Maternal Nighttime Sleep and Infant–Mother Attachment Security: The Mediating Role of Maternal Parenting Quality during Bedtime and Free Play

Abstract

The present study examined mothers’ emotional availability (EA) during daytime free play and bedtime as a mediator of linkages between maternal nighttime sleep and infant–mother attachment. Participants included 153 mothers (85% White) with infants (53% female). When infants were 1, 3, 6, 9, and 12 months, maternal sleep was assessed using actigraphy and daily sleep diaries for seven consecutive days. At each time point, mothers’ EA was scored from one observation of daytime free play and from one evening observation of infant bedtime by trained observers who were blind to all other participant information. Average scores were created for maternal sleep and EA across the five occasions in the first year. At 12 and 18 months, infant–mother attachment security in the home was scored by blind observers using the Attachment Q-Set, averaged across the two age points, and used in analyses. Mediational analyses revealed that mothers who experienced highly variable sleep and had poor sleep quality were less emotionally available with infants at bedtime during infants’ first year of life, which in turn was predictive of lower infant–mother attachment security in the second year, supporting mediation. Linkages between maternal sleep characteristics and daytime EA were less evident. Later maternal sleep timing was also directly predictive of low attachment security, after accounting for maternal EA. Findings emphasize that poor parental sleep places both parenting and infant socioemotional development at risk, and that parental sleep hygiene and sleep habits should be a salient focus of parenting intervention.

The family is the first and most important microsystem for development during infancy, and caregiving environments created by parents during interactions with infants have significant organizing influences on child developmental outcomes. It has been well-documented that parenting quality is predictive of infant adjustment, including physiological stress response (Philbrook et al., 2014), attachment security (Kim et al., 2017; van Bakel & Riksen-Walraven, 2002), sleep development (Philbrook & Teti, 2016), and cognitive development (e.g., van Bakel & Riksen-Walraven, 2002).

Parent Sleep and Parenting Quality

A large data base broadly supports the conclusion that parenting quality is multiply determined (Belsky, 1984; Teti et al., 2017). Parents with more depressive symptoms, highly reactive infants, and low marital and network support are more likely to have difficulty providing sensitive parenting to their infants (e.g., Bernard et al., 2018; Perry et al., 2018; Zhou et al., 2017). Curiously, one factor that has been largely overlooked in this literature is parental sleep. Only a few recent studies have provided evidence that parental sleep is predictive of parenting quality in parents with young children (Bai et al., 2020; King et al., 2020; McQuillan et al., 2019). However, no published study to date has investigated whether the effects of sleep on parenting quality matters, in terms of child developmental outcomes. The present study attempted to examine whether and how maternal sleep influences maternal parenting quality and infant–mother attachment security, a key formative outcome in infancy that has been associated with later child adjustment (e.g., Fearon et al., 2010; Groh et al., 2014; Groh et al., 2017; Madigan et al., 2013, Tabachnick et al., 2020). We were especially interested in examining whether the quality of mothering would mediate linkages between maternal sleep and infant attachment security to mothers.

Nighttime sleep has been linked with individual daily functioning and family relationships that have direct links to parenting quality. Specifically, insufficient and poor sleep has been associated with more depressive and anxiety symptoms (e.g., McDaniel & Teti, 2012; Gueron-Sela et al., 2021), poor performance and negative bias in attention and memory (e.g., Banks & Dinges, 2007), poor self-regulatory abilities (e.g., Gruber & Cassoff, 2014), as well as negative evaluations of marital relationships and coparenting relationship (e.g., Maranges & McNulty, 2017; McDaniel & Teti, 2012). Because of its ubiquitous effects, it was reasonable to expect that disrupted or insufficient parental sleep would undermine parenting quality during parent-child interactions, which in turn would be expected to be predictive of insecure attachment (Verhage et al., 2016).

As mentioned, only a few empirical studies to date have examined sleep–parenting linkages in parents with young children (Bai et al., 2020; King et al., 2020; McQuillan et al., 2019). In a recent cross-sectional study with mothers of toddlers, McQuillan et al. (2019) found that mothers with greater sleep deficits (i.e., later, shorter, and more variable sleep), higher levels of nighttime activity, and longer sleep onset latency were less likely to show positive parenting during observations of mother-child interaction. In addition, mothers who required longer time to fall asleep were more likely to report dysfunctional parenting (McQuillan et al., 2019). Similarly, in a longitudinal study, Bai et al. (2020) found that maternal irregular sleep patterns (i.e., late fall asleep times, great and increased variability in sleep, and short nighttime sleep) were predictive of poor bedtime parenting quality, from video-recorded observations, across infants’ first six months. King et al. (2020) found that poorer actigraphy-assessed sleep continuity (i.e., a composite score of number of wakings and wake time after sleep onset) was predictive of decreased maternal sensitivity during free-play interaction with infants. However, maternal self-reported sleep disturbance and actigraphic sleep duration were not associated with maternal sensitivity. Although limited, existing evidence supports the hypothesis that quality of parenting for young children becomes compromised in parents who sleep late and experience poor quality and insufficient sleep at night.

To our knowledge, Bai et al. (2020) and King et al. (2020) are the only studies demonstrating relations between maternal sleep and parenting quality during infancy. Although Bai et al. (2020) examined linkages between maternal sleep and parenting quality longitudinally, maternal sleep was assessed via daily diaries, which, albeit more rigorous than questionnaire assessments, nevertheless constitute subjective measurements of sleep. The disagreements between self-report (e.g., daily diaries) and objective measurements (e.g., actigraphy) of sleep have been reported across several studies, and these discrepancies may be due to potential biases in individuals’ subjective estimate of sleep (Girschik et al., 2012; Jackson et al., 2020). Indeed, King et al. (2020) found no association between maternal subjective sleep (i.e., nighttime sleep duration, number of night wakings, wake time after sleep onset, and perceived sleep quality) and maternal sensitivity towards infants during free-play interaction. However, they found actigraphic sleep continuity but not sleep duration to be predictive of changes in maternal sensitivity during interactions. Thus, studies applying objective measurements of sleep are warranted to understand more comprehensively the contributions of parental sleep to parenting and parent–child relationships. The present study, therefore, examined predictive linkages between maternal sleep and parenting quality across the infants’ first years, and in turn between maternal sleep and parenting quality and quality of infant–mother attachment in infants’ second year, taking a multi-method approach to the assessment of maternal sleep that integrated information from sleep diaries and actigraphy. Further, because sleep is a multi-dimensional construct, the present study included multiple sleep indicators (e.g., mean sleep duration, variability in sleep duration, wake time after sleep onset, sleep fragmentation, fall asleep time) to help determine if specific aspects of maternal sleep were more predictive of parenting quality and infant attachment than others.

