Abstract
We examined recall of events by children 4 to 11 years to inform patterns of retention of autobiographical memories as well as factors that predict their survival. 101 children participated in a 4-year prospective study. At study inception, children were 4, 6, and 8 years. They were tested annually for three more years for a total of four waves of data collection. At each wave, we obtained narrative reports of recent (all waves) and distant (Waves 2–4) events, resulting in virtually continuous sampling of memories formed by 4- to 11-year-olds and recalled after 1–3-year delays. We also measured children’s language, and domain-general and memory-specific cognitive skills. Multi-level modeling revealed age-related increases in the likelihood of survival of memories over the delays. Critically, the rate of increase in retention of individual memories was the same across the cohorts. In addition to age, thematic coherence of original memory reports predicted memory survivability. Other factors were not predictive. The dense sampling and prospective tracking of memories across the 4–11-year age period permitted an especially strong test for continuity versus discontinuity in autobiographical memory across the second half of the first decade of life. The data are strongly indicative of continuity and gradual change.
Keywords: autobiographical memory, childhood amnesia, development, longitudinal study, narrative coherence, prospective study
The phenomenon of childhood amnesia (the relative paucity of autobiographical memories from the first years of life available for recall in adulthood; Freud, 1905/1953) has shaped the literature on the study of the development of memory in profound ways. Search for explanations of childhood amnesia has focused research on two questions in particular: the earliest age at which children remember specific past events, and the “fates” of early memories as children get older and begin to exhibit a more adult-like distribution of autobiographical memories, including loss of access to early memories. As a result of these foci, there is a sizeable literature on the development of autobiographical memory in early childhood and even infancy (see Bauer, 2015; Fivush & Zaman, 2014; Lukowski & Bauer, 2014, for reviews), and a smaller yet rich literature on the survival of early memories into later childhood and adolescence (e.g., Bauer & Larkina, 2014b, 2016; Peterson, Hallett, & Compton-Gillingham, 2018; Reese, Jack & White, 2010; Wang & Peterson, 2016). Yet occupation (if not preoccupation) with explanation of the phenomenon of childhood amnesia has meant that other periods in the development of autobiographical memory have been relatively neglected. One such period is the time between when earliest memories are formed (in the first 2–3 years of childhood) and the time of tests of their survival, around the end of the first decade of life. The focus of the present research is on this period of roughly 4 to 11 years of age, to inform patterns of retention of autobiographical memories formed in these years, and the factors that contribute to their survival over time. The data have implications for explanations of childhood amnesia.
Childhood Autobiographical Narratives and Memory: Open Questions
The period of roughly 4 to 11 years of age has not been wholly overlooked in the study of autobiographical memory. Research has revealed a number of changes in the narratives that children produce about their memories, as well as informed some of the factors that relate to developmental differences in narrative skills. Over childhood, there are increases in the length as well as the breadth or completeness of children’s narrative reports (i.e., they include more information about the who, what, where, when, why, and how of events), and in the coherence and complexity of their accounts (e.g., Habermas, Negele, & Mayer, 2010). The amount of information that children include nearly doubles over this period (Van Abbema & Bauer, 2005), as does the temporal organization of the reports that children produce (Morris, Baker-Ward, & Bauer, 2010). Children also include more orientation to the time and place of events, and they more effectively maintain and elaborate on topics (e.g., O’Kearney, Speyer, & Kenardy, 2007). Their reports also feature more causal connections (e.g., because, so that) that help to explain why events unfolded as they did (e.g., Bauer, Stark, Lukowski, Rademacher, Van Abbema, & Ackil, 2005; Habermas et al., 2010; see Bauer, 2015; and Reese, Haden, Baker-Ward, Bauer, Fivush, & Ornstein, 2011, for reviews). In longitudinal research, these changes have been found to be predicted by (a) non-autobiographical story recall; (b) domain-general cognitive abilities, such as speed of processing, working memory, and attention; and (c) memory-specific skills, including use of memory strategies and metamemory understanding (Bauer & Larkina, 2019).
In addition to changes in autobiographical narrative skills, the second half of the first decade of life is witness to gains in the length of time over which autobiographical memories are retained. It is clear that even 3-year-olds verbally report on specific past events after delays of 18 months and even longer (e.g., Bauer & Larkina, 2014b; Hamond & Fivush, 1991; nonverbal evidence of recall is apparent even in the first year of life: see Lukowski & Bauer, 2014, for a review). Yet across the first decade of life, there are increases in the likelihood that a given event will be recalled after a long delay. For example, in a prospective study, Bauer and Larkina (2016) found that over 1- to 3-year delays, adults were more likely to recall specific past events relative to children who had been 4, 6, and 8 years of age at the time of experience of the events. That is, multi-level modeling revealed that memories formed by adults had a higher likelihood of being recalled 1 to 3 years later, relative to all three child groups. As well, children who had been 8 years of age at the time of the events were more likely to recall them than children 4 and 6 years of age at the time of experience (see also Morris, et al., 2010). The differences were especially pronounced (i.e., higher odds ratios) in open-ended recall as opposed to recall that was prompted by wh- questions used as cues (e.g., who was there?). Indeed, when recall was supported by prompts and cues, age-related differences all but disappeared (see Bauer & Larkina, 2016, for details). How the likelihood that a given event will be recalled after a delay changes as children move through the elementary school years, as narrative ability develops, is unknown. This is an important question because it is over this period that the distribution of autobiographical memories becomes more adult-like, characterized by accumulation of recent memories and loss of access to early memories (Bauer & Larkina, 2014a).
Along with research on the developmental changes in the likelihood of recall of events over long delays, there is need for research on the factors that predict long-term retention of memories. The way parents, and in particular, mothers, talk with their children is a consistent determinant of children’s contributions to co-constructed autobiographical narratives in early childhood (e.g., Larkina & Bauer, 2010; Reese, Haden, & Fivush, 1993; see Fivush & Zaman, 2014, for a review), and it is a predictor of the age of earliest memory of events recalled by adolescents (Reese et al., 2010). Yet for children in the target age range, it has not proven to be an especially strong predictor of independently produced narratives (Bauer & Larkina, 2019; Larkina & Bauer, 2012; Peterson, Sales, Reese, & Fivush, 2007).
In contrast to maternal narrative style, research has revealed that characteristics of to-be-remembered events and of the narratives children produce to describe them are predictive of children’s independent recall after long delays. In a longitudinal investigation of 4- to 13-year-olds’ retention of their three earliest memories, Peterson and colleagues found that emotional events were more likely to be remembered (Peterson, Morris, Baker-Ward & Flynn, 2014). The survival of memories over a 2-year delay also was predicted by the coherence of children’s original memory narratives. To our knowledge, there have been no investigations of whether the factors that predict narrative skill, including memory-specific and domain-general cognitive abilities (Bauer & Larkina, 2019), also contribute to developmental changes in the likelihood of long-term recall. Moreover, how the pattern of prediction may change over the second half of the first decade of life also is unknown.
Informing Conceptualizations of Childhood Amnesia
Examination of patterns of change in remembering and forgetting over the second half of the first decade of life stands to inform our conceptualization of childhood amnesia. Consider that in both the adult and developmental autobiographical memory literatures, it is common to conceptualize childhood amnesia as having an “offset” around 5 to 7 years of age (e.g., Cleveland & Reese, 2008; Eacott & Crawley, 1998; Nelson & Fivush, 2004; Pillemer & White, 1989; Rubin, 2000). This conceptualization implies that at roughly this time, there are changes in memory (and/or associated conceptual domains) that permit children to emerge from a period during which few or no autobiographical memories were formed, into a period in which there is a steady accumulation of autobiographical memories that can be retained over long periods of time (e.g., Fivush, 2014; Perner & Ruffman, 1995; Suddendorf, Nielson, & van Gehlen, 2011; Tulving, 2002, 2005; Wheeler, 2000). If this is the case, then we should observe different patterns of remembering and forgetting early in the 4- to 11-year age range and later in the period. Specifically, at ages 4, 5, and 6 years, few memories should be formed and they should be highly vulnerable to forgetting. In contrast, at ages 8, 9, and 10 years, more memories should be formed and they should be more resistant to forgetting. The age of 7 years is considered the “inflection point” (Wetzler & Sweeney, 1986)—the age at which the distribution starts to become more adult-like.
An alternative conceptualization of childhood amnesia is that it is better described in terms of an “onset,” rather than an offset (Bauer, 2014, 2015). In this view, the ability to form and retain autobiographical memories is an early rather than a later achievement (Bauer, 2007, 2015). The relative paucity of early life events accessible to recall by older children and adults is not because few or none were formed but because early-formed memories are forgotten over time (whether memories are lost from storage or are preserved but become inaccessible to recall is a matter of long-standing debate, resolution of which is beyond the scope of the present research; see Bauer & Larkina, 2014a, for discussion). In Bauer and Larkina (2014b), for example, memories formed at age 3 years were well preserved over delays of 2, 3, and 4 years, but more significant levels of forgetting were apparent by 5 and 6 years after the events (see also Cleveland & Reese, 2008; Fivush & Schwarzmueller, 1998; Peterson, Warren & Short, 2011). Indeed, throughout the first decade of life, relative to adults, children experience an accelerated rate of forgetting (Bauer, Burch, Scholin, & Güler, 2007; Bauer & Larkina, 2014a). The result is the “childhood amnesia” component of the autobiographical memory distribution (Pillemer & White, 1989)—fewer memories from early in life relative to the number expected by normal forgetting alone (with normal forgetting defined in terms of the adult rate of forgetting). This conceptualization is consistent with observations that relative to younger children, older children have later “earliest” memories (Peterson, Grant, & Boland, 2005; Tustin & Hayne, 2010), and that within subjects, over time, the age of earliest memory that children report edges upward, to older ages (e.g., Peterson et al., 2011; Wang & Peterson, 2016). This conceptualization implies that rather than discontinuous patterns, we should observe continuous patterns of remembering and forgetting across the 4- to 11-year age range. That is, there should be similarity in the slopes of forgetting of memories formed by children 4, 5, and 6 years and by children 8, 9, and 10 years. Examination of these competing conceptualizations requires dense sampling and prospective tracking of memories across the 4- to 11-year age range.
Present Research
In the present research, we used data from a 4-year prospective study (Bauer & Larkina, 2016, 2019) to further our understanding of developmental changes in children’s long-term recall of autobiographical events and experiences in two ways. First, we examined patterns of remembering and forgetting over delays of 1, 2, and 3 years, by children 4 to 11 years of age. As described in Bauer and Larkina (2016), the children were enrolled in the prospective study at the ages of 4, 6, and 8 years (Wave 1). They were tested annually for three subsequent years (Waves 2, 3, and 4). Thus the 4-year-old cohort was tested at ages 4, 5, 6 and 7 years; the 6-year-old cohort was tested at ages 6, 7, 8, and 9 years; and the 8-year-old cohort was tested at ages 8, 9, 10, and 11 years. In Bauer and Larkina (2016), we reported on the children’s recall of events experienced at 4, 6, or 8 years, after delays of 1, 2, and 3 years. Not included in that report were additional data on children’s long-term recall of events experienced at each of the subsequent waves of testing. That is, each year, in addition to recall of events experienced at the time of enrollment (4, 6, or 8 years), children also were asked to produce autobiographical narratives about new events from the recent past (within 4 months). In subsequent years, in addition to events experienced at the time of enrollment, children were tested for recall of events experienced 1 year and 2 years after enrollment. The net effect was an almost continuous sampling of recall of events from 1 year, 2 years, and 3 years in the past, by children 4 to 11 years of age. This period spans that of the densest amnesia for events from childhood (i.e., before 5 to 7 years), as well as the period of development of a more adult-like distribution of autobiographical memories (Bauer & Larkina, 2014a).
