Advertisement

A Cross-Country Network Analysis of Adolescent Resilience

Published:September 10, 2020DOI:https://doi.org/10.1016/j.jadohealth.2020.07.010

      Abstract

      Purpose

      In situations of adversity, young people draw on individual, relational, and contextual (community and cultural) resources to foster their resilience. Recent literature defines resilience as a capacity that is underpinned by a network of interrelated resources. Although empirical studies show evidence of the value of a network approach, little is known regarding how different country contexts influence which resources are most critical within a resource network and how resources interact for adolescent resilience.

      Methods

      Network analysis was conducted with data from studies that had used the Child and Youth Resilience Measure. Regularized partial correlation networks of 17 resources were estimated for 14 countries (Botswana, Canada, China, Colombia, Equatorial Guinea, India, Indonesia, Italy, Jordan, New Zealand, the Philippines, Romania, South Africa, and Syrian refugees living in Jordan). The sample size was 18,914 (mean age = 15.70 years, 48.8% female).

      Results

      We observed mostly positive associations between the resources of interest. The salience and strength of associations between resources varied by country. The most central resource across countries was having supportive caregivers during stressful times because this resource had the most and strongest positive associations with other resources.

      Conclusions

      This study gives first empirical evidence from multiple countries that an interplay of social–ecological resources (such as individual skills, peer, caregiver and community support, and educational aspirations and opportunities) matter for adolescent resilience. Across countries, caregiver support appears to be most central for adolescent resilience. Future resilience interventions might apply this network approach to identify important, contextually relevant resources that likely foster additional resources.

      Keywords

      Implications and Contribution
      Resilience draws on a network of interrelated resources whose relations vary across countries. Caregiver support has been identified as the most important resource to have access to other meaningful resources across countries. Resilience interventions should aim at fostering such central resources in addition to the most protective resources.
      Resilience is defined as the capacity of individuals and collectives (such as families and communities) to navigate to and negotiate for meaningful social–ecological resources (e.g., caregivers, peer groups, or institutional supports) that can protect their well-being and development in times of stress [
      • Ungar M.
      • Theron L.
      Resilience and mental health: How multisystemic processes contribute to positive outcomes.
      ]. Resilience relies, therefore, on personal strengths as well as the resources that are provided by a facilitative environment [
      • Lerner R.M.
      Resilience as an attribute of the developmental system: Comments on the papers of Professors Masten & Wachs.
      ]. Which personal strengths and assets are available and used is influenced by the ecological context an individual lives in (such as cultural values or sociostructural characteristics of a community or country) [
      • Panter-Brick C.
      Culture and resilience: Next steps for theory and practice.
      ,
      • Liu J.J.
      • Reed M.
      • Girard T.A.
      Advancing resilience: An integrative, multi-system model of resilience.
      ]. Furthermore, these resources do not function in isolation. Resilience is promoted by dynamic processes of interaction between personal strengths and relevant ecological resources that help individuals and collectives to successfully deal with a specific stressor [
      • Lerner R.M.
      Resilience as an attribute of the developmental system: Comments on the papers of Professors Masten & Wachs.
      ,
      • Masten A.S.
      • Best K.M.
      • Garmezy N.
      Resilience and development: Contributions from the study of children who overcome adversity.
      ]. For example, child and youth resilience across diverse stressful contexts relies on a mixture of rearing practices by caregivers, social relationships and responsibilities, available protective institutions such as schools or places of worship, and personal strengths such as self-efficacy, cognitive skills, and competence [
      • Masten A.S.
      • Best K.M.
      • Garmezy N.
      Resilience and development: Contributions from the study of children who overcome adversity.
      ,
      • Masten A.S.
      Global perspectives on resilience in children and youth.
      ].
      This dynamic, ecological systems theory conceptualizes resilience as a network of mutually dependent, interacting resources from different systems such as psychological, social, institutional, cultural, and environmental [
      • Janssen M.
      • Ö Bodin
      • Anderies J.
      • et al.
      Toward a network perspective of the study of resilience in social-ecological systems.
      ,
      • Ungar M.
      Systemic resilience: Principles and processes for a science of change in contexts of adversity.
      ]. Resources are thought to influence each other in several ways. The accessibility and quality of resources can depend on other resources (e.g., caregivers with higher socioeconomic status can give their children access to higher quality education if it is provided by societal institutions). Resources can be positively related and reinforce each other (e.g., children learn social skills from their caregivers and peers; these social skills attract positive responses from caregivers and peers) or they can be negatively related and hinder each other (caregivers who deny children contact with peers or access to education). This network perspective adds novel characteristics: it draws attention to how interconnected a resource network is and which resources are most influential on other resources in a resilience network [
      • Fried E.I.
      • Eidhof M.B.
      • Palic S.
      • et al.
      Replicability and generalizability of posttraumatic stress disorder (PTSD) networks: A cross-cultural multisite study of PTSD symptoms in four trauma patient samples.
      ]. This may help to better design resilience interventions: from solely improving or sustaining protective resources to also targeting the resources that exert (the most) positive influences on other resources.
      With the emergence of network analysis (NA) [
      • Epskamp S.
      • Fried E.I.
      A tutorial on regularized partial correlation networks.
      ], recent studies have been able to examine not just resources but networks of resources. For example, a study with adolescents from the United Kingdom showed that experiences of early life adversity can lead to negative interactions between personal and ecological resources and associated negative mental health effects [
      • Fritz J.
      • Stochl J.
      • Fried E.I.
      • et al.
      Unravelling the complex nature of resilience factors and their changes between early and later adolescence.
      ]. Furthermore, a study with a population sample of university students in Belgium showed that psychological and familial resources are generally positively related [
      • Briganti G.
      • Linkowski P.
      Item and domain network structures of the Resilience Scale for Adults in 675 university students.
      ].
      How resources are related to each other, however, could be significantly influenced by specific social contexts, such as the community in which a young person grows up [
      • Ungar M.
      • Ghazinour M.
      • Richter J.
      Annual research review: What is resilience within the social ecology of human development?.
      ]. When communities prioritize specific resources over others, they influence resource availability and accessibility and potentially which resources matter more, or less, for resilience [
      • Panter-Brick C.
      Culture and resilience: Next steps for theory and practice.
      ,
      • Southwick S.M.
      • Bonanno G.A.
      • Masten A.S.
      • et al.
      Resilience definitions, theory, and challenges: Interdisciplinary perspectives.
      ]. Caution is, therefore, necessary when generalizing the importance of resources for the functioning of a resilience network within different ecological levels (across communities, countries, or cultures). Thus, although there are resources that are common across different cultures at different levels of stress exposure [
      • Ungar M.
      • Liebenberg L.
      Assessing resilience across cultures using mixed methods: Construction of the child and youth resilience measure.
      ,
      • Windle G.
      • Bennett K.M.
      • Noyes J.
      A methodological review of resilience measurement scales.
      ], the context has the potential to influence resilience networks by impacting the importance, availability, and interrelatedness of these resources [
      • Ungar M.
      • Ghazinour M.
      • Richter J.
      Annual research review: What is resilience within the social ecology of human development?.
      ].

