Validation of the Malay Version of the Affiliate Stigma Scale among Caregivers of Patients with Mental Illness

Background Caregivers of patients with mental illness are exposed to stigma. The internalisation of this stigma among caregivers is known as affiliate stigma and can be measured by the Affiliate Stigma Scale (ASS). The aim of this study was to validate the Malay version of the ASS. Methods A cross-sectional study was performed from May to December 2017 with 372 caregivers of patients with mental illness. The ASS was first translated into Malay using standard forward and backward translation procedures. The final version of the ASS-Malay (ASS-M) was completed by participants. The data analyses involved assessment of construct validity by exploratory factor analysis, confirmatory factor analysis and construct reliability. Results The final model of the ASS-M consists of four factors with 21 items, as compared to the original version, which has three factors with 22 items. The results showed that the final model has good model fit based on RMSEA (0.065) and SRMR (0.055) and a satisfactory composite reliability (affective = 0.827, cognitive = 0.857, behaviour = 0.764, self-esteem = 0.861). Conclusion The study showed that the four-factor, 21-item ASS-M model has good psychometric properties. The scale is valid and reliable for measuring affiliate stigma among caregivers of patients with mental illness in Malaysia.


Introduction
Stigma is defined as a set of prejudicial attitudes, negative stereotypes, discrimination and biased social structures towards a certain group of people (1). The process of stigma starts with labelling and stereotyping, which in turn lead to separation, status loss and discrimination (2).
There are many ways of looking at stigma. Public stigma focuses on a community's discrediting response to the stigmatised person, while holding a negative attitude or prejudice towards oneself is known as self-stigma (3)(4). Courtesy stigma is the stigma experienced by family members or caregivers of a stigmatised person (5).
Affiliate stigma is the internalisation of negative stigma-related experiences by the family members of the stigmatised person (6). Affiliate stigma indirectly covers aspects of the caregiver's self-stigma and their subsequent psychological responses of the associates. The result of this internalisation process affects the person's agree) (6), and includes 22 items assessing three domains (or subscales) of affiliate stigma: affective, cognitive and behavioural. The affective subscale consists of seven items (item 1, 4, 7, 10, 13, 16 and 19); the cognitive subscale also includes seven items (item 3, 6, 9, 12, 15, 18 and 21); and the behavioural subscale contains eight items (item 2, 5, 8, 11, 14, 17, 20 and 21). A higher mean score of the 22 items indicates a higher level of affiliate stigma. The ASS has good internal consistency [α = 0.94] for caregivers of mental illness and for exploratory factor analysis (6).

Instrument translation
The original English language version of the ASS was translated into the Malay language using forward and backward translation by bilingual experts of Malay and English (see Figure 1). Two psychiatrists, who were competent bilingual speakers, reviewed both backward and forward translations, comparing each item in Malay to the corresponding item in the original English version. Expert panels assessed the contents of the questionnaire to be culturally appropriate to the Malaysian population. The final version in the Malay language, ASS-Malay (ASS-M), was pre-tested among 10 caregivers of patients with mental illness for clarity and comprehension. The participants were asked to answer the questions and to comment on the wording and presentation of the questionnaire. The results of the pre-test were found to be good, and therefore no modifications were necessary.

Statistical analysis
The Statistical Package for Social Sciences version 22.0 was used to analyse the data which included descriptive statistics of the respondents' sociodemographic characteristics, EFA and internal consistency reliability. The acceptable cut-off value for the internal consistency reliability based on Cronbach's alpha coefficient was ≥ 0.70 (12). A factor loading less than 0.3 was considered for removal of an item, and a factor with an Eigenvalue > 1.0 was accepted (12). The final model found in EFA was then confirmed by using CFA via Mplus 8 software (13).
The fitness of the model was assessed by the following indices: root mean square error of approximation (RMSEA) with an acceptable level of < 0.08; standardised root mean square residual (SRMR) with an acceptable level of < 0.08; Tucker-Lewis fit index (TLI) with an cognition, affect, behaviour, self-esteem and self-efficacy.
The Affiliate Stigma Scale (ASS) consists of 22 items that measure the cognitive, affective and behavioural components of affiliate stigma (6). This scale was developed to study affiliate stigma among caregivers of people with intellectual disabilities or mental illness. The scale has been shown to have good psychometric properties, and its use has increased over the last few years. Furthermore, the ASS has already been validated and translated into different languages, including Chinese (6), Urdu (7), Hebrew (8), Hindi (9), Persian (10) and Amharic (11).
While stigma is commonly experienced by caregivers of patients with mental illness around the world, including Malaysia, a validated measurement scale is needed to assess affiliate stigma among caregivers in Malaysia. Thus, the aim of this study was to validate the Malay version of the ASS among caregivers of patients with mental illness in Kelantan, Malaysia.

Study Design and Procedures
This cross-sectional study was conducted in the psychiatric clinic at Universiti Sains Malaysia (USM) from May to December 2017. A total of 372 caregivers aged 18 and above consented to participate in the study. The estimated sample size for exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) was calculated according to a standard size per domain set by Heir and colleagues (12). An estimated 20% non-response rate was also included for both EFA and CFA sample size determination. At the time of the study, the caregivers had been taking care of patients with mental illness (schizophrenia, mood disorder, anxiety disorder and intellectual disability) for at least 6 months. Participants who had major psychiatric illness were excluded from the study. The participants were recruited using non-probability convenience sampling. The study protocol was approved by the Human Research Ethics Committee of USM [USM/ JEPeM/16120605].