The second way in which the present research furthers our understanding of developmental changes in children’s long-term recall of autobiographical events and experiences is by permitting examination of the unique and combined variance in the likelihood of survival of memories over the long term contributed by measures of the quality of the narrative used to describe the event, and both domain-general and memory-specific abilities, and how the patterns of prediction change over the second half of the first decade of life. At each wave of data collection, in addition to recall of recent (all waves) and distant (Waves 2–4 only) events, we also obtained measures of domain-general (e.g., speed of processing, working memory, attention) and memory-specific (e.g., deliberate and strategic remembering, metamemory ability, non-autobiographical story recall) cognitive skills. We also assessed children’s language skills at the first wave of data collection. In Bauer and Larkina (2019), we used the measures to predict the qualities of the narratives children used to describe events from the recent past (within 4 months). In the present research, we used the measures to predict whether the memories of the events survived to be recalled 1 year, 2 years, or 3 years after the events took place, under the assumption that factors that impact whether an event is remembered over the short term also might impact recall over longer delays. None of the data from this analysis have been published previously.
Finally, the present research also permits test of the unique and combined variance in the likelihood that autobiographical memories will survive over the long term contributed by (a) children’s language, (b) both domain-general and memory-specific abilities, and (c) measures of the quality of autobiographical narratives, and how the patterns of prediction change over the second half of the first decade of life.
Method
The present analysis is of unique data collected in the course of a 4-year prospective study. Most of the methods and procedures have been described in prior reports (Bauer & Larkina, 2016, 2019). For this reason, they are presented in abbreviated fashion here.
Participants
The participants were 101 children (53 female, 48 male). At the beginning of the study, children were ages 4 (n =37), 6 (n = 34), and 8 (n = 30) years (see Table 1, Panel a, for sample demographic information). The children were drawn from a participant pool maintained by the Institute of Child Development, University of Minnesota, USA. The participant pool was comprised entirely of volunteers. The families were primarily Caucasian (94%); none was of Hispanic descent. Based on parental report of occupation and education, the families were of middle to upper-middle socioeconomic status. At the time of the inception of the study, the sample was reflective of the community from which it was drawn. An additional 34 children were enrolled but did not complete all points of data collection and thus are not included in the analyses. The demographic characteristics of the families lost to attrition did not differ from those of the families retained (see Bauer & Larkina, 2016, for details).
Table 1.
Sample Characteristics (Panel a) and Average Delays between Waves of Data Collection (Panel b)
| Overall | 4-year-olds | 6-year-olds | 8-year-olds | |
|---|---|---|---|---|
| Panel a: Sample characteristics | ||||
| Number of participants | 101 | 37 | 34 | 30 |
| Participant gender (F/M) | 53/48 | 16/21 | 19/15 | 18/12 |
| Age at initial assessment (in years): M (SD), range | n/a | 4.18 (0.06) | 6.19 (0.05) | 8.20 (0.04) |
| 4.07–4.40 | 6.10–6.28 | 8.12–8.28 | ||
| Panel b: Delays between waves of data collection (in days): M (SD), range | ||||
| Wave 1 to Wave 2 | 364 (34) | 369 (34) | 360 (37) | 364 (34) |
| 270–442 | 308–442 | 270–440 | 300–440 | |
| Wave 2 to Wave 3 | 329 (26) | 326 (24) | 330 (24) | 331 (29) |
| 264–403 | 264–366 | 278–365 | 292–403 | |
| Wave 3 to Wave 4 | 362 (17) | 362 (18) | 363 (18) | 360 (18) |
| 305–427 | 305–396 | 343–427 | 319–391 | |
At each of the four waves of data collection, each spaced approximately 1 year apart (see Table 1, Panel b), children took part in two test sessions, approximately 1 week apart. The University of Minnesota Institutional Review Board approved the protocol prior to the start of the study. At each wave of data collection, written parental consent was obtained for each child. Children ages 7 years and older provided written assent for their participation; children younger than 7 gave verbal assent. At the end of the second session at each wave, children received a toy, and parents were given a gift certificate to a local merchant.
Materials and Procedure
All testing took place in a university laboratory setting. Participants visited the laboratory annually for four years. At each wave, the children took part in two 1–1.5 hour sessions, approximately 1 week apart. Over the four years of data collection, nine female experimenters administered the tasks. The two sessions within a wave were conducted by the same experimenter; participants were tested by different experimenters at each wave. Task procedures were outlined in a written protocol, and the experimenters regularly reviewed and discussed videotaped sessions to ensure protocol fidelity.
Event selection and elicitation of narrative descriptions.
As described in Bauer and Larkina (2016), at each wave, events were drawn from calendars on which children’s parents recorded at least one unique event per week for a period of 4 months prior to the session. To establish whether children remembered the events and the quality of the narratives they used to describe them, children were interviewed about these events when they visited the laboratory. As depicted in Figure 1, Panel a, at Wave 1 (children were 4, 6, and 8 years of age), children were tested on nine events drawn at random from the calendars. These events then were randomly assigned to be tested again 1 year (Events 1–3), 2 years (Events 4–6), or 3 years (Events 7–9) later. Results on recall of the nine original events were reported in Bauer and Larkina (2016). The data on events initially assessed at Waves 2 and 3 and recalled after 1 year at Waves 3 and 4 (Panel b), are unique to this report. At Wave 2 (children were 5, 7, and 9 years of age), children were tested on six new events drawn at random from calendars maintained for the 4 months prior to the Wave 2 sessions (Events 10–15). These events were randomly assigned to be tested again 1 year (Events 10–12) or 2 years (Events 13–15) later. At Wave 3 (children were 6, 8, and 10 years of age), children were tested on three new events drawn at random from calendars maintained for the 4 months prior to the Wave 3 sessions (Events 16–18). The events were tested again 1 year later, at Wave 4.
Figure 1.
Schematic representation of the schedule on which events were tested.
At each wave, to initiate the interviews about the new events (those drawn from the calendar for that wave), the experimenter began by saying “Your mom/dad wrote down some things you’ve done and it’s my turn to talk with you about them. Some of the things happened more recently and some of them happened a long time ago. Since I wasn’t with you during these times, it’s up to you to tell me everything you can about each one. When you’re done, I’ll ask you a few questions.” For each event in turn, the experimenter then provided a general prompt in the form of a title for the event (taken from the participants’ calendars): “What can you tell me about X?” If this prompt failed to elicit a report, the experimenter provided additional cues, which were drawn from parents’ descriptions of the events (see Bauer & Larkina, 2016, for details). The experimenter encouraged participants to continue their reports using generic prompts, such as “Tell me more.” After the participant’s unstructured report had been exhausted, the experimenter asked a series of seven direct wh- memory probes about the event: who was there, what else did you do, where did you X, when did you X, why did you X, how did you X, and how did you feel about X? Participants were instructed to answer these questions even if they had already provided this type of information during the open-ended portion of the interview.
At each of Waves 1 and 2, the new events were randomly assigned to be tested after delays of 1 year (both waves), 2 years (both waves), or 3 years (Wave 1 only); at Wave 3, the new events were tested 1 year later (at Wave 4). For an event to be included in the analysis for long-term recall, at the initial assessment for that event, children were required to have provided at least two unique pieces of information about the event (see Bauer & Larkina, 2014b; Fivush & Schwarzmueller, 1998; Reese et al., 1993, for a similar criterion). When in the course of the initial interview, participants seemingly failed to meet this criterion, additional events were sampled from the calendars. As explained in detail in Bauer and Larkina (2016), although children always were tested for long-term recall of the designated number of three events from each of the previous waves of testing (see Figure 1), only events that were considered recalled at the initial assessment for that event were included in the analyses of long-term recall.
Test for long-term recall.
The delayed recall memory interviews were administered following the same procedure as described for the initial assessment, with the exception that after the general prompt in the form of a title for the event (the same titles used in the initial assessment, taken from the participants’ calendars), the experimenter provided a cue to aid in determination of the time-slice in which the event occurred, such as, “It happened when you were about 4 years old,” or “It was two years ago.” The balance of the interview proceeded as at the initial assessment.
Coding and Reliability and Data Reduction
The interviews were videotaped and later transcribed verbatim. All transcripts were reviewed for accuracy. All coding took place from the written transcripts. For all waves of data collection, all on-task contributions were parsed into propositional units defined as a unit of meaning that included subject-verb construction and either contained unique information about the events (e.g., “I went to the park”) or no content (e.g., “I don’t know”). Reliability of propositions was calculated on 25% of the transcripts, and averaged 94%, 91%, and 92% at Wave 1, 2, and 3, respectively (range 83–99%). The primary coders’ judgments were used in all analyses.
Initial assessment.
For events that met the criterion for recall (the participant provided at least two unique pieces of information), the narrative reports were coded for length, breadth, and coherence. For length and breadth, the entire memory interview was coded. Separate scores were calculated for the open-ended portion of the interview and for the entire memory report, including responses following additional cues and to wh- questions. The measure of length was the total number of propositions provided by the participant (reliability of coding indicated above). The measure of breadth reflects the completeness of the narrative. It was quantified by tallying the number of different narrative categories included in the report: information about (1) who participated (who), (2) the actions involved (what-action), (3) the objects involved (what-object), (4) when the event took place (when), (5) where the event took place (where), (6) why the event occurred or unfolded as it did (why), (7) description of physical attributes of the event (how-description), and (8) evaluation or subjective perception of the event (how-evaluation). The narrative categories are the same as those used in prior related research (e.g., Bauer & Larkina, 2014a, 2014b). For each event, children received 1 point for inclusion of a token reflective of the category, regardless of the number of tokens provided. The maximum narrative breadth score was 8. At Wave 1, all transcripts were coded by the same individual; for the reliability of coding, 25% of the transcripts were coded by an independent rater. Interrater agreement was 95% (range 91–99%). For Waves 2 and 3, the master coder trained four (Wave 2) and three (Wave 3) individuals, each of whom coded an approximately equal number of transcripts. Reliability was assessed on 25% of the transcripts and averaged 87% and 90% for Waves 2 and 3, respectively (range 80–97%). The primary coders’ judgments were used in all analyses.