Quality of Parenting in Context

Although the fundamental role of parenting quality in promoting secure attachment relationships has been well established theoretically and empirically, the magnitude of the effect of parental sensitivity on attachment security varies across studies (Ainsworth, 1973; Ainsworth et al., 1978; Bowlby, 1969/1982; see De Wolff & van Ijzendoorn, 1997; van Ijzendoorn et al., 2004; Verhage et al., 2016; Zeegers et al., 2017, for reviews). These varying effect sizes may, in part, be explained by taking into account the different contexts in which parenting quality has been assessed (Goldberg et al., 1999; Thompson, 1997). From a methodological perspective, brief, relatively low-stress, structured observations of parenting, frequently used in many previous studies, may have low ecological validity, and these observations may constrain researchers’ ability to assess meaningful individual differences in parenting quality. Several studies have suggested that parenting in response to infant distress may be a more salient predictor of infant attachment than parenting assessed in lower-stress contexts (Leerkes, 2011; McElwain & Booth-LaForce, 2006). However, despite acknowledgments that parenting within specific family contexts may be differentially predictive of attachment, it is rare for studies to have actually examined parenting quality between different contexts in predicting infant attachment. The present study examined maternal parenting quality during a semi-structured, daytime free play and in a naturalistic, unstructured, daily-occurring context common to all infants, infant bedtime.

Bedtime as an alternative context to free play was examined for several reasons. First, preparing the infant for sleep requires the parent to downregulate the infant’s emotions and attention, which may demand greater effort from the parent than stimulating the child for play. In addition, parents may find infant bedtimes to be more challenging and potentially stressful than a daytime free play for the simple reason that parents are likely to be tired at the end of the day, and especially so if they have experienced poor and/or insufficient sleep. We thus expected maternal sleep patterns to have a more evident influence on parenting quality observed during infant bedtime than during a daytime free play. We also hypothesized that the greater challenges imposed by the bedtime context would more accurately reflect mothers’ capacity to be sensitive and attuned to their infants and in turn render maternal parenting quality at infant bedtime to be a more salient and robust predictor of infant attachment security and mediator of maternal sleep-infant attachment linkages, compared with maternal parenting observed during daytime free play.

The Present Study

The purpose of the current longitudinal study was to examine relations among first-year maternal nighttime sleep, first-year maternal parenting quality, and second-year infants’ attachment security to mothers, and to determine the extent to which any relations found between maternal sleep and infant–mother attachment would be mediated by the quality of mothering (see Figure 1 for the conceptual model). In the current study, maternal parenting quality was observed during a daytime free play and during infant bedtime using the emotional availability (EA) framework. EA of mothers toward their infants was conceptualized in terms of mothers’ ability to understand infants’ signals and provide appropriate emotional responses to infants’ needs along four parental dimensions that, together, contributed to mothers’ emotional availability: sensitivity, structuring, nonhostility, and nonintrusiveness (Biringen et al., 1998). Mothers’ EA in turn was used to predict infant secure base behavior in the home, averaged across 12 and 18 months, using the Attachment Q-Set (AQS; Waters et al., 1995; Waters et al., 2021).

Figure 1.

The Conceptual Mediational Model Examining the Indirect Effect of Maternal Nighttime Sleep on Infant Attachment Security through Maternal Bedtime and Daytime Emotional Availability.

Note. Mos = months; EA = maternal emotional availability. Model for each sleep component (i.e., sleep variability, interrupted sleep, sleep timing, sleep duration) was examined separately.

To our knowledge, this is the first study to examine whether maternal parenting quality mediates the association between maternal sleep and infant attachment security, and to do so with examinations of parenting quality in two family contexts, one a naturalistic context common to all infants (bedtime) and a semi-structured context in which mothers were asked to play with their infants (free play). In addition, the present study included multiple characteristics of maternal sleep, including sleep quantity, quality, variability, and timing, to provide a more comprehensive examination of whether and how maternal sleep impacts maternal parenting quality and infant–parent attachment security.

The following hypotheses were addressed:

1. Later sleep timing, shorter sleep durations, greater variability in sleep, and lower quality sleep will predict poorer quality of maternal parenting, and these linkages would be especially evident at infants’ bedtimes.

2. Infants of mothers who provided higher quality of parenting across the first year of life would be more securely attached to their mothers in their second year. Left open was whether this prediction would be as equally or more strongly supported with parenting observed at infant bedtimes vs. earlier in the day.

3. Mothers who experienced late, poor, variable, or insufficient sleep would be more likely to have lower quality of parenting, particularly at infant bedtimes, which in turn, would be linked with less secure infant–mother attachment.

Method

Participants

This study was not pre-registered. The current study drew data from a larger project on infants’ and their parents’ sleep trajectories, parenting, and developmental outcomes across the first two years of life. All procedures of the project were approved by the Institutional Review Board of the Office of Research Protections at the Pennsylvania State University (IRB ID #PRAMS00042695, study title: Study of Infants’ Emergent Sleep TrAjectories). Mothers who had given birth to healthy, full-term infants were recruited from local medical centers immediately following birth and were given complete study information via phone call 2–3 weeks later. Families who were interested were visited again at their homes when the infants were one month old, to obtain consent and other demographic information. At one month, participants included 167 mother-infant dyads. Mothers had a mean age of 29.43 (SD = 5.27) with a range of 18 to 43 years of age, were mostly married or cohabitating (97.0%), and most had at least a high school degree or more (98.2%). Over half (61.7%) were employed. Eighty-three percent of the mothers were White, 3.6% were Black, 3.6% Asian, 5.4% Hispanic/Latinx, 3.6% Other, and 1.2% did not report. The median yearly family income was $60,000 (ranged from $1,538 to $350,000). According to the family income-to-needs ratio, 31.7% of families’ financial circumstances were identified as poverty (ratio < 1) or low income (1< ratio <2; Diemer et al., 2013). Approximately half of the infants were biologically female (53%), and 37% were firstborns. Twenty percent of infants consistently slept in a separate room from parents (i.e., solitary sleep) across the first year, 31% switched to solitary sleep at 3 or 6 months, 8% switched to solitary sleep at 9 to 12 months, 13% consistently shared bed or room with their parents across the first year, 11% had sleep arrangements that varied across the night during the year, and 17% of families did not report their sleep arrangements.

