Auburn University

Auburn University

Duke University

Indiana University

Duke University

The delayed entry into marriage that characterizes modern society raises questions about young adults' romantic relationship trajectories and whether patterns found to characterize adolescent romantic relationships persist into young adulthood. The current study traced developmental transitions into and out of romantic relationships from age 18 through age 25 in a sample of 511 young adults. The developmental antecedents of these different romantic relationship experiences in both distal and proximal family and peer domains were also examined. Analyses included both person-oriented and variable-oriented approaches. Findings show 5 distinct clusters varying in timing, duration, and frequency of participation in romantic relationships that range from those who had only recently entered into a romantic relationship to those who had been in the same relationship from age 18 to age 25. These relationship outcome trajectory clusters were predicted by variations in competence in early relationships with family and peers. Interpersonal experiences in family and peer contexts in early childhood through adolescence thus may form a scaffold on which later competence in romantic relationships develops. Findings shed light on both normative and nonnormative developmental transitions of romantic relationships in young adulthood.

The achievement of intimacy in a romantic relationship is considered to be one of the critical developmental tasks marking one's entry into adulthood (Conger, Cui, Bryant, & Elder, 2000). Failure to establish and sustain a committed intimate relationship during this time is thought to not only hinder development (Erikson, 1968) but have serious negative implications for well-being across the life span (Kiecolt-Glaser & Newton, 2001). Thus, given the crucial role romantic relationships play in both short- and long-term adjustment (Lehnart, Neyer, & Eccles, 2010; Schulenberg, Bryant, & O'Malley, 2004), researchers have begun to focus on the roots of these relationships. By understanding the normative developmental sequence of romantic relationships, it is thought that programs can be better shaped to facilitate successful transitions into adulthood and adult relationships for a number of young adults who otherwise might struggle to achieve these developmental tasks (Karney, Beckett, Collins, & Shaw, 2007).

Though this goal is laudable, researchers face two challenges in trying to operationally define normative romantic development in young adulthood. First, earlier entry into romantic relationships in adolescence coupled with the rising age at first marriage means the period in which individuals are having premarital relationships has lengthened to over a decade for many young adults (Carver, Joyner, & Udry, 2003; U.S. Census Bureau, 2005). This extended period of time allows for greater variability in young adults' romantic experiences (Furman, Brown, & Feiring, 1999). The second challenge is a consequence of this prolonged period, namely that adulthood itself has become a moving target for most individuals, with greater interindividual variability emerging in the timing and content of developmental milestones (Cohen, Kasen, Chen, Hartmark, & Gordon, 2003; Schulenberg et al., 2004). Efforts to identify a normative romantic relationship profile for young adults today are therefore inherently difficult, as there is likely no single typical pattern. This is particularly true for establishing intimacy, because, unlike other developmental tasks, few individuals initially feel up to this challenge (Prager, 1986). For many young adults, the capacity for intimacy is fine-tuned only with substantial experience (Cantor, Acker, & Cook-Flanagan, 1992). The nature of the experience required before one can achieve a committed relationship, however, remains unclear but is likely to vary based on groundwork laid earlier in the life span (Conger et al., 2000). Thus, the premise of the current study is that understanding the nature of young adult romantic relationship progression requires drawing upon both a person-centered approach that acknowledges the likelihood of variability of romantic relationship experiences during this period and a variable-centered approach that helps identify the developmental antecedents associated with these different experiences.

Theorists have proposed that the development of early romantic relationships follows a phase-based approach, whereby adolescents begin with fairly short-term, shallow romantic connections primarily occurring in peer groups that develop into multiple shorter relationships that occur less in the group context and are defined more by emotional intimacy. In late adolescence, romantic development is thought to culminate in a single, committed intimate relationship of extended duration (Brown, 1999; Collins, 2003; Seiffge-Krenke, 2003). Using data from the National Longitudinal Study of Adolescent Health, Meier and Allen (2009) found this “progression” from limited or no romantic relationship experience to more serious romantic relationship experience was more prevalent than “regression” (i.e., going from a steady relationship to no relationship). This progression from less committed romantic relationships to a single, committed “adult-like” intimate relationship is thought to be not only normative but essential for achieving adult status (Lanz & Tagliabue, 2007).

Consistent with this view, Schulenberg and colleagues (2004) defined romantic success as being in a committed relationship by age 26 (e.g., married, engaged) with no prior divorce. Maintaining was defined as having a less committed relationship (e.g., cohabiting) or having a divorce. Young adults were thought to have stalled if they had limited romantic experience (e.g., not married, dating once a month or less). This suggests stability is a critical measure of romantic success for young adults. Though there are many different dimensions by which to judge intimate relationships (Conger et al., 2000), focusing on relationship stability is sound because it promotes mental and physical well-being throughout the life span (Waite, 1995). Further, the dissolution of a close romantic relationship is thought to be one of the most traumatic events individuals experience (Simpson, 1987), a conclusion bolstered by a large literature on the effects of divorce (see Amato, 2000). Thus, to capture the stability of romantic relationships in young adulthood, the current study examines the amount of romantic involvement and turnover experienced across this period. Not only is this approach in line with previous work on the key structural features of early romantic relationships (Davies & Windle, 2000), but it allowed us to determine whether young adults are achieving stable, long-term relationships as theorized.