To assess the coherence of the narrative reports, we used the Narrative Coherence Coding Scheme (NaCCs) developed by Reese et al. (2011). Narrative coherence is the overall quality of the narrative in terms of how well the story can be understood by a naive listener or reader. Because they are external to the participants’ own narrative, responses to wh- questions were not coded. As explained in Bauer and Larkina (2016), only the theme dimension (maintaining and elaborating on a topic) was used in analyses. For this dimension, we used all of the participants’ responses following the open-ended prompt (but before the wh- questions). The dimension was coded on a 4-point scale, from 0–3. A rating of 0 indicated that the narratives were substantially off-topic, a score of 1 was assigned to the minimally developed narratives, and narratives rated as 2 were substantially developed through evaluations, interpretations or causal links. Narratives coded as 3 included all of the previous characteristics with additional links to other autobiographical experiences and/or self-concept. At Wave 1 and 2, reliability of coding of the children’s reports originally was established in Morris et al. (2010), based on approximately 10% of the children’s transcripts. For present purposes, we recalculated reliability on the subset of the child sample included in the current research (including only those participants for whom we had all four waves of data collection). The intraclass correlation was .90. At Wave 3, one individual was trained on the Wave 1 and 2 transcripts to a criterion of .85 (intraclass correlation) and then coded all transcripts. Reliability was established on 25% of transcripts coded by independent rater. The intraclass correlation was .89. For purposes of analysis, we re-coded the theme dimension into two levels: low (ratings of 0 and 1) and high (ratings of 2 and 3). We collapsed the 4-point scale into two categories because of low percentages of narratives coded as 0 (8.5%) and as 3 (1.9%). With thematic coherence represented as two levels, roughly half of all memories received a low score (0–1; 58%) and half received a high score (2–3; 42%).
Long-term recall.
The procedure to determine whether a memory was recalled after the delay is provided in detail in Bauer and Larkina (2016). The first criterion was whether the child provided two or more unique pieces of information about the event. If fewer than two unique pieces of information were provided, the event was considered not-recalled. For reports that met this basal criterion, we further asked whether the narrative featured evidence that the memory was of a specific instance of an event, as opposed to a general category of events. Memories for which the narratives featured little or no evidence of recollection of a specific past event were considered not-recalled. Finally, for memories that featured specific details, we compared the 1-year, 2-year, or 3-year memory report with the initial assessment report of the nominally same event to determine whether the memory was the same as the target event, or a non-target event (examples of narratives that did and did not meet these criteria are provided in Bauer and Larkina, 2016).
For purposes of scoring and reliability, we made separate assessments for open-ended testing and across the open-ended, additional cue, and wh- phases of testing. The open-ended portion of the interview was that for which the experimenter provided only the title of the event and either the participant’s age or information about how long ago in time the event took place. We also scored the number of memories for which the narrative report met the above criterion when we considered not only the information provided in open-ended testing, but also the information provided after additional cuing and in response to the wh- questions. One highly experienced individual coded all transcripts. For purposes of estimating the reliability of coding, an independent rater coded 20% of the transcripts with approximately equal representation of waves and age groups. Interrater agreement on whether a memory was recalled in open-ended testing was 87% (intraclass correlation .85). Interrater agreement on whether a memory was recalled based on all sections of the interview (overall recall) was 83% (intraclass correlation .73). For both open-ended and overall recall, coding disagreements on the 20% of the transcripts that were double coded were resolved by discussion.
Potential correlates.
In addition to age and delay interval, potential correlates of long-term recall and of age-related differences therein were drawn from two categories: (a) the quality of the description of the event at the initial assessment, in terms of the length, breadth, and thematic coherence of the narrative; and (b) domain-general and memory-specific predictors. Table 2 provides abbreviated descriptions of the measures used to assess children’s (i) language comprehension; (ii) domain-general cognitive abilities (speed of processing, working memory, and attention); (iii) non-autobiographical story recall; (iv) use of memory strategies and metamemory understanding; and (v) memory for items and source of new factual knowledge (see Bauer & Larkina, 2019, for details). In Bauer and Larkina (2019), we used the domain-general and memory-specific measures to predict the quality of the narrative descriptions of the events children provided at the initial assessment of them. In the present manuscript, we report original analyses of the unique and combined variance in long-term recall (as opposed to the initial assessment) explained by the domain-general and memory-specific predictors, as well as the quality of the narrative description. As explained in Bauer and Larkina (2019), other tasks also were administered but because they did not load on any of the factors of cognitive ability and were not correlated with autobiographical narrative performance, they are not discussed.
Table 2.
Dependent Measures of Children’s Language (Panel a), Domain-general Cognition Functions (Panel b), Non-autobiographical Story Recall (Panel c), Deliberate and Strategic Remembering and Metamemory (Panel d), and Memory for Source (Panel e)
| Dependent Measure(s) | ||
|---|---|---|
| Panel a: Language (obtained at Wave 1 only) | ||
| WJ Language | The four subscales of the Test of Verbal Comprehension: Picture Vocabulary, Synonyms, Antonyms, and Verbal Analogy were combined into a total score. | |
| Panel b: Domain-general cognitive functions | ||
| WJ Visual Matching (VM) | VM1 at 4 years; VM2 at 5 years and older. Total score = number of rows (VM1–items) correctly completed in 3 minutes (VM1–2 mn) | |
| WJ Numbers Reversed | Children given a series of numbers (2–9) and instructed to repeat the numbers in reverse order. Total score (max = 30) | |
| WJ Pair Cancellation | Children asked to circle a pair of objects (ball and dog) when they appear in a certain order on the response sheet. The number of correct pairs completed in 3 minutes (max = 69) | |
| Panel c: Non-autobiographical story recall | ||
| Story Recall | Encoding (Session 1) | Experimenter read a story to the child using an illustrated book: “I’m going to tell you a story and show you some pictures. Next time you come, I’m going to see if you can tell the story back to me without looking at the pictures.” Different stories were used at each wave. No measures obtained at Session 1 |
| Retrieval (Session 2) | Child was shown the cover of the story book: “Last week, we read the story about_____. Tell me what happened in the story.” If child did not respond in free recall, prompted by reading 1st page of story. Number of parses, verbatim units, and gist units | |
| Panel d: Deliberate and strategic remembering and metamemory | ||
| Sort Recall | Experimenter presented 18 picture cards (6 from each of 3 categories; 15 at age 4: 5 from each of 3 categories), labeling and arranging them on the table. Instruction: “Study the cards and do whatever you want with them so you can remember them later.” After a 2 mn study period, the cards were removed; child was engaged in a 30 mn buffer activity, and then asked to “Recall as many cards in any order.” Proportion of items recalled. Strategic behavior: total number of organizational strategies (sorting, category naming, clustering) | |
| Metamemory | Children asked 8 metamemory questions regarding the sort-recall task (e.g., Did you do anything to help you remember?). Total metamemory score (max= 8) | |
| Panel e: Memory for source | ||
| Source Memory | Encoding (Session 1) | Children were presented with 12 novel facts by either a person or a puppet (on a computer monitor). Experimenter instructed children to watch the videos and learn the new facts because they would be asked about them later. No measures obtained at Session 1 |
| Retrieval (Session 2) | Children were asked to recall (a) what they learned (items), and (b) the source (person/puppet) that presented the item. Proportion of correctly recalled items and for correct items, proportion of correctly recalled sources | |
Notes. WJ = Woodcock-Johnson Tests of Cognitive Abilities (Woodcock, McGrew, & Mather, 2001); ARC = Adjusted Ratio of Clustering (Roenker, Thompson, & Brown, 1971)
As described in Bauer and Larkina (2019), at each wave of data collection, the domain-general and memory-specific measures were subjected to factor analysis. The resulting factors are reflected in Table 3. The factors were largely consistent across the waves. At all three waves, the factors that emerged were reflective of (a) non-autobiographical story recall, (b) domain-general cognitive abilities and item memory, and (c) deliberate and strategic remembering and metamemory.
Table 3.
Results of Factor Analyses for Each of Waves 1, 2, and 3
| Phase/ Predictor |
Factor | ||
|---|---|---|---|
| Factor 1 | Factor 2 | Factor 3 | |
| Wave 1 (72% variance) | |||
| Story Parse | .92 | ||
| Story Verbatim | .87 | ||
| Story Gist | .91 | ||
| WJ Visual Matching | .68 | ||
| WJ Numbers Reversed | .75 | ||
| WJ Pair Cancellation | .31 | .66 | .43 |
| Number Recalled | .33 | .35 | .61 |
| Deliberate Strategies | .39 | ||
| Metamemory | .38 | .44 | |
| Source Item Memory | .70 | ||
| Memory for Source | .48 | ||
| Eigenvalue | 5.65 | 1.39 | 1.14 |
| % Variance | 48.5% | 12.6% | 10.4% |
| Wave 2 (76% variance) | |||
| Story Parse | .92 | ||
| Story Verbatim | .92 | ||
| Story Gist | .94 | ||
| WJ Visual Matching | .30 | .81 | .36 |
| WJ Numbers Reversed | .79 | ||
| WJ Pair Cancellation | .83 | .32 | |
| Number Recalled | .64 | .52 | |
| Deliberate Strategies | .53 | ||
| Metamemory | .37 | .58 | |
| Source Item Memory | .58 | ||
| Memory for Source | .39 | ||
| Eigenvalue | 5.47 | 1.89 | 1.01 |
| % Variance | 49.7% | 17.2% | 9.14% |
| Wave 3 (74% variance) | |||
| Story Parse | .89 | ||
| Story Verbatim | .86 | ||
| Story Gist | .31 | .90 | |
| WJ Visual Matching | .88 | ||
| WJ Numbers Reversed | .74 | ||
| WJ Pair Cancellation | .82 | .33 | |
| Number Recalled | .60 | .55 | |
| Deliberate Strategies | .64 | ||
| Metamemory | .54 | .39 | |
| Source Item Memory | .60 | .30 | |
| Memory for Source | |||
| Eigenvalue | 5.52 | 1.61 | 1.04 |
| % Variance | 50.2% | 14.6% | 9.5% |
Analytic Approach
In the present research, the unit of analysis was the individual memory. The first major question was whether the likelihood that an individual memory would survive over time differed as a function of age cohort (i.e., age at the beginning of the study: 4-, 6-, 8-year-olds). The second major purpose of the present research was to determine the predictors of the survival of individual memories, including the characteristics of the memory reports provided at the time of the events (initial assessment) and children’s language and domain-general and memory-specific abilities. We also tested how the pattern of findings related to the length of the delay over which recall was tested (i.e., 1, 2, and 3 years). These purposes are best addressed with logistic multilevel modeling (e.g., Guo & Zhao, 2000; Morris et al., 2010; Peterson et al., 2014; Raudenbush & Bryk, 2002).
Logistic multilevel modeling permits examination of how predictors measured at various levels of data and cross-level interactions affect the outcome variable (e.g., Guo & Zhao, 2000). Specifically, it permits predictors at the level of the individual memory (e.g., narrative characteristics at the initial memory report) and at the level of the person (e.g., age cohort, child’s language comprehension). In addition, multilevel modeling takes into account the interdependency of multiple observations per person (e.g., up to nine memories per child for the 1-year delay), correcting for the biases in parameter estimates resulting from dependency of the observations (Wright, 1998). Moreover, multilevel modeling allows for variation across participants in the number of observations and also allows for missing data without excluding participants (Raudenbush & Bryk, 2002). Accordingly, we conducted logistic multilevel modeling using a PROC GLIMMIX procedure using SAS for Windows software (Version 9.4). The estimation method was based on residual pseudo-likelihood techniques for generalized linear mixed models. Model fit of reported models was assessed by subtracting the −2 restricted log pseudo-likelihood estimations of models and examining differences on a chi-square distribution with degree of freedom equaling the change in number of parameters (Singer & Willett, 2003). We also report fit statistics AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion), to estimate the relative quality of the models. Note that the reported fit statistics are Pseudo-AIC and Pseudo-BIC because they are based on pseudo-likelihoods and are not useful for comparing models that differ in their pseudo-data.
