The way mothers reminisce with their young children is critical for children's autobiographical memory development, as well as for a host of other skills (see Fivush, Haden, & Reese, Reference Fivush, Haden and Reese2006; Reese, Reference Reese2013; Salmon & Reese, Reference Salmon and Reese2016, for reviews). The long-term benefits of an elaborative maternal reminiscing style for a range of developmental outcomes have now been established through longitudinal correlational and experimental studies. When adults discuss shared past events with young children in elaborative ways, by providing many factual and emotional details about the past event, and by asking open-ended questions, children later remember past events in more detail (Haden, Ornstein, Rudek, & Cameron, Reference Haden, Ornstein, Rudek and Cameron2009; Jack, MacDonald, Reese, & Hayne, Reference Jack, MacDonald, Reese and Hayne2009; McGuigan & Salmon, Reference McGuigan and Salmon2004, Reference McGuigan and Salmon2006; Reese, Haden, & Fivush, Reference Reese, Haden and Fivush1993; van Bergen & Salmon, Reference van Bergen and Salmon2010). Children of mothers coached in elaborative reminiscing develop more accurate and detailed autobiographical memories, better emotion understanding, and more advanced theory of mind abilities (Reese & Newcombe, Reference Reese and Newcombe2007; Taumoepeau & Reese, Reference Taumoepeau and Reese2013; van Bergen, Salmon, Dadds, & Allen, Reference van Bergen, Salmon, Dadds and Allen2009). They also tell higher quality narratives about their own and others’ experiences (Peterson, Jesso, & McCabe, Reference Peterson, Jesso and McCabe1999; Reese, Leyva, Sparks, & Grolnick, Reference Reese, Leyva, Sparks and Grolnick2010; Reese & Newcombe, Reference Reese and Newcombe2007).

Mothers do adapt their reminiscing styles to individual children's attention and language skill and to their temperaments, such that mothers engage in more elaborative and emotional reminiscing with children who are more attentive during conversations, who possess better language skills, and who have high levels of effortful control (attentional self-regulation; Bird, Reese, & Tripp, Reference Bird, Reese and Tripp2006; Farrant & Reese, Reference Farrant and Reese2000; Laible, Panfile Murphy, & Augustine, Reference Laible, Panfile Murphy and Augustine2013; Rothbart, Ahadi, Hershey, & Fisher, Reference Rothbart, Ahadi, Hershey and Fisher2001). Mothers also differ in their elaborative reminiscing as a function of children's attachment security, such that mothers later become more elaborative and emotional in their reminiscing with children who were securely attached as toddlers (e.g., Newcombe & Reese, Reference Newcombe and Reese2004; Raikes & Thompson, Reference Raikes and Thompson2006). These differences in maternal reminiscing as a function of attachment security are more pronounced for negative than positive emotions (Laible, Reference Laible2004). It is maternal reminiscing about negative events that is most closely linked to children's socioemotional functioning, both concurrently (Laible, Reference Laible2011) and predictively (Laible et al., Reference Laible, Panfile Murphy and Augustine2013).

Despite these adaptations to children's characteristics and skills, mothers largely adopt a stable style of reminiscing across their different children (Haden, Reference Haden1998), with the same children over time (Haden et al., Reference Haden, Ornstein, Rudek and Cameron2009; Jack et al., Reference Jack, MacDonald, Reese and Hayne2009; Reese et al., Reference Reese, Haden and Fivush1993), and across different types of past events (Reese & Brown, Reference Reese and Brown2000; Reese & Neha, Reference Reese and Neha2015). Mothers who are more elaborative when reminiscing with their children, however, are not necessarily more elaborative in their talk about the here-and-now (Haden & Fivush, Reference Haden and Fivush1996) or in other more abstract conversations, such as their extratextual talk during shared book reading (Laible, Reference Laible2004; Leyva, Sparks, & Reese, Reference Leyva, Sparks and Reese2012). Mothers are, however, consistently elaborative in their talk about past and future events (Hudson, Reference Hudson2002). This pattern suggests that maternal reminiscing style is specific to talk about events displaced in time, both past and future.

