Bridging the gaps: Comparing structural equation models to network analysis models of depression, anxiety, and perfectionism

Network models of psychopathology can identify specific items/symptoms that explain the connections among broader constructs such as depression, anxiety, and perfectionism. In two studies, we examine the dynamic interplay between depression, anxiety, and perfectionism symptoms among undergraduates using structural equation modeling (SEM) and network analysis. Participants in two independent samples (N = 774 and N = 759) completed online, cross-sectional questionnaires including measures of anxiety, depressive symptoms, and perfectionism (i.e., concerns over mistakes, doubts about actions, and personal standards). When analyzing data in the traditional fashion using SEM as a point of comparison, results from both samples were consistent with the existing literature. After controlling for all other perfectionism variables in the model, concerns over mistakes and doubts about actions were positively associated with depressive and anxiety symptoms (βs from .21 to .46), while personal standards showed negative associations with depressive symptoms (β = -.20 both samples) and non-significant associations with anxiety symptoms (βs from -.09. to -.03). Nonetheless, model fit for the confirmatory factor model was below ideal cutoffs in the second sample, suggesting other structures (e.g., a network model) might better represent the data. Network analyses revealed associations be-tween constructs at the item level across both samples. Four key symptoms emerged as central nodes linking depression, anxiety, and perfectionism: difficulty taking initiative to do activities, feeling worthless, feeling close to panic, and doubts about simple everyday activities. This study underscores the importance of investigating item-level associations for a nuanced interpretation of these constructs.

Depressive symptoms include persistent sadness, reduced interest/pleasure in once-enjoyable activities, and impairments in concentration, sleep, and appetite; in contrast, anxiety symptoms are characterized by excessive worry and heightened physiological arousal (American Psychiatric Association, 2013).Both psychological symptoms are prevalent worldwide, especially in the face of stressful events (Racine et al., 2021).Moreover, they contribute to decreased quality of life, chronic illnesses, and economic burden (Greenberg et al., 2021;Whiteford et al., 2013;Zhang et al., 2018).Although conceptually distinct, depressive and anxiety symptoms also tend to co-occur within people (Kessler et al., 2015); however, the reasons for their co-occurrence are not fully understood.
Psychopathology has been traditionally conceptualized using reflective models (Schmittmann et al., 2013), such as latent variables in structural equation models.In reflective models, mental disorders are conceptualized as underlying entities causing symptoms.Consequently, individual symptoms like feeling worthless and feeling life is meaningless should positively correlate because they are caused by the same underlying latent variable, namely depression.However, in reflective models, these correlations are spurious: symptom A, feeling worthless, does not directly cause symptom B, feeling life is meaninglesstheir relationship is the result of a third, unmeasured variable (depression) that affects both symptoms A and B. Therefore, reflective models by their very nature preclude the possibility that symptoms might influence each other.
An alternative approach to psychopathology is emerging.The network model describes psychopathology as the dynamic interplay of symptoms that worsen and maintain one another (Borsboom & Cramer, 2013;Epskamp et al., 2018;Jones et al., 2021).For example, feeling worthless is understood to reinforce the feeling that life is meaningless, and vice versa, in the network model of depressive symptoms (Van den Bergh et al., 2021).In other words, when an individual's competence is undermined, their ability to derive meaning from life is similarly diminished.This finding aligns with the basic psychological needs portion of self-determination theory (Ryan & Deci, 2000), which proposes that the basic psychological need for competence, such as mastering challenges and achieving success, is integral to eudaemonic well-being (i.e., finding meaning in life).
One notable feature of network models is their ability to explain comorbidities through "bridge symptoms" (Cramer et al., 2010).Bridge symptoms are those that connect two different disorders.In the network model of depressive and anxiety symptoms, worries about panicking and appearing foolish in public (anxiety symptom) seem to trigger feeling worthless (depressive symptom), and vice versa (Van den Bergh et al., 2021).The negative social repercussions of panicking in public may intensify feeling worthless, leading to avoidance behaviors like social withdrawal and to reinforcing worries about future public panic episodes (Ike et al., 2020).In this way, the presence of these bridge symptoms serves as the mechanism that connects symptoms of depression and anxiety, explaining their comorbidity.Furthermore, when bridge symptoms are present, it increases the likelihood that the activation of symptoms within one cluster of symptoms will spread and eventually activate another cluster of symptoms (Jones et al., 2021).Identifying bridge symptoms has significant clinical implications, such as targeted treatment of bridge symptoms.While network analysis studies have examined depressive and anxiety symptoms (e.g., Van den Bergh et al., 2021), none have done so including putative predictors like perfectionism (Kim, Sherry, et al., 2024;Smith et al., 2018Smith et al., , 2021)).
Perfectionism is often conceptualized as a multidimensional construct (Hewitt & Flett, 1991).Frost et al. (1990) operationalizes some of perfectionism's facets by assessing concern over mistakes (being extremely concerned about making errors and failing to comply with set standards), doubts about actions (uncertainties about one's own performance), and personal standards (setting of high standards for one's own behavior).Concern over mistakes and doubts about actions are generally regarded as maladaptive aspects of perfectionism linked to various psychopathologies, such as depression, anxiety, interpersonal conflict, and binge eating (Damian et al., 2017;Kim, Sherry, et al., 2023;Mackinnon et al., 2012).In comparison, some researchers have suggested that personal standards can be adaptive or neutral, promoting resiliency that buffers against depressive symptoms (Enns et al., 2005;Stoeber & Otto, 2006;Wu & Wei, 2008).However, these claims have not gone unchallenged.For instance, meta-analytic research has found that personal standards leads to more, rather than less, depressive symptoms (Smith et al., 2016).
Extensive research has examined the associations between symptoms of depression, anxiety, and perfectionism (Egan et al., 2013;Enns et al., 2002;Smith et al., 2015), often relying on statistical methods like structural equation modeling (SEM).However, the SEM approach treats perfectionism dimensions as unobservable latent variables, essentially ruling out the possibility of direct associations between individual questionnaire items by the very nature of the model.In contrast, a network approach could offer insights into the specific items that comprise concern over mistakes, doubts about actions, and personal standards and how they might contribute to specific symptoms of depression and anxiety in the context of a complex network of interrelationships.Despite the growing popularity of network models in personality research (Costantini et al., 2015;Di Fabio et al., 2022), no studies have yet explored the dynamic interplay of concern over mistakes, doubts about actions, 1 All core analyses (the SEM results and the network analysis were re-run on the open-access dataset with the 22 participants redacted, and results were virtually identical to the results presented in this paper.Because the findings were so similar, we do not present them in the paper but interested readers can review the open-access data and syntax for themselves at https://osf.io/sfwxn/.The full dataset is available upon request by emailing the corresponding author.
personal standards, depressive symptoms, and anxiety symptoms using network analysis.To address this research gap, the present study aimed to address the following research questions: Research Question 1: How do the findings differ when comparing conclusions from SEM and network analysis in examining concern over mistakes, doubts about actions, and personal standards as predictors for depressive and anxiety symptoms?
Research Question 2: Additionally, which bridge symptoms play a significant role in understanding the associations between concern over mistakes, doubts about actions, and personal standards, depressive symptoms, and anxiety symptoms?

