Parenting Stress and Attachment Insecurity in Young Adulthood: A Social Relations Model

Abstract

Objective:

We used the Social Relations Model to inspect the individual- and dyad-specific components of attachment among young adults and their parents, and examined relations between these components and parenting stress.

Background:

Young adulthood is a transitional period in which the whole family is concerned with “launching” the young adult and exploring new ways to interact with and attach to one another. However, research on young adulthood attachment has primarily focused on young adults’ attachment style rather than reciprocal attachments among family members.

Method:

When the young adults were age 23, mothers, fathers, and young adults from 156 families reported their mutual attachment security. At ages 18 and 23, parents of the adolescent/young adult reported their parenting stress in interparental and parent-child relationship domains.

Results:

Attachment in the families of young adults can be separated into three components: 1) actor effects (each family member’s internal working model of attachment), 2) partner effects (characteristics of each family member as an attachment figure), and 3) relationship effects (dyad-specific attachment between family members). Increase of parenting stress in a family subsystem (dyad of family members) predicted attachment insecurity within the subsystem. Additionally, compensatory effects across family subsystems were observed.

Conclusion:

Attachment in the family during young adulthood is explained by family members’ own characteristics as well as dyad-specific interactions and is predicted by parenting stress in family subsystems.

Young adulthood is a period of dramatic change, during which young adults face many life transitions, such as leaving home, entering college, developing a career, and forming long-term romantic relationships (Arnett, 2014; Cohen et al., 2003). During this period, attachments with parents are also typically renegotiated as young adults seek to separate and form deeper attachments with others outside of the family (Buist et al., 2004). Compared with adolescents, young adults are less emotionally close to parents and more attached to romantic partners (Markiewicz et al., 2006; Rosenthal & Kobak, 2010; Whiteman et al., 2011). Although attachments to parents are renegotiated and deemphasized from adolescence to young adulthood, the degree to which attachments remain secure is strongly linked to young adults’ future well-being in terms of both individual adjustment (Mattanah et al., 2004) and healthy romantic relationships (Steinberg et al., 2006). Thus, the extent to which young adults and their parents successfully navigate this transitional period, such that attachments are renegotiated yet remain intact, is crucial for a healthy and fulfilling adulthood. The relationships between young adults and their families of origin are particularly important in the contemporary context of extended transition to adulthood (e.g., later home-leaving and marriage; Furstenberg, 2010).

A notable limitation of research on young adult attachment is that it typically only relies on young adults’ self-reports of their general attachment style. According to attachment theory (Bowlby, 1969/1982; Bretherton & Munholland, 2008; Cook, 2000), however, attachment is a function of both individual characteristics (e.g., attachment styles, internal working models, characteristics as attachment figures) and dyadic processes (i.e., unique interactions between parent and child). In other words, the renegotiation of attachments between young adults and a given parent does not occur independent of that parent. Thus, existing research, which primarily focuses on the attachment style of the young adult, misses important and illuminating complexities regarding interdependencies in attachments among young adults and their parents. This dyadic and dynamic view on young adulthood attachment has been reiterated by family theorists. Specifically, family life cycle theory (Carter & McGoldrick, 2005) suggests that the nuclear family often transits a series of changes related to their children’s development. One major change takes place when children grow up and leave the family – typically in late adolescence to young adulthood – during which the family needs to adapt to new dynamics, roles, and relationships. Adaptation to young adulthood is considered a family-level process that impacts not just the young adult’s, but all family members’ mutual attachments (e.g., developing more independent and adult-to-adult parent-child relationships, renegotiating the marital relationship; Bornstein et al., 2012; Carter & McGoldrick, 2005). To date, however, such family system perspectives have not been well integrated into young adulthood attachment research.

Some methodological issues may have hindered the application of family perspectives to young adult attachment research. First, studying young adult attachment in a family context requires data from multiple family members. Such study designs are difficult to implement. Second, family members may agree or disagree on the quality of their relationships, which makes the data analysis statistically challenging. To address these issues, methods have been developed for analyzing relationship data reported by multiple people. For example, the Social Relations Model (SRM; Kenny & La Voie, 1994, described in more detail in the next sections) partitions the variance in multiple family members’ reports into several individual-based and relationship-specific components, which provides insights on the organization of relationship dynamics in the family system and each subsystem (e.g., marital relationship, parent-child relationship). Perhaps due to the scarcity of data involving multiple family members, however, only a few attachment studies have implemented the SRM (Buist et al., 2004; Buist et al., 2008; Cook, 2000; De Meulenaere et al., 2021), none of which focused on young adulthood. Thus, our understanding of family attachment during young adulthood is limited.

The family system perspectives may also help answer a long-standing question in attachment research – that is, how attachment is “updated” through life experiences. Attachment is often conceptualized as a product of early parent-child interactions: Offspring who receive unresponsive, insensitive treatments from their parents in infancy and early childhood develop insecure attachment styles which persist to adulthood (Fraley et al., 2013; Haydon et al., 2012). However, the stability of attachment from infancy/early childhood to late adolescence/young adulthood is often weak to moderate and varies across studies (Groh et al., 2014; Hamilton, 2000; Waters et al., 2000; Weinfield et al., 2000), indicating that later personal and social factors also contribute to attachment security.

Specifically, the literature has consistently suggested that stress is a key factor that contributes to changes in attachment relationships (Hamilton, 2000; Waters et al., 2000; Weinfield et al., 2000). Indeed, heightened stress often results in more family conflict and less warm, responsive parenting (Putnick et al., 2008), which is detrimental to the development of secure attachment (Atkinson et al., 2000; Hamilton, 2000; Kor et al., 2012; Waters et al., 2000). Attachment researchers have primarily focused on the impact of negative life events (e.g., illness, loss of parent, divorce) on attachment (Hamilton, 2000; Waters et al., 2000), but stress can also be introduced by individual and family life transitions. For instance, family life cycle theory views the major developmental changes or transition periods in the family (e.g., the “launching” of the adolescent/young adult) as stressful to the family (Carter & McGoldrick, 2005). Accordingly, parents may experience increased stress related to their roles as parents (i.e., parenting stress) when their children reach late adolescence/young adulthood, especially when family members have difficulty adjusting to the new roles and dynamics (Cui et al., 2019; Dankoski, 2001). The increasing parenting stress may reflect maladjustment to the life changes at individual, dyadic, and familial levels, and lead to negative interactions among family members as well as deteriorating attachment security.

The aims of the current study are twofold. One, we use the Social Relations Model (SRM; Kenny & La Voie, 1994) to examine the organization of attachment in the families of young adults and to identify the components of attachment specific to each family member as well as dynamics between family members. Two, we elucidate the life-course-related contributors to adulthood attachment by associating parenting stress during late adolescence to young adulthood with the organization of attachment in young adulthood. Below, we elaborate on the theoretical background of the structure of attachment in the family, the role of parenting stress in the development of attachment, and the Social Relations Model.

