Adding fuel to the fire: How childhood and adolescent histories of interparental conflict amplify sleep problems during the transition to emerging adulthood

Abstract

The study examined whether change in childhood interparental conflict (IPC) moderated links between change in adolescent IPC and sleep from adolescence to emerging adulthood. Participants (N = 351, 52% female, 67% European American, 32% African American, 1% biracial) from diverse socioeconomic backgrounds in the Southeastern United States completed five waves of data collection (M_ages = 8, 10, 15, 17, 22) between 2005 and 2022. Actigraphs measured sleep duration and quality. Participants reported on parents’ IPC. Latent change score models examined change-on-change in IPC and sleep. Among those with histories of increasing childhood IPC, increases in adolescent IPC predicted increases in sleep activity (β = .24) and long-wake episodes (β = .22) and decreases in sleep efficiency (β = −.20) from adolescence to emerging adulthood. Findings are consistent with stress sensitization.

Sleep is a key bioregulatory process, and growing evidence highlights its importance as a fundamental domain of development (Boatswain-Jacques et al., 2023; Park et al., 2019). Sleep problems—characterized by short duration and poor quality—in youth increase the risk for adverse developmental outcomes (Cooper et al., 2023; Scott et al., 2021). To better identify youth at risk for sleep problems, developmental models and research are needed to examine their antecedents. Exposure to family conflict has been associated with poor sleep (Gordon et al., 2021; Khor et al., 2021). A growing literature has considered the role of the interparental relationship as a unique source of stress that may interfere with youths’ sleep. Although previous research indicates that destructive interparental conflict (IPC; i.e., intense and frequent psychological and physical aggression between parents or guardians) predicts sleep problems in youth (El-Sheikh et al., 2006; Kelly et al., 2024; Nowakowski et al., 2016), the modest effect sizes point to substantial individual differences and emphasize the need to identify sources of variability to better understand which youth are at heightened risk for worsening sleep problems (e.g., El-Sheikh et al., 2015; Kelly & El-Sheikh, 2013). Moreover, most studies have used cross-sectional or short-term longitudinal designs, and none have assessed how change in IPC is associated with change in sleep—a critical gap in the field (El-Sheikh & Kelly, 2017). Our study utilized latent change score (LCS) modeling to examine individual differences in intraindividual change over time to identify moderators of the prospective association between changes in adolescent IPC (from age 15 to age 17) and changes in sleep occurring across the transition from adolescence to emerging adulthood (from age 17 to age 22).

A small but growing literature suggests that distal IPC histories in childhood act as an emotional primer or sensitization factor, increasing vulnerability to IPC during subsequent developmental periods. Early exposure to intense and frequent IPC can establish enduring emotional, cognitive, and physiological response patterns that persist even if later IPC decreases (Rudolph & Flynn, 2007; van Eldik et al., 2020). Accordingly, IPC histories may heighten children's vulnerability to conflict in later developmental periods, as evidenced by their increased emotional, cognitive, and physiological reactivity to later inter-adult anger and family conflict (Cummings et al., 1989; El-Sheikh, 1994; Sloan et al., 2022). To our knowledge, only one study has assessed childhood IPC histories as a moderator of the link between recent IPC and negative emotional reactivity during a later developmental period (Davies et al., 2021). Little attention has been given to how childhood IPC histories affect longitudinal pathways between later IPC and sleep. Addressing these gaps, we examined whether changes in IPC histories during childhood (from ages 8 to 10) and IPC in adolescence interact to predict change in sleep from adolescence to emerging adulthood.

Sleep is a multifaceted construct (Sadeh, 2015), and its parameters have distinct developmental and clinical implications (Boatswain-Jacques et al., 2023; Thompson et al., 2024). Based on links between IPC and youths’ sleep duration and quality (Kelly et al., 2024; Nowakowski et al., 2016), we selected established actigraphy-based measures to assess these parameters. For sleep duration, we assessed sleep minutes—the total number of minutes scored as sleep between sleep onset and wake time. For sleep quality, we examined sleep efficiency (the percentage of time between sleep onset and wake time scored as sleep), sleep activity (the percentage of epochs with activity), and long-wake episodes (LWE; the number of wake episodes lasting ≥ 5 min).

Interparental conflict and sleep

Sleep problems in childhood and adolescence persist across developmental periods (Hysing et al., 2020), and IPC is a known risk factor for sleep problems. Conceptually, IPC could disrupt youths’ sleep by undermining their sense of safety and security in the interparental subsystem, increasing hypervigilance and emotional arousal to IPC (El-Sheikh & Kelly, 2017; Kelly & El-Sheikh, 2013). Research shows that IPC has both concurrent and longitudinal predictive effects on sleep in youth. At age 8, IPC was concurrently associated with short, poor-quality sleep (El-Sheikh et al., 2006) and prospectively predicted sleep problems two years later (Kelly & El-Sheikh, 2011). In adolescence, IPC at age 15 predicted shorter sleep one year later (Nowakowski et al., 2016). Recent evidence further suggests that IPC in adolescence (ages 16–18) predicted poor sleep quality into emerging adulthood (Kelly et al., 2024). Many of these associations, however, are small to modest and vary based on factors such as children's emotional security (Kelly & El-Sheikh, 2013), autonomic nervous system activity (El-Sheikh et al., 2015), and reporter of conflict (El-Sheikh et al., 2006). Collectively, heterogeneity in the predictive effects of IPC on sleep highlights the continued need to identify which youth are most vulnerable to worsening sleep problems.

