Broken Bonds, Lonely Hearts? A mediation analysis of childhood adversity, relationship quality, and loneliness in later life

Abstract

Objectives:

This study examined whether the link between adverse childhood experiences (ACEs, including child maltreatment, financial hardship, and household dysfunction) and later-life loneliness was mediated by quality of social relationships.

Methods:

Data came from the Health and Retirement Study (N = 11,700). Five path models were tested: four individual relationship types (spouse, children, relatives, and friends) and one combining all relationship types.

Results:

In the spouse model, child maltreatment and financial hardship increased loneliness via lower support and higher strain; household dysfunction showed no significant indirect effect. In the children and other relatives models, child maltreatment and financial hardship again showed significant indirect effects via lower support and higher strain, whereas household dysfunction had a small indirect effect via higher strain only. In the friendship model, only child maltreatment had a significant indirect effect on loneliness. In the combined model, child maltreatment was consistently associated with lower support and higher strain across relationship types, except for strain with friends.

Conclusions:

ACEs increased loneliness partly by reducing social support and increasing social strain, with patterns varying by ACE and relationship type.

Clinical implications:

Loneliness interventions should address skills for rebuilding trust and supportive social ties.

Introduction

Loneliness is a subjective internal state of feeling isolated and disconnected, a common result of a discrepancy between an individual’s desired and actual experience with social connection (Perlman & Peplau, 1981). Loneliness tends to increase in the mid-50s and then again, particularly after age 75 (Hawkley et al., 2022), with 30-40% of older adults reporting feeling lonely and isolated in national polls and surveys (Kullgren et al., 2023). A robust body of literature has demonstrated the detrimental effects of loneliness on health and longevity. Systematic reviews have found that loneliness increased the risk of heart disease, stroke, and other physical health conditions (Valtorta et al., 2016), as well as depression, anxiety, and other mental health problems (Park et al., 2020). A meta-analytical review found that strong social relationships increased survival by 50% and equated the health damages of loneliness to smoking 15 cigarettes a day (Holt-Lunstad et al., 2010).

Despite growing awareness and efforts to reduce loneliness, current interventions have produced inconsistent effects and modest improvements (Mann et al., 2017; O’Rourke et al., 2018). Most intervention programs focus on increasing social interactions, operating on the premise that loneliness is caused by deficits in social contact (O’Rourke et al., 2018). However, this approach overlooks the cognitive and emotional aspects of loneliness, particularly in chronic loneliness (Domènech-Abella & Domènech, 2024). Research has already established that social isolation and loneliness are distinctive constructs (Cornwell & Waite, 2009), and emerging evidence suggests that simply increasing the number of social contacts is not sufficient to reduce loneliness (Y. Lee & Ko, 2018). Broadening our understanding of the causes of loneliness beyond structural social contact is thus critical for developing more effective interventions.

One potential, yet underexplored risk factor for later-life loneliness is adverse childhood experiences (ACEs). ACEs is an umbrella term describing a broad set of early-life events and experiences that cause severe stress or overwhelm children’s capacity to cope (Anda et al., 2006; McLaughlin et al., 2014). The original ACEs focused on child maltreatment and household dysfunction (Felitti et al., 1998). The expanded ACE framework added household financial hardship (Afifi, 2020). A large body of research has linked ACEs to a wide range of poor physical and health outcomes, unhealthy behaviors, and social relationships in adulthood, including in later life (Godoy et al., 2021; Yu et al., 2024). Although most existing studies used a sum score (i.e., summing up different types of ACEs) to quantify ACEs, research has shown that a sum score approach can obscure the differential health consequences of ACE dimensions (Barboza, 2018; Lacey et al., 2020). Increasingly, scholars have recommended examining distinct ACE dimensions to more clearly inform targeted intervention strategies (McLennan et al., 2020).

Conceptual framework

The attachment theory (Bowlby, 1969) and life course perspective (Elder et al., 2003) provide a strong theoretical foundation for understanding how ACEs influence loneliness in later life. From an attachment perspective, early experiences with caregivers, peers, and romantic partners shape our internal working models, which guide how we expect, interpret, and respond in social situations. When these experiences are inconsistent or harmful, individuals are more likely to develop insecure attachment styles and view themselves as incompetent and others as untrustworthy, cold, and rejecting. This makes it harder for them to form and sustain supportive relationships, increasing one’s susceptibility to loneliness (DiTommaso et al., 2015). Empirical studies lend support to this connection and have linked ACEs to insecure attachment styles (Riggs & Kaminski, 2010; Unger & De Luca, 2014), poor marriage satisfaction (Wheeler et al., 2020), and higher rates of marriage dissolution (Whisman, 2006). The life course perspective complements attachment theory by emphasizing that human development is a lifelong process and that early life experiences can shape later-life outcomes. Furthermore, the linked lives concept highlights the interconnections of lives, particularly within families. Integrating the attachment theory with the life course perspective allows us to conceptualize ACEs as disruptors of attachment formation, emotional regulation, and social skills. These negative consequences can track across the life span, leading to transient and chronic loneliness.

