Trends in adolescent emotional problems in Wales between 2013 and 2019: the contribution of peer relationships

,

Emotional problems (anxiety and depression) are associated with distress, impairment and poor longterm outcomes.For many individuals, these problems are first experienced in childhood and adolescence (Kessler et al., 2005;Uhlhaas et al., 2023).There has been a marked rise in the prevalence of emotional problems in recent decades observed in epidemiological studies in a number of countries using symptom scales and structured diagnostic assessments (Collishaw, 2015;Patalay & Gage, 2019;Pitchforth et al., 2019;Sadler et al., 2018;Wilson & Dumornay, 2022).Two important and related questions remain largely unresolved.The first question is what explains the rise in emotional problems in young people.The second is whether increases in emotional problems have impacted different sociodemographic groups to differing degrees.
Only a few studies have examined possible explanations for temporal increases in youth emotional problems using linked data on trends in emotional problems and possible explanatory factors.Studies have tested changes in family life, parenting practices and parent mental health (Collishaw, Gardner, Maughan, Scott, & Pickles, 2012;Collishaw, Goodman, Pickles, & Maughan, 2007;Schepman et al., 2011), exam-related stress (H€ ogberg, Strandh, & Hagquist, 2020;West & Sweeting, 2003), healthrelated behaviours (Patalay & Gage, 2019) and the increasing use of digital technology (Twenge, 2020).Findings suggest that changes in some of these areas of young people's lives might have contributed to trends in mental health, but factors considered to date do not fully explain the substantial increases in youth emotional problems over recent decades (Collishaw & Sellers, 2020).It is therefore important to consider other factors that may play a role.
Developmental theory emphasises the importance of social relationships in adolescence (Blakemore & Mills, 2014).Good quality friendships provide support and enjoyment and are linked with young people's self-efficacy, self-esteem and better mental health (van Harmelen et al., 2016).In contrast, adverse social experiences, such as frequent bullying, are associated with a wide range of adverse health and psychosocial consequences (Arseneault, 2018;Moore et al., 2017;Singham et al., 2017); a recent metaanalysis of quasi-experimental studies provides evidence consistent with a causal effect of bullying on Conflict of interest statement: No conflicts were declared.emotional problems (Schoeler, Duncan, Cecil, Ploubidis, & Pingault, 2018).Efforts to reduce bullying victimisation have shown promise in recent studies from a number of countries (Bonell et al., 2018;Tiiri et al., 2020); however, studies tracking trends in bullying provide mixed evidence on population-level trends across different countries (Cosma, Whitehead, Neville, Currie, & Inchley, 2017;Molcho et al., 2009;Smith et al., 2023).In the UK, rates of bullying victimisation remained fairly stable between 2002 and 2010 and increased when assessed again in 2014 (Cosma et al., 2020).In contrast to studies tracking bullying, little is known about trends in youthreported friendship quality (Currie & Morgan, 2020).In summary, it remains uncertain whether and how young people's social relationships have changed over recent decades, and we are not aware of any studies testing whether changes in social relationships have contributed to trends in emotional problems.More evidence is needed that examines trends in adolescent social relationships encompassing both positive and negative experiences, and importantly whether any change in social relationships has impacted population trends in mental health.
When considering secular trends in youth mental health and social relationships, it is important to recognise that trends may vary across sociodemographic factors.For example, long-term increases in youth emotional problems appear more pronounced for girls compared to boys, and for adolescents compared to children (Collishaw, 2015).In addition, mental health problems are more common among young people from poorer backgrounds (Bor, Dean, Najman, & Hayatbakhsh, 2014;Sadler et al., 2018), and inequalities in young people's mental health have likely increased (Collishaw, Furzer, Thapar, & Sellers, 2019;Royal College of Paediatrics and Child Health, 2017).An additional aim, therefore, is to test differences in trends in youth mental health and social relationships by age, sex and family affluence.
To directly test changes in population prevalence, it is important to compare unselected population cohorts using the same instruments and methodology on each occasion.The School Health Research Network has undertaken a national, school-based survey of young people in Wales, and this presents an unrivalled opportunity to examine trends in youth mental health at scaleutilising very large representative cohorts of children aged 11-16 years assessed in comparable ways across three occasions -2013, 2017, 2019.Using data from these representative surveys of young people in Wales, UK we tested the following research questions:

