Aim and Setting of the Study
Given the above, we set out to: (a) investigate the risk factors associated with emotional symptoms among girls aged 11–12 years, examining these jointly to isolate their unique contributions; and (b) assess whether exposure to a greater number of risk factors corresponds to higher levels of symptoms in this population. Such investigation contributes to knowledge by isolating unique risk associations with emotional symptoms, overcoming various methodological challenges present in prior evidence, and offering population-specific estimates of risk within a vulnerable group. We focus on girls specifically, rather than seeking to establish gender differences, given consistent evidence of high rates of symptoms among girls and women and indications of early adolescence as a vulnerable period. Thus, investigation of the particular factors contributing to symptoms among early adolescent girls offers insights into a specific phenomenon within a vulnerable group, rather than offering a “comparative” discussion. We draw on data collected in 2017 for the evaluation of HeadStart, a large-scale programme exploring ways to improve young people’s mental health and wellbeing. Use of secondary data offers several strengths; this dataset comprises a variety of explanatory variables and a large sample spread across a range of settings in England. We also note a key limitation inherent to all secondary analyses: as study variables were predetermined, we were unable to capture all factors of potential interest (e.g., biological factors, such as adrenal hormones).
Participants
The sample comprised 8,327 girls aged 11–12 years (M = 12.04, SD = 0.29) across 100 English education settings. Ethnicity was similar to the national secondary school composition (28); most participants were White (n = 6,217; 75.9%), followed by Asian (n = 885; 10.8%), Black (n = 472; 5.8%), mixed (n = 344; 4.2%), other (n = 131; 1.6%), and Chinese (n = 15; 0.2%). The remaining 1.5% (n = 122) had incomplete ethnicity information. Free school meal (FSM) eligibility (n = 1,436; 17.2%) was higher than national levels (14% (28)).
Measures
Emotional symptoms
The self-report Strengths and Difficulties Questionnaire (SDQ) emotional symptoms subscale (29) was used. There are five items (e.g., “I worry a lot”) with three responses: “not true” (0), “somewhat true” (1), and “certainly true” (2). Summed scores range from 0–10, with higher scores indicating greater symptoms. Research has indicated acceptable psychometric properties for this subscale (30). Here, Cronbach’s α was .72 and confirmatory factor analysis indicated acceptable fit: c2 (5) = 255.28, p < .001; root mean square error of approximation (RMSEA) = .08, 90% CI [.07, .09], p < .001; comparative fit index (CFI) = .98, Tucker-Lewis Index (TLI) = .95.
Risk variables
Table 1 shows the measure used for each candidate risk factor, along with the approach to dichotomising data; all risk variables were obtained from the National Pupil Database (NPD), except caregiving responsibilities, which was self-reported.
[Insert Table 1 here]
Procedure
Ethical approval was granted for the HeadStart evaluation by University College London’s ethics committee (reference 8097/003). Participants completed self-report measures in March–July 2017. Information sheets were provided to parents/carers and opt-out parental consent was used (114 girls opted out). Participants were presented with age-appropriate information and gave informed assent prior to completing by ticking a box to proceed. Surveys were administered online in teacher-facilitated sessions in participating schools.
Statistical Analysis
Analysis was undertaken using structural equation modelling in Mplus 8.1, using a robust weighted least squares (WLSMV) estimator to model emotional symptoms as a latent variable with categorical indicators (31). As data were gathered from participants across 100 settings (mean cluster = 83), clustering was controlled for using Type = Complex (intracluster correlation coefficients = .00–.40). RMSEA values below .06 and/or with 90% confidence intervals below 1.0, and CFI and TLI values above .95, indicated acceptable model fit (32,33). First, a linear multiple regression model was specified with risk variables predicting emotional symptoms. Variables were confirmed as risk factors where coefficients were positive and significant (p < .05).
Next, confirmed risk factors were collated to create a CRE index, in line with guidance that a CRE index should comprise only empirically confirmed sample-specific risk factors (rather than all theorised variables) given the contextual specificity of risk (10). Factors are coded as “1 = risk present” and “0 = risk absent” and summed (wherein a score of 1 denotes exposure to one risk factor, a score of 2 exposure to two risk factors, etc.). This index was then modelled as a predictor of symptoms to examine whether greater CRE is associated with increased symptomatology (10). Risk factors were then added in turn as covariates to confirm that any effects were not driven by any one factor (10).