Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2026 Jan 21;65(1):e70116. doi: 10.1111/famp.70116

Autonomy and Relatedness in Mother‐Adolescent Interactions: An Investigation Using Exploratory Graph Analysis

Natasha A Bailey 1,✉, Hudson F Golino 1, Joseph P Allen 1
PMCID: PMC12824057  PMID: 41566608

ABSTRACT

Mother‐adolescent interactions are important contexts for teens to develop essential autonomy and relatedness skills. The Autonomy and Relatedness Coding System was designed to measure these behaviors and is based on four a priori theoretical categories, including behaviors promoting autonomy, behaviors undermining autonomy, behaviors promoting relatedness, and behaviors undermining relatedness. The current study used Exploratory Graph Analysis (EGA) to examine the underlying dimensional structure of autonomy and relatedness behaviors in mother‐adolescent interactions and compare this structure to the theoretical categories. Participants were 184 mother‐adolescent dyads participating in a larger longitudinal study of adolescent social development. Mothers and adolescents (M age = 13.35, SD = 0.64) discussed an area of disagreement. These interactions were coded for nine different autonomy and relatedness behaviors displayed by mothers and adolescents. EGA results revealed a three‐dimensional structure for both adolescents' behaviors toward mothers and mothers' behaviors toward adolescents. These three‐dimensional models fit the data significantly better than the theoretical four‐dimensional model. Bootstrap EGA results further replicated the three‐dimensional structure. These findings suggest that EGA is a useful tool for examining the dimensional structure of autonomy and relatedness behaviors in mother‐adolescent interactions and provide more nuanced insights into the developmental differences of these behaviors in mothers versus teens.

Keywords: autonomy and relatedness, exploratory graph analysis, mother‐adolescent interactions

1. Introduction

A key developmental task of adolescence is establishing a sense of autonomy while maintaining a sense of relatedness, or a positive relationship with close others (Allen, Hauser, Bell, and O'Connor 1994; Steinberg 1990). The success with which this task is handled has been linked to a variety of outcomes, including lower levels of internalizing and externalizing problems, higher‐quality peer and romantic relationships, and better physical health (Allen, Hauser, Bell, and O'Connor 1994; Cook et al. 2015; Oudekerk et al. 2015). Conceptualizations of the structure of autonomy and relatedness in parent‐adolescent relationships have been theoretical to date (Allen, Hauser, Bell, and O'Connor 1994; Allen, Hauser, Eickholt, et al. 1994; Cook et al. 2015; Inguglia et al. 2015). Thus, a statistical approach with the ability to identify the dimensional structure of autonomy and relatedness in parent‐adolescent interactions would prove useful in deepening our understanding of these behaviors. Exploratory Graph Analysis (EGA; Golino and Epskamp 2017) is a statistical method that uses psychometric network models to identify underlying dimensions within a dataset and to visualize the relationships between variables as a network. Unlike latent variable models, which assume that observed variables are manifestations of underlying constructs and rely on local independence, EGA models conceptualize psychological constructs as systems of interacting components. Dimensions, or “communities,” emerge from these patterns of interconnectedness rather than from an underlying latent cause (Christensen and Golino 2021a; Christensen et al. 2025). This technique is exploratory in nature and has the potential to help researchers uncover unexpected structures that may or may not align with existing theories, which makes it a well‐suited method for evaluating the empirical organization of psychological constructs (Christensen et al. 2019). Further, EGA has been shown to outperform factor analysis and other latent variable models in accurately determining dimensions (Christensen et al. 2024; Christensen and Golino 2021b; Golino et al. 2020; Golino and Demetriou 2017; Golino and Epskamp 2017; Jamison et al. 2021). Given the advantages of this method, the current study used EGA to examine the dimensional structure of autonomy and relatedness behaviors in parent‐adolescent interactions and compared that structure to a theoretical conceptualization of autonomy and relatedness.

1.1. Theoretical Structure of Autonomy and Relatedness

Mother‐adolescent interactions play a pivotal role in adolescents' ability to develop a sense of autonomy. Interactions that occur within the context of a disagreement or difference of opinions are particularly important settings for adolescents to learn autonomy‐relatedness skills (Branje 2018; Phinney et al. 2005). Mothers who engage in behaviors that promote autonomy, such as encouraging discussions that allow adolescents to state their opinions, as well as modeling these behaviors themselves, can help adolescents engage in self‐exploration (Grotevant and Cooper 2013). However, maternal behaviors that inhibit the autonomy of their adolescent by undercutting their statements or avoiding the opportunity to discuss their reasons behind their differing opinion can send the message to the adolescent that their opinions are incorrect or unimportant (Allen, Hauser, Eickholt, et al. 1994). Mothers also play a role in modeling behaviors that help adolescents understand how to maintain positive relationships with others, even in the context of a disagreement, such as by being supportive and validating of their adolescent (Allen, Hauser, Bell, and O'Connor 1994). In contrast, behaviors that convey hostility are likely to contribute to negativity in the relationship and increase the likelihood that adolescents will display these behaviors in future interactions with their mother and close others (Allen et al. 2002; Soenens and Vansteenkiste 2020).

The Autonomy and Relatedness (AR) Coding System was developed to examine autonomy and relatedness behaviors in mother‐adolescent relationships (Allen, Hauser, Bell, and O'Connor 1994; Allen, Hauser, Eickholt, et al. 1994; Allen et al. 2000). This system has been translated into four languages, cited more than 1300 times, and used empirically in a range of samples internationally to predict a variety of outcomes (Allen et al. 2012; Becker‐Stoll et al. 2008; Samuolis et al. 2005; Zhang and Slesnick 2017). Four a priori theoretical clusters were established to describe the extent to which both members of the mother‐adolescent dyad engage in autonomy and relatedness behaviors: (a) behaviors exhibiting autonomy (Positive Autonomy); (b) behaviors undermining autonomy (Negative Autonomy); (c) behaviors exhibiting relatedness (Positive Relatedness); and (d) behaviors undermining relatedness (Negative Relatedness). According to the coding system, positive autonomy behaviors are thought to involve making clear and confident statements related to one's thoughts and opinions, while negative autonomy behaviors are those that attempt to reduce the validity of or change the other's thoughts and opinions. Behaviors thought to promote relatedness are those that express interest and engagement in the interaction. Finally, negative relatedness behaviors are those that reduce the positive relationship between the dyad, such as by making hostile statements or distracting/ignoring the other person.