Despite compelling theoretical foundations linking ACEs to later-life loneliness, empirical studies remain limited. To date, most studies have focused on adolescents or young adulthood, and only a few studies have directly examined this linkage in older adults (Ejlskov et al., 2020; Furuya & Wang, 2023; Kamiya et al., 2014; Lampraki et al., 2022; Liu & Liu, 2023). The most comprehensive study to date, conducted by Furuya and Wang (2023), found that ACEs (labeled as “home environment”) significantly increased the risk of loneliness in later life. The authors also found that only about one-fifth of this association was explained by adulthood socioeconomic and health conditions, leaving other potential pathways unexplored. In a UK-based study, Ejlskov et al. (2020) examined the cumulative impact of social relationship difficulties across childhood, mid-adulthood, and later adulthood on loneliness at age 68. However, they did not focus explicitly on ACEs and instead included only two ACE-related indicators (i.e., parental death and divorce before age 18) and assessed overall social relationship difficulties (Ejlskov et al., 2020). Importantly, none of the previous studies examined the mediating role of relationship quality, a potentially key missing link in understanding how ACEs influence loneliness over time.

Outside of the ACEs context, a separate and more extensive body of research has examined the impact of relationship quality on loneliness. Overall, these studies suggest that relationship quality has a positive and negative aspect (H. J. Lee & Szinovacz, 2016), and that negative aspects of relationship (social strains) increase loneliness, whereas positive aspects (social support) reduce loneliness (Kim & Jung, 2025; Norlin et al., 2025; Teater et al., 2021). Previous research also suggests that the impact of relationship quality on loneliness varies by relationship type. While some inconsistencies exist, studies have generally found that social support from a spouse and friends is more strongly linked to lower loneliness, whereas social support from children and other family members shows less consistent connection to loneliness across studies (Chen & Feeley, 2014; Norlin et al., 2025; H. O. Taylor, 2020).

While existing research suggests that ACEs increase loneliness in later life and that relationship quality influences loneliness, we are not aware of any study that has explicitly tested whether relationship quality mediates the connection between ACEs and loneliness in later life across relationship types. Testing this mediation pathway can advance our understanding of how early adversity shapes social connections in later life, either supporting or challenging the integration of attachment theory and the life course perspective in explaining loneliness. Practically, this research can inform more effective interventions by identifying relationship quality and specific relationship types as key targets for reducing loneliness.

Objectives and hypotheses

This study aims to answer the following research question: Is the relationship between ACEs and later-life loneliness mediated by quality of social relationships? We propose the following hypothesis: The relationship between ACEs and loneliness is mediated by the quality of relationships with spouse, children, family, and friends, respectively. Specifically, we hypothesize that ACEs increase loneliness by reducing social support (positive aspect of relationship quality) and increasing social strain (negative aspect) within each relationship type.

Methods

Data

The Health and Retirement Study (HRS) is a nationally representative study of people over age 50 in the United States, sponsored by the National Institute on Aging (NIA U01AG0097) and housed at the University of Michigan Institute for Social Research. HRS employs a steady-state design that refreshes the sample every six years. Eligible participants undergo biennial core interviews. HRS has also administered the Psychosocial and Lifestyle Leave-Behind Questionnaire (LBQ) regularly since 2006 (Smith et al., 2023). The LBQ, conducted with a rotating random 50% of the core panel, collects rich data on social relationships, loneliness, and other psychosocial outcomes. The LBQ is repeated every 4 years for the same 50% of the core panel.

In this study, the sample consisted of 11,700 adults aged 51 or older who completed the LBQ in either 2016 or 2018, excluding proxy respondents. We combined responses from two years of LBQs to construct a full panel, as is often done in prior research (Smith et al., 2023). While the mediators and outcome were measured in the 2016/2018 LBQ, information on ACEs was reported retrospectively and compiled from across HRS waves. The present study analyzed publicly available, de-identified data and was determined as “not regulated as human subjects research” by the authors’ IRB.