METHOD
Data for Sample 1 relied on existing open-access data (osf.io/aec25/)and analyses were not preregistered.Sample 2 analyses were preregistered and can be found on the Open Science Framework (https://doi.org/10.17605/OSF.IO/-VDKX2).Following institutional Research Ethics Board procedures, Sample 2 participants were asked for consent to use their data for research and consent to share their data in an open-access repository separately, with some (n = 22) granting consent for the former but not the latter.Results presented in this paper for Sample 2 includes these 22 participants.Thus, Sample 2's open-access dataset has 22 participants redacted 1 .Copies of all study materials, the open access datasets, and all statistical syntax for both studies can be found at https://osf.io/sfwxn/.

Participants and procedure
This study used two separate undergraduate samples recruited through online ads, flyers, and the undergraduate research participant pool at a Canadian university (N = 774 for Sample 1; N = 759 for Sample 2).Participants completed cross-sectional online surveys and were mostly women (79.2%; 80.6%), young (M age = 21.1,SD = 6.0;M age = 20.2,SD = 3.3), and White (72.1%;72.3%).There were no inclusion/exclusion criteria.Sample 1 was collected between February 2018 and February 2019.The data were drawn from a study examining response-order effects, where the order of item responses (ascending vs. descending) was manipulated but showed no significant differences in participants' responses (Mackinnon & Wang, 2020), so this manipulation is unlikely to influence the analyses.To ensure the robustness and reliability of Sample 1's network model (i.e., to reduce Type 1 errors), we also collected a replication sample (Sample 2) between October 2023 and April 2024.

Materials
Participants from both samples responded to two virtually identical measures of depression, anxiety and perfectionism, except that for depression and anxiety in Sample 1, an adapted response instruction was used.

