Participation‐related constructs and participation of children with additional support needs in schools

To investigate associations between participation‐related constructs and participation frequency and involvement in inclusive schools.

Children with additional support needs, disabilities, and/or chronic conditions are at risk for participation restrictions. The World Health Organization's International Classification of Functioning, Disability, and Health defines participation as 'involvement in a life situation', 1 alongside 'body functions and structures', and 'activities'. These exist in dynamic interaction with 'health conditions' (disease/symptoms) and 'personal and environmental factors'. Participation itself has two important elements: attendance (presence, frequency, or amount) and involvement (experience, intensity, or engagement). 2,3 Factors that influence participation have been termed participation-related constructs. 2,3 These include environment (e.g. physical and social structures) and withinperson factors (e.g. interests, preferences, sense of self). 2,3 Environments provide (or do not provide) spaces, objects, relationships, activities, and opportunities 4 which influence choices, behaviours, and feelings around participation. No single aspect controls participation outcomes 5 and participation research should consider the simultaneous contributions of these interrelated factors. 6 Within-person processes interact with environments 2,3 and there may be differential influences of personal versus environmental determinants on different facets of participation. 7 Environment and within-person factors should therefore always be considered together. [2][3][4][5][6][7] Previous research has clustered around leisure or community participation in clinical samples. Yet children spend significant time in school, and a focus on participation here provides insights into needs and supports. 8 Children with additional support needs can experience participation restrictions. [9][10][11][12][13][14] However, it remains difficult to know the prevalence of restrictions because of a lack of representative research with contemporary measures. 15 Often, research is completed on the basis of parent or clinician reports, whereas teachers are seldom involved. This does not reflect the contemporary shift from individual 'health' interventions towards collaboration with educators. [16][17][18] This study aimed to investigate school participation (hereafter referred to as participation) using teacher report. We focus on participation in typical activities: classroom activities, field trips and school events, teams, clubs and organizations, getting together with peers, and special roles. Multilevel linear mixed-effects regression models were used. We expected, on the basis of our previous research, 5-7 that participation would be associated with participation-related constructs (identity, competence, symptoms, and environment) while controlling for a range of confounding variables.

Context and sampling
This study was conducted in primary schools randomly sampled in an urban area of Scotland (details in Appendices S1 and S2). To avoid overlapping samples, we excluded schools included in our previous studies. 6,7 Twenty-two schools were recruited, each providing 25 to 35 children. Such targets had previously been identified as feasible. 6,7 In Scotland, children begin primary school at 4 years 6 months to 5 years 6 months, attending for 7 years before starting secondary education at 11 years 6 months to 12 years 6 months. Schooling is inclusive, with most children attending general schools and a minority attending 'special' schools; inclusive practices led by teachers are therefore commonplace. 19 Inclusion Eligible children were in inclusive primary schools and had a need as reflected in national census records: 19 learning disability; dyslexia; other specific learning difficulty (e.g. numeric); other moderate learning difficulty; visual impairment; hearing impairment; deafblind; physical or motor impairment; language or speech disorder; autism spectrum disorder; social, emotional, and behavioural difficulty; physical health problem; or mental health problem. 19 Children could have multiple conditions/needs. The school leadership team identified children who met inclusion criteria and selected participants using a lottery (aiming to provide equal proportions across each of the seven school years/grades). Sample representativeness was explored through checks against a population census. 19 On review, the final sample demonstrated acceptable representativeness compared with census records (Appendix S3).

Ethics
Queen Margaret University Ethics Committee and the City of Edinburgh Local Authority (local government) Research Access Committee provided approval. Each school's head teacher provided written informed consent. Participating teachers provided informed consent. Participation was voluntary and schools and teachers were given the opportunity to opt out at any time. Parental consent was not sought, as teachers completed measures based on professional

What this paper adds
• Across a series of models, participation-related constructs were associated with frequency and intensity of participation. • Only participation-related constructs were associated with participation intensity. • Demographic and diagnostic variables were associated with frequency, not intensity, of participation. • Teacher assessment is valid for assessment of participation and participation-related constructs.
knowledge and school-held records. Children were not directly involved and data were anonymized before release.

Measures
Participation and Environment Measure for Children and Youth Participation was assessed using the Participation and Environment Measure for Children and Youth (PEM-CY). 20 The PEM-CY is based on a contemporary model of participation and has moderate to good reliability and validity. 20 Frequency and involvement scores from the school module were used. Respondents scored participation frequency (from 0 'never' to 7 'daily') and involvement (from 1 'minimally' to 5 'very involved'). Frequency was calculated as the average of ratings. This calculation can be inclusive of the items that are scored 'never' (0). 20 Involvement was calculated as the average of all ratings except those marked 'never'.

