Adolescent Mental Health Disorders and Upper Secondary School Completion – The Role of Family Resources

ABSTRACT

This article investigates the role of socioeconomic family resources in modifying the relationships with upper secondary school completion (SSC) for three mental health dimensions, i.e., externalizing, internalizing and substance use disorders. Using data from administrative registers, we follow a cohort in Norway born in 1996 into early adulthood. We find that having a mental health disorder in adolescence was associated with 12–17 percentage points lower SSC rate after adjustment for demographic and household factors, comorbidity and educational performance. In girls, high family income attenuated the negative relationships between all three mental health dimensions and SSC, while in boys, this was true only for substance use disorders. The paper concludes that access to family socioeconomic resources “buffers” the negative impact of mental health disorders on SSC, but less so in boys, contributing to aggravate social and health-related inequalities in SSC.

KEYWORDS:

• Adolescent
• mental health disorders
• upper secondary school completion
• socioeconomic status
• Norwegian Patient Registry
• the compensatory advantage model
• Blaxter hypothesis

Introduction

Non-completion of upper secondary school is a major concern at the national (St. meld. nr. 21 (2016–2017)) and international (OECD, 2012) levels and has consequences for individuals and society (Lillejord et al., 2015). Young adults who do not complete secondary education more often experience difficulties in entering the labour market, and thus, face poorer long-term socioeconomic outcomes than their peers. Societal consequences include increasing social expenditure (Brekke, 2014; De Ridder et al., 2012) and possible marginalization and social exclusion. Furthermore, upper secondary school completion (SSC) varies greatly by parents’ socioeconomic class, i.e., adolescents who have parents higher up the socioeconomic ladder are more likely to complete secondary school. As an educational achievement in turn influences occupational careers, health-exposures and income over the life course, socioeconomic disparities in SSC likely acts as a vehicle for intergenerational reproduction of social inequality. This was acknowledged in a recent Norwegian government white paper: “The Government wants everyone to become the best possible version of themselves. The effort, hard work and talents – not your family background or where you live – should be decisive of one’s opportunities in life.” (Meld. St. 21 (2016–2017), p. 5). Thus, to reduce personal and societal costs, and promote equal opportunity in education, it is important to understand why students fail to complete upper secondary school, and in particular, how this is structured by parental socioeconomic status (SES) alongside individual traits.

Reports of increasing mental health problems (Mojtabai et al., 2016; Potrebny et al., 2019) and increasing rates of disability pensioning in young people in Norway (Bragstad, 2018) have attracted the interest of educational and public health researchers (Breslau et al., 2011; De Ridder et al., 2013; Esch et al., 2014; Evensen et al., 2016; Melkevik et al., 2016; Sagatun et al., 2014). The onset of mental disorders often happens in adolescence during the formative years of education (Isohanni et al., 2001). Unequal access to family resources during this period of life may create disparities in how mental health disorders affect SSC and subsequent transition to adulthood. Yet only a few studies have examined this question, i.e., how mental health disorders and parental SES interact in producing educational disparities (Brekke & Reisel, 2017; Evensen et al., 2016; Jackson, 2009; Mikkonen et al., 2020).

Thus, the aim of this paper is to investigate the relationship between adolescent mental health disorders, parental SES and SSC. We pursue following research questions: (1) To what extent are adolescents’ mental health disorders associated with elevated risk of not completing upper secondary school? (2) To what extent does socioeconomic background moderate the relationship between mental health disorders and upper secondary school completion? Unlike previous studies, which has mostly relied on survey data and self-reported symptoms, we use population-covering administrative data with objective mental health diagnoses extracted from the specialist healthcare. This approach circumvents common problems in survey analysis, in particular non-response related to mental illness and low SES, which might bias results and affect statistical power. Another advantage is that we have reliable information on parental education, occupational class and income, as well as grade point average (GPA) from primary education, which might otherwise be at risk for reporting errors.

Previous Research

The relationship between adolescent mental health and SSC seems to vary by type of mental health problem. Externalizing problems, characterized by symptoms directed outwards such as conduct problems and disruptive behaviour (Liu et al., 2011), have a strong negative association with SSC even after adjustment for family background and other mental health problems (Breslau et al., 2011; Esch et al., 2014; Evensen et al., 2016; Melkevik et al., 2016; Sagatun et al., 2014; Simson et al., 2021). For internalizing problems, reflecting symptoms directed inwards such as anxiety and depression (Liu et al., 2011), current evidence is more ambiguous (Esch et al., 2014; Melkevik et al., 2016).

