The total student population at the time of data collection was 650. Of the 600 questionnaires administered, 555 were returned, representing a response rate of 92.5%. Two substantially uncompleted questionnaires were subsequently excluded.

Ghana’s educational system operates on a 6-3-3-4 system (i.e. primary school – 6 years, junior high school – 3 years, senior high school – 3 years and university bachelor’s degree – 4 years). The school was selected using a multi-stage sampling technique, involving a random selection of Eastern region out of the 10 regions in Ghana. This was followed by a random selection of a district in the region and one senior high school in the district. Information relating to the various schools in the selected district was obtained from the Ghana Education Service. English is the official language of instruction at the various levels of education (Adjorlolo, 2016). More than half of the participants were girls (n = 322, 58.1%) with an average age of 16.85 (standard deviation [s.d.] = 1.32).

A self-report methodology using a cross-sectional survey design was employed to gather data from senior high school students recruited from a school in the Eastern region of Ghana. A questionnaire consisting of a demographics section, the GAD-7, PHQ, WHO-5 and SBQ-R was used.

Generalised Anxiety Disorder-7 items are rated on a 4-point Likert scale, ranging from 0 (not at all) to 3 (nearly always). The seven items are summed to generate a total score that ranges from 0 to 20 in the present study. Higher scores indicate more severe symptoms of GAD.

Patient Health Questionnaire-9 (PHQ; Kroenke, Spitzer, & Williams, 2001) was administered to measure depressive symptoms in the participants. The PHQ-9 items are rated on a four-point scale ranging from ‘not at all’ (0) to ‘nearly every day’ (3). Higher scores, obtained by summing the participants’ responses, indicate more depressive symptoms. The Cronbach’s alpha recorded in the present sample was 0.71.

World Health Organization Well-Being Index (WHO-5; Topp, Østergaard, Søndergaard, & Bech, 2015) is a five-item scale used to index subjective, positive well-being. The items are scored from 5 (all of the time) to 0 (none of the time), with a total score ranging from 0 (absence of well-being) to 25 (maximal well-being). A Cronbach’s alpha of 0.70 was observed for the present sample.

Suicidal Behaviour Questionnaire-Revised (SBQ-R; Osman et al., 2001) is a four-item scale administered to screen for suicidal tendencies. The items are rated on Likert scales as follows: item 1 = 1–4, item 2 = 1–5, item 3 = 1–3 and item 4 = 0–6. The SBQ-R total score ranges from 3 to 18, with higher scores reflecting greater risk for suicidal tendencies. Prior studies have found positive and significant correlations between suicidal tendencies and anxiety or GAD (Balázs et al., 2013; Nepon, Belik, Bolton, & Sareen, 2010). The Cronbach’s alpha reported in the present study was 0.78.

Data were collected from the students in their respective classes. In each class, the research team briefed the participants on the purpose of the study and their responsibilities as participants. They were encouraged to ask questions to allay any fear and anxiety pertaining to participating in the study. Ethical issues, particularly those relating to confidentiality, anonymity and withdrawal from the study without being penalised, were communicated to the participants and ensured. To maintain anonymity, for instance, the questionnaires were devoid of identifying information of the participants such as name and student numbers. The participants were also informed that their responses will be treated strictly confidentially, and that they can withdraw from the study anytime without being penalised, or the research team can terminate their participation without their explicit consent. Those expressing interest and willing to participate in the study were handed a pack of questionnaires described previously. The questionnaires were handed over to the research team in each class upon competition. Data were collected from the second- and third-year students because the first-year students were yet to commence school at the time of data collection.

Missing data analysis revealed that less than 5% of cases have data points missing on the SBQ-R (2.4% – 3.1%), PHQ (2% – 3.3%), GAD-7 (1.3% – 2.7%) and WHO-5 (1.3% – 2.4%). Further analyses revealed that data were missing completely at random (Little’s chi-square > 0.05). Consequently, the missing data point was imputed using the expectation-maximisation algorithm (Adjorlolo & Watt, 2017). Confirmatory factor analysis, using the maximum likelihood method, was used to test for the factorial validity of the GAD-7. A multi-group CFA was conducted to determine the invariance of the associations between the observed items and the latent factor. This was performed sequentially in the following ways (see Adjorlolo & Watt, 2017; Byrne, 2010; Steinmetz et al., 2009). Firstly, separate models were estimated for boys and girls, called the baseline model. Secondly, an unconstrained model was established whereby the fixed and free parameters were estimated simultaneously for the groups (i.e. configural invariance). Thirdly, the factor loadings were constrained across the groups to establish metric invariance. Next, the factor variance was constrained, in addition to the factor loadings to investigate invariance of factor variance. Lastly, the factor loadings, factor variance and error variances were constrained to be equal across the groups to determine the invariance of error variances. Constraining error variances is important for testing the equality of reliability of the GAD-7 across the groups (Byrne, 1988).

Evidence of gender invariance was evaluated using the difference in comparative fit index (CFI; ΔCFI) and chi-square (c²; Δc²). A non-significant Δc² and ΔCFI ≥ −0.01 between the restrictive and less restrictive or unconstrained models indicate the attainment of gender invariance. In case of discordance between the Δc² and ΔCFI, the estimates provided by the latter were deemed reliable. This is because, unlike the Δc², the ΔCFI is independent of the sample size and model complexity, and is also uncorrelated with the overall fit measures (Cheung & Rensvold, 2002). Model fit was determined using c² and the following commonly used fit indicators: CFI, Tucker–Lewis index (TLI), adjusted goodness of fit index (AGFI) and a non-centrality-based index, the root mean square error of approximation (RMSEA). Comparative fit index and TLI values close to 0.95, or greater, and RMSEA value close to 0.06 or below indicate good model fit (Hu & Bentler, 1999). The CFA and multi-group CFA were performed using Analysis of Moment Structures (AMOS) version 21.