Attachment in Young Adulthood: A Family Perspective

Attachment theorists have frequently used the term “internal working model” to refer to a person’s mental representations of close relationships (Bowlby, 1969/1982; Bretherton & Munholland, 2008). Adults often hold internal working models that are generalized across various relationship contexts, but they also respond uniquely to each relationship based on its specific dynamics (Bretherton & Munholland, 2008). As such, attachment in the family has been theorized to comprise a few components based on the characteristics of each family member as well as interdependencies between family members (Cook, 2000): First, each person’s attachments to different family members tend to be similar to some extent because they share the same basis of the person’s generalized internal working model (Bretherton & Monholland, 2008; Cook, 2000). Second, attachment is fostered through interactions with attachment figures, and adults tend to “update” their internal working model based on the partner’s availability to provide love and support (Booth-LaForce & Roisman, 2021), which means family members’ attachment to the same person may be similar to some extent based on the person’s characteristics as an attachment figure. Third, attachment is also subject to unique dynamics within a dyad (Bradley & Cafferty, 2001), which means that there should be dyad-specific components in addition to the individual-based components. Finally, attachment in the family is based on the same family context and may be generalizable to some degree across all family members. Although these components are generally theorized to play roles in family attachment, the salience of each component may vary across the family’s life course (Buist et al., 2008).

Social Relations Model (SRM) is a statistical method designed to disentangle the aforementioned components of relationships among multiple family members. Specifically, by incorporating mutual ratings (i.e., round-robin design) of family members, SRM statistically separates the variances of attachment among observed relationships into four components: 1) the actor effect, which is the general level of how a given family member rates his/her attachment relationship with the other family members based on the given family member’s internal working model; 2) the partner effect, which is the general level of how the other family members rate their unique attachment to a given family member, and reflects the given family member’s characteristics as an attachment figure; 3) the relationship effect, which is how a given family member rates attachment to another specific family member that is not explained by the more generalized actor and partner effects (i.e., unique attachment in the dyadic relationship); and 4) the family effect, which is the average level of attachment in the whole family system.

By partitioning attachments in the family into multiple components, SRM helps elucidate the structure of attachment during adolescence and young adulthood, a question of theoretical importance but not well answered. Compared to childhood, adolescence and adulthood is marked by the maturation of cognitive abilities and changes in parent-child dynamics, which can evolve attachment relationships in multiple ways. First, adolescents and young adults are capable of more abstract, generalized thinking about relationships, which may lead to a more coherent and integrated attachment style across different relationships (Fivush, 2006; Thompson & Raikes, 2005). Second, adolescents and young adults develop more complex and nuanced cognitive schemas, which enables more specific reactions to unique dynamics in each relationship (Fivush, 2006; Fraley & Roisman, 2019). Third, the “launch” of young adults from parents fosters more specific, “adult-to-adult” parent-child relationships, and provides parents a new opportunity to “revisit” family relationship dynamics (Bouchard, 2013). These developments suggest that, during adolescence to young adulthood, SRM actor effects (i.e., generalized working models across attachment relationships) and relationship effects (i.e., nuanced, dyad-specific attachment relationships) may serve as particularly important components of attachment in the family.

The SRM studies of family attachment, however, have produced mixed findings. Consistent with theory, studies using SRM generally support that attachment in the family can be allocated to actor, partner, relationship, and family effects in adolescence (Buist et al., 2004; Buist et al., 2008; Cook, 2000; De Meulenaere et al., 2021). SRM actor effects have usually been found to be salient. By contrast, partner, relationship, and family effects have shown varying significance across studies and occasionally explain little variance in reported attachments (Cook, 2000; Buist et al., 2008; De Meulenaere et al., 2021). Buist and colleagues (2008) found that adolescents’ attachment to family members became more relationship-specific over time, suggesting adolescence may be a period for gaining dyad-specific flexibility rather than generalization across relationships. De Meulenaere et al. (2021) also reported non-significant partner effects in adolescence, suggesting family members’ generalization about attachment figures may be a non-salient process. By contrast, other studies suggest that generalization across dyadic attachments (i.e., actor and partner effects) stably exists in the family (e.g., Cook, 2000). Thus, more research is needed to examine the relative salience of components of attachment.

Although these studies provide knowledge about the organization of family attachment, significant questions remain unanswered. First, extant SRM attachment studies have focused on adolescence. More work is needed to understand attachments of young adults, who, compared with adolescents, are more advanced in terms of cognitive flexibility and integration and farther along in the process of separating from their parents.

Second, the SRM attachment studies have uniformly used a four-person design. However, recruiting four participants per family poses challenges. Additionally, because SRM requires the roles of the four participating family members to be consistent across the sample (e.g., all previous studies required that every family must have a mother, a father, and two participating children), the four-person design places more constraints on the family structure.

A three-person design reduces recruitment challenges, is more flexible and inclusive regarding family structures (e.g., a single parent and two children; a parent, a grandparent/ extended family member, and a child), and thus has broader applications than the four-person design. However, the three-person design brings an extra statistical challenge – that is, unlike the four-person design, simultaneously estimating all the actor, partner, relationship, and family effects in the three-person SRM is mathematically impossible (the model is underidentified due to too many parameters being estimated; Kenny et al., 2006). Statisticians have recommended either omitting the family effect and focusing on individual- and dyad-specific processes (Kenny et al., 2006) or using a series of procedures to empirically retain the most salient SRM effects (Eichelsheim et al., 2009). However, no study has empirically examined attachment in the family using a three-person SRM, and the feasibility of such models needs to be determined.

In this study, we used a convenience sample to conduct a three-person SRM for attachment in the family. The sample consisted of ethnically homogenous European American families and was not representative of the U.S. population (see the Method section for the history of the sample and the recruitment rationale). Although the specific findings (e.g., the significance of specific attachment components) may not generalize to more diverse populations, the findings do generalize to a known population (Jager et al., 2017), and the data provided an opportunity to evaluate the feasibility of the three-person design for attachment research.

Parenting Stress during Adolescence to Adulthood and Family Attachment

As discussed earlier in this paper, parenting stress is a risk factor for insecure attachment due to its negative effects on interactions and communications among family members (Atkinson et al., 2000; Kor et al., 2012; Putnick et al., 2008). Parenting stress may be especially relevant to attachment in the family during late adolescence to young adulthood because it likely coincides with a family’s navigating the transition (Dankoski, 2001). Although most studies of the link between parenting stress and insecure attachment focus on infancy to childhood, as a major indicator of family maladjustment, parenting stress likely also plays a role in attachment in late adolescence to adulthood. Indeed, parents’ adjusting to their adolescent’s transition to young adulthood generally predicts attachment insecurity in the family system. Hannum and Dvorak (2004) found that family conflict during the transition into college was negatively associated with young adults’ attachment to parents. Further, parents’ overinvolvement in and overcontrol of their adult child’s life – a sign of parents being overly stressed – are related to lower family life satisfaction and greater attachment anxiety in parents (Brenning et al., 2017; Segrin et al., 2012). However, the dyadic nature of attachment was not fully disaggregated in those studies. In addition, because most studies are cross-sectional, longitudinal studies are needed to understand stability and change in familial adjustment to this transition (Bornstein et al., 2017).