Measures

Exposure variables – ACEs.

We extracted HRS questions that aligned with Felitti et al. (1998)’s original ACEs classification and the expanded framework (Afifi, 2020). We classified ACEs into three domains: child maltreatment (including physical abuse and neglect, and emotional abuse and neglect), family dysfunction, and financial hardship. Each domain was represented by a binary indicator denoting whether the participant endorsed at least one item within that domain. Child maltreatment included five binary items: physical abuse by either parent, poor relationship with mother, poor relationship with father, little or no time and attention from mother when needed, and little or no effort from mother in providing supervision and ensuring a good upbringing before age 18. Although we initially attempted to distinguish between physical abuse and neglect versus emotional abuse and neglect, we found that their associations with both the mediator and the outcome were similar in magnitude. Therefore, we combined them into a single domain of child maltreatment.

Family dysfunction included two binary indicators: having a parent who drank or used drugs so often that it caused problems in the family and experiencing the death of either parent before age 18. Financial hardship consisted of three binary items: rating their family as “poor”, relocation due to financial difficulties, and receiving assistance from relatives due to financial difficulties before age 16.

ACEs were assessed retrospectively and only once, typically when participants first entered the HRS, thus warranting data extraction across HRS waves. Specifically, data on parental death before age 18 were extracted from HRS core interview files (1992-2020). Items on financial hardship were obtained from the Cross-Wave Childhood Health and Family Aggregated Data file. The remaining ACEs variables were obtained from the Cross-wave Lifecourse Experiences of Adversity and Trauma Aggregated Data (V1.0).

Outcome variables – Loneliness.

The HRS LBQ includes the 11-item revised UCLA loneliness scale, with each item rated on a 3-point scale (J. Lee & Cagle, 2017; Russell, 1996). Following the scoring approach described in Smith et al. (Smith et al., 2023), we reverse-coded negatively worded items and then calculated the average of 11 items. The final score was a continuous score ranging from 1 to 3, with higher scores indicating greater loneliness. The final score was set to missing if more than five items were missing. The scale had excellent internal consistency (Cronbach’s alpha =.88 in 2016) (Smith et al., 2023).

Mediators – Quality of social relationships.

The HRS LBQ collects information on both positive (social support) and negative (social strain) aspects of social ties across four types, including spouse, children, other relatives, and friends. Seven items were asked for each category, including three items for social support (e.g., “How much can you rely on them if you have a serious problem?”) and four items for social strain (e.g., “How much do they criticize you?”). Each item was rated on a four-point scale. These questions were asked only if the participant reported any social ties in each respective relationship category.

Following HRS guidelines, all items were reverse-coded, and an index of social support and an index of social strain were created for each relationship category by averaging the scores within each dimension (positive vs negative) (Smith et al., 2023). The final scores for each index ranged from 1 to 4, which we treated as an approximately continuous variable, with a higher social support score indicating greater perceived social support and a higher social strain score indicating greater strain. The final score was set to missing if there is more than one item with missing values for the positive dimension and more than two items with missing values for the negative dimension. All subscales have good internal consistency (Cronbach’s alpha ranged from .76 to .86 in 2016).

Covariates.

We adjusted for participant demographics (age, sex, race/ethnicity). Moreover, childhood conditions included (1) parental education, measured as the highest grade completed by either parent (if both values were available, the average was used; if only one was available, that value was used; if neither was available, the value was coded as 0); (2) a binary indicator of whether the respondent lived with a grandparent in the same household for one year or more before age 17; (3) self-rated health before age 16, reported on a five-point scale ranging from excellent (coded as 1) to poor (coded as 5); and (4) a binary indicator of whether the respondent reported having depression, emotional, or psychological problems before age 16. These childhood condition items were obtained from the Cross-Wave Childhood Health and Family Aggregated Data file.

We included social network size as a covariate, defined as a count variable representing the number of “yes” responses to four items: living with a spouse or partner, having any living children, having other family members, and having friends. These questions were asked in the LBQ, with responses from 2016 and 2018 combined to represent the full panel. The final network size score ranged from 0 to 4. Although social network size may also act as a mediator, we included it as a covariate to help isolate the unique contribution of relationship quality by accounting for the structural differences in social contacts.