For instance, a Norwegian study using data from a large regional population-based health survey, the Young HUNT Study, found a consistent negative relationship between externalizing problems and educational outcomes, but no association for internalizing problems when controlling for sibling fixed-effects (Evensen et al., 2016). Similarly, a register-based study using mental health diagnosed by general practitioners, found that although both externalizing and internalizing problems were related to secondary school dropout, there were doubts about whether internalizing problems had an impact independently of externalizing problems adjusted for parental education (Hetlevik et al., 2018). Simson et al. (2021), however, using administrative data on five full Norwegian birth cohorts, found comparable SSC estimates for both externalizing and internalizing disorders, even after extensive control for background factors and cohort and school fixed effects.

Substance use disorders in adolescence appear to be a strong indicator of non-completion. Results from a community-based longitudinal study showed that adolescent with substance use disorders were twice as likely to drop out of high school compared to those without this disorder, adjusted for e.g., parental education and family income (Johnson et al., 1999). A cross-national study in the World Mental Health Survey Initiative, carried out in sixteen countries, found strong associations between prior substance use disorders and non-completion in secondary school after adjustments for childhood adversities (e.g., parental psychopathology and family economic adversities) (Lee et al., 2009).

There are clear gender differences in mental disorders. Internalizing disorders are more common in girls from puberty and onward, whereas boys more often are diagnosed with externalizing disorders (Reneflot et al., 2018). However, only a few studies have addressed the relationship between mental health and SSC from a gender perspective, and results are mixed (Esch et al., 2014). Fletcher (2008) using data from a longitudinal survey of U.S. adolescents (Health Add) found that depressive symptoms in high school were significantly related to drop out only for female adolescents after adjustments of e.g., mothers education and family income. Also Sagatun et al. (2014) found that internalizing problems in the 10th grade was a significant predictor of non-completion of upper secondary school among girls but not boys controlled for parental education, income and marital status. Contrary, Miech et al. (1999) using longitudinal data from the Dunedin study did not find internalizing mental disorders to impede educational attainment measured at age 21 for neither boys nor girls, adjusted for family background factors and other mental disorders.

Few studies have investigated the possible interaction effects between health and parental SES on SSC, the main interest of this article. Evensen et al. (2016), using sibling fixed-effects models, effectively controlling for all time-invariant heterogeneity related to siblings’ shared family background, revealed that the negative effect of mental health on SSC did not vary by social origin or by parental education. Another Norwegian study, linking a health survey to longitudinal registry data did not find significant relationships between psychological distress and SSC in neither low- nor high-income families when controlled for all other included variables (e.g., parental education and health-related behaviour) (Brekke & Reisel, 2017). Conversely, a recent Finnish register study (Mikkonen et al., 2020) investigated the role of parental education and social origin in (mental) health selection in educational attainment using six birth cohorts. Sibling fixed-effects models showed that high parental education buffered against the negative impact of mental health disorders on SSC.

In the United States, Jackson (2009) used the National Longitudinal Survey of Youth 1997 data and found that the decrease in the odds of timely high school graduation associated with self-reported general health was greatest for non-Hispanic white adolescent compared to black peers, controlling household fixed effects. Thus, contrary to Mikkonen et al. (2020), the study does not support the idea that a favourable social background protects against the educational consequences of poor health, when defined by race/ethnicity. None of the studies above investigated how the interplay among adolescent mental health disorders, SES and SSC varies by gender.

Theoretical Framework

The compensatory advantage model (CAM) predicts that high-SES families use their resources to counteract their children’s disadvantages and thereby avoiding educational failure. While a disadvantage may accumulate over the school career for children from low-SES families, this may be less the case for children from more privileged families (Bernardi, 2014; Bernardi & Cebolla-Boado, 2014). Advantaged families can pay for facilities and assistance, e.g., private tutoring (Bernardi, 2012), or purchase quicker and better health services in the private market (Barneombudet, 2020).

Class-related values and practices in parenting, particularly related to education, have been studied extensively (Irwin & Elley, 2011). A Norwegian study found that high-SES parents respond to and intervene earlier in problems than low-SES parents. As a result, it seems that low-SES parents intervene after the problem has manifested (Stefansen et al., 2017). Thus, the different intervention styles across SESs might lead to disparities in the timing of professional help, amplifying existing inequalities in completion rates and subsequent social mobility patterns among those with mental health problems, in congruence with the CAM.