Analysis was performed separately for each level of delay (1-, 2-, 3-year). Fully unconditional models were estimated for each delay and both open-ended and overall recall to ensure variability between children (i.e., Level 2) and within children (i.e., Level 1). The intraclass correlation was calculated with an intercept variance of τ00 = 0.50, 0.44, and 0.44 for the open-ended recall model and τ00 = 0.54, 0.50, and 0.61 for the overall recall model for 1-, 2-, and 3-year delay models, respectively; the variance of the standard logistic distribution σ2, where σ2 = π2 / 3 = 3.29 (Guo & Zhao, 2000; Snijders & Bosker, 2004). The results of this analysis indicated that in open-ended recall, 13%, 12% and 12% of the variability in remembering was between children; in overall recall, 14%, 13% and 16% of the variability was between children for 1-, 2-, and 3-year delay, respectively. Thus the values for within-children variability ranged from 84–88%. The results of the unconditional models established that there was sufficient variability on both levels for further MLM analyses. The specific equations are provided in the Appendix.
Results
We present the results in two sections. We first characterize recall of events at the initial assessments, which took place at Waves 1, 2, and 3. The events queried at these assessments had taken place within 4 months of the laboratory visits. In the second section, we present the results of analyses predicting the survival of individual memories of the events after delay intervals of 1 year, 2 years, and 3 years. We present analyses for open-ended recall and for overall recall, including information provided in response to prompts and wh- questions.
Recall at the Initial Assessment
Overall, the total number of events available for recall across the waves was 1818 (101 subjects × 18 events/subject). Of these, 1767 events (97.19%) were coded as recalled and were included in further analysis (see above and Bauer & Larkina, 2016, for procedures for determining whether an event was recalled).
Characteristics of the narratives the children produced to describe the events at the initial assessments are reported in Table 4 (the values for the predictor measures are provided in Bauer & Larkina, 2019). Descriptive statistics were calculated on nine events per child for Wave 1, six events for Wave 2, and three events for Wave 3. As described in the Method section, at each of Waves 1 and 2, the new events were randomly assigned to be tested after delays of 1 year (both waves), 2 years (both waves), or 3 years (Wave 1 only). Preliminary analyses indicated that at the initial assessment, there were no chance differences between the events randomly assigned to the different waves of later testing. Specifically, for each narrative characteristic, we conducted a multilevel model with Delay as a Level 1 (the individual memory report) predictor. Delay was not a significant predictor in any model: for Wave 1 events, Fs(2, 197) = 1.51, 0.90, and 0.56, ps > .20; for Wave 2 events, Fs(1, 100) = 0.46, 1.00, and 0.78, ps > .30, for length, breadth, and thematic coherence (respectively).
Table 4.
Descriptive Statistics of Narrative Measures at Initial Assessments
| Study Phase | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| Wave 1 | Wave 2 | Wave 3 | |||||||
| Cohort | 4-year-olds | 6-year-olds | 8-year-olds | 4-year-olds | 6-year-olds | 8-year-olds | 4-year-olds | 6-year-olds | 8-year-olds |
| Age | 4 years | 6 years | 8 years | 5 years | 7 years | 9 years | 6 years | 8 years | 10 yrs |
| M (SD) | M (SD) | M (SD) | M (SD) | M (SD) | M (SD) | M (SD) | M (SD) | M (SD) | |
| a Length in propositions | 24.41 (13.31) |
26.83 (13.21) |
37.23 (21.36) |
32.83 (15.67) |
37.68 (18.17) |
57.68 (35.65) |
34.02 (20.33) |
43.51 (22.17) |
57.20 (35.98) |
| a Breadth (max=8.0) | 6.16 (1.72) |
6.92 (1.30) |
7.51 (0.84) |
6.50 (1.50) |
7.20 (1.12) |
7.67 (0.82) |
6.65 (1.51) |
7.51 (0.74) |
7.74 (0.80) |
| b Thematic Coherence | 1.07 (0.74) |
1.25 (0.59) |
1.56 (0.62) |
1.26 (0.67) |
1.55 (0.58) |
1.88 (0.48) |
1.05 (0.05) |
1.24 (0.49) |
1.38 (0.57) |
Note.
For length and breadth, the entire memory interview was coded;
for thematic coherence, only the open-ended portion of the interview was coded.
Long-term Recall as a Function of Cohort and Delay
We conducted separate analyses for each level of delay (1-, 2-, and 3-year). For each delay level, we first present results of the likelihood of children’s recall of the memory during open-ended testing, and then for their overall recall (open-ended plus responses to cues and wh- questions); open-ended and overall recall were tested in separate analyses. For each delay, we tested three models using a multi-level modeling approach. Bonferroni adjustments were applied for multiple comparisons in these models. We report fixed effects (odds ratio) in Tables 5, 6, and 7, for delay intervals of 1, 2, and 3 years, respectively.
Table 5.
Fixed Effects (Odds Ratio) for Multilevel Models of Long-term Recall for One-year Delay for Open-ended Recall (Panel a) and Overall Recall (Panel b)
| Model | |||
|---|---|---|---|
| Model 1 (Basic) |
Model 2 (Partially adjusted) |
Model 3 (Fully adjusted) |
|
| Odds Ratio (95% CI) | Odds Ratio (95% CI) | Odds Ratio (95% CI) | |
| Panel a: Open-ended recall | |||
| Age Cohort | |||
| 6 year vs 4 year | 1.96** (1.18–3.25) | 1.87** (1.14–3.06) | 1.32 (0.65–2.69) |
| 8 year vs 4 year | 3.89*** (2.25–6.73) | 3.41*** (1.97–5.88) | 1.74 (0.55–5.56) |
| 8 year vs 6 year | 1.98** (1.13–3.49) | 1.82* (1.04–3.14) | 1.32 (0.62–2.78) |
| Wave at Long-term Recall | |||
| Wave 4 vs Wave 3 | 1.59* (1.02–2.49) | 1.80** (1.15–2.839) | 1.78** (1.13–2.80) |
| Wave 4 vs Wave 2 | 2.20*** (1.41–3.44) | 2.28*** (1.47–3.52) | 2.28*** (1.47–3.53) |
| Wave 3 vs Wave 2 | 1.38 (0.91–2.11) | 1.26 (0.82–1.94) | 1.28 (0.83–1.97) |
| Thematic Coherence | |||
| High vs Low | 1.55** (1.11–2.15) | 1.46* (1.05–2.04) | |
| Language | 1.02 (0.98–1.05) | ||
| Factor 1 | 1.20 (0.96–1.49) | ||
| Factor 2 | 1.03 (0.79–1.35) | ||
| Factor 3 | 1.18 (0.93–1.51) | ||
| −2LL | 3915.26 | 3881.38 | 3905.36 |
| −2LL change | 33.88*** | 23.98*** | |
| Pseudo-AIC | 3917.26 | 3883.38 | 3907.36 |
| Pseudo-BIC | 3919.87 | 3886.00 | 3909.98 |
| Panel b: Overall recall | |||
| Age Cohort | |||
| 6 year vs 4 year | 2.60*** (1.39–4.86) | 2.32*** (1.27–4.23) | 1.89 (0.76–4.72) |
| 8 year vs 4 year | 4.42*** (2.14–9.11) | 3.64*** (1.75–7.58) | 2.53 (0.54–11.75) |
| 8 year vs 6 year | 1.70 (0.77–3.74) | 1.57 (0.72–3.42) | 1.33 (0.48–3.73) |
| Wave at Long-term Recall | |||
| Wave 4 vs Wave 3 | 1.42 (0.78–2.59) | 1.33 (0.77–2.30) | 1.35 (0.78–2.34) |
| Wave 4 vs Wave 2 | 1.89^ (0.997–3.60) | 2.25** (1.27–3.99) | 2.32** (1.30–4.15) |
| Wave 3 vs Wave 2 | 1.33 (0.69–2.56) | 1.69 (0.92–3.11) | 1.72 (0.93–3.19) |
| Thematic Coherence | |||
| High vs Low | 2.16*** (1.36–3.40) | 1.98** (1.24–3.15) | |
| Language | 1.01 (0.97–1.05) | ||
| Factor 1 | 1.16 (0.86–1.56) | ||
| Factor 2 | 0.89 (0.62–1.28) | ||
| Factor 3 | 1.40* (1.002–1.96) | ||
| −2LL | 4402.04 | 4396.27 | 4431.67 |
| −2LL change | 5.77* | 35.40*** | |
| Pseudo-AIC | 4404.04 | 4398.27 | 4433.67 |
| Pseudo-BIC | 4406.65 | 4400.89 | 4436.28 |
Note:
p <.10;
p < .05;
p < .01;
p < .001.
Fit Statistics: −2LL = −2 Res Log Pseudo-Likelihood, Pseudo-AIC = Akaike information criterion; Pseudo-BIC = Bayesian information criterion. Bonferroni adjustments were applied for multiple comparisons.
Table 6.
Fixed Effects (Odds Ratio) for Multilevel Models of Long-term Recall for Two-year Delay for Open-ended Recall (Panel a) and Overall Recall (Panel b)
| Model | |||
|---|---|---|---|
| Model 1 (Basic) |
Model 2 (Partially adjusted) |
Model 3 (Fully adjusted) |
|
| Odds Ratio (95% CI) | Odds Ratio (95% CI) | Odds Ratio (95% CI) | |
| Panel a: Open-ended recall | |||
| Age Cohort | |||
| 6 year vs 4 year | 1.45 (0.78–2.70) | 1.42 (0.76–2.65) | 1.13 (0.45–2.86) |
| 8 year vs 4 year | 2.48** (1.32–4.67) | 2.30** (1.18–4.50) | 1.44 (0.31–6.82) |
| 8 year vs 6 year | 1.71 (0.91–3.21) | 0.95 (0.85–3.11) | 1.28 (0.49–3.37) |
| Wave at Long-term Recall | |||
| Wave 4 vs Wave 3 | 1.42* (1.01–2.00) | 1.33 (0.93–1.89) | 1.34 (0.94–1.90) |
| Thematic Coherence | |||
| High vs Low | 1.45^ (0.99–2.13) | 1.46^ (0.99–2.17) | |
| Language | 1.04^ (0.99–1.09) | ||
| Factor 1 | 1.05 (0.82–1.35) | ||
| Factor 2 | 0.90 (0.57–1.42) | ||
| Factor 3 | 0.80 (0.57–1.10) | ||
| −2LL | 2543.22 | 2515.21 | 2536.22 |
| −2LL change | 28.01*** | 21.01*** | |
| Pseudo-AIC | 2545.22 | 2517.21 | 2538.22 |
| Pseudo-BIC | 2547.84 | 2519.83 | 2540.83 |
| Panel b: Overall recall | |||
| Age Cohort | |||
| 6 year vs 4 year | 1.42 (0.74–2.74) | 1.33 (0.68–2.57) | 0.90 (0.33–2.46) |
| 8 year vs 4 year | 2.17* (1.09–4.34) | 1.84 (0.88–3.82) | 0.82 (0.15–4.50) |
| 8 year vs 6 year | 1.52 (0.75–3.09) | 1.39 (0.67–2.88) | 0.91 (0.31–2.69) |
| Wave at Long-term Recall | |||
| Wave 4 vs Wave 3 | 1.52* (1.06–2.19) | 1.48* (1.02–2.14) | 1.48* (1.02–2.15) |
| Thematic Coherence | |||
| High vs Low | 1.73** (1.15–2.61) | 1.72* (1.13–2.62) | |
| Language | 1.04 (0.99–1.09) | ||
| Factor 1 | 1.15 (0.87–1.51) | ||
| Factor 2 | 1.06 (0.64–1.73) | ||
| Factor 3 | 0.88 (0.62–1.25) | ||
| −2LL | 2593.61 | 2574.60 | 2594.59 |
| −2LL change | 19.01*** | 19.99*** | |
| Pseudo-AIC | 2595.61 | 2576.60 | 2596.59 |
| Pseudo-BIC | 2598.22 | 2579.22 | 2599.20 |
Note:
p <.10;
p < .05;
p < .01;
p < .001.