While the a priori categories make theoretical sense and have been shown to predict a range of developmental outcomes (Allen et al. 2002; Niolon et al. 2015; Shah et al. 2023), the validity of these categories has actually never been empirically assessed, and cogent arguments can be made for other organizations of these behaviors. For example, hostility within the dyad could inhibit the other's autonomy by reducing opportunities for them to voice their opinions. Additionally, there may be developmental differences in autonomy and relatedness behaviors between mothers and adolescents. In general, the development of autonomy skills has been shown to increase across childhood and adolescence, suggesting that adolescents may struggle to utilize these skills to the same effectiveness as mothers (Wray‐Lake et al. 2010). Autonomy and relatedness behaviors may also be less integrated in younger teenagers (Inguglia et al. 2015). Therefore, certain autonomy and relatedness behaviors may manifest differently depending on whether they are being displayed by mothers versus adolescents, and behaviors that are autonomy‐supporting when exhibited by mothers may be autonomy‐inhibiting when displayed by adolescents or vice versa. For example, question‐asking in one context could come across as showing positive interest in the conversation, or it could convey negative or repeated requests for information that undermine the autonomy of the other. As a result, examining both mothers' and adolescents' behaviors during disagreements can yield greater insight into developmental differences in autonomy and relatedness behaviors. A statistical approach that allows for a formal examination of the theory behind the AR Coding System has promise in allowing deeper exploration of the underlying organization of these constructs.

1.2. Exploratory Graph Analysis (EGA)

Network psychometrics is a growing field that defines constructs as complex systems that arise from mutually reinforcing interactions between variables, which makes it an ideal approach for examining the multifaceted nature of human interactions. EGA, a technique within network psychometrics, estimates a graphical model and then applies cluster detection to estimate the number of dimensions in a dataset, as well as determine which items best fit within each dimension (Golino and Epskamp 2017). Network models are visualized as a web or graph, where each item or variable represents a node and the lines (or edges) that connect them represent the strength of the connection between them. These edges are created using partial correlations to isolate the unique association between two variables while accounting for the influence of all other variables in the network. Researchers can then apply regularization or other techniques to the partial correlation coefficients to create conditional independence and reduce the likelihood of spurious correlations. While regularization techniques are most commonly used in network psychometrics, other non‐regularization techniques are also available (see Isvoranu and Epskamp 2023 for a full review). The resulting network structure consists of a certain number of dimensions, where items from the same dimension form groups of densely connected nodes (Golino and Epskamp 2017). This clustering approach allows for the visualization of potential latent dimensions in the dataset, thereby yielding a more detailed understanding of the relationships between variables. EGA differs from traditional factor analysis in that it allows for a nuanced examination of individual relationships between variables instead of just grouping variables into broader, latent categories (Golino 2024) and has also been found to present higher accuracy in detecting communities in the network structure compared to parallel analysis and factor analysis, even when the data come from a factor model (Christensen et al. 2024; Christensen and Golino 2021b; Golino et al. 2020; Golino and Demetriou 2017; Golino and Epskamp 2017; Jamison et al. 2021).

1.3. The Current Study

The potential insights into specific pathways of influence within a network and underlying structures in the data that can be gained through EGA make this approach a useful tool for investigating the constructs within the AR Coding System. Additionally, due to the emergent structure of networks from the relationships among the observed data, EGA can explore whether these structures align with the existing theory underlying the AR Coding System or whether an alternative organization of autonomy‐relatedness behaviors might be more appropriate. Given that the current study was exploratory, we did not have specific hypotheses. Generally, we expected that EGA would replicate the theoretical structure identified by the AR Coding System, with communities representing positive autonomy behaviors, negative autonomy behaviors, positive relatedness behaviors, and negative relatedness behaviors. We also expected that there may be variation in the organization of these behaviors onto communities depending on whether they are exhibited by adolescents versus mothers.

2. Method

2.1. Participants

Data for this investigation were drawn from a larger study of adolescent social development in familial and peer contexts. Participants included 184 adolescents (85 male and 99 female) assessed at age 13 (M = 13.35, SD = 0.64) and their mothers. The sample was racially/ethnically and socioeconomically diverse: 107 (58%) identified as White, 53 (29%) as African American, 15 (8%) as of mixed race/ethnicity, and 9 (5%) as being from other minority groups. Adolescents' parents reported a median family income in the $40,000 – $59,999 range.

The current study received approval from the Institutional Review Board for Social and Behavioral Research at the University of Virginia. Adolescents were recruited from the seventh and eighth grades of a public middle school drawing from suburban and urban populations in the Southeastern United States. Students were recruited via an initial mailing to all parents of students in the school along with follow‐up contact efforts at school lunches. Families of adolescents who indicated they were interested in the study were contacted by telephone.

2.2. Procedure

Mother‐adolescent interactions took place in private offices within a university academic building. Confidentiality was assured to all study participants in the study session. All adolescents provided assent before the interview session, and parents provided informed consent for adolescents. If necessary, participants were provided with transportation and childcare. Adolescent participants and their parents were paid for their participation.

2.3. Measures

2.3.1. Observed Autonomy and Relatedness

Adolescents and their mother participated in a revealed differences task in which they discussed a family issue that they had separately identified as an area of disagreement. Discussions began with the adolescent playing a previously recorded audiotape in which they stated the problem, their perspective on it, and what the adolescent thought their mother's perspective was on the topic. Typical topics ranged from money, grades, household rules, friends, and sibling issues. Interactions lasted 8 min and were videotaped and then transcribed.