These intervention styles are related to the concept of health literacy dealing with how differences in knowledge, skills and attitudes toward health and health behaviour contribute to shape health inequality (Wharf Higgins et al., 2009). Low SES appear to be associated with low health literacy (Yin et al., 2009). Low parental health literacy is a concern, as adolescents often depend on their parent’s ability to stand up for them when facing a health problem (Lee et al., 2020) or to encourage a healthy lifestyle (Wharf Higgins et al., 2009). Therefore, adolescents from advantaged families might possess and have access to higher health literacy compared to adolescents from disadvantaged families, helping them to cope with their mental health issues and reducing the negative impact of poor health on SSC.

In contrast, the so-called “Blaxter hypothesis” builds on the assumption that the educational attainment of advantaged adolescents is equally or even more adversely affected by poor health than that of their peers (Jackson, 2009). A strong positive relationship between parent’s educational expectations and social class is well documented (Andres et al., 2007). These expectations may operate as a stressor (Banks & Smyth, 2015). In a recent Norwegian qualitative study, a group of adolescents from the financial middle class talked about their parents’ explicit demands for academic achievements and tied parental pressure to their achievement-related mental health problems. These adolescents portrayed vulnerable future selves that are at a high risk of failure unless they can constantly be “the best” (Eriksen, 2021). Mentally ill adolescents with advantaged parents may be more likely to feel like failures if they are not able to fulfil parental academic expectations. Thus, in line with the Blaxter hypothesis, a high-SES background do not buffer against the educational consequences of mental illness, but rather leads to a double burden.

Data and Methods

Data Sources and Sample Selection

This study uses administrative data for the entire Norwegian population drawn from the Norwegian Patient Registry (NPR), the Norwegian National Education Database, FD-Trygd and Statistics Norway. The NPR is managed by the Directorate of Health and contains data at the individual level from 2008 and onward (Bakken et al., 2020). Access to data has been approved in a Data Protection Impact Assessment performed by Oslo Metropolitan University and in ethical vetting by the Regional Committees for Medical and Health Research Ethics and the various register owners.

This study uses individual-level data on an entire birth cohort born in Norway in 1996, entering upper secondary school in 2012. We followed them through the registries from 2009 until 2017, that is, from the year they entered secondary school until the last year they were entitled to upper secondary education. The following individuals were excluded: those who were registered as deceased (N = 523) or migrated (N = 1 838) in 2017 and those with missing education (N = 981), GPA (N = 1 693) or household income (N = 160) data. Also, individuals registered in NPR with a mental health disorder not being internalizing, externalizing or substance use disorders were excluded (N = 1834). This left us with a study population of 53 187 individuals, who were linked to their biological parents through their national identity numbers.

Upper Secondary School Completion

The dependent variable was defined as whether the participant graduated from upper secondary school in 2017 at the age of 21 years. In Norway, students are entitled to three years of upper secondary education within five years.

Mental Health Disorders

The main independent variables are the following three health variables: internalizing disorders, externalizing disorders and substance use disorders. The three variables were coded as indicator variables, taking the value 1 if the individual was registered in the NPR with relevant psychiatric diagnoses between 2009 and 2015, when the individuals were between 13 and 19 years old.

The variable, Internalizing disorders, includes diagnoses of depression (ICD-10 codes F32-F34 and F38) and anxiety (F40-44 and F48). The measure for externalizing disorders included attention-deficit hyperactivity disorder (F90) and conduct disorder (F91). For the measure of substance use disorders, we included all mental and behavioural disorders due to psychoactive substance use (F10-19).

It is well recognized that many adolescents experience more than one mental health disorder (Costello et al., 2003). Thus, to provide unbiased results, it is important to account for comorbidity when examining the relationship between mental health disorders and educational achievement (Melkevik et al., 2016). Therefore, the three mental health variables are not mutually exclusive, meaning that individuals can appear in several disease categories simultaneously. Hence, we do not treat individuals with multiple disorders as a separate category, but focus on the independent effect of internalizing, externalizing and substance use disorders, respectively, on SSC.¹

Number of Months in Psychiatric Services

We identified the number of months the individual was registered in the NPR from 2009 to 2015 and constructed a continuous variable used in the regression analysis. Next, we constructed a dichotomy variable used in the descriptive analysis: “long-term contact with psychiatric services” (higher than the average number of observations by month) and “short-term contact with psychiatric services” (lower than the average number of observations by month).