School Participation Questionnaire
Participation-related constructs were measured by the School Participation Questionnaire (SPQ), a 44-item teacher-report measure 21 (items in Appendix S4). The SPQ draws on a novel conceptual framework of participation determinants in the school setting. 5

Confounders
We controlled for factors that might affect observed relationships: age (months); sex (male/female); English as an additional language (yes/no); autism (yes/no); level of school support (three levels, with level III representing most intensive support); and looked-after status (yes/no) (government terminology for children in state care). We selected autism as these children were more likely to have participation restrictions 9,10 allowing us to test the relative importance of participation-related constructs alongside a known determinant.

Statistics
Bivariate analysis (Spearman's rank correlation) was initially used to explore the association between SPQ scores (participation-related constructs) and PEM-CY scores (participation), as well as the association of age with PEM-CY and SPQ scores. A Wilcoxon rank-sum test with continuity correction was used to measure the association of the following demographic variables with both the PEM-CY and SPQ: sex, looked after, language, and autism. A Kruskal-Wallis test was used to measure associations between school support level with PEM-CY and SPQ. Correlation coefficients (ρ) and effect size with p-values were obtained. Regression models examined how participation-related constructs affected participation. Our rationale was that the SPQ measures factors that influence participation. Hence participation was the dependant variable, while SPQ scales were independent variables, with other independent variables treated as confounders. Outcomes were observed for children (level 1 units) within schools (level 2 units) leading to non-independent data. A class of multilevel linear mixedeffects regression models 22 taking into account the hierarchical nature of the data were fitted, wherein t-tests using Satterthwaite approximation were used for the fixed effects and likelihood ratio tests for the random effects. Maximum likelihood estimates and bootstrap confidence intervals with p-values were obtained. 23 The fixed effects part of the model illustrated the impact of independent variables (e.g. participation-related constructs, demographics) on the outcome (e.g. participation intensity), whereas the random effects part included the grouping factor 'school', and allowed for a random intercept for each school to control for the variation between children across schools. Throughout, separate models were run for participation frequency and involvement, as the literature indicates they are distinct. For each outcome, three sets of models were built. Model A included only participation-related constructs. Model B investigated confounding variables. Model C examined all variables, exploring associations between participation-related constructs and outcome while adjusting for confounders. Multicollinearity was detected in models including both SPQ identity and SPQ competence scales (Spearman's ρ = 0.84; variance inflation factor >3). Therefore, separate models were run including each scale.
Model robustness was determined through verifying the assumptions of linearity, normality, and homoscedasticity, and through the random effects block bootstrap technique. 24 Missing data were less than 10% (5.2%), and complete case analysis was conducted. Statistical tests were conducted at the 5% level of significance. SPQ ratings for each subscale were transformed to Rasch measures (Appendix S5). Analyses were conducted in R (R Foundation, Vienna, Austria).

Model validity
Models satisfied the assumptions of linearity, homoscedasticity, and normality for fixed and random effects (Appendices S6 and S7). The residual plot did not indicate deviations from a linear form, its variance was constant and did not depend on the fitted values, and both the residuals and random intercepts followed a normal distribution. A non-significant result was found for the Shapiro-Wilk normality test. The random effects block bootstrapping technique showed negligible bias for parameter estimates, wherein the observed values of regression coefficients from the models were very close to the average value of the same obtained from 1000 replicates (Appendices S6 and S7).

R E SU LTS
Two hundred and fifty-two teachers from 22 schools administered measures for 688 children.

Bivariate analysis
Most confounders were significantly but weakly associated with participation frequency and involvement. Significantly higher participation frequency and involvement was observed for females (vs males); non-autistic children (vs autistic); across school support levels I, II, and III; and for age ( Table 2). Participation frequency and involvement had moderate to strong significant positive relationships with all participation-related constructs, except for a weaker association between symptoms and frequency (Appendix S8). All participation-related constructs (identity, competence, environment, and symptoms) were significantly but weakly associated with autism and with school support level, with higher scores observed for non-autistic children (vs autistic children) and school support levels I, II, and III. Identity, competence, and symptom scores were significantly higher in females. Identity and competence scores also significantly increased with age. Symptoms were significantly associated with looked-after status, with better scores for those not looked after. Competence and symptoms scores were significantly higher for those with English as an additional language (Table 3).