We did not adjust for adulthood conditions, such as socioeconomic status, physical health, or other psychosocial factors, because these conditions occurred after ACEs and are plausibly influenced by them, and thus may lie on the causal pathway from ACEs to later-life loneliness (Furuya & Wang, 2023). Treating such adulthood conditions as covariates would block part of the pathway and introduce overadjustment bias (Elwert, 2013; Schisterman, Cole, & Platt, 2009), misleading estimates of the total effect of ACEs and distracting from our central focus on the relationship pathways.

Statistical analysis

The hypothesized pathways linking ACEs and loneliness via relationship quality were tested using manifest path analysis, treating the key variables as observed composite variables rather than specifying separate latent measurement models for each construct. Five models were estimated, including four individual relationship models examining social support and social strain in each relationship type (spouse, children, other relatives, and friends), and one combined model including all relationship types in a single model. Social network size was included as a covariate in the individual relationship models but excluded from the combined model, as it would create a perfect collinearity conceptually.

In addition to the main mediation pathways of interest (i.e., ACEs -> Relationship Quality -> Loneliness), each of the path models contains direct paths from covariates to loneliness and the mediators, as well as covariance between the mediators. To balance parsimony and model fit, insignificant paths from covariates to mediators were removed from the final models.

Because not all participants reported at least one social tie in each relationship category, the effective sample size differed across individual relationship models. For example, the spouse/partner model contained only individuals who reported having a spouse or partner. For the combined model, the effective sample included participants who reported having a spouse/partner and at least one living child, a relative, and a friend. As a result, the sample for the combined model represented older adults who were less socially isolated overall. Despite this limitation to generalizability, the combined model allows us to compare the impacts of relationships across different categories more directly.

All models were estimated using Mplus 8 (Muthen & Muthen, 2017) using Maximum Likelihood (ML) estimators with Bootstrap confidence intervals (n = 5000). All final models had acceptable model fit, indicated by a Root Mean Square Error of Approximation (RMSEA) < .06, Comparative Fit Index (CFI) > .95, and Standardized Root Mean Square Residual (SRMR) < .08 (Hooper et al., 2008).

HRS cross-sectional weights are constructed independently for each wave and post-stratified to represent the age-eligible U.S. population in that specific year (HRS Staff, 2019). As such, in their original form they are not intended to be directly combined in pooled analyses across waves without constructing new pooled weights. Therefore, in the main manuscript, we conducted unweighted descriptive statistics for the pooled 2016–2018 analytic sample (Table 1) and did not apply weights in the pooled path analysis. To facilitate comparison with other HRS-based studies and to provide population-level context, we additionally present wave-specific weighted descriptive statistics in Supplementary Tables S1.

Table 1.

Unweighted descriptive statistics for the pooled 2016–2018 study sample

	N	All	Any childhood adversity
			No	Yes	P-value
Sample characteristics
Age in years	11,605	67.1 (10.5)	66.9 (10.7)	67.2 (10.4)	.151
Sex (%)					<.001
Female	6,975	59.6	56.6	61.3
Male	4,725	40.4	43.4	38.7
Race/ethnicity (%)					<.001
White, non-Hispanic	7,547	64.6	70.6	61.4
African American, non-Hispanic	2,118	18.1	14.4	20.1
Other, non-Hispanic	492	4.2	4.0	4.3
Hispanic	1,522	13.0	11.0	14.2
Parental education in grades	11,700	9.5 (4.3)	10.3 (4.1)	9.1 (4.4)	<.001
Lived with a grandparent (%)	2,978	25.5	20.5	28.1	<.001
Childhood psychological problem (%)	594	5.1	2.1	6.7	<.001
Childhood self-rated health (%)					<.001
Excellent	6,219	53.5	60.0	50.1
Very good	2,890	24.9	24.3	25.2
Good	1,730	14.9	11.7	16.6
Fair/Poor	779	6.7	4.0	8.1
Social network size	11,618	3.1 (0.9)	3.2 (0.8)	3.1 (0.9)	<.001
UCLA loneliness score	11,516	1.5 (0.4)	1.5 (0.4)	1.6 (0.5)	<.001
Quality of social relationships
Spouse/partner - social support	7,598	3.5 (0.7)	3.6 (0.6)	3.4 (0.7)	<.001
Spouse/partner - social strain	7,581	1.9 (0.7)	1.9 (0.6)	2.0 (0.7)	<.001
Children - social support	9,862	3.2 (0.8)	3.3 (0.7)	3.2 (0.8)	<.001
Children - social strain	10,230	1.7 (0.6)	1.6 (0.6)	1.7 (0.7)	<.001
Other relatives - social support	10,840	2.9 (0.9)	3.0 (0.8)	2.8 (0.9)	<.001
Other relatives - social strain	10,833	1.6 (0.7)	1.5 (0.6)	1.7 (0.7)	<.001
Friends - social support	10,383	3.1 (0.7)	3.1 (0.7)	3.0 (0.8)	<.001
Friends - social strain	10,381	1.4 (0.5)	1.4 (0.5)	1.4 (0.5)	<.001