Of the 167 families that were originally recruited, 156 mothers (85% White) and their infants (53% female) remained in analyses for the current study. Eleven families were excluded because they had no data on any study variables. A priori power analysis conducted for the larger project suggested that a sample size of 120 was sufficient to provide high power (95%) with as many as nine predictors in the regression analyses. Therefore, we regarded our final sample size (N = 156 families) to be adequate to achieve sufficient power to find effects. Using chi-square test and one-way analysis of variance (ANOVA) in IBS SPSS Statistics 26, the final sample was compared to the 11 excluded families on sociodemographic variables such as maternal age, maternal race/ethnicity, yearly income, education, marital/cohabitation status, maternal employment status, and the number of children. Mothers in the final sample did not significantly differ from those excluded on any sociodemographic variables.

Procedure

Data for the current study were collected when infants were 1, 3, 6, 9, 12, and 18 months. At each time point over a seven-day period, parents were visited multiple times in their homes for about 30 minutes to two hours to obtain demographic, sleep (parent report and actiwatch), and bedtime observational data. At the first visit, parents completed demographic questionnaires and were given the wearable AW-64 actiwatches (Philips/Respironics, Inc.) to record their nighttime sleep behavior for the next seven consecutive days. Participants were visited again on the 6th day to set up video and audio equipment around the home where infant-parent interactions at bedtime and nighttime typically took place. Between two to four cameras were set up around the home, with at least one being above the infant’s crib. Infrared Color CCD night-vision security cameras by ARM Electronics (Model #C420BCVFIR) and Channel Vision microphones (Model #CV-5104MIC) were used to record video and audio data, the Bosch Divar XF Digital Versatile Recorder (Model #DHR-0800B-150A) for video and audio storage, and the Audiovox D9000 portable DVD player (Model #A-299–1040) to view video recordings for observational coding. Parents were instructed to turn on the video cameras about one hour before the infants’ typical bedtime routine, and turn off the cameras when the infant woke up the next morning. The last visit occurred on the 7th day to collect the actiwatches, video/audio equipment, and the parent-reported questionnaires. Families were visited again during the daytime to obtain video recordings of free-play interaction. These visits usually occurred between late morning and afternoon, planned around the times infants were likely to be awake, per parent report. At 12 and 18 months, researchers separately visited the participants on a different occasion for about two hours for an assessment of infant attachment.

Measures

First-Year Maternal Nighttime Sleep

Actigraphy.

On each occasion (1, 3, 6, 9, 12 months), mothers wore the Respironics AW-64 actiwatch on their wrists for seven consecutive nights, which provided information about mothers’ sleep-wake activity throughout the night. The Sadeh algorithm (Sadeh et al., 1994), which identifies sleep onset as the first of at least three consecutive minutes of inactivity and sleep offset as the last of at least five continuous minutes of inactivity, was used to capture maternal night wakings. Mothers’ fall asleep time (i.e., sleep onset time), midpoint sleep (the midpoint between sleep onset and sleep offset assessed by actigraphy), sleep duration (i.e., the total time from onset to offset of sleep after subtracting awake time after sleep onset), sleep fragmentation (i.e., percentage of inactive sleep time across the designated sleep period), sleep efficiency (i.e., percentage of total sleep time across the designated sleep period), and wake after sleep onset (i.e., total wake time within the sleep period) were calculated from the actiwatch data at each of the seven nights. Reasons for not scoring a single day of actigraphic sleep data included watch errors, incongruence between diaries and actigraphy data, and the watch being off wrist. Then, an average score across seven consecutive nights for each sleep variable was obtained at each time point. The night-to-night standard deviations across the seven days were also calculated for fall asleep time, midpoint sleep, and sleep duration to capture the variability in fall asleep time, variability in midpoint sleep, and variability in sleep duration. At each occasion, average scores and standard deviations of sleep variables were calculated for the mother if there were at least five of the seven days of data (Acebo et al., 1999). In the final sample, the percentage of mothers who provided information on actigraphic sleep variables on at least five days was 92.3% at 1 month, 83.3% at 3 months, 84.6% at 6 months, 82.7% at 9 months, and 74.4% at 12 months. For analysis purposes, sleep variables were each averaged across the five time points in infants’ first year of life in the current study.

Sleep daily diary.

Mothers also reported their daily sleep across the same seven consecutive days at each time point using a sleep daily diary (24-Hour Sleep Patterns Interview [24-HSPI]; Meltzer et al., 2007). This measure showed good reliability and validity (Meltzer et al., 2007). In the current study, maternal bedtime on each night was obtained using a single item (i.e., “What time did you try to fall asleep last night?”). Items assessing maternal wake time (i.e., “What time did you wake up this morning”) and sleep onset latency (i.e., “How long did it take you to fall asleep last night?”) were used with bedtime to calculate maternal subjective fall asleep time (i.e., bedtime plus the sleep onset latency), and time spent in bed (i.e., the elapsed time from bedtime to wake time on each night), sleep period (i.e., the elapsed time from fall asleep time to wake time on each night). At each time point, mean scores and standard deviations across the seven consecutive nights were obtained for bedtime, fall asleep time, time spent in bed, and sleep period. Calculations of sleep variables were conducted for mothers only if at least five days of data were available. The compliance rates on sleep daily diaries were high. In the final sample, the percentage of mothers who provided information on sleep variables on at least five days was 98.7% – 100% at 1 month, 94.9% – 96.2% at 3 months, 94.9% at 6 months, 93.6% – 94.2% at 9 months, and 91.7% at 12 months. Missing data in sleep diaries were due to item nonresponses. The average score of each sleep variable across five occasions in infants’ first year of life was used in further analyses.