Modelling
Owing to multicollinearity, separate models were run including SPQ identity or competence as a covariate (Appendix S9). For brevity, results are presented here for models including competence, with any salient differences in models highlighted. Assumptions were tested and validation performed for all models.

Participation involvement
In the intermediate model containing only participationrelated constructs (Table 4, model A), all (competence, environment, and symptoms scores) were significantly associated with participation involvement, where increases contributed towards higher participation involvement. In the intermediate model containing confounding variables (Table 4, model B), sex (male), autism (yes), and school support (level III compared with level II) were significantly associated with lower participation involvement. Increasing age was significantly associated with higher participation involvement. In the final model (Table 4, model C), participation-related constructs (competence, environment, and symptoms scores) remained significantly associated with participation involvement after controlling for confounding variables. No confounding variables were significantly associated with the outcome. When analyses were replicated with identity as a covariate, findings remained the same. Identity scores were significantly associated with participation involvement across all analyses, and participation-related constructs remained the only significant associations in the model. Wilcoxon rank-sum test with continuity correction was used to measure the association of demographic variables sex, looked after, language, and autism with PEM-CY. A Kruskal-Wallis test was used to measure the association of support level with PEM-CY. ρ denotes Spearman's ρ, which was used to measure the association between age and PEM-CY.
Respondents scored PEM-CY participation frequency (from 0 'never' to 7 'daily') and involvement (from 1 'minimally' involved to 5 'very involved).   Wilcoxon rank-sum test with continuity correction was used to measure the association of demographic variables sex, looked after, language, and autism with SPQ. A Kruskal-Wallis test was used to measure the association of support level with SPQ. ρ denotes Spearman's ρ, which was used to measure the association between age and PEM-CY. All SPQ items are rated on a 4-point Likert scale (from 1 'disagree' to 4 'agree'). Higher scores represent a more facilitative school environment (environment scale), favourable child characteristics (identity scale, competence scale), and fewer symptoms (symptoms scale).

Participation frequency
In the intermediate model containing only participationrelated constructs (Table 5, model A), competence and environment scores were significantly associated with participation frequency, where increases contributed towards higher frequency. In the intermediate model containing confounding variables (Table 5, model B), sex (male), language (not English), autism (yes), and school support (level III compared with level II) were significantly associated with lower participation frequency. Increasing age was also significantly associated with higher participation frequency. In the final model (Table 5, model C), after controlling for confounding variables, competence and environment scores remained significantly associated with participation frequency. Sex (male) and autism (yes) were no longer significantly associated after adjusting for participationrelated constructs. Language and school support, however, did remain significantly associated. When models A, B, and C were replicated with identity, the findings were similar. Identity scores were significantly associated with participation frequency across all analyses. However, autism (yes) remained significantly associated in the final model involving identity scores, whereas age and school support levels were not significantly associated.