Moreover, studies of the association between parenting stress and attachment often only consider one parent (typically the mother; see Atkinson et al., 2000). However, the few studies that involve fathers have shown that both mothers’ and fathers’ parenting stress contribute to the child’s attachment insecurity (e.g., Jarvis & Creasey, 1991), supporting the necessity of taking a more integrative family perspective. Further, parenting stress has been considered a multifaceted construct that does not only depend on parent-child dynamics but also the whole family system. Specifically, researchers have distinguished parenting stress due to the marital dyad (i.e., co-parenting stress) and parent-child dyads (i.e., stress in parent-child interactions; Sheras et al., 1998). According to family systems theory, parenting stress specific to each family subsystem (e.g., parent-child dyad, the marital dyad) may primarily impact the corresponding dyadic relationship, but the effect may also extend to other dyadic relationships in the family (Nelson et al., 2009). Studies of the effects of parenting stress on parent-child interactions and family relationships generally support three types of effects: 1) subsystem-specific effects, when stress in one dyad is linked with maladjustment in the specific dyad (e.g., Nelson et al., 2009; Ponnet et al., 2013; Zeng et al., 2021); 2) spillover effects, when stress in one dyad “transfers” into poor adjustment in another dyad (e.g., stress in the marital dyad worsens parent-child relationships; Ponnet et al., 2013), and 3) compensatory effects, when stress in one dyad (e.g., the mother’s stress due to their relationship with the young adult) elicits positive dynamics in the other dyads (e.g., better support from the marital relationship) as a form of compensation (e.g., Nelson et al., 2009). Although these three types of effects of parenting stress have been discussed in the family literature, it is unknown whether or how these effects apply to the context of attachment in young adulthood. Examining the effects of dyad-specific parenting stress is necessary to advance understanding of how attachment in the family evolves in the transition to young adulthood.

The Present Study

The first aim of the present study was to describe attachment relationships in the family during early adulthood at individual, dyadic, and family levels by attributing family members’ reported attachment insecurity to the four SRM components using a three-person design. As attachment theory suggests, attachment takes place at multiple levels of the family system (Cook, 2000). Consistent with the SRM attachment literature, we expected that attachment insecurity among family members can be attributed to characteristics of the reporter (i.e., actor effects), the attachment figure (i.e., partner effects), each dyad (i.e., relationship effects), and the general atmosphere in the family (i.e., family effect). Because those SRM components cannot be estimated simultaneously in a three-person design (Kenny et al., 2006), and the salience of the SRM components (i.e., their capacities of significantly explaining variances in the reported attachment relationships) is inconsistent in previous studies (Buist et al., 2004; Buist et al., 2008; Cook, 2000; De Meulenaere et al., 2021), we used procedures recommended in the SRM literature (Eichelsheim et al., 2009) to empirically explore which SRM attachment effects are the most salient and should be retained in the model. We also compared the relative importance of the SRM components in each dyadic attachment relationship.

The second aim of the study was to examine how parenting stress from late adolescence to young adulthood predicts the SRM components of attachment insecurity. Specifically, we separately examined the effects of parenting stress related to the marital dyad (i.e., co-parenting stress) and the parent-child dyads (i.e., specific stress in parent-child interactions) with both mothers and fathers as reporters. Crucially, examining relations between parenting stress in different family subsystems and the SRM components of attachment insecurity will clarify how parenting stress informs attachment insecurity within and across family subsystems (e.g., whether co-parenting stress primarily contributes to attachment insecurity between mother and father, or spills over to attachment insecurity between parents and the child as well).

Because parenting stress has been theorized to indicate parenting difficulties and serve as a mechanism that “updates” attachment (Atkinson et al., 2000; Kor et al., 2012), we expected parenting stress to positively predict attachment insecurities in the SRM. Based on family life cycle theory, instability in parenting stress (increase in the rank order from late adolescence to young adulthood) is an indication of maladjustment during transition. Thus, we expected instabilities of parenting stress across adolescence and young adulthood to have a larger effect on family attachment than the initial level of parenting stress at late adolescence.

Specifically, the research on parenting stress (Nelson et al., 2009; Ponnet et al., 2013; Zeng et al., 2021) has suggested three types of effects of parenting stress on the family system – subsystem-specific, spillover, and compensatory. Because no study has examined relations between parenting stress and attachment in the family system during young adulthood, our analyses are exploratory. However, we anticipated that all three types of effects would emerge in this sample. Specifically, subsystem-specific effects (i.e., stress in each dyad is positively associated primarily with relationship-specific attachment insecurity within the dyad and the actor and partner effects that involve dyad members) were expected because they are the most consistent effects in the literature. In addition, parenting stress in one dyad may spillover to attachment relationships in other dyads. Compensatory effects are also possible; that is, heightened parenting stress in one dyad may predict better attachment security between dyad members and other family members because dyad members would turn to other family members as an alternative secure base, and other family members may try to intervene in the problematic dyadic relationship. Because different reporters may adopt different roles in, and perspectives on, family dynamics (e.g., Jager et al., 2012), we anticipated the effects of parenting stress on attachment to differ across reporters.

Method

Participants and Procedure

Data in the study were part of a longitudinal project that started in 1988 and lasted for over 20 years. At the beginning of the longitudinal project, mothers and infants (healthy and full-term) from a U.S. East Coast metropolitan area were invited to participate in the study through hospital or government birth notifications, patient lists of medical groups and well-baby clinics, early education coordinators, media advertisements and birth announcements, and targeted mailings. We initially included only the first-born infants in the recruitment, but extended the sample to include some second-born children at later times. Fathers were also invited to the study when the children were 10 years old. For the broader goal of the longitudinal project (elucidating family processes specific to sociodemographic characteristics and conducting cross-cultural comparisons), we intentionally recruited a sample of ethnically homogenous European American families. The cross-cultural comparison was not a focus of the current paper as the data used in this paper were not paralleled in other cultural groups.

In this study, we used data when the target child was in late adolescence (M_age = 18.19 years, SD = 0.34) and young adulthood (M_age = 23.52 years, SD = 0.56). At these time points, families that stayed in the study were mostly of middle- to upper-socioeconomic status (mean Hollingshead SES score calculated from parents’ occupations, education, and socioeconomic status = 56.8, range = 35 to 66; see Hollingshead, 2011) and well educated. The highest level of maternal education was high school or partial college for 20.7%, a four-year college degree for 38.0%, and a graduate or professional degree for 41.3%. Fathers’ education was similar: 21.7%, 31.3%, and 46.1%, respectively. The average ages of mothers and fathers, respectively, were 49.91 years (SD = 3.80) and 51.73 years (SD = 5.58) when the target adolescent was 18 years old, and 54.70 years (SD = 4.43) and 56.80 years (SD = 5.71) when the target young adult was 23 years old. Analyses were limited to 156 families with attachment security data provided by all three family members (i.e., mother, father, and the target child), for the purpose of assessing the feasibility of the three-person model without the impact of missing data (note that SRM can be applied with the presence of missing data, but this analysis falls outside the scope of this paper). Families with complete data did not differ from families with missing attachment data in terms of demographics, except that families with missing data were less likely to be intact, t(65.93) = −2.87, p = .006. Of the 156 families, 86.5% were intact (based on mothers’ reports about living with their child’s biological father), and 51.9% of the target young adults were male.