Data reduction for sleep variables.

The 16 sleep variables mentioned above were submitted to a Principal Component Analysis (PCA) with oblique (Promax) rotation to reduce sleep data, following the recommendation of previous research (Schoch et al., 2020; Staples et al., 2019). This approach allowed us to comprehensively examine the effect of maternal sleep on maternal parenting and infant attachment without re-running analyses for multiple sleep variables that could be highly correlated with each other. An eigenvalue greater than one was used as the criterion to determine the principal sleep components (Kaiser, 1960). A 4-component solution explaining 85.45% of the variance in total was identified: 1) sleep variability, 2) sleep timing, 3) interrupted sleep, and 4) sleep duration (explaining 40.79 %, 18.25 %, 16.55%, and 9.86% of the variance, respectively), which was consistent with previous research (Schoch et al., 2020; Staples et al., 2019). The component of sleep variability included three actigraphic sleep variables: variability in sleep duration, variability in fall asleep time and variability in midpoint sleep, and four mother-reported sleep variables: variability in bedtime, variability in fall asleep time, variability in time spent in bed, and variability in sleep period. The sleep timing component included four variables: average midpoint sleep assessed by actigraphy, average fall asleep time assessed by actigraphy and mothers’ reports, as well as mother-reported average bedtime. Interrupted sleep was a composite of three actigraphic variables, including average sleep efficiency (reverse scored), average sleep fragmentation, and average wake after sleep onset. The component of sleep duration included actigraphy-assessed average sleep duration, mother-reported average sleep period and average time spent in bed. Using the solution from the PCA, a composite score of each sleep component was created by taking the mean of standardized z-scores of the sleep variables in each component. The internal reliabilities of all component were high: α = .928 for sleep variability, α = .984 for sleep timing, α = .934 for interrupted sleep, and α = .887 for sleep duration. Composite scores for sleep components were also calculated at each time point. Correlational analyses suggested that the sleep components were relatively stable across time points (rs = .21 to .59, ps < .01 for sleep variability, rs = .33 to .59, ps < .001 for interrupted sleep, rs = .64 to .84, ps < .001 for sleep timing and rs = .33 to .70, ps < .001 for sleep duration), supporting our decision to use average scores across the first year.

First-Year Maternal Emotional Availability

Daytime emotional availability.

Video-recorded daytime interactions at 1, 3, 6, 9, and 12 months were used to assess mothers’ daytime emotional availability (EA) towards infants. During the daytime visit, mothers and infants were instructed to play as they normally would for 10-to-15 minutes. Two coders who were blind to other measurements (i.e., bedtime EA, maternal sleep, and infant–mother attachment security) coded maternal daytime EA. The coders participated in the training workshop, completed reliability cases, and received certification for EA Scales from the developer’s laboratory after achieving an acceptable level of reliability with their expert coders. To score emotional availability, we used the Emotional Availability Scales (EAS; Biringen et al., 1998), which consists of four parental dimensions: sensitivity, structuring, non-intrusiveness, and non-hostility. Sensitivity is characteristic of parental behaviors that are prompt, responsive, accurate, and consistent to the child’s needs, both in times of distress and nondistress. Structuring refers to the parent’s ability to balance authority and independence to guide the child to achieve appropriate developmental milestones or learning experiences. Non-intrusiveness is the parent’s ability to engage positively with the child in a manner that is absent of overstimulating, highly arousing physical activities or interferences. Non-hostility refers to the parent’s positive regard for the child and the parent’s capacity to regulate their negative affect toward the child. Non-hostile parents avoid expressing anger, disapproval, impatience, or giving harsh punishments to the child. These four dimensions were scored separately and each standardized into z-scores, which were then combined to create a composite daytime emotional availability score for each participant (the hostility dimension was excluded due to the low reliability; see below). This procedure was repeated at all time points. Correlations of EA amongst the time points (rs = .22 to .53) and the internal consistency of the Emotional Availability Scales (α = .74 to .77) were acceptable (Tavakol & Dennick, 2011). Thus, to reduce measurement error introduced by a single observation (Belsky & Fearon, 2008) and get stable between-person differences in first-year maternal parenting, a mean EA score per mother was created across all first-year time points (1, 3, 6, 9, and 12 months). A mean was obtained only if there were data from at least two (out of five) time points.

In the current project, interrater reliability was performed between two coders on videos of different sets of families at different time points to rule out observer bias created by observing the same families at different points (six from 1 month, six from 3 months, 18 from 6 months, six from 9 months, six from 12 months). Thus, 42 families were included in reliability assessment across the five occasions and comprised about 35% of the total number of families on whom we had video data of daytime free play. In the current sample, intraclass correlations were high across the time points, ranging from .89 to .96, with the exception of non-hostility which was .24. This was due to the fact that hostility occurred infrequently and at low levels, thus creating a range restriction problem. However, there was enough shared variance here that justified creating an average score for daytime EA even after excluding the hostility dimension.

Bedtime emotional availability.

Video-recorded bedtime interactions at 1, 3, 6, 9, and 12 months were used to assess mothers’ bedtime emotional availability (EA) towards infants. Two certified coders who were blind to other measurements (i.e., daytime EA, maternal sleep, and infant–mother attachment security) coded maternal EA. Bedtime observations were typically scored beginning the moment the infant appeared on the camera screen and concluded after five consecutive minutes of the infant falling asleep. Like daytime interactions, we used the Emotional Availability Scales (EAS; Biringen et al., 1998) and the same four dimensions (sensitivity, structuring, non-intrusiveness, and non-hostility) to score emotional availability during bedtime interactions. However, the definition for the structuring dimension was adapted to fit the bedtime context. Structuring during bedtime referred to the parent’s ability to use calming bedtime routines to successfully guide infants toward sleep, while following infant’s leads but setting appropriate limits at the same time (Teti et al., 2010). Like daytime EA, the standardized z-scores from these dimensions were combined to create a composite bedtime emotional availability score for each mother at each time point.