Although more highly educated mothers tend to engage in more elaborative reminiscing, some mothers with lower levels of education are highly elaborative in their reminiscing (Reese & Newcombe, Reference Reese and Newcombe2007). Moreover, mothers in a diverse range of cultures reminisce with their children (Miller, Potts, Fung, Hoogstra, & Mintz, Reference Miller, Potts, Fung, Hoogstra and Mintz1990), although at times they elaborate about different aspects and types of past events. For instance, Māori mothers are particularly elaborative about children's birth stories in comparison to their talk about everyday past events (Reese, Hayne, & MacDonald, Reference Reese, Hayne and MacDonald2008), and Chinese mothers are particularly elaborative about children's role in appropriate social interactions compared to European American mothers (Wang & Fivush, Reference Wang and Fivush2005). Individual differences in mothers’ reminiscing style exist within all cultures studied to date (e.g., Reese & Neha, Reference Reese and Neha2015; Schröder et al., Reference Schröder, Keller, Kärtner, Kleis, Abels, Yovsi and Papaligoura2013; Tõugu, Tulviste, Schröder, Keller, & De Geer, Reference Tõugu, Tulviste, Schröder, Keller and De Geer2011; Wang & Fivush, Reference Wang and Fivush2005).

Thus, reminiscing appears to be a special context of conversation that is distinct from other forms of talk. Although maternal reminiscing is shaped by cultural values and by education levels, individual differences exist within cultures and within social classes. Given the apparent importance of reminiscing for children's development, it is vital for us to better understand the origins of maternal reminiscing style, and in this way shed light on the reasons for individual differences among mothers in their reminiscing style. Little research to date has explored maternal characteristics linked to mothers’ reminiscing style. In a notable exception, Laible et al. (Reference Laible, Panfile Murphy and Augustine2013) investigated the role of mothers’ personality traits and caregiving representations for their reminiscing style with preschoolers. Mothers who reported higher levels of openness and more positive caregiving representations were more elaborative when reminiscing with their preschoolers at ages 3.5 and 4 years. The main aim of the present study was to identify even earlier origins of mothers’ reminiscing styles with their very young children, after controlling for critical child characteristics and skills.

Several factors arise as potential explanations of individual differences in maternal reminiscing style. First and foremost, mothers’ own mental states and moods may contribute to their reminiscing style. In particular, mothers who are depressed engage in less sensitive and more negative interactions with their children (see Lovejoy, Graczyk, O'Hare, & Neuman, Reference Lovejoy, Graczyk, O'Hare and Neuman2000, for a meta-analysis). With respect specifically to reminiscing, Wareham and Salmon (Reference Wareham and Salmon2006) theorized that maternal depression would be associated with less frequent, sensitive, and elaborative reminiscing with children, accompanied by higher levels of negative affect. To our knowledge, only one reminiscing study has explored the role of maternal depression. In a low-income sample, Raikes and Thompson (Reference Raikes and Thompson2006) found that maternal depression when children were 2 years old was directly associated with children's lower emotion understanding at age 3, but this association was not mediated by mother–child talk about past emotions of happiness, anger, and sadness. Maternal depression was not significantly associated at either time point with mother–child references to past emotions. Instead, mother–child emotion talk was predicted by attachment security at age 2. However, this study did not examine maternal and child elaborative reminiscing, but only the frequency of the dyads’ emotion words in the conversations.

In related research, several studies have been conducted on mothers’ and fathers’ reminiscing with anxious and nonanxious preadolescent children about worried, angry, and happy past events (Suveg, Zeman, Flannery-Schroeder, & Cassano, Reference Suveg, Zeman, Flannery-Schroeder and Cassano2005; Suveg et al., Reference Suveg, Sood, Barmish, Tiwari, Hudson and Kendall2008). The findings varied as a function of the valence of the event. Overall, parents of anxious and nonanxious children were similar when discussing a happy event, but parents of anxious children talked less elaboratively about negative events of worry and anger. Maltreating mothers also exhibit lower levels of elaborative reminiscing with their preschool children about emotional events compared to nonmaltreating mothers (Valentino et al., Reference Valentino, Hibel, Cummings, Nuttall, Comas and McDonnell2015), but maltreating mothers can be coached to engage in more elaborative and emotional reminiscing with their children (Valentino, Comas, Nuttall, & Thomas, Reference Valentino, Comas, Nuttall and Thomas2013). To extrapolate from these related studies, we predicted that depressed mothers would reminisce less elaboratively than nondepressed mothers about emotional events, and that these differences would be exacerbated when discussing a negative experience. In community samples, this exploration entails examining the extremes of the distribution in order to compare depressed mothers to nondepressed mothers. In the present sample, we were also able to track changes in depressive symptoms over time. Those mothers who experience depressive symptoms at only one time point may be quite different from those mothers who experience stably high or increasing/decreasing levels. We wanted to capture these patterns in a person-centered way by examining stability of depression over time. Thus, it is the most comprehensive examination to date of the role of maternal depression in mother–child reminiscing.