Despite growing evidence that the progression to a single, stable relationship is optimal, this is not a path taken by all. For example, though Meier and Allen (2009) provided evidence for a normative romantic sequence in adolescence, their findings suggest romantic relationships are rather diverse. Six unique sequences emerged over the two waves (T1: ages 11–18; T2: 13–20): (a) no relationships at either age (38% of sample); (b) no relationships to casual relationships (10%); (c) stability in casual relationships (3%); (d) one steady relationship to no or casual relationships (18%); (e) no or casual relationships to a steady relationship (14%); and (f) stable in a steady relationship (18%). Thus, only a third of the sample was in a steady relationship at T2 (Groups 5 and 6), with most of those individuals being females. Males, minorities, and low-income adolescents were more likely to have had no relationship experience. Similarly, Wood, Avellar, and Goesling (2008)— drawing upon the National Longitudinal Survey of Youth, 1997— found only 33% of young adults between the ages of 20 and 25 reported being in a committed relationship (married or cohabiting), with the remaining individuals either dating (27%) or not in any romantic relationship (40%). Again, females were more likely to be in a committed relationship, as were individuals whose romantic and sexual experiences started earlier in adolescence. Finally, Schulenberg and colleagues (2004) found that only 51% of their sample had achieved a committed relationship by age 26. Though being in a committed relationship in young adulthood may have been normative in previous cohorts (Cherlin, 2009), these studies call into question how pervasive commitment is at this stage of development for the current young adult cohort, particularly for certain groups of young adults, and suggest the disparate patterns Meier and Allen (2009) found to characterize adolescence may persist into young adulthood.

In light of accumulating evidence of alternative pathways toward long-term commitment, conceptual frameworks that accommodate diversity in romantic relationship experiences could prove useful. Arnett's (2000) theory of emerging adulthood offers such a framework, predicting continuing diversity in romantic experiences and a delaying of commitment well into the 20s. In this theory, the period from 18 to 25 is a time of exploration and instability, more characterized by a self-focus than a focus on establishing a lasting connection with someone else. Thus, we would expect multiple romantic relationship sequences that would likely parallel Meier and Allen's (2009) patterns. Whether this diversity in romantic relationship experiences comes at the expense of young adults' eventual romantic success appears to depend on how stability is conceptualized. Though Seiffge-Krenke (2003) proposed that greater involvement, be it with one partner or many, early on leads to later positive romantic outcomes, the work on romantic dissolutions suggests high amounts of partner turnover could be problematic (Amato, 2000; Simpson, 1987). Davies and Windle (2000) found adolescent romantic relationships with high involvement but high turnover had different effects on adjustment than did relationships characterized by high involvement with a steady partner. Thus, although early romantic involvement and turnover are related, the two pieces of romantic stability appear to have distinct outcomes. The question of central interest in the current study is whether they have distinct antecedents as well, and whether these antecedents represent coherent pathways through which the key features of romantic relationship stability may develop. Given the importance of establishing a committed intimate relationship for achieving adult status (Lehnart et al., 2010), it is critical to know what kinds of early experiences appear to better equip young adults to achieve romantic relationship stability.

The capacity for intimacy in adulthood is thought to evolve from earlier interpersonal experiences that accumulate across childhood and adolescence (Collins, Hennighausen, Schmit, & Sroufe, 1997). As to what experiences are most critical and when, a rich body of literature now indicates that individual differences in romantic relationships accrue through both positive and negative interactions with family and peers (Collins & van Dulmen, 2006; Conger et al., 2000). To help determine how these interactions might accrue over multiple developmental periods, we draw upon the developmental cascades model (Masten & Cicchetti, 2010), which suggests that competence in early relationships with family and peers becomes the scaffold on which later competence in newly emerging domains such as romantic relationships develop.

Within the family domain, longitudinal research reveals that the history of parent–child relationships early in life significantly predicts an individual's ability to initiate and maintain romantic relationships as an adult (Collins & van Dulmen, 2006; Franz, McClelland, & Weinberger, 1991). Collins and Sroufe (1999) suggested that caregiver relationships may influence romantic development by shaping children's relational abilities and expectancies. As to what features of the caregiver relationship are important, sensitivity to developmental context requires a consideration of which measures might best represent key relationship experiences at each period (Pettit et al., 2006). Early on, parents who are overly punitive or harsh teach children that connecting to others can be risky, which explains why early harsh parenting has been associated with later challenges in establishing healthy, stable romantic relationships as a young adult (Conger et al., 2000; Franz et al., 1991). In contrast, parents who are warm and proactive in their parenting teach children that relationships can be rewarding and fulfilling. This appears to be the case, as greater parental monitoring and higher quality parent– child relationships as the child matures have been linked to greater competence in romantic relationships later in life (Longmore, Manning, & Giordano, 2001; Overbeek, Stattin, Vermulst, Ha, & Engels, 2007; Pettit et al., 2006). Although these studies provide persuasive evidence of predictive links between parent– child relationships and later romantic development in young adulthood, Seiffge-Krenke (2003) found their influence may begin to wane as romantic relationships deepen. This is not surprising in light of the developmental cascade model, as one would expect other domains of influence to emerge as individuals mature.

As children develop, the peer domain begins to take on greater importance for romantic development (Collins et al., 1997), so peers may supplant some of parents' influence on romantic relationships (Seiffge-Krenke, 2003). Peers' growing influence is not surprising, as the peer network is often the pool from which romantic partners are chosen (Furman, 1999). Socially skilled and connected children have greater opportunities for initiating romantic relationships than do more isolated or socially awkward children (Simpson, Collins, & Salvatore, 2011). Because friendships involve intimacy needs and skills, they may be a root of later romantic relationships (Collins & van Dulmen, 2006; Sullivan, 1953). Peer relationships may then act as a bridge between parents and romantic relationships, as learning to meet the need for intimacy through friendships gives adolescents the confidence and skills to go outside the caregiver relationship to fill this need. However, characteristics of the friends may be important in shaping adolescents' expectations and abilities in later romantic relationships. Research bears this out, as peer networks characterized by deviance predict less healthy romantic development (Whitbeck, Yoder, Hoyt, & Conger, 1999), whereas having a close group of supportive peers positively affects the timing and stability of romantic relationships (Connolly, Furman, & Konarski, 2000; Dhariwal, Connolly, Paciello, & Caprara, 2009). Thus, it appears relationships with both parents and peers work together to shape the course of romantic relationship development in young adulthood (Simpson et al., 2011), leading to likely variations in romantic pathways as individuals accumulate different experiences both across and within these critical domains over time.