The moderating role of childhood histories of interparental conflict

One potential source of heterogeneity is variability in the nature and frequency of IPC across developmental periods. The current investigation is guided by a stress sensitization framework (Monroe et al., 2019; Morris et al., 2010), which posits that increased exposure to IPC in childhood may heighten sensitivity to subsequent IPC in adolescence, thereby contributing to worsening sleep problems from adolescence to emerging adulthood. These long-term alterations in sleep-wake processes may arise, in part, through the accumulation of allostatic load—the multisystemic wear and tear on physiological and biological stress response systems that depletes regulatory resources and contributes to dysregulation and heightened reactivity to future stressors (Evans, 2003; Kelly & El-Sheikh, 2019). Allostatic load carries long-term consequences for the body's capacity for restoration and repair (Monroe, 2008) and is specifically proposed to alter sleep-wake processes (Juster & McEwen, 2015), which play a central role in maintaining homeostasis (McEwen, 2006). Although stress sensitization models have been applied to other developmental outcomes, they have not been previously examined in relation to sleep. Thus, we extend prior work to determine whether IPC in childhood sensitizes youth to later IPC in ways that increase risk for worsening sleep problems.

Within stress sensitization frameworks, childhood stress is hypothesized to increase vulnerability in three primary ways (Morris et al., 2010). The stress amplification model posits that increases in childhood stress heightens vulnerability to psychopathology as subsequent stressors intensify (Morris et al., 2010; Rudolph & Flynn, 2007). Increases in childhood stress heighten vulnerability to the negative effects of escalating proximal stress in adolescence, whereas decreases (or stability) in early stress exposure predicts corresponding decreases (or stability) in maladaptation across rates of change in recent stress levels (Figure 1). In our study, stress amplification would be supported if increases in childhood IPC histories forecast increases in sleep problems from adolescence to emerging adulthood amid escalating adolescent IPC, whereas decreases (or stability) in childhood IPC predicts decreases (or stability) in sleep problems regardless of rates of change in adolescent conflict.

Figure 1.

Prototypical depictions of stress sensitization models.

Likewise, the stress activation model proposes that increases in childhood stress strengthen the association between proximal stress and maladjustment, regardless of rates of change in recent stress (Morris et al., 2010; Rudolph & Flynn, 2007). Increases in childhood stress are posited to increase vulnerability across all rates of change in recent stress, thus lowering the threshold of proximal stress needed to precipitate change in maladaptation (Figure 1) (Morris et al., 2010; Rudolph & Flynn, 2007; Stroud, 2020). Accordingly, the stress activation model hypothesizes greater differences between youth with increasing versus decreasing (or stable) childhood stress across increases and decreases in recent stress. The stress amplification model, in contrast, predicts larger differences specifically as recent stress intensifies. In the current study, stress activation would be supported if increases in childhood IPC strengthen the association between changes in adolescent IPC and sleep problems from adolescence to emerging adulthood, regardless of rates of change in adolescent conflict.

Last, the risk saturation model theorizes that as childhood stress increases, vulnerability increases across all rates of change in proximal stress, similar to the stress activation model (Morris et al., 2010). Conversely, the risk saturation model hypothesizes a weakened or negligible association between changes in proximal stress and changes in maladjustment in these youth (Figure 1). Among youth with decreasing (or stable) childhood stress, maladjustment increases as stress in adolescence increases, resulting in larger differences between youth with increasing versus decreasing (or stable) childhood stress specifically amid declines in adolescent stress. Applied to this study, youth with increasing childhood IPC are at risk for worsening sleep problems from adolescence to emerging adulthood regardless of shifts in adolescent IPC.

Although prior studies have yet to examine change-on-change processes in IPC-related stress sensitization, evidence from prior work still offers relevant insight. Supporting stress sensitization, studies indicate that children with high IPC exhibit heightened emotional, behavioral, and physiological reactivity, as well as more negative appraisals and mood, in response to real-time simulated inter-adult conflict (Cummings et al., 1989; El-Sheikh, 1994) or daily IPC (Chiang et al., 2023). More robust analyses (e.g., simple slopes, regions of significance [RoS]) are needed to differentiate among the three stress sensitization models. Preliminary support exists for stress amplification models; adolescents reporting high but not low global IPC reported increased threat with high but not low daily IPC (Sloan et al., 2022). Notably, global IPC assessments were utilized as proxies for understanding how IPC histories operated as a source of variability in later responses. Addressing this gap, Davies and colleagues (2021) found that adolescents with high childhood IPC histories exhibited greater increases in negative emotional reactivity regardless of concurrent IPC. Adolescents with low childhood IPC histories showed greater emotional reactivity only under high concurrent IPC conditions. Consistent with the risk saturation model, the association between concurrent IPC and emotional reactivity was significant for adolescents with low but not high childhood IPC histories. Collectively, these studies provide evidence for sensitivity processes in youths’ reactivity to IPC. Little is known about how childhood IPC histories inform the predictive effects of proximal IPC in adolescence on sleep during the transition to emerging adulthood. Expanding this literature, we examined change in childhood IPC histories as a source of individual differences in the prospective pathway between changes in proximal IPC in adolescence and changes in sleep—a key bioregulatory process—from adolescence to emerging adulthood.

Present study

Translating theory into analytic models requires attention to the inherently dynamic nature of development. The stress sensitization model, for example, emphasizes the severity and frequency of stressors required to elicit maladaptation change over time (Monroe et al., 2019; Stroud, 2020). Extending this perspective to the current study, we examine parallel developmental processes in IPC and sleep—both of which have been conceptualized as dynamic and fluctuating across development (Fincham et al., 1994; Thompson et al., 2024). Indeed, it is change, or intensification, in IPC over time through which stress sensitization is posited to operate, rather than conflict occurring within any given static developmental window (Davies et al., 2018). Stress sensitization frameworks, therefore, caution against inferring sensitization from static between-person differences in vulnerability (Monroe et al., 2019; Stroud, 2020). Yet few studies have employed statistical methods capable of capturing change-on-change processes in IPC and sleep (El-Sheikh & Kelly, 2017) or sensitization processes (Monroe et al., 2019), instead relying on residualized change models that cannot assess how upstream changes in one construct relate to downstream changes in another. This distinction is critical when modeling dynamic developmental processes. As described by Fincham and colleagues (1994), changes in conflict are theorized to be particularly informative for understanding changes in adaptation. For example, large changes in IPC should correspond to relatively large changes in sleep, whereas small changes in IPC are expected to predict small changes in sleep. Importantly, the magnitude of change in IPC does not necessarily relate to poor or optimal sleep. To address these gaps and align our analytic approach with the dynamic nature of our constructs and guiding conceptual model, we used LCS modeling to capture individual differences in intraindividual change in IPC and sleep. Specifically, we tested whether change in childhood IPC serves as a vulnerability factor that alters the predictive effect of change in adolescent IPC on changes in sleep during the transition from adolescence to emerging adulthood.