Socioeconomic Background and Family Factors

Information about household income was taken from administrative registry data. The household was constructed using a household identification number. Using the EU standard (first adult = 1.0, next adult = 0.5 and children under 17 in the household = 0.3), equalized household income was computed for a five-year period (2010–2014). The average household income over the five years was computed as quartiles: high income, middle-high income, middle-low income and low income. Parental education was obtained from Statistics Norway’s education registry and measured when the participant was 16 years old. Parental education is the educational level of the parent with the highest education or of the only parent who is present. The variable was divided into three categories: low (compulsory), intermediate (upper secondary and post-secondary non-tertiary) and high (tertiary). Parents with missing information were included in the low category. Information about parental occupation was retrieved from FD-Trygd and coded into the variable Occupational Class using the European Socio-economic Classification (ESeC class).² If the parents were not registered in any occupation, due to e.g., unemployment, being a homemaker or disabled, they were classified as “Missing Class”.

Information about whether the adolescent in 2012 lived with both parents, with one parent or alone was constructed using household identification number retrieved from Statistics Norway. The variable was dichotomized into “living with both parents” or “living with one parent or alone”. The variable parent’s country of birth was grouped into three categories: one or two Norway-born parents, two foreign-born parents from EU/EØS, the United States, Canada, Australia and New Zealand or two foreign-born parents from Asia, Africa, Latin America and Europe except EU/EØS. The information was provided by Statistics Norway.

Educational Performance

Final GPAs in compulsory school in mathematics, and in Norwegian and English, were retrieved from the Norwegian National Education Database in 2012 and constructed as a continuous measure ranging from 1 to 6. In addition, a categorical variable used in the cross-tabulation analysis was defined by high or intermediate grades≥ 3 and low grades <3.

Statistical Methods

With a dichotomous dependent variable, logistic regression has been the standard approach. However, as logistic regressions models does not allow comparisons across models and samples, due to the influence of unobserved heterogeneity (Mood, 2010), we opted for linear probability models (LPMs). The LPM also has the additional benefit of substantively interpretable coefficients on the probability scale (Hellevik, 2009). Our stepwise regression analysis consists of three models, run separately for boys and girls. First, we estimate the mutually controlled association between the three mental health variables and SSC (Model 1) adjusted for family income, number of months in psychiatric services and GPAs. Model 2 introduces interaction terms between each mental health disorder and household income to examine whether SES modifies the relationship between mental health and SSC. Finally, Model 3 adds parents’ country of birth, parental education, family cohabitation status and parental occupational class, to account for possible confounding of the interaction effects presented in Model 2. As financial mechanisms like purchasing additional tutoring or mental health services may be particularly important, we use family income as our main measure of family SES. In a sensitivity analysis, however, we also present results using parental education, to better account for the multidimensional resources available to high-SES families (Bernardi, 2014). Table 1

Table 1. Descriptive statistics for full cohort, boys and girls (N = 53 187).

	Total cohort	Boys	Girls
Upper secondary school completion	42 517 (80)	20 976 (76)	21 541 (84)
Individual health factors
No mental health disorder	46 147 (87)	24 460 (89)	21 687 (84)
Externalizing disorder	2017 (4)	1313 (5)	704 (3)
Internalizing disorder	4800 (9)	1449 (5)	3351 (13)
Substance use disorder	892 (2)	471 (2)	421 (2)
Number of months in psychiatric services (mean)	10.7 (std. = 10.8)	9.8 (std. = 10.0)	11.4 (std. = 11.3)
Family factors
Family income
Low income (Q1)	13 298 (25)	6806 (25)	6492 (25)
Middle-low income (Q2)	13 296 (25)	6929 (25)	6367 (25)
Middle-high income (Q3)	13 297 (25)	6840 (25)	6457 (25)
High income (Q4)	13 296 (25)	6844 (25)	6452 (25)
Parents’ highest educational level
Compulsory	5543 (10)	2802 (10)	2741 (11)
Intermediate	22 621 (43)	11 690 (43)	10 931 (42)
Tertiary	25 023 (47)	12 927 (47)	12 096 (47)
Family cohabitation status
Living with both parents	33 608 (63)	17 417 (64)	16 191 (63)
Living with one parent or alone	19 579 (37)	10 002 (36)	9577 (37)
Parents’ country of birth
One or both parents born in Norway	50 679 (95)	26 142 (95)	24 537 (95)
Both parents born in EU/EØS, USA, Canada, Oceania	258 (1)	149 (1)	109 (1)
Both parents born in Asia, Africa, Latin America, European countries outside EU	2250 (4)	1128 (4)	1122 (4)
Parents’ occupational class
Higher salariat	8829 (16)	4585 (16)	4244 (16)
Lower salariat	12 140 (23)	6277 (23)	5863 (23)
Higher grade white collar workers	9750 (18)	5027 (18)	4723 (18)
Petite bourgeoisie or independents	1051 (2)	550 (2)	501 (2)
Petite-bourgeoisie or independents in agriculture	163 (0.5)	78 (0.5)	85 (0.5)
Higher grade blue collar workers	163 (0.5)	76 (0.5)	87 (0.5)
Lower grade white collar workers	11 242 (21)	5777 (21)	5565 (21)
Skilled workers	2023 (4)	1011 (4)	1012 (4)
Semi- and non-skilled workers	3065 (6)	1566 (6)	1499 (6)
Missing class	4761 (9)	2472 (9)	2289 (9)
Final GPA in compulsory school (mean)	3.9 (std. = 0.92)	3.7 (std. = 0.92)	4.1 (std. = 0.87)
Observations	53 187	27 419	25 768