The Institutional Review Board at the Eunice Kennedy Shriver National Institute of Child Health and Human Development approved this study (protocol #88-CH-0032). We obtained informed consent from all parents and young adults. Family members independently completed a battery of questionnaires online and were modestly compensated for participation.

Measures

Attachment insecurity.

In a round-robin design (i.e., mutual ratings among family members), at 23 years all three family members (mother, father, and young adult) were asked to report their attachment to each of the other two family members using the Adult Attachment Scale (AAS; Collins & Read, 1990). The AAS uses a 5-point Likert scale; higher scores indicate higher attachment insecurity. Two subscales were used: Depend (5 items; e.g., “I find it difficult to allow myself to depend on [family member]”) and Anxiety (5 items; e.g., “Often I worry that [family member] would like to avoid me”). AAS has proved reliable and valid in previous studies using SRM (Cook, 2000). Cronbach’s alphas of the subscales in this study ranged between .79 and .92 for different family members. Composites were formed using subscale means for each rater-ratee dyad, with higher scores reflecting higher attachment insecurity.

Parenting stress.

When the target adolescent/young adult was 18 and 23 years old, mothers and fathers reported their parenting stress on a 5-point Likert scale using the Stress Index for Parents of Adolescents (SIPA; Sheras et al., 1998). In this study, we used data from the interparental stress subscale (9 items; e.g., “I frequently argue with my spouse about how to raise my son/daughter”) and the adolescent-parent relationship distress subscale (16 items; e.g., “I cannot get my son/daughter to listen to me”). This measure has proved reliable and valid in previous studies (Sheras et al., 1998) and showed good reliability in the current sample (αs = .87 to .91 for interparental stress and .87 to .91 for adolescent-parent relationship distress). We calculated subscale sums for each reporter, with higher scores reflecting higher levels of stress.

Results

We utilized the structural equation framework to analyze the organization of family attachment (i.e., estimate the SRM model; Aim 1) and its association with previous parenting stress (Aim 2). In the following sections, we first delineate the model configurations and then continue to our findings. All analyses were conducted with Mplus 7.31 (Muthén & Muthén, 1998–2021) using a maximum likelihood estimator that is robust to nonnormality (MLR). Because 16 families had two young adult children who participated in the study, a multilevel model was estimated using the “TYPE = COMPLEX” option in Mplus to account for nonindependence in the family data. We used Kline’s (2015) guidelines to assess model fit to the data: Root Mean Square Error of Approximation (RMSEA) smaller than .05 represents close fit, between .05 and .08 represents fair fit, between .08 and .10 represents mediocre fit, and larger than .10 represents poor fit; Comparative Fit Index (CFI) larger than .90 indicates acceptable fit, and larger than .95 indicates excellent fit; Standardized Root Mean Squared Residual (SRMR) smaller than .08 suggests good fit of the model to the data. The descriptive statistics of attachment insecurity among family members are presented in Table 1.

Table 1.

Descriptive Statistics of Parenting Stress and Attachment Insecurity in the Family

	Mean	SD	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20
Age 18 Parenting Stress
1. M→Y Stress	31.33	8.24	.38**	.43**	.35**	.60**	.34**	.33**	.38**	.48**	.22*	.43**	.23*	.32**	.28**	.31**	.28**	.12	.10	.17*	.03
2. M→F Stress	19.56	7.42		.15	.58**	.19	.60**	.25*	.55**	.31**	.56**	.26**	.52**	.34**	.40**	.36**	.42**	.24**	.16	.16	.10
3. F→Y Stress	34.38	9.02			.47**	.20	.14	.61**	.36**	.24*	.00	.13	.16	.52**	.21*	.56**	.23*	.05	.14	.04	.15
4. F→M Stress	20.60	7.42				.10	.45**	.43**	.72**	.15	.33**	.07	.44**	.38**	.48**	.51**	.47**	.25*	.25*	.18	.24*
Age 23 Parenting Stress
5. M→Y Stress	31.09	8.06					.28**	.40**	.27**	.69**	.12	.74**	.14*	.37**	.18*	.35**	.21**	.41**	.18**	.45**	.13
6. M→F Stress	17.19	6.71						.30**	.60**	.31**	.66**	.31**	.68**	.28**	.35**	.33**	.38**	.15*	.28**	.11	.19**
7. F→Y Stress	32.75	10.10							.52**	.34**	.09	.32**	.33**	.77**	.30**	.83**	.27**	.28**	.40**	.26**	.42**
8. F→M Stress	17.93	7.09								.30**	.42**	.29**	.55**	.47**	.66**	.54**	.65**	.33**	.30**	.30**	.36**
Age 23 Attachment Insecurity
9. M→Y Depend	1.93	0.75									.12	.65**	.12	.36**	.17*	.28**	.08	.30**	.22**	.37**	.20*
10. M→F Depend	1.68	0.82										.13	.71**	.03	.37**	.11	.50**	.09	.19*	.03	.11
11. M→Y Anxiety	1.60	0.74											.24**	.25**	.15	.31**	.12	.51**	.25**	.44**	.19*
12. M→F Anxiety	1.50	0.73												.22**	.42**	.33**	.51**	.15	.17*	.03	.16
13. F→Y Depend	1.95	0.81													.32**	.78**	.22**	.25**	.36**	.24**	.38**
14. F→M Depend	1.64	0.75														.34**	.74**	.24**	.18*	.16*	.19*
15. F→Y Anxiety	1.59	0.70															.31**	.27**	.37**	.24**	.43**
16. F→M Anxiety	1.67	0.84																.19	.10	.12	.14
17. Y→M Depend	1.73	0.80																	.50**	.78**	.39**
18. Y→F Depend	1.75	0.87																		.36**	.81**
19. Y→M Anxiety	1.32	0.55																			.40**
20. Y→F Anxiety	1.49	0.71

Note. M = Mother. F = Father. Y = Young Adult. For parenting stress, letters before the arrow represent the reporter, and letters after the arrow represent the family member involved in the stress (i.e., stress due to the adolescent/young adult vs. the other parent). For attachment insecurity, letters before the arrow represent the reporter, and letters after the arrow represents the attachment figure. Both Depend and Anxiety were coded in the same direction – higher scores represent higher attachment insecurity.

^*p < 0.05.

^**p < 0.01.

Aim 1: Organization of Attachment Insecurity among Families of Young Adults: A Social Relations Model Approach

Analytic Plan.