Note. Values are presented as mean (SD) or percent. Estimates are unweighted and based on the pooled 2016 and 2018 leave-behind respondents. “Any childhood adversity” refers to reports of any of the following childhood adversity items: 1) physical abuse by a parent, 2) poor relationship with mother, 3) poor relationship with father, 4) little or no time and attention from mother when needed, 5) little or no effort from mother in providing supervision and ensuring a good upbringing, 6) having a parent who drank or used drugs so often that it caused problems in the family, 7) death of either parent, 8) family as “poor”, 9) relocation due to financial difficulties, and 10) received assistance from relatives due to financial difficulties.

Results

Participant characteristics

Most of the study sample endorsed at least one ACE item (65.0%). The top three reported specific ACEs were having a poor relationship with father (44.0%), growing up in a “poor” family (43.5%), and having a poor relationship with mother (30.1%). Physical abuse by a parent was the least reported ACE (11.3%).

As shown in Table 1, those with ACEs were more likely to be women, racial and ethnic minorities, and to report lower parental education, more childhood emotional problems, poorer childhood self-rated health, smaller social networks, lower social support, and higher social strain across all relationship domains.

Relationship-specific mediation path models

Results from the four relationship-specific mediation models (Table 2) supported our hypothesis, showing significant indirect effects through which ACEs increased loneliness by reducing social support and increasing social strain. However, the specific patterns differed depending on the type of ACEs and, to a lesser degree, relationship category. Overall, child maltreatment showed the strongest and most pervasive indirect effects, followed by childhood financial hardship; household dysfunction had weaker and less consistent indirect effects.

Table 2.

Direct and indirect effects of individual relationship path model linking childhood adversity, loneliness, and relationship quality

Mediation path model	Direct effect	Indirect effect
		via social support	via social strain	Total indirect
Model 1: Relationship with spouse (N=7,470)
Child maltreatment to loneliness	.067 (.046, .088) *	.024 (.016, .032) *	.008 (.004, .012) *	.032 (.022, .042) *
Financial hardship to loneliness	.012 (−.009, .033)	.008 (.000, .015) *	.004 (.000, .008) *	.012 (.002, .021) *
Household dysfunction to loneliness	−.016 (−.037, .004)	.004 (−.003, .012)	.003 (−.001, .007)	.008 (−.002, .018)
Model 2: Relationship with children (N=10,079)
Child maltreatment to loneliness	.063 (.044, .082) *	.029 (.023, .035) *	.008 (.005, .011) *	.037 (.030, .044) *
Financial hardship to loneliness	.014 (−.005, .032)	.009 (.004, .015) *	.003 (.000, .005) *	.012 (.005, .019) *
Household dysfunction to loneliness	−.016 (−.035, .003)	.004 (−.002, .010)	.003 (.001, .006) *	.007 (.000, .015) *
Model 3: Relationship with non-child relatives (N=10,687)
Child maltreatment to loneliness	.056 (.038, .075) *	.031 (.026, .037) *	.007 (.005, .010) *	.039 (.032, .045) *
Financial hardship to loneliness	.019 (.001, .037) *	.006 (.001, .011) *	.004 (.001, .006) *	.010 (.003, .016) *
Household dysfunction to loneliness	−.018 (−.037, .000) *	.005 (.000, .010)	.003 (.000, .006) *	.008 (.001, .014) *
Model 4: Relationship with friends (N=10,241)
Child maltreatment to loneliness	.074 (.055, .092) *	.015 (.008, .022) *	.005 (.002, .008) *	.020 (.012, .028) *
Financial hardship to loneliness	.012 (−.006, .030)	.005 (−.002, .012)	.003 (.000, .006)	.007 (−.001, .015)
Household dysfunction to loneliness	−.005 (−.023, .013)	−.004 (−.011, .003)	.001 (−.002, .004)	−.003 (−.011, .005)

Notes: Standardized estimates presented, with 95% bootstrap confidence intervals in parentheses.