The Autonomy and Relatedness (AR) Coding System examines behavior that promotes or undermines autonomy and relatedness in a family interaction task. Adolescents and mothers are rated on their overall behavior toward the other in the interaction across a variety of categories. These ratings are molar in nature and represent behavior across the entire interaction. Each item is scored on scales of 0–4 with half‐point intervals and behavioral anchors of each full point for a given code. The AR Coding System takes into consideration both the frequency and intensity of the behavior over the duration of the interaction. For example, a single but particularly hostile remark might substantially increase the score for the hostility scale, while a digression where the mother and adolescent discuss plans for later in the day might not influence any scores. Interactions were coded by psychology graduate students who were blind to other participant data. Students were trained and then supervised as a group until they were able to code reliably (i.e., scores were within one point of each other on 85% of comparisons). Coders then coded independently, and reliabilities were assessed randomly thereafter. Each interaction was coded by two trained coders and the intraclass correlation between the two sets of codes was calculated for each behavior. ICCs from 0.50 to 0.59 were considered to be in the fair range, those between 0.60 and 0.74 in the good range, and 0.75 or higher in the excellent range (Cicchetti and Sparrow 1981).

In the original coding system, ten categories of behavior were summed into four scales as described below, with intraclass correlation coefficients (ICCs) listed after each: (1) promoting autonomy, which combined subscales of “reasons” (clearly stating reasons for disagreeing; ICC = 0.78 for adolescents' behavior and 0.89 for mothers' behavior) and “confidence” (confidence when stating one's thoughts and opinions; ICC = 0.66 for adolescents' behavior and 0.86 for mothers' behavior); (2) undermining autonomy, which combined subscales of “recanting/placating” (saying a statement one does not mean in order to placate the other person or de‐escalate the argument; ICC = 0.31 for adolescents' behavior and 0.69 for mothers' behavior), “overpersonalizing/blurring the boundary between the person and their position” (enlisting an outside person's opinion or behavior, characterizing the other person, attacking the speaker; ICC = 0.70 for adolescents' behavior and 0.82 for mothers' behavior), and “pressuring” (pushing the other person to take one's side or to agree; ICC = 0.78 for adolescents' behavior and 0.84 for mothers' behavior); (3) promoting relatedness, which consisted of subscales of “querying” (asking the other person questions to seek information; ICC = 0.78 for adolescents' behavior and 0.77 for mothers' behavior), “validating” (agreeing/positively reacting to the other person; ICC = 0.80 for adolescents' behavior and 0.64 for mothers' behavior), and “engagement” (being involved in the interaction; ICC = 0.81 for adolescents' behavior and 0.79 for mothers' behavior); and (4) undermining relatedness, which comprised subscales of “distracting” (making comments to bring the other's attention away from their statements or ignoring/cutting off the other person; ICC = 0.61 for adolescents' behavior and 0.82 for mothers' behavior) and “hostility” (making negative or devaluing statements toward the other; ICC = 0.52 for adolescents' behavior and 0.80 for mothers' behavior). Given the low ICC values for recanting/placating, as well as very low base rates of this behavior for both adolescents and mothers, this variable was dropped from further analyses, resulting in a total of nine variables.

2.4. Data Analysis

To investigate the number of dimensions in the AR discussion task, we applied EGA using the EGAnet package (Version 2.1.1, Golino et al. 2025) in R (Version 2024.12.0, R Core Team 2024). We estimated two EGA models; one representing adolescents' behavior toward their mothers in the AR task and the other representing mothers' behavior toward their adolescents in the AR task. In the present study, we opted to use the EBIC graphical least absolute shrinkage and selection operator (GLASSO) to estimate a Gaussian graphical model (for more details, see: Epskamp and Fried 2018). This commonly used regularization technique shrinks parameter estimates to zero to achieve a sparse network model where non‐relevant edges are removed. The network with the lowest EBIC is selected. We elected to use EBICglasso to estimate a regularized network due to clear evidence of the strength of this method for dimensionality assessment (Christensen et al. 2024; Golino and Epskamp 2017; Golino et al. 2020). This was followed by the use of a community detection algorithm to identify the dimensions of the network (Golino et al. 2020; Golino et al. 2021). The EGAnet package has several community detection methods: Louvain (Blondel et al. 2008), Leiden (Traag et al. 2019), and Walktrap (Pons and Latapy 2005). We constructed EGA models using each algorithm. The Louvain algorithm was selected for the main analyses due to evidence that it is among the most effective approaches in identifying network community structures when paired with GLASSO (Christensen et al. 2023; Gates et al. 2016). A consensus clustering approach was used in the Louvain algorithm introduced by Lancichinetti and Fortunato (2012), as is the default option in the EGAnet package. The consensus method used is called the “most common” method, which selects the community solution that appears the most frequently across the applications.

We then used bootstrap EGA to estimate the stability of dimensions identified in our EGA models. This approach also provides a more detailed examination of the dimensionality of items and is particularly important in the context of small samples, as was the case in the current study (Christensen and Golino 2019). Bootstrap EGA provides a measure of item stability for each item, which represents the proportion of times that an item is replicated in each dimension across bootstrap samples. Proportions greater than at least 0.70 suggest the item is stable and consistently identified in its confirmatory dimension, while items with stability values less than 0.70 are considered to be multidimensional and may be problematic for accurately estimating dimensions (Christensen and Golino 2021b). Bootstrap EGA also provides a measure of structural consistency, or the proportion of times each dimension is replicated in the bootstrap samples. Applying the Louvain algorithm to the bootstrap EGA analyses resulted in relatively higher item stability values compared to the Walktrap algorithm.