Interrater reliability was performed on randomly selected bedtime videos from each time point (eight from 1 month, eight from 3 months, nine from 6 months, nine from 9 months, eight from 12 months). Thus, 42 families were included in reliability assessment across the five occasions and comprised about 35% of the total number of families on whom we had bedtime video data. Intraclass correlations were high across the time points (ranging from .80 to 1.00) in the current sample. The composite bedtime EA scores were significantly correlated amongst the time points (rs = .25 to .61) and thus were averaged to create a mean bedtime EA score across the first year (1, 3, 6, 9, and 12 months) in order to reduce measurement error introduced by a single observation (Belsky & Fearon, 2008) and get stable between-person differences in first-year maternal parenting. Similar to daytime EA, a mean was obtained if there were data from at least two (out of five) time points.

In the final sample, 88.46% of families provided available bedtime EA, and 74.36% of families provided daytime EA for at least two time points. The missing data of EA was due to reasons such as attrition, parents not consenting to be video recorded, and low video quality for coding. Although we considered using the two child dimensions of emotional availability in our scoring (involvement and responsiveness), it was difficult to score these dimensions at bedtime because the goal of bedtime parenting was to down-regulate infant arousal (rather than stimulate it) in preparing the infant for sleep. Thus, we used parent EA dimensions for both the bedtime and daytime contexts, for consistency and to enable straightforward bedtime-daytime comparisons.

Infant–Mother Attachment

Infants’ quality of attachment to mothers at 12 and 18 months was assessed using the Attachment Q-Set (AQS; Waters et al., 1995). Trained observers, who were blind to the observations of emotional availability and maternal sleep, visited the participants’ homes at each time point and observed mother–infant interactions in an unstructured home setting for approximately 1.5 to 2 hours. Different coders did the AQS observations at both age points. Based on the infants’ behavior observed with mothers during this interaction, the observers scored the 90 items in the AQS into nine categories, from “very much unlike the child” to “very much like the child.” Then, each infant’s sorting was correlated with a security criterion sort (security score of the hypothetically most secure child, based on the Q-sorts of 8 experts on attachment), and this correlation coefficient was used as a final security score for each infant. Examples of items include “If held in mother’s arms, child stops crying and quickly recovers after being frightened or upset” (security criterion of 8.80) and “when something upsets the child, he stays where he is and cries” (security criterion of 1.20).

Compared with the Strange Situation procedure, which assesses secure-based behaviors in a stressful, structured laboratory context with a shorter period of observation time, the observer-rated AQS offers an opportunity to assess attachment behaviors objectively in a naturalistic home environment with a much longer observation (i.e., about 2 hours; Waters et al., 2021). In addition, the AQS conceptualizes and rates attachment security on a continuum, and thus could help reduce potential measurement errors and increase statistical power (Cummings, 2003), compared with categorical assessments of attachment (e.g., the Strange Situation procedure).

In the current study, the final AQS scores for 12 months ranged between −.38 and .75 (M = .39; SD = .23), and between −.31 and .72 (M = .39; SD = .23) for 18 months. To obtain reliability, 30.7% of observations were double coded by a second observer. The intraclass correlation (ICC) for rater pairs was .87 to .96 across the two time points. Because there was a strong association between the two time points (r = .53), AQS scores were averaged across 12 and 18 months to create one attachment variable. In the final sample, AQS was obtained from 91% families at 12 months and 84% families at 18 months. Missing data was caused by families dropping out of the study.

Analytical Strategy

Descriptive statistics (means and standard deviations) and correlations among study variables were calculated using IBM SPSS Statistics 26 in the preliminary analyses. Then, path analyses using the lavaan package (Rosseel, 2012) in R program, were used to test the hypothesized direct and indirect relations between maternal sleep, maternal emotional availability, and infant–mother attachment. Specifically, parallel mediation analyses with two mediators (i.e., bedtime EA and daytime EA) were conducted with each of the maternal sleep components (i.e., sleep variability, interrupted sleep, sleep timing, and sleep duration) as the predictor, maternal bedtime and daytime EA as the mediators, and the average of infant–mother attachment at 12 and 18 months as the outcome (Figure 1). Missing rates of study variables were 5.1% to 25.6% in the final sample. Although Little’s MCAR test was significant, χ²(69) = 90.57, p = .042, the normed chi-square was 1.31, which was acceptable (χ²/df < 2; Ullman, 2007), suggesting our data were likely missing at random. Thus, full information maximum likelihood (FIML) estimation strategy was applied to adjust for missing data to include all available data in analyses (Enders & Bandalos, 2001). Following the advice of Preacher and Hayes (2004) and Shrout and Bolger (2002), we tested the indirect effects using the bias-corrected bootstrapped 95% confidence intervals (CIs) based on 10,000 samples. A significant indirect effect was determined if the confidence interval did not include zero. Unstandardized path coefficients with standard errors (SEs), 95% CIs of unstandardized estimates, and standardized coefficients were reported.

Post-hoc analyses were conducted to compare the magnitude of pathways from maternal sleep to attachment through bedtime EA and daytime EA, when significant direct or indirect effects were detected. Specifically, to test whether the effect of maternal sleep on bedtime EA was different from that on daytime EA (Hypothesis 1), a contrast was made by subtracting the sleep effect on daytime EA from that on bedtime EA. To examine the difference in magnitudes of bedtime and daytime EA effect on attachment (Hypothesis 2), a contrast was made by subtracting the daytime EA effect from bedtime EA effect on attachment. The magnitudes of the indirect effect of maternal sleep on attachment through bedtime EA (bedtime indirect effect) and daytime EA (daytime indirect effect) were also compared with a contrast which is the subtraction of daytime indirect effect from bedtime indirect effect (Hypothesis 3). In all post-hoc analyses, a significant contrast (p < .05) indicates the magnitudes of two effects are not the same, with a positive value suggesting a stronger effect for bedtime EA than daytime EA. Access to the data for the present study can be requested by emailing the corresponding author.