^*p < .05

In the spouse (Figure 1) and children (Figure 2) models, child maltreatment was associated with both lower support and higher strain, and these paths jointly contributed to higher loneliness. Financial hardship showed smaller but still significant indirect effects through reduced support and increased strain in both models. Household dysfunction showed only modest indirect effects, mainly through strain with children. In both models, child maltreatment also retained a significant direct association with loneliness, consistent with partial (rather than full) mediation.

Figure 1.

Path model linking adverse childhood experiences and loneliness via relationship quality with spouse.

Path analysis showing the association between adverse childhood experiences and loneliness in later life, mediated by relationship quality with a spouse or partner, including both social support and social strain. The analytical sample was older adults who reported having a spouse or partner (N=7,443). Standardized coefficients are shown, with 95% bootstrap confidence intervals in brackets. The model is adjusted for direct paths from covariates to loneliness and for significant paths from covariates to the mediators; however, these covariate paths are not displayed in the figure for clarity. Solid lines indicate statistically significant paths (p < .05), while dashed lines indicate non-significant paths. Model fit indices: χ²(3) = 4.51 (p=.211), CFI = 1.00, RMSEA = .01, SRMR = .00. The model explained 25.3% of the variance in loneliness.

Figure 2.

Path model linking adverse childhood experiences and loneliness via relationship quality with children.

Path analysis showing the association between adverse childhood experiences and loneliness in later life, mediated by the quality of relationships with children. The analytical sample was older adults who reported having at least one living child (N=10,052). Standardized coefficients are shown, with 95% bootstrap confidence intervals in brackets. The model is adjusted for direct paths from covariates to loneliness and for significant paths from covariates to the mediators; however, these covariate paths are not displayed in the figure for clarity. Solid lines indicate statistically significant paths (p < .05), while dashed lines indicate non-significant paths. Model fit indices: χ²(7) = 20.43 (p=.005), CFI = .99, RMSEA = .01, SRMR = .01. The model explained 22.3% of the variance in loneliness.

In the relatives (Supplementary Figure S1) and friends models (Supplementary Figure S2), child maltreatment again showed consistent indirect effects through both lower support and higher strain, whereas financial hardship and household dysfunction had smaller and more domain-specific indirect effects. For example, financial hardship was indirectly related to loneliness mainly through reduced support from children and relatives, while household dysfunction had a small indirect effect through increased strain with relatives. Overall, the four models explained 20–25% of the variance in loneliness.

Combined model

The combined model including all relationship domains simultaneously (Table 3) showed that child maltreatment remained robustly associated with higher loneliness through lower support and higher strain across nearly all relationship types, with a significant total indirect effect. Financial hardship had a smaller total indirect effect, driven primarily by reduced support from children, whereas household dysfunction showed only a modest indirect effect through reduced support from non-child relatives. The combined model explained 31.1% of the variance in loneliness.

Table 3.

Indirect effects of combined path model linking childhood adversity, and loneliness via relationship quality

	Standardized Estimates (95% bootstrap confidence interval)
Child maltreatment to loneliness
Total indirect	.063 (.048, .079) *
via social support from spouse	.019 (.012, .027) *
via social support from children	.010 (.006, .014) *
via social support from non-child relatives	.008 (.005, .012) *
via social support from friends	.013 (.006, .021) *
via social strain from spouse	.005 (.002, .008) *
via social strain from children	.003 (.000, .005) *
via social strain from non-child relatives	.003 (.001, .006) *
via social strain from friends	.002 (.000, .003)
Financial hardship to loneliness
Total indirect	.021 (.006, .037) *
via social support from spouse	.005 (−.002, .012)
via social support from children	.004 (.001, .006) *
via social support from non-child relatives	.001 (−.001, .003)
via social support from friends	.006 (−.002, .013)
via social strain from spouse	.002 (−.001, .005)
via social strain from children	.001 (−.001, .002)
via social strain from non-child relatives	.002 (.000, .003)
via social strain from friends	.001 (−.001, .003)
Household dysfunction to loneliness
Total indirect	.006 (−.010, .021)
via social support from spouse	.005 (−.002, .012)
via social support from children	−.001 (−.003, .002)
via social support from non-child relatives	.002 (.000, .005) *
via social support from friends	−.005 (−.012, .002)
via social strain from spouse	.001 (−.001, .004)
via social strain from children	.000 (−.001, .002)
via social strain from non-child relatives	.002 (.000, .004)
via social strain from friends	.000 (−.002, .002)

Notes: Path analysis of the combined model, showing association between adverse childhood experiences and loneliness in later life, mediated by the quality of relationships with spouse, children, family, and friends. The analytical sample was older adults who reported having a spouse/partner, living children, non-child relatives, and friends (N=5,832). Model fit indices: χ²(24) = 394.01 (p <.001), CFI = .97, RMSEA = .05, SRMR = .04.