Second, mothers who are highly elaborative when talking about past events with their children may be better at taking the perspective of others. When talking about a past event, one must rely on an internal representation of the event in order to discuss it with another person. Ideally one would be able to imagine the other person's perspective on an event, for both objective and subjective details (Fivush & Nelson, Reference Fivush and Nelson2006; Reese & Cleveland, Reference Reese and Cleveland2006). One marker of mothers’ ability to take the perspective of their children is mind-mindedness (Meins, Reference Meins1997). Mind-mindedness is assessed in infancy in terms of caregivers’ tendency to comment in an appropriate versus nonattuned manner on their infants’ internal states (Meins et al., Reference Meins, Fernyhough, de Rosnay, Arnott, Leekam and Turner2012; Meins, Fernyhough, Fradley, & Tuckey, Reference Meins, Fernyhough, Fradley and Tuckey2001). Appropriate mind-related comments indicate an accurate interpretation of the infant's putative internal state (e.g., “You want that toy” if the infant gestures or reaches toward it), whereas nonattuned comments index misinterpretations of the infant's thoughts or feelings (e.g., “You're bored with that toy” while the infant is still actively involved in playing with it). Individual differences in mothers’ mind-mindedness are relatively stable across early childhood (McMahon, Camberis, Berry, & Gibson, Reference McMahon, Camberis, Berry and Gibson2016; Meins, Fernyhough, Arnott, Turner, & Leekam, Reference Meins, Fernyhough, Arnott, Turner and Leekam2011; Meins et al., Reference Meins, Fernyhough, Wainwright, Clark-Carter, Das Gupta, Fradley and Tuckey2003) and may emerge even before the infant is born (Arnott & Meins, Reference Arnott and Meins2008; McMahon et al., Reference McMahon, Camberis, Berry and Gibson2016). Thus, individual differences in mind-mindedness predate any engagement in mother–child reminiscing, which begins around the middle of the second year of life when children start verbally referencing the past (Reese, Reference Reese1999; Sachs, Reference Sachs1983). Mothers who are more likely to impute mental states to their infants may be more willing later on to elaborate on their children's personal experiences, including their subjective experiences.

Third, elaborative reminiscing also entails a certain level of sensitivity about what the child does and does not wish to remember or talk about. Parents and children are often interested in dramatically different parts of the same experiences (e.g., Reese, Reference Reese1999). For instance, when discussing a past visit to the zoo, the parent may want to talk about the exotic animals that they saw, but the child is only interested in talking about a crushed worm she found on the footpath. Parents will be more successful in reminiscing with their children if they are sensitive to the aspects of the events that children want to discuss, and if they follow in on these aspects in their conversations. Children recall the greatest detail in past event conversations when mothers elaborate in an autonomy-supportive rather than a controlling way (Cleveland & Reese, Reference Cleveland and Reese2005). Thus, mothers who are generally more autonomy supportive and sensitive to their children in their everyday interactions from infancy may later become highly elaborative in their reminiscing. Nevertheless, although maternal mind-mindedness is correlated with maternal sensitivity, the two are distinct constructs, with mind-mindedness a better predictor than maternal sensitivity of children's social understanding (Meins, Fernyhough, Arnott, Leekam, & de Rosnay, Reference Meins, Fernyhough, Arnott, Leekam and Rosnay2013; Meins et al., Reference Meins, Fernyhough, Wainwright, Das Gupta, Fradley and Tuckey2002, Reference Meins, Fernyhough, Wainwright, Clark-Carter, Das Gupta, Fradley and Tuckey2003).