Results

Preliminary Analyses

Table 1 presents sample descriptive statistics and zero-order, bivariate correlations of sleep components, maternal bedtime and daytime emotional availability, and infant–mother attachment. Later maternal sleep timing was correlated with greater variability in sleep (r = .54, p < .001) and shorter sleep duration (r = −.55, p < .001). Greater sleep variability was also significantly associated with shorter sleep duration (r = −.36, p < .001). First-year maternal emotional availability at daytime and bedtime, as well as infant–mother attachment across 12 and 18 months, were all significantly intercorrelated (rs = .36 to .50, p < .001). Table 1 also shows that maternal bedtime EA was significantly linked with all four maternal sleep dimensions, such that higher maternal bedtime EA was correlated with lower sleep variability, lower interrupted sleep, earlier sleep timing and longer sleep duration. Maternal daytime EA was only significantly associated with one maternal sleep dimension, such that higher daytime EA was correlated with lower interrupted sleep. Also, daytime EA’s association with mothers’ sleep variability approached significance, r = −.17, p = .075. More secure attachment was only significantly associated with earlier sleep timing across the first year. The association between second-year attachment and first-year sleep variability approached significance (r = −.16, p = .063). All correlations were in the expected directions.

Table 1.

Descriptive Statistics and Bivariate Correlations among Study Variables

Variable	1	2	3	4	5	6	M	SD
1. Sleep variability, 1 to 12 mos	---						.053	.943
2. Interrupted sleep, 1 to 12 mos	−.025	---					.000	.940
3. Sleep timing, 1 to 12 mos	.535***	−.014	---				.043	1.022
4. Sleep duration, 1 to 12 mos	−.363***	.052	−.553***	---			−.017	.871
5. Bedtime EA, 1 to 12 mos	−.366***	−.218*	−.269**	.187*	---		−.044	3.183
6. Daytime EA, 1 to 12 mos	−. 169†	−.221*	−.088	−.031	.503***	---	−.079	3.015
7. AQS, 12 to 18 mos	−.158†	−.123	−.301***	.112	.359***	.426***	.388	.210

Note. AQS = Attachment Q-Set; Mos = months; EA = maternal emotional availability.

^†p < .10

^*p < .05.

^**p < .01

^***p < .001, two-tailed

Main Analyses

As shown in Figure 1, parallel mediational models were conducted to test whether first year bedtime and daytime EA mediated the association between maternal sleep across infants’ first year and infant attachment security averaged across 12 and 18 months. The model for each sleep component was tested separately.

Sleep Variability

The mediational model with first-year maternal sleep variability as the predictor and first-year bedtime and daytime EA as the mediators in predicting infant attachment security across 12 and 18 months was conducted. Results suggested that greater variability in maternal sleep was associated with lower bedtime EA (b = −0.24, SE = 0.08, 95% CI = [−0.387, −0.088], β = −.36), but not daytime EA (b = −0.08, SE = 0.05, 95% CI = [−0.187, 0.009], β = −.13). After controlling for EA in the other context and sleep variability, maternal bedtime and daytime EA were not significantly associated with infant attachment (b = 0.06, SE = 0.05, 95% CI = [−0.011, 0.165], β = .20; b = 0.10, SE = 0.05, 95% CI = [−0.001, 0.183], β = .28, respectively). However, bedtime EA significantly mediated the association between maternal sleep variability and infant attachment across 12 and 18 months (the indirect effect, see Table 2), such that greater sleep variability in mothers was associated with poorer maternal EA at infant bedtime, which was in turn related to less secure attachment at 12 and 18 months. The indirect effect involving daytime EA was not significant. The direct effect of maternal sleep variability on infant attachment was not significant, while the total effect was significant (Table 2). The final model explained 13.0% of the variance in bedtime EA, 1.6% of the variance in daytime EA, and 18.5% of the variance in attachment.

Table 2.

Estimates of Mediating Effects of the Associations Between Maternal Sleep and Infant Attachment through Maternal Emotional Availability

IV: Sleep Component	Mediator (EA)	Mediated (indirect) effecta			Direct effectb			Total effectc
		b (SE)	95% CI	β	b (SE)	95% CI	β	b (SE)	95% CI	β
Sleep Variability	bed	−0.02 (0.01)	[−0.055, −0.001]	−0.07	−0.01 (0.02)	[−0.049, 0.025]	−0.05	−0.04 (0.02)	[−0.066, −0.005]	−0.16
	day	−0.01 (0.01)	[−0.029, 0.001]	−0.04
Interrupted sleep	bed	−0.01 (0.01)	[−0.036, −0.001]	−0.05	−0.00 (0.02)	[−0.062, 0.033]	−0.02	−0.03 (0.02)	[−0.084, 0.011]	−0.12
	day	−0.01 (0.01)	[−0.035, 0.000]	−0.05
Sleep timing	bed	−0.01 (0.01)	[−0.039, 0.002]	−0.05	−0.05 (0.02)	[−0.104, −0.015]	−0.23	−0.06 (0.02)	[−0.111, −0.034]	−0.30
	day	−0.00 (0.01)	[−0.025, 0.006]	−0.02
Sleep duration	bed	0.01 (0.01)	[−0.002, 0.045]	0.04	0.02 (0.02)	[−0.014, 0.057]	0.09	0.03 (0.02)	[−0.011, 0.063]	0.11
	day	−0.00 (0.01)	[−0.021, 0.008]	−0.01

Note. IV = independent variable; EA = maternal emotional availability; bed = maternal EA during bedtime interaction; day = maternal EA during daytime free play; SE = standard error; CI = confidence interval.

^aThe mediated effect presented the effect of maternal sleep on attachment through maternal EA.

^bThe direct effect presented the effect of maternal sleep on attachment after controlling maternal bedtime and daytime EA.

^cThe total effect presented the summary of the two mediated effects (through bedtime and daytime EA) and direct effect of maternal sleep on attachment.

Significant estimates were bolded.