Figures are numbers (percentages), unless stated otherwise.

Results

Descriptive Statistics

Table 1A (Supplementary Material) lists the completion rate in 2017 of students starting upper secondary school in 2012 stratified by gender. Of the study population, 76 percent of the boys and 83 percent of the girls completed upper secondary school within five years after finishing compulsory school. The completion rate was lowest for boys (38 percent) and girls (42 percent) diagnosed with externalizing disorders. In addition, adolescents diagnosed with internalizing and substance use disorders had significantly lower completion rates compared to healthy peers. Not surprisingly, the completion rate was two times higher for adolescents with high or intermediate grades relative to students with low grades.

For both genders, the completion rate was lower among those who had parents with low educational levels and among those from the lowest family income group, compared to those with parents with higher education and income.

Upper Secondary School Completion

Table 2 presents the regression results for SSC in 2017 for each of the three mental health measures stratified by gender. Starting with externalizing disorders, Model 1a and Model 1b show that boys and girls with this type of disorder, on average, were 17 percentage points less likely to complete upper secondary school compared to peers without any mental disorder, adjusted for comorbidity of other mental health disorders, number of months in psychiatric services, GPA and family income.

Table 2. Mental disorders and SSC. Gender stratified LPM regression analyses (N = 53 187).

	Boys	Girls
	Model 1a	Model 1b
Externalizing	-.17 (.000)	-.17 (.000)
Internalizing	-.13 (.000)	-.16 (.000)
Substance use	-.17 (.000)	-.12 (.000)
Constant	.76 (.000)	.77 (.000)
R^2	22%	24%
P-values in parenthesis

Model 1 controls for comorbidity of other mental health disorders, number of months in psychiatric services, GPA and family income.

Surprisingly, the results from Table 2 show that internalizing disorders are strongly related to SSC when adjusted for externalizing and substance use disorders. Boys with internalizing disorders were, on average, 13 percentage points less likely to complete upper secondary school compared to peers without a disorder, net of other variables (Model 1a). Girls with internalizing disorders were on average 16 percentage points less likely to complete upper secondary school, net of other included factors (Model 1b)

Looking at substance use disorders, boys and girls were at significantly greater risk (boys = 17 percentage points and girls = 12 percentage points) of not completing upper secondary school compared to healthy students, net of other variables.

Upper Secondary School Completion and Interaction with Family Income

Table 3 includes interaction terms between each mental health measures and family income in addition to adjustments of family factors. As the influence of controlling for family factors in Model 3 on the interaction terms was negligible, we refrain from detailed descriptions of the coefficients for Model 2. Starting with boys in Model 3a, family income did not moderate the relationship between externalizing and internalizing disorders and SSC, as none of the interaction terms are statistically significant. However, the interaction term between substance use disorder and highest family income was statistically significant, net of other factors. This means that the probability of SSC for boys diagnosed with a substance use disorder from the highest income group was 0.74.³ In contrast, boys with a substance use disorder from the lowest income group have a probability of 0.58 to complete upper secondary school.

Table 3. Mental disorders, family resources and SSC. Gender stratified LPM regression analyses (N = 53 187).

