Secondary data analysis using data from an interview study [18] nested within phase three of the King’s Centre for Military Health Research Health and Wellbeing Cohort Study [19] were conducted. In short, participants were sampled into the interview study from phase three of the King’s Centre for Military Health Research Health and Wellbeing Cohort Study based on self-reporting a mental health or stress problem within the last three years and subsequently were invited to complete a structured telephone clinical interview covering measures of mental health symptomatology (e.g., depression, anxiety, post-traumatic stress disorder [PTSD], and alcohol misuse), stigma, barriers to care, and help-seeking behaviour associated with their mental health. Participants received £25 for their participation. Data collection took place between February 2015 and December 2016. Further details on data collection procedures have been described elsewhere [18]. The study received ethical approval from the UK Ministry of Defence Research Ethics Committee (ref: 535/MODREC/14). The final sample included 1448 serving and ex-serving members of the UK Armed Forces. Of these, 1229 (84.9%) were men and 219 (15.1%) were women.

An a-priori power analysis using G*power [20] determined that a total sample of 496 participants would be required given the observed difference in group size (85/15) to detect gender differences in stigma scores with a small effect size (d = 0.30) with 80% power using a two-sample t-test, suggesting the current study was sufficiently powered to detect potential gender differences in barriers to mental healthcare.

Participants completed questions from the Perceived Stigma and Barriers to Care for Psychological Problems– Stigma Subscale (PSBCP-SS; [21]), the Barriers to Access Care Evaluation measure (BACE; [22]), and the Self-Stigma of Seeking Help Scale (SSSHS; [23]). Participants responded to statements relating to (i) access to mental health services (4 items from the PSBCP-SS), (ii) self-stigma of mental illness (7 items from the PSBCP-SS and BACE), (iii) self-stigma of mental health treatment (5 items from the SSSHS), and (iv) public stigma of mental health care/providers (9 items from the PSBCP-SS and BACE). Each item was rated using a 5-point response scale that ranged from (1 = strongly agree to 5 = strongly disagree). All items were reverse-scored. Scores on each subscale were averaged with higher scores reflecting greater stigma/lower access. Similar scoring approaches have been used in previous research [24]. Additionally, we used an item-response-theory (IRT) approach to generate total scores using a response style model to capture latent variables, as IRT scoring approaches have been found to perform better than mean scores when using Likert-style responses [25].

To adjust for co-occurring mental health problems, we included information on the participant’s current mental health. This included questions on depression (Patient Health Questionnaire [PHQ-9; [26]]), generalized anxiety disorder (GAD-7 scale; [27]), post-traumatic stress disorder (20-item PTSD Checklist [PCL-5; [28]]), and alcohol misuse (3-item Alcohol Use Disorders Identification Test [AUDIT-C; [29]]). In line with previous work [18], we used the following cut-offs to indicate the presence of a probable disorder: a cut-off score of 15 on the PHQ-9 was used to identify probable depression (scores range from 0 to 27), a cut-off score of 10 or greater on the GAD-7 was used to identify probable generalised anxiety disorder (scores range from 0 to 21), a score of 38 or greater on the PCL-5 was used to identify probable PTSD (scores ranged from 0 to 78), and an AUDIT-C score of 10 or more was used to indicate alcohol misuse (scores ranged from 0 to 12).

Participants completed a range of demographic variables as part of the King’s Centre for Military Health Research Health and Wellbeing Cohort Study [19], from which participants were screened into the clinical interview study. Gender was assessed using a binary approach (e.g., man/woman). Additionally, participants were asked about their current serving status (i.e., serving or ex-serving) during the interview study. For the purpose of this study, we included data on gender, age at interview, serving status at interview, engagement at screening, service branch at screening, and rank at screening.

All analyses were conducted in Stata version 17. The analysis plan was pre-registered (https://osf.io/zntdu/?view_only=3efd0cef07664fc693d96cf95efa4bf3) and all deviations from this planned analysis are explicitly outlined below.

First, men and women were compared on demographic variables to identify potential confounding variables Although it was initially planned to examine relationship status, number of children, and length of service as potential confounding variables, these variables were only available in the original cohort study, and so it was decided not to include these, as they could have changed. Additionally, in line with previous studies, we decided to include rank as a potential confounding variable [18]. Differences were examined using Pearson’s chi-squared test to account for sampling weights. Secondly, men and women were compared on the variables of interest (stigma and access to care subscales, self-stigma) using univariate analyses. Lastly, using regression analysis, men and women were compared on the variables of interest adjusting for demographic variables that were identified as differing significantly between men and women in the first step of the analyses, as well as current mental health disorders based on screening measures. No issues with multicollinearity were detected.

All analyses were adjusted for sampling weights using the svy command. We had originally planned to deal with missing data using the full information maximum likelihood framework through the sem-suite in Stata. However, as no data was missing on the variables of interest (i.e., stigma variables, access to care, and gender), we employed a regression framework instead.