The current study drew upon data from the “School Absence in Finland” project.

Youth were recruited via 15 Finnish schools agreeing to participate in the project. Schools were recruited via numerous channels: (1) the webpages and email lists of organizations for teachers and principals; (2) teacher organizations' social media channels; and (3) various Facebook groups for school professionals. All channels shared information about the study and invited school personnel to participate in an online seminar in which the project was presented. The online seminar was conducted in January 2021, comprising presentations about school absence and information about the project. The 15 participating schools were located in southern and western Finland. Participating schools emailed parents with information about the project and requested consent for their child's participation. The parents of youth below 15 years were asked to complete an informed consent for their child's participation in the study. In addition, youth of all ages provided informed written consent at the time of data gathering.

Survey data were collected from youth during school time in May 2021. A research assistant or teacher was present in class during data collection to answer any questions and provide technical assistance if needed. To obtain data from persistently absent youth, school personnel engaged in individual outreach—contacting absent youth directly, visiting their homes if necessary, and inviting them to complete the questionnaire. The responses of 42 youth were collected in this way.

Data were initially collected from 2,137 youth, through both school-based administration of the questionnaire and individual outreach. Of these, 271 cases (12.7%) were excluded principally due to lack of consent, implausible age, or excessive missing data. From the remaining 1,866 respondents, we selected a subsample of 370 youth (19.8%) who self-reported school absence of ≥10% over the past 12 school weeks. This subsample was inspected for patterns of missing data, which were imputed where possible (see Data Analysis section for details). In the present analyses, we focused on youth with ≥10% school absence. In an earlier study using the full sample, where “days absent” was included as a separate node, the overall network structure was highly similar (32). In the current models, we did not include “days absent” as a node, because the questionnaire items were explicitly framed in relation to school absence, making its inclusion redundant. Importantly, the similarity of the network structures across the full and high-absence samples suggests that the associations identified are highly similar. An additional item in the ISAP asked youth to estimate how many days they had been absent from school during the past 12 school weeks. The researchers did not ask participants to distinguish between authorized and unauthorized absences. Response options were: 1 = Not at all, 2 = Sometimes (up to 4 school days missed), 3 = Often (5–12 school days missed), 4 = Very often (13–36 school days absent), 5 = Most of the time (37–48 days missed), and 6 = (Almost Always (more than 48 days missed). Responses ≥3 were selected. The included options correspond to absence ≥8.3% of the previous 12 weeks (where 12 weeks corresponds to 60 school days).

The average age of participants in the final sample was 14.91 years (SD = 0.84, range 12–17). Participant gender was male (40%), female (57%), or other (3%). According to the adolescents, their parent's highest educational level was university or higher education (51%), upper secondary school or vocational studies (25%), comprehensive school or equivalent (4%), with the rest of the youth either not knowing their parents' educational level or indicating none of the above (20%). Adolescents lived with both parents (76%), interchangeably with two parents (14%), with one parent (10%), or in residential childcare (under 1%).

The study was approved by the research ethics committee of Åbo Akademi University.

To measure symptoms and their function in relation to absence, participants were administered the Inventory of School Attendance Problems [ISAP; (30)]. This 48-item questionnaire was constructed as a screening tool for identifying problems linked to emerging and existing SAPs. The ISAP contains both a symptom scale (ISAP-S) and a function scale (ISAP-F). For each item (e.g., “Before or at school, I feel down or depressed”; “Before or at school, I'm afraid of exams”), the respondent is asked to report: (a) the extent to which the item applies to them (i.e., how true it is for them or how often it occurs); and (b) the extent to which it is why they miss school or find it hard to attend school (i.e., how true it is for them or how often this is the reason). Both questions are answered on a 4-point Likert scale (from “not true at all”/“never” to “very much true”/“very often”). Responses to part (a) of each item are used to form the symptom factors (ISAP-S), and responses to part (b) of each item are used to form the function factors (ISAP-F).

The 48 items of the ISAP reflect 13 factors. The items associated with each factor can be found in Appendix 1, and the questionnaire can be found at INSA.network. The 13 factors relate to the following subscales (reliability estimates in the current sample for each symptom subscale reported in parentheses): (1) Depression (α = .89, ω_t = .95), (2) Social Anxiety (α = .85, ω_t = .88), (3) Separation Anxiety (α = .82, ω_t = .86), (4) Performance Anxiety (α = .85, ω_t = .85), (5) Agoraphobia/Panic (α = .82, ω_t = .87), (6) Somatic Complaints (α = .71, ω_t = .81), (7) School Aversion/Attractive Alternatives (α = .77, ω_t = .81), (8) Aggression (α = .82, ω_t = .83), (9) Problems with Peers (α = .81, ω_t = .86), (10) Problems with Teachers (α = .77, ω_t = .78), (11) Dislike of the Specific School (α = .86, ω_t = .86), (12) Problems Within the Family (α = .90, ω_t = .90), and (13) Problems with Parents (α = .84, ω_t = .86).