Finally, we compared the EGA models to the theoretical four‐factor structure using EGA's fit index, the Total Entropy Fit Index (TEFI: Golino et al. 2021). The TEFI assesses the degree of uncertainty in a set of variables and is calculated by taking the average Von Neumann entropy of the factors and subtracting the total entropy of the matrix. Lower TEFI values indicate lower uncertainty of a system of variables, suggesting a greater likelihood that a given structure represents the most optimal partition of multidimensional space (Golino et al. 2021). Nonparametric bootstrap hypothesis tests were used to compare TEFI values for the empirical and theoretical models. In this test, TEFI values for two different structures using bootstrapped correlation matrices from Bootstrap EGA are obtained and compared using a nonparametric bootstrap test. The null hypothesis is that the TEFI values obtained in the bootstrapped correlation matrices for the empirical EGA result are greater than the TEFI values obtained in the bootstrapped correlation matrices for the theoretical structure. Therefore, the p‐value in this bootstrap test can be interpreted as follows: If the p‐value is less than 0.05, TEFI values for the empirical structure tend to be lower than the theoretical structure, indicating that the former provides a better fit (lower entropy) than the latter. If the p‐value is greater than 0.05, the TEFI values for the empirical structure are not significantly lower than the theoretical structure, suggesting that both structures may provide similar fits or that the comparison might fit better. For more details on nonparametric bootstrap tests, see Chihara and Hesterberg (2022).

3. Results

Descriptive statistics for the autonomy and relatedness behaviors can be found in Table 1.

TABLE 1.

Means and standard deviations of the nine autonomy and relatedness behaviors.

Adolescent behaviors Maternal behaviors
M SD M SD
Reasons 1.83 0.76 2.73 0.75
Confidence 2.46 0.92 3.36 0.54
Overpersonalizing/blurring 0.60 0.74 0.93 0.80
Pressuring 0.66 0.74 1.59 0.81
Querying 1.01 0.79 2.02 0.83
Validating 0.87 0.60 1.26 0.73
Engagement 2.14 0.60 2.53 0.69
Distracting 1.02 0.78 0.72 0.57
Hostility 0.21 0.48 0.28 0.46
Open in a new tab

3.1. Adolescent‐To‐Mother Autonomy and Relatedness

EGA was conducted using the Louvain community detection algorithm 1 to examine the network structure of adolescents' behaviors toward their mother during the AR task. The EGA analyses revealed three dimensions (Figure 1). “Reasons” and “confidence” represented one dimension (Positive Autonomy), “pressures,” “queries,” “blurs,” “distracting,” and “hostile,” comprised the second dimension (Negative Autonomy and Relatedness), and “engaged” and “validates” made up the third dimension (Positive Relatedness). Thus, the first dimension replicates the theoretical category of behaviors thought to represent promoting autonomy; the second appears to be a combination of behaviors from the undermining autonomy and undermining relatedness dimensions in addition to “Queries,” and the third likely reflects the positive relatedness dimension.

FIGURE 1.

FIGURE 1

Open in a new tab

Adolescent‐to‐mother and mother‐to‐adolescent EGA models. Green lines represent positive partial correlations, red lines represent negative partial correlations. Line thickness indicates the strength of the partial correlation. Individual partial correlation values are not displayed due to evidence that individual edge weights are not reliable indices (Huth et al. 2025).

To examine the stability of the dimensions identified in the first EGA model, we conducted a parametric bootstrap EGA using the Louvain algorithm 2 with 500 bootstrap samples. Results revealed that the item stability for all items was 0.86 or higher (Figure 2). Structural consistency values for the three dimensions were also good: 1.00 for the Positive Autonomy dimension, 0.83 for the Negative Autonomy and Relatedness dimension, and 0.86 for the Positive Relatedness dimension, indicating a relatively high replication of these communities across bootstrap samples. Structural consistency is the extent to which a dimension is interrelated (internal consistency) and homogeneous (test homogeneity) in the presence of other related dimensions (Christensen and Golino 2021b). Item stability, on the other hand, quantifies the robustness of each item's placement within each empirically derived dimension (Christensen and Golino 2021b). Items and dimensions should have stability indices equal to or greater than 0.70 to be considered stable.

FIGURE 2.

FIGURE 2

Open in a new tab

Bootstrap EGA results for the adolescent‐to‐mother model. Values represent the proportion of times each item was assigned to its specific empirical community in the bootstrap samples. Community 1 = Positive Autonomy, Community 2 = Negative Autonomy and Relatedness, Community 3 = Positive Relatedness.

According to the theoretical AR Coding System categories, “Queries” falls within the Positive Relatedness dimension. To explore the change of this behavior to the Negative Autonomy and Relatedness dimension in the EGA model, we examined the dimensional stability of this item. “Queries” was categorized into the Negative Autonomy and Relatedness dimension in 85.8% of the adolescent‐to‐mother bootstrap models, the Positive Autonomy dimension in 3.4% of the bootstrap models, and the Positive Relatedness dimension in 10.8% of the bootstrap models. This instability was also reflected in a low network loading for “Queries” of 0.14, which is equivalent to a factor loading of 0.34 (Christensen et al. 2025). All other items had moderate or high network loadings (0.32–0.64) within their respective dimensions except for “Queries.”

Finally, we compared the fit of the three‐dimensional EGA model with the four theoretical categories. The TEFI of the EGA model was −4.09 versus a TEFI of −2.01 for the theoretical model, indicating that the empirical EGA result fits the data better than the theoretical structure. A nonparametric bootstrap hypothesis test within the EGAnet package revealed that the difference between these two TEFI values was statistically significant (p = 0.002), further indicating that the EGA model was a better fit to the data than the theoretical model (Figure 3).

FIGURE 3.

FIGURE 3

Open in a new tab

Nonparametric hypothesis test comparing the fit of the empirical and theoretical adolescent‐to‐mother models. TEFI= Total Entropy Fit Index.

3.2. Mother‐To‐Adolescent Autonomy and Relatedness

We next used EGA to examine the dimensional structure of mothers' behavior toward their adolescent during the AR task using the Louvain algorithm. EGA analyses also revealed a three‐dimensional structure for these variables (Figure 1). All items emerged on the same dimensions as the adolescent‐to‐mother variables, with the exception of “Queries,” which loaded onto the “Positive Relatedness” community in the mother‐to‐adolescent model, thereby replicating the theoretical categorization of this dimension. Thus, the same three dimensions were largely consistent between the adolescent‐to‐mother and mother‐to‐adolescent models, representing Positive Autonomy, Negative Autonomy and Relatedness, and Positive Relatedness dimensions.