      Meaningful Resources for Adolescent Resilience Across Countries and Contexts

      A social–ecological model of resilience also stresses that the developmental period influences which resources are available and used in stressful circumstances [
      • Ungar M.
      The social ecology of resilience: Addressing contextual and cultural ambiguity of a nascent construct.
      ]. A multicountry study identified resources that are meaningful to the resilience of adolescents across nations and for manifold stressful contexts such as war, poverty, dislocation, and marginalization [
      • Ungar M.
      • Liebenberg L.
      Assessing resilience across cultures using mixed methods: Construction of the child and youth resilience measure.
      ,
      • Windle G.
      • Bennett K.M.
      • Noyes J.
      A methodological review of resilience measurement scales.
      ]. This resulted in the culturally and contextually sensitive Child and Youth Resilience Measure (CYRM-28) that assesses resources on the individual (personal and social skills and peer support), caregiver (psychological and physical caregiving), and contextual levels (spirituality, education, and culture) [
      • Ungar M.
      • Liebenberg L.
      Assessing resilience across cultures using mixed methods: Construction of the child and youth resilience measure.
      ]. However, follow-up studies found that the measured resources might not operate equivalently across different countries [
      • Sanders J.
      • Munford R.
      • Thimasarn-Anwar T.
      • Liebenberg L.
      Validation of the child and youth resilience measure (CYRM-28) on a sample of at-risk New Zealand youth.
      ,
      • van Rensburg A.C.
      • Theron L.
      • Ungar M.
      Using the CYRM-28 with South African young people: A factor structure analysis.
      ,
      • Zand B.K.
      • Liebenberg L.
      • Shamloo Z.S.
      Validation of the factorial structure of the child and youth resilience measure for use with Iranian youth.
      ]. This indicates structural and processual differences in resilience networks between countries.

      The present study

      Few empirical studies have investigated the network of resources that support adolescent resilience or considered how specific country contexts influence the characteristics of resource networks. The present study aims to use NA to compare the same network of individual and social–ecological resources across different countries to identify country-specific as well as common characteristics of adolescent resilience networks (e.g., which resources are positively/negatively related and which resources have the most associations with others).

      Methods

      Measures

      The analysis was based on the CYRM-Revised (CYRM-R) [
      • Jefferies P.
      • McGarrigle L.
      • Ungar M.
      The CYRM-R: A Rasch-validated revision of the child and youth resilience measure.
      ], which is a 17-item Rasch-validated revision of the widely used CYRM-28 measure [
      • Ungar M.
      • Liebenberg L.
      Assessing resilience across cultures using mixed methods: Construction of the child and youth resilience measure.
      ]. The CYRM-R assesses cross-culturally meaningful resilience resources of young people for manifold stressful situations and consists of two subscales: intra/interpersonal resilience and caregiver resilience (Table S1 provides reliability statistics). Intra/interpersonal resilience relates to internal skills as well as resources of the environment, such as peers, school, or learning opportunities. Caregiver resilience relates to psychological and physical resources provided by caregivers, such as safety, nutrition, or support. The measure can be administered with a three- or five-point Likert-scale. The data analysis for the present study was based on the five-point Likert-scale. One data set for India with 62 participants (Table S1) had used the three-point Likert-scale and was linear transformed to a five-point Likert-scale (1 remained 1, 2 into 3, and 3 into 5). Higher scores indicate higher resilience.