Survey procedure and ethics
The 2013 survey was paper-based and supervised by researchers, whereas the 2017 and 2019 surveys were delivered electronically and supervised by teachers.Schools were asked to organise classrooms to ensure privacy for students, and supervising staff were asked to remain at the front of the classroom unless a student asked for help.
Students were provided with information about the surveys, and the first question asked for their consent to take part which if they declined closed the survey.Students also had the opportunity to withdraw from data collection at any time.All questions included an additional response option, 'I do not want to answer', except the consent and year group questions.
To maximise the capacity for a broader range of questions, both the 2017 and 2019 surveys had a number of different pathways (or 'routes'), with some common items included within all routes (asked of all students) and some included only in certain routes (asked of subsamples of students).In the current study, all questions presented in the main analyses were included across all survey routes, with the exception of friendship quality which was included for 28% of students in 2017 and 50% in 2019.Details of routings are included in Figure S1.Family affluence was assessed using the Family Affluence Scale III, which comprises six items about family assets reported by young people (Hartley, Levin, & Currie, 2016), and which has been validated cross-nationally against measures of household income (Torsheim et al., 2016).Respondents reported the household's number of cars (none, one, two or more), computers (none, one, two and more than two) and bathrooms (none, one, two and more than two), whether they had their own bedroom (no and yes), whether they had a dishwasher (no and yes) and the number of holidays spent abroad in the past year (not at all, once, twice and more than twice).The scores for each item were summed to give a total affluence score ranging from 0 to 13 and then split into three groups (low -0 to 8, medium -9 to 10 and high affluence, 11 to 13).
Emotional problems.The Health Behaviour in Schoolaged Children Symptom Checklist (HBSC-SCL) assesses the occurrence of eight somatic and emotional problems.We focus on the 4-items psychological sub-scale (Gariepy, McKinnon, Sentenac, & Elgar, 2016).Participants were asked about the frequency with which they had felt each of the following over the past 6 months: (a) Feeling low, (b) Irritability or bad temper, (c) Feeling nervous and (d) Difficulties in getting to sleep (the items were reverse coded and responses included: (4) 'About every day', (3) 'More than once a week', (2) 'About every week', (1) 'About every month', (0) 'Rarely or never').Complete case responses were summed to form a scale score ranging from 0 to 16.The internal consistency of the items was adequate across the surveys (Cronbach's a = .77in 2013; a = .75 in 2017; a = .77in 2019).Other studies have reported high reliability and validity with respect to measures of emotional problems of this 4-item subscale for adolescents between 11 and 15 years (Gariepy et al., 2016).Additionally, our own analysis provided evidence of good validity with respect to a well-established measure of depression symptoms in the SHRN 2019 sample, and of DSM-IV major depressive disorder in a sample of young people with a parent with a history of major depressive disorder (The Early Prediction of Adolescent Depression study; EPAD; Mars et al., 2012).See Appendix S1 for full details of validation.Main analyses use the mean emotional problem score, but for descriptive purposes, we also examine the proportion with high emotional problem scores using a threshold of SCL4 > 7 suggested by the validation analyses (see Appendix Figures S1 and S2).
Social relationships.Friendship quality: Young people were asked to indicate their agreement or disagreement with four items on a 7-point Likert scale, relating to the extent to which they felt that friends tried to help them, could be counted on when things go wrong, could share in joys and sorrows, and that they could talk easily to friends.Response options ranged from very strongly disagree (1) to very strongly agree (7).Due to survey routing (see above), these questions were asked to 28% of respondents in 2017 and 50% in 2019.
Items were converted to a mean scale score ranging from 1 to 7 for respondents who answered at least three out of four items.Scale items demonstrated good internal consistency (Cronbach's a = .95 in 2013; a = .96in 2017; a = .96in 2019).
Experience of bullying (face to face): To address research question 2, analyses were stratified by gender, age and family affluence.Differences in trends in emotional problems for each demographic variable were then examined by testing two-way interactions of survey year 9 each demographic.
To address research question 3, we tested for differences between surveys in mean friendship quality using linear regression and rates of frequent bullying and cyberbullying using logistic regression, with survey year entered as a categorical predictor (2013 as the reference).All analyses were conducted without covariates (unadjusted model) and again controlling for age, gender and family affluence (adjusted model).
Next, to address research question 4, analyses of social relationship outcomes were repeatedly stratified by gender, age and family affluence.In addition, two-way interactions of survey year 9 each demographic were tested.
To address research question 5, we compared unadjusted and adjusted (for changes in social relationships) estimates of change in mean emotional problems by survey year.
All models were conducted using complete case analysis and included cluster-robust standard errors to account for schoollevel clustering.
Sensitivity analyses to account for changes in school sampling and participation between 2013 and 2017/2019.First, analyses were repeated including only the common schools in all three surveys (N = 82 schools, N = 22,671 students) to assess whether findings were affected by changes in school selection and participation between 2013 and 2017/2019.Second, weighted regression models were estimated using survey weights for the 2013 S1  and S2).