The internal consistency of the subscales was deemed to be adequate based on administration with the original sample [.75 ≤ α ≤ .88, 3 testlets/scale; (30)] and the current sample (.71 ≤ α ≤ .90,.72 ≤ ω_t ≤ .95).

Materials and analysis code for this study are available upon request.

The large majority of respondents described themselves as being absent frequently: 80% selected “often,” 15% “very often,” 3% “mostly,” and 2% “always.”

For each of the 13 ISAP-S scales and 13 ISAP-F scales, sum score variables were calculated for the items within a given scale, and these sum score variables were used in the network model. Means and standard deviations for each sum score variable are displayed in Table 1. Visual inspection indicates that the mean values for the symptom scales were consistently higher than the mean values for the equivalent function scales. For example, the mean symptom score for the six items related to Depression was higher than the mean function score for the same six items. Across all 13 symptom scales, the scale with the highest mean score was Depression. The same is true for the 13 function scales.

To obtain a clear understanding of elevated scores in the subscales, we also report descriptive statistics of the proportion of participants who scored higher than a cut-off value of 1.5 on the ISAP subscales, consistent with the definition employed for descriptive purposes (i.e., mean scale value above or equal to half of the maximum of 3) in the original study conducted by Knollmann et al. (30). Of the youth who scored ≥1.5 on any of the subscales, symptoms of School Aversion (42%) and Depression (25%) were the most common, whereas functions of School Aversion and Somatic Complaints (both reported by 14%), followed by Depression (11%) were the most common (see Table 1).

We estimated a network structure based on the 13 sum score variables (nodes) for the symptom scales, yielding partial correlations (edges) between the nodes. All reported correlations are partial correlations. There were 31 edges with non-zero weight. (See also Table 2 for partial correlation coefficients.) Some nodes were more connected than others, as can be seen in Figure 1. The redundancy analysis revealed several overlapping nodes. For the symptoms network, Problems with Teachers and Aggression, Dislike of Specific School and Problems with Teachers, Agoraphobia and Social Anxiety, Problems with Peers and Social Anxiety, and Dislike of Specific School and Aggression shared significant overlap. These overlaps indicate that these combinations of nodes may measure the same underlying construct. However, we chose not to combine the overlapping variables to keep the results comparable with other studies using the ISAP questionnaire. The symptoms network showed high stability, with CS-coefficients of .673 and .593 for edges and strength centrality, respectively. The strongest edge was found between Problems within Family and Problems with Parents, with a partial correlation of .37 (see Table 2). Other strong edges were found for Social Anxiety with Peer Problems (r = .24), and with Agoraphobia/Panic (r = .25) and Depression (r = .25); and for Depression with Performance Anxiety (r = .26) and with Aggression (r = .28).

Figure 2 shows the centrality estimate strength. The strength measures the sum of all absolute partial correlations that are associated with the edge. The strength is also a measure of centrality in the model. Depression is the most central node according to this measure, while Agoraphobia/Panic was the next most central node in the network, although they did not differ in strength based on the bootstrapped significance test (bootstrapped confidence interval: −0.50, 0.10).

When we estimated a network structure based on the 13 sum score variables (nodes) for the function scales, we found a pattern similar to the symptom network, with some small differences. The redundancy analysis revealed several overlapping nodes. In the functions network, Problems with Teachers and Problems with Peers as well as Problems with Peers and Performance Anxiety shared significant overlap. The functions network showed high stability, with CS-coefficients of .593 and .593 for edges and strength centrality, respectively. The strongest correlations (see Table 3) were between Depression and School Aversion (r = .32), Depression and Social Anxiety (r = .25), Social Anxiety and Agoraphobia/Panic (r = .26), and Separation Anxiety and Problems Within the Family (r = .23). A strong link, similar to that found in the symptoms network, was observed between Problems with Parents and Problems Within the Family (r = .31) (Figure 3).

Figure 4 shows the centrality estimate strength for the ISAP function scales. Depression is the most central node, according to this measure, and Social Anxiety was the next central node in the network, although they did not differ in strength based on the bootstrapped significance test (bootstrapped confidence interval: −0.57, 0.01).