      Procedure

      The CYRM-R is a free tool and can be accessed via the homepage of the Resilience Research Centre (www.resilienceresearch.org). To download the CYRM-R, users are asked to give their contact information. E-mail addresses were used to contact anyone who had accessed the CYRM up to mid-April 2019 and used it to study children and youth that live in stressful circumstances. Hence, the risk was implicitly included in the current analysis by the characteristics of each sample. Data sets from the same country were merged for ease of analysis, thereby operationalizing social context as a country-specific condition. By the time the data sets were collected, a total of 43 studies that have used the CYRM were published in scientific journals. This analysis included a total of 23 data sets from 16 different principal investigators from 14 countries, of which not all had published their data (Table S1 provides each sample's details).
      Each respective study that was included in this analysis was approved by the institutional review board of each study author's institution where the original study was conducted.

      Data analysis

      Individuals with missing data were excluded from the analysis (Table S2 provides characteristics of the included and excluded participants). The present analyses were performed using R version 3.6 in RStudio 1.2.1335 (R Core Team, Vienna, Austria).
      The goldbricker method was used to test for topologically overlapping items before the network models were estimated using networktools [
      • Jones P.J.
      Networktools: Assorted tools for identifying important nodes in networks. R Package version 1.1.2.
      ]. Overall, eight countries showed no overlapping items, whereas six countries showed country-specific overlapping items (see R-script for results). Hence, to be able to compare the countries, no items were merged.

      Network estimation

      NA shows significant conditional associations between all resources in a network [
      • Epskamp S.
      • Fried E.I.
      A tutorial on regularized partial correlation networks.
      ,
      • Costantini G.
      • Epskamp S.
      • Borsboom D.
      • et al.
      State of the aRt personality research: A tutorial on network analysis of personality data in R.
      ]. By controlling for the influence of all other resources, it indicates what unique association exists between two resources, how they are related (valence), and how strong this relation is (weight). Although a network model based on cross-sectional data (as in the present analysis) does not imply causal influence, it can serve to form hypotheses about causal relationships.
      Regularized partial correlation networks were estimated for each country using EBICglasso in bootnet [
      • Epskamp S.
      • Borsboom D.
      • Fried E.I.
      Estimating psychological networks and their accuracy: A tutorial paper.
      ]. The huge method was applied because multivariate normality was not given (Figure S1). The tuning parameter for EBIC (Extended Bayesian Information Criterion) was set to lead to the network with the highest possible specificity for each country. Networks were visualized using qgraph [
      • Epskamp S.
      • Cramer A.O.
      • Waldorp L.J.
      • et al.
      qgraph: Network visualizations of relationships in psychometric data.
      ].

      Network inference

      Expected influence (EI) centrality is related to strength centrality and indicates the importance of a resource for the structure and functioning of the network [
      • Robinaugh D.J.
      • Millner A.J.
      • McNally R.J.
      Identifying highly influential nodes in the complicated grief network.
      ]. It is based on the sum of all regularized partial correlations/edge weights of a resource to its connected resources. EI, therefore, indicates if a resource has an overall activating (more positive than negative connections) or deactivating (more negative than positive connections) influence on a network. Resources with a high negative or positive EI have a high importance for a network. A study by Epskamp et al [
      • Epskamp S.
      • Borsboom D.
      • Fried E.I.
      Estimating psychological networks and their accuracy: A tutorial paper.
      ] has shown that strength centrality (and thus EI) is the most reliable indicator of the commonly investigated centrality indices next to closeness and betweenness centrality. For this reason, this study focused on EI centrality. Analyses of the other centrality indices can be found in the Supplemental Material (Figure S2). Furthermore, the average predictability of a whole network was estimated to indicate how well a network can predict itself and how much it is dependent on unknown variables via mgm [
      • Haslbeck J.M.B.
      • Borsboom D.
      • Waldorp L.J.
      Moderated network models.
      ].

      Network comparison

      A moderated network model was estimated via glasso and EBIC to examine differences between two networks by analyzing how many edge weights significantly differed between each pair of country networks [
      • Haslbeck J.M.B.
      • Borsboom D.
      • Waldorp L.J.
      Moderated network models.
      ]. To get an insight into how similar the countries were in the importance of each resource for the interrelatedness and functioning of the network, correlations were used to compare their EI profiles [
      • Fried E.I.
      • Eidhof M.B.
      • Palic S.
      • et al.
      Replicability and generalizability of posttraumatic stress disorder (PTSD) networks: A cross-cultural multisite study of PTSD symptoms in four trauma patient samples.
      ]. Also, an average network was estimated by averaging the edge weights and node EI centralities across all country-specific networks [
      • Fried E.I.
      • Eidhof M.B.
      • Palic S.
      • et al.
      Replicability and generalizability of posttraumatic stress disorder (PTSD) networks: A cross-cultural multisite study of PTSD symptoms in four trauma patient samples.
      ]. In addition, a variability network across the included countries was estimated where each edge and node represent their standard deviation (SD) in edge weight and EI across all networks [
      • Fried E.I.
      • Eidhof M.B.
      • Palic S.
      • et al.
      Replicability and generalizability of posttraumatic stress disorder (PTSD) networks: A cross-cultural multisite study of PTSD symptoms in four trauma patient samples.
      ]. This gives an insight into where the included samples show the most differences and similarities.

