52 3 CHAPTER 3 score was used. Previous experience Participants were asked whether they had had previous experiences and/or had witnessed others’ experiences with treatment seeking (in general). If they responded with ‘yes’, they were asked whether this experience was positive or negative. A dummy variable was used to compare the reference (no experience) to positive and negative experiences. Unit cohesion A three-item measure was used to measure perceived unit cohesion (22). The items were ‘The members of my unit are… cooperative with each other/know they can depend on each other/stand up for each other’. Items were measured on a 5-point scale ranging from ‘Completely disagree’ to ‘Completely agree’. The mean value was used as the final measure. Participants with mental illness were asked about unit cohesion at the time they had experienced their illness. Statistical analyses To answer research questions 1 and 2 on attitudes, beliefs, and needs of military personnel regarding not seeking treatment, descriptive analyses were performed. Comparisons were made between the decision (intention) to seek treatment or not, using Chi-square tests and Mann-Whitney U-tests, as variables, were not normally distributed. For comparisons between those who intended to seek treatment, and those who did not, ‘very-unlikely’ and ‘unlikely’ were combined, just as ‘likely’ and ‘very-likely’ were combined. For research question 3 on factors associated with behaviour and intentions of not seeking treatment, two separate analyses were performed. For personnel with mental illness, a firth logistic regression was performed, as it corrects for quasi-separation in the data and the small number of people who did not seek treatment (23). Treatment-seeking decision was entered as the dependent variable (0 = treatment seeking, 1 = not seeking treatment). Ordinal regression was performed for personnel without mental illness. The assumption of proportional odds was violated. Therefore, for analysis, the categories ‘very unlikely’ and ‘unlikely’ were merged, resulting in the dependent variable, ‘not seeking treatment intention’, with categories 1 = Very likely; 2 = Likely; 3 = (Very) unlikely, meeting the assumption. To prevent information loss, the categories ‘likely’ and ‘very-likely’ were not combined. To decrease the number of predictors, fear of negative career consequences, social rejection, discrimination, self-stigma, shame, and fear of being blamed were combined into one measure of stigma, as they are all aspects of stigma (24). Together, these items formed a reliable scale (apersonnel with mental illness = .882, apersonnel without mental illness =
RkJQdWJsaXNoZXIy MjY0ODMw