Our study offers the first assessment of change-on-change dynamics in IPC and sleep to date. We examined IPC during middle childhood and adolescence for several reasons. Middle childhood is a sensitive period when environmental variations interact with emerging individual differences (e.g., gene expression; hormone-sensitive neural pathways), shaping enduring regulatory capacities, and adolescence allows for refining these regulatory strategies prior to maturation (Del Giudice, 2018). Findings on sleep changes from adolescence to emerging adulthood are mixed. Some studies report improvements in sleep duration and quality (Chen & Chen, 2021; Maslowsky & Ozer, 2014), while others find decrements (Keyes et al., 2015; Park et al., 2019). This inconsistency underscores the need to identify sources of variability in sleep problems during this transition. Moreover, emerging adulthood presents additional sleep barriers due to the convergence of biological (e.g., circadian phase delay, sleep-wake homeostasis) and psychosocial (e.g., academic and work schedules, lifestyle choices) processes (Carskadon & Tarokh, 2013).

The study incorporates a balanced blend of confirmatory and exploratory elements. Reflecting the confirmatory component, the study is grounded in stress sensitization models and their empirical application to IPC. Accordingly, we hypothesized that increasing childhood IPC would moderate the link between changes in adolescent IPC and changes in sleep from adolescence to emerging adulthood in a manner consistent with stress sensitization. However, extending these models to IPC and youths’ bioregulatory functioning—namely their sleep—is novel and untested, and limited empirical research differentiates how stress sensitization processes operate specifically in the context of IPC. Therefore, the precise nature of this moderating effect remained exploratory, and we did not hypothesize about its specific form.

Method

Participants

Data for this study were derived from an ongoing longitudinal study focused on family stress and youth development among rural and suburban families in the Southeastern United States. To be eligible, children had to be from two-parent households and attending public elementary schools within the region. Children whose parents divorced during the study were not excluded; 62 parents reported a divorce throughout the study. Children with intellectual disabilities or chronic health conditions (e.g., ADHD, sleep apnea) were ineligible to reduce potential confounds. The current study includes five waves of data spanning childhood (Waves 1 & 3), adolescence (Waves 4 & 6), and emerging adulthood (Wave 7). Data were collected between 2005 and 2022. For the current study, waves within each developmental period were collected approximately two years apart whereas waves between developmental periods were separated by approximately five years. The current study used data from Waves 1, 3, 4, 6, and 7 to capture the first and last waves of childhood and adolescence, providing a broader developmental window to detect meaningful change and variability—an important consideration given the relative stability of family environments (Repetti et al., 2011). Waves 6 and 7 additionally capture the transition from adolescence to emerging adulthood.

Demographic information about participants is provided in Table 1. At the beginning of the study (Wave 1), 251 children were enrolled. Of these, 194 (77%) returned for Wave 3. At Wave 4, 53 families similar in demographics and from the same school districts as the original cohort were added to account for study attrition due to factors such as family relocation or inability to contact families, bringing the total to 247 participants. These 53 participants did not differ from the original cohort based on age, sex, race, or family socioeconomic status (SES). Wave 6 included 220 participants, 210 (85%) of whom returned from Wave 4. Wave 7 included 204 returning participants, 185 of whom (84%) had participated at Wave 6. At Wave 7, 47 individuals were added to account for study attrition, bringing the total Wave 7 sample to 251. These 47 individuals were similar in sex, race, and family SES to the original cohort but were slightly older (M = 23.89, SD = .63) than the existing cohort (M = 22.76 years, SD = .70), t(249) = −10.15, p < .001).

Table 1.

Demographic and sleep information for participants at each wave.

	Wave 1	Wave 3	Wave 4	Wave 6	Wave 7
	(N = 251)	(N = 194)	(N = 247)	(N = 220)	(N = 251)
Predictor	M (SD) or %	M (SD) or %	M (SD) or %	M (SD) or %	M (SD) or %
Age in years	8.23 (0.72)	10.26 (0.78)	15.27 (0.87)	17.19 (0.79)	22.47 (0.76)
Race
African American/Black	36%	36%	34%	30%	28%
European American/White	64%	64%	66%	70%	69%
Mixed	—	—	—	—	3%
Sex
Male	49%	49%	47%	45%	43%
Female	51%	51%	53%	55%	57%
SES/INRa	37.38 (9.92)	37.87 (10.40)	2.39 (1.30)	3.03 (1.83)	2.71 (2.16)
Parents divorcedb (n)	—	6	46	10	—
Sleep data
Nights Available	—	—	—	5.37 (1.61)	5.70 (1.73)
Missing	—	—	—	4%	6%

^aSocioeconomic status was measured using self-report at Waves 1 & 3 and income-to-needs ratio at Waves 4 & 6.

^bParental divorce rates account for new divorces between waves included in the current study; no parental reports are available at Wave 7.

