The Stigma of Suicide Scale-Short Form (SOSS-SF)

The original 16-item Stigma of Suicide Scale-Short Form (SOSS-SF) was developed by Batterham et al to determine the stigma towards people with a history of suicide attempts. ^{Reference Batterham, Calear and Christensen14} It consists of three subscales, each item of which includes a positive or negative phrase/statement (one or two words) about suicide and people who take their own lives. ^{Reference Batterham, Calear and Christensen5} The subscales are: (a) Stigma (eight items, e.g. shallow, stupid, irresponsible); (b) Isolation or depression (four items, e.g. lonely, isolated, disconnected); and (c) Glorification/Normalisation (four items, e.g. strong, noble). Participants are required to rate how much they agree with each item using a 5-point Likert scale (1 = strongly disagree, 2 = disagree, 3 = neutral, 4 = agree, 5 = strongly agree). Higher scores denote higher levels of stigma towards people who kill themselves. The SOSS-SF has been translated into several languages, such as Chinese, ^{Reference Han, Batterham, Calear, Wu, Shou and van Spijker18} Arabic, ^{Reference Aldalaykeh, Dalky, Shahrour and Rababa6} Bangla (Bengali), ^{Reference Arafat, Hussain, Hossain, Islam and Menon19} Hong Kong Chinese ^{Reference Fong, Yip, Chan and Ho20} and Nepali. ^{Reference Gupta, Sharma, Sah, Sharma, Jha and Chapagai21} The SOSS-SF is deemed a short, focused and easily administrable tool for large-scale research. ^{Reference Wu, Chen, Batterham and Han22}

The subscales of the original SOSS-SF have been demonstrated to have high internal consistency, with overall Cronbach’s α values of 0.89 (Stigma subscale), 0.88 and 0.78 (Isolation or depression subscale), 0.78 and 0.82 (Glorification/Normalisation subscale), and 0.70 for the overall scale in a sample size of 1286 Australians aged 18 years and above. ^{Reference Batterham, Calear and Christensen14}

The Attitudes Towards Suicide (ATTS) questionnaire

We used the adapted Attitudes Towards Suicide (ATTS) questionnaire to test the discriminant validity of the translated SOSS-SF. The ATTS was developed by Renberg & Jacobsson, ^{Reference Renberg and Jacobsson23} who aimed to introduce an instrument to measure attitudes towards suicide from a broader perspective. It consists of 37 items rated on a 5-point Likert scale (1 = totally disagree, 5 = totally agree). The questionnaire has been translated into Malay and validated among healthcare personnel in Malaysia. ^{Reference Siau, Wee, Ibrahim, Visvalingam and Wahab13} The ATTS comprises ten factors, such as Suicide as a right, Non-communication, Preventability and Tabooing. Higher scores indicate agreement with the factor or domain. The Malay ATTS demonstrated acceptable reliability (Cronbach’s α = 0.72) and adequate validity within Malaysia. ^{Reference Siau, Wee, Ibrahim, Visvalingam and Wahab13} In our study, the researchers focused on four distinct domains: Factor 1, ‘Ability to understand and accept suicide’, with six items; Factor 3, ‘Believability of suicidal threats’, consisting of five items; and Factor 10, ‘Normality of suicide’, comprised of two items.

Translation of the SOSS-FF into Malay

For the translation and cross-cultural adaptation process, the SOSS-SF was forward-translated into Malay, following the recommended guideline outlined by Cruchinho et al. ^{Reference Cruchinho, López-Franco, Capelas, Almeida, Bennett and da Silva24} A Malay linguistic expert was recruited who is also bilingual in both the source language and Malay to ensure that the SOSS-SF’s meaning was retained as much as possible. The Malay version was then back-translated into English by a counsellor registered with the Board of Counsellors (Malaysia), who is a native speaker of Malay. For some items, direct translation was not achievable since some phrases required adjustment to the local context.

We convened a harmonisation meeting with a panel of clinical psychologists, psychologists, counsellors and linguistic experts, with the aim of comparing the forward- and back-translations. In the meeting, the panel reviewed and discussed potentially unclear phrases and agreed on the first Malay version of SOSS-SF. For example, ‘isolated’ and ‘disconnected’ presented inconsistencies during translation. The direct translations were discarded to align with local understanding of ‘isolation’ and ‘feeling disconnected from society’, and the disparities were reconciled: a five-word descriptor was utilised for ‘disconnected’ and a two-word descriptor was agreed on for ‘isolated’.