      Network stability

      To estimate the accuracy of the edge weights, 95% bootstrapped confidence intervals were derived (for results, see Figure S3) [
      • Epskamp S.
      • Borsboom D.
      • Fried E.I.
      Estimating psychological networks and their accuracy: A tutorial paper.
      ]. Second, the stability of EI was indicated via case-dropping subset bootstraps and the correlation stability coefficient (for results, see Figure S4). These measures indicate if the order of the EI of the resources remains the same in a network that is based on a smaller sample. The correlation between the original sample and a subset should preferably be above .5 and at least .25 [
      • Epskamp S.
      • Borsboom D.
      • Fried E.I.
      Estimating psychological networks and their accuracy: A tutorial paper.
      ]. The EBIC tuning parameter needed to be lowered for the six countries with the smallest sample sizes to conduct this analysis (see R-script). These analyses were conducted using bootnet [
      • Epskamp S.
      • Borsboom D.
      • Fried E.I.
      Estimating psychological networks and their accuracy: A tutorial paper.
      ].
      In summary, NA results mainly in descriptive metrics that are used to describe the characteristics of a network. The network estimation identifies edges that are meaningful for the network independent of how much they differ from each other.

      Results

      Descriptive statistics

      The overall sample consisted of 18,914 participants, with 48.8% of females and a mean age of 15.70 years (SD = 3.37). There was a large variance in the sample sizes for the 14 countries, ranging from n = 200 for Romania to n = 3,868 for Italy (Table 1 and Table S1 provide more details on included data sets). Most samples had a lower percentage of females. The country that scored highest on resilience-enabling resources was Botswana, followed by Colombia and India. The countries with the lowest resources were Equatorial Guinea, Italy, and Indonesia. Botswana, Colombia, Equatorial Guinea, and Italy showed significant differences to all other countries in relation to their overall resource level, whereas the other countries showed similar levels.
      Table 1Sample characteristics
      CountryNAge, M (SD) [Range]Gender (%♀)Intra/interpersonal resilience, M (SD)Caregiver resilience, M (SD)CYRM-R total, M (SD)
      1. Botswana51114.88 (1.14) [12–19]4445.81 (4.91)32.13 (3.69)77.9 (7.60)
      2. Canada3,09015.47 (2.47) [11–25]51.941.68 (5.96)28.92 (5.81)70.61 (10.39)6,9
      3. China1,42516.09 (1.62) [13–19]57.440.58 (6.13)28.87 (5.67)69.45 (10.08)9,12,13
      4. Colombia1,31515.82 (1.84) [13–19]50.142.99 (4.64)30.18 (4.40)73.18 (7.51)
      5. Equatorial Guinea35312.54 (1.95) [10–16]53.837.35 (5.87)24.86 (4.63)62.21 (9.10)
      6. India96615.12 (1.21) [12–19]50.441.37 (5.21)29.48 (4.64)70.86 (8.61)2,9
      7. Indonesia80110.88 (0.89) [10–14]47.838.44 (7.44)27.9 (5.47)66.33 (12.03)11,14
      8. Italy3,86816.68 (1.44) [14–21]41.337.45 (5.67)26.5 (5.49)63.96 (9.77)
      9. Jordan27214.32 (1.62) [11–19]41.240.71 (6.30)29.82 (4.27)70.53 (9.32)2,3,6
      10. New Zealand1,18615.08 (1.23) [12–17]39.240.26 (5.84)27.23 (6.00)67.49 (9.79)11,12,14
      11. The Philippines82519.97 (1.03) [17–23]63.540.42 (4.83)26.82 (4.36)67.24 (8.21)7,10,12,14
      12. Romania20012.54 (1.59) [9–17]46.340.57 (6.25)28.16 (4.35)68.72 (9.89)3,10,11,13,14
      13. South Africa3,80715.60 (2.21) [11–26]53.439.68 (7.06)29.55 (5.48)69.23 (11.10)3,12
      14. Syrian refugees living in Jordan (Syria)29514.06 (1.89) [10–19]40.338.39 (6.77)28.92 (4.33)67.31 (9.67)7,10,11,12
      Total18,91415.70 (3.37)48.840.08 (6.35)28.44 (5.53)68.52 (10.47)
      CYRM-R = Child and Youth Resilience Measure-Revised; M = mean, SD = standard deviation.
      Superscript numbers refer to the country with nonsignificant differences in CYRM-R total score (p > .05).
      As can be seen in Table 2, the most commonly reported available resource on average was I2 (“Getting an education is important to me,” M = 4.47,) and the least commonly reported available resource was C4 (“I talk to my caregiver(s) about how I feel,” M = 3.36). Country-specific distributions of resources were also present (Table S3), and the resource profiles of two countries correlated on average by r = .44 (Table S4).
      Table 2CYRM-R item characteristics across countries
      Short nameItemMScore (SD)MEIHighest score per countryHighest centrality per country
      I1I cooperate with people around me3.95 (1.02)−.05
      I2Getting an education is important to me4.47 (.90)−1.18BWA, COL, GNQ, IND, PHL, ZAF
      I3I know how to behave in different social situations4.12 (1.01)−.85
      I4People think I am fun to be with3.92 (1.06)−1.07
      I5I feel supported by my friends3.93 (1.13)1.03
      I6I feel I belong at my school3.80 (1.23).20
      I7My friends stand by me during difficult times3.87 (1.17)1.13JOR, PHL, SYR
      I8I am treated fairly in my community3.88 (1.15).08ROUIDN
      I9I am given opportunities to show others that I am becoming an adult and can act responsibly3.98 (1.07)−.06
      I10I have opportunities to develop skills that will be useful later in life (like job skills and skills to care for others)4.15 (1.02).18
      C1My caregiver(s) watch me closely4.07 (1.15)−.71JOR
      C2My caregiver(s) know a lot about me3.90 (1.23).49
      C3If I am hungry, there is enough to eat4.38 (1.00)−1.68CAN, ITA
      C4I talk to my caregiver(s) about how I feel3.36 (1.44)−.54
      C5My caregiver(s) stand(s) by me during difficult times4.20 (1.12)2.21BWA, CAN, CHN, GNQ, IND, ITA, ROU, ZAF
      C6I feel safe when I am with my caregiver(s)4.44 (.96)1.15CHN, IDN, NZL, SYRCOL
      C7I enjoy the family traditions of my caregiver(s)4.09 (1.17).14
      I = items of the intra/interpersonal resilience subscale; C = items of the caregiver resilience subscale; MScore = average item score; SD = standard deviation of average item score; MEI = average standardized expected influence; BWA = Botswana; CAN = Canada; CHN = China; COL = Colombia; GNQ = Equatorial Guinea; IND = India; IDN = Indonesia; ITA = Italy; JOR = Jordan; NZL = New Zealand; PHL = The Philippines; ROU = Romania; SYR = Syria; ZAF = South Africa.