Fit Statistics: −2LL = −2 Res Log Pseudo-Likelihood, Pseudo-AIC = Akaike information criterion; Pseudo-BIC = Bayesian information criterion. Bonferroni adjustments were applied for multiple comparisons.
Table 7.
Fixed Effects (Odds Ratio) for Multilevel Models of Long-term Recall for Three-year Delay for Open-ended Recall (Panel A) and Overall Recall (Panel b)
| Model | |||
|---|---|---|---|
| Model 1 (Basic) |
Model 2 (Partially adjusted) |
Model 3 (Fully adjusted) |
|
| Odds Ratio (95% CI) | Odds Ratio (95% CI) | Odds Ratio (95% CI) | |
| Panel a: Open-ended recall | |||
| Age Cohort | |||
| 6 year vs 4 year | 2.26 (0.88–5.81) | 2.41^ (0.90–6.48) | 2.72 (0.69–10.78) |
| 8 year vs 4 year | 2.46^ (0.94–6.44) | 1.92 (0.68–5.36) | 2.22 (0.18–26.93) |
| 8 year vs 6 year | 1.09 (0.46–2.58) | 0.80 (0.31–2.04) | 0.81 (0.15–4.40) |
| Thematic Coherence | |||
| High vs Low | 2.66** (1.43–4.95) | 3.03** (1.56–5.89) | |
| Language | 0.99 (0.93–1.05) | ||
| Factor 1 | 1.01 (0.64–1.59) | ||
| Factor 2 | 1.23 (0.59–2.59) | ||
| Factor 3 | 0.66 (0.39–1.14) | ||
| −2LL | 1299.64 | 1298.82 | 1311.02 |
| −2LL change | 0.82 | 12.20* | |
| Pseudo-AIC | 1301.64 | 1300.82 | 1313.02 |
| Pseudo-BIC | 1304.25 | 1303.43 | 1315.62 |
| Panel b: Overall recall | |||
| Age Cohort | |||
| 6 year vs 4 year | 2.43* (1.05–5.61) | 2.57* (1.09–6.03) | 2.22 (0.69–10.78) |
| 8 year vs 4 year | 3.41** (1.43–8.13) | 2.71* (1.10–6.66) | 2.22 (0.18–26.93) |
| 8 year vs 6 year | 1.41 (0.61–3.25) | 1.05 (0.44–2.55) | 0.81 (0.15–4.40) |
| Thematic coherence | |||
| High vs Low | 2.51** (1.41–4.45) | 3.03** (1.56–5.89) | |
| Language | 0.99 (0.93–1.05) | ||
| Factor 1 | 1.01 (0.64–1.59) | ||
| Factor 2 | 1.23 (0.59–2.59) | ||
| Factor 3 | 0.66 (0.39–1.14) | ||
| −2LL | 1244.01 | 1242.15 | 1259.92 |
| −2LL change | 1.86 | 17.77** | |
| Pseudo-AIC | 1246.01 | 1244.15 | 1261.92 |
| Pseudo-BIC | 1248.62 | 1246.76 | 1264.53 |
Note:
p <.10;
p < .05;
p < .01;
p < .001.
Fit Statistics: −2LL = −2 Res Log Pseudo-Likelihood, Pseudo-AIC = Akaike information criterion; Pseudo-BIC = Bayesian information criterion. Bonferroni adjustments were applied for multiple comparisons.
Model 1 tested whether the likelihood of recall of an individual memory differed as a function of age cohort (i.e., the children’s ages at the beginning of the study: 4-, 6-, 8-year-olds) and Wave at long-term recall (for delay of 1 year and 2 years only), as well as a function of the interaction of the factors. Because in all models, the interaction effect of Cohort x Wave was not significant, we removed it from all further models.
Model 2 and Model 3 were conducted to identify predictors beyond the cohort to which the child belonged and the wave of testing, by including other potential predictor variables. In Model 2, we included variables that captured the quantity and quality of the narratives children produced about the events at the initial assessments. Potential predictors were the number of propositions (as a measure of narrative length), narrative breadth (as a measure of the completeness of the narrative), and thematic coherence of the narrative. Previous research has identified thematic coherence as a reliable predictor of long-term recall, therefore we always included thematic coherence in Model 2. Preliminary analysis revealed that none of the Theme x Cohort and Theme x Wave interactions was significant (ps >.75). Therefore the interaction terms were not included in the final models. To decide whether narrative length and breadth should be included in Model 2, we conducted preliminary analyses. Specifically, in separate models, we tested whether these variables were significant predictors when cohort and wave of testing (delay of 1 year and 2 years only) were already included. When either length or breadth was a significant predictor, we then tested whether it remained a significant predictor with thematic coherence also included.
Model 3 included the factor scores derived from the initial wave of testing (i.e., Wave 1, 2 and 3), as well as children’s verbal comprehension scores from the time of initial enrollment in the study. All of these potential predictors were entered together in the final fully adjusted models.
One-year delay.
Each child provided up to nine memories to be tested for long-term recall after 1 year. As reflected in Figure 1, three of the nine memories were initially assessed at Wave 1 (represented by Events 1–3), three were initially assessed at Wave 2 (represented by Events 10–12), and three were initially assessed at Wave 3 (represented by Events 16–18). Thus all nine memories were tested after the same delay between initial assessment and long-term recall. The difference is the wave of the study at which the events were experienced (and thus the age of the child at the time of experience) and the wave of the study at which long-term recall was assessed. A total of 892 memories were included in the analyses (98% out of 909 possible events; 17 events were excluded because they were coded as not-recalled at the initial assessment).
Open-ended recall.
The percentages of memories recalled in open-ended testing after the delay interval of 1 year as a function of age cohort and wave at long-term recall (represented by children’s age at the time of recall) are provided in Figure 2, Panel a. As is apparent from the figure, for each age cohort, there were steady increases in the numbers of memories recalled after 1 year, as a function of increases in the children’s ages at the time of long-term recall.
Figure 2.
Percentage of events recalled in open-ended testing after delays of 1 year (Panel a), 2 years (Panel b), and 3 years (Panel c), by children in the 4-, 6-, and 8-year-old cohorts.
Note: Percentage recalled is calculated at the level of the event; after 3-year delay, children were 7-, 9-, and 11-year-olds at long-term recall for 4-, 6-, and 8-year-old cohorts, respectively.
Model 1 (see Table 5, Panel a) indicated that there were reliable cohort differences in the likelihood that an individual memory would survive to be recalled after 1 year, F(2, 785) = 17.87, p < .001. Across all waves, the memories of children in the 8-year-old cohort were almost 2 and 4 times more times more likely to be recalled than the memories of children in both the 6- and 4-year-old cohorts, respectively. In addition, the memories of children in the 6-year-old cohort were almost 2 times more likely to be recalled than the memories of children in the 4-year-old cohort. The likelihood that an individual memory would be recalled after the 1-year delay also differed as a function of the wave of testing, F(2, 785) = 9.12, p < .001. Memories were more likely to be recalled when children originally experienced the events at the later waves (when they were older) relative to the earlier waves. The effects were reliably different for events experienced at the first versus the third wave (and tested at Waves 2 and 4) and events experienced at the second versus the third wave (and tested at Waves 3 and 4). Importantly for present purposes, the interaction of Cohort x Wave was not statistically significant (p = .71). This implies that the effect of cohort was similar across waves. In other words, the rate of increase in retention was the same for the children in the 4-, 6-, and 8-year cohorts; the slopes or rates of change did not differ across the second half of the first decade of life.
For Model 2, the preliminary analysis indicated that the number of propositions (narrative length) was not a significant predictor in the model. Although breadth was a significant predictor (p < .05) when added to the basic model, it became non-significant when thematic coherence was included. The final Model 2 thus included only thematic coherence as a Level 1 predictor. Individual events that were initially described in narratives that were more thematically coherent were almost 1.5 times more likely to be recalled after the 1-year delay (odds ratios = 1.55). Model 3 included the factor scores and children’s verbal comprehension scores. These factors did not make significant contributions to the models. However, inclusion of them in the model rendered cohort non-significant. In sum, the final fully adjusted model indicated that individual memories were more likely to be recalled when children originally experienced the events at the later waves (when they were older) and when they were initially narrated with higher levels of thematic coherence. Importantly, after controlling for differences in children’s language comprehension and other cognitive abilities, cohort became a non-significant predictor in recall of memories after 1-year delays.
Overall recall.
The percentages of memories recalled with the additional support of prompts and wh- questions as a function of age cohort and wave at long-term recall (represented by children’s age at the time of recall) are provided in Figure 3, Panel a. Not surprisingly, levels of recall were (nominally) higher when children received additional support for recall. Age-related increases in the numbers of memories recalled after the delays nevertheless are apparent. The basic model indicated reliable cohort differences, F(2, 785) = 14.56, p < .001 (see Table 5, Panel b). The memories of 6- and 8-year-olds were roughly 2.5 and 4 times more likely to be recalled than the memories of 4-year-olds. There was no difference between 6- and 8-year-olds. With the addition of prompts and wh- questions, there were not pronounced differences in the likelihood of recall of the memories as a function of the wave of testing. Children were more likely to recall memories encoded at the third than the first wave (and tested at Waves 4 and 2); the effect fell just below the level of statistical significance (p = .051). Importantly, as was the case for open-ended recall, the interaction of Cohort x Wave was not statistically significant (p = .43).
Figure 3.
Percentage of events recalled across all phases of testing (openended, prompted, and in response to wh-questions) after delays of 1 year (Panel a), 2 years (Panel b), and 3 years (Panel c), by children in the 4-, 6-, and 8-year-old cohorts.
Note: Percentage recalled is calculated at the level of the event; after 3-year delay, children were 7-, 9-, and 11-year-olds at long-term recall for 4-, 6-, and 8-year-old cohorts, respectively.
The results of Model 2 (Table 5, Panel b) indicated that thematic coherence was the only narrative characteristic that predicted overall recall (number of propositions and narrative breadth, ps > .12). Similarly to open-ended recall, in the fully adjusted Model 3, cohort no longer predicted the survival of individual memories over the 1-year delay, and children’s verbal comprehension scores were not a significant predictor. However, Factor 3 (deliberate and strategic remembering) contributed significant variance to the models. Memories from children with higher deliberate and strategic memory scores were 1.4 times more likely to be recalled than memories from children with low scores. No other factors were significant predictors of recall. In sum, after additional support in the form of prompts and wh- questions, memories were more likely to be recalled when they were initially narrated with higher thematic coherence, were tested for recall at Wave 4, and were from children who had higher deliberate and strategic memory scores.