The current study was sampled from a wide range of socioeconomic backgrounds. SES was measured using a standardized aggregate of SES across all included waves of the study, including Hollingshead SES scores (i.e., SES based on parental education level and job type; Hollingshead, 1975) at Waves 1 and 3 and income-to-needs ratio (INR; total family income in relation to household size) at Waves 4, 6, and 7. Information to calculate INR was not available at Waves 1 and 3. Parents reported on SES in Waves 1–6, and participants reported on household SES at Wave 7. On average, participants from Waves 1 and 3 were from lower-middle class socioeconomic backgrounds (M = 37.59, SD = 10.12, range = 11–66), with most participants reporting a family household income between 25 and 50 k. This was comparable to the reported family SES in Waves 4 and 6, wherein the average income-to-needs ratio was 2.69 (SD = 1.60, range = 0–8.91), corresponding to lower-middle class. At Wave 7, when participants were in emerging adulthood, the average income-to-needs ratio was similar (M = 2.71, SD = 2.16, range = 0.04–18.75), but there was considerably more variability.

Participants who had data available on the main variables at any time point for this study were included in the final sample, resulting in a final analytic sample of 351 (52% female, 32% AA, 67% EA, and 1% Biracial). Note, 123 (35%) individuals participated in all five waves of data collection. See Table 1 for Ns during each wave.

Procedures

Participants were recruited through public school systems based on eligibility criteria and contacted by research personnel. Youth assent and parental consent were obtained for Waves 1–6; for Wave 7, consent was obtained from participants. During Waves 1 and 3, parents and children completed questionnaires in the lab. During Waves 4–7, participants were mailed an actigraph and daily diaries to complete for seven nights to validate actigraphy data. After this period, participants were invited to the lab to complete a series of questionnaires, often within a few days of completing actigraphy wear (M = 2.48, SD = 10.69). Parents also completed questionnaires for Waves 4 and 6. The option to complete questionnaires online was provided for Waves 4, 6, and 7. All study procedures were approved by the Auburn University Institutional Review Board.

Measures

Interparental conflict

IPC was measured at each wave using participants’ reports of the Conflict Tactics Scale (CTS2; Straus et al., 1996). Participants responded to how often they were exposed to parents’ (mother and father) physical and psychological aggression toward each other within the past year. Participants responded to each item on the Physical Aggression (12 items; e.g., “Slammed against a wall”) and Psychological Aggression (8 items; e.g., “Insulted or swore at”) subscales using a 7-point scale from 1 (Once in the past year) to 7 (Not in the past year, but it did happen before). In the case of parental divorce, participants reported on current parents in the home in childhood and adolescence. For each wave, scores were aggregated across type of aggression as well as across parent to create an overall IPC score. Internal consistency was good across all waves (αs ranged from .84 to .96). Overall IPC scores were used to create latent change scores (see Analytic Plan for details) that capture changes in IPC during childhood and adolescence for the predictors of interest.

Sleep

To measure sleep during the transition from adolescence to emerging adulthood (i.e., Waves 6 and 7), objective sleep parameters were obtained and calculated using Octagonal Basic Motionlogger actigraphs and ACTme software (Action W2, Ambulatory Monitoring, Ardsley, NY, USA). Actigraphy was not utilized in childhood. Sleep was scored as motion captured within 1-minute epochs with zero crossing mode. Per established guidelines and to address changes in sleep with age, we used Sadeh scoring algorithms in adolescence (Sadeh et al., 1994) and Cole-Kripke scoring algorithms in emerging adulthood (Cole et al., 1992). Sleep duration and quality were derived based on the manual that accompanied the software and were similar across both scoring algorithms. Sleep duration was assessed using sleep minutes (i.e., total number of minutes scored as sleep from onset to wake time). Sleep quality was indexed by sleep efficiency (i.e., sleep percent from onset to wake time), sleep activity (i.e., percentage of epochs with motion activity), and LWE (i.e., number of wake episodes that were equal to or exceeded 5 minutes). Sleep diaries were used to corroborate sleep onset and wake times.

Participants wore actigraphs for a maximum of seven nights. To be eligible for consideration in sleep analyses, participants were required to have a minimum of three nights of actigraphy data; otherwise, their data were classified as missing. Nights when participants reported using medications due to acute illness were removed. The average number of nights of sleep data and percentages of missing data per wave are available in Table 1. Across both waves, participants had on average 5.55 (SD = 1.68) nights of sleep data, and the mode of nights available per wave ranged from 6 to 7. The percentage of missing sleep data ranged from 4–6% across waves, often due to the participants forgetting to wear the actigraph watch. Actigraph malfunctions and discrepancy between sleep diaries and actigraphy-based sleep onset also occurred but were rare.

Covariates

To reduce potential confounds, four covariates were included in all models: race (0 = EA, 1 = AA, 2 = Biracial); biological sex (0 = female, 1 = male); SES; and IPC in emerging adulthood (i.e., Wave 7). SES was standardized to account for different measures across waves (i.e., self-report and income-to-needs ratio) and aggregated across included waves of the study.

Analytic plan

Prior to analyses, all main variables of interest were screened for potential outliers at each wave. Outliers exceeding ± 3 SDs above the mean were winsorized, thereby rescoring outliers to the values corresponding to ±3 SD. Following preliminary outlier screening, analyses for the current study were conducted in two phases. First, following well-established procedures (McArdle, 2009), LCS modeling was used to calculate variability in change (i.e., individual differences in intraindividual change) among main variables of interest. LCS modeling captures both the average change and the variability around the average level of change for predictors and outcomes of interest. Two unconditional LCS models were estimated to create latent change scores for IPC corresponding to childhood and adolescent developmental periods. Each model used the first and last waves of its respective developmental period (childhood: Waves 1 and 3; adolescence: Waves 4 and 6) to provide a broader window for detecting meaningful change and variability. For our outcomes of interest, four additional unconditional LCS models were conducted to assess latent change in sleep minutes, sleep efficiency, sleep activity, and LWE. To examine the transition from adolescence to emerging adulthood and create a lagged effect between IPC in adolescence and change in sleep, we utilized the last wave in adolescence (Wave 6) and the first wave in emerging adulthood (Wave 7). Per established guidelines (McArdle, 2009), each unconditional LCS model regressed the later measurement of a construct on both its prior level and latent change factor. Both paths were fixed to 1 to define the later measurement as the sum of its prior level and latent change factor, thereby ensuring that the change factor represents true intraindividual change and that its variance reflects individual differences in that change. Figure 2 presents the conceptual analytic model.