Depressive and anxiety symptoms
Depressive and anxiety symptoms were measured using the Depression Anxiety Stress Scale-21 (DASS-21; Lovibond & Lovibond, 1995).The DASS-21 aligns with the threefactor structure of depression (e.g., "I felt down-hearted and blue"), anxiety (e.g., "I felt close to panic"), and stress (e.g., "I found it hard to wind down") symptoms (Lee et al., 2019).Only the depression and anxiety subscales (seven items each) were analyzed in this study, as general stress is common to all emotional disorders and not specific to either depression or anxiety (Clark & Watson, 1991).Moreover, our research questions pertained to anxiety and depression symptoms specifically (rather than stress more generally), and we wanted to reduce the number of nodes in the model given the sample size and large number of items already included.The DASS-21 is a widely used measure in the field, with established reliability and validity in normative samples (Henry & Crawford, 2005).Research has found medium-to-strong correlations between DASS-21 depression and anxiety (Osman et al., 2012), as well as the predictive validity of each construct (Kim et al., 2022;Kim, Merlo, et al., 2023).
The response instruction adaptation for Sample 1 concerned the following: a 5-point Likert response format (i.e., 1 strongly disagree to 5 strongly agree) and a longer measurement period of "the past year" was used to allow for more variation in responses and to maintain consistency with other measures in the parent study (Mackinnon & Wang, 2020).In Sample 2, we used DASS-21's original rating scale of 0 ("did not apply to me at all") to 3 ("applied to me very much or most of the time") and the original reference period of the "past 7 days" to align with how it is typically used throughout the literature (Henry & Crawford, 2005).

Perfectionism
Perfectionism was measured using the Frost Multidimensional Perfectionism Scale (Frost et al., 1990), including three short-form subscales developed by Cox et al. (2002): a 5-item Concern Over Mistakes (COM) subscale (e.g., "If I fail at work/school, I am a failure as a person"), a 4-item Doubts About Actions (DAA) subscale (e.g., "Even when I do something very carefully, I often feel that it is not quite right"), and a 4-item Personal Standards (PS) subscale (e.g., "I set higher goals than most people.").Participants responded to items on a 5-point Likert scale from 1 ("strongly disagree") to 5 ("strongly agree").Items referred to the measurement period of "the past several years".Cox et al. (2002) demonstrated the factor structure of these short forms2 .

Data analytic strategy
Both samples were analyzed using the same data analysis strategy with the R software (version 4.2.1).The SEM analysis used the lavaan package (Rosseel, 2012), beginning with a 5-factor confirmatory factor analysis (CFA).Then, we conducted a structural model with all three perfectionism subscale measures predicting both depressive and anxiety symptoms.Items were specified as ordinal variables.Thus, for both the CFA and structural model, the diagonal weighted least squares (DWLS) estimator and robust estimates of standard errors were used.A well-fitting model has a Confirmatory Fit Index (CFI) and Tucker-Lewis Index (TLI) around .95 or higher, Root Mean Square Error of Approximation (RMSEA) of around .06 or lower, and Standardized Root Mean Squared Residual (SRMR) around .08 or lower (Kline, 2005).Though we report the χ 2 goodness of fit statistic, we did not use the p-value as a measure of model fit because the chi-square is overly sensitive to large sample sizes (Bergh, 2015).
The network analysis was analyzed using the bootnet package (Epskamp et al., 2018).The network models were estimated using least absolute shrinkage and selection operator (LASSO) with the Extended Bayseian Information Criterion (EBIC) to aid with model selection.As in the SEM analyses, the items were treated as ordinal.The correlation matrix was estimated with the polychoric method using the corMethod = "cor_auto" argument.Thresholding was set to "FALSE", which results in strong Type II error control, but may lead to higher Type I error rates; however, we mitigate this problem by collecting a preregistered replication sample.Listwise deletion was used for missing data, except for analyses of averaged total scores in Table 1, where scale totals were calculated by taking the average of all completed items for any given participant.The number of complete observations divided by the total number of observations yielded the proportion of missing data.Sample 1 displayed 5.0% missing data (N = 735 complete observations), and Sample 2 had 2.5% missing data (N = 740 complete observations).
Some terminology must be parsed to interpret network model results."Nodes" refer in this case to individual questionnaire items."Edges" refer to the partial polychoric correlations between items.A "community" is defined as a group of theoretically similar nodes, in this case: perfectionism, anxiety, and depression."Bridge centrality statistics" are summary statistics that attempt to quantify which nodes most strongly connect communities together (Jones et al., 2021).There are many different types of bridge centrality statistics (e.g., betweenness, closeness), but we focus on two in the present paper that are more relevant and applicable within the context of psychopathology networks (Jones et al., 2021): "Bridge strength" refers to the absolute value of a node's total edges with other communities, ignoring the direction of effects, and "Bridge expected influence" refers to a summed edges between communities without taking the absolute value.Bridge expected influence can be computed for 1-step (one edge away) or 2step (up two edges away).Internal reliability of the network was assessed using procedures outlined in Epskamp et al. (2018); these analyses calculated how much the results are affected by sampling variation (i.e., accuracy) and how much the results change if observations are removed (i.e., stability).We looked at the stability of centrality indices using a case-dropping bootstrap with 2,500 resamples.This produces the CS-coefficient, which should ideally be around .50 or more (Epskamp et al., 2018) as well as a plot of how centrality indices change as a function of sample size.Accuracy of all edges were also described with all bootstrapped edge weights.

