental Psychology: General, 137, 591–614. doi: 10.1037/a0013165 Rubin, D. C., Dennis, M. F., & Beckham, J. C. (2011). Autobiographical memory for stressful events: The role of autobiographical memory in posttraumatic stress disorder. Consciousness and Cognition, 20, 840–856. doi: 10.1016/j.concog.2011.03.015 Schlagman, S., & Kvavilashvili, L. (2008). Involuntary autobiographical memories in and outside the laboratory: how different are they from voluntary autobiographical memories? Memory & Cognition, 36, 920–932. doi: 10.3758/MC.36.5.920 Tulving, E. (1983). Elements of episodic memory. Oxford: Oxford University Press. Tulving, E. (2002). Episodic memory: From mind to brain. Annual Review of Psychology, 53, 1–25. doi: 10.1146/annurev.psych.53.100901.135114 Uzer, T., & Brown, N. R. (2017). The effect of cue content on retrieval from autobiographical memory. Acta Psychologica, 172, 84–91. doi: 10.1016/j.actpsy.2016.11.012 25 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitiv e version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Uzer, T., Lee, P. J., & Brown, N. R. (2012). On the prevalence of directly retrieved autobiographical memories. Journal of Experimental Psychology. Learning, Memory, and Cognition, 38, 1296–308. doi: 10.1037/a0028142 Watson, L. A., Berntsen, D., Kuyken, W., & Watkins, E. R. (2012). The characteristics of involuntary and voluntary autobiographical memories in depressed and never depressed individuals. Consciousness and Cognition, 21, 1382–1392. doi: 10.1016/j.concog.2012.06.016 Williams, J. M. G., Chan, S., Crane, C., Barnhofer, T., Eade, J., & Healy, H. (2006). Retrieval of autobiographical memories: The mechanisms and consequences of truncated search. Cognition and Emotion, 20, 351–382. doi:10.1080/02699930500342522 Wright, D. B. (1998). Modelling clustered data in autobiographical memory research: the multilevel approach. Applied Cognitive Psychology, 12, 339–358. doi: 10.1002/(SICI)10990720(199808)12 26 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitive version was subsequently published in Consciousne ss and Cognition. DOI:10.1016/j.concog.2019.102793. Table 1. Qualities of Elicited Memories, By Retrieval Type and Cue Type Retrieval Cue Type Type Direct Latency (seconds) Concrete Emotion Personal Overall Generative Concrete Emotion Personal Overall Specificity Specificity (prop) Age Clarity Rehearsal Significan Emotion Emotion – Coded (years (1-7) (1-7) ce Valence Intensity (prop) ago) (1-7) (1-7) (0-3) 5.60 0.70 0.73 5.31 4.75 3.16 3.38 4.59 1.43 (6.20) (0.46) (0.44) (6.39) (1.55) (1.65) (1.74) (1.65) (1.00) 9.36 0.69 0.74 2.60 5.50 4.81 4.92 3.69 2.22 (12.51) (0.46) (0.44) (4.97) (1.49) (1.81) (1.87) (2.37) (0.85) 4.25 0.77 0.78 3.75 5.45 4.35 5.01 5.29 2.09 (4.16) (0.42) (0.42) (4.66) (1.37) (1.72) (1.68) (1.86) (0.86) 5.86 0.73 0.76 3.96 5.25 4.10 4.49 4.70 1.92 (7.76) (0.44) (0.43) (5.40) (1.49) (1.84) (1.90) (2.03) (0.96) 13.52 0.80 0.84 4.62 4.45 3.02 2.97 4.50 1.20 (14.85) (0.40) (0.37) (5.35) (1.53) (1.63) (1.70) (1.48) (1.01) 18.76 0.79 0.79 2.49 5.16 4.12 4.31 4.16 1.92 (17.01) (0.41) (0.41) (3.96) (1.63) (1.65) (1.86) (2.17) (1.02) 10.66 0.80 0.87 4.09 5.13 4.00 4.40 5.26 1.83 (11.38) (0.40) (0.34) (4.86) (1.34) (1.58) (1.72) (1.62) (0.93) 14.91 0.80 0.83 3.66 4.89 3.68 3.83 4.54 1.63 (15.38) (0.40) (0.38) (4.82) (1.56) (1.70) (1.89) (1.86) (1.05) Note: values are means across elicited memories, and standard deviations in parentheses. 27 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitive version was subsequently published in Consciousne ss and Cognition. DOI:10.1016/j.concog.2019.102793. Table 2. Parameters for Reported Multilevel Model Comparing Directly vs. Generatively Retrieved Memories on Each Quality Latency Specificity Specificity Age Clarity Rehearsal Significance Emotion Emotion (seconds) Rated Coded (years ago) (1-7) (1-7) (1-7) Valence Intensity (1-7) (0-3) (proportion) (proportion) Model Estimate SE Model df 5.81 0.04 0.05 0.25 0.28 0.36 0.49 4.57 0.26 1.08 0.02 0.02 0.25 0.08 0.09 0.09 0.09 0.05 1486.76 1480.80 1468.04 1456.70 1468.61 1472.38 1460.77 1467.55 1483.07 Model t 5.38 2.20 2.57 1.02 3.59 4.24 5.26 1.37 4.96 Model p < .001 .028 .010 .306 < .001 < .001 < .001 .170 < .001 95% CI of Difference 3.69-7.93 .005-.09 .01-.09 -0.74-0.23 0.13-0.43 0.19-0.53 0.31-0.67 -0.33-0.06 0.16-0.36 28 Notice: This is the author’s version of a work that was accepted for publication Consciousness and Cognition A definitive version was subsequently published in Consciousness and Cognition. DOI:10.1016/j.concog.2019.102793. Figure 1. Distribution of retrieval latencies for directly and generatively retrieved memories. 29