      Cross-country network of resilience

      Edge weight

      Figure 1A shows that all edges were positive on average, and the strongest association (edge weight = .37) was found between “I feel supported by my friends” (I5) and “My friends stand by me during difficult times” (I7). The second strongest association (edge weight = .23) was found between “My caregiver(s) stand(s) by me during difficult times” (C5) and “I feel safe when I am with my caregiver(s)” (C6).
      Figure thumbnail gr1
      Figure 1Cross-country analyses. Note. (A) Cross-country regularized partial correlation network. White nodes: intra/interpersonal subscale of CYRM-R. Gray nodes: caregiver subscale of CYRM-R. Item wordings in . Solid edges indicate positive relations, and dashed edges negative relations. The thicker the edge, the larger the edge weight and stronger the unique association between two resources. (B) Expected influence centrality profiles. Black line represents cross-cultural network (black line). Gray lines represent the countries. Higher positive values indicate higher expected influence centrality. (C) Cross-country variability network. Edge thickness indicates the standard deviation of the edge weights across all countries. Node size indicates the standard deviation of the EI centralities across all countries. The thicker the edge/node, the higher its standard deviation and variation across countries.

      Expected influence

      On average, the resources with the highest EI were C5 (“My caregiver(s) stand(s) by me during difficult times”), C6 (“I feel safe when I am with my caregiver(s)”), and I7 (“My friends stand by me during difficult times”) with standardized EI estimates of 1.41, .75, and .68, respectively (Figure 1B). The resources C3 (“If I am hungry, there is enough to eat”), I2 (“Getting an education is important to me”), and I4 (“People think I am fun to be with”) showed the lowest standardized EI with estimates of −1.03, −.77, and −.69.
      The average predictability of the entire network was R2 = .26.

      Country-specific resilience networks

      Edge weight

      Figure 2 indicates that the edge connecting I5 (“I feel supported by my friends”) and I7 (“My friends stand by me during difficult times”) showed a moderate to strong connection for most networks except for Romania where no connection was estimated and for Jordan and Indonesia where rather weak connections were estimated. Also, an edge that showed a relatively moderate connection in all countries apart from Equatorial Guinea was the one between C5 (“My caregiver(s) stand(s) by me during difficult times”) and C6 (“I feel safe when I am with my caregiver(s)”). An edge that showed one of the lowest connections for all countries was the one between C3 (“If I am hungry, there is enough to eat”) and C4 (“I talk to my caregiver(s) about how I feel”), whereas an edge that showed a high variability across countries was the one connecting I2 (“Getting an education is important to me”) and I6 (I feel I belong at my school). These resources were connected in all countries but weakly in South Africa and India, moderately in Botswana and Canada, and strongly in Jordan, China, and New Zealand.
      Figure thumbnail gr2
      Figure 2Country-specific regularized partial correlation networks. Note. White nodes: intra/interpersonal subscale of CYRM-R. Gray nodes: caregiver subscale of CYRM-R. Item wordings in . Solid edges indicate positive relations, and dashed edges negative relations. The thicker the edge, the larger the edge weight and stronger the unique association between two resources. See for taller figures.
      All networks showed significant differences in their edge weights, and thus, no two countries showed the same network (Table 3). On average, 15.01% of edges significantly differed between the countries. South Africa showed the most different edge weights to all other countries (M = 23.38%) followed by Italy (M = 20.00%), and Equatorial Guinea (M = 17.23%), whereas Syria (M = 7.92%), Jordan (M = 9.54%), and Romania (M = 10.15%) showed the least. Figure 1C gives an overview of how much the edge weight of each edge varied across all countries. The edges that differed the most between the countries were between I5 (“I feel supported by my friends”) and I7 (“My friends stand by me during difficult times”; SD = .16), between I2 (“Getting an education is important to me”) and I6 (“I feel I belong at my school”; SD = .13), and between C2 (“My caregiver(s) know a lot about me”) and C4 (“I talk to my caregiver(s) about how I feel”; SD = .11).
      Table 3Percentage of significantly different edge weights between countries
      BWACANCHNCOLGNQINDIDNITAJORNZLPHLROUZAFSYR
      BWA24151098192111181272210
      CAN101315142925771115324
      CHN1013132129410129264
      COL11122630121719133213
      GNQ1228321026225329
      IND24317172372912
      IDN2451574127
      ITA5111510243
      JOR111510243
      NZL1513325
      PHL8821
      ROU256
      ZAF6
      SYR
      Numbers indicate how many edges significantly differ between two countries in percent based on moderated network models.
      BWA = Botswana; CAN = Canada; CHN = China; COL = Colombia; GNQ = Equatorial Guinea; IND = India; IDN = Indonesia; ITA = Italy; JOR = Jordan; NZL = New Zealand; PHL = The Philippines; ROU = Romania; SYR = Syria; ZAF = South Africa.

