Previous studies suggest that public stigma toward groups with implicit marks, such as people with mental illness (PMI), is harmful [1]. For PMI, perceived public stigma may exclude them and devalue their social status and networks [2–4]. This loss may cause their normal lives to further deteriorate and make them believe that they deserve the stigma they experience. The general public may feel unsafe with community psychiatric institutions near their residence [5] and support keeping a larger social distance from PMI [6]. Even mental health staff and medical students may express relatively high levels of stigma toward PMI [7,8]. When it comes to severe mental illnesses, such as schizophrenia and bipolar disease, these negative attitudes become even worse [9]. This situation occurs in the Chinese context as well. Owing to culturally embedded notions of mental illness, PMI and their families often opt to conceal the condition, apprehensive that disclosure could precipitate social exclusion and impute diminished social and moral competence. Although individuals with frequent and sustained contact with PMI often display a more accurate understanding of the disorder, such improved knowledge is not necessarily accompanied by more favorable attitudes toward this population [10]. Consequently, these deeply entrenched stigmatizing attitudes toward PMI are likely to persist over extended periods [11].

In order to tackle the negative consequences of public stigma toward PMI in the Chinese context, developing validated measurements to examine the degree of public stigma is the first step. One locally developed scale, the Stigma Scale towards PMI [12], is the most widely used in the Chinese context. The 26-item scale is rated by five Likert points and includes three perspectives: social distance, danger, and competence. Although the whole scale has high internal consistency, with a Cronbach’s alpha of 0.89, the sample in Zeng and colleagues’ study mainly comprised medical professionals and students, and family members of PMI. Its application to the general public has not been validated. Scales translated from the English version and adapted in a Chinese context have also been used to measure public stigma toward PMI. An official scale, Attitudes towards Mental Illness [13], measuring public stigma toward PMI, is adapted from Link and colleagues’ [14] Devalued and Discrimination Scale (DDS). The adapted version includes 12 items and has been used in several surveys in China. However, reliability and validity information were not reported publicly [15]. The Community Attitudes towards The Mentally Ill (CAMI) scale has also been validated in China [16]. The adaptive version kept three factors from the original scale, with 16 items retained. However, it was only tested by 100 medical professionals, and its reliability was not reported.

After briefly reviewing the instruments measuring public stigma toward PMI, we found that instruments applicable to Chinese stigma research are quite limited. Instead of having a concrete focus on components of public stigma, current Chinese scales only provide preliminary descriptions regarding mental illness and PMI. Especially under the promotion of the de-institutionalization of PMI, residents’ attitudes toward the reintegration of PMI into the community and society as a whole are still not accurately measured and discussed in the existing literature. In other words, whether or not the general public in Chinese society can accept rehabilitation and recovery services for PMI “in their backyards” has not been examined.

As a result, to measure community attitudes toward PMI, the CAMI scale became the first choice of this research. The scale was developed from the perspective of the general public, which satisfied our requirements in regard to measuring public stigma. Moreover, the inclusion of attitude toward community mental health services reflects the dilemma in the Chinese context in a timely manner. Participants might be able to comply with social desirability when answering emotion-related questions, but their attitudes toward community services could help unveil the participants’ underlying ideas.

However, the scale cannot be used directly in Chinese research for the following reasons. First, the 40-item scale is relatively long, as repeated measurements are required in academic research. Second, the positively and negatively worded items in the original scale might actually measure the same thing under one factor, which might have the potential to guide participants’ attitudes. Moreover, it might not be feasible to say that the denial of negatively worded items indicates the opposite meaning of the acceptance of positively worded items. For example, one of the negatively worded items in the authoritarian subscale is: “The mentally ill should not be treated as outcasts of society”. Even if participants do not accept this statement, this does not necessarily mean that they support the inclusion of the mentally ill in their community. After examining the factor loadings of the negatively worded items under each subscale in the original scale, the results showed that these items had rather low loadings, compared with the positively worded items. As a result, there is potential to further improve the validity of the scale and examine its structure by omitting the negatively worded items. Third, the content of some statements under one factor might be repeated. For example, when measuring benevolence, attitudes reflected by tax money allocation were described twice in the scale. Finally, the scale was first developed in the 1980s and so the expressions and social context need to be examined under the current context. Although the CAMI scale has been translated to different languages, including Chinese, and validated in different contexts [16–19], validation with an adequate sample size has not been recently carried out in China.

Consequently, the original 40-item CAMI scale will be translated, shortened, and validated in the current study. The main task of this research is to develop a shorter version of the CAMI scale that can be used in the Chinese context, focusing on the direct descriptions of each subscale. The new scale is named the Shortened Chinese Version of the Community Attitudes towards the Mentally Ill Scale (C-CAMI-SF).

For sample-size estimation in the present study, we integrated recommendations from existing discussion. Early conventions for exploratory factor analysis (EFA) propose that the required number of respondents equals to the number of response categories per item [20]. A minimum ratio of 10 participants per item is also widely accepted [21–23], whereas sample size exceeding 300 respondents yield more stable factor solutions [24]. Synthesizing these perspectives, we adopted the 1:10 participant-to-item ratio, thereby targeting a minimum of 400 participants to ensure robust psychometric evaluation.

Participants were recruited through a convenience-sampling strategy. Invitations were disseminated on Chinese-predominant social-media platforms such as WeChat and WhatsApp between September and October 2018. The inclusion criteria were: (1) Chinese; (2) aged 18 years and above; and (3) able to complete the Chinese questionnaire in a self-reported manner. Consent to participate in the study was obtained before the start of the online survey. Participants could quit the online survey at any time without any negative consequences. The Human Research Ethics Committee of the University of Hong Kong (No. EA1807014) approved this study.