Exploratory Factor Analysis (EFA)
Principal axis factoring analysis with Promax rotation was conducted and resulted in a total of four factors with eigenvalues greater than one. The value of the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy test was excellent, at 0.92. Bartlett's test of sphericity was significant, with x² (231) = 2170.164, P < 0.01. All items loaded in a single dimension, with the value of factor loading higher than 0.30 (Table 2). Item 2 was deleted, as its factor loading was lower than 0.3.
All items were arranged based on the factor loading under the four factors extracted in this study (Table 3). Items with cross loading results were rearranged under the related factor after discussion with experts from the research team. Item 1, 'I feel inferior ……', had a factor loading that was slightly lower for the affective factor [0.328] than for the behaviour factor [0.418]. However, the research team decided to put item 1 under the affective factor, as this item was related to emotion more than to behaviour. Similar to item 1, both items 9 and 21 were placed under the cognitive factor despite their factor loadings [0.452; 0.487] being lower than those for self-esteem [0.456] and behaviour [0.564].

Confirmatory Factor Analysis (CFA)
The four-factor model extracted from EFA was tested, and each item was allowed to load on its corresponding factor. The results of CFA are shown in Tables 4 and 5.
The initial model for the ASS-M had a good fit to the data based on fit indices of RMSEA and SRMR except for CFI and TLI (see Table 5). Further modification to the model was done to improve the fit indices.
The final ASS-M model displayed the following fit indices: RMSEA = 0.065, SRMR = 0.055, CFI = 0.904, TLI = 0.888. These results showed that two out of four fit indices (i.e., acceptable level of > 0.95; and finally, the comparative fit index (CFI) with an acceptable level of > 0.95 (12).
The construct reliability (CR) of the ASS-M was estimated by Raykov's rho. A reliability based on Raykov's rho of ≥ 0.70 was considered both reliable and acceptable (12). The acceptable cut-off value for the Cronbach's alpha coefficient was also similar: ≥ 0.70 (12).

Socio-demographic Characteristics of the Respondents
The participants (n = 132 for EFA; n = 240 for CFA) were mostly married (n = 86, 65.2%; n = 169, 70.4%) and female (n = 90, 68.2%;  No further modification was done because all the factor loadings (see Table 4) were above the recommended value and the items were found to be important, to remain in the constructs.
communicating with a family member having mental illness/intellectual disability'. Q2 was considered inappropriate for the Malaysian population because relevant data have shown that most Malaysians prefer to talk than to keep silent when faced with a problem (16). Thus, it is likely that almost all caregivers would have responded to this item with 'strongly disagree', i.e., a score of 1.
The four factors differed from the original ASS with respect to item clustering. This could be due to the cultural adaptation of the original validated scale (17). Thus, the four factors are affective (item 1,4,13,19), cognitive (item 3,9,11,15,16,18,21,22), behaviour (item 5,14,17,20) and self-esteem (item 6, 7, 8, 10, 12). The research team named the fourth factor 'selfesteem' based on the meanings of all items under it. This factor also correlates with many studies which have shown that self-stigma plays an important role in a stigmatised person's selfesteem (18)(19).
This demonstrates adequate evidence for the reliability of the Malay version of the ASS. The CR for the ASS-M was more than 0.7, as required (12).

Discussion
In this study, exploratory and confirmatory factor analyses for the factor structure of the ASS-M were conducted. EFA was performed to extract the new factor structure from the dataset and compare it with the three-factor model, while CFA was conducted to assess the fitness of the new model.
The prominent sociodemographic characteristics of the participants in this study, i.e., female, middle-aged, married and parental caregivers, was a finding similar to that generated in studies on related subjects (14)(15).
In comparison to the original ASS model, which has a three-factor construct, EFA in the present study produced a four-factor construct with the removal of one item: Q2 'I avoid  recommended that stringent criteria should be applied with caution (21). All items achieved a satisfactory factor loading to their respective factors. Most items achieved a loading of more than 0.5-except for item 3, whose value [0.49] was slightly lower than 0.5. Based on these indices, this study had an acceptable fit to the four-factor model.
The four-factor ASS-M model proposed in this study has good psychometric properties and will be useful for researchers and health care providers in Malaysia to study the association and impact of affiliate stigma on caregivers of patients with mental illness, whose main spoken language is Malay. The ASS-M model will enable more interventions in the future to reduce the harmful effects of affiliate stigma.
The standardised factor loadings yielded from EFA ranged between 0.32 and 0.86. These were higher than those in the original model's factor loading, which ranged between 0.42 and 0.79 (6). The internal consistency of the ASS-M with a Cronbach's alpha of 0.92 was comparable to the Cronbach's alpha of 0.94 for the ASS (6). This was good, as its value was more than 0.7, as required (12).
Several fit indices were used to evaluate the goodness of fit of the model. For the final ASS-M model, the values of both RMSEA and SRMR were lower than 0.08 (12), indicating a good fit. The CFI and TLI indices did not reach the 0.95 cut-offs suggested by Hair et al. (12). However, Maiyaki (20)   We acknowledge that there were some limitations in the present study. The majority of participants in this study was Malays and using only one study site may hinder the generalisation of the results to other ethnicities in Malaysia. The data collection process, which relied on self-reported questionnaires, may have led to response bias. Additionally, no comparisons were made with other tools that also assess affiliate

Conclusion
This study showed that the four-factor, 21-item ASS-M model has good psychometric properties. The scale is valid and reliable for measuring affiliate stigma among caregivers of patients with mental illness in Malaysia.