We estimated the actor, partner, relationship, and family effects of attachment insecurity among young adults (Y), mothers (M), and fathers (F) by attempting to form a series of latent factors using the SRM. The latent factors were: 1) Actor effects were estimated by loading each family member’s attachment insecurity to the other two family members onto one latent factor. Three actor effect factors were estimated, one for each family member (i.e., Y, M, and F). 2) Partner effects were estimated by loading each two family members’ attachment insecurity to the third family member onto one factor. Three partner effect factors were estimated, one for each family member (i.e., Y, M, and F). 3) Relationship effects were estimated by loading each family member’s attachment insecurity to another family member (two indicators) onto one factor. As the relationship effect was estimated for each attachment relationship from one family member to another family member, six relationship effect factors (i.e., Y→M, Y→F, M→Y, F→Y, M→F, and F→M) were estimated in total. 4) The family effect (when specified; see explanations below) was estimated by loading all family members’ attachment security indicators onto a single factor. Following the guidelines by Kenny et al. (2006), all the factor loadings mentioned above were fixed at 1.

Following conventions for SRM models (Kenny et al., 2006), we also specified generalized reciprocities and dyadic reciprocities. Generalized reciprocity (i.e., the extent to which a person’s actor effect – attachment insecurity to other family members, corresponds with their partner effect – other family members’ attachment insecurity to them) was estimated by covarying each family member’s actor and partner effect. Dyadic reciprocity (i.e., the extent to which the unique attachment relationships in each dyad correspond with one another) was estimated by covarying relationship effects within each dyad. Finally, across-reporter residual variances of the same subscale were allowed to covary with one another to account for method variance due to using the same items (Kenny et al., 2006).

For three-person SRMs, estimating the family effect along with all of the actor, partner, and relationship effects would result in model underidentification and nonconvergence, requiring exclusion of some factors from the model (Kenny et al., 2006). Researchers have recommended eliminating the family effect and focusing on the more important actor, partner, and relationship effects (Kenny et al., 2006), but more nuanced steps have been proposed for determining which effect factors are the most salient and should be kept in the model. Following the suggestions by Eichelsheim et al. (2009), we utilized a three-step sequence to determine which factors to include or exclude: 1) Estimate SRM with actor, partner, and relationship effects but no family effect, 2) simplify the model by removing the SRM effects with non-significant variances, and 3) add a family effect to the simplified model. We then compared fit indices of these models and empirically determined the best model for the data. After the best model was determined, we further analyzed the relative importance of the SRM components in each reported attachment relationship by calculating the proportions of variances accounted for by each SRM factor.

Results.

Comparison of alternative models and model selection.

Below we report the three-step sequence for three-person SRM model selection (Eichelsheim et al., 2009). For the first step (Model 1a; see Figure 1 for the initial model configuration), the residual variance of young adults’ attachment anxiety towards mothers was negative but non-significant (a not unusual occurrence with three-person SRM because all factor loadings were fixed at 1 for model identification purposes; Kenny et al., 2006), as the indicators may not have equal variances empirically. The negative residual variance no longer existed when the covariances between the young adults’ attachment anxiety towards mothers and other observed indicators of attachment insecurity were set to zero (Model 1b); this additional step did not impact model fit (Satorra-Bentler adjusted comparison of Model 1a and Model 1b was Δχ²(5) = 8.36, p = .14). Model 1b fit the data well: χ²(23) = 34.54, p = .06; CFI = .984; RMSEA = .057; SRMR = .053.

Figure 1.

Illustration of the proposed Social Relations Model for attachment insecurity among young adults, mothers, and fathers. Because estimating all the SRM effect factors in the figure would result in model underidentification, following the steps suggested by Eichelsheim, Deković, Buist, and Cook (2009), we utilized the following three-step sequence to determine which factors to include or exclude: 1) Estimate SRM with actor, partner, and relationship effects but no family effect (i.e., Model 1a; only the white latent variables were estimated), 2) simplify the model by removing the SRM effects with non-significant variances (i.e., Model 2, removing the dashed latent factors), and 3) add a family effect to the simplified model (i.e., Model 3; adding the gray latent factor to Model 2). Model 2 proved to be the best model (see the result section for more information). All factor loadings were constrained to 1. Although not depicted, error variances of the observed variables and covariances between across-family member reports of the same observed variable were estimated (see the result section for exceptions). The letters before the arrow represent the reporter, and the letters after the arrow represent the target person/relationship being reported. M = Mother; Y = Young Adult; F = Father.

For the second step, building on Model 1b and following the suggestions of Eichelsheim et al. (2009) about estimating the family effect in three-person models, we removed the non-significant SRM variances one by one, starting from the one with the highest p value. After removing the first two non-significant SRM factors (i.e., Y→M and F→Y relationship effects), all remaining SRM variances were significant except for the father partner effect which was marginally significant, p = .06 (considered sufficient for producing reliable variances in the SRM literature; see Kenny et al., 2006). Model 2 with these two non-significant SRM factors removed fit the data well, χ²(27) = 37.08, p = .06; CFI = .986; RMSEA = .049; SRMR = .069; and removal of the non-significant SRM factors did not affect the model fit (Satorra-Bentler adjusted comparison of Model 2 and Model 1b: Δχ²(4) = 3.76, p = .44).

For the third step, we added the family effect into the model (Model 3; see Figure 1) by loading all the observed variables onto one additional latent factor, with the factor loadings fixed at 1.0. The variance for the family effect factor turned out to be small in size and non-significant (variance = 0.012, p = .82). Adding the family effect did not improve model fit (Satorra-Bentler adjusted comparison of Model 3 and Model 2: Δχ²(5) = 4.28, p = .51). Therefore, Model 2 (i.e., the model that excludes Y→M and F→Y relationship effects and the family effect) proved to be the final model, as it was the most parsimonious and best described the data. The variances of the SRM components of attachment insecurity for Model 2 are shown in Figure 2.

Figure 2.

The final version of Social Relations Model for attachment insecurity among young adults, mothers, and fathers (Model 2). All factor loadings were constrained to 1. Although not depicted, error variances of the observed variables and covariances between across-family member reports of the same observed variable were estimated. The letters before the arrow represent the reporter, and the letters after the arrow represent the target person/relationship being reported. M = Mother; Y = Young Adult; F = Father. Results are unstandardized except for the reciprocities (i.e., correlations; in parentheses). ^†p < .10; *p < .05; **p < .01; ***p < .001.

Results of the final model.

The final model is depicted in Figure 2. The actor effects were significant for all family members, indicating that for every family member, there was common variance in their reports of attachment insecurity to other family members. Thus, each family member’s attachment insecurity towards the other two family members can be generalized to some degree. A similar pattern was observed for the partner effects, suggesting that there was common variance in other family members’ reports of attachment insecurity towards young adults, mothers, and fathers (i.e., their characteristics as attachment figures were generalizable in the eyes of other family members). Note that the father partner effect was only marginally significant; although the literature suggests such effects should be included in SRM analyses (Kenny et al., 2006), this effect should be interpreted with caution.

Out of the six relationship effects, all but the Y→M and F→Y relationship effects, proved significant and merited inclusion in further analyses. In most cases, family members were able to form dyad-specific attachments, and these specific relationships persisted over-and-above the generalized commonalities about each person across dyads (i.e., the actor and partner effects). The non-significant (and thus omitted) Y→M and F→Y relationship effect factors, however, suggested that the reported Y→M and F→Y attachments were determined by dyad member’s own characteristics, not their dyad-specific interactions (more on this in the next sections).