Descriptive statistics
Table 1 presents the means, standard deviations, and bivariate correlations of the variables in both samples, derived from the total mean scores of COM, DAA, PS, depressive symptoms, and anxiety symptoms.Supplementary Table S1 reports the means and standard deviations of measures at the item-level for both samples.
In Sample 1, all variables showed significant positive correlations (rs from .10 to .60), with the strongest correlation observed between depressive and anxiety symptoms.PS showed weak positive correlations with depressive and anxiety symptoms.In Sample 2, all variables, except PS with depressive symptoms (r = .01,p = .774)and anxiety symptoms (r = .06,p = .135),showed significant positive correlations (rs from .10 to .62).COM and DAA demonstrated moderate correlations with depressive and anxiety symptoms across both samples, while PS showed a moderate correlation with COM and a weak correlation with DAA.Polychoric correlations at the item-level for Sample 1 and Sample 2 are reported in Supplementary Figures S1 and S2, respectively.

Structural model
Although it is not typically advised to estimate a structural model when the measurement model lacks good fit (which may apply to Sample 2), we will proceed with the structural model to facilitate comparison with network analysis results later (Figure 2).In both samples, the results indicated positive and significant paths from COM and DAA to depressive and anxiety symptoms, even after controlling for PS.Additionally, PS demonstrated a negative effect on depressive symptoms and a non-significant effect on anxiety symptoms after controlling for COM and DAA.This phenomenon is known as a "suppressor effect" (Martinez Gutierrez & Cribbie, 2021) where the magnitude of β for PS when predicting depressive and anxiety symptoms reverses in sign or becomes nonsignificant compared to their positive bivariate correlations with depressive and anxiety symptoms after controlling for COM and DAA .

Network diagram
Figures 3 and 4 illustrate the graphical representation of the estimated polychoric correlation network for Sample 1 and 2, respectively.In general, most edges showed positive associations, except for some edges connecting PS and depressive symptoms, which showed negative associations (e.g., depression item 2 with PS items 2 and 4).Items within each construct demonstrated the strongest positive connections, indicated by the presence of thick green edges.Though edge strengths were generally strongest within communities, the strongest edge connecting different communities consistently across both samples appeared to be between depression item 4 ("I felt down-hearted and blue") and anxiety item 5 ("I felt I was close to panic").Numerical coefficients and bootstrapped confidence intervals for every edge in the models are plotted 3 in Supplementary Figure S3 for Sample 1 and Supplementary Figure S4 for Sample 2.
Figures 5 and 6 present the bridge strength and expected influence statistics for Sample 1 and 2, respectively.In both samples, anxiety item 5 ("I felt I was close to panic") showed high coefficients across the three categories of bridge strength, bridge expected influence (1-step), and bridge expected (2-step).This suggests that anxiety item 5 serves as the connection between the community of anxiety symptoms with depressive symptoms 3 Because of the sheer number of edges, a supplementary digital pdf file is needed because these results must exceed the boundaries of a printed page to be legible. 4Additionally, Sample 1's CS-coefficients for closeness (.44) and bridge closeness (.52) were only adequate, whereas the CS-coefficients for betweenness were relatively poor, with values of .13.Similarly, Sample 2's CS-coefficients were .28 and .36 for closeness and bridge closeness, respectively, as well as.13 for both betweenness and bridge betweenness, respectively.As a result, we do not discuss these centrality statistics in the manuscript as they would not necessarily be stable, generalizable results.and perfectionism.Furthermore, depression item 2 ("I found it difficult to work up the initiative to do things") and depression item 6 ("I felt I wasn't much worth as a person") exhibited high coefficients in all three bridge statistic categories across both samples.Therefore, these two depression items connect the community of depressive symptoms to anxiety symptoms and perfectionism.Among perfectionism items across both samples, DAA item 2 ("I usually have doubts about the simple everyday things I do") displayed high coefficients across the three bridge statistic categories.This underscores DAA item 2 as a link between the perfectionism community with depressive and anxiety symptoms.
Concerning the stability of the network, both Sample 1 and Sample 2's CS-coefficients were high for bridge strength (.67 for Sample 1 and .75 for Sample 2) and bridge expected influence (.75 for both samples).Therefore, the results of the three bridge statistic categories reported in Figures 5 and 6 can be reliably interpreted given their sample sizes (see Supplementary Figures S5 and S6 for a plot of how centrality indices change as a function of sample size for Samples 1 and 2, respectively) 4 .Although not the focus of this paper, both strength and expected influence (not bridge) statistics showed high CS-coefficients for Sample 1 (.75) and Sample 2 (.75).Supplementary Figure S7 and S8 depicts the strength and expected influence statistics for each node among Sample 1 and 2, respectively.These statistics consider all the node's connections, including within the same community, unlike bridge statistics.With regards to edge weight accuracy, bootstrapped confidence intervals around the estimated edge-weights indicated a reasonable precision, as shown in Supplementary Figure S3 (Sample 1) and S4 (Sample 2).