Boys
	Externalizing		Internalizing		Substance use
	Model 2a	Model 3a	Model 2a	Model 3a	Model 2a	Model 3a
Mental disorder	-.16 (.000)	-.16 (.000)	-.14 (.000)	-.13 (.000)	-.19 (.000)	-.18 (.000)
Interaction terms with household income (ref: Quartile 1)
* Quartile 2	-.03 (.229)	-.03 (.191)	.02 (.411)	.02 (.331)	-.02 (.644)	-.01 (.781)
* Quartile 3	-.01 (.804)	.00 (.843)	.00 (.998)	.00 (.925)	.04 (.336)	.04 (.408)
* Quartile 4	.05 (.126)	.04 (.161)	.01 (.652)	.01 (.674)	.10 (.036)	.10 (.041)
Household income (ref: Quartile 1)
Quartile 2	.06 (.000)	.04 (.000)	.06 (.000)	.04 (.000)	.06 (.000)	.04 (.000)
Quartile 3	.08 (.000)	.05 (.000)	.08 (.000)	.05 (.000)	.08 (.000)	.05 (.000)
Quartile 4	.10 (.000)	.06 (.000)	.10 (.000)	.06 (.000)	.10 (.000)	.06 (.000)
Constant	.76 (.000)	.76 (.000)	.76 (.000)	.76 (.000)	.76 (.000)	.76 (.000)
Adjusted R2	22%	23%	22%	23%	22%	23%
Girls
	Externalizing		Internalizing		Substance use
	Model 2b	Model 3b	Model 2b	Model 3b	Model 2b	Model 3b
Mental disorder	-.24 (.000)	-.22 (.000)	-.20 (.000)	-.19 (.000)	-.18 (.000)	-.17 (.000)
Interaction terms with household income (ref: Quartile 1)
* Quartile 2	.06 (.067)	.05 (.098)	.04 (.013)	.03 (.032)	.06 (.121)	.06 (.156)
* Quartile 3	.11 (.001)	.10 (.002)	.05 (.003)	.04 (.014)	.07 (.091)	.06 (.173)
* Quartile 4	.17 (.000)	.16 (.000)	.09 (.000)	.08 (.000)	.17 (.000)	.16 (.000)
Household income (ref: Quartile 1)
Quartile 2	.05 (.000)	.04 (.000)	.05 (.000)	.04 (.000)	.05 (.000)	.04 (.000)
Quartile 3	.07 (.000)	.05 (.000)	.06 (.000)	.05 (.000)	.07 (.000)	.05 (.000)
Quartile 4	.07 (.000)	.05 (.000)	.06 (.000)	.04 (.000)	.07 (.000)	.05 (.000)
Constant	.79 (.000)	.77 (.000)	.79 (.000)	.77 (.000)	.79 (.000)	.77 (.000)
Adjusted R2	24%	25%	24%	25%	24%	25%

P-values in parenthesis

Model 2 includes interaction terms. Model 3 includes additional controls for parental education, parental occupational class, immigrant status and family cohabitation status

For girls, the interaction terms between externalizing disorders and high and middle-high income were statistically significant, net of other variables (Model 3b). Girls with externalizing disorders from high-income families had a 0.76 probability of completing upper secondary school; girls from middle-high income have a 0.70 probability whereas girls with externalizing disorders from the lowest income group have a probability of 0.55.

All interaction terms between internalizing disorders and family income in girls were statistically significant, net of other variables (Model 3b). The average probability of SSC for girls with internalizing disorders from the highest income group was 0.70; girls from middle-high income group 0.67; from middle-low income 0.65 and girls from lowest income group 0.58.

Finally, the interaction term between substance use disorders and highest family income was significant, adjusted for other variables (Model 3b). The average probability of SSC was 0.81 for girls with a substance use disorder from the highest income group. In contrast, the probability of SSC was 0.60 for girls with a substance use disorder from the lowest income group.

We also examined the moderating effect of parental education (low parental education as the reference category) on the relationship between mental health and SSC (Table 5A in Supplementary Material). For boys, two interaction terms were significant: externalizing disorder and parental tertiary education (0.70 probability of SSC with high parental education in contrast to 0.56 probability with low parental education) plus internalizing disorder and parental intermediate education (0.72 probability of SSC with intermediate parental education in contrast to 0.58 probability with low parental education). For girls, the interaction between all three health measures and parental tertiary education were significant.⁴ The average probability of SSC for girls with externalizing disorders with high parental education was 0.75 in contrast to 0.57 with low parental education; girls with internalizing disorders with high parental education was 0.70 in contrast to 0.56 with low parental education; girls with substance use disorders with high parental education was 0.78 in contrast to 0.63 with low parental education. These results underscore the protective effect of family resources on SSC for girls, whereas for boys the differences seem less strong.