Two-year delay.
Each child provided up to six memories to be tested for long-term recall after 2 years. Three of the memories initially were assessed at Wave 1 (long-term recall at Wave 3) and the other three memories initially were assessed at Wave 2 (long-term recall at Wave 4; see Figure 1, represented by Events 4–6 and 13–15, respectively). A total of 587 memories were included in the analyses (97% out of 606 possible events; 19 events were excluded because they were coded as not-recalled at the initial assessment). The analytic approach was the same as used to predict recall after 1-year delay (see above).
Open-ended recall.
The percentages of memories recalled in open-ended testing after the delay interval of 2 years as a function of age cohort and wave at long-term recall (represented by children’s age at time of recall) are provided in Figure 2, Panel b. Results of the basic Model 1 (Table 6, Panel a) indicated that both predictors were significant: cohort, F(2, 483) = 5.96, p < .01; and wave, F(1, 483) = 4.06, p < .05. Across waves, the memories of children in the 8-year-old cohort were roughly 2.5 times more likely to be recalled than the memories of children in 4-year-old cohort. Memories were more likely to be recalled when children originally experienced the events at Wave 2 (when they were older) relative to Wave 1. Although the graph in Figure 2, Panel b suggests differences in the rates of changes across cohorts, the interaction of Cohort x Wave was not statistically significant, F(2, 483) = 2.30, p = .10.
Preliminary analyses of the narrative characteristics of the initial reports showed that only thematic coherence approached significance (p = .059) as a predictor when added to the basic model (Model 2 in Table 6, Panel a). In the final fully adjusted model (Model 3), none of the predictors was significant; thematic coherence and language comprehension approached significance (ps = .057 and .067, respectively). In sum, after delays of 2 years, recall of memories in the open-ended phase of the interview did not depend either on children’s age (measured by cohort or wave of testing) or on children’s performance on cognitive or other tasks.
Overall recall.
The percentages of memories recalled with the additional support of prompts and wh- questions as a function of age cohort and wave are provided in Figure 3, Panel b. The results of multilevel models are in Table 6, Panel b. In the basic model, cohort, F(2, 483) = 3.64 and wave, F(1, 483) = 5.17 were significant predictors (ps < .05). The memories of 8-year-olds were roughly 2 times more likely to be recalled than the memories of 4-year-olds. Memories that were initially assessed at Wave 2 (and tested at Wave 4) were 1.5 times more likely to be recalled than memories that were first reported at Wave 1 (and tested at Wave 3). The interaction of Cohort x Wave was not statistically significant, F(2, 483) = 1.34, p = .26.
After preliminary analysis of the narrative measures, only thematic coherence was included in Model 2. In the model including thematic coherence, wave remained a significant predictor but cohort became non-significant (p = .14). In the final fully adjusted model, wave and thematic coherence were the only significant predictors: memories initially reported at Wave 2 were more likely to be recalled after 2 years than memories initially reported at Wave 1 (when the children were younger), and memories initially narrated with higher thematic coherence were more likely to be remembered relative to memories with lower thematic coherence.
Three-year delay.
Each child provided up to three memories at Wave 1 to be tested for long-term recall after 3 years at Wave 4 (Figure 1, represented by Events 7–9). A total of 288 memories were included in analysis (95% out of 303 possible events; 15 events were excluded because they were coded as not-recalled at the initial assessment).
Open-ended recall.
The percentages of memories recalled in open-ended testing after the delay interval of 3 years as a function of age cohort are provided in Figure 2, Panel c. Because all memories available for the 3-year recall test came from Wave 1, only cohort was included in the basic Model 1 (Table 7, Panel a). The effect of cohort fell just below the conventional level of statistical significance, F(2, 188) = 2.98, p = .053. Based on the results of preliminary analysis only thematic coherence was included in Model 2: memories narrated with high thematic coherence at the initial assessment were 2.6 more likely to be remembered after 3 years than memories narrated with low thematic coherence. In Model 3, thematic coherence was the only significant predictor.
Overall recall.
The percentages of memories recalled with the additional support of prompts and wh- questions as a function of age cohort and wave are provided in Figure 3, Panel c. The results of multilevel models are in Table 7, Panel b. In Model 1, cohort was a strong predictor of overall recall after a delay of 3 years, F(2, 188) = 6.29, p < .01: the memories of 6- and 8-year-olds were 2.4 and 3.4 times more likely to be recalled than memories of 4-year-olds; there was no difference between 6- and 8-year-olds. In Model 2, thematic coherence and cohort were significant predictors. After adding children’s language comprehension scores and factors in Model 3, cohort became non-significant. Thus, in the fully adjusted model, only thematic coherence predicted overall long-term recall after the delay of 3 years.
Discussion
The present research had two major purposes. The first was to examine patterns of change in remembering and forgetting over delays of 1, 2, and 3 years, by children 4 to 11 years of age. The second major purpose of the present research was to test the unique and combined variance in long-term recall contributed by measures of the narrative characteristics of memory reports at the time of the events (initial assessment) and by both domain-general and memory-specific abilities, and how the patterns of prediction change over the second half of the first decade of life. The vehicle was a 4-year prospective study that permitted virtually continuous sampling across the entire age period. Our approach was to test children’s recall of a corpus of events that had been experienced prior to each of Waves 1, 2, and 3 of the study. The initial assessments established that the children had encoded the events and that they were remembered by the children, at least over the brief delay since they had occurred (within 4 months). We then tested unique subsets of the memories after delay intervals of 1, 2, and 3 years, to determine which of the individual memories had survived (see Figure 1 for a schematic diagram of the design).
Patterns of Change in Remembering and Forgetting
Overall, the analyses revealed a pattern of age-related increases in the likelihood of survival of memories. The pattern was apparent in both open-ended and overall recall (including information provided in response to prompts and wh- questions). The pattern was apparent when comparisons were between subjects, such that the memories of children in the older cohorts were more likely to survive than the memories of children in the younger cohorts. Differences were especially pronounced for memories of children in the 8-year-old relative to the 4-year-old cohort. The pattern also was apparent when comparisons were within subjects, such that individual memories were more likely to survive when the events were experienced when children were older than when the children were younger. Critically, the between- (cohort) and within- (wave) subjects variables did not interact. Thus the rate of increase in retention of individual memories was the same for the children in the 4-, 6-, and 8-year cohorts; the slopes or rates of change did not differ across the second half of the first decade of life.
Predictors of the Survival of Memories over Time
The second major purpose of the present research was to test the variance in long-term recall contributed by measures of the narrative characteristics of memory reports at the time of the events (initial assessment) and by both domain-general and memory-specific abilities. The results of the analyses most relevant to this question were strikingly consistent. At all of the delay intervals, memories originally narrated with higher degrees of thematic coherence (scores of 2 and 3) were more likely to be recalled than memories originally narrated with lower degrees of thematic coherence (scores of 0 and 1). The effect was apparent in open-ended testing, although for events tested after delays of 2 years, it fell just below the conventional level of statistical significance. The effect remained apparent even when recall was supported by additional prompts and wh- questions. In contrast, neither the length of the original narrative (measured in terms of the number of propositions included in the report) nor the completeness of the original narrative (measured in terms of the number of wh- categories included) contributed significant variance to prediction of the survival of memories. The overall pattern is similar to the findings of Peterson and colleagues (2014). In that study, narrative coherence was a significant predictor of the likelihood of survival of early memories over time, whereas the length of children’s memory reports was not.
The critical importance of the narrative coherence of the original memory report for the long-term survivability of event memories in this age period was further reinforced by the statistical models that included children’s language skills and both domain-general abilities (speed of processing, working memory, sustained attention) and memory-specific abilities (non-autobiographical story recall, deliberate and strategic remembering, metamemory, memory for source). In only one case did one of these variables contribute to prediction of the survival of memories over the long term. That is, the measure of deliberate and strategic remembering contributed variance to overall recall of events tested after 1-year delays. This effect can be interpreted to suggest that children with greater control over their mnemonic abilities were better able to take advantage of the prompts and wh- questions provided by the experimenters (see Roebers, 2014, for discussion of developmental change in deliberate and strategic remembering).
The paucity of relations with measures of language skills and both domain-general and memory-specific abilities stands in sharp contrast to the findings of Bauer and Larkina (2019). In that article, we used the same measures to predict the narrative qualities of the children’s initial reports of their memories at each of Waves 1, 2, and 3. That is, we used the measures to predict the length, breadth, and thematic coherence of the children’s initial memory reports. We found that the measures of domain-general cognitive abilities, of deliberate and strategic remembering and metamemory, and of non-autobiographical story recall all contributed variance in prediction of both the breadth and thematic coherence of autobiographical narratives in free recall. Children’s own language abilities predicted significant variance in their narrative reports when it was the only factor in the model. As above, with the single exception of deliberate and strategic remembering, in overall recall, for the 1-year delays, none of these measures contributed significant variance to prediction of the survival of memories over time.
The findings of the present research lead to the conclusion that whereas a variety of memory-specific and domain-general cognitive abilities contribute to the quality of narrative description of a recent past event, only the thematic coherence of the narrative report predicts the long-term survival of the memory—at least as measured in the present study. As discussed in Bauer and Larkina (2019), general cognitive and memory-specific abilities may contribute to the quality of narrative descriptions in a number of ways. For example, non-autobiographical story recall shares demands with autobiographical memory reports, such as the requirement to detail who did what to whom, and when. More general cognitive abilities, such as speed of processing, may contribute to the efficiency of encoding and perhaps even consolidation. These skills and abilities are essential to support formation of an organized memory trace, expressed through a thematically coherent autobiographical narrative. Yet the findings of the present research suggest that once the organized memory trace is instantiated and becomes integrated into the child’s developing autobiography (in the case of scores of 3 on the thematic coherence scale), it is that representation—rather than the skills and abilities that contributed to its formation—that supports later retrieval. The finding of the importance of thematic coherence as a predictor of long-term recall is consistent with prior reports (e.g., Baker-Ward, Ornstein, & Principe, 1997; Peterson et al., 2014; Reese et al., 2011). At the same time, there also is evidence from Peterson and colleagues that whereas thematic coherence was predictive after 2 years (Peterson et al., 2014), after a longer delay than assessed in the present research (i.e., after 8 years), it no longer had significant predictive utility. In contrast, contextual coherence (orienting an event in time and place) was predictive after both 2 and 8 years (Peterson, Hallett, & Compton-Gillingham, 2018). In the present research (as well as prior reports from this sample: Bauer & Larkina, 2016, 2019), we were not able to assess contextual coherence because the experimenters often provided this information in their prompts for recall). Sorting out the relative predictive utility of different dimensions of narrative coherence, over various retention intervals, will require further research.