Considering the findings from both networks—the network of symptom scales and the network of function scales—depression emerges as an important construct in understanding SAPs. Specifically, Depression was the most central node in both networks. However, although Depression was descriptively the most central symptom, bootstrap difference testing indicated that it was not significantly more central than Agoraphobia/Panic in the symptoms network, or Social Anxiety in the functions network. This pattern suggests that a cluster of symptoms, including Depression, Social Anxiety, and Agoraphobia/Panic, may represent particularly important processes in SAPs.

Prior research on SAPs has highlighted the prominent role of depression (48). Building on this, the present study distinguished between Depression as a symptom and as a function, allowing for a more nuanced examination of the role of Depression in SAPs. Depression as a symptom reflects the young person's experience of depressive feelings and emotional distress within the school-related emotional landscape. In contrast, Depression as a function—or reason for absence—suggests that absence from school may serve to manage depressive symptoms, or that the young person's overall functioning is severely affected by depressive symptomatology.

In the current study, both the symptom and function network analyses revealed that Depression was strongly associated with School Aversion/Attractive Alternatives—a construct resembling truancy (2). The association between Depression and School Aversion was stronger in the function network than in the symptom network, suggesting heightened functional difficulties related to depressive symptomatology. This finding adds nuance to earlier evidence from Finning et al.'s (48) meta-analysis, which showed that depression was associated not only with overall absence but also with specific types of absence, including emotion-related absence and truancy. Moderate-to-large associations were observed between depression and emotion-related absence, whereas small-to-moderate associations were found for unexcused absences typically equated with truancy (48). Consistent with this pattern, a review by Heyne et al. (49) suggested that depression is more characteristic of emotion-related absence than of truancy.

A further pattern observed in the current study was the association between Depression and Aggression, consistent with previous research (50, 51). These findings suggest that some youth with SAPs who ostensibly display aggressive behavior may also experience depressive symptoms that are less readily visible (mixed disorder of conduct and emotions; ICD-10: F92x). Understanding this association is important, because among youth with SAPs—especially when the absence is characteristic of truancy—depressive symptomatology may go unnoticed, being overshadowed by behavioral issues. The redundancy analysis revealed that Aggression overlapped with having Problems with Teachers and Dislike of the Specific School. This pattern suggests that these factors account for similar variance in absence, possibly reflecting a broader pattern of school-related conflict and disengagement. However, results from the confirmatory factor analysis based on the same dataset suggest that they represent distinct latent constructs (32, 33). That is, rather than indicating that they are identical constructs, the overlap highlights how aggressive behavior, strained teacher relationships, and negative attitudes toward school may co-occur and reinforce one another in ways that contribute to problematic absence.

From a network perspective, symptoms may form self-sustaining loops, whereby strong interconnections help maintain them even after the original trigger has subsided (35, 52). However, because our data are cross-sectional, such feedback processes cannot be directly tested and should therefore be considered speculative. Consistent with prior network studies of adolescent depression (53, 54), we found that Depression and specific symptoms such as sadness and loneliness occupied central positions, supporting the view that depressive symptoms often function as structurally central features within broader symptom networks. This centrality may reflect the frequent co-occurrence of depressive symptomatology among adolescents experiencing persistent difficulties or comorbidity (55, 56). Nevertheless, potential causal or sustaining dynamics should be interpreted with caution when based on cross-sectional designs (57).

Agoraphobia/Panic emerged as another central node in the symptoms and functions networks. Prior research has consistently shown associations between anxiety and school absence (58). A meta-analysis by Finning et al. (48) identified social anxiety, separation anxiety, and generalized anxiety as commonly associated with absence, and these forms of anxiety also showed many edges in the current symptoms and functions networks. While Agoraphobia/Panic has received comparatively less attention in the field of school attendance, a study by Hella and Bernstein (59) demonstrated its relevance to absence. In the present analyses, Agoraphobia/Panic overlapped considerably with Social Anxiety in the redundancy analysis, suggesting shared explanatory variance. However, the CFA based on the same dataset suggested that they remain separable constructs (32, 33).

Social Anxiety also emerged as a central node in the functions network. It was linked to Depression, Agoraphobia/Panic, and Problems with Peers—a pattern consistent with prior meta-analytical evidence (48). Studies investigating interventions for emotion-related absence have shown less favorable outcomes when young people experience social anxiety (1). Social anxiety may contribute to strong avoidance of the school environment as a means of evading peer interactions. Ingul and Nordahl (21) similarly reported higher levels of social anxiety and peer problems among non-attending youth in Norway, and problematic social functioning appears to be a distinguishing feature of emotion-related absence in adolescence (60).