Results
The Emotional problem scores are summarised by gender, age and family affluence in Figures 1-3 respectively (and in Table S4).Stratified linear regression analyses (adjusting for other demographic covariates) estimated the effect of survey year separately for each subgroup.As shown, emotional problem scores increased between 2013 and 2019 for both boys and girls (Figure 1; Table 4, Models 1a, b), for all age groups (Figure 2; Table 4, Models 1c-1g) and each of the three family affluence groups (Figure 3; Table 4, Models 1h-j).
Next, a multivariable regression analysis tested for variation in change over time by testing two-way interactions of age, gender and affluence by survey year (Table 4, Model 2).A significant interaction between survey year and gender showed that the increase in emotional problems between 2013 and   S7).
There was no evidence that trends in face-to-face bullying differed by gender or affluence, but a significant interaction between survey year and age (OR [95% CI] = 1.059 [1.004, 1.118]) suggested a steeper increase in face-to-face bullying between 2013 and 2019 for older children (Table S9).There was no evidence that trends in cyberbullying differed by age or affluence.However, a significant interaction between survey year and gender suggests that the increase in cyberbullying over time was less marked for girls than boys between 2013 and 2019 (OR [95% CI] = 0.694 [0.513, 0.941]; Table S11).

Research question 5:
To what extent do changes in social relationships account for changes in young people's emotional problems?
To test if changes in social relationships contributed to trends in emotional problems, analyses compared estimates of the effect of survey year in an unadjusted model (  .117 [1.938, 2.296]) were each independently associated with worse emotional problems in the multivariable analysis (Table 5c,Model 3c).Table S5 also shows univariable associations with emotional problems, with comparison of effect sizes and their confidence intervals indicating a strengthening of associations between face-to-face bullying and friendship quality with emotional problems.

Discussion
This large whole-nation study provides further evidence for an increase in adolescent emotional problems over the second decade of the 21st Century.Findings suggest that this is not a trivial increase in difficultiesmean scores increased by more than a third of a standard deviation over a 6year period, and the proportion with high emotional problem scores increased from 22.84% in 2013 to 37.69% in 2019.Findings are consistent with other data from the United Kingdom, United States and high-income European countries (Bor et al., 2014;Collishaw, 2015;Cybulski et al., 2021;H€ ogberg, Strandh, Johansson, & Petersen, 2023;Thorisdottir, Asgeirsdottir, Sigurvinsdottir, Allegrante, & Sigfusdottir, 2017) suggesting a substantial longterm increase in adolescent emotional problems predating the COVID-19 pandemic.While not directly assessed in our study, findings also accord with UK data showing an increasing rate of emotional disorders (Sadler et al., 2018), referrals to Child and Adolescent Mental Health Services (CAMHS; Grimm et al., 2022) and youth self-harm and suicide (Cybulski et al., 2021;Padmanathan, Bould, Winstone, Moran, & Gunnell, 2020) over this period.
One aim of our study was to test whether trends varied by gender, age or family affluence.As expected, we found higher emotional problems among girls than boys and in older children in each survey.In line with some previous findings (Keyes, Gary, O'Malley, Hamilton, & Schulenberg, 2019;Potrebny et al., 2019;Riecher-R€ ossler, 2017;Slee, Nazareth, Freemantle, & Horsfall, 2021), the study suggests an overall greater increase in emotional problems between 2013 to 2019 for girls than boys (though results comparing 2013 and 2017 were contradictory).Previous evidence has suggested that secular increases in emotional problems have been more pronounced for adolescents than for children (Armitage et al., 2023;Sadler et al., 2018).The current study found no differences in trends according to age.Emotional problems increased over time for all levels of family affluence; however, those from poorer families experienced a steeper increase between 2013 and 2019.These findings provide evidence that inequalities in youth mental health have further increased in the UK, continuing a longer-term trend observed over previous decades (Collishaw et al., 2019;Reiss, 2013).
The study is one of the first to directly examine whether the quality of young people's social relationships has changed over time and to test the implications for trends in youth emotional problems.Despite increased recognition of its harmful effects, bullying remains a common experience in Wales.One in seven young people reported being bullied at least two to three times a month.Frequent face-toface bullying was more common among girls, younger children and those from lower affluent family backgrounds.Frequent face-to-face bullying showed only minor differences in prevalence between 2013, 2017 and 2019 and broadly similar trends for different demographic groups.These results extend previous evidence showing a rise in the prevalence of bullying in Wales between 2002 and 2014 (Cosma et al., 2020).
Around 1 in 20 children in Wales reported frequent (at least 2 to 3 times per month) experiences of cyberbullying, with rates higher in girls, older children and those from less affluent backgrounds.Cyberbullying rates showed a small increase over this periodin line with international evidence (Kennedy, 2021;Trompeter et al., 2022) and little evidence that trends varied markedly by demographic group.
This study is among the first to examine trends in friendship quality.Developmental theory and research highlight the importance of friends as a source of enjoyment, intimacy and support (Blakemore & Mills, 2014;van Harmelen et al., 2016).Positive friendships are associated with better mental health, well-being and can promote resilience in the context of adversity (Collishaw et al., 2016).In general, girls, younger pupils and those from higher affluent backgrounds reported higher quality friendships.There was also a modest reduction in reported friendship quality between 2013 and 2019, and this trend particularly affected older pupils who reported a steeper decline in perceived friendship quality over the study period.