Implications for Conceptualizations of Childhood Amnesia
The findings of the present research are important for our conceptualization of childhood amnesia—the relative paucity among adults and older children of memories of events from the early years of life (e.g., Pillemer & White, 1989). In both the adult and developmental literatures, it is suggested that with the “offset” of childhood amnesia at roughly 5 to 7 years of age, there are marked changes in autobiographical memory (e.g., Nelson & Fivush, 2004; Perner & Ruffman, 1995; Tulving, 2002; Wetzler & Sweeney, 1986; Wheeler, 2000). At this time, children emerge from a period during which few or no autobiographical memories were formed, into a period in which there is a steady accumulation of autobiographical memories that can be retained over long periods of time. This conceptualization implies that there should be different patterns of remembering and forgetting early versus later in the 4- to 11-year age range. Early in this period, few memories should be formed and they should be highly vulnerable to forgetting. In contrast, later in this period, more memories should be formed and they should be more resistant to forgetting. In contrast to this expectation, we observed consistent patterns across the entire age range of 4 to 11 years. That is, there was no discontinuity in the slope of forgetting.
The pattern of results in the present research is consistent with an alternative conceptualization of childhood amnesia—characterization of the amnesia in terms of an “onset,” rather than an offset (Bauer, 2014, 2015). In this view, the ability to form and retain autobiographical memories emerges early and undergoes gradual development (for similar views, see for example Harley & Reese, 1999; Howe & Courage, 1997; Peterson & Rideout, 1998; Tustin & Hayne, 2010). The relative paucity of memories of early life events accessible to recall by older children and adults is not because few or none were formed but because early-formed memories are forgotten at an accelerated rate, relative to later-formed memories (e.g., Bauer, et al., 2007; Bauer & Larkina, 2014b). Indeed, throughout the first decade of life, the distribution of autobiographical memories obtained from retrospective studies is better fit by the exponential function (Bauer & Larkina, 2014b) whereas in adulthood, it is better fit by the power function (Bauer & Larkina, 2014b; Wetzler & Sweeney, 1986). As discussed in Bauer (2015), these functions differ both in terms of the initial rate of forgetting (steeper at younger ages) and in terms of the number of memories lost from the corpus with each unit of time (more memories lost at younger ages; see Bauer, 2015, Figure 5). This helps to explain observations that older children have older earliest memories, relative to younger children (Peterson et al., 2005; Tustin & Hayne, 2010), and that over time, the age of earliest memory that children report edges upward, to older ages (e.g., Peterson et al., 2011; Wang & Peterson, 2016). Both of these patterns could result from a pool of memories that shrinks over time.
For present purposes, the important point is that the change in rate of forgetting of memories from childhood is gradual over the first decade of life. This finding from the present prospective study complements previous research using a retrospective method that also indicates gradual changes over the first decade of life and even into young adulthood (Bauer & Larkina, 2014b). In the present research, this was evidenced in comparable slopes of forgetting of children 4 to 6 years and children 7 to 10 years. The continuity was revealed due to the dense sampling and prospective tracking of memories across the “boundary” of childhood amnesia. Prospective tracking is especially important for eliminating concerns that sometimes stem from retrospective studies, regarding when events occurred, and whether the subject of the report is the individual’s own memory, or is a trace resulting from a “family story,” for example (see Larkina & Bauer, 2012, for a controlled investigation of family stories). Such speculation has led to dismissal of adults’ reports of early memories they date from the age of 2 to 3 years, on the grounds that they are “fictional” (Akhtar, Justice, Morrison, & Conway, 2018). As argued by Bauer, Baker-Ward, Krøjgaard, Peterson, and Wang (in press), such dismissal is unfounded, based on converging evidence that autobiographical memory emerges early and undergoes gradual development. The present research provides an especially compelling illustration of both of these trends.
Limitations and Directions for Future Research
The present research is not without limitations. One limitation is that the pool of memories available to be included in the analyses of long-term recall varied as a function of the delay interval, such that the pool of memories for the 1-year, 2-year, and 3-year delays was 909, 606, and 303 (respectively). As a result, the power and thus the sensitivity of the predictive models was diminished at the longer relative to the shorter delays. Second, although we measured the breadth of children’s narratives, we did not conduct a full content analysis of the descriptions children provided of their memories. It is possible that a more fine-grained analysis of the amount of detail—as opposed to narrative breadth alone—would suggest a larger predictive role for this aspect of narrative, relative to that estimated in the present research. Third, we did not assess aspects of the events that gave rise to the memories, such as their emotional qualities. The emotional quality of events has been found to predict the likelihood of recall after delays (e.g., Peterson et al., 2014). We also did not assess the number of times the memories had been rehearsed or shared (which might be measured through a parent questionnaire, for example), which also could contribute to the likelihood of their long-term survival (although see Larkina & Bauer, 2012, and Peterson et al., 2014, for evidence to the contrary). Importantly, all of these limitations were equally true for the 4-year-olds, 6-year-olds, and 8-year-olds in the study. As such, although such measurements may prove to modify conclusions regarding the narrative predictors of the long-term survivability of memories, they would not be expected to modify the major conclusion that the slopes or rates of change in retention did not differ across the second half of the first decade of life.
The findings of the present research compel additional research on changes in autobiographical narrative throughout middle childhood and into early adolescence, a period still relatively neglected in the existing literature (although see Bauer, Hättenschwiler, & Larkina, 2015; Bauer & Larkina, 2019; and Larkina, Merrill, & Bauer, 2016). Also, as discussed in Bauer and Larkina (2016), it will be important to evaluate other potential predictors of autobiographical memory, such as self-referential variables, including self-concept (Harley & Reese, 1999; Howe & Courage, 1997) and subjective perspective (Fivush, 2014), which have been revealed to be important determinants of autobiographical memory. Indeed, self-concept assessed in infancy has been found to interact with maternal narrative style (i.e., elaborative talk; see Fivush & Zaman, 2014, for a review), such that maternal style is an important determinant of earliest memories for adolescents who had lower levels of self-concept in infancy, but not for adolescents who had higher self-concept in infancy (Reese & Robertson, 2018). Findings such as these highlight the importance of further research on self-referential variables, as well as the need for additional research that spans childhood and adolescence.
Conclusions
In conclusion, in the present research, we addressed the question of potential developmental changes in the likelihood of recall of events over long delays. Across the delay intervals of 1, 2, and 3 years, events were more likely to be recalled if they were experienced by older children. Yet importantly, the rates of increase in retention was the same for the children in the 4-, 6-, and 8-year cohorts. That is, the slopes or rates of change did not differ across the second half of the first decade of life. This implies continuity and gradual change in autobiographical memory. The present research also provided important findings regarding the predictors of the survival of individual memories over time. Whereas domain-general and memory-specific cognitive abilities relate to the breadth and thematic coherence of initial memory reports (Bauer & Larkina, 2019), they do not contribute unique variance to prediction of the survival of memories over long delays. That important role is reserved for the thematic coherence of initial memory reports, which was a consistent predictor.
Acknowledgments
Support for this research was provided by HD28425 and HD42486 to Patricia J. Bauer, and by Emory College of Arts and Sciences. The authors also thank the many members of the Cognition in the Transition (University of Minnesota) and Memory at Emory (Emory University) laboratories for help at various stages of the research. They extend a special note of gratitude to the children and families who generously gave of their time to participate in this research over a 4-year period. Their contributions made this work possible.
Appendix
An example of multilevel model with thematic coherence Theme) as a Level 1 predictor and age at the initial assessment (codes as a categorical variable Age Cohort) as a Level 2 predictor:
In each equation, the indices i and j are used to denote individual participants and memories, respectively. In Level 1, the intercept, β0ij is defined as the expected probability of remembering for memory j of participant i, and β1 slope represents the associated change in log odds of remembering. The error term, rit, represents a unique effect associated with participant i (i.e., how much an individual fluctuates in remembering over multiple events). The individual intercept β0i and slope β1i become the outcome variables in the Level 2 equations, where γ00 represents the overall mean probability of remembering for the sample. Further, γ01 corresponds to the effects of age cohort on the log odds of remembering above and beyond the effects of thematic coherence; γ10 corresponds to the effect of thematic coherence on remembering; γ11 is the cross-level interaction testing whether the relationship between thematic coherence and remembering varied as a function of participants’ age cohort; u0i represents the degree to which individuals vary from the sample as a whole.
Additional Level 1 (within-person) predictors were added in the different models, including wave at long-term recall (for delay of 1 year and 2 years only), narrative breadth, length, and factors. Children’s verbal comprehension scores were treated as Level 2 (between-person) because it was assessed only one at the time of initial enrollment in the study. In the models with multiple Level 1 predictors, the appropriate slopes were added.
Contributor Information
Patricia J. Bauer, Department of Psychology, Emory University;
Marina Larkina, Department of Psychology, Emory University;.
Evren Güler, Department of Psychology, Augsburg University;.
Melissa Burch, School of Cognitive Science, Hampshire College..