      Expected influence

      Figure 1B indicates that the resources differed in their EI between countries, whereas others were also similar (Figure S3 shows the country-specific standardized and unstandardized centrality profiles). The countries with the most similar EI profiles were Canada and the Philippines (r = .79), India and China (r = .77), and New Zealand and Colombia (r = .75); whereas South Africa and Jordan (r = −.30), New Zealand and Jordan (r = −.27), and Colombia and Jordan (r = −.20) showed the most differences (Table S5).
      Although C5 (“My caregiver(s) stand(s) by me during difficult times”) was found to be the most central resource for most countries, for Philippines, Romania, and Syria, it was I7 (“My friends stand by me during difficult times”), for Indonesia I8 (“I am treated fairly in my community”), for Colombia C6 (“I feel I belong at my school”), and for Jordan I6 (“I feel I belong at my school”). Also, the least central resource for Equatorial Guinea and India was I4 (“People think I am fun to be with”), for Jordan and Romania I2 (“Getting an education is important to me”), for Indonesia and South Africa C4 (“I talk to my caregiver(s) about how I feel”), and for Botswana and Equatorial Guinea C1 (“My caregiver(s) watch me closely”), whereas C3 (“If I am hungry, there is enough to eat”) was the least central resource for the remaining countries. Figure 1C gives an overview of how much each node's EI varied across all countries. The resources with the highest EI SDs were C1 (“My caregiver(s) watch me closely,” 1.15), C3 (“If I am hungry, there is enough to eat,” .97), and C6 (“I feel safe when I am with my caregiver(s),” .88).
      The average predictability ranged from .08 for Equatorial Guinea to .38 for Canada (Figure S2 provides average predictability of all countries).

      Stability analyses

      Most networks showed a sufficient stability of their EI (above .5). Botswana, India, and Jordan showed acceptable stability above .25. However, Equatorial Guinea showed an insufficient stability of .21.

      Discussion

      Using data from multiple administrations of the Child and Youth Resilience Measure (CYRM) in 14 countries, this study investigated commonalities and differences in adolescent's resilience networks. No two countries showed the exact same network. This contributes further evidence for the significant influence of the environment on resilience [

      Rutter, M. Resilience: Some conceptual considerations. J Adolesc Health 1993;14:626–631.

      ]. Most networks were characterized by mostly positively interrelated resources. Some countries showed a high similarity in their resource centrality profiles, whereas others tended to show opposite profiles. In addition, resources were identified that were strongly connected in the resilience network of most countries such as having caregivers or friends that give support during stressful times, although there were also country-specific patterns. Finally, this study showed that the rank order of resources according to their score must not be equal to their rank order according to their connectedness in a network.

      Resilience as a network

      The following interpretation of the results is mainly based on resource associations and the expected influence centrality of resources. Currently, there is no straightforward interpretation of centrality that is based on cross-sectional data. This makes it necessary to tailor the interpretation of the studied network to the research question and to choose a hypothetical direction of influence [
      • Bringmann L.F.
      • Elmer T.
      • Epskamp S.
      • et al.
      What do centrality measures measure in psychological networks?.
      ]. Hence, based on a social–ecological definition of resilience [
      • Ungar M.
      The social ecology of resilience: Addressing contextual and cultural ambiguity of a nascent construct.
      ], central resources are expected to significantly influence how much other related resources will be available, even if the direction of these network associations has yet to be fully assessed.
      The study provides the first multicountry empirical evidence that adolescent resilience can be operationalized as a network of interconnected multisystemic resources. This adds new and important evidence to the literature that stresses that the accessibility of a resource could depend on other resources, from the same and/or different ecological systems. The general positive associations that have been identified between the studied resources of the CYRM indicate that social–ecological resources condition each other: they either facilitate or limit each other. This speaks in favor of the selected resources that are assessed by the CYRM because one does not need to choose between associated resources in the prospect of losing another.
      However, the studied CYRM-R captures only a limited number of the social–ecological resources found to be important for an individual's resilience [
      • Ungar M.
      • Liebenberg L.
      Assessing resilience across cultures using mixed methods: Construction of the child and youth resilience measure.
      ]. The average predictability of each country's network shows that resilience relies on more resources. For example, socioeconomic resources such as governmental cash transfers to families living in poverty have been found to exert resilience-enhancing effects on children [
      • Cluver L.D.
      • Orkin F.M.
      • Campeau L.
      • et al.
      Improving lives by accelerating progress towards the UN sustainable development goals for adolescents living with HIV: A prospective cohort study.
      ]. Sufficient financial resources could play a crucial role in navigating to other social–ecological resources, such as schools, caregiver time, and nutrition. Such additional resources could, however, also have negative associations with the studied resources because acquaintance with some resources might come at the expense of other resources.