Figure 2.

Illustration of the conceptual analytic model. (A) illustrates the measurement of two separate latent change score models for interparental conflict (IPC): one estimating change in IPC across childhood and a second estimating change in IPC across adolescence. These latent change scores were estimated independently, saved, and subsequently used as observed variables in multiple regression models. As shown in (B), multiple regression models tested whether the interaction between change in childhood IPC and change in adolescent IPC predicted latent changes in sleep outcomes from adolescence to emerging adulthood.

In the second phase, latent change scores were used to conduct moderation analyses. Following established guidelines and prior work (DiStefano et al., 2009; Ng & Chan, 2020; Pearson et al., 2023; Skrondal & Laake, 2001), we tested study hypotheses in multiple regression models that incorporated saved latent change scores for IPC (Figure 2A) while retaining full LCS modeling for sleep outcomes. Moderation analyses via multiple regression were used to determine whether the interaction between latent changes in childhood IPC and latent changes in adolescent IPC significantly predicted latent changes in sleep outcomes from adolescence to emerging adulthood (Figure 2B). We examined each sleep outcome in a separate model for a total of four moderation models. We further specified race, sex, average SES, and IPC in emerging adulthood as covariates. Models were run with and without the addition of covariates. Model results were generally consistent across inclusion and exclusion of covariates—the pattern of main effects, moderation effects, and simple slopes did not differ. Models including covariates were reported. All predictors and covariates were grand-mean centered to facilitate interpretation. Significant interactions were further probed using simple slopes analyses (Aiken & West, 1991) and region of significance calculations (Roisman et al., 2012).

All analyses were conducted in Mplus v 8.9. Prior to running LCS models, data were assessed for patterns of missingness. Little's MCAR tests indicated that the data were not missing completely at random, χ² (366) = 421.37, p = .02. Greater rates of missing data were associated with being male (r = .12, p = .02), lower average SES (r = −.19, p < .001), and higher levels of IPC at Wave 6 (r = .20, p = .004). Across all study variables, 33.39% of data were missing. Therefore, we proceeded to use full-information maximum likelihood to capitalize on all data available when conducting analyses (Little et al., 2014).

Results

Descriptive statistics and correlations among main variables of interest can be found in Table 2. Results from bivariate correlations revealed that IPC and sleep outcomes were not correlated across time points. Higher SES was positively associated with Wave 6 sleep minutes and efficiency and negatively associated with Wave 6 LWE. Being AA was associated with shorter sleep duration at Waves 6 and 7, lower sleep efficiency at Wave 6, and more long-wake episodes at Wave 6. Being male was also associated with shorter sleep duration at Waves 6 and 7, lower sleep efficiency at Wave 6, as well as higher sleep activity at Wave 6.

Table 2.

Correlations and descriptives for study variables.

	M	SD	1	2	3	4	5	6	7	8	9	10	11	12	13
Covariates
1. Sex	.48	.50	—
2. Race	.34	.50	−.08	—
3. SES	−.06	.78	.08	−.29***	—
4. IPC—emerging adulthood	1.09	2.19	−.18**	−.05	−.07	—
Predictors
5. IPC—childhood	−.02	3.25	−.07	.05	.00	−.04	—
6. IPC—adolescence	−.64	2.96	.01	−.11	.05	−.08	−.06	—
Sleep outcomes
7. Sleep minutes—Wave 6	394.34	60.68	−.26***	−.28***	.14*	.10	.02	.03	—
8. Sleep minutes—Wave 7	406.49	70.20	−.22**	−.23**	.13	.00	.21*	.05	.20*	—
9. Sleep efficiency—Wave 6	91.04	7.41	−.18*	−.18**	.16*	.08	.00	−.11	.61***	.01	—
10. Sleep efficiency—Wave 7	93.83	5.49	−.01	−.13	.13	−.04	.00	−.02	.10	.28***	.26**	—
11. Sleep activity—Wave 6	34.88	12.38	.17*	−.04	−.11	−.06	.05	.08	−.44***	.05	−.74***	−.19*	—
12. Sleep activity—Wave 7	40.26	13.98	.10	−.07	−.09	.03	.03	.04	−.07	−.22**	−.11	−.74***	.30***	—
13. Long wake episodes—Wave 6	2.25	1.92	.10	.15*	−.19**	−.06	.01	.11	−.46***	.02	−.94***	−.27***	.72***	.10	—
14. Long wake episodes—Wave 7	1.76	1.69	−.03	.12	−.12	−.02	.03	.05	−.08	−.15*	−.29***	−.94***	.20*	.71***	.29***

Note. Childhood: Δ from Waves 1 to 3; Adolescence: Δ from Waves 4 to 6; Emerging Adulthood: Wave 7; SES = socioeconomic status; IPC = interparental conflict. * p < .05, ** p < .001, *** p < .001.

Following preliminary descriptives and correlations, unconditional LCS models were conducted to determine whether there was significant variability among the predictors and outcomes of interest. Results for all unconditional LCS models can be found in Table 3. As all unconditional LCS models were fully saturated, no model fit statistics were reported. There was no significant average change in childhood IPC, but there was a significant decrease in adolescent IPC. There was significant variability in average change. In other words, participants demonstrated significant variation in how their exposure to IPC changed within each developmental period when compared to the overall mean level of IPC.

Table 3.

Unstandardized parameter estimates from unconditional latent change score models.