Previous studies employing path analytical methods have documented several pathways akin to those that emerged in the current study, such as connections between variables like depression and anxiety (22), separation anxiety and problems with parents (61), and social anxiety and problems with peers (62). The distinctive strength of network analysis lies in its ability to transcend predefined (a priori) pathways, estimating all potential pathways and subsequently regularizing them to zero if they prove too weak. This unique feature allows for the identification of pathways that might otherwise be overlooked. Given the relative newness of the ISAP, considering the patterns observed across its symptom and function subscales—although not part of a formal research question—offers valuable insight into how SAP-related processes may interact.

The intricate relationships among symptoms reported by young people displaying SAPs, and among the functions of these symptoms, are reflected in the complexity of interconnectedness between the nodes presented in Figures 1, 3, respectively. The symptoms network reveals 31 significant associations, and the functions network reveals 36 significant associations. This suggests that when adolescents with school absence of 10% or more have high scores on a symptoms and/or functions subscale, they are likely to also experience problems related to another subscale. The effect sizes on most nodes were large, indicating that when accounting for the effect of all factors in the symptoms model simultaneously and all factors in the functions model simultaneously, these factors collectively account for a substantial portion of the variance attributed to the respective node.

wThe strongest edge in both the symptoms network (.37) and the functions network (.31) was between Problems within the Family and Problems with Parents, after controlling for all other nodes. This moderate association is theoretically consistent, given that both constructs refer to challenges situated within the family domain, albeit with slightly different emphases. Problems within the Family contains specific reference to events/specific problems within the family, whereas Problems with Parents includes items about feeling rejected or overlooked by parents. The former node may be thought to include contextual factors, such as parental unemployment, or family dysfunction or conflicts. This could also be reflected in the association of the node with separation anxiety—cases when the anxiety is related to worry about parent wellbeing or home environment stability while the young person is at school. Conversely, Problems with Parents may reflect the quality of the parent-child relationship and parenting style, as evidenced by the emotional responses of the young person. The findings suggest a bidirectional influence between family dynamics and parent-child relationships.

Most associations at both the symptom and function levels were small to moderate in size and largely involved the same nodes across the symptoms and functions networks. Thus, the rationale for the differentiation between symptoms and the function of symptoms needs to be considered. The mean values for the symptoms scales were consistently higher than the mean values for the equivalent functions scales, possibly reflecting the generally high level of functioning in the current sample. That is, the participants may report symptoms, but may not experience functional difficulties related to the symptoms (i.e., report no function of the node). Furthermore, at the individual level, a separate analysis of symptoms and functions is still warranted for treatment planning purposes.

The current study capitalizes on the strengths of network analysis, a robust statistical approach enabling the simultaneous estimation of all potential connections. In this context, we were able to examine connections between symptoms often associated with SAPs, and the connections between the functions those symptoms serve with respect to absence from school. This methodological choice yielded valuable insights, highlighting the central position of depression among symptoms of SAPs and the reasons for absence from school. Additionally, a notable strength of this study is the incorporation of a sizable community sample, enhancing the generalizability of the study's findings. The network showed high stability in our analyses, meaning that it is likely that the networks would replicate in other similar samples.

Our use of the ISAP to measure 48 symptoms potentially associated with SAPs, along with the function each symptom serves in youth' absence, can be considered both a strength and limitation of the study. On the one hand, the ISAP allows for the simultaneous measurement of symptoms and the function of each symptom, surpassing the scope of other questionnaires that focus on either symptoms or functions, but not both. On the other hand, the novelty of the ISAP presents a potential limitation, given the limited—albeit growing –information on its psychometric properties. Overall, we suggest that the strength of using the ISAP outweighs the limitation, as no other available instrument provides information about the function of each specific symptom assessed.

In general, all ISAP constructs had acceptable to excellent internal reliability in the current study. Several subscales of the ISAP were identified as redundant in the networks. Whereas redundancy analysis identified overlap between several factors, the CFA models of ISAP using the same sample (32) and other samples (33) suggested that they represent distinct latent constructs. This difference arises because redundancy analysis is outcome-oriented, highlighting predictors that account for similar variance in the dependent variables, while CFA evaluates the measurement structure of the constructs themselves. Thus, factors can appear redundant in terms of explanatory power without being indistinguishable at the latent construct level.

Six limitations merit consideration. It is important to adopt a cautious approach in interpreting the findings in light of these limitations.

First, the study's reliance on self-reported data introduces potential recall and response biases, as participants' perceptions and memories may not fully capture their actual experiences or behaviors.