Interrupted Sleep

The mediational model predicting infant attachment security from maternal interrupted sleep (predictor) and bedtime and daytime EA (mediators) was then conducted. The direct effect of maternal interrupted sleep on maternal bedtime and daytime EA were both significant (b = −0.14, SE = 0.07, 95% CI = [−0.294, −0.020], β = −.20; b = −0.13, SE = 0.06, 95% CI = [−0.260, −0.021], β = −.20, respectively). High bedtime EA was linked with more secure infant attachment across 12 and 18 months even after controlling for daytime EA (b = 0.08, SE = 0.04, 95% CI = [0.010, 0.159], β = .24), but daytime EA’s link with attachment was not significant. As shown in Table 2, the indirect effect of maternal interrupted sleep on infant attachment through bedtime EA, but not daytime EA, was significant, identifying bedtime EA as a mediator of the link between interrupted maternal sleep and infant–mother attachment across 12 and 18 months. The direct and total effects of maternal interrupted sleep on attachment were not significant. The final model explained 4.2% of the variance in bedtime EA, 4.0% of the variance in daytime EA, and explained 19.0 % of the variance in attachment averaged across 12 and 18 months.

Sleep Timing

The mediational model predicting infant attachment security from maternal sleep timing (predictor) and bedtime and daytime EA (mediators) was then conducted. Later maternal sleep timing was significantly associated with lower bedtime EA (b = −0.18, SE = 0.09, 95% CI = [−0.341, −0.003], β = −.29) but not associated with daytime EA (b = −0.04, SE = 0.06, 95% CI = [−0.170, 0.084], β = −.08), and only daytime EA was associated with infant-mother attachment (b = 0.10, SE = 0.04, 95% CI = [0.015, 0.184], β = .29). As presented in Table 2, the direct effect of sleep timing on infant attachment security was significant, such that mothers with later sleep timing across the infants’ first year had infants with less secure attachment across 12 and 18 months. The total effect of maternal sleep timing on attachment was also significant (see Table 2). However, the indirect (mediated) effects of maternal sleep timing on infant attachment via bedtime EA and daytime EA were not significant. This final model explained 8.2% of the variance in first-year bedtime EA, 0.6% of the variance in first-year daytime EA, and 23.9% in attachment across 12 and 18 months.

Sleep Duration

The final mediational model examining bedtime and daytime EA was conducted to examine whether bedtime and daytime EA mediated the association between maternal sleep duration and infant attachment security across 12 and 18 months. The effects of maternal sleep duration on bedtime (b = 0.13, SE = 0.13, 95% CI = [−0.127, 0.360], β = .18) and daytime EA were not significant (b = −0.03, SE = 0.06, 95% CI = [−0.147, 0.080], β = −.04). Maternal bedtime EA was not associated with infant attachment security, but daytime EA was significantly linked to attachment (b = 0.10, SE = 0.05, 95% CI = [0.007, 0.191], β = .29), after controlling for maternal sleep duration. The indirect (mediated) effects of maternal sleep duration on infant attachment security through bedtime and daytime EA were not significant (Table 2), indicating that both bedtime and daytime EA were not significant mediators. No significant direct or total effect of sleep duration on attachment was found. The final model explained 3.3% of the variance in bedtime EA, 0.2% in daytime EA, and 19.7% of the variance in attachment across 12 and 18 months.

Post-hoc Analyses Comparing Daytime and Bedtime Contexts

Post-hoc analyses were conducted to compare the magnitude of pathways from maternal sleep to attachment through bedtime EA and daytime EA. The results revealed that the effect of maternal sleep on bedtime EA was not significantly stronger than that on daytime EA for any sleep dimensions, although the difference approached significance for sleep variability (p = .052). The magnitude of effects of maternal bedtime EA and daytime EA on attachment also were not statistically different from each other, and the bedtime indirect effect was not statistically different from daytime indirect effect (ps > .10). Nevertheless, the present results overall indicated that only in the bedtime context was EA identified as a significant mediator (i.e., statistically different from 0) of the linkage between first-year maternal sleep (i.e., sleep variability and interrupted sleep) and second-year infants’ attachment security to mothers¹.

Discussion

The present study explored longitudinal relations among first-year mothers’ sleep, first-year mothers’ quality of parenting during the day and around bedtime, and infants’ quality of attachment to the mother across 12 and 18 months. Taking a multi-method approach by assessing maternal sleep with objective actigraphy in conjunction with self-reported sleep daily diaries, this study considered multiple characteristics of maternal sleep including sleep variability, timing, duration, and interrupted sleep to conduct a comprehensive investigation of the effects of mothers’ sleep on the quality of mothering and infants’ attachment to the mother.

Consistent with attachment theory (Bowlby, 1969/1982) and prior empirical work (e.g., Kim et al., 2017), mothers’ EA across the year at both daytime and bedtime was predictive of infants’ quality of attachment to the mother at 12 and 18 months. Additionally, mothers’ sleep patterns were predictive of their EA with their infants, but particularly so at bedtime. The poorer the quality (i.e., more interrupted sleep), the later the sleep timing, and the more variability that characterized mothers’ sleep in the first-year postpartum, the lower the independent ratings of mothers’ EA at bedtime with their infants. By contrast, mothers’ daytime emotional availability was predicted by only one maternal sleep index, interrupted sleep averaged across the first year. The more robust associations between maternal sleep and bedtime EA may be because poor, irregular sleep can have debilitating effects on attention, memory, and self-regulation (Banks & Dinges, 2007; Gruber & Cassoff, 2014) that become even more pronounced as the day wears on. Infant bedtime routinely takes place at the end of a long day. It is normal and expected for parents to experience some fatigue by this time, but parents with poor sleep habits may be particularly vulnerable to heightened fatigue and poorer self-regulation at day’s end. When putting their infants to bed, such parents may be less capable of attending to their infants’ needs, less able to maintain consistent bedtime routines, and perhaps less tolerant of and more frustrated by a fussy, uncooperative infant (or uncooperative partner) compared to parents with better sleep habits (Bai et al., 2020).

That mothers’ sleep characteristics were more robustly linked with bedtime EA than with daytime EA in turn helps to explain why bedtime EA more consistently mediated linkages between mothers’ sleep and infant–mother attachment across 12 and 18 months. Bedtime EA significantly mediated the association between mothers’ variability in sleep and interrupted sleep and infant-mother attachment. By contrast, maternal EA during daytime free play did not mediate the association between mothers’ sleep and infant attachment. Again, the more consistent mediation observed with bedtime EA may be because mothers with insufficient and irregular sleep may be more fatigued by the end of the day than parents with healthier sleep habits. Sleep patterns in adulthood tend to be quite persistent (Gaines et al., 2015; Knutson et al, 2007). What the present study indicates is that, among parents of infants, poor sleep habits can lead to problems in parenting, especially when the debilitating effects of poor sleep habits are combined with the effects of fatigue that typically come at the end of a long day, with negative consequences for infant socioemotional development.