Variable	Average change (µ)	Variability in change (σ2)
Δ IPC—childhood	−0.02	13.06***
Δ IPC—adolescence	−0.64**	10.50***
Δ Sleep minutes	12.29*	6798.38***
Δ Sleep efficiency	2.82***	60.77***
Δ Sleep activity	5.29***	237.81***
Δ Long-wake episodes	−0.53**	4.34***

Note. IPC = interparental conflict. * p < .05, ** p < .01, *** p < .001.

Regarding changes in sleep during the transition from adolescence to emerging adulthood, participants generally experienced an improvement in sleep. Specifically, we observed a significant increase in sleep minutes and sleep efficiency as well as a significant decrease in LWE. Conversely, we found a significant increase in sleep activity from adolescence to emerging adulthood. There was significant variability around the mean level of change for all sleep variables of interest, suggesting significant variability around the mean for sleep minutes, sleep efficiency, sleep activity, and LWE during the transition to emerging adulthood. Given that significant variability in latent change scores is recommended for using LCS models (Hinnant et al., 2021), we proceeded to use LCS scores for IPC and sleep in multiple regression models. Results for each model can be found in Table 4.

Table 4.

Regression results for sleep outcomes of interest.

Predictor	Sleep minutes	Sleep efficiency	Sleep activity	Long-Wake episodes
	β (SE)	β (SE)	β (SE)	β (SE)
Intercept	−.05 (.06)	−.02 (.05)	.02 (.07)	−.01 (.05)
Sleep Wave 6	−.67 (.05)***	−.75 (.06)***	−.52 (.07)***	−.65 (.07)***
Race	−.21 (.06)**	−.08 (.05)	−.04 (.06)	.07 (.05)
Sex	−.18 (.06)**	−.01 (.05)	.07 (.07)	−.02 (.05)
SES	.05 (.05)	.03 (.04)	−.06 (.05)	−.02 (.05)
IPC—emerging adulthood	−.01 (.06)	−.04 (.06)	.03 (.07)	−.01 (.06)
Δ IPC—childhood	.19 (.07)**	−.10 (.06)	.16 (.09)	.14 (.08)
Δ IPC—adolescence	−.01 (.06)	−.08 (.06)	.10 (.07)	.11 (.06)
IPC—childhood × adolescence	−.03 (.06)	−.20 (.07)**	.24 (.08)**	.22 (.08)**

Note. SES = socioeconomic status; IPC = interparental conflict. ** p < .01, *** p < .001.

Sleep minutes

Regarding main effects, change in childhood IPC significantly predicted change in sleep minutes such that increases in childhood IPC predicted an increase in sleep minutes from adolescence to emerging adulthood. Change in adolescent IPC was not associated with change in sleep minutes. Additionally, there was no moderation effect between changes in childhood IPC and adolescent IPC for change in sleep minutes.

Sleep efficiency

There were no main effects for either change in childhood or adolescent IPC on change in sleep efficiency. However, the interaction between adolescent IPC and childhood IPC was significant. Simple slopes analyses at ±1 SD revealed that having a history of increases in childhood IPC exacerbated the association between increases in IPC in adolescence and decreases in sleep efficiency (+1 SD, b = −.65, p = .009) when compared to those with a history of constant (b = −.22, p = .17) or decreasing (−1 SD, b = .22, p = .13) IPC in childhood. In other words, adolescents who had been exposed to increasing childhood IPC experienced the greatest declines in sleep efficiency during the transition to emerging adulthood (Figure 3A). To further understand this significant moderation effect, we examined the regions of significance on the moderator (i.e., RoS on Z) (Figure 3A). The RoS on Z test measured the significance of IPC in adolescence as a predictor of changes in sleep efficiency along the full range of possible childhood IPC values rather than probing only two specific values (e.g., childhood IPC values at ±1 SD of the mean). Findings from the RoS on Z test confirmed simple slope findings. The positive association between changes in IPC in adolescence and changes in sleep efficiency was significant as changes in IPC scores in childhood reached values of .001 or higher (i.e., .0003 SD above the mean or greater). Given that regions of significance plots can sometimes present ambiguity concerning plot formation, a complementary test of the regions of significance on the predictor (i.e., RoS on X) was examined to confirm the form of the interaction (Figure 4A) (Roisman et al., 2012). Inverting the RoS on Z test, the RoS on X test examined the significance of the association between our moderator, childhood IPC, and sleep efficiency along all possible values of our predictor, IPC in adolescence. Changes in childhood IPC were positively associated with changes in sleep efficiency when changes in IPC in adolescence were .0004 or higher (i.e., .0001 SD above the mean or greater).

Figure 3.

Interaction of interparental conflict (IPC) in adolescence and childhood predicting sleep quality during the transition to emerging adulthood. Panel A shows the simple slopes and regions of significance (RoS) on Z test for change in sleep efficiency. Panel B depicts the simple slopes and RoS on Z test for change in sleep activity. Panel C illustrates the simple slopes and RoS on Z test for change in long-wake episodes. The shaded regions represent the range of childhood IPC change values where the association between change in adolescent IPC and change in sleep is statistically significant.

Figure 4.

Region of significance (RoS) on X for the interaction of interparental conflict (IPC) in adolescence and childhood predicting sleep quality during the transition to emerging adulthood. Panel A shows the RoS on X test for change in sleep efficiency. Panel B depicts the RoS on X test for change in sleep activity. Panel C illustrates the RoS on X test for long-wake episodes. The shaded regions represent the range of adolescent IPC change values where the association between change in childhood IPC and change in sleep is statistically significant.