How the actor, partner, and relationship components accounted for variances in each family member’s reported attachments.

The relative importance of the SRM factors for explaining each family member’s reported attachment is illustrated in Figure 3. For the young adult’s reported attachment to father (Y→F), mother’s reported attachment to the young adult (M→Y), and mother’s reported attachment to father (M→F), relationship effects (solid gray portions in Figure 3) accounted for larger proportions of variance in family members’ reports of attachment insecurity than did either actor (solid white portions in Figure 3) or partner effects (solid black portions in Figure 3) alone. For the father’s reported attachment to mother (F→M), the relationship effect also accounted for a considerable proportion of variance (only slightly smaller than the variance accounted for by father actor effect). Although each family member’s reported attachment insecurity towards other family members was partially explained by their generalized attachment representation and the attachment figure’s characteristics (i.e., actor and partner effects), it seems attachment insecurity was primarily a function of dyad-specific relationships (i.e., relationship effects).

Figure 3.

Percentage of variance of attachment insecurity indicators, by reporter and family member for which attachment insecurity was reported. Percentages not in parentheses are percentage of total variance (i.e., variance explained by SRM factors plus residual error); Percentages in parentheses are percentage of variance explained by SRM factors.

Two exceptions were the young adult’s reported attachment to mother (Y→M) and the father’s reported attachment to the young adult (F→Y), where the relationship effects were non-significant and removed from the model. For these two observed attachments, actor effects explained the largest proportion of variance, but the mother partner effect also accounted for a considerable proportion of variance in the young adult’s reported attachment to mother (Y→M). Thus, the young adult’s attachment insecurity with mother appears primarily a function of the young adult’s generalized attachment representation (i.e., actor effect) and mother’s role as an attachment figure (i.e., partner effect), whereas father’s attachment insecurity with the young adult seems determined more by father’s general attachment style.

Reciprocities of attachment relationships.

Although not a main focus of the present study, findings for the generalized reciprocities (captured in Figure 2 by the correlations between a given family member’s actor and partner effect) indicated that attachment insecurity in the family was reciprocal (i.e., those who were insecurely attached to the other family members were also insecurely attached to by the other family members), rs = .68 to .73, ps < .001. Because the non-significant Y→M and F→Y relationship effects were removed from model, only the dyadic reciprocity between M→F and F→M relationship effects was estimated, and the correlation was significant, r = .52, p < .05. This association indicates that mother’s unique attachment insecurity to father corresponded with father’s unique attachment insecurity to mother.

Summary.

The results generally supported our hypotheses that the reported attachments among three family members can be partitioned into individual (i.e., actor and partner effects) and dyad-specific (i.e., relationship effects) components. Specifically, relationship-specific effects accounted for a considerable proportion of variance in family members’ reported attachment insecurity, and actor and partner effects explained smaller but significant variances. Two exceptions were for the Y→M and F→Y attachment reports, in which the relationship-specific effects were non-significant and excluded from the model, and only the actor and partner effects explained the reported attachments. After accounting for the actor, partner, and relationship effects, the family-level variance in attachment was negligible. The non-significant family-level variance was contradictory to our hypotheses, but was consistent with the literature that the family-level dynamic is usually well explained by the individual and dyadic components of relationships among family members (Kenny et al., 2006).

Aim 2: Predicting SRM Attachment Components from Earlier Parenting Stress

Analytic plan.

After disaggregating attachment insecurity among family members into individual and dyadic level components, we further examined how these components of attachment insecurity were associated with changes in family relationships from late adolescence to early adulthood. Specifically, we attempted to understand how 18-year parenting stress and instabilities (i.e., changes in the rank order) of parenting stress in different domains from 18 to 23 years uniquely predicted different SRM components of attachment insecurity.

Based on the suggestions of Lanz and Tagliabue (2014), we used factor scores estimated by Mplus 7.31 to quantify the SRM effects. To examine the unique contribution of parenting stress in the co-parenting and parent-child domains, we included interparental parenting stress and adolescent/young adult-parent relationship distress as separate predictors, and we utilized both mothers’ and fathers’ reports of parenting stress to understand reporter effects. Each SRM component was predicted by the initial level (i.e., age 18 parenting stress) and instabilities of parenting stress from age 18 to 23 (i.e., residual of age 23 parenting stress regressed on age 18 parenting stress, which reflects relative change across ages 18 to 23, or the extent to which one’s position relative to others in the sample changed from ages 18 to 23 years). SES and family intactness at age 18 were included as covariates. Missing data were handled using full-information maximum likelihood (FIML). We predicted each SRM component in separate models using the same predictors. The models fit the data well: χ²(8) = 11.19 – 13.83, ps = .09 – .19; CFI = .982 – .991; RMSEA = .051 – .068; SRMR = .034 – .042.

Results.

Results for Aim 2 are shown in Table 2. In general, instabilities in parenting stress rather than initial levels of parenting stress in adolescence more strongly predicted components of attachment, but the effects varied by domains and reporters of parenting stress. Two predominant types of effects emerged from the pattern of predictions. First, parenting stress specific to a dyad predicted worse attachment specific to the dyad (i.e., relationship effects) as well as worse attachment characteristics among the actor and partner effects of the dyad members. For example, instabilities in interparental (M→F and F→M) parenting stress predicted worse attachment characteristics of mother and father (i.e., mother and father actor and partner effects as well as M→F and F→M relationship effects). However, instabilities in mother-reported interparental (M→F) parenting stress were specifically associated with the father’s role as an attachment figure (i.e., father partner effect) and mothers’ unique attachment insecurity to father (i.e., M→F relationship effect). By contrast, instabilities in father-reported interparental (F→M) parenting stress were broadly associated with mother and father actor and partner effects, which suggested mother- and father-reported interparental parenting stress have different meanings. In addition, instabilities in mother- and father-reported stress due to the adolescent/young adult (M→Y and F→Y) predicted worse individual-level attachment of the parent and the young adult (i.e., the actor and partner effects of the parent and the young adult).

Table 2.

Predictive effects of Parenting Stress on SRM Attachment Insecurity Factors

	SRM Factors
Predictor	M Actor	F Actor	Y Actor	M Partner	F Partner	Y Partner	M→F Rel	F→M Rel	Y→M Rela	Y→F Rel	M→Y Rel	F→Y Rela
Initial F→M stress	−0.18	−0.04	0.07	−0.09	−0.03	−0.07	−0.06	−0.16	-	0.02	−0.28*	-
Instability F→M stress	0.41***	0.38***	0.04	0.55***	0.27**	−0.02	0.09	0.78***	-	0.13	0.20	-
Initial M→F stress	0.21	0.02	−0.02	0.14	0.06	0.05	0.24*	0.04	-	−0.18	0.00	-
Instability M→F stress	0.11	−0.02	−0.14	−0.17	0.25**	−0.11	0.54***	0.02	-	0.15	0.07	-
Initial F→Y stress	−0.21	0.00	−0.08	−0.35**	−0.05	0.12	−0.08	−0.27*	-	−0.18	0.07	-
Instability F→Y stress	−0.37***	0.54***	0.36***	−0.32**	0.52***	0.51***	−0.11	−0.35***	-	0.11	−0.31**	-
Initial M→Y stress	0.07	0.02	−0.20	0.04	0.00	−0.08	0.02	0.15	-	0.02	0.18	-
Instability M→Y stress	0.46***	−0.27**	0.43***	0.43***	−0.33***	0.42***	−0.20*	−0.03	-	−0.08	0.52***	-