The objective of the current study was to identify and describe variations in romantic relationship experiences in young adulthood and their antecedents in a longitudinal, multisite study of males and females. Beginning at age 18 and continuing to age 25, participants were asked about their romantic relationships and whether they were with the same or a new partner. The current study is well positioned to address whether patterns of romantic involvement and stability in young adulthood map onto patterns found earlier in adolescence (Meier & Allen, 2009). Use of a person-oriented approach allows for the possibility these features of romantic involvement may be connected in different ways for different young adults, which can augment traditional variable-centered methods with their focus on more aggregate-level associations (Zarrett et al., 2009). Finally, the current study draws upon multidimensional (parents, peers), multiple-informant (participant, parents, teachers, peers, observers) data spanning 12 years of development in early childhood, middle childhood, and adolescence (ages 5–16) to explore the possible antecedents of these different young adult romantic relationship experiences.

Several questions were of interest in the current investigation. To what extent are young adults establishing and maintaining committed intimate relationships? Further, what kinds of configurations of romantic stability/instability characterize this period? Based on work on the variability of early romantic relationships coupled with the instability that characterizes young adulthood (Arnett, 2000; Wood et al., 2008), we hypothesized young adults would vary in both the extent to which they were involved in romantic relationships and how much partner turnover they experienced. Similar to Meier and Allen's (2009) groups, we expected to find a group of young adults who were already in a single, long-term relationship. We next expected to find two groups that demonstrated progression to a committed relationship—the first having more consistent romantic involvement characterized by a few long-term relationships and the second, reflecting that this progression may take longer for some individuals, having less overall involvement but still reporting a relationship by the end of the study period. Capturing the nonprogressing groups, we expected a group of young adults with both high involvement and high turnover. For the fifth and final group, we expected to find young adults with little to no romantic involvement.

Finally, we drew upon the developmental cascade model to address what leads young adults to have different pathways, examining positive and negative experiences within the family and peer domains at multiple stages of development as predictors of romantic involvement and turnover. We utilized person-centered and variable-centered approaches to identify a cumulative progression of influences starting with the most distal influences in early childhood (proactive parenting, harsh discipline), continuing to middle childhood (physical discipline, parental monitoring, peer competence), and then to the proximal influences in adolescence (parent–child relationship quality, friends' deviance and support) on both the number of waves young adults were in a relationship from ages 18 to 25 and the number of partners they had during this time. The current study not only sheds light on young adult romantic relationship development but also begins to link patterns of developmental influences over time to understand why some young adults progress to more committed relationships, whereas others diverge from this path.

Data for this project were drawn from an ongoing, multisite longitudinal study of child development (Pettit, Bates, & Dodge, 1997). Children entering kindergarten were recruited from two cohorts—one in 1987 (n = 308) and one in 1988 (n = 277)—from three sites: Knoxville and Nashville, Tennessee, and Bloomington, Indiana. The sample consisted of 585 families at the first wave; this sample was demographically representative of the communities from which it was drawn. Males comprised 52% of the sample; 81% of the sample was European American, 17% was African American, and 2% was from other groups. Follow-up assessments were conducted annually through age 25 through face-to-face interviews, telephone interviews, or questionnaire mail-outs. To have complete data for the cluster analyses, analyses for the present study were based on 87% (n = 511) of the original 585 participants who provided data on both romantic relationship variables (number of partners, number of waves in a relationship) between ages 18 and 25. Within this subsample, 51% of the participants were male and 16% were minorities. By age 25, 14% of the sample had not graduated from high school, 19% were high school graduates, 32% had some college, and 35% had graduated college. Beginning at 15, parenthood status was assessed annually using a dichotomous score to indicate if participants had become a parent (1) or not (0) by age 25. The participants included in the analyses were of higher socioeconomic-status families than were the 73 original participants not included in the analyses, F(1, 568) = 4.98, p < .001; were more likely to be female, χ2(1) = 5.65, p < .05; and were more likely to be European American, χ2(2) = 13.40, p < .001; but these two groups did not differ by parents' marital status changes or by mother-rated internalizing or externalizing behavior problems at age 5.

From ages 18 to 25, interviews were conducted to assess romantic relationship involvement. Participants were first asked “Do you currently have a romantic partner?” Those who responded “yes” were asked a series of questions about their relationship, including its duration and their partner's name. From these answers, we derived two indices of romantic involvement: the number of romantic partners identified across the eight waves and the amount of romantic involvement, as indicated by the number of waves the participant reported being in a romantic relationship during the study period. Numbers of participants reporting involvement in a relationship can be seen in Table 1.

To assess the discriminant validity of the clusters, four relationship variables were assessed. First, individuals reported the duration of their romantic relationship in months. A variable averaging across these relationships was created to capture the average length of relationships during the study period. The stability of these relationships was captured with a variable reflecting the number of waves that individuals listed the same partner. The variable ranged from 1 to 7, with 1 indicating a new partner and 7 signifying a relationship that lasted the entire study period. To examine the timing of these relationships, a variable ranging from 18 to 25 was created that reflected the first age an individual reported having a romantic relationship. Age at first sexual intercourse was assessed with in-home interviews at age 16 and continuing through each subsequent year until participants were 23. In each year, participants were asked whether they had ever had sexual intercourse. Those who responded that they had had sex at or by age 16 were grouped together, as in past studies (e.g., Fergusson & Woodward, 2000).

