Gender inequalities in trajectories of depressive symptoms among young people in London and Tokyo: a longitudinal cross-cohort study

Study design and participants

TTC is a population-based cohort study of adolescent health and development based in three districts of Tokyo (Setagaya, Mitaka, Chofu) (http://ttcp.umin.jp/).¹³ Households with a child born between September 2002 and August 2004 were identified from resident registries and recruited via random sampling (with oversampling of low-income households in anticipation of lower participation). From 2012-14 (T1), trained interviewers visited participant households every two years to administer questionnaires. Participating young people were approximately age 9-10 (T1, 2012-14), 11-12 (T2, 2014-16), 13-14 (T3, 2016-18), and 15-16 (T4, 2018-20) years old. 3,171 child-parent pairs entered the study at T1; 3,007 participated at T2 (94·8%), 2,667 at T3 (84·1%), and 2,616 at T4 (82·5%). Caregivers and young people gave informed written consent/assent at each wave. All procedures are approved by the Ethics Committees of the Tokyo Metropolitan Institute of Medical Science (no.: 12-35), the University of Tokyo (10057), and SOKENDAI (the Graduate University for Advanced Studies; 2012002).

REACH is an accelerated cohort study of adolescent mental health based in two inner-London boroughs, Lambeth and Southwark (www.thereachstudy.com).¹⁴ Three representative cohorts of young people – initially aged 11-12 (Cohort 1), 12-13 (Cohort 2), or 13-14 (Cohort 3) years – were recruited from 12 mainstream secondary schools (T1, 2016-17) and followed-up one (T2, 2017-18) and two (T3, 2018-19) years later. At each wave, young people provided informed written assent before completing computerised questionnaires. 4,353 participated at T1; 4,005 (92%) participated at T2 and/or T3. All procedures are approved by the Psychiatry, Nursing and Midwifery Research Ethics Subcommittee, King’s College London (15/16-2320).

We used the three waves of data from each study that maximise comparability of calendar years and ages, i.e., TTC: T2-T4; REACH T1-T3.

Choice of primary measure

The Short Mood and Feelings Questionnaire (SMFQ) is a free, widely used, unidimensional 13-item self-report measure of depressive symptoms in the past two weeks.¹⁶ Example items include: ‘I cried a lot’ and ‘I felt miserable or unhappy’. Each item has three response options: true (0), sometimes true (1), not true (2), from which a total score (0-26) is calculated. Higher scores indicate greater depressive symptoms. The SMFQ was originally developed in English and validated against diagnoses of depression in US paediatric samples.¹⁶ The Japanese version was developed for TTC using back-translation. (To aid readers’ interpretation of effect size, longitudinal within-person analyses of REACH data suggest SMFQ scores increase by around two points on average in those who become a victim of bullying (unpublished).)

Gender

In TTC, gender/sex (boy, girl) was obtained via resident register at recruitment. In REACH, gender/sex (boy, girl) was self-reported.

Data analysis

First, we examined distributions and missingness on included variables, and produced summary statistics overall and by cohort, gender, and gender-within-cohort, at each wave. Second, we fit a series of ordinal-indicator multigroup and longitudinal confirmatory factor analysis (CFA) models to examine configural, metric (loadings), and scalar (thresholds) invariance across cohorts/places, genders, and ages/waves (Figures S1 and S2).¹⁷ To assess model fit, we considered fit indices in combination rather than in isolation, gave weight to indices that are not sensitive to large sample size, and used proposed cut-offs as a guide (i.e., RMSEA≤0·08, CFI,TLI>0·90, and SRMR<0·08 as acceptable fit; RMSEA<0·06, CFI,TLI≥0·95 as good fit).^17,18 With the addition of each set of equality constraints, we considered a change in RMSEA (ΔRMSEA) <0·015, ΔCFI ≤0·010, and ΔSRMR ≤0·030 as support for invariance.¹⁹ This is consistent with the approach used by others with similar measures, ordinal indicators, large samples, and cross-cohort designs.^9,20 As the 13 SMFQ items have 3-level ordinal response options, we used Weighted Least Squares Mean and Variance-adjusted (WLSMV) estimation (polychoric correlations) to estimate covariance matrices.