Implications
We found strong associations between social relationship measures and emotional problems in line with other studies (K€ allm en & Hallgren, 2021).However, there were only modest changes in rates of bullying and friendship quality over time, and these did not explain the population-level increases in emotional problems.
Population health efforts have often focused on the development, implementation and evaluation of antibullying strategies (Bonell et al., 2018;Tiiri et al., 2020).Some evidence suggests that where this is successful this could impact population trends in youth emotional problems.For example,  et al., 2020).Whole school interventions aim to provide a supportive context for healthy relationships in schools more generally, and it will be interesting to consider their success in supporting young people's positive peer relationships and friendships.This study found no evidence that trends in social relationships accounted for the observed increase in youth emotional problems, but it supports the view that improving the quality of young people's social relationships and reducing their exposure to bullying are important population health priorities.

Strengths and limitations
A main strength of this study is its large, robust, national and comparable repeated cross-sectional samples, which allowed us to monitor temporal change in young people's mental health on a population-wide scale.This study is based on three comparable and well-designed surveys conducted in 2013, 2017 and 2019 in mainstream secondary schools in Wales.Moreover, young people's emotional problems were assessed using the same measure, and we provide evidence of the validity of this measure against established measures of adolescent mental health included in the 2019 survey and in a separate (older) high-risk sample (Powell et al., 2021) as shown in Appendix S1.Additionally, this study is one of the largest contemporary studies to assess both positive and negative experiences of young people's peer relationships and one of the few studies to our knowledge to examine how these experiences have changed over time and how they relate to trends in mental health.There are also important potential limitations.Emotional problems were assessed using a brief measure (four items) not designed as a diagnostic tool.Caution is needed in extrapolating to trends in clinical need or clinical outcomes.Nevertheless, the findings are in accord with previous epidemiological evidence in the UK showing an increase in emotional disorders in this age group between 1999 and 2017 (Sadler et al., 2018).Measures of social relationships were relatively crude, and several key aspects were not covered, including loneliness, emotional support, and characteristics (including mental health) of peers.The most recent surveys (2017 and 2019) had greater coverage across schools (and classrooms within schools) in Wales, and one possibility is that differences between the schools included on each occasion might have contributed to the observed findings.However, the sensitivity analysis showed similar findings for analysis restricted to common schools in all three surveys.In addition, it is possible that the different modes of delivery across surveys (on paper supervised by researchers in 2013 vs. on a computer supervised by teachers in 2017 and 2019) may have differentially influenced young people's responses.However, in all surveys, schools were instructed to ensure that young people could complete questionnaires in privacy.We excluded students who stated that neither gender best described them given that this categorisation was only introduced in the 2019 survey.This is important to note as this population has particularly poor mental health and well-being (Ghassabian et al., 2022;Page et al., 2021), thus their exclusion may have led to an underestimation of emotional problems for the 2019 survey.Further population-based research is needed to examine the mental health of this underrepresented group, including trends over time, and relevant risk and protective factors.More generally, despite overall high school and pupil participation rates, there were non-trivial differences in response rates to the three surveys.Whilst school-level participation improved between 2013 and 2019, pupil participation rates dropped over this period.There was also modest item-level non-response particularly in relation to more sensitive questions (e.g.bullying), see Table S3.Previous research demonstrates that survey non-response is often greater for more vulnerable groups (Mostafa et al., 2020).As such, rates of emotional problems might be underestimated, and cross-survey estimates of increases in emotional problems may be conservative given increasing levels of pupil non-response over time.A further important consideration is that the survey did not include children outside of mainstream educational settings.Finally, we note the large sample size and the number of statistical tests performed.