We then conducted a parametric bootstrap EGA using the Louvain algorithm with 500 bootstrap samples. All item stability values were 0.87 or higher (Figure 4). Structural consistency values for the three dimensions were 0.95 for the Positive Autonomy dimension, 0.85 for the Negative Autonomy and Relatedness dimension, and 0.98 for the Positive Relatedness dimension, indicating a high replication of these communities across bootstrap samples.

FIGURE 4.

FIGURE 4

Open in a new tab

Bootstrap EGA results for the mother‐to‐adolescent model. Values represent the proportion of times each item was assigned to its specific empirical community in the bootstrap samples. Community 1 = Positive Autonomy, Community 2 = Negative Autonomy and Relatedness, Community 3 = Positive Relatedness.

Notably, “Queries” was much more stable when displayed by mothers versus adolescents, loading onto the Negative Autonomy and Relatedness community in only 0.6% of bootstrap models and the Positive Autonomy community in 1% of models. The network loading of this variable in the mother‐to‐adolescent model was also higher than the adolescent‐to‐mother model, with a value of 0.28 in the Positive Relatedness community.

The TEFI of the mother‐to‐adolescent EGA model was −4.56 versus a TEFI of −2.68 for the theoretical model. A nonparametric bootstrap hypothesis test indicated that the difference between these two TEFI values was also statistically significant (p = 0.002), suggesting that the EGA model was a better fit to the data than the theoretical model (Figure 5).

FIGURE 5.

FIGURE 5

Open in a new tab

Nonparametric hypothesis test comparing the fit of the empirical and theoretical mother‐to‐adolescent models. TEFI= Total Entropy Fit Index.

4. Discussion

The current study used Exploratory Graph Analysis (EGA) to examine the dimensional structure of autonomy and relatedness behaviors in mother‐adolescent interactions when teens were 13 years of age and compared these structures to the theoretical categories developed by the Autonomy and Relatedness (AR) Coding System. EGA identified three dimensions across variables representing both adolescents' behaviors toward their mother and mothers' behaviors toward their adolescent. The first dimension consisted of behaviors related to promoting autonomy, the second comprised behaviors corresponding to both undermining autonomy and relatedness, and the third consisted of behaviors promoting relatedness. Slight variations in the number and strength of connections between variables existed across the two models, and one variable loaded onto a separate community in the adolescent versus mother model. Bootstrap analyses confirmed the stability of items and dimensions. Finally, in both models, the EGA‐identified model was estimated to be a better fit to the data than the theoretical model.

The use of EGA in examining autonomy and relatedness behaviors between mothers and adolescents yielded novel insights that advance beyond purely theoretical conceptualizations. A key finding was the emergence of a three‐dimensional model rather than the theoretical four‐dimensional structure, with negative autonomy and relatedness behaviors merging into a single dimension. While the theoretical model treated hostile/devaluing statements and distracting behaviors as distinct elements undermining dyadic relationships, the EGA model revealed these clustered together with autonomy‐undermining behaviors like overpersonalizing and pressuring for agreement. This suggests that in practice, behaviors undermining relatedness inherently undermine autonomy and vice versa. Negative behaviors may actually overlap more in terms of their function than they appear. For example, psychologically controlling parental behavior serves to undermine both autonomy and relatedness due to parental love being conditional on the extent to which children act and think in the way dictated by the parent (Soenens and Vansteenkiste 2010). On the other hand, positive autonomy and relatedness behaviors may have more clearly differentiated functions than behaviors that undermine them. In a study of dimensions of parental behavior, parental autonomy support uniquely predicted the satisfaction of adolescents' need for autonomy, while parental warmth uniquely predicted the satisfaction of adolescents' need for relatedness. Negative parental autonomy and relatedness behaviors did not clearly predict teens' autonomy and relatedness needs but rather were linked to overall negative affect in adolescents (Costa et al. 2019). The consolidation of the negative autonomy and relatedness dimension in the EGA models provides a better understanding of how negative behaviors manifest in parent‐adolescent interactions and challenges previous theoretical distinctions between autonomy and relatedness undermining behaviors. Thus, the presence of two positive communities and only one negative community in the EGA result suggests that negative behaviors may be less distinguishable in their effect on autonomy and relatedness in the interaction than positive behaviors.

These findings are useful from a clinical perspective, as well as provide more information about the behaviors that may help versus hinder adolescent development. For example, both parents and adolescents experience high levels of distress in the context of conflict, even when positive behaviors are displayed, suggesting that negative behaviors may be more salient (Silva et al. 2020). Additionally, adolescents who have experienced a history of negative and controlling parenting are likely to be desensitized to autonomy‐supportive behaviors from their parent (Soenens and Vansteenkiste 2020; van Petegem et al. 2017). Negative behaviors may therefore have a more globally negative effect on parent‐adolescent interactions than positive behaviors. Some work has already considered examining negative autonomy and relatedness behaviors together as a whole (Loeb et al. 2020; Niolon et al. 2015), and these findings further suggest the validity of this approach.

The current findings also point to developmental differences in autonomy and relatedness behaviors between mothers and adolescents. Specifically, the transferring of querying from the Negative Autonomy and Relatedness dimension in the adolescent‐to‐mother model to the Positive Relatedness dimension in the mother‐to‐adolescent model suggests that this behavior presents differently when displayed by mothers versus teens. Perhaps when adolescents are exhibiting this behavior, the questions they are asking their mother could be perceived as pressuring or otherwise serve to have a negative effect on the interaction. This is supported by the connection between “Queries” and “Pressuring” in the adolescent‐to‐mother model, while “Queries” and “Engaged” were connected in the mother‐to‐adolescent model. Therefore, mothers and adolescents may have different developmental goals when engaging in this behavior. Given that teens are striving for autonomy during adolescence, they may be less likely to draw their parents out and be more focused on establishing their own position within the discussion, which is consistent with research suggesting autonomy and relatedness behaviors are less integrated in younger adolescents (Inguglia et al. 2015). When mothers are asking questions, these questions may serve the function of showing interest in their adolescent and promoting a positive relationship between the two (Beveridge and Berg 2007). Additionally, a follow‐up t‐test conducted to investigate the mean difference in adolescents' versus mothers' querying revealed that adolescents asked significantly fewer questions overall during the interaction (t(329) = −11.33, p < 0.001). While this further supports the instability of this variable, “Queries” still had relatively high item and dimensional stability in the adolescent‐to‐mother model and clearly loaded onto the Negative Autonomy and Relatedness community in the majority of bootstrap models. Thus, misclassifying this behavior into the Positive Relatedness dimension for teens may obscure important developmental differences between mothers and adolescents. These findings further underscore the importance of examining mothers' and adolescents' behaviors separately in the context of parent‐teen interactions and continuing to investigate developmental differences in autonomy and relatedness.