Discussion

Increasing the upper secondary school completion rate has been on the political agenda in Norway for years (Sletten et al., 2019). To achieve this goal, it is important to understand why young people fail to complete school. Previous researchers have documented a strong impact on SSC of parental SES and health, but the possible interaction between these factors has seldom been considered. Furthermore, few studies have investigated gender differences in these associations. This paper contributes to fill these gaps by investigating the role of different mental health disorders in SSC among boys and girls, and whether the associations vary by SES. In doing so, we shed light on two competing theoretical approaches, the CAM and the Blaxter hypothesis. The former approach holds that the unequal distribution of various resources amplifies socioeconomic disparities in the face of mental illness, as advantaged families are able to draw on these resources to facilitate academic performance. The latter approach, in contrast, predicts that vulnerable adolescents from advantaged families are prevented from exploiting resources, which compromise educational achievement to the same or even greater degree than disadvantage peers. In addition, privileged youth might blame themselves of not being able to live up to parental educational expectations creating a double burden.

The results show that, first, all three mental health measures are strongly associated with SSC after controlling for family income, comorbidity and GPAs. This result is consistent with previous research showing that externalizing and substance use disorders have a strong negative impact on SSC (Breslau et al., 2011; Esch et al., 2014; Evensen et al., 2016; Johnson et al., 1999; Melkevik et al., 2016; Sagatun et al., 2014; Simson et al., 2021). However, we also identified strong effects of internalizing disorders after control for externalizing and substance use disorders. This finding contradicts previous research which showed that the effect disappears when externalizing disorders are controlled for (Miech et al., 1999) or found significant effects for girls only (Fletcher, 2008; Sagatun et al., 2014). Conversely, the results suggest that anxiety and depression impact educational attainment in both genders, regardless of the presence of a comorbidity.

Second, we found that family income moderated the relationship between all three mental health disorders and SSC in girls, whereas in boys, family income only moderated the association between substance use disorder and SSC. Additional analyses showed that a high level of parental education had a protective effect for SSC for girls with mental health disorders, while the moderating effect of parental education was less pronounced for boys. In sum, high family income and a high level of parental education seem to protect against the detrimental effects of mental health disorders on SSC for girls. For boys, the picture is more ambiguous, indicating that boys do not benefit from family resources to the same degree as girls. Thus, for girls, our study corroborates the findings in Mikkonen et al. (2020) and provides support for the CAM. For boys, however, our results are more in line with the Blaxter hypothesis and previous research showing that mental health problems and SSC do not vary by parental SES (Brekke & Reisel, 2017; Evensen et al., 2016). In comparison to these previous papers, an advantage of our study is the gender stratified analyses, that perhaps explain some of the disagreements in previous research using mixed samples.

A possible explanation for the gender difference could be linked to the fact that adolescence boys seem to experience more social stigma and shame related to mental illness relative to female peers (Rice et al., 2018). Boys are less open about their mental health problems compared to girls (Grace et al., 2018), which might also increase the barrier to seek support in the family. Consequently, boys with mental health issues from advantaged homes could, in contrast to female counterparts, miss out on assets and support available in the family which otherwise would have contributed positively to educational attainment.

Related to this, research also suggest that boys are less likely to seek professional help for mental health disorders than girls (Granrud et al., 2020; Rice et al., 2018). Thus, it is possible that boys’ mental health is poorer when they finally get professional help, reducing the scope for family resources to make a difference, which might explain our results. Interestingly, this explanation seems at odds with the finding in Table 1 that boys spend less time in psychiatric services than girls, assuming that time in services is indicative of severity. However, as time in services may be influenced by other factors, such as help-seeking behaviour, diagnostic and therapeutic procedures and administrative decisions, we have doubts about the validity of this measure as an indicator of severity. Further research should investigate the mechanisms behind the gendered results found here in greater depth.

Overall, the results highlight family background as an important social context regarding SSC. This is in line with a large body of research documenting a strong relationship between SSC and parental SES (Bergsli, 2013; Lamb & Markussen, 2011). The strong relationship arguably reflects a process where parents and adolescents value education differently depending on social position (Boudon, 1974). According to this perspective, adolescents and their families make different academic choices according to social background in terms of avoiding downward social mobility (Hansen, 2005). In this respect, it is possible to assume that adolescents from advantaged homes strive for academic success and have high academic ambitions despite mental illness as well as the parents have strong incentives to compensate for their children’s disadvantages that could jeopardize educational attainment.

Our results suggest that financial resources shape social inequalities in SSC. Previous researchers have shown that the waiting time for children and adolescents who need psychiatric help is suboptimal in Norwegian specialist healthcare (Agledal et al., 2006). As private mental health services are also available, this situation creates leeway for economically advantaged families, more than low-income families, to address the needs of their children.