DISCUSSION
The present study included two large independent samples, improving the reliability of findings through replication.We investigated the associations between depressive symptoms, anxiety symptoms, and perfectionism among undergraduates using SEM and network analysis.Traditionally, most researchers use multiple regression or SEM-style models to test whether perfectionism predicts depressive and anxiety symptoms (Egan et al., 2013;Enns et al., 2002;Smith et al., 2015), so these results were included as a point of comparison to how analyses are typically done.
The SEM results partially corroborated existing literature, revealing positive associations between COM and DAA with both depressive and anxiety symptoms (Damian et al., 2017;Mackinnon et al., 2012), while PS demonstrated a negative association with depressive symptoms (Enns et al., 2005;Stoeber & Otto, 2006;Wu & Wei, 2008).On the other hand, Sample 2's factor structure was somewhat weaker than anticipated, which is generally obscured in multiple regression-style research using sum scores.Note.Green edges represent positive polychoric correlations, red edges represent negative polychoric correlations.Thicker edges correspond to stronger associations.The layout is based on a forced-directed algorithm to produce visually appealing plots.Therefore, the positioning of the nodes cannot be interpreted.Note.Green edges represent positive polychoric correlations, red edges represent negative polychoric correlations.Thicker edges correspond to stronger associations.The layout is based on a forced-directed algorithm to produce visually appealing plots.Therefore, the positioning of the nodes cannot be interpreted.
The network analysis provided greater insights into these associations at the item level.For instance, across both samples, depression item 2 ("I found it difficult to work up the initiative to do things") was negatively associated with PS item 2 ("I set higher goals than most people") and PS item 4 ("I expected higher performance in my daily tasks than most people").Furthermore, the bridge strength of depression item 2 (i.e., sum of absolute edge values) surpassed its expected bridge influence (i.e., sum of edges without taking absolute values), indicating the substantial impact of the negative associations between depression item 2 and the PS items.Therefore, upholding excessive personal standards may be protective against depressive symptoms by promoting initiative and engagement in activities.This interpretation aligns well with the principals of behavioral activation, a highly effective component of cognitivebehavioral therapy for depression focused on increasing adaptive and pleasurable activities while reducing activities that maintain its symptoms (Ekers et al., 2014).Additionally, having high performance expectations promoting activity engagement might lend to competence in activities and provide meaning in life (Ryan & Deci, 2000), thus buffering against depression symptoms.
Past research has identified feeling worthless as a key node in depression and anxiety networks (Langer et al., 2019;Van den Bergh et al., 2021).Similarly, both of our  study samples found high bridge strength and bridge expected influence for feeling worthless (depression item 6; "I felt I wasn't much worth as a person"), confirming and extending past research to include perfectionism.One possible interpretation is that feelings of worthlessness arise when an individual's sense of self-worth depends on success at work/school (e.g., COM item 1; "If I fail at work/school, I am a failure as a person").Thus, understanding feeling worthless as a symptom of depression should consider the broader context of perfectionism and situations that induce fears of failure.
Across both samples, feeling close to panic (anxiety item 5; "I felt I was close to panic") displayed high bridge strength and bridge expected influence among all nodes, which can be partly attributed to its strong positive connection with low mood (depression item 4; "I felt downhearted and blue").This underscores the importance of considering low mood and panic within the depressionanxiety link.Specifically, low mood may prompt avoidance behaviors, such as social withdrawal, that reinforce fear and panic in certain situations (Ike et al., 2020).Additionally, reflecting on the inner experiences of panic could trigger low mood, as observed in individuals with heightened anxiety sensitivity (Kim, Stewart, et al., 2024).Moreover, recurring panic episodes without resolution may lead to learned helplessness, exacerbating and maintaining low mood (Maier & Seligman, 2016).
Regarding perfectionism, doubts about doing simple everyday activities (DAA item 2; "I usually have doubts about the simple everyday things I do") had the highest bridge strength and bridge expected influence across both samples, signifying its significant contributions to depression and anxiety symptoms.For perfectionists, whose self-worth is contingent on being successful, feeling incapable of even simple activities may be devastating.Moreover, perfectionists tend to engage in high levels of rumination, characterized by repetitive negative thinking (Olson & Kwon, 2008).Doubts about doing simple activities encountered in daily life might manifest as rumination, an established risk factor for depression and anxiety (Aldao et al., 2010;Connolly & Alloy, 2017;Michl et al., 2013).Future research should examine whether rumination is a mediating link between these everyday performance doubts and symptoms of depression and anxiety.