The current study provided support for the use of EGA to determine the dimensional structure of the AR Coding System. While the dimensions identified by EGA differed somewhat from the theoretical dimensions, the two are not necessarily incompatible. Instead, the EGA dimensions largely support the construct validity of the theoretically‐derived dimensions and provide more nuanced insights into the latent structures underlying behavior in the AR task, as well as the connections between variables. Another major strength of the study was the use of observed interactions between parents and teens. This measure likely provided greater objectivity in the assessment of autonomy and relatedness than self‐report questionnaires. Finally, the AR Coding System was developed by a team of experts and its usage in a range of international samples to successfully predict a variety of outcomes increases our confidence in the use of this system to categorize behaviors in parent‐adolescent interactions (Allen et al. 2012; Becker‐Stoll et al. 2008; Niolon et al. 2015; Samuolis et al. 2005; Zhang and Slesnick 2017).

There are several limitations of the study that are important to note. First, we examined only mother‐adolescent autonomy and relatedness. Adolescents' relationships with fathers, grandparents, extended family members, and peers are all potential contexts through which adolescents develop and hone their autonomy and relatedness skills. Our findings provide information about autonomy and relatedness behaviors in only one of adolescents' relationships within their broader social system. Additionally, given that EGA is an exploratory technique, the dimensions identified in the current study should not be seen as definitive. Future studies should seek to replicate these findings in a different sample.

5. Conclusion

The current study suggests that EGA is a useful tool for understanding autonomy and relatedness behaviors in mother‐adolescent disagreements. These findings have implications for the study of autonomy and relatedness by suggesting that positive autonomy and relatedness behaviors may each be more distinct from one another in comparison to negative autonomy and relatedness behaviors and by underscoring developmental differences in these behaviors. This study also contributes to burgeoning evidence that EGA is widely applicable to a range of psychological data and should continue to be used to examine behavior in parent‐adolescent interactions.

Funding

This work was supported by the National Institutes of Health, R01‐MH58066. Eunice Kennedy Shriver National Institute of Child Health and Human Development, R37HD058305. National Science Foundation Graduate Research Fellowship Program, 2234693.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Bootstrap EGA Results for the Teen‐to‐Mother Model Using the Walktrap Algorithm.

Figure S2: Bootstrap EGA Results for the Teen‐to‐Mother Model Using the Leiden Algorithm.

Figure S3: Bootstrap EGA Results for the Teen‐to‐Mother Model Using the Walktrap Algorithm.

Figure S4: Bootstrap EGA Results for the Teen‐to‐Mother Model Using the Leiden Algorithm.

FAMP-65-0-s001.docx (495.2KB, docx)

Acknowledgments

Preparation of this manuscript was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development, the National Institutes of Health, and the National Science Foundation Graduate Research Fellowship Program (R37HD058305, R01‐MH58066, and 2234693).

Bailey, N. A. , Golino H. F., and Allen J. P.. 2026. “Autonomy and Relatedness in Mother‐Adolescent Interactions: An Investigation Using Exploratory Graph Analysis.” Family Process 65, no. 1: e70116. 10.1111/famp.70116.

Endnotes

1

All three community detection algorithms (Louvain, Leiden, and Walktrap) produced the same community structure for both the adolescent‐to‐mother and mother‐to‐adolescent models.

2

Item stability values varied across the three community detection algorithms (Louvain, Leiden, and Walktrap) for both the adolescent‐to‐mother and mother‐to‐adolescent bootstrap models. See Data S1 for the bootstrap results for the Walktrap and Leiden algorithms.

Data Availability Statement

The data that support this paper are available from the corresponding author on reasonable request. The R code used in the manuscript is available at the following OSF page: https://doi.org/10.17605/OSF.IO/79CG3.