^*p < .05

Discussion

Using data from a nationally representative study of older adults in the U.S., we found that the quality of social relationships partially mediated the link between ACEs and loneliness, as some direct associations between ACEs and loneliness remained statistically significant. However, the patterns varied depending on the ACE domain and relationship type.

Across the relationship-specific and combined models, the proportion of variance in loneliness explained ranged from about 20% to 31%, which are in line with prior psychosocial models of loneliness (Hill & MacGillivray, 2023) and indicate that ACEs and relationship quality account for a meaningful share of the variability in later-life loneliness. Our findings provide empirical support for the integrated framework of attachment theory and life course perspective in understanding the long-term consequences of ACEs on loneliness in later life.

Child abuse and neglect, among the most disruptive ACEs for early attachment formation, consistently showed the strongest association with lower social support and higher social strain across relationship types. Prior research has shown that children who experience abuse and neglect are more likely to develop insecure attachment styles, which can lead to relationship difficulties in young adulthood and midlife (Baer & Martinez, 2006; Riggs & Kaminski, 2010; Taussig & Culhane, 2010; Unger & De Luca, 2014). Our study extends this literature by demonstrating that the detrimental impact of child maltreatment continues into later life, shaping interpersonal dynamics well beyond early and midlife stages. Unlike household dysfunction and childhood financial hardship, the impact of which depends on the specific relationship type, abuse and neglect appear to disrupt fundamental relational capacities, such as trust, emotional regulation, and expectations in relationships. As a result, they do not just affect specific relationship types but instead appear to shape a generalized pattern of social difficulties that permeate all kinds of social bonds. These findings echo studies that have demonstrated the differential impacts of different forms of ACEs (Wang et al., 2024), highlighting the distinct and severe impact of child abuse and neglect (Herrenkohl & Herrenkohl, 2007; Wang et al., 2023).

Although its impact was not as strong as that of child maltreatment, childhood financial hardship was also significantly associated with loneliness in later life, when relationships were analyzed in separate models, through decreased social support and increased social strain across nearly all types of relationships. Revealing the negative impact of childhood financial hardship is important given that most ACEs research continues to overlook financial hardship as a form of early adversity (Wang et al., 2025). Our study well aligns with prior literature suggesting childhood financial hardship as a chronic stressor that undermines the development of key psychological resources, such as self-regulation and coping capacity (Evans & Kim, 2013; McEwen & McEwen, 2017; R. L. Taylor & Barch, 2022) that are essential for building healthy interpersonal relationships (Conger et al., 2010).

Among the three ACE categories, household dysfunction had the weakest and most context-specific association with loneliness and relationship quality. Experiencing parental death or parental substance use during childhood may not fundamentally alter the internal schemas that guide emotional and cognitive interactions with others to the same degree as abuse and neglect (Amos et al., 2011; Moehler et al., 2007). While family dysfunction can negatively impact various aspects of human development, it may not always act as a trauma trigger that fundamentally reshapes interpersonal relationship patterns. This again underscores the need for future studies to analyze different forms of ACEs separately rather than relying on a cumulative score that treats them as equivalent. Interestingly, the total indirect effect of household dysfunction on loneliness was significant only in the model focused on relationships with other relatives, but not in models examining the other types of relationships. This finding aligns with the linked lives concept, suggesting that shared exposure to a dysfunctional household environment may create long-lasting relational difficulties within the family unit.