Geographic proximity to private and public health services is likely to be better in urban than in rural areas. As the urban–rural divide includes socioeconomic differences (i.e., families with higher income and education tend to cluster around the big cities⁵), adolescents from these families may have better access to healthcare. Thus, socioeconomic residential patterns may have biased the results. Fortunately, the data allow us to assess the impact of this on our findings using a pre-coded variable from Statistics Norway on geographic centrality. Results from this sensitivity analysis were not statistically significantly different from the main findings.

To our knowledge, this is the first Norwegian study examining the relationship between adolescent mental health, SSC and SES using population covering register data, and one of very few in the Nordic context (Mikkonen et al., 2020). While this approach has many advantages compared to survey-based studies, such as including an entire birth cohort, zero attrition and long follow-up, a limitation is that a register-based measure of health may describe the use of health services and not necessarily the actual prevalence of a condition or the extent of disease in the population in a given period. However, adolescent mental health disorders are mainly diagnosed in the sector of mental health facilities for children and adolescents (BUP); thus, the NPR provide good estimates for the proportions with diagnoses (Reneflot et al., 2018).

A second limitation is that high-SES individuals consult specialist healthcare to a higher degree than low-SES individuals, causing underreporting of mental disorders in the latter group (Godager & Iversen, 2013). Thus, the severity of the health problems we observe may differ in the two groups. If high-SES adolescents more often contact the specialist mental health services for milder disorders, this may have biased the estimated role of family resources in buffering SSC, found in girls. Unfortunately, our data does not allow adequate control for the severity of disorder; as pointed out earlier, the variable “number of months in psychiatric services” is probably not sufficient.

Third, NPR includes linkable information only from 2008 onward. Thus, there is a risk of diagnoses with early debut, such as ADHD, being underestimated in the cohort as these early debuts may be followed up in primary healthcare, rather than in secondary specialist healthcare, from which our data are drawn. Since more boys than girls are diagnosed with ADHD in childhood (Reneflot et al., 2018), the observed association between externalizing disorders and SSC for boys might be underestimated as boys with ADHD more often are included in the control group.

Fourth, even if we include control for academic achievements in compulsory school, we cannot completely rule out reversed causality, i.e., that negative experiences at school might have affected mental health, leading to an overestimation of the link between mental health and SSC.

Finally, we observed the sample population for only a short time after standard education completion. The effect will likely change in the longer term, as many complete educational training as adults, especially during their 20s (Halvorsen et al., 2012). This calls for future research investigating longer-term effects of the interplay between mental health, educational attainment and SES.

Conclusion

In this paper, we investigated the toll on SSC of internalizing, externalizing and substance use mental health disorders, and how these relationships are modified by SES for boys and girls. Using high-quality administrative data, we found substantially poorer outcomes associated with all three mental health disorders, in the range of 12 to 17 percent lower SSC rates, after controlling for GPA’s and other covariates. We also found that parental income and educational level amplifies existing inequalities in SSC among girls with mental health disorders, whereas for boys the protective effect of family resources were more ambiguous. The apparent lack of an impact of family resources on the relationship between mental health and SSC among boys may support the Blaxter hypothesis that mental illness hinders adolescents from advantaged families in utilizing resources available within the family. They might even experience a “double burden” as they portray themselves as failures if they do not fulfil parental academic expectations. But we see no reason why this should be true only for boys. This calls for future research investigating the mechanisms driving these gendered results.

Acknowledgements

Data from Statistics Norway and the Norwegian Patient Registry has been used in this publication. The interpretation and reporting of these data are the sole responsibility of the authors, and no endorsement by Statistics Norway or the Norwegian Patient Registry is intended nor should be inferred. The authors would like to thank participants at the Nordic Working Life Conference in 2020 for valuable comments.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The work was undertaken as part of the C-LIFE and WELLIFE projects, funded by Nordforsk [grant numbers 75970 and 83540].

Notes

1 627 individuals were registered with multiple psychiatric disorders in the sample

2 https://www.iser.essex.ac.uk/archives/esec/user-guide/the-european-socio-economic-classification

3 The probability is calculated by adding the constant, mental disorder, household income and interaction term (p = a + b1 + b2 + (b1 × b2)).

4 Note that the interaction term between substance use disorder and tertiary educational level was significant at a 5,5% level, so slightly above the 5 percent.

5 https://ssb1.maps.arcgis.com/apps/MapSeries/index.html?appid=2e2d96014fb846938d10847b3f5fc4ca