Third, we used latent growth curve models (LCMs) to (a) determine the best-fitting functional form (i.e., intercept-only vs linear growth), and (b) estimate mean trajectories of SMFQ scores from age 11 to age 16 years in boys and girls in each cohort. For each growth parameter (i.e., intercept and linear slope), we added a gender*cohort interaction to formally test our hypothesis that gender inequalities are larger among young people in London than in Tokyo. We used full information robust maximum likelihood to accommodate missing data and the non-normal distribution of SMFQ scores. In sensitivity analyses, we (a) repeated the LCM with square root transformed SMFQ scores, because kurtosis in the raw score distribution was high and toward or above generally accepted levels, and (b) re-ran the LCM with the REACH cohorts treated as three distinct groups (given the accelerated design). Analyses were conducted in Stata 17 and Mplus 8.

Youth involvement

To co-write this paper with young people (see author list), we (1) held several meetings and workshops with five young people in London (age 18-19) and five in Tokyo (age 19-20) to discuss findings and explore possible interpretations, explanations, and implications, (2) integrate their perspectives throughout, and (3) present their ideas in the Supplement (S6).

Role of the funder

The funders had no role in study design, data collection, analyses, interpretation, writing, or the decision to publish.

Descriptive information

In total, 7,100 young people (TTC, 2,813; REACH, 4,287) were included in this analysis. Half were boys (n=3,587, 50·5%), half girls (n=3,513, 49·5%). In TTC, reasons for being excluded from analyses (n=358 of 3,171 who took part at T1) were: lost to follow-up after the first wave (n=164) or did not complete the SMFQ (n=194). In REACH, reasons for being excluded from analyses (n=658 of 4,945 invited to take part at T1) were: opted-out by parent/carer (n=160): child declined to participate (n=20); persistent absence from school (n=80); or did not complete the SMFQ (n=398). The included samples remain highly representative of the target populations based on socio-demographic indicators.^13,14 The REACH sample comprises high proportions from ethnically minoritised groups (n=3613, 84%), over a quarter eligible for free school meals (an indicator of low household income; n=1,183, 28%), around half boys (n, 2097, 49%), and roughly even proportions from each of the three REACH cohorts (Cohort 1: n=1,580; Cohort 2: n=1,382; Cohort 3: n=1,325). The TTC sample comprises just over half boys (n, 1490, 53%), 9% (n=242) with a household income <4 million yen (an indicator of low household income; the tax-free threshold is just under 2 million yen per person), and most had two Japanese parents (n, 2616, 97%). Summary statistics for SMFQ items and scores at age 11–12, by cohort and gender, are presented (Table 1, Figure 1).

Measurement invariance

The unidimensional factor structure was well supported in both cohorts and genders, and at all waves/ages. For example, in the cross-cohort multigroup CFA at age 11-12 years, the configural model fit well (CFI, 0·970; TLI, 0·965; SRMR, 0·048; RMSEA, 0·064; Table 2) and all item loadings were high (0·6 to 0·9) in both cohorts (Table S3). Overall, loading and threshold (i.e., scalar) invariance across cohorts and genders was well supported (Table 2). For instance, constraining item loadings and thresholds to be equal across cohorts at age 11-12 led to change in fit that was within the acceptable range across all fit indices (e.g., SRMR changed from 0·048 [configural], to 0·050 [metric], to 0·051 [scalar]), and the overall fit of the scalar model was good (CFI 0·964, RMSEA 0·065). Support for scalar invariance was generally strong across all fit indices in each CFA, with two exceptions: (a) in cross-cohort invariance tests at age 15-16, ΔCFI and ΔSRMR were comfortably within the acceptable range, whereas ΔRMSEA for the metric model was at the upper limit of what is generally considered acceptable (ΔRMSEA: 0·016); and (b) in models examining invariance by gender within REACH at age 15-16, the overall fit of the scalar model was good according to most indices (i.e., CFI and TLI 0·969, SRMR 0·066) but at the upper limit based on the RMSEA (0·090, 90% CI: 0·083, 0·097)). Longitudinal invariance of loadings and thresholds over time within each cohort was also well-supported (Table 3; Mplus syntax provided in S4). For example, in TTC, change in fit was comfortably within the acceptable range when constraining loadings and thresholds to be equal over time (e.g., ΔCFI 0·004; ΔRMSEA 0·002).