A battery of questionnaires was created on an online survey website and could be accessed by the participants through an online survey link. The main scale to be validated was the original CAMI scale. Adaptation and translation were performed in advance by linguistic and mental health professionals. Another three scales were included to test the convergent and discriminant validity of the C-CAMI-SF: the 7-item Social Distance Scale (SDS), the Attribution Questionnaires-9-Item Version (AQ-9), and the Interpersonal Support Evaluation List-12 (ISEL-12). Self-reported demographic characteristics included age, gender, education level, marital status, previous diagnosis of mental illnesses, and the mental health status of family members; these were collected before the intervention.

All data analyses were carried out using the statistical software SPSS 23 (IBM Corp., Armonk, NY, USA). Principal component analysis was conducted to explore the factor structure of the C-CAMI-SF. Factors with a criterion of an eigenvalue >1 will be retained [33]. As the four subscales of the original CAMI scale indicate four independent aspects of stigma attitudes, and the scores of the four subscales cannot be added up directly, factor analyses were conducted respectively for the four subscales. Prior to factor analysis, the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity were performed to assess data suitability. A KMO value > 0.50 and a Bartlett’s test significant at p < 0.05 were required to confirm that factor analysis was appropriate for the dataset [22,23]. Internal consistency was examined using the Cronbach’s alpha coefficient. A value of 0.70 or above was accepted as a satisfactory level of scale reliability [34]. The convergent and discriminant validity were examined using the correlations between the C-CAMI-SF and the other three validity check scales. It was expected that the AT and SR subscales would have positive associations with the SDS and seven factors in the AQ-9, excluding the pity and help factors. Positive associations were also expected between the BN and CMHI subscales and the pity and help factors in the AQ-9. No correlation was expected between the C-CAMI-SF and the ISEL-12 scale.

The reduction of the number of items per subscale was carried out on the basis of the exploratory factor analysis. Items loaded on more than one factor were excluded. If several items expressed similar content, the one with the higher factor loading and effect size was retained. Items with corrected inter-item total correlation loadings less than 0.40 were deleted as well, as loadings of less than 0.40 are considered “weak” [35].

A total of 400 participants satisfied the inclusion criteria among the 455 participants who consented to participate; the responses of these 400 participants were used in the current study. The mean age was 31.40 years (SD = 8.87), and 65% of the participants were women. The majority of participants (69%) were full-time employees and had tertiary education levels (93%). Single (48%) and married (47%) participants were relatively equal. Only 25 of the 400 participants reported had previous diagnoses of mental disorders. Table 1 summarizes the demographic data of the sample.

Factor analyses were conducted respectively for the four subscales. All four subscales satisfied the requirements for factor analysis. The results of the KMO test of the four subscales were 0.70, 0.62, 0.75, and 0.71, respectively. And the results of the Bartlett’s test of the four subscales were all significant with p < 0.001.

EFA focused on the positively worded items of the CAMI scale. Using principal component analysis with a criterion of an eigenvalue >1, all subscales satisfied the single-factor structure. The four-factor orthogonal solution accounted for 53.11% of the variance. Variance accounted for by items in each subscale was 38.25%, 54.83%, 57.66%, and 44.74%, respectively. Factor loadings under each subscale are listed in Table 2. With a significant loading set at 0.40, 17 items in total were retained for the Chinese CAMI-SF Scale.

The Cronbach’s alpha coefficients of the four subscales were 0.58, 0.57, 0.75, and 0.68, respectively. Inter-factor correlations were lower, ranging between −0.16 and 0.52, which indicates the independence of each subscale.

Correlations between the C-CAMI-SF and the validation scales are summarized in Table 3. All four subscales have significant correlations in the expected directions with the SDS and AQ-9. The AT and SR subscales had positive correlations with the SDS, while the BN and CMHI subscales had negative correlations with the SDS. Each item in the AQ-9 represents one aspect of emotional reactions or behaviors in the stigma process [24]. Pity (AQ1) may lead to helping (AQ7) behavior, while blame (AQ4), anger (AQ6), danger (AQ2), and fear (AQ3) may predict avoidance (AQ8), segregation (AQ5), and coercion (AQ9). As expected, the AT and SR subscales had positive correlations with negative emotional responses (AQ2, 3, 4, 6) and discriminative behaviors (AQ5, 8, 9), and negative correlations with supporting emotional responses (AQ1) and behavior (AQ7). The other two subscales, the BN and CMHI subscales, presented the opposite result. All four subscales had insignificant correlations with the ISEL-12.

The mean scores of the four subscales in the current sample were 13.79, 11.69, 11.59, and 18.05, respectively. The theoretical ranges of each subscale were 5–25, 5–15, 5–20, and 5–25, accordingly. The means and standard deviations of each subscale score in different strata of each demographical variable are listed in Table 4. For each variable, an ANOVA was conducted to compare the subscale scores among different strata. All factors except education level had significant influences on the subscale scores.

Female participants presented less restrictive attitudes (p = 0.02) and more welcoming attitudes in regard to integrating PMI in the community (p = 0.03) than male participants. Participants in the 18–29 age group had significantly lower AT (p = 0.013) and SR (p = 0.001) scores than those in the 40–49 age group. Their AT scores were also lower than those of participants aged over 50 (p < 0.001). The SR scores of participants in the 30–39 age group were also lower than those of participants in the 40–49 age group (p = 0.02). Participants with previous diagnoses of mental illness presented less authoritative (p = 0.02) and restrictive attitudes (p = 0.002) toward PMI than those without diagnoses. Participants with family members with mental illnesses showed more tolerant attitudes, with higher BN scores (p = 0.009), than participants without such family members.