DISCUS SION
This study contributes to our understanding of participation and significantly extends previous research using the SPQ by using more sophisticated modelling. The study also benefits from a large representative sample. Using the PEM-CY, participation involvement and frequency were measured. This is important as there may be differences in these outcomes. Involvement in particular is complex, with few available measures to capture it, or understanding of what influences it. For these reasons, it is essential to make a distinction between frequency and involvement. After controlling for confounders, participationrelated constructs (as measured by the SPQ) were associated with participation while controlling for confounders. For frequency, a greater range of factors were associated with outcomes, while for involvement only participation-related constructs were important. As part of a body of evidence, this replicates findings that frequency and involvement are T A B L E 4 Multilevel linear mixed-effects regression models for participation involvement  *p < 0.05, **p < 0.01, ***p < 0.001.
Abbreviations: AIC, Akaike information criterion; CI, confidence interval; ICC, intraclass correlation coefficient; SD, standard deviation. a Support level III vs I was also tested and found to be non-significant in all cases.
uniquely enabled by environment and within-person factors, as found in early ground-breaking studies, 25 previous research using the SPQ model, 6,7 and wider participation measurement research. 12,13,26,27 This study allows us to explicate the influence of children's identity and competence on participation. This includes a child's ability to do things such as follow rules and routines, their belief in themselves, and their understanding of and ability to meet responsibilities. While findings suggest these constructs are associated with more frequent and more intensive participation, they may also be strengthened through participation. 3 That is, participation is a means by which children acquire an identity and sense of belonging, and gain an understanding of their responsibilities and contributions to social groups. Participation also affords the opportunity for practice, further building competence.
Our findings confirm previous indications that the relationship between environment and participation differs across settings. Coefficients for environment factors, while generally remaining statistically significant, were lower than for child factors. Previous research has confirmed environmental factors to be a strong predictor of community participation, 27 and comparisons between community and school settings show more pronounced effects of environment in the community over schools. 26 It may be that schools are safe and supported environments where teachers compensate for needs, in contrast to less supportive community settings. Nonetheless, it remains incumbent for practitioners to ensure that children receive supports that include modifications to the physical and social environment, in all settings, including the school. The SPQ measurement model 5-7 includes child and environment factors, from the perspective of teachers, and has demonstrated consistent associations with participation. The current literature looking at participation has largely relied on operationalizing from 'objective' variables (e.g. income, functional status) or from parents/family members or the person with the disability. Very few or none have asked teachers. This is a useful contribution to show expected patterns of relationships hold even with a reporter that has typically not been used in the literature thus far. It also demonstrates that teachers can validity assess participation and participation-related constructs. *p < 0.05, **p < 0.01, ***p < 0.001.
Abbreviations: AIC, Akaike information criterion; CI, confidence interval; ICC, intraclass correlation coefficient; SD, standard deviation. a Support level III vs I was also tested and found to be non-significant in all cases.
This study contributes understanding about the relationship between children's demographic characteristics, support needs, participation, and participation-related constructs that have implications for policy and practice. We found, after controlling for a range of variables, that there was a relationship between having English as an additional language and reduced participation frequency. Although this is a novel finding in participation research, the 'double disadvantage' of possessing multiple needs has been previously identified. 28 Targeted participation support for children who do not speak the dominant language may therefore be helpful. We also found a relationship between the highest and lowest levels of school support and reduced participation frequency. This may reflect the fact that children have higher levels of need, but do not have supports in place, that their needs have not been identified, or that supports are insufficient. Overall, as highlighted by educationalists and multidisciplinary commentators, 8,16,29 teachers require support in working with these learners, with the potential for participation to be negatively impacted.
When adjusting for participation-related constructs, several demographic characteristics did not maintain significant associations with outcomes. The implication is that, although factors such as language, sex, age, or diagnosis are important, it is the interaction of these with participationrelated constructs that determines outcomes. For example, previous research has shown participation restrictions among autistic children. 9,10 In our study, although autistic children did demonstrate reduced participation frequency and involvement, and greater levels of need as measured by the SPQ, autism was not consistently associated with participation in most of the final models. This supports the idea that the level of functioning (as captured by the SPQ) is more important than diagnosis, since diagnosis does not capture the person-environment complexity which determines participation outcomes.
The strengths of this study include the use of robust sampling, measures conceptually matched to evidence-based definitions, and the inclusion of several correlates. The sample was large enough to produce meaningful estimates, and was non-overlapping with previous research. However, it was located in one city, and there are important factors, such as income 26 and mother's educational status, 14 which were not included.
The ultimate goal is to facilitate implementation of evidence-informed approaches to participation. Interventions provided by therapists 30 and teachers 17 are necessary. For teachers, examining the child and environment to understand how and why participation varies is important. Such investigations have practical implications and highlight learners requiring support. This is important as there are children for whom it is difficult to identify what extra provision is needed. 8,29 Collaboration and innovation is required. 16 Over and above children being offered, or having available to them, a range of activities, there are child and environment factors that restrict participation. Therefore, focusing on placement, 'inclusion', or presence is insufficient if increased participation is desired. Finally, research is needed to ascertain which participation differences are meaningful to teachers and families. Currently available measures have little established evidence for what is considered an important difference according to these stakeholders.

CONCLUSION
Participation-related constructs as measured by the SPQ remained associated with participation after controlling for age, sex, level of school support, and autism. For participation frequency, a greater range of factors were associated with outcomes, while for participation involvement only participation-related constructs were important. This suggests that a focus on typical variables of interest may yield insights associated with participation frequency (e.g. how much or how often children participate), but an understanding of participation involvement requires analysis of participation-related constructs. The findings demonstrate that more focus is required on addressing these factors to support children's participation. ers, and clinicians and can be used to improve the participation of children in inclusive settings. The SPQ is free and available online (www.third space.scot). The authors have stated they had no interests that might be perceived as posing a conflict or bias.

F U N DI NG I N FOR M AT ION
The City of Edinburgh Council, National Health Service Lothian, and Scottish Government.

DATA AVA I L A BI L I T Y S TAT E M E N T
The data that support the findings of this study are available on reasonable request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.