Depressive symptom trajectories

Key findings

We found strong evidence that gender inequalities in depressive symptoms are larger, and may emerge earlier, among young people in London than among young people in Tokyo. Notably, the mean rate of increase in depressive symptoms per year was around four times steeper among girls in London than among girls in Tokyo. We found little evidence to suggest these differences are due to incomparable measurement. Our work extends and strengthens a growing body of evidence on the context dependent nature of gender inequalities in emotional health in several important ways: robust consideration of measurement equivalence; the longitudinal design spanning the age at which gender inequalities in emotional health typically emerge; large representative samples of young people in places that, at nation-level, rank far apart on societal gender equity metrics; and the involvement of young people in writing this paper.

Limitations

Nonetheless, several limitations should be noted. First, gender is not a binary construct, however the available cross-cohort data limited our investigation to comparisons of boys and girls. We use the term gender, rather than sex, because our hypotheses are grounded in social (not sex-linked biological) causes of gender inequalities in mental health. Second, being limited to three waves meant we were only able to model linear trajectories. With additional waves of data, it will become possible to model non-linear growth, which may provide a better fit in TTC. Third, while measurement invariance tests can identify individual items or groups of items that function differently across groups, it cannot detect systematic under-reporting on each and every item. As highlighted by young people in our team (see S6), this is an important caveat. The extent to which this potential issue contributes to the differences we report, and in which direction(s), is unclear; cross-cohort qualitative work may be a useful next step.

Findings in context

Our findings are consistent with a growing body of evidence on the context-dependent nature of gender inequalities in mental health.^5,7,9 Using cross-sectional data on over 500,000 adolescents aged 15 years from 73 countries, Campbell and colleagues (2021) found substantial variation in the size of gender inequalities in mental distress and mental wellbeing across countries.⁹ Consistent with our findings, they also found that gender inequalities in mental distress tend to be larger in countries that perform better on formal indices of societal gender equity. In their analyses, the UK (currently ranked 15^th in the GGGI¹⁵) had one of the largest gender distress gaps, whereas Japan (ranked 125^th) had one of the smallest.

Collectively, the evidence suggests there are social and structural drivers underpinning elevated rates of emotional distress among women and girls. But how might we explain the paradoxical finding that ostensibly more gender-equal societies have greater gender inequalities in depression? In relation to adults, leading theories include ‘mismatch between reality and expectation’ and ‘role overload’. While social expectations for women to succeed in education and employment have greatly increased, traditional attitudes placing domestic work and childcare responsibilities predominantly on women remain firmly entrenched.²¹ As noted by our Young Researchers, indices like the GGGI are limited as they only capture certain aspects of gender inequity; they do not paint the full picture. The GGGI assesses gender (in)equity in health, education, economic participation and political representation, but does not include things like unpaid work, gender-norm attitudes, and gender-based violence. Sexual harassment, assault, and intimate partner violence (IPV), are strongly associated with mental ill-health and are common in countries that perform well on the GGGI¹⁵, including among young people.⁸ Though comparable data is scarce, there are indeed indications that prevalence of and attitudes toward gender-based violence is better in Japan than in the UK/Great Britain (GB), e.g., lifetime prevalence of IPV is lower, as is the proportion of women who agree that IPV is ‘justified in certain circumstances’.²²