Both positive and negative aspects of parenting were of interest in the early childhood period. To capture positive parenting, mothers completed the Concerns and Constraints Questionnaire (Pettit et al., 1997) when participants were 5. Mothers were presented with five hypothetical stories depicting a child misbehaving with peers and asked what they would do if their child behaved in that manner and what they might have done to prevent the misbehavior. Responses were coded as doing nothing, after-the-fact punishment, after-the-fact guidance and reasoning, before-the-fact preventive but general, and before-the-fact preventive and situation specific. Mothers who used either of the latter two approaches received a score of 1; those using any other approach received a score of 0. Scores were summed across the five stories to create a measure of proactive parenting (α = .70). To capture negative parenting, mothers were asked a number of open-ended questions for each era (age 1 to 4 and age 4 to 5) during the initial interview: “Who usually disciplines your child?” “How?” “Was your child ever physically punished?” “How often?” Based on these answers, the interviewer rated the use of discipline from nonrestrictive, mostly positive guidance (1) to severe, strict, often physical (5). These ratings were averaged across the two eras to produce an overall harsh discipline score. The correlation among independent raters was r = .80, with an alpha coefficient of .61 across eras.

Both parenting and peer experiences were of interest in middle childhood. For the negative aspect of parenting at age 12, mothers were asked about their use of physical discipline when dealing with misbehavior during the last year, including (a) slapping or hitting with their hand, (b) spanking, and (c) using a belt or paddle. The frequency of each type of discipline was rated on a 4-point scale (1 = never, 2 = rarely, 3 = sometimes, and 4 = frequently) and the items were averaged to reflect the frequency with which mothers used physical discipline (α= .71). For positive parenting, mothers were asked about their monitoring and supervision of their child (Pettit, Bates, Dodge, & Meece, 1999). Nine items rated on 5-point scales were averaged to reflect parents' awareness of their children's activities, whereabouts, and friendships (e.g., “When your child is not at home, do you know who he/she is with?”; α = .69).

To assess children's peer competence, sociometric measures and teacher ratings in Grades 1, 2, and 3 were used (see Pettit et al., 2006). Sociometric nomination data were used to create three indices of classroom peer experience: average social preference (difference between standardized liked most and liked least nominations), reciprocated friendships (average number of reciprocated 5s on a 5-point liking scale), and mutual disliking (number of reciprocated 1s on the 5-point peer liking scale). Teacher ratings on the Teacher Checklist (Pettit et al., 1997) provided a measure of children's social skill with peers. This checklist contains a set of items that assess teacher judgments of children's social skillfulness on 5-point scales (from very poor to very good) and includes items such as “understands others' feelings.” The seven-item composite was highly reliable (αs > .90 in all grades). The peer-relationship scores were correlated within and across years and were averaged across years (reverse-scored as needed) to form a psychometrically reliable index of middle childhood average peer competence (α = .73).

As with the middle childhood period, both peer and parenting experiences were examined in adolescence. At age 16, parenting was observationally assessed in a 45-min mother–teen interaction focused on parent–teen communication about rules, roles, and responsibilities and conflict resolution strategies. Using a modified version of the Melby and Conger (2001) coding protocol, coders rated individual and dyadic behavior on a set of 9-point rating scales. A superordinate 9-point rating summarized overall observed relationship quality. A low score indicated an unhappy, emotionally unsatisfying, or brittle relationship. A high score indicated the relationship was warm, open, and emotionally satisfying. Interrater agreement was computed for 20% of the total cases (average intraclass r = .70; p < .001).

Both positive and negative aspects of peer relations were assessed at age 16. A measure of friends' deviance was derived from participants' reports on the level of antisocial behavior within their peer group (e.g., “Get into fights with other kids,” “Like to do things that make you scared or uncomfortable”; Dishion, Patterson, Stoolmiller, & Skinner, 1991). Each behavior was rated on a 5-point scale ranging from never to very often, and values were then averaged (α = .74). An index of friendship support was derived from six items (e.g., “If you have a problem at school or at home you can talk to your friend about it”; Bukowski, Hoza, & Boivin, 1994). Items were rated on a 5-point scale from strongly disagree (1) to strongly agree (5) and were then averaged (α = .88).

Based on recommendations by Henry, Tolan, and Gorman-Smith (2005), we used both hierarchical and nonhierarchical clustering methods to determine if we could find evidence for Meier and Allen's (2009) romantic relationship patterns, using both the number of partners they had and the total number of waves they reported being in a relationship across the study period. In comparison to other clustering techniques (e.g., latent profile analysis), cluster analysis offers the advantage that it does not require that the clustering variables be independent (Zhang, 2004). Using the hierarchical agglomerative technique with the average linkage method, we determined how many clusters to expect and where to place the initial cluster centers. Using nonhierarchical clustering methods involving an iterative process, observations were assigned to the clusters using the predetermined number of clusters and the hierarchically determined cluster centers. We next assessed the cluster solution reliability by performing cluster analyses separately on each of the two cohorts of participants and comparing them with the clusters obtained from the full sample to evaluate the replicability of the assignments. Finally, with theoretically and empirically related variables not used to form the clusters, the discriminant validity of the cluster solution was examined with analysis of variance (ANOVA) and chi-square analyses. These analyses were undertaken to bolster confidence that meaningful clusters depicting patterns of romantic relationships in the transition from late adolescence to early adulthood can be empirically derived from continuous measures of (a) amount of relationship involvement (i.e., number of waves in a relationship) and (b) turnover across relationships (i.e., number of partners).