      Centrality of resilience resources and resource associations across countries

      The study showed that there is a difference between the distribution of resources as indicated by their score on the CYRM and the centrality of resources. The resources with the highest scores were not necessarily found to be the resources with the highest centrality. The same pattern was found for resources with lesser scores and centralities across countries. This indicates that a highly available resource must not be the resource that gives access to relatively many other resources. This finding has important practical implications: interventions should focus on strengthening or sustaining resources that might not be important for the functioning of a resource network but nevertheless be fundamentally important for one's resilience and well-being (e.g., having enough to eat or valuing education showed the highest levels on average but were the least central resources) and focus on strengthening/sustaining central resources to quickly increase the resource portfolio of a person.
      Across countries, the support by caregiver(s) and friends during difficult times as well as feeling safe when a child is with its caregiver(s) were found to be the most central resources. Social support is known to be an important global resource for child and youth resilience [
      • Armstrong M.I.
      • Birnie-Lefcovitch S.
      • Ungar M.
      Pathways between social support, family well being, quality of parenting, and child resilience: What we know.
      ], whereas the literature on attachment shows the importance of a secure attachment to a caregiver for a child's resilience [
      • Riley J.R.
      • Masten A.S.
      Resilience in context.
      ,
      • Selcuk E.
      • Zayas V.
      • Günaydin G.
      • et al.
      Mental representations of attachment figures facilitate recovery following upsetting autobiographical memory recall.
      ]. However, although such former studies have usually shown that robust social support leads to better outcomes in stressful situations [
      • Werner E.E.
      The children of Kauai: Resiliency and recovery in adolescence and adulthood.
      ], the present study shows that social support from caregiver and friends is key for adolescent resilience, likely because it offers access to other resources.
      Besides these common central resources, the analysis also showed country-specific centralities. For example, being treated fairly by one's community was the resource with the highest centrality in Indonesia. This could indicate that the availability of several resilience resources is strongly influenced by one's community in this country. Furthermore, the importance of getting an education was the second most central resource for India and South Africa, which could reflect the usefulness of education for making further resources available in those countries. For example, studies in South Africa have shown that education is culturally valued, more particularly in contexts of structural disadvantage where education is strongly associated with upward life trajectories [
      • van Breda A.D.
      • Theron L.
      A critical review of South African child and youth resilience studies, 2009–2017.
      ]. These and the other country-specific resource centralities reinforce for the importance of a socioecological perspective on resilience [
      • Ungar M.
      The social ecology of resilience: Addressing contextual and cultural ambiguity of a nascent construct.
      ].
      Cross-country and country-specific resource associations were also identified. A strong interdependency between the feeling of belonging at one's school and valuing education was found for some countries (e.g., Jordan, China, and New Zealand), whereas in other countries, these resources were almost independent of each other (e.g., South Africa, Romania, and India). This country-specific, nonexistent association might be explained by reports of schools being rather hostile places in more disadvantaged communities [
      • Romero R.H.
      • Hall J.
      • Cluver L.
      Exposure to violence, teacher support, and school delay amongst adolescents in South Africa.
      ]. Furthermore, the connection between feeling supported by one's friends and the certainty that friends help in difficult times showed the strongest variability across countries. Although this connection would be expected to be strong across all countries, for two countries, it was rather low (Jordan) or unconnected (Romania). This could point to the fact that different countries have different reference points for what support and difficult times mean.

      Conclusion

      This study adds to the literature by contributing a more nuanced understanding of adolescent resilience as a complex network of interacting resources that shows country-specific characteristics. It demonstrates that it is worthwhile endeavor to go behind the usual practice of using total scores of instruments that assess different resources to get a deeper understanding of the processes that constitute resilience. NA is a promising means to further existing understandings of resilience; in combination with longitudinal designs, this tool promises new and significant information for general as well as context-specific resilience interventions.

      Funding Sources

      The position of J.H. was funded by the Swiss National Science Foundation (grant number P2ZHP1_184004 ).