Direct Links Between Maternal Sleep and Infant Attachment

Interestingly, the present study found some direct linkages between maternal sleep and infant–mother attachment. Specifically, we found that mothers who went to bed later in infants’ first year had infants who were less securely attached in the second year, after bedtime and daytime maternal EA were controlled. This finding was unexpected, given that there is no theoretical basis for a direct link between maternal sleep and infant–mother attachment. In the present study, daytime EA was only weakly linked with mothers’ sleep, and even though bedtime EA was a more robust mediator between mothers’ sleep and infant attachment, it may be unreasonable to expect that maternal EA in the two contexts studied herein would fully and completely account for links between maternal sleep and infant attachment. Indeed, an infant’s attachment relationship with a caregiver is likely to be the cumulative product of many mother–infant interactions occurring across many contexts across days and weeks. These include, for example, mornings when the infants first wake up, feeding times, and other naturalistic contexts which provide many opportunities for parents to engage in interactions with their infants that could be impacted by variations in maternal sleep. Additional research is needed to understand and identify those naturalistic contexts and situations in which quality of parenting is vulnerable to poor parental sleep, and to assess whether and how sleep-linked differences in parenting quality within and across these contexts are predictive of infant–parent attachment.

We must also consider the possibility that, whereas mothers’ emotional availability was examined only on one day at 1, 3, 6, 9, and 12 months postpartum, mothers’ sleep was assessed across seven consecutive days at each time point. Because of privacy concerns about placing cameras in homes across multiple days and nights, we were not permitted to conduct multiple recordings of parent–infant interaction beyond one day and night at each time point. Thus, to minimize measurement error, the average score across the five time points were used in the current study. It was nevertheless the case that maternal sleep was more comprehensively assessed at each occasion, involving a larger window of time and more data points that contributed to the summary measures, than mothers’ emotional availability. This could have potentially yielded a more reliable and representative measure of maternal sleep than of emotional availability, which in turn could have reduced our ability to demonstrate full mediation by maternal sleep–infant attachment links by maternal emotional availability. In addition, because one’s sleep quality can have differential effects on individual functioning on a daily basis, our inability to assess daily maternal EA meant we were unable to examine the impact of maternal sleep on parenting quality on a day-to-day basis, and the cumulative effects of these daily occasions on 12-month infant attachment.

Strengths and Limitations

The present study has several strengths. First, we employed a longitudinal design, measuring first-year maternal sleep, first-year maternal parenting, and attachment across 12 and 18 months, enabling more clearly defined pathways of influence than those afforded by a cross-sectional approach. Next, we included comprehensive measures of sleep and parenting quality. Sleep was measured extensively—we examined quality, quantity, variability, and timing of mothers’ sleep, relying on both actigraphy data and mothers’ self-reports. On a related note, our study minimized shared method variance by including multiple methods of measurement (sleep actigraphy, self-reported sleep, observer ratings of video-recorded parenting, live Q-sort evaluations of attachment behavior) and separate teams of trained observers to rate all observational measures (mothers’ emotional availability at bedtime, daytime, and infant–mother attachment). The observers were blind to each other’s assessments.

We also acknowledge several limitations. First, our sample was relatively homogeneous in terms of race, consisting of mainly of White families, which limits the generalizability of our study. Second, we again point out the mismatch between the timing of our measures of sleep and emotional availability, with emotional availability being observed only on one day and night at each time point. Due to this limitation, we could not assess the day-to-day consistency of mothers’ parenting quality and determine how representative one evening’s assessment of parenting quality was of parenting quality across multiple nights and multiple contexts. Third, the current study only focused on how mothers’ sleep and parenting influenced infants’ quality of attachment to mothers. We were not able to assess infants’ attachment to fathers because of the limited information on fathers’ parenting due to low rates of participation. This limited our ability to examine linkages between parental sleep, parenting and attachment at a family level, or make comparisons between fathers and mothers.

Conclusion and New Directions

The results of our study emphasize the importance of maternal sleep habits and the impact they can have on the quality of mothering and infant–mother attachment quality. We believe the present findings have significant implications for future studies. To our knowledge, our study is the first to investigate the role that maternal sleep plays in the development of infant–mother attachment. Additional research is clearly needed to examine these linkages more comprehensively. We offer that, in addition to bedtimes and nighttimes, linkages between parent sleep and parenting should be measured more intensively across additional naturalistic contexts in future work, particularly those that may be inherently more stressful and present challenges to parents, such as mornings and mealtimes, when parenting needs to be coordinated with other potentially competing activities. Also, infancy is a period of time during which sleep develops rapidly (Mindell et al., 2016), and maternal sleep can be influenced by infant sleep directly (McDaniel & Teti, 2012) or through sleep arrangements (e.g., bed sharing or room sharing; Teti et al., 2016). Other individual and ecological factors, such as household chaos, family socioeconomic risk, depression, marital discord and other children in the family, can also influence maternal sleep. Given the multiple determinants of parental sleep, there is a need for future studies to identify the plethora of factors that can impact the sleep of mothers during the transition to parenthood. Future work is also needed to examine whether and how these contextual and individual factors influence parenting and parent-child relationship during infancy and beyond through parental sleep.

The present study also emphasizes the need for more parental interventions that specifically target parent sleep as a potential domain for improvement. To our knowledge, parent intervention programs do not commonly target parent sleep hygiene and habits as a means of improving parenting quality. Indeed, evidence is accumulating that parents whose sleep is compromised because of inadequate sleep hygiene are at higher risk for parenting problems than parents who sleep well (Bai et al., 2020; King et al., 2020; McQuillan et al., 2019). The present study further contributes evidence to that effect, and to our knowledge is the first study to demonstrate that sleep-induced parenting problems can impact children’s socioemotional development. We believe that such a focus in parenting intervention is needed and overdue.