Sleep activity

Similar to sleep efficiency, there were no main effects of change in childhood or adolescent IPC on change in sleep activity; yet there was a significant moderation effect such that the interaction between change in adolescent IPC and childhood IPC significantly predicted change in sleep activity. Simple slopes analyses at ±1 SD were conducted to clarify the pattern of moderation effects (Figure 3B). A history of increasing IPC in childhood intensified the association between increases in IPC in adolescence and increases in sleep activity from adolescence into emerging adulthood (+1 SD, b = 1.53, p = .008). The simple slopes for constant (b = .50, p = .13) and decreasing (−1 SD, b = −.52, p = .10) IPC in childhood were not significant. A RoS on Z plot corroborated these simple slope findings (Figure 3B). The upper bound of the RoS on Z plot demonstrated that the positive association between IPC in adolescence and changes in sleep activity was significant as changes in childhood IPC increased from scores of .003 or higher (i.e., .001 SD above the mean or greater). An RoS on X plot was also conducted to further confirm the form of the interaction (Figure 4B). The positive association between changes in childhood IPC and changes in sleep activity from adolescence to emerging adulthood is significant when changes in IPC in adolescence are valued at .001 or higher (i.e., .0003 SD above the mean or greater).

Long-wake episodes

Consistent with patterns observed for sleep efficiency and activity, there were no main effects of change in childhood or adolescent IPC on change in LWE, and the interaction between adolescent IPC and childhood IPC was significant. Simple slopes analyses at ±1 SD revealed that having a history of increases in childhood IPC exacerbated the association between increases in IPC in adolescence and increases in LWE (+1 SD, b = .21, p = .005) when compared to those with a history of constant (b = .07, p = .10) or decreasing (−1 SD, b = −.06, p = .14) IPC in childhood. In other words, adolescents who experienced increases in IPC during childhood exhibited the greatest increases in LWE during the transition to emerging adulthood (Figure 3C). RoS on Z (Figure 3C) and RoS on X (Figure 4C) plots corroborated these simple slope findings. The RoS on Z plot showed that the positive association between changes in IPC in adolescence and changes in LWE was significant as changes in IPC scores in childhood were valued at .000 or higher (i.e., .000 SD above the mean or greater). The RoS on X plot illustrated that changes in childhood IPC were positively associated with changes in LWE when changes in IPC in adolescence were .000 or higher (i.e., .000 SD above the mean or greater).

Discussion

IPC is an established risk factor for sleep problems, but associations exhibit considerable heterogeneity. Childhood IPC histories may alter vulnerability to more proximal adolescent IPC, yet research has not assessed how these processes may operate as a source of variability that increases later sleep problems. Providing the longest longitudinal investigation and the first test of change-on-change processes in IPC and sleep, we examined whether changes in childhood IPC moderated the link between changes in adolescent IPC and four sleep parameters from adolescence to emerging adulthood. Utilizing a multi-method design, LCS modeling showed that increases in adolescent IPC predicted declines in sleep quality (efficiency, activity, and LWE) from adolescence to emerging adulthood for participants with a history of increasing childhood IPC but not for those without such a history. Specifically, increases in IPC across adolescence corresponded to increases in sleep activity and LWE and decreases in sleep efficiency from adolescence to emerging adulthood but only for those who experienced increases in IPC across childhood. These findings held even after controlling for acute exposure in emerging adulthood.

Findings are consistent with stress amplification models in which increases in childhood IPC amplified the negative effects of adolescent IPC, whereas these negative effects were blunted among those with decreasing or stable rates of childhood IPC (Morris et al., 2010; Rudolph & Flynn, 2007). Simple slope analyses and RoS tests provided further confirmation that childhood IPC operated as a vulnerability factor, amplifying the association between increases in adolescent IPC and worsening sleep problems across adolescence and emerging adulthood. Notably, the relatively small values for RoS tests may partly reflect grand-mean centering but also highlight that even minor elevations in IPC are sufficient to elicit significant increases in later sleep problems.

Toward understanding why increases in childhood IPC may amplify proximal effects of increases in adolescent IPC, emotional security theory underscores the role of insecure response strategies (e.g., involvement in IPC, fearful distress) as a key underlying mechanism (Davies & Martin, 2013; Kelly & El-Sheikh, 2013). Childhood IPC may foster early patterns of hypervigilance and insecurity, which then may serve as a guide for responding to subsequent threats and potentially sensitize children to future IPC (Davies et al., 2014). Over time, repeated activation of insecure responses consolidates maladaptive stress regulation across emotional, attentional, and physiological systems. As these systems become increasingly interconnected, the wear and tear on the stress response system may trigger a “domino effect” in which dysregulation in one system cascades to others and ultimately depletes resources needed for effective regulation (Evans, 2003; Juster & McEwen, 2015; Kelly & El-Sheikh, 2019; Repetti et al., 2011). Persisting alterations to the stress response system may be accompanied by disruptions to the arousal regulatory system (e.g., wake-regulating orexin and noradrenergic systems), creating long-lasting difficulties with downregulation that increase vulnerability to sleep problems (Altena et al., 2016; Palagini et al., 2015).

Our findings are also consistent with cumulative stress models in that repeated exposure to the same stressor over time is detrimental to long-term adjustment (Slopen et al., 2018). Cumulative stress models have been applied to both additive and multiplicative risk (Bagley et al., 2014; Gillis et al., 2023). Consistent with the multiplicative effects component, we found that youth exposed to increasing rates of IPC in both childhood and adolescence were most at risk for increasing sleep problems during the transition to emerging adulthood. Those with increases in IPC limited to either childhood or adolescence tended to have more stability in sleep problems across adolescence and emerging adulthood. However, we did not find unique effects for both childhood and adolescent IPC (i.e., additive effects) which may indicate that findings align more closely with stress amplification.