While such factors are evidently relevant to older adolescents and young adults (e.g., 16-25-year-olds engaged in unpaid care are disproportionately female²³ and, as discussed earlier, sexual violence explains part of the gender mental health gap in mid-to-late adolescence⁸), it is less clear how such factors could explain the emergence of gender inequalities in mental health by age 11. In part, this is because we lack data on relevant social experience (e.g., domestic work, gender-based violence, gendered parenting, etc.) among girls of this age; most of the research in this area has focussed on biological explanations, particularly those related to puberty.^1,2 Evidence that sex hormones are responsible for the divergence of emotional health between boys and girls is inconsistent, correlational at best, and the mechanisms are unknown.²⁴ Very few studies have investigated the possibility that the mechanism linking puberty to the divergence of emotional health between boys and girls in early adolescence could be exogenous, e.g., driven by the way society treats girls as they develop.²⁵ Consequently, potentially modifiable targets to prevent gender inequalities in mental health from emerging at this age remain under-investigated, under-theorised, and unidentified.

Among our Young Researchers, there was a strong sense that young people in London may assume ‘adult’ roles and responsibilities at a younger age compared with young people in Tokyo who may typically be more ‘protected’ from such responsibilities during high school. In the UK, young girls and those in low income and financially insecure families do assume ‘adult’ responsibilities such as unpaid care at younger ages than do boys and those from more affluent backgrounds.²³ Given the relative social disadvantage of young people in REACH compared with TTC, it is possible that (gendered) differences in young people’s roles and responsibilities in the home and the community contribute to disproportionately high depressive symptoms among girls in London. Considering Japan’s high gender equity in health and secondary education outcomes but low equity in economic and political participation,¹⁵ we speculate that in Tokyo the gender depression gap will widen later, around early adulthood, as young women who experience comparatively equal treatment throughout compulsory education are confronted with structural barriers in work, career advancement, and childcare. We will test this hypothesis when the cohort reaches adulthood.

In the UK, there is much concern about the possible impacts of social media use and academic pressure on the mental health of young people, particularly girls. We were unable to find comparable data for our two cities/countries, but discussions with our Young Researchers suggest social media use is ubiquitous – and perhaps similar in nature – in both places, and more broadly we note that gender inequalities in emotional health predate the invention of social media and smartphones.¹ However, we cannot rule out that differences in use and impacts may contribute to our findings. A recent systematic review of qualitative evidence in the UK suggests that academic expectations and pressures are greater among girls than among boys.²⁶ Our Young Researchers suggest academic pressure is universally high in Japan, potentially reducing the gender disparity in comparison, though again comparable data is scarce.

Finally, societal and structural differences may contribute to the overall higher levels of depressive symptoms among young people in London compared with Tokyo: violence and interpersonal crime is significantly lower in Tokyo;²⁷ experience of racism is common among young people in London²⁸ (and London is far more ethnically diverse than Tokyo); and the UK’s prolonged period of austerity – and related increases in child poverty and inequality and chronic underfunding of public services supporting young people – is not mirrored in Japan. These interlinked structural harms are strongly associated with mental distress among young people.^28,29 Whether they also contribute to larger gender inequality in depressive symptoms and steeper increases in emotional distress among girls in London is unclear. We will explore the possibility of harmonising data on social conditions and experiences in TTC and REACH to investigate the likely confluence of factors that may explain our findings.

Implications

In public debate, differences in emotional health between teenage boys and girls are often considered a fact of life, an inevitable part of teenage development. Our own and others’ work increasingly suggests this may not be case. If high and rising rates of emotional ill-health among girls and young women³⁰ are largely driven by social and structural factors, then individual-focussed interventions (e.g., mindfulness, talking therapies, mental health first aid) – which currently dominate the UK’s response to the problem – are unlikely to be an effective solution.

Conclusions

Our findings suggest gender inequalities in emotional health are context dependent and the steep increases in emotional distress experienced by many girls in the UK are not an inevitable part of development. There is an urgent need to understand the contexts and conditions that enable young girls to thrive.