References
- Akhtar S, Justice LV, Morrison CM, & Conway MA (2018). Fictional first memories. Psychological Science, 1–8. doi: 10.1177/0956797618778831 [DOI] [PubMed] [Google Scholar]
- Baker-Ward L, Ornstein PA, & Principe GFA (1997). Revealing the representation: Evidence from children’s reports of events In van den Broek PW, Bauer PJ, & Bourg T (Eds.), Developmental spans in event comprehension and representation: Bridging fictional and actual events (pp. 79–107). Hillsdale, NJ, US: Lawrence Erlbaum Associates, Inc. [Google Scholar]
- Bauer PJ (2007). Remembering the times of our lives: Memory in infancy and beyond. Mahwah, NJ: Erlbaum. [Google Scholar]
- Bauer PJ (2014). The development of forgetting: Childhood amnesia In Bauer PJ & Fivush R (Eds.), The Wiley-Blackwell handbook on the development of children’s memory (pp. 519–544). West Sussex, United Kingdom: Wiley-Blackwell. [Google Scholar]
- Bauer PJ (2015). A complementary processes account of the development of childhood amnesia and a personal past. Psychological Review, 122, 204–231. 10.1037/a0038939 [DOI] [PubMed] [Google Scholar]
- Bauer PJ, Baker-Ward L, Krøjgaard P, Peterson C, & Wang Q (in press). Evidence against depiction as fiction: A comment on “Fictional First Memories” (Akhtar et al., 2018). Psychological Science. [DOI] [PubMed] [Google Scholar]
- Bauer PJ, Burch MM, Scholin SE, & Güler OE (2007). Using cue words to investigate the distribution of autobiographical memories in childhood. Psychological Science, 18, 910–916. 10.1111/j.1467-9280.2007.01999.x [DOI] [PubMed] [Google Scholar]
- Bauer PJ, Hättenschwiler N, & Larkina M (2016). “Owning” the personal past: Adolescents’ and adults’ autobiographical narratives and ratings of memories of recent and distant events. Memory, 24, 165–183. doi: 10.1080/09658211.2014.995673 [DOI] [PubMed] [Google Scholar]
- Bauer PJ, & Larkina M (2014a). Childhood amnesia in the making: Different distributions of autobiographical memories in children and adults. Journal of Experimental Psychology: General, 143(2), 597–611. DOI: 10.1037/a0033307. [DOI] [PubMed] [Google Scholar]
- Bauer PJ, & Larkina M (2014b). The onset of childhood amnesia in childhood: A prospective investigation of the course and determinants of forgetting of early-life events. Memory, 22, 8, 907–924. DOI: 10.1080/09658211.2013.854806 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bauer PJ, & Larkina M (2016). Predicting remembering and forgetting of autobiographical memories in children and adults: A 4-year prospective study. Memory, 24 (10), 1345–1368. DOI: 10.1080/09658211.2015.1110595 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bauer PJ, & Larkina M (2019). Predictors of age-related and individual variability in autobiographical memory in childhood. Memory, 27, 63–78. doi: 10.1080/09658211.2017.1381267 [DOI] [PubMed] [Google Scholar]
- Bauer PJ, Stark EN, Lukowski AF, Rademacher J, Van Abbema DL, & Ackil JK (2005). Working together to make sense of the past: Mothers’ and children’s use of internal states language in conversations about traumatic and non-traumatic events. Journal of Cognition and Development, 6, 463–488. 10.1207/s15327647jcd0604_2 [DOI] [Google Scholar]
- Cleveland ES, & Reese E (2008). Children remember early childhood: Long-term recall across the offset of childhood amnesia. Applied Cognitive Psychology, 22, 127–142. 10.1002/acp.1359 [DOI] [Google Scholar]
- Eacott MJ, & Crawley RA (1998). The offset of childhood amnesia: Memory for events that occurred before age 3. Journal of Experimental Psychology: General, 127, 22–33. 10.1037/0096-3445.127.1.22 [DOI] [PubMed] [Google Scholar]
- Fivush R (2014). Maternal reminiscing style: The sociocultural construction of autobiographical memory across childhood and adolescence In Bauer PJ & Fivush R (Eds.), The Wiley-Blackwell handbook on the development of children’s memory (pp. 568–585). West Sussex, United Kingdom: Wiley-Blackwell. [Google Scholar]
- Fivush R, & Schwarzmueller A (1998). Children remember childhood: Implications for childhood amnesia. Applied Cognitive Psychology, 12, 455–473. [Google Scholar]
- Fivush R, & Zaman W (2014). Gender, subjective perspective, and autobiographical consciousness In Bauer PJ & Fivush R (Eds.), The Wiley-Blackwell handbook on the development of children’s memory (Vol. II, pp. 586–604). West Sussex, United Kingdom: Wiley-Blackwell. [Google Scholar]
- Freud S (1905/1953). Childhood and concealing memories In Brill AA (Trans. & Ed.), The basic writings of Sigmund Freud (pp. 62–68). New York, NY: The Modern Library. [Google Scholar]
- Guo G, & Zhao H (2000). Multilevel modeling for binary data. Annual Reviews of Sociology, 26, 441–462. doi: 10.1146/annurev.soc.26.1.441. [DOI] [Google Scholar]
- Habermas T, Negele A, & Mayer FB (2010). Honey, you’re jumping about: Mothers’ scaffolding of their children’s and adolescents’ life narration. Cognitive Development, 25, 339–351. 10.1016/j.cogdev.2010.08.004 [DOI] [Google Scholar]
- Hamond NR, & Fivush R (1991). Memories of Mickey Mouse: Young children recount their trip to Disneyworld. Cognitive Development, 6, 433–448. 10.1016/0885-2014(91)90048-I [DOI] [Google Scholar]
- Harley K, & Reese E (1999). Origins of autobiographical memory. Developmental Psychology, 35, 1338–1348. 10.1037/0012-1649.35.5.1338 [DOI] [PubMed] [Google Scholar]
- Howe ML, & Courage ML (1997). The emergence and early development of autobiographical memory. Psychological Review, 104, 499–523. 10.1037/0033-295X.104.3.499 [DOI] [PubMed] [Google Scholar]
- Larkina M, & Bauer PJ (2010). The role of maternal verbal, affective, and behavioral support in preschool children’s independent and collaborative autobiographical memory reports. Cognitive Development, 25, 309–324. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Larkina M, & Bauer PJ (2012). “Family stories” and their implications for preschoolers’ memories of personal events. Journal of Cognition and Development, 13, 473–504. doi: 10.1080/15248372.2011.591295 [DOI] [Google Scholar]
- Larkina M, Merrill N, & Bauer PJ (2017). Developmental changes in consistency of autobiographical memories: Adolescents’ and young adults’ repeated recall of recent and distance events. Memory, 25, 1036–1051. 10.1080/09658211.2016.1253750. [DOI] [PubMed] [Google Scholar]
- Lukowski AF, & Bauer PJ (2014). Long-term memory in infancy and early childhood In Bauer PJ & Fivush R (Eds.), The Wiley-Blackwell handbook on the development of children’s memory (Vol. I, pp. 230–254). West Sussex, United Kingdom: Wiley-Blackwell. [Google Scholar]
- Morris G, Baker-Ward L, & Bauer PJ (2010). What remains of that day: The survival of children’s autobiographical memories across time. Applied Cognitive Psychology, 24, 527–544. [Google Scholar]
- Nelson K, & Fivush R (2004). The emergence of autobiographical memory: A social cultural developmental theory. Psychological Review, 111, 486–511. 10.1037/0033-295X.111.2.486 [DOI] [PubMed] [Google Scholar]
- O’Kearney R, Speyer J, & Kenardy J (2007). Children’s narrative memory for accidents and their post-traumatic distress. Applied Cognitive Psychology, 21, 821–838. 10.1002/acp.1294 [DOI] [Google Scholar]
- Perner J, & Ruffman T (1995). Episodic memory and autonoetic consciousness: Developmental evidence and a theory of childhood amnesia. Journal of Experimental Child Psychology, 59, 516–548. 10.1006/jecp.1995.1024 [DOI] [PubMed] [Google Scholar]
- Peterson C, Grant VV, & Boland LD (2005). Childhood amnesia in children and adolescents: Their earliest memories. Memory, 13, 622–637. 10.1080/09658210444000278 [DOI] [PubMed] [Google Scholar]
- Peterson C, Hallett D, & Compton‐Gillingham C (2018). Childhood amnesia in children: A prospective study across eight years. Child development, 89(6), e520–e534. [DOI] [PubMed] [Google Scholar]
- Peterson C, Morris G, Baker-Ward L, & Flynn S (2014). Predicting which childhood memories persist: Contributions of memory characteristics. Developmental Psychology, 50(2), 439. [DOI] [PubMed] [Google Scholar]
- Peterson C, & Rideout R (1998). Memory for medical emergencies experienced by 1- and 2-year-olds. Developmental Psychology, 34(5), 1059. [DOI] [PubMed] [Google Scholar]
- Peterson C, Sales JM, Rees M, & Fivush R (2007). Parent–child talk and children’ memory for stressful events. Applied Cognitive Psychology, 21, 1057–1075. [Google Scholar]
- Peterson C, Warren KL, & Short MM (2011). Infantile amnesia across the years: A 2-year follow-up of children’s earliest memories. Child Development, 82, 1092–1105. 10.1111/j.1467-8624.2011.01597.x [DOI] [PubMed] [Google Scholar]
- Pillemer DB, & White SH (1989). Childhood events recalled by children and adults In Reese HW (Ed.), Advances in child development and behavior (Vol. 21, pp. 297–340). Orlando, FL: Academic Press; 10.1016/S0065-2407(08)60291-8 [DOI] [PubMed] [Google Scholar]
- Raudenbush SW, & Bryk AS (2002). Hierarchical linear models: Applications and data analysis methods, (2 ed.), Sage, Thousand Oaks, CA. [Google Scholar]
- Reese E, Haden CA, Baker-Ward L, Bauer PJ, Fivush R, & Ornstein PA (2011). Coherence of personal narratives across the lifespan: A multidimensional model and coding method. Journal of Cognition and Development, 12, 424–462. 10.1080/15248372.2011.587854 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Reese E, Haden CA, & Fivush R (1993). Mother-child conversations about the past: Relationships of style and memory over time. Cognitive Development, 8, 403–430. 10.1016/S0885-2014(05)80002-4 [DOI] [Google Scholar]
- Reese E, Jack F, & White N (2010). Origins of adolescents’ autobiographical memories. Cognitive Development, 25, 352–367. [Google Scholar]
- Reese E & Robertson S (2019). Origins of adolescents’ earliest memories. Memory, 27, 79–91. 10.1080/09658211.2018.1512631 [DOI] [PubMed] [Google Scholar]
- Roebers C (2014). Children’s deliberate memory development: the contribution of strategies and metacognitive processes In Bauer PJ & Fivush R (Eds.), The Wiley-Blackwell handbook on the development of children’s memory (pp. 865–894). West Sussex, UK: Wiley-Blackwell. [Google Scholar]
- Roenker DL, Thompson CP, & Brown SC (1971). Comparison of measures for the estimation of clustering in free recall. Psychological Bulletin, 76, 45–48. [Google Scholar]
- Rubin DC (2000). The distribution of early childhood memories. Memory, 8, 265–269. 10.1080/096582100406810 [DOI] [PubMed] [Google Scholar]
- Singer JD, & Willett JB (2003). Applied longitudinal data analysis: Modeling change and event occurrence. New York, NY: Oxford University Press; 10.1093/acprof:oso/9780195152968.001.0001 [DOI] [Google Scholar]
- Snijders TAB, & Bosker RJ (2004). Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. London etc.: Sage Publishers. [Google Scholar]
- Suddendorf T, Nielsen M, & von Gehlen R (2011). Children’s capacity to remember a novel problem and to secure its future solution. Developmental Science, 14, 26–33. 10.1111/j.1467-7687.2010.00950.x [DOI] [PubMed] [Google Scholar]
- Tulving E (2002). Episodic memory: From mind to brain. Annual Review of Psychology, 53, 1–25. 10.1146/annurev.psych.53.100901.135114 [DOI] [PubMed] [Google Scholar]
- Tulving E (2005). Episodic memory and autonoesis: Uniquely human? In Terrace HS & Metcalfe J (Eds.), The missing link in cognition (pp.3–56). Oxford: Oxford University Press. [Google Scholar]
- Tustin K, & Hayne H (2010). Defining the boundary: Age-related changes in childhood amnesia. Developmental Psychology, 46, 1049–1061. 10.1037/a0020105 [DOI] [PubMed] [Google Scholar]
- Van Abbema DL, & Bauer PJ (2005). Autobiographical memory in middle childhood: Recollections of the recent and distant past. Memory, 13, 829–845. 10.1080/09658210444000430 [DOI] [PubMed] [Google Scholar]
- Wang Q, & Peterson C (2016). The fate of childhood memories: Children postdated their earliest memories as they grew older. Frontiers in Psychology, 6, 2038. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wetzler SE, & Sweeney JA (1986). Childhood amnesia: An empirical demonstration In Rubin DC (Ed.), Autobiographical memory (pp. 191–201). New York, NY: Cambridge University Press; 10.1017/CBO9780511558313.017 [DOI] [Google Scholar]
- Wheeler MA (2000). Episodic memory and autonoetic awareness In Tulving E & Craik FIM (Eds.), The Oxford handbook of memory (pp. 597–608). New York, NY: Oxford University Press. [Google Scholar]
- Woodcock RW, McGrew KS, & Mather N (2001). Woodcock-Johnson-III Test of Cognitive Abilities. Itasca, IL: Riverside Publishing. [Google Scholar]
- Wright DB (1998). Modeling clustered data in autobiographical memory research: The multilevel approach. Applied Cognitive Psychology, 12, 339–357. [Google Scholar]