Limitations
This study has several limitations.Firstly, the crosssectional nature of the data prevents establishing the direction or temporal precedence of connections in the network model.Secondly, both samples mostly consist of young White undergraduate women, limiting generalizability to diverse populations.Thirdly, there may be other symptoms, such as binge eating (Kim, Sherry, et al., 2023), that could help explain the connections between perfectionism, depression, and anxiety among undergraduate women.Including these symptoms might reveal new strong edges in the network, reducing the significance of some original edges (Haslbeck & Fried, 2017).Lastly, findings are based on group-level averages and may not readily apply to individual cases; using a nomothetic approach could obscure symptom associations for certain individuals.

Conclusion
In this replication and extension of cross-sectional data among undergraduates, the four key symptoms identifieddifficulty taking initiative to do activities, feeling worthless, feeling close to panic, and doubts about simple everyday activitiesappear to be crucial in the spread of activation across symptoms of depression, anxiety, and perfectionism.

Figure 1 .
Figure 1.Five-factor measurement model using confirmatory factor analysis with standardized factor loadings for Samples 1 and 2. Note.DEP = depression; ANX = anxiety; COM = concern over mistakes; DAA = doubts about actions; PS = personal standards.Ovals represent latent variables.Rectangles represent observed variables.The black, straight double-headed arrows represent a significant latent correlation, and the grey dotted double-headed arrow represents a non-significant latent correlation.All factor loadings and correlations were significant (p < .001).

Figure 2 .
Figure 2. Structural models for Samples 1 and 2. Note.DEP = depression; ANX = anxiety; COM = concern over mistakes; DAA = doubts about actions; PS = personal standards.Ovals represent latent variables.The double-headed arrows represent a significant latent correlation and single-headed black arrows represent significant paths (p < .05).Gray dotted arrows represent nonsignificant paths (p > .05).Path coefficients are standardized.Factor loadings are omitted for clarity in the diagram, but are located in Figure 1.

Figure 3 .
Figure 3. Graphical representation of Sample 1's network model for depression, anxiety, and perfectionism.

Figure 4 .
Figure 4. Graphical representation of Sample 2's network model for depression, anxiety, and perfectionism.

Figure 6 .
Figure 6.Bridge centrality estimates for each node in Sample 2's network, ordered in descending order.

Figure 5 .
Figure 5. Bridge centrality estimates for each node in Sample 1's network, ordered in descending order.

Table 1 .
Descriptive statistics: means, standard deviations, and zero-order correlations of total scores for Samples 1 and 2 Note: Sample 1's N = 773 and Sample 2's N = 759, larger than the samples used in item-level analyses (SEM and network analysis) because scale averages were calculated by taking the average of all completed items.Note that in Sample 1, depressive and anxiety symptoms used a 1-5 scale, whereas Sample 2 used a 0-4 scale.Both samples used a 1-5 scale for other measures.***p<.001;**p<.01.2-tailed.