      References

        • Ungar M.
        • Theron L.
        Resilience and mental health: How multisystemic processes contribute to positive outcomes.
        Lancet Psychiat. 2019; 7: 441-448
        • Lerner R.M.
        Resilience as an attribute of the developmental system: Comments on the papers of Professors Masten & Wachs.
        Ann NY Acad Sci. 2006; 1094: 40-51
        • Panter-Brick C.
        Culture and resilience: Next steps for theory and practice.
        in: Theron L. Liebenberg L. Ungar M. Youth Resilience and Culture: Commonalities and Complexities. Springer, Dordrecht, NL2015: 233-244
        • Liu J.J.
        • Reed M.
        • Girard T.A.
        Advancing resilience: An integrative, multi-system model of resilience.
        Pers Indiv Differ. 2017; 111: 111-118
        • Masten A.S.
        • Best K.M.
        • Garmezy N.
        Resilience and development: Contributions from the study of children who overcome adversity.
        Dev Psychopathol. 1990; 2: 425-444
        • Masten A.S.
        Global perspectives on resilience in children and youth.
        Child Dev. 2014; 85: 6-20
        • Janssen M.
        • Ö Bodin
        • Anderies J.
        • et al.
        Toward a network perspective of the study of resilience in social-ecological systems.
        Ecol Soc. 2006; 11: 15
        • Ungar M.
        Systemic resilience: Principles and processes for a science of change in contexts of adversity.
        Ecol Soc. 2018; 23: 34
        • Fried E.I.
        • Eidhof M.B.
        • Palic S.
        • et al.
        Replicability and generalizability of posttraumatic stress disorder (PTSD) networks: A cross-cultural multisite study of PTSD symptoms in four trauma patient samples.
        Clin Psychol Sci. 2018; 6: 335-351
        • Epskamp S.
        • Fried E.I.
        A tutorial on regularized partial correlation networks.
        Psychol Methods. 2018; 23: 617-634
        • Fritz J.
        • Stochl J.
        • Fried E.I.
        • et al.
        Unravelling the complex nature of resilience factors and their changes between early and later adolescence.
        BMC Med. 2019; 17: 203
        • Briganti G.
        • Linkowski P.
        Item and domain network structures of the Resilience Scale for Adults in 675 university students.
        Epidemiol Psych Sci. 2020; 29: e33
        • Ungar M.
        • Ghazinour M.
        • Richter J.
        Annual research review: What is resilience within the social ecology of human development?.
        J Child Psychol Psych. 2013; 54: 348-366
        • Southwick S.M.
        • Bonanno G.A.
        • Masten A.S.
        • et al.
        Resilience definitions, theory, and challenges: Interdisciplinary perspectives.
        Eur J Psychotraumato. 2014; 5: 25338
        • Ungar M.
        • Liebenberg L.
        Assessing resilience across cultures using mixed methods: Construction of the child and youth resilience measure.
        J Mix Methods Res. 2011; 5: 126-149
        • Windle G.
        • Bennett K.M.
        • Noyes J.
        A methodological review of resilience measurement scales.
        Health Qual Life Out. 2011; 9: 8
        • Ungar M.
        The social ecology of resilience: Addressing contextual and cultural ambiguity of a nascent construct.
        Am J Orthopsychiat. 2011; 81: 1-17
        • Sanders J.
        • Munford R.
        • Thimasarn-Anwar T.
        • Liebenberg L.
        Validation of the child and youth resilience measure (CYRM-28) on a sample of at-risk New Zealand youth.
        Res Social Work Prac. 2017; 27: 827-840
        • van Rensburg A.C.
        • Theron L.
        • Ungar M.
        Using the CYRM-28 with South African young people: A factor structure analysis.
        Res Social Work Prac. 2019; 29: 93-102
        • Zand B.K.
        • Liebenberg L.
        • Shamloo Z.S.
        Validation of the factorial structure of the child and youth resilience measure for use with Iranian youth.
        Child Indic Res. 2017; 10: 797-809
        • Jefferies P.
        • McGarrigle L.
        • Ungar M.
        The CYRM-R: A Rasch-validated revision of the child and youth resilience measure.
        J Evid Inf Soc Work. 2018; 16: 70-92
        • Jones P.J.
        Networktools: Assorted tools for identifying important nodes in networks. R Package version 1.1.2.
        2017 (Available at:) (Accessed August 12, 2020)
        • Costantini G.
        • Epskamp S.
        • Borsboom D.
        • et al.
        State of the aRt personality research: A tutorial on network analysis of personality data in R.
        J Res Pers. 2015; 54: 13-29
        • Epskamp S.
        • Borsboom D.
        • Fried E.I.
        Estimating psychological networks and their accuracy: A tutorial paper.
        Behav Res Methods. 2017; 50: 195-212
        • Epskamp S.
        • Cramer A.O.
        • Waldorp L.J.
        • et al.
        qgraph: Network visualizations of relationships in psychometric data.
        J Stat Softw. 2012; 48: 1-18
        • Robinaugh D.J.
        • Millner A.J.
        • McNally R.J.
        Identifying highly influential nodes in the complicated grief network.
        J Abnorm Psychol. 2016; 125: 747-757
        • Haslbeck J.M.B.
        • Borsboom D.
        • Waldorp L.J.
        Moderated network models.
        Multivar Behav Res. 2019; https://doi.org/10.1080/00273171.2019.1677207
      1. Rutter, M. Resilience: Some conceptual considerations. J Adolesc Health 1993;14:626–631.

        • Bringmann L.F.
        • Elmer T.
        • Epskamp S.
        • et al.
        What do centrality measures measure in psychological networks?.
        J Abnorm Psychol. 2019; 128: 892-903
        • Cluver L.D.
        • Orkin F.M.
        • Campeau L.
        • et al.
        Improving lives by accelerating progress towards the UN sustainable development goals for adolescents living with HIV: A prospective cohort study.
        Lancet Child Adolesc Health. 2019; 3: 245-254
        • Armstrong M.I.
        • Birnie-Lefcovitch S.
        • Ungar M.
        Pathways between social support, family well being, quality of parenting, and child resilience: What we know.
        J Child Fam Stud. 2005; 14: 269-281
        • Riley J.R.
        • Masten A.S.
        Resilience in context.
        in: Peters R. Leadbeater B. McMahon R.J. Resilience in Children, Families, and Communities. Linking Context to Practice and Policy. Springer, Boston, MA2005: 13-25
        • Selcuk E.
        • Zayas V.
        • Günaydin G.
        • et al.
        Mental representations of attachment figures facilitate recovery following upsetting autobiographical memory recall.
        J Pers Soc Psychol. 2012; 103: 362-378
        • Werner E.E.
        The children of Kauai: Resiliency and recovery in adolescence and adulthood.
        J Adolesc Health. 1992; 13: 262-268
        • van Breda A.D.
        • Theron L.
        A critical review of South African child and youth resilience studies, 2009–2017.
        Child Youth Serv Rev. 2018; 91: 237-247
        • Romero R.H.
        • Hall J.
        • Cluver L.
        Exposure to violence, teacher support, and school delay amongst adolescents in South Africa.
        Br J Educ Psychol. 2019; 89: 1-21