Since the beginning of the pandemic, people with HIV have been stigmatized. Mahajan [1
] defines HIV-related stigma as when people living with HIV are labeled, stereotyped, experience separation and status loss, and become discriminated both on an individual and structural level. Across different contexts perceived HIV-related stigma is associated to poor mental and physical health for persons living with HIV [2
]. Even in the current era of efficient treatment, making HIV a chronic illness with normal life expectancy where treatment is generally available, people living with HIV are exposed to and relate to HIV-related stigma [6
] and stigma has been found to be a common barrier to treatment and prevention [8
Stigma has sometimes been understood as an individual process, where stigma is constituted of what some individuals do to others [10
], whereas other scholars have argued that an individual perspective of stigma only may be relevant in highly individualized countries (as the US or some parts of Europe) [11
]. Parker and Aggleton [11
] argue that stigma, where it appears, is strongly related to the specific context of culture and power, and that the social and cultural phenomenon of stigma may be linked to actions of whole groups of people. Some known aspects of HIV-related stigma are also clearly cultural specific, as they, for example, relate to specific religious beliefs [12
The efforts to reduce HIV-related stigma have yet not matched the magnitude of the problem [13
]. Parker and Aggleton suggest that the diversity and complexity of HIV-related stigma makes it difficult to grasp in a programmatically useful way [11
]. Ogden and Nyblade, on the other hand, argue that differences in HIV-related stigma across cultures are largely superficial and that stigma is expressed remarkably consistent across contexts [13
]. Valid and reliable instruments for measuring HIV-related stigma are essential for stigma research, for evaluation of stigma-reducing interventions, and for monitoring and understanding experiences of HIV-related stigma [10
]. Based on findings that suggest consistencies in HIV-related stigma across cultures [13
], it would be valuable to have measures of HIV-related stigma that are valid across cultures, thus enabling measurement of changing trends in HIV-related stigma over time and across contexts.
One of the many instruments designed to measure HIV-related stigma perceived by persons living with HIV is the HIV Stigma