A pilot test of this first (‘harmonised’) version was conducted in a public university, involving 30 undergraduate students recruited through convenience sampling. Following this, the translation was amended and proceeded to validation in a larger sample (Supplementary Fig. S1, available at https://doi.org/10.1192/bji.2026.10132), as described in this paper.

Factor structure

Previous studies have demonstrated 3 to 4 structures of the SOSS-SF in different contexts: for example, in Hong Kong, a 4-factor structure with 12 items was recommended rather than the original 3-factor structure with 16 items. ^{Reference Fong, Yip, Chan and Ho20} Therefore, exploratory factor analysis (EFA) was again employed as an exploratory approach in our study. This method was used to determine the number of latent factors and to evaluate which observed variables serve as meaningful indicators of these latent dimensions, based on the magnitude and variability of the factor loadings. ^{Reference Brown25} We used EFA used principal component analysis and oblimin rotation to determine the number of factors and similarities between three-factor domains. We chose oblimin rotation from the oblique rotation methods because the factors in the SOSS-SF were estimated to correlate with one another. ^{Reference Batterham, Han, Calear, Anderson and Christensen4} Factor loading of an item had to be above a threshold level of 0.35 on the factors on which it loaded the strongest, and the item factor loading on others should not exceed 0.30. ^{Reference Hair, Anderson, Tatham and Black15}

Discriminant validity

The scarcity of validated psychometric instruments in Malaysia for directly assessing stigma towards suicide underscores the importance of synthesising the discriminant validity of the SOSS-SF and comparing it with other instruments. Discriminant validity helps to determine whether the Malay versions of the SOSS-SF and ATTS measure distinct constructs. Discriminant validity refers to two instruments demonstrating validity by showing that their scores have low correlations with factors that, in theory, they should not be strongly associated with. ^{Reference Reber, Reber and Allen26} This weak correlation indicates that the measures are indeed capturing different constructs. ^{Reference Rönkkö and Cho27} As part of the evaluation of discriminant validity, the total scores on the SOSS-SF and subscores on the ATTS were assessed by Pearson correlations. Furthermore, each latent variable was assessed for internal consistency reliability using Cronbach’s α coefficients. Cronbach’s α values between 0.70 and 0.95 were considered to indicate adequate reliability of the questionnaire. ^{Reference Tavakol and Dennick28}

Invariance

We used several models to assess the invariance of the Malay version of the SOSS-SF across age and gender groups. ^{Reference Tan and Pektaş29} First, we established the configural invariance model, allowing for differences in all parameters, including factor loadings, intercepts and error terms for age and gender groups. The focus was on determining whether the latent variable structure replicated consistently across groups. Second, the metric invariance model aimed to assess whether the factor loadings were the same across groups by allowing the differences in intercepts and error terms. This model also assessed whether the strength of relationships between observed variables and latent variables was equivalent across groups. Its model fit was compared with the configural invariance model fit to determine whether a metric invariance model was achieved. Third, the scalar invariance model assessed whether factor loadings and intercepts were at the same value across groups by allowing for differences in intercepts and error terms. This step makes meaningful comparisons of the means for each latent variable across groups. Its model fit was compared with the metric invariance model fit to determine whether a scalar invariance model was achieved. The measurement invariance of the instrument was subsequently confirmed when the model fit indices were satisfactory, and nested model comparisons met the criteria of difference in the comparative fit index (ΔCFI ≤ 0.010), in the Tucker–Lewis index (ΔTLI ≤ 0.010) and in the root mean square error of approximation (ΔRMSEA ≤ 0.010), indicating that the invariance model was supported. ^{Reference Andre, Maulana, Helms-Lorenz, Telli, Chun and Fernández-García30,Reference Yue, Yue, Liu, Li, Dong and Bao31} The difference in the chi-squared statistic (Δχ²), the difference in the degree of freedom (Δd.f.) and the p-value were not used as criteria owing to sensitivity to sample size and the limited ability to differentiate between invariant and non-invariant models. ^{Reference Andre, Maulana, Helms-Lorenz, Telli, Chun and Fernández-García30,Reference Yue, Yue, Liu, Li, Dong and Bao31}