Specificity in sleep parameters

Increases in childhood IPC histories amplified relations between changes in adolescent IPC and the three indices of sleep quality but not sleep duration. These findings contribute to evidence suggesting possible specificity in associations between IPC and sleep (Kelly & El-Sheikh, 2011; Kelly et al., 2024). Consistent with these studies, we found stronger associations between IPC and sleep quality parameters, including indices of continuity (i.e., efficiency), restlessness (i.e., activity), and fragmentation (i.e., LWE). Although little research has tested why IPC may more strongly forecast sleep quality than duration, the broader literature has found more robust effects for quality (Chung et al., 2025; Heissel et al., 2018; Kelly et al., 2014). Below, we offer possible explanations that may help guide future research.

First, sleep duration reflects an overall assessment of sleep length and does not account for night waking (Sadeh et al., 1994). Thus, individuals may have sufficient sleep duration but of poor quality. For example, an individual may exhibit substantial movement indicative of transient awakenings or experience multiple nighttime wake episodes that do not ultimately affect sleep duration scoring (Carskadon & Rechtschaffen, 2005).

Second, poor quality sleep may reflect hyperarousal. Heightened physiological arousal and vigilance, frequently associated with IPC (El-Sheikh et al., 2009; Harold & Sellers, 2018), are linked with sleep quality indices but not sleep duration (El-Sheikh et al., 2013). These sleep problems may further compromise sleep architecture (Altevogt & Colten, 2006; Burman & Muzumdar, 2020). Because arousal thresholds are lower in early sleep stages, frequent movement and wakings may reduce time in slow wave sleep and rapid eye movement sleep (Altevogt & Colten, 2006; Carskadon & Rechtschaffen, 2005). Disruptions of these deeper sleep stages, which support stress adaptation through emotional processing and reactivity (Altena et al., 2016; Kahn et al., 2013), may further reinforce physiological arousal and sleep fragmentation, creating a potentially vicious cycle (Kahn et al., 2013).

Limitations and future directions

The study should be interpreted in the context of its methodology, strengths, and limitations. First, hypotheses were tested in a sociodemographically and racially diverse sample from small towns and may not generalize to other populations. Second, we hypothesized that stress amplification may be explained by emotional insecurity insofar as insecure responses to childhood IPC persist into adolescence, increasing vulnerability to worsening sleep problems. However, associations between emotional regulation and sleep may be bidirectional (Altena et al., 2016; Kahn et al., 2013), warranting future research on reciprocal processes. Third, although stress amplification models have preliminary support (Sloan et al., 2022), prior work on childhood IPC histories and developmental psychopathology aligns more with risk saturation models (Davies et al., 2021). These differences in findings may stem from stress timing, outcomes, or time scales. For instance, increases in childhood IPC may initially sensitize youth across all rates of change in adolescent IPC, consistent with risk saturation. Over time, repeated IPC may lead to adaptation, specifically requiring progressive increases to trigger a stress response. Although speculative, stress amplification may emerge gradually through a combination of risk saturation and habituation.

Fourth, it is important to consider the potential influence of emerging adults’ structural home environments and the possibility that ongoing, acute IPC exposure could disrupt their concurrent sleep environment. However, several findings weaken this possibility. At Wave 7, most (73%) emerging adults reported no longer living with their parents. Although those living independently may be less likely to be exposed to IPC, bivariate correlations showed that IPC in emerging adulthood was unrelated to residential status. This finding underscores that IPC exposure can happen regardless of whether emerging adults live with their parents. Additionally, we controlled for ongoing, acute IPC exposure during emerging adulthood and demonstrated that stress amplification processes operated above and beyond these effects. Taken together, these findings suggest that acute IPC exposure in emerging adulthood is unlikely to be the primary mechanism driving the observed effects, though this possibility cannot be entirely ruled out.

Lastly, LCS modeling allows for an examination of change-on-change processes, capturing average change and variability around average change (i.e., individual differences in intraindividual change). This approach provides a nuanced perspective, suggesting that the predictive effect of IPC is not purely a function of average levels but also the intensification of IPC over time and the extent to which these changes correspond with changes in sleep. Nevertheless, there are two key considerations. Although we specified correlations among childhood IPC change scores, adolescent IPC change scores, and emerging adulthood IPC—which were non-significant—LCS models do not explicitly capture interdependence of change across developmental periods despite potential continuity over time. Additionally, saved factor scores are treated as perfectly measured, error-free values, even though they inherently contain measurement error (Ng & Chan, 2020; Skrondal & Laake, 2001). Alternative statistical approaches (e.g., latent growth modeling), may provide additional perspectives.

Conclusions

The findings are consistent with stress amplification models and show that increases in childhood IPC exacerbate the proximal association between changes in IPC in adolescence and changes in sleep problems from adolescence to emerging adulthood. The findings underscore the long-term effects of intensifying IPC on biological regulation predicting worsening sleep quality during the transition to emerging adulthood—after many youth have left home. These lasting effects underscore the importance of early intervention programs aimed at disrupting the cascade linking IPC to sleep problems in emerging adulthood as well as the need to evaluate the long-term efficacy of such programs.

Data availability

The data necessary to reproduce the analyses presented here are not publicly accessible. The analytic code necessary to reproduce the analyses presented in this paper is not publicly accessible. The materials necessary to attempt to replicate the findings presented here are not publicly accessible. Data will be made available after completion of each longitudinal study in accordance with National Institutes of Health data sharing guidelines.

Author contributions

Morgan J. Thompson (Conceptualization, Formal analysis, Methodology, Writing—original draft, Writing—review & editing [lead], Data curation, Supervision, Validation, Visualization [equal]), Alexandra D. Ehrhardt (Data curation, Formal analysis, Validation, Visualization, Writing—review & editing [equal], Methodology, Writing—original draft [supporting]), Ryan J Kelly (Conceptualization, Methodology [supporting], Writing—review & editing [equal]), and Mona El-Sheikh (Conceptualization [supporting], Funding acquisition, Resources [lead], Supervision, Writing—review & editing [equal]).

Funding

This research was supported by Grant R01-HD046795 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development awarded to Mona El-Sheikh. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Statement of preregistration

The analyses presented here were not preregistered.