Participants

The PISA 2018 survey was conducted by the Organisation for Economic Cooperation and Development (OECD). We used the PISA survey data collected between March 2018 and August 2018. More than 600,000 adolescents aged 15–16 years attending secondary education participated worldwide in the 2018 survey [21]. In recent years, the OECD has put more focus on carrying out research on adolescents’ wellbeing (e.g., life satisfaction) and social factors (e.g., exposure to bullying). PISA 2018 adopted a two-stage stratified sample design where schools were sampled systematically with probabilities proportional to the number of adolescents enrolled in the school. The second stage involved adolescents being randomly sampled within those schools and weights allocated to ensure the surveyed sample was representative of adolescents in the population. More details on PISA 2018 and the sampling method can be found in the technical report and user guide [21]. In the present sample, we excluded 10 countries due to missingness of the outcome and/or the exposure variables of interest. We excluded 7 cities and economic regions to ensure cross-national comparisons. In total, we excluded 17 countries resulting in a sample comprising of 479,685 adolescents from 63 countries. A participant flow diagram is given in Supplemental Fig. 1.

Data and measuresOutcome variables – psychological distress and life satisfaction

Psychological distress was measured using 4 items that asked the adolescents how often they felt (1) sad, (2) miserable, (3) scared, or (4) afraid on a 4-point frequency scale that ranged from never to always. We included adolescents who answered a total of 2 or more items. Mean imputation was applied for missing items. The overall score for psychological distress ranged from 4 to 16 after multiplying the score by 4 to give the original range. Life satisfaction was assessed with the question ‘overall, how satisfied are you with your life as a whole these days?’. Life satisfaction was measured on a frequency scale from 0 to 10 where 0 was ‘not at all satisfied’ and 10 ‘completely satisfied’. We excluded adolescents who did not answer the life satisfaction question.

Exposure variable – bullying victimisation

Adolescents were asked ‘During the past 12 months, how often have you had the following experiences in school? (some experiences can also happen in social media)’. Adolescents completed six items including ‘I got hit or pushed around by other students’, ‘I was threatened by other students’ and ‘Other students left me out of things on purpose’. The six items were grouped into corresponding subtypes to allow analysis by subtypes (full set of bullying items and subtypes are given in Supplemental Table 1).

Frequency was assessed on a 4-point scale: never or almost never (1), a few times a year (2), a few times a month (3), and once a week or more (4). We included adolescents who answered 3 or more items out of 6 for the total scale, and at least 1 out of 2 items per subtype i.e., those that answered less than 50% were excluded. Mean imputation was used for missing items. To create the ‘total bullying victimisation’ (hereafter referred to as ‘bullying victimisation’) score, responses to all six options were summed (range: 6–24). To create scores for each subtype, the two items were summed (range: 2–8). All sum scores were standardised to a mean of zero and a standard deviation of one.

Country-level factors:Bullying prevalence

We defined high bullying prevalence as exposure to bullying a few times a month and/or once a week or more [22]. A binary variable was created for each respondent (i.e., bullied or not bullied). The mean prevalence score was calculated per country.

Income inequality

We used the Gini index as the country-level income inequality predictor for this study (henceforth, we use income inequality and the Gini index interchangeably). The Gini index is a widely used measure of inequality in income distribution and ranges from 0 (complete equality) to 100 (complete inequality). For countries with missing Gini index values in the year 2018, we used available data within a 10-year period (2008–2018), retrieved from the World Bank dataset on the 13th January 2023 [23].

National wealth

We used the Gross domestic product per capita based on purchasing power parity (henceforth GDP) as the national wealth indicator for this study. GDP was selected as the best means of comparing country wealth as it accounts for differences in price levels between countries [24]. For each country included in this study, we sourced the GDP from the World Bank 2018 dataset, retrieved 13th January 2023 [23].

The sample characteristics are given in Supplemental Table 2.

Covariates

Gender (girl = 0, boy = 1) was adjusted for as there are gender differences in both bullying victimisation [25] and mental health [26].

Socioeconomic status was adjusted for using the Economic Social and Cultural Status (ESCS) PISA index variable derived from measures of parental education, highest parental occupation and home possessions [21]. There is evidence to suggest associations between increased likelihood of bullying for both low parental education [27] and low parental occupation [28] during childhood and adolescence. Research also reports associations between mental health problems in children and adolescents and low socioeconomic status, indicated by household income, parental education and parental unemployment [29].

Analyses

We conducted multilevel regression analyses within the statistical software R version 4.1.2, package lme4 [30]. The PISA-recommended weight variable ‘SENWT’ was used to allow inferences to the target population in each country.

Supplemental Table 3 details the multi-level modelling approach taken to investigate how the relationships between bullying victimisation, psychological distress and life satisfaction vary between countries. Gender and ESCS were controlled for in all models. Firstly, a Baseline Model was created which included only the outcome variable and covariates. This model acted as a reference to estimate the magnitude of variation at each level. We used the Baseline Models for each outcome to calculate the Intraclass Correlation Coefficient (i.e. the proportion of the total variance explained by the variation between countries) [31]. For psychological distress, 7.8% of the variation was explained by cross-country variation and for life satisfaction this variation was 5.9%. In Model 2, bullying victimisation was added as a predictor. This model tested the fixed effects of bullying victimisation on psychological distress and life satisfaction. Bullying victimisation was added as a random slope in Model 3. We tested the significance of the random slope by comparing the model fit between Model 2 and Model 3 (i.e., with or without a random slope) using a log-likelihood difference test. This tested the hypothesis that the strength of the associations between bullying victimisation and poor mental health varies across countries. We included the moderators (i.e., country-level factors) in Models 4, 5 and 6. In Model 4, we included bullying prevalence as a fixed main effect and an interaction term with bullying victimisation. Models 5 (the Gini index) and 6 (GDP) took the same approach.