Fourth and finally, we acknowledge the important role that children play in shaping mothers’ reminiscing style. Although mothers are stable in their reminiscing over time and across different children, they are more elaborative when children have higher language levels (Farrant & Reese, Reference Farrant and Reese2000). Some research also demonstrates differences as a function of child gender, with mothers reminiscing more elaboratively with girls than with boys (e.g., Reese & Fivush, Reference Reese and Fivush1993; but cf. Laible, Reference Laible2004). Children's interest in reminiscing and in conversing in general is apparent from a young age. For instance, mothers became more elaborative over time when their toddlers (19 months) were more attentive during the conversations, sharing eye contact with their mothers and engaging in appropriate conversational turns during reminiscing (e.g., “Mmm” and “I dunno”; Farrant & Reese, Reference Farrant and Reese2000). Farrant and Reese hypothesized that this attentive turn taking was akin to joint attention in reminiscing, such that both mother and child were jointly attending to the memory conversation, and perhaps to the memory representation. This shared interest in maintaining the conversation may have its roots in joint attention processes in infancy, with mother and child sharing a gaze upon objects and referencing them nonverbally and verbally. Yet no research to date has explored the link between joint attention in infancy and later mother–child reminiscing. Thus, we hypothesized that another important child factor in shaping maternal reminiscing style is the level of joint attention in infancy.

Results

At Phase 1, 5 mothers did not complete the depression measure. Mind-mindedness and sensitivity data were not available for 1 participant due to a technical recording difficulty, and 3 Strange Situations were terminated due to the infant becoming overly distressed. The joint attention measure was also missing for those 3 children. Three children did not complete the language assessment due to attention difficulties. A total of 19 positive and 27 negative event conversations were missing because the task was not administered (10 positive/10 negative), because of video and audio recorder malfunction (1 positive/1 negative), because the dyad couldn't generate an event to discuss (0 positive/6 negative), or because the child was uncooperative (3 positive/4 negative). Other conversations were excluded (5 positive/6 negative) because the dyads discussed emotions other than happy and scared. Analyses comparing dyads who did and did not have reminiscing data revealed no significant differences on any of the maternal or child variables (all ps > .06).

Our main aim was to explore the predictors of maternal and child elaborations when reminiscing about a positive (happy) and a negative (scared) event, with a specific focus on maternal depression. In this community sample, we needed to look at the extremes of the distribution, and to capture patterns of change in depressive symptoms over time, in order to compare depressed to nondepressed mothers. We used full information maximum likelihood because we used raw data with some missingness at Phases 3 and 4. The full information maximum likelihood techniques provide less biased estimates than listwise or pairwise deletion, and are used even when data are not missing at random (Schafer & Graham, Reference Schafer and Graham2002). Schafer and Graham argue that one can achieve adequate performance from likelihood estimates in psychology without necessarily modeling probabilities of missingness, because the consequences from these departures from random missingness tend not to be serious when compared to other research fields; however, admittedly this depends on the sample size, where larger samples are preferred. Yet, in Mplus, we examined proportions of missing data via a covariance “coverage,” and coverage ranged from .95 to 1.00; the minimum coverage is recommended at .10 (Muthén & Muthén, Reference Muthén and Muthén2013). In addition, we examined whether missingness on reminiscing was related to our early measures of attachment, mind-mindedness, and depression, and none were significant predictors of missingness. In preliminary analyses, child gender was significantly associated with three variables: dyadic joint attention, child language, and child reminiscing for the negative event, with girls demonstrating higher levels than boys for these variables. In contrast, SES was not significantly associated with mother–child reminiscing (rs between SES and reminiscing variables ranged from .02 to .14, ns). Thus, in the final model, we did not include SES, but we retained child gender. See Table 1 for descriptive statistics for all variables included in analyses, and Table 2 for bivariate correlations among variables in the final model.