Second, the use of a non-random, self-selected sample limits the generalizability of the findings. It is possible that schools already engaged with attendance-related issues were more inclined to participate, resulting in a sample that may not fully represent all school contexts. Nevertheless, within participating schools and classes, most students took part, which helps mitigate—but does not eliminate—these sampling concerns.

Third, youth experiencing more severe SAPs may have been absent on data collection days. Given the limited outreach capacity for this extensively absent group, their underrepresentation introduces the likelihood of non-representative responses.

Fourth, the categorical measurement of “days absent” may have reduced the specificity of subsequent analyses. The ISAP questionnaire's non-intuitive categorical response options may also have been difficult for respondents to interpret.

Fifth, the study did not include some symptoms previously associated with SAPs, such as neuropsychiatric conditions such as autism spectrum disorder (63) and attention deficit hyperactivity disorder (64, 65), or symptoms associated with eating disorders (64).

Sixth, data collection occurred during the latter phase of the COVID-19 pandemic, potentially leading to distinct network structures compared with pre- or post-pandemic contexts. Although schools in Finland had resumed in-person instruction by that time, students' routines, well-being, and perceptions of school were likely still influenced by earlier disruptions caused by school closures and public health measures. These lingering effects may have shaped both the experiences reported by students and the observed network structures, thereby limiting the generalisability of the findings to fully post-pandemic conditions. Nevertheless, all students had returned to at least partial in-person instruction during data collection, marking a period of relative normalisation. The data therefore likely capture enduring effects of the pandemic without being confounded by the acute disruptions of earlier phases. This contextualises the findings and underscores the need for future replication studies in clearly post-pandemic settings.

At the same time, important shifts in school attendance research were already underway prior to the pandemic (77). Specifically, there has been a growing move from focusing primarily on individual and family determinants of absence toward considering broader systemic and structural influences (78), alongside a shift from addressing absence to actively promoting attendance (66). While the pandemic drew increased attention to school attendance issues, it is best understood as having accelerated—rather than initiated—these ongoing developments.

Adolescence is a pivotal period for identity development (67) and the safeguarding of mental well-being (68), making it a critical time for identifying and addressing SAPs, which tend to peak during this stage of development (1, 69). This underscores the importance of developmentally sensitive interventions that promote school re-entry, engagement, and well-being (1).

Given that SAPs are prone to become chronic when co-occuring with mental health symptoms (1, 70), timely and focused intervention is essential. Intervention efforts should strategically address symptoms that impede progress toward increased school attendance (1, 71, 72). The findings suggest the importance of ensuring that depressive symptoms are included in the assessment process—even when other mental health challenges initially appear salient. This emphasis does not minimize the relevance of other symptoms but acknowledges depression as a potentially central contributor to SAPs that might otherwise be overlooked. Depending on assessment outcomes, interventions may need to address depressive symptoms alongside social anxiety, school aversion, bullying, and related concerns. Several existing intervention protocols, including @school (73, 74), Back2School (75), and Modular Treatment (72), already include modules targeting depression. Future research could investigate whether prioritising central nodes (e.g., depression, social anxiety) leads to measurable reductions in absence. Combining network analysis with intervention trials may ultimately yield more efficient and personalised treatment strategies.

The results also indicated that symptoms and their functions tend to cluster, suggesting that certain emotional and contextual difficulties commonly co-occur. For example, problems within the family, difficulties in the parent-child relationship, and separation anxiety often appeared together. Recognizing these clusters highlights how interconnected factors may maintain or exacerbate SAPs. This has important implications for both assessment and intervention, as practitioners can target groups of related factors rather than addressing symptoms in isolation. Interventions that strengthen family relationships, improve communication, and reduce anxiety around separation may therefore alleviate multiple, interlinked challenges.

Future research should take into account the research context—whether community-based or specialist-referred. SAPs vary along a continuum of severity (76), and the current study included youth with absences ranging from 5 to over 48 days, most at the lower end. Thus, the findings likely reflect network patterns characteristic of milder forms of SAPs, which may differ from those observed in more severe or clinically referred samples. Investigating whether similar network structures of symptoms and functions emerge in such samples represents a valuable next step.

In addition, future studies should consider the broader context in which data are collected. The present study was conducted during the latter phase COVID-19 pandemic, a period marked by unique disruptions to students' social and school lives. Replicating the current analyses in post-pandemic conditions would help clarify the extent to which the observed patterns reflect enduring features of SAPs vs. context-specific patterns shaped by the pandemic environment.