Conclusion
An accurate understanding of trends in adolescent psychopathology is essential for planning population health priorities.Our findings highlight a growing need for mental health support for young people, as indicated by a substantial increase in emotional problems over the past decade, particularly among those from lower affluent families.Friendship quality and experiencing any form of bullying were strongly associated with higher emotional problems in all three surveys, but trends in social relationships did not account for the increase in emotional problems over time.

Figure 3
Figure 3 Mean emotional problems score by family affluence, 2013-2019.Error bars indicate 95% confidence intervals.FAS = Family affluence.The 2015 survey represents a predicted midpoint figure The surveys were each approved by the Cardiff University School of Social Sciences Ethics Committee.Ó 2023 The Authors.Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for Child and Adolescent Mental Health.
Students were asked 'how often have you been bullied at school' in the past 2 months.Response options included 'I have not been bullied', 'it has happened once or twice', '2 or 3 times a month', 'about once a week' and 'several times a week'.Response options were combined into 'frequently bullied' if they reported being bullied 2 or 3 times a month or more in the past 2 months.Experience of cyberbullying:Participants were asked if in the last 2 months they had ever been cyberbullied 'e.g., someone sent mean instant messages, email or text messages about you, wall postings, created a website making fun of you, posted unflattering or inappropriate pictures of you online without permission or shared them with others'.As with faceto-face bullying, response options included 'I have not been bullied', 'it has happened once or twice', '2 or 3 times a month', 'about once a week' and 'several times a week'.Response options were combined into 'frequently cyberbullied' if they reported being cyberbullied 2 or 3 times a month or more in the past 2 months.Descriptive data are presented as frequencies and percentages.Emotional problem scores and associated standard errors among the whole population and stratified by survey year are presented for years 2013, 2017 and 2019.To address research question 1, linear regression analysis examined categorical survey years as a predictor(2017 vs.  2013 and 2019 vs. 2013)of emotional problem scores.Analysis was conducted first with no covariates (unadjusted) and then repeated while adjusting for child gender, age, and family affluence to account for any cross-survey differences in response rate by gender, age or affluence.
HBSC data to correct for disproportionate stratification by local health board when sampling schools, varying numbers of classes per year Ó 2023 The Authors.Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for Child and Adolescent Mental Health.Downloaded from https://acamh.onlinelibrary.wiley.com/doi/10.1111/jcpp.13924 by Welsh Assembly Government, Wiley Online Library on [13/12/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)onWileyOnline Library for rules of use; OA articles are governed by the applicable Creative Commons License group, varying class sizes, and correcting for different levels of response among particular groups(Welsh Government, 2015).All sensitivity analyses were consistent with substantive conclusions, showing only minor variations in findings.Findings are presented in Supporting Information (Tables 95% CI [.957, 1.349], p < .001) in mean SCL-4 emotional problem scores between 2013 and 2017 and an increase of 1.573 (95% CI [1.380, 1.765], p < .001) in scores between 2013 and 2019 (see Table3, Model 1a).Accounting for differences in the demographic profiles of the three samples showed closely comparable results (Table3, Model 1b).

Table 2
Emotional problems and social relationship variables, 11-16-year olds participating in Welsh HBSC/SHRN SHW surveys between 2013 and 2019 a a The mean friendship quality score measure was asked to all participants in the 2013 survey, 28% in the 2017 survey and 50% in the 2019 survey due to survey routing.Ó 2023 The Authors.Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for Child and Adolescent Mental Health.Research question 3: Has the quality of young people's social relationships changed from 2013 to 2019 (friendship quality, face-to-face bullying, cyberbullying)?Table 2 summarises the peer relationship measures by survey year.Friendship quality reduced slightly

Table 3
Coefficient estimates for change over time in emotional problem scores among students in Wales between 2013 and 2019, unadjusted and adjusted for age, gender and family affluence Downloaded from https://acamh.onlinelibrary.wiley.com/doi/10.1111/jcpp.13924 by Welsh Assembly Government, Wiley Online Library on [13/12/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Ó 2023 The Authors.Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for Child and Adolescent Mental Health.

Table 4
Coefficient estimates for emotional problems among students in Wales between 2013 and 2019; stratified by age, gender and family affluence.Adjusted for socio-demographics Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for Child and Adolescent Mental Health.

Table 5
Coefficient estimates for emotional problems among students in Wales between 2013 and 2019, adjusting for social relationships comparison of trends in emotional problems unadjusted or adjusted for trends in peer victimisation (which declined between 2005 and 2017 in the Netherlands) supports this conclusion (De Looze Ó 2023 The Authors.Journal of Child Psychology and Psychiatry published by John Wiley & Sons Ltd on behalf of Association for Child and Adolescent Mental Health.a