Table 1. Descriptive statistics for maternal and child variables

Table 2. Bivariate correlations among variables in final model

Note: C_Happy = child elaborations about happy event (44 months); C_Scared = child elaborations about scared event (44 months); M_Happy = mother elaborations about happy event (44 months); M_Scared = mother elaborations about scared event (44 months); M_Mind = maternal mind-mindedness (8 months); C_Lang = children's language (26 months); C_Gender = child gender, 0 = male and 1 = female; JA = joint attention with mother (15 months); M_Sens = maternal sensitivity (8 months); Att = attachment security with mother (15 months), 0 = insecurely attached and 1 = securely attached; M_Dep= depression status based on latent profile analysis, 0 = nondepressed and 1 = increasing depression; correlations below the diagonal based on n = 140 with complete reminiscing data; correlations above the diagonal based on pairwise deletion with full sample. *p < .05; **p < .01.

Latent profile analysis (LPA)

LPA is a group-based trajectory model, which is based on approaches developed by Nagin (Nagin, Reference Nagin1999, Reference Nagin2005; Nagin & Tremblay, Reference Nagin and Tremblay1999). It is an extension of latent class analysis but accommodates continuous indicators. LPA identifies heterogeneous latent classes by decomposing the covariance matrix to create distinct subgroups that are assumed to be homogenous with regard to change over time: individuals are assumed to follow a particular trajectory together that indicates a shared psychopathological longitudinal trajectory for a subgroup of individuals (Bauer & Curran, Reference Bauer and Curran2004). In this study, we were interested in tracing the trajectories of mothers based on maternal depression levels reported from the time their children were 8 to 44 months. In this community sample, we expected to find groups of mothers who reported low levels of maternal depression across time, relatively high levels across time, and possibly decreasing or increasing levels. At the start of the study, mothers’ scores on the BDI were distributed into the following categories: 166 (82%) minimal depression, 24 (12%) mild depression, 4 (2%) moderate depression, and 7 (4%) severe depression.

To identify the number of groups or classes deemed to have separate trajectories, we specified models with different numbers of expected classes using Mplus 7.3 (Muthén & Muthén, Reference Muthén and Muthén2013). We then tested the fit and likelihood ratios to guide our selection of the best model. Separate LPA models that differ in the number of classes were specified, which allows for the identification of the optimal number of groups with different trajectories (e.g., high, low, decreasing, and increasing). We used the Bayesian information criterion (BIC), the Lo–Mendel–Rubin (LMR) statistic, and the entropy value as statistical criteria for model comparisons (Nylund, Asparouhov, & Muthén, Reference Nylund, Asparouhov and Muthén2007). The model with lower BIC is preferred. The LMR statistic tests k – 1 classes against k classes, and a nonsignificant chi-square value (p > .05) suggests that a model with one fewer class is preferred (Lo, Mendell, & Rubin, Reference Lo, Mendell and Rubin2001).

We estimated four separate LPA models to compare from one to four classes of trajectories. Each model specified how many classes to extract: one through four. The models were bootstrapped at 100 initial stage random starts and 20 final stage optimizations. Vuong–Lo–Mendell–Rubin likelihood ratio for two versus three indicated that two classes could not be rejected as being favored over three classes: –2,513.927. –2LLΔ = 62.462, df = 3, p = .6616; LMR = 58.785, p = .6744. However, the Vuong–Lo–Mendell–Rubin likelihood ratio for one versus two classes indicated two classes was favored: –2,585.744. –2LLΔ = 80.973, df = 3, p = .0288; LMR = 76.205, p = .0349. As shown in Table 3, the BIC statistic decreased from testing one-class models to three-class models, and the LMR statistic fell below significance for the three-class model, again along with Vuong–Lo–Mendell–Rubin likelihood ratio, suggesting the model with only two classes (i.e., low and increasing) was preferred. Thus, the two-class model fit the data better than the one-class, homogenous-sample model where no distinct classes exist. Although the three-class model had a lower BIC and better entropy than the two-class model, these improvements in fit were not significant, as indicated by a nonsignificant LMR. The LMR also suggests that the three-class solution is not a significant improvement over the two-class solution. Thus, the two-class model was retained as the entropy indicated good cohesion within classes/profiles and good separation among classes/profiles. The average probability scores for the two identified groups were .91 and .99, respectively, with an entropy value of .91. The classes were well separated (heterogeneous) from each other, yet homogenous within their grouping. The two classes identified are shown in Figure 1. One group typified mothers who were stably low on depressive symptoms (91.8%) and the other typified mothers who were increasing in depressive symptoms over time (8.2%). For this latter group, mothers’ scores on the BDI were in the “moderate depression” range by Phases 3 and 4 of the study.