References

  1. Allen, J. P. , Chango J., Szwedo D., Schad M., and Marston E.. 2012. “Predictors of Susceptibility to Peer Influence Regarding Substance Use in Adolescence.” Child Development 83, no. 1: 337–350. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Allen, J. P. , Hauser S. T., Bell K. L., et al. 2000. The Autonomy and Relatedness Coding System. University of Virginia. [Google Scholar]
  3. Allen, J. P. , Hauser S. T., Bell K. L., and O'Connor T. G.. 1994. “Longitudinal Assessment of Autonomy and Relatedness in Adolescent‐Family Interactions as Predictors of Adolescent Ego Development and Self‐Esteem.” Child Development 65, no. 1: 179–194. [DOI] [PubMed] [Google Scholar]
  4. Allen, J. P. , Hauser S. T., Eickholt C., Bell K. L., and O'Connor T. G.. 1994. “Autonomy and Relatedness in Family Interactions as Predictors of Expressions of Negative Adolescent Affect.” Journal of Research on Adolescence 4, no. 4: 535–552. [Google Scholar]
  5. Allen, J. P. , Hauser S. T., O'Conner T. G., and Bell K. L.. 2002. “Prediction of Peer‐Rated Adult Hostility From Autonomy Struggles in Adolescent–Family Interactions.” Development and Psychopathology 14, no. 1: 123–137. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Becker‐Stoll, F. , Fremmer‐Bombik E., Wartner U., Zimmermann P., and Grossmann K. E.. 2008. “Is Attachment at Ages 1, 6 and 16 Related to Autonomy and Relatedness Behavior of Adolescents in Interaction Towards Their Mothers?” International Journal of Behavioral Development 32, no. 5: 372–380. [Google Scholar]
  7. Beveridge, R. M. , and Berg C. A.. 2007. “Parent–Adolescent Collaboration: An Interpersonal Model for Understanding Optimal Interactions.” Clinical Child and Family Psychology Review 10, no. 1: 25–52. [DOI] [PubMed] [Google Scholar]
  8. Blondel, V. D. , Guillaume J. L., Lambiotte R., and Lefebvre E.. 2008. “Fast Unfolding of Communities in Large Networks.” Journal of Statistical Mechanics: Theory and Experiment 2008, no. 10: P10008. [Google Scholar]
  9. Branje, S. 2018. “Development of Parent–Adolescent Relationships: Conflict Interactions as a Mechanism of Change.” Child Development Perspectives 12, no. 3: 171–176. [Google Scholar]
  10. Chihara, L. M. , and Hesterberg T. C.. 2022. Mathematical Statistics With Resampling and R. John Wiley & Sons. [Google Scholar]
  11. Christensen, A. , and Golino H.. 2019. “Estimating the Stability of the Number of Factors via Bootstrap Exploratory Graph Analysis: A Tutorial.” PsyArXiv: 1–32. [Google Scholar]
  12. Christensen, A. P. , Garrido L. E., and Golino H.. 2023. “Unique Variable Analysis: A Network Psychometrics Method to Detect Local Dependence.” Multivariate Behavioral Research 58, no. 6: 1165–1182. [DOI] [PubMed] [Google Scholar]
  13. Christensen, A. P. , Garrido L. E., Guerra‐Peña K., and Golino H.. 2024. “Comparing Community Detection Algorithms in Psychometric Networks: A Monte Carlo Simulation.” Behavior Research Methods 56, no. 3: 1485–1505. [DOI] [PubMed] [Google Scholar]
  14. Christensen, A. P. , and Golino H.. 2021a. “On the Equivalency of Factor and Network Loadings.” Behavior Research Methods 53, no. 4: 1563–1580. [DOI] [PubMed] [Google Scholar]
  15. Christensen, A. P. , and Golino H.. 2021b. “Estimating the Stability of Psychological Dimensions via Bootstrap Exploratory Graph Analysis: A Monte Carlo Simulation and Tutorial.” Psychiatrist 3, no. 3: 479–500. [Google Scholar]
  16. Christensen, A. P. , Golino H., Abad F. J., and Garrido L. E.. 2025. “Revised Network Loadings.” Behavior Research Methods 57, no. 4: 1–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Christensen, A. P. , Gross G. M., Golino H. F., Silvia P. J., and Kwapil T. R.. 2019. “Exploratory Graph Analysis of the Multidimensional Schizotypy Scale.” Schizophrenia Research 206: 43–51. [DOI] [PubMed] [Google Scholar]
  18. Cicchetti, D. V. , and Sparrow S. A.. 1981. “Developing Criteria for Establishing Interrater Reliability of Specific Items: Applications to Assessment of Adaptive Behavior.” American Journal of Mental Deficiency 86, no. 2: 127–137. [PubMed] [Google Scholar]
  19. Cook, E. C. , Chaplin T. M., and Stroud L. R.. 2015. “The Relationship Between Autonomy and Relatedness and Adolescents' Adrenocortical and Cardiovascular Stress Response.” Journal of Youth and Adolescence 44, no. 11: 1999–2011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Costa, S. , Sireno S., Larcan R., and Cuzzocrea F.. 2019. “The Six Dimensions of Parenting and Adolescent Psychological Adjustment: The Mediating Role of Psychological Needs.” Scandinavian Journal of Psychology 60, no. 2: 128–137. [DOI] [PubMed] [Google Scholar]
  21. Epskamp, S. , and Fried E. I.. 2018. “A Tutorial on Regularized Partial Correlation Networks.” Psychological Methods 23, no. 4: 617–634. [DOI] [PubMed] [Google Scholar]
  22. Gates, K. M. , Henry T., Steinley D., and Fair D. A.. 2016. “A Monte Carlo Evaluation of Weighted Community Detection Algorithms.” Frontiers in Neuroinformatics 10: 1–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Golino, H. , Christensen A. P., and Moulder R.. 2025. “EGAnet: Exploratory Graph Analysis: A Framework for Estimating the Number of Dimensions in Multivariate Data Using Network Psychometrics.” R Package Version 2, no. 1: 1. [Google Scholar]
  24. Golino, H. , Moulder R., Shi D., et al. 2021. “Entropy Fit Indices: New Fit Measures for Assessing the Structure and Dimensionality of Multiple Latent Variables.” Multivariate Behavioral Research 56, no. 6: 874–902. [DOI] [PubMed] [Google Scholar]
  25. Golino, H. , Shi D., Christensen A. P., et al. 2020. “Investigating the Performance of Exploratory Graph Analysis and Traditional Techniques to Identify the Number of Latent Factors: A Simulation and Tutorial.” Psychological Methods 25, no. 3: 292–320. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Golino, H. F. 2024. Modern Psychometrics With Exploratory Graph Analysis Using R: Structural Validity, Item Analysis, and Beyond [Book Chapter]. Unpublished Manuscript .
  27. Golino, H. F. , and Demetriou A.. 2017. “Estimating the Dimensionality of Intelligence Like Data Using Exploratory Graph Analysis.” Intelligence 62: 54–70. [Google Scholar]