Consistent with previous research (Chen & Feeley, 2014), results from the combined model indicated that relationship quality with spouses and friends showed a stronger association with loneliness than relationships with children or other family members. According to the socioemotional selectivity theory (Carstensen et al., 1999; Charles & Carstensen, 2010), as people perceive their remaining time as more limited, they increasingly prioritize emotionally meaningful goals and relationships. In later life, this may lead older adults to invest more heavily in close ties such as a spouse, who is a primary source of physical and psychological intimacy. It is possible that as individuals hold strong expectations for attachment-related support from their intimate partners, and when these expectations are unmet, the resulting disappointment may intensify feelings of loneliness. Likewise, people may intentionally invest in friendships to derive emotional meaning from life as they age (Carstensen et al., 2003). Particularly in later life, friendships are often associated with more positive experiences than family relationships (Huxhold et al., 2014) as family relationships are frequently experienced as obligatory (Roberts & Dunbar, 2011). Further, friendships can be especially beneficial for older adults, as friends may offer crucial support when family ties weaken or end, especially after the loss of a spouse (Gupta & Korte, 1994). At the same time, social strain in friendships can have a significant negative impact on the health of older adults (Chopik, 2017). These findings reinforce prior evidence that the impact of relationship quality on loneliness varies by relationship type (Chen & Feeley, 2014; Norlin et al., 2025; H. O. Taylor, 2020), and underscore the importance of avoiding overgeneralization across relationship domains.

Limitations

This study has several limitations. Relationship quality and loneliness were measured at the same wave of LBQ, preventing us from establishing a clear time order between the mediators and the outcome. Although the temporal order between ACEs and the outcome is clear, and we adjusted for several childhood conditions, unmeasured confounding may still be present due to other early-life factors not accounted for in our models (Furuya et al., 2024). Moreover, ACE variables were aggregated from multiple waves, and occasional data reconciliation or participant reporting errors may have occurred. In addition, sample size varied across models, as different subsamples were analyzed depending on the availability of specific relationship data. Because HRS cross-sectional weights are wave-specific and not designed for pooled analyses, we estimated unweighted models; this and the varying sample sizes across relationship models should be considered when assessing the generalizability of our findings. While our model conceptualized relationship quality as a mediator, it can act as a potential moderator. Lastly, we used manifest path models with observed composite scores rather than latent-variable structural equation modeling; As such, we did not explicitly model measurement error, which may modestly attenuate some effect sizes and R² values.

Implications

Our findings highlight the critical importance of trauma-informed care for older adults. Trauma-informed care involves programs, organizations, or systems that understand the widespread impact of trauma, recognize its signs and symptoms, respond by integrating trauma knowledge into policies and practices, and actively work to prevent re-traumatization (SAMHSA, 2014). Given the long-term impact of childhood trauma on loneliness in later life, practitioners working with older adults should consider prioritizing efforts to address enduring relational difficulties that are more common among individuals with histories of ACEs, such as insecure attachment, emotion regulation problems, and heightened mistrust in close relationships (Snyder et al., 2024; Kascakova et al., 2025; Porat-Butman et al., 2025; Murphy et al., 2015) rather than focusing solely on increasing the number of social contacts. In addition, interventions for older adults affected by early-life household dysfunction may benefit from incorporating family-based approaches, such as sibling therapy or structured support for improving strained family relationships.

Second, relationship type matters. Interventions aimed at reducing later-life loneliness may be more effective if they prioritize strengthening spousal and friendship relationships, as these appear to have the greatest influence on loneliness in later life. Trauma-focused couple therapy that focuses on later-life romantic relationship development, as well as peer support groups that foster friendships in later life, are emerging areas for advancement in gerontology. In addition to prioritizing certain relationship types, it may be more effective to focus on enhancing the positive aspects of relationships rather than reducing social strain. This is based on the findings that, although not directly tested, the protective effects of social support appeared to be stronger than the harmful effects of social strain on loneliness, as indicated by the relatively larger standardized coefficient.

Our findings suggest that relational difficulties may be a key mechanism linking ACEs to loneliness in older adulthood, extending previous research that has primarily focused on socioeconomic and health pathways (Furuya & Wang, 2023). These results highlight the value of an integrated framework combining attachment theory and the life course perspective to understand better the long-term impact of early adversity on social and emotional well-being in later life. Future research should continue to examine relationship-specific pathways, the interplay between social support and strain, and potential protective factors that may help break the cycle between early adversity and loneliness in later life. Interventions aimed at reducing loneliness should consider strategies that not only address emotional trauma but also equip individuals with the skills to rebuild trust and form meaningful, supportive relationships.

Clinical Implications.

• Practitioners working with lonely older adults should consider addressing maladaptive relational patterns rather than focusing solely on increasing social contacts.

• Loneliness interventions may improve effectiveness by strengthening spousal and friendship relationships, as these types of relationships appear to have the most significant influence on loneliness in later life.

Data Availability Statement

The study used de-identified, publicly available data from the Health and Retirement Study. Interested readers can request data access at https://hrs.isr.umich.edu/about. The study reported in the manuscript was not pre-registered.