Note. Standardized coefficients are reported in this table. Some coefficients are not reported (i.e., represented by the dashes) because the variance of the SRM effect was not significant. M = Mother. F = Father. Y = Young Adult. Actor = SRM actor effect. Partner = SRM partner effect. Rel = SRM relationship effect. Initial = Age 18 parenting stress. Instability = Residual score of Age 23 parenting stress regressed on Age 18 parenting stress. F→M stress = Father reported interparental parenting stress. M→F stress = Mother reported interparental parenting stress. F→Y stress = Father reported adolescent/young adult-related parenting stress. M→Y stress = Mother reported adolescent/young adult-related parenting stress.

^aFactor omitted from analysis due to non-significant variance.

^*p < 0.05.

^**p < 0.01.

^***p < .001.

Second, we note compensatory effects such that, when parenting stress in one dyad increased, the dyad members would develop better attachments to, and/or be attached to more securely by, the third family member. For example, instabilities in parenting stress in the father-young adult dyad (F→Y) predicted lower attachment insecurity (i.e., better attachment security) between the father/young adult and the mother (i.e., mother actor and partner effects), father’s lower attachment insecurity unique to mother (i.e., F→M relationship effect), and mother’s lower attachment insecurity unique to the young adult (i.e., M→Y relationship effect). The same compensatory pattern was observed for parenting stress in the mother-young adult dyad. Fewer compensatory effects were found for interparental (M→F and F→M) parenting stress, but higher levels of father-reported interparental (F→M) stress at age 18 also were associated with a higher level of mother attachment security unique to the young adult (i.e., M→Y relationship effect).

Summary.

Parenting stress – and particularly instabilities (i.e., increase in rank order) in parenting stress between ages 18 and 23 - positively predicted young adulthood attachment insecurity, and the associations between parenting stress and attachment insecurity primarily occurred within each family subsystem. Family members may also serve a compensatory role when the tension between the other two family members increased.

Discussion

Among a European American, largely middle- to upper-class sample, we used the Social Relations Model to examine the organization of attachments in the family of young adults (i.e., mother, father, and young adult child) by separating familial attachment actor effects (how each family member generally attaches to other family members), partner effects (how each family member is generally attached to by other family members), and relationship effects (the dyad-specific attachment between family members). We also systematically examined relations between earlier parenting stress and each component of attachment.

This study demonstrated that the three-person SRM design can be applied to research on family attachment with a few statistical cautions. As the three-person design is easier to implement than four-or-more-person designs, our findings broaden the application of the SRM in attachment research. Attachment in the family, as hypothesized, can be separated into person- and dyad-specific components, but some of the components may not be salient in young adulthood. The components of attachment are also differentially associated with prior parenting stress in late adolescence to young adulthood. Below we discuss implications of these findings.

How Is Attachment Organized in the Family in Young Adulthood?

The first aim of this study was to examine the organization of attachment in the family during young adulthood. Consistent with our expectations and some literature on adolescence (e.g., Buist et al., 2004; Cook, 2000), our findings indicate that attachment in the family during young adulthood can be productively separated into actor, partner, and relationship effects. Specifically, actor, partner, and relationship effects significantly explained four out of six observed attachment relationships (F→M, M→F, M→Y, and Y→F) in the family, and relationship effects showed the largest relative contribution to attachment compared to actor and partner effects (except for F→M attachment in which the relationship effect and the father actor effect accounted for similar proportions of variance in attachment). Thus, the most powerful driver for these four attachment relationships is specific to each relationship. The remaining two attachment relationships (Y→M and F→Y) were mostly explained by the general characteristics of dyad members, but not dyad-specific interactions.

These findings are partially consistent with, but more complicated than, previous findings described in the literature. As in all previous SRM attachment studies (Buist et al., 2004; Buist et al., 2008; Cook, 2000; De Meulenaere et al., 2021), our study supported the salience of all the actor effects, which makes sense as adults are theorized to form generalized internal working models across relationships (Bretherton & Munholland, 2008). However, some SRM studies suggest that adolescents’ relationship-specific component of attachment increases with age (Buist et al., 2004; Buist et al., 2008), whereas the partner effects (usually for the parents but sometimes adolescents too) are less salient, indicating that family members may have very different views on a specific family member’s role as an attachment figure (Buist et al., 2008; De Meulenaere et al., 2021). We were not able to replicate those findings in young adulthood. Specifically, our findings suggest that partner effects explain significant proportions of variance of the reported attachments, and attachments between young adults and their parents are not universally more dyad-specific. Instead, the organization of attachment appears to reflect family members’ different roles and emergent interaction patterns in families of young adults. Although unique responses to different attachment figures may increase with age (Buist et al., 2004; Buist et al., 2008), young adults’ increasing independence from parents may diminish the frequency of dyad-specific interactions, and consequently result in more generalized attachments between young adults and parents, as indicated by our findings. The extent of generalization further differs between parents, as evidenced by the mother’s reported attachment to the young adult (M→Y) being more dyad-specific than the father’s reported attachment to the young adult (F→Y), while the young adult’s reported attachment to the mother (Y→M) was less dyad-specific than young adult’s reported attachment to the father (Y→F). Attachment with mothers and fathers has been a complicated topic as the assessment methods do not always align and research findings are not fully consistent (Paquette et al., 2013). Measures of attachment are centered on mother-child dyads and may not apply equally to father-child dyads (Bornstein, 2015; Cabrera & Volling, 2019); additionally, the meanings of young adults’ attachment to their parents and parents’ attachment to young adults may be different. Despite the complexities, compared with fathers, mothers tend to play the caregiving role in the family and attend to the needs of each family member (Bornstein, 2015; Palkovitz et al., 2014). Mothers may cater to their relationships with young adults more specifically than fathers do, resulting in unique dyadic adjustments (i.e., relationship effects) in M→Y, but not F→Y, attachments. In turn, young adults are more likely to rely on mothers than fathers as the primary attachment figure and source of support (Freeman & Almond, 2010). Their relationships with mothers may be symbolic of the “default” general secure base in the family, whereas their attachment with their fathers is more dyad-specific and variable, depending on fathers’ roles in the family and resources (Schoppe-Sullivan et al., 2021). Although speculative, differences in mother and father roles may make young adults view their relationships with their fathers as “additional” to their relationships with their mothers, resulting in more dyad-specific variance in Y→F, but not Y→M, attachments.

Notably, we found very little variation in the average level of attachment insecurity in the family after accounting for individual and dyadic level variations (i.e., small and non-significant family effect which was removed from the model), which is consistent with the literature that the family atmosphere component may not be salient in attachment research, especially for three-person designs (Cook, 2000; Kenny et al., 2006).