  28. Golino, H. F. , and Epskamp S.. 2017. “Exploratory Graph Analysis: A New Approach for Estimating the Number of Dimensions in Psychological Research.” PLoS One 12, no. 6: e0174035. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Grotevant, H. D. , and Cooper C. R.. 2013. “Patterns of Interaction in Family Relationships and the Development of Identity Exploration in Adolescence.” In Adolescents and Their Families, 103–116. Routledge. [PubMed] [Google Scholar]
  30. Huth, K. , Haslbeck J. M. B., Keetelaar S., van Holst R., and Marsman M.. 2025. Statistical Evidence in Psychological Networks: A Bayesian Analysis of 294 Networks From 126 Studies. Nature Human Behaviour. [DOI] [PubMed] [Google Scholar]
  31. Inguglia, C. , Ingoglia S., Liga F., Lo Coco A., and Lo Cricchio M. G.. 2015. “Autonomy and Relatedness in Adolescence and Emerging Adulthood: Relationships With Parental Support and Psychological Distress.” Journal of Adult Development 22, no. 1: 1–13. [Google Scholar]
  32. Isvoranu, A. M. , and Epskamp S.. 2023. “Which Estimation Method to Choose in Network Psychometrics? Deriving Guidelines for Applied Researchers.” Psychological Methods 28, no. 4: 925–946. [DOI] [PubMed] [Google Scholar]
  33. Jamison, L. , Christensen A. P., and Golino H.. 2021. Optimizing Walktrap's Community Detection in Networks Using the Total Entropy Fit Index.
  34. Lancichinetti, A. , and Fortunato S.. 2012. “Consensus Clustering in Complex Networks.” Scientific Reports 2, no. 1: 336. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Loeb, E. L. , Davis A., Costello M., and Allen J. P.. 2020. “Autonomy and Relatedness in Early Adolescent Friendships as Predictors of Short‐ and Long‐Term Academic Success.” Social Development 29, no. 3: 818–836. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Niolon, P. H. , Kuperminc G. P., and Allen J. P.. 2015. “Autonomy and Relatedness in Mother–Teen Interactions as Predictors of Involvement in Adolescent Dating Aggression.” Psychology of Violence 5, no. 2: 133–143. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Oudekerk, B. A. , Allen J. P., Hessel E. T., and Molloy L. E.. 2015. “The Cascading Development of Autonomy and Relatedness From Adolescence to Adulthood.” Child Development 86, no. 2: 472–485. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Phinney, J. S. , Kim‐Jo T., Osorio S., and Vilhjalmsdottir P.. 2005. “Autonomy and Relatedness in Adolescent‐Parent Disagreements: Ethnic and Developmental Factors.” Journal of Adolescent Research 20, no. 1: 8–39. [Google Scholar]
  39. Pons, P. , and Latapy M.. 2005. “Computing Communities in Large Networks Using Random Walks.” In Computer and Information Sciences‐ISCIS 2005: 20th International Symposium, Istanbul, Turkey, October 26–28, 2005. Proceedings 20, 284–293. Springer. [Google Scholar]
  40. R Core Team . 2024. R: A Language and Environment for Statistical Computing (Version 2024.12.0). R Foundation for Statistical Computing. https://www.R‐project.org/. [Google Scholar]
  41. Samuolis, J. , Hogue A., Dauber S., and Liddle H. A.. 2005. “Autonomy and Relatedness in Inner‐City Families of Substance Abusing Adolescents.” Journal of Child & Adolescent Substance Abuse 15, no. 2: 53–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Shah, E. N. , Szwedo D. E., and Allen J. P.. 2023. “Parental Autonomy Restricting Behaviors During Adolescence as Predictors of Dependency on Parents in Emerging Adulthood.” Emerging Adulthood 11, no. 1: 15–31. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Silva, K. , Ford C. A., and Miller V. A.. 2020. “Daily Parent–Teen Conflict and Parent and Adolescent Well‐Being: The Moderating Role of Daily and Person‐Level Warmth.” Journal of Youth and Adolescence 49, no. 8: 1601–1616. [DOI] [PubMed] [Google Scholar]
  44. Soenens, B. , and Vansteenkiste M.. 2010. “A Theoretical Upgrade of the Concept of Parental Psychological Control: Proposing New Insights on the Basis of Self‐Determination Theory.” Developmental Review 30, no. 1: 74–99. [Google Scholar]
  45. Soenens, B. , and Vansteenkiste M.. 2020. “Taking Adolescents' Agency in Socialization Seriously: The Role of Appraisals and Cognitive‐Behavioral Responses in Autonomy‐Relevant Parenting.” New Directions for Child and Adolescent Development 2020, no. 173: 7–26. [DOI] [PubMed] [Google Scholar]
  46. Steinberg, L. 1990. Interdependency in the Family: Autonomy, Conflict, and Harmony in the Parent‐Adolescent Relationship. At the Threshold: The Developing Adolescent.
  47. Traag, V. A. , Waltman L., and van Eck N. J.. 2019. “From Louvain to Leiden: Guaranteeing Well‐Connected Communities.” Scientific Reports 9, no. 1: 1–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. van Petegem, S. , Zimmer‐Gembeck M. J., Soenens B., et al. 2017. “Does General Parenting Context Modify Adolescents' Appraisals and Coping With a Situation of Parental Regulation? The Case of Autonomy‐Supportive Parenting.” Journal of Child and Family Studies 26: 2623–2639. [Google Scholar]
  49. Wray‐Lake, L. , Crouter A. C., and McHale S. M.. 2010. “Developmental Patterns in Decision‐Making Autonomy Across Middle Childhood and Adolescence: European American Parents' Perspectives.” Child Development 81, no. 2: 636–651. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Zhang, J. , and Slesnick N.. 2017. “Discrepancies in Autonomy and Relatedness Promoting Behaviors of Substance Using Mothers and Their Children: The Effects of a Family Systems Intervention.” Journal of Youth and Adolescence 46: 668–681. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Figure S1: Bootstrap EGA Results for the Teen‐to‐Mother Model Using the Walktrap Algorithm.

Figure S2: Bootstrap EGA Results for the Teen‐to‐Mother Model Using the Leiden Algorithm.

Figure S3: Bootstrap EGA Results for the Teen‐to‐Mother Model Using the Walktrap Algorithm.

Figure S4: Bootstrap EGA Results for the Teen‐to‐Mother Model Using the Leiden Algorithm.

FAMP-65-0-s001.docx (495.2KB, docx)

Data Availability Statement

The data that support this paper are available from the corresponding author on reasonable request. The R code used in the manuscript is available at the following OSF page: https://doi.org/10.17605/OSF.IO/79CG3.

Articles from Family Process are provided here courtesy of Wiley

RESOURCES