Design and study population
Data stem from the Danish arm of the international Health Behaviour in School-aged Children (HBSC) study (Inchley et al. 2020). The study design included nationally representative samples of students in the age groups, 11-, 13- and 15-year-olds. We pooled data from the last three surveys (2010, 2014 and 2018) to increase the statistical power. The three surveys were comparable because they adhered to similar protocols for sampling, measurement, and data collection. The surveys used cluster sampling, i. e. sampling of schools. We invited random samples of schools, a new sample in each survey, drawn from complete lists of public and private schools in Denmark. In each school we invited all students in the fifth, seventh, and ninth grade (corresponding to the age groups 11, 13 and 15 years) to participate and complete the internationally standardized HBSC questionnaire in the classroom (Rogers et al. 2009). The participation rate among students across all three surveys was 85.7% (n = 15,310), calculated as the percentage of students enrolled in the participating classes who completed the questionnaire. Students in the fifth grade in 2010 (n = 1,839) were not asked about loneliness, resulting in an eligible study population of 13,471. The analyses in this manuscript included students with complete information about loneliness, close relationships, and socio-demographic co-variates (n = 10,425).
Measurements
The study measured loneliness with one item, “Do you feel lonely?” (never, sometimes, often, very often), dichotomized into no (never, sometimes) and yes (often, very often). In adolescent samples, this one-item measure correlated strongly with the four-item version of the University of California Los Angeles (UCLA) Loneliness Scale (Eccles et al. 2020, Mund et al. 2023). Further, the one-item measure and the four-item scale showed remarkably similar patterns of associations with measures of health, sleep, and scholastic self-beliefs (Eccles et al. 2020, Eccles et al. 2021). A study by Mund et al. (2023) also showed that such a direct single-item measure correlated highly with other measures of loneliness. These findings suggest that the measure is valid for the purpose of our study.
The study used the HBSC Ease of Communication Measure to define close relationships. This measure has been used for analyses of young people’s social life (Arnarsson et al. 2019, Brooks et al. 2015, Damsgaard et al. 2014, De Looze et al. 2020, Levin et al. 2012) and asks: “How easy is it for you to talk to the following persons about things that really bother you? … a) father, b) stepfather (or mother’s boyfriend), c) mother, d) stepmother (or father’s girlfriend), e) best friend, f) friends of the same sex, and g) friends of the other sex” (response categories: very easy, easy, difficult, very difficult, don’t have or see this person). We recoded the responses into three measures: first, responses to communication with parents (items a-d above) were categorized into four levels: 1) having very easy communication with at least one parent, 2) having at easy communication with any parent, 3) having difficult communication with any parent, and 4) having very difficult communication with any parent. Second, communication with friends, based on item e-g and categorized into four levels as above. Third, a combination of the two measures with four categories: having very easy or easy communication with at least one parent and at least one friend; having very easy or easy communication with at least one parent but no friend; having very easy or easy communication with at least one friend, but with no parent; and not having very easy/easy communication with any parent or friend.
The analyses included four socio-demographic co-variates: sex, age group (11, 13, 15 years), survey year (2010, 2014, 2018), and socioeconomic status defined as the parents’ occupational social class (OSC). OSC was measured by eight items: “Does your father/mother have a job?”, “If no, why does he/she not have a job?”, “If yes, please say in what place he/she works (for example: hospital, bank, restaurant)” and “Please write down exactly what job he/she does there (for example: teacher, bus driver)”. The research group coded the answers following the Danish Occupational Social Class measure (Christensen et al. 2014). Each participant was categorized by the highest-ranking parent into four levels of family OSC: 1) high, e. g. professionals and managerial positions, large-scale business owners and managers, 2) middle, e. g. technical and administrative staff, small-scale business owners and managers, skilled workers, 3) low (unskilled and semi-skilled workers and economically inactive, i. e. parents receiving transfer income) and 4) unclassifiable due to insufficient information. Most participants (88.1%) gave sufficient information for the coding of OSC. Several studies show that most schoolchildren in these age categories can report their parents' occupation with a high agreement with parents’ own information (Ensminger et al. 2000, Lien et al 2001, Pu et al. 2011, Pueyo et al. 2007, West et al. 2001). Pförtner et al. (2015) showed that OSC is a suitable variable for studies of social inequality in adolescents’ health.
Statistical procedures
The first step was inspection of data by crosstabulations and use of chi2-test for homogeneity. The second step was logistic regression analyses to examine the association between ease of communication and loneliness, adjusted by sex, age group, survey year, and OSC. The third step was examination of potential effect modification by means of stratified logistic regression analyses, supplemented by test for statistical interaction. The logistic regression analyses accounted for the cluster sampling by multilevel modelling (PROC GLIMMIX in SAS). Prevalence rates in Figs. 1–3 were standardized for sex and age group.