Zero-order correlations with measures of anxiety, mental well-being and suicidal tendencies were performed to examine construct validity of GAD-7, whereas internal consistency was investigated using Cronbach’s alpha (a) and mean inter-item correlation (MIC; Clark & Watson, 1995). The discriminant validity of the GAD-7 was determined using the receiver operating curve (ROC), with the area under the curve (AUC) of the ROC indicating the overall discriminant accuracy. The AUC value ranges between 0.5 (no discriminative power) and 1.0 (maximum discriminative power). The zero-order correlations, internal consistency and ROC analyses were conducted using IBM Corp. SPSS version 23.

The Ethics Committee for Humanities of the University of Ghana granted ethical approval for the study (Ethical clearance number: ECH 165/17-18). The school management granted institutional permission for the study.

Participants endorsing the highest response options on the items of the various measures were designated as high risk. Using this criterion, 4.2% (n = 23), 17% (n = 95) and 18% (n = 102) of the participants were designated as at high risk for GAD, suicidal tendencies and depression, respectively.

The results of the factorial validity and gender invariance analyses of the GAD-7 are summarised in Table 1. The GAD-7 evidenced a unidimensional factor structure, providing a good fit to the data in the full sample (AGFI = 0.98; TLI = 0.97; CFI = 0.98; RMSEA = 0.03), boys subsample (AGFI = 0.96; TLI = 0.97; CFI = 0.98; RMSEA = 0.04) and girls subsample (AGFI = 0.96; TLI = 0.96; CFI = 0.98; RMSEA = 0.03). Measurement invariance analyses indicate that configural invariance (GFI = 0.97; TLI = 0.96; CFI = 0.97; RMSEA = 0.03), metric invariance (ΔCFI = −0.002), invariance of factor variance (ΔCFI = 0.000) and invariance of error variances (ΔCFI = 0.001) have been attained. As shown in Table 2, the GAD-7 items loaded satisfactorily and significantly onto the unidimensional factor structure. A two-factor model was also tested in accordance with previous studies (Beard & Björgvinsson, 2014; Ito et al., 2018). It was observed that the model for the full sample did not provide a good model fit to the data (i.e. AGFI = 0.88; TLI = 0.74; CFI = 0.67; RMSEA = 0.09). Further analyses reveal that the girls (AGFI = 0.75; TLI = 0.92; CFI = 0.64; RMSEA = 0.11) and boys (AGFI = 0.89; TLI = 0.94; CFI = 0.76; RMSEA = 0.09) subsamples did not yield good model fits. Inspection of the modification indices and coefficient suggests that no alteration of the models will cause improvement to the model. Given that these initial model assessments are critical to measurement invariance analysis, a decision was reached not to investigate the models further by constraining the model parameters. It was therefore concluded a unidimensional model fit the data for Ghanaian adolescents’ sample.

The GAD-7 correlated significantly and positively with measures of depression (r = 0.67, p < 0.001; r = 0.68, p < 0.001; r = 0.67, p < 0.001), suicidal behaviour (r = 0.40, p < 0.001; r = 0.44, p < 0.001; r = 0.37, p < 0.001) and negatively with mental well-being (r = −0.35, p < 0.001; r = −0.32, p < 0.001; r = −0.37, p < 0.001) in the full sample, boys and girls subsamples, respectively.

Likewise, the internal consistency of the GAD-7 for the full sample, based on the Cronbach’s alpha (a), was 0.69, whereas the MIC was 0.24 for the full sample. Similar result was obtained for boys (a = 0.72; MIC= 0.26) and girls (a = 0.66; MIC = 0.22) subsamples.

The GAD-7 significantly discriminated between participants designated as high and low or no risk for depression (AUC = 0.70; z = 6.77, p < 0.001; CI [0.66, 0.74]), as well as high and low or no risk of suicidal tendencies (AUC = 0.71; z = 7.70, p < 0.001; CI = [0.67, 0.75]).

Firstly, the study did not include a gold standard measure of GAD, thereby making it impossible to assess the diagnostic properties (e.g. sensitivity and specificity) of the GAD-7 for GAD, in addition to depression and suicidal tendencies. The use of self-report measures provides no mechanism to verify the accuracy of the participants’ responses. Thus, inaccurate responses (underreporting or over-reporting) are possibilities. The generalisability of the study findings to adolescents with dissimilar background characteristics (e.g. those out of school) may be limited. Again, using a clinical sample would have provided an added layer of validity, and therefore it would be useful in future studies to examine adolescents diagnosed with GAD. Future studies addressing the limitations noted here will help to further illuminate the psychometric appropriateness and utility of the GAD-7 in adolescents in Ghana and other LMICs.

The results suggest that the GAD-7 is a reliable and valid measure to screen and identify adolescents at risk of GAD in Ghana, an SSA state. The findings have contributed to the repertoire of the cross-cultural literature on the assessment of psychopathologies or mental disorders. While the influence of cultural factors on psychopathological behaviours should not be discounted, it is equally important to acknowledge that measures developed to assess behaviours that supposedly have roots in Western, individualistic cultures could prove useful in non-Western, collectivists’ cultures.