How Do Stability and Change in Parenting Stress Predict Attachment in the Family?

The second aim of this study was to predict attachment in the family using the initial level and instabilities (i.e., changes of the rank order) of parenting stress from late adolescence to early adulthood. Consistent with the literature (Kor et al., 2012), heightened parenting stress predicted higher attachment insecurity. The instabilities of parenting stress from age 18 to 23 had broader impacts on the family system than 18-year levels did, which is consistent with family life cycle theory’s (Carter & McGoldrick, 2005) position that increases in parenting stress reflect maladjustment to family life cycle transitions and thus contribute more to attachment problems in the family. This finding suggests that it is important to go beyond viewing parenting style or parental beliefs as a static characteristic of parents when examining parental effects on families and children, as parenting can be “updated” by the family life course as well as by external factors, and changes in parenting in response to life events and stress can play more powerful predictive roles than previous levels.

Our findings supported family subsystem-specific effects of parenting stress on attachment. As expected, stress within each dyad primarily predicted dyad-specific components of attachment insecurity (i.e., relationship effects) as well as the actor and partner effects of the dyad members. Such findings were replicated for parenting stress related to the mother-young adult, father-young adult, and mother-father dyads. In other words, increased stress in a family subsystem during the transitional period of late adolescence to young adulthood is generally associated with attachment problems within the specific subsystem.

Moreover, evidence emerged that stress in one subsystem of the family was related to attachment in other family subsystems. Specifically, consistent with the literature (e.g., Nelson et al., 2009), we observed some compensatory effects within the family system. For example, when fathers reported increasing stress due to interaction with the adolescent/young adult, the generalized attachment between father/young adult and mother (i.e., mother actor and partner effects) became more secure, and father and young adult also developed more secure dyad-specific relationships with mother (i.e., F→M and M→Y relationship effects). Similar findings were observed for mother-reported parenting stress due to the adolescent/young adult. However, compensatory effects occurred mostly for parenting stress due to the adolescent/young adult. For interparental/co-parenting stress, father-reported high initial levels of parenting stress at late adolescence predicted better security in the unique attachment from mother to the young adult in early adulthood, but no other compensatory effect was found. In general, these results suggest that, when there is tension between two family members, the third family member (usually the other parent but not the young adult) may intervene or serve as a secure base of the family.

In addition to the compensatory effects, we also anticipated that stress within one family subsystem/dyad would spill over to another family subsystem. However, the findings did not support this hypothesis. Because our sample consisted mostly of middle- to upper-class and low- risk families, perhaps the participants had more resources to respond to maladjustment in other family subsystems in healthy and positive ways (e.g., stepping in to provide a secure base), rather than become negatively influenced by tensions arising in other subsystems. Indeed, compensatory effects, rather than spillover effects, are more commonly found among families with higher socioeconomic status (e.g., Nelson et al., 2009). It is possible that families that face more risks in their lives may have more challenges and fewer resources when adjusting to the transitional period, which may make stress in one dyad more likely to spill over to other dyads.

Different informants’ reports may reflect different problems in the family system. For example, father-reported parenting stress due to mother contributed broadly to all aspects of attachment insecurity in the marital dyad (i.e., positively predicted mother and father actor and partner effects as well as the F→M relationship effect), whereas mother-reported parenting stress due to father was only associated with father partner effects and the M→F relationship effect. It seems that mother-reported co-parenting stress specifically reflects father’s attachment problems, whereas father-reported co-parenting stress indicates attachment problems of both the mother and father. Alternatively, fathers may have more difficulty handling increasing co-parenting stress than mothers, which negatively influences the attachment of themselves and mothers.

In summary, we found that rank-order increases in parenting stress from late adolescence to early adulthood are predictive of later attachment problems. Specifically, stress in each dyad of the family is associated with higher attachment insecurity within the dyad, but when one parent is distressed about the adolescent/young adult, that stress often draws attention of the other parent and encourages the other parent to play positive roles in the family. Therefore, when parenting stress and attachment problems are observed between the young adult and one of the parents, clinicians and counselors in the intervention need to focus on problems within the dyad and emphasize the potential of positive roles of the other parent or other family members.

Limitations and Future Directions

This study has a few limitations. First, although the sample size was considered good for Social Relations Model (Kenny et al., 2006), it was not large enough to allow multiple-group analysis that would compare, for example, potential young adult gender differences. The possibility that young men and women differ in their attachments to their mothers and fathers should be examined in future studies with larger samples. Second, and more importantly, the sample mostly consisted of middle- to upper-SES European American intact families who may be more capable of coping adaptively with family tension in this transitional period, as evidenced by their relatively low self-reported attachment insecurity. The findings may not generalize to culturally different or at-risk samples. Parental roles, involvement, and parent-child relationships vary greatly across cultural and ethnic groups, as well as across socioeconomic statuses (e.g., DeBell, 2008), which means the structure of attachment relationships is also likely to vary. For example, cultural values such as familism may set up alternative independence/interdependence goals for young adults and impact attachments in the family. Less-intact families may be more prone to experience lower integration of actor and partner effects in their attachments. Further, the meaning and sources of parenting stress during young adulthood may differ across social groups, and the effects of parenting stress could differ as well (e.g., negative spillover effects, which we did not find in the current sample, may be more prominent in at-risk samples). Considering all these complexities, future studies need to either utilize diverse samples or utilize homogenous samples of other sub-populations (Jager et al., 2017), such as samples limited to ethnically minority groups or samples limited to non-intact families, to achieve a more inclusive understanding of parenting stress and attachment. Finally, the associations between parenting stress and attachment insecurity may be bidirectional, such that earlier attachment insecurity may also contribute to parenting stress. Future studies should incorporate longitudinal data assessed at multiple times to examine the reciprocal interplay between parenting stress and attachment.

Conclusion

This study expands the body of research using the SRM to examine attachment in the family with a three-person design. It demonstrates attachment in the family can be attributed to the characteristics of family members and dyad-specific dynamics between family members and extends previous SRM attachment research on adolescence to young adulthood. By predicting the components of attachment using dyad-specific indicators of parenting stress, this study further demonstrates that parenting stress in a family subsystem primarily contributes to attachment problems within subsystems, but family members are able to react positively and serve as sources of secure attachment when they sense stress in other family subsystems. Specifically, the findings highlight the role of increasing parenting stress on family relationships, and suggest that changes in stress and family dynamics may play a major role in the family’s adjustment. Such findings correspond with the applied literature that counselors should take family systems into account when working with adolescents and young adults (Cretzmeyer, 2003; Kenny & Rice, 1995). Intervention programs may focus on individuals’ internal working models (e.g., cognitive behavioral therapy) as well as the renegotiation of dyad-specific relationships; in addition, interventions that involve multiple family members may be helpful in improving dyadic adjustments as well as emphasizing each family member’s positive impact on other dyads in the family. Indeed, therapeutic approaches that target family dynamics have yielded promising results for adolescents’ adjustment (Diamond et al., 2002) and may have further implications for the transition to young adulthood (Cretzmeyer, 2003).