Two types of analyses were used to investigate how early interpersonal experiences may shape romantic outcomes. First, we examined whether the antecedent variables differentiated the clusters using multivariate analyses of variance (MANOVAs). This analysis identified whether different romantic profiles had unique developmental antecedents. Moreover, by entering more proximal antecedents subsequent to distal antecedents in a series of multivariate analyses of covariance (MANCOVAs), we could determine whether the impact of distal experiences on romantic outcomes was accounted for by more proximal ones. Second, we conducted a path analysis in MPlus Version 6.0 (Muthén & Muthén, 2007) to examine how positive and negative qualities of relationships with parents and peers across three periods (early childhood, middle childhood, and adolescence) were associated with subsequent amounts of romantic involvement in young adulthood and number of romantic partners during this time. In this way, we were able to trace overlapping and distinct predictors of romantic involvement and turnover simultaneously. Together, these two sets of analyses not only shed light on the unique developmental antecedents of different romantic experiences in young adulthood but also began to delineate a sequence of distal to proximal influences that might explain how the key features of romantic stability may develop.

Table 2 presents the means, standard deviations, and intercorrelations for all study variables. The clustering variables of number of waves in a relationship and number of partners were moderately correlated, with the former displaying more and generally larger correlations with the discriminant validity variables than did the latter. Number of waves in a relationship and number of partners were significantly correlated with several of the developmental antecedents, which themselves were correlated both within and across domain and developmental period. These and other correlations demonstrate that structural relationship indices cohere in meaningful ways and appear to have unique as well as overlapping antecedents.

Note. EC = early childhood indicator; MC = middle childhood indicator; A = adolescence indicator.

Using the standard criteria (Sarle's cubic clustering criterion, pseudo-F statistic, pseudo-T2 statistic) to determine whether the profiles in young adults' romantic relationships mapped onto those identified in adolescence by Meier and Allen (2009), a 5-cluster solution was determined to be the most appropriate1 (see Figure 1). ANOVAs showed the cluster solution significantly distinguished between the groups on the clustering variables (see Table 3).

Romantic relationship profiles.

Note. Means with different subscripts within a row are significantly different from one another (p ≤ .10) as tested with a Tukey's post hoc comparison.

Cluster 1 (n = 112; 21.9%) was characterized by significant involvement in romantic relationships with two partners over the study period. This cluster was named the steady involvement cluster based on their tendency to form and maintain long-lasting relationships and its alignment with those identified by Meier and Allen (2009) as progressing. Cluster 2, the largest group (n = 145; 28.4%), had the fewest partners and the least romantic involvement, with nearly one third (n = 48) not yet in a relationship. Cluster 2, labeled the later involvement cluster, appears to be a continuation of the group identified by Meier and Allen as having no romantic experience—this was also the largest group in their sample as well. Cluster 3, the smallest cluster (n = 64; 12.5%), had two partners and they were involved in romantic relationships for less than half of the waves. Cluster 3 did not differ from Cluster 1 in the number of partners but did differ in the duration and spacing of their relationships and were labeled the sporadic involvement cluster. Cluster 3 matches the group Meier and Allen suggested was progressing more slowly in their romantic development. Cluster 4 (n = 84; 16.4%) was characterized by the highest number of partners coupled with a high number of waves of romantic participation. This cluster, named the frequent involvement cluster based on participants' frequent and continuous involvement in short-term relationships, appears to combine two groups identified by Meier and Allen—those stably involved in multiple relationships and those who regressed to multiple relationships. Finally, Cluster 5 (n = 106; 20.7%) was characterized by having a single romantic relationship partner over the entire study period. Cluster 5 did not differ from Cluster 1 in the number of waves participants were involved in a romantic relationship, but Cluster 5 was notable in that this participation was limited to a single relationship. Based on the long-term commitment to a single partner, Cluster 5 was labeled the long-term committed cluster. This cluster closely parallels Meier and Allen's final group of individuals, who maintained a steady relationship across adolescence.

Following procedures outlined by Good, Willoughby, and Busseri (2011) to evaluate cluster reliability, cluster assignments from the full sample were cross-tabulated with the cluster assignments obtained with each cohort subsample to estimate the level of agreement. Averaging agreement between the full sample and each subsample, the kappa coefficient was .95, indicating that such romantic relationship patterns were replicable to those found in the larger population from which they were drawn. All five of the clusters found in the full sample were closely matched to clusters identified in the separate cohort samples, as indicated by the similarity in the magnitude and direction of the cluster centroids across samples.

Next, a series of ANOVAs and chi-square analyses confirmed the discriminant validity of the cluster solution (see Table 3). The long-term committeds had the longest relationships and were with their current partners the longest, followed by those with steady involvement, and then the other three clusters. Those with steady and frequent involvement reported first being in relationships during the study period, followed by those in the long-term committed and sporadic involvement clusters, with those in the later involvement cluster not reporting a relationship until nearly two to three years later. These individuals also reported having sex significantly later than did those in any other cluster. Although there were no differences in race, education, or income based on cluster membership, those in the steady involvement, frequent involvement, and long-term committed clusters were more likely to be female; those with later and sporadic involvement were mostly male. Long-term committeds were more likely to be married and to be parents in comparison to the other clusters. Not surprisingly, those with later involvement were the least likely to be married and have children.