Figure 1. Latent profile analysis of maternal depression over time (8 to 44 months).

Table 3. Results of latent profile analysis of maternal depression across infant ages of 8, 15, 24, and 44 months with Bayesian information criteria (BIC) and the Lo–Mendel–Rubin (LMR) likelihood ratio test as main criteria for selection of the number of latent classes to retain

We tested the validity of the classes extracted and retained. A fully saturated model (χ² = 0.00, df = 0, p = .00, comparative fit index = 1.00, Tucker–Lewis index = 1.00, root mean square error of approximation = 0.00) with no latent variables estimated, only observed variables, was tested using binary regression in Mplus 7.3. We present unstandardized estimates, standard errors (SEs), and 95% confidence intervals (CIs) as indicators of the size of the effect. As expected, the maternal depression classes identified (low vs. increasing) differed on maternal sensitivity, estimate = –1.07, SE = 0.40, p = .008, 95% CI [–1.86, –0.28], and mind-mindedness, estimate = –2.20, SE = 0.99, p = .027, 95% CI [–4.15, –0.25], with the increasing group showing less sensitivity and fewer appropriate mind-related comments when infants were 8 months old than the low group. However, they did not differ significantly on children's attachment security at 15 months, estimate = –0.23, SE = 0.12, p = .07, 95% CI [–0.47, 0.02], although the association was in the direction of insecure attachment for children whose mothers were in the increasing depression group. This pattern suggests the classes identified are valid subgroups that show meaningful differences in the quality of mother–infant interaction, despite the overall low levels of depression in this sample.

To test reminiscing by depressive grouping, we conducted two regressions. In the first, we included only the depression classes (but controlling for child gender). Thus, we regressed the four reminiscing variables (maternal elaborations and child elaborations for scared and happy events) onto the depressive classes, controlling for gender. The total R ² values for the happy event were .04 (SE = 0.03, p = .23) for children's elaborations and .05 (SE = 0.04, p = .16) for mothers’ elaborations. The total R ² values for the scared event were .11 (SE = 0.05, p < .05) for children's elaborations and .09 (SE = 0.05, p = .06) for mothers’ elaborations. We found that the group with increasing maternal depression was associated with fewer child elaborations for scared events at 44 months, estimate = –1.86, SE = 0.94, p = .047, 95% CI [–3.70, –0.03]. No other significant effects were found.

In the second regression, we tested whether the depression classes would still be significant predictors of reminiscing when included in the same model as the other hypothesized predictor variables: maternal sensitivity, mind-mindedness, joint attention, attachment security, and language. We regressed the four reminiscing measures onto child gender, child language, the dichotomous depression group variable, maternal sensitivity, mind-mindedness, attachment, and joint attention. Figure 2 shows the results of this final model including all measures. The total R ² values for the happy event were .04 (SE = 0.03, p = .20) for children's elaborations and .05 (SE = 0.04, p = .16) for mothers’ elaborations. The total R ² values for the scared event were .16 (SE = 0.06, p < .01) for children's elaborations and .15 (SE = 0.06, p < .01) for mothers’ elaborations. The depression classes were no longer significant predictors. Instead, as expected, children's language significantly predicted child and maternal elaborations, specifically for scared events, estimate = 0.04, SE = 0.02, p = .035, 95% CI [0.00, 0.07]; estimate = 0.09, SE = 0.03, p = .010, 95% CI [0.02, 0.15], respectively. Child elaborations for scared events were additionally predicted by joint attention, estimate = 0.22, SE = 0.10, p = .028, 95% CI [0.02, 0.42], and maternal sensitivity, estimate = 0.46, SE = 0.19, p = .012, 95% CI [0.10, 0.83]. Maternal sensitivity also predicted maternal elaborations for scared events, estimate = 0.90, SE = 0.36, p = .013, 95% CI [0.19, 1.61]. Maternal sensitivity thus predicted greater mother–child elaboration for scared events, and this effect remained when including other mother–child relationship variables, child variables, and depressive class trajectories.

Figure 2. Results of full model predicting mother–child reminiscing.