Thus, the cluster solution found here generally replicates Meier and Allen's (2009) groupings, suggesting the variability found to characterize romantic development in adolescence may persist well into young adulthood. The findings presented here suggest that meaningful romantic-relationship groupings can be derived from continuous measures of involvement and turnover. We now turn to an examination of developmental factors (i.e., earlier interpersonal experiences) that may predispose individuals to be more (vs. less) involved in romantic relationships during young adulthood and to experience many (vs. few) romantic turnovers.

We examined whether the romantic relationship clusters had unique developmental antecedents using a series of MANOVAs and MANCOVAs. We first used MANOVAs to examine whether the antecedents within each developmental period differed based on cluster membership. Analyses revealed significant multivariate effects for cluster membership for each set of antecedents (see Table 4). We next used MANCOVAs to examine whether distal antecedents continued to differentiate the clusters once more proximal antecedents were included. To note, no significant gender by cluster interactions emerged for any of these analyses, and thus the parsimonious single-group analysis of 511 individuals is presented here.

Note. Means with different subscripts within a row are significantly different from one another (p ≤ .10).

Looking first at the early childhood antecedents (see Table 4), there were no differences in harsh discipline based on cluster membership. Analyses did reveal that those with steady involvement had the most proactive parents in early childhood, whereas those with sporadic romantic involvement had the least proactive parents. Once later antecedents were taken into account in a MANCOVA, these differences were no longer significant, suggesting the effects of earlier experiences on romantic relationships may operate through more proximal experiences.

Turning to middle childhood, post hoc analyses revealed that those with later or sporadic romantic involvement were more likely to have been physically disciplined, but there were no differences based on parental monitoring. The long-term committeds and those with frequent involvement were seen as more socially competent than were those with later or sporadic involvement. However, once we included the adolescent antecedents in a MANCOVA, only the cluster differences based on harsh discipline remained, suggesting that later experiences explained some of the effects of peer competence on romantic relationship development in young adulthood.

Finally, for the adolescent antecedents, post hoc analyses revealed marginally significant differences in relationship quality with mothers. Those with sporadic involvement were rated as having the lowest quality mother–teen relationships. For peers, the long-term committeds and those with later involvement had fewer deviant friends. Those with frequent involvement had the most supportive friends, followed by the long-term committeds, those with steady involvement, and then those with sporadic and later involvement. Although the effects sizes are modest, these analyses begin to paint a picture of developmental influences on romantic relationships extending back to early childhood. To clarify if these antecedents represent coherent pathways through which romantic involvement and turnover develop, we next turn to the path analysis.

A path analysis that included all antecedent variables (two early childhood, three middle childhood, three adolescence) was conducted with both outcomes (involvement and turnover) considered simultaneously to account for their linked nature (see Figure 2). The model provided an excellent fit to the data, χ2(16) = 20.58, p = .20; comparative fit index (CFI) = .99; root-mean-square error of approximation (RMSEA) = .02; standardized root-mean-square residual (SRMR) = .03, explaining 6% of the variance in the number of waves young adults had been in relationships and 9% of the variance in how many partners they had during this time. To test for gender differences, we compared two path models—one with free parameters and one with parameters fixed across groups (Muthén & Muthén, 2007). A chi-square difference test revealed no difference between these models, indicating the path model fit men and women equally well.

Model linking romantic relationship experiences in young adulthood to the developmental antecedents. Path coefficients are standardized. Significant pathways are presented, while nonsignificant pathways and cross-sectional correlations are not shown, to improve the clarity of presentation. χ2(16) = 20.58, p = .20; comparative fit index = .99; root-mean-square error of approximation = .02; standardized root-mean-square residual = .03; Rel. = relationship. *p ≤ .05. **p ≤ .01.

We found evidence for both common and distinct antecedents of young adult romantic experiences. Number of waves in a relationship was predicted by high-quality interactions with mothers and more friendship support in adolescence, whereas number of partners was predicted by more friendship support and also by having more deviant friends. In line with the prior analyses, the earlier antecedents (early and middle childhood) operated through the more proximal adolescent variables to influence young adult romantic relationships. For example, harsh discipline in early childhood forecasted lower peer competence and parental monitoring and more harsh discipline in middle childhood. Peer competence and parental monitoring then predicted more friendship support and mother–teen relationship quality, as well as lower levels of friendship deviance, in adolescence. These three adolescent variables were associated in both distinct and overlapping ways with the two romantic relationship dimensions. Similar pathways were found for physical discipline and parental monitoring in middle childhood. In contrast to harsh discipline, early proactive parenting was associated with later variables only through parental monitoring in middle childhood. These findings suggest that, whereas earlier experiences with parents and peers individually (i.e., at the bivariate level) forecast individual differences in key dimensions of romantic relationship experiences in young adulthood, their developmental “effects” are indirect, operating through later, developmentally more proximal experiences.

Although there has been growing concern that the institution of marriage is becoming obsolete in younger generations, finding a long-term partner remains an important goal for most young adults (Lehnart et al., 2010; Scott, Schelar, Manlove, & Cui, 2009), with many of them considering this a critical marker of adult status (Lanz & Tagliabue, 2007). Despite these beliefs in the importance of attaining a committed relationship, the current study suggests that the paths young adults take to achieving this goal vary greatly. Closely aligning with Meier and Allen's (2009) work on adolescent relationships, we identified five unique young adult romantic relationship profiles. Our efforts at understanding romantic experiences in young adulthood shed light on the normative and nonnormative patterns of romantic relationship involvement in this sample and begin to offer a perspective on why they may develop in this manner.

The challenge of determining what constitutes normative romantic development in young adulthood is recent. In previous decades, this development was both clear and uniform, with young adults launched from their family of origin into a committed relationship that represented their entry into adulthood (Duvall, 1962). Whereas marriage used to represent a first step into adulthood, dramatic changes in the sequencing and timing of interpersonal partnerships and parenthood over the last few decades of American family life coupled, with the rising focus on self-development, have pushed the marriage milestone further into adulthood (Cherlin, 2009). Though young adults still value marriage (Scott et al., 2009), most no longer appear to be willing to commit to a relationship before they fully explore their options. This has not only given rise to unprecedented variability in young adult romantic experiences, a trend clearly illustrated by the five unique romantic profiles found here, but it suggests that the definition of normative needs to be revised.

Perhaps most illustrative of the sociodemographic shifts experienced in today's society was that the most prevalent romantic relationship profile in this sample was the later involvement cluster—young adults who postponed the pursuit of a serious relationship well into their midtwenties, if not later for some (9.4% of the sample had yet to report a romantic relationship of at least 3 months in duration by age 25). This pattern's prevalence, which corresponds with its prevalence in Meier and Allen's (2009) study of adolescent relationships, dovetails nicely with Cherlin's (2009) assertion that a committed relationship now often represents the last step into adulthood rather than the first. Although such individuals appear to be fairly normative both in the current sample and according to Cherlin, their delayed entry into a committed relationship is nonnormative according to previous studies proposing these individuals should have achieved this milestone years earlier (Collins, 2003; Seiffge-Krenke, 2003). This later involvement, however, may reflect a focus on achieving other developmental tasks before this one. For example, although women who pursue higher education are more likely to delay marriage, they are also more likely to eventually marry (Goldstein & Kenney, 2001). This delay may simply be that and not reflective of an overall reduced capacity for intimacy. However, recent work has suggested that individuals who delay romantic activity in young adulthood are less likely to achieve success in other key domains (Lehnart et al., 2010; Seiffge-Krenke, 2010).

Possibly shedding light on whether this lack of serious romantic relationship involvement is problematic are the developmental antecedents of noninvolvement. Romantic activity delays may signify the larger problem of a difficulty relating to other individuals, as less involvement in romantic relationships, which characterized those in both the later and sporadic involvement clusters, was related to both lower observed mother–child relationship quality and less friendship support in adolescence. Our analyses suggest these represent persistent patterns of difficulty in the family and peer domains that go back to early childhood. This history may have made these individuals feel less confident to establish healthy, outside relationships (Englund, Kuo, Puig, & Collins, 2011). It may also be the case that these individuals had personality or social characteristics that made it difficult for them to establish and maintain healthy romantic relationships. Thus, although young adults in the sporadic involvement cluster initiated romantic relationships earlier than did those with later involvement and thus could be considered more successful by some standards (Seiffge-Krenke, 2010), they had difficulty maintaining this involvement across young adulthood. It may be, then, that it is not the delay in romantic relationship activity itself that is problematic but rather that individuals who had difficulty in connecting competently with others earlier in life continue to have these difficulties in the romantic domain as adults, especially if these problems with parents and peers persist across multiple developmental periods.

The rather abbreviated romantic relationship history of those in the lower involvement clusters may also owe to the disproportionate number of males in these clusters. A national study revealed that young men are not only less likely to be in a romantic relationship in comparison to young women, but they are also less likely to consider lifelong commitment as a critical component of a successful relationship (Scott et al., 2009). Thus, it is less surprising that young men may not be as interested in pursuing a long-term relationship (Wood et al., 2008). Though these delays in romantic involvement may be expected, Seiffge-Krenke (2010) found men who stayed single from ages 20 to 28 experienced substantial declines in their self-esteem. Interestingly, women who remained single had relatively stable self-esteem. This suggests that although young men profess to not be as interested in serious, committed relationships as are women and behave accordingly, spending a significant portion of one's young adulthood single may have consequences down the road. Even for men, romantic relationships represent an important and vital connection to others.

It thus appears that the timing of romantic relationship activity cannot be the only measure by which we evaluate romantic success. Delays in commitment for those who established healthy connections with others early in life may be of an entirely different nature, as indicated by those with frequent and steady involvement. These young adults had not yet progressed to the serious commitment seen in the long-term committeds but they did have a high level of involvement in romantic relationships, which may reflect a history of high-quality relationships with family and peers. Interestingly, despite more consistent participation in romantic relationships, these individuals switched partners across the study period as much as did those with sporadic involvement and far more often than did those with later involvement or the long-term committeds. Though romantic turnover can be traumatic (Simpson, 1987), these dissolutions did not appear to deter these young adults from the pursuit of intimacy. Looking at the antecedents, it may be that these individuals are more embedded in their peer networks, not only giving them access to a wider range of potential romantic partners (Furman, 1999) but also perhaps reflecting a tendency to gravitate toward relationships of all types (Cavanagh, 2007). The fact that these seemingly less stable relationships are likely to become more common in the future (Wood et al., 2008) does not necessarily warrant heightened concern, because a lack of early commitment may be problematic only when it represents a full abstention from romantic relationship activity and other key developmental tasks. This echoes Seiffge-Krenke's (2003) conclusion that greater involvement in early romantic relationships, regardless of how many partners it involves, should lead to eventual positive romantic outcomes, as it provides individuals with valuable learning experiences they can apply to later romantic relationships.

If delaying serious romantic involvement now appears to represent the normative course of development for young adults, what about those who adhere more closely to the theoretically expected progression to a single, committed relationship (Seiffge-Krenke, 2003)? Representing approximately one fifth of the analytic sample, the long-term committeds appeared to buck the trend of postponing key markers of adult status, as evidenced by their earlier entry into marriage and parenthood compared to other young adults in this sample and to national trends (Mathews & Hamilton, 2009). What contributed to this earlier entry into adulthood? Strong relationships with family and peers earlier in life coupled with less deviant friends may have fostered earlier commitment. Attachment theory suggests that the high-quality parent– child and peer relationships they experienced earlier in life likely enabled them to not only establish but also maintain healthy, committed romantic relationships as young adults (Hazan & Shaver, 1987). Thus, what differentiates these young adults from those with later and sporadic involvement is that although none may be very comfortable interacting with multiple romantic partners, the long-term committeds have developed the social skills early on that are necessary to sustain a romantic relationship once they do initiate romantic activity with their chosen partner. It is also possible that personality differences in social anxiety and preference for stability contributed to the divergent paths of the two groups, a question addressable in future studies.

Perhaps the most interesting contrast to emerge is between the long-term committeds and those with frequent involvement. Demographically, both clusters were indistinguishable, being nearly equally female, white, educated, and well off. Where these clusters diverge is in stability, with the long-term committeds seemingly emerging from this comparison as more successful. On the one hand, the long-term committeds did achieve a critical developmental task earlier (Conger et al., 2000). On the other hand, both clusters had consistent romantic involvement, which bodes well for their romantic futures (Seiffge-Krenke, 2003). These young adults were equally socially skilled with their peers, indicating the lack of commitment of the frequent involvement cluster does not appear to reflect a lack of ability due to social experiences. It may reflect, as previously mentioned, differences at the level of the individual, rather than in the early social environment, whereby different young adults have varying levels of interest in commitment at this stage of development. For example, personality differences may engender different types of romantic involvement and comfort levels with commitment and dependency (Lehnart & Neyer, 2006). Cherlin (2009) might suggest their multiple partnerships in fact reflect how seriously these young adults take commitment, as they are unwilling to settle for anything less than the right partner to ensure that once they do choose, it will last. This comparison perhaps best reveals the importance of exercising caution when proposing theories of normative romantic development and promoting relationship education programs in line with these theories. Though settling down is an important goal for most young adults (Lehnart et al., 2010; Scott et al., 2009), recent sociohistorical shifts may have permanently altered the way in which this goal is achieved (Cherlin, 2009). The current study suggests diversity and delays may be more common now and variations from the progression to a single, committed relationship may not be problematic, let alone worthy of intervention to promote such a progression at this stage of life.

Our confidence in these results is enhanced by a number of strengths in our study design. First, our use of a prospective, longitudinal study allowed us to examine stability and change in romantic relationships across young adulthood, when the pursuit of intimacy is the critical task. Second, combining person-centered and variable-centered approaches allowed us to understand both complex configurations at the level of the individual and the pathways that may have shaped the underlying dimensions. Further, our replication of the cluster analyses using latent profile analyses both underscored the robustness of our cluster solution and is an example of the internal replication recently called for in the field (Duncan, Engle, Claessens, & Dowsett, 2012). Finally, we capitalized on a rich array of antecedents that spanned multiple developmental periods (early and middle childhood, adolescence), surveyed multiple contexts (personality, parents, peers), and drew from multiple informants (participants, parents, teachers, peers, observers), allowing us to identify how both distal and proximal experiences in multiple domains of interest were related to later romantic development.

Despite these strengths, several limitations suggest these findings should be interpreted with caution. Given that cluster analysis is sample-dependent, findings need to be replicated beyond our community sample of predominantly European Americans and African Americans in heterosexual relationships to more diverse samples, as research has found class-graded special populations (e.g., sexual minorities) face unique circumstances that shape romantic development (Meier & Allen, 2008). Second, although our definition of a relationship as being one of at least 3 months' duration is consistent with Meier and Allen (2009), only assessing current status may have masked even greater diversity for those whose relationships began and ended between waves or for those who were dating multiple partners simultaneously (Davies & Windle, 2000). Further, though our study provided rich structural information about romantic involvement, it was limited by the lack of information about the quality of the romantic partners or of the relationships. It is likely that whom one is partnered with affects the seriousness of that relationship and how likely an individual is to pursue another relationship. Perhaps the long-term committeds enjoyed stability as a result of having higher quality partners. Given research suggesting that the characteristics of early romantic partners have serious implications for later development (Giordano, Phelps, Manning, & Longmore, 2008), future research should explore how characteristics of early dating partners and relationships shape later romantic relationship development. Finally, although we had a wealth of measures for our developmental antecedents and used well-established measures whenever possible, some of the internal consistencies for our measures were rather low.

In conclusion, research on romantic relationship development has historically assumed there is a linear progression from short-term, shallow relationships to a single, committed relationship in early adulthood. However, that progression may have been profoundly altered by recent sociohistorical shifts. As seen in the present study of a sample of young adults coming of age in the 21st century, there is great diversity in romantic relationship experiences in emerging adulthood, with the roots of this diversity traceable to family and peer relationships earlier in life.

The Child Development Project has been funded by Grants MH42498, MH56961, MH57024, and MH57095 from the National Institute of Mental Health; HD30572 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development; and DA016903 from the National Institute on Drug Abuse. We are grateful to Anna Swanson and the parents, children, and teachers who participated in this research.

1Supplemental latent profile analyses confirmed that a five-class solution, comparable to the five-cluster solution reported here, provided a significantly better model fit than did models with different numbers of classes. For example, a chi-square difference test comparing the four-class solution to a five-class solution revealed the former to be a significantly poorer fit, Δχ2(3) = 35.876, p < .001. Further, the five-class solution greatly overlapped with the five-cluster solution reported in the article, enhancing our confidence in our original cluster solution.

Amy J. Rauer, Auburn University.

Gregory S. Pettit, Auburn University.

Jennifer E. Lansford, Duke University.

John E. Bates, Indiana University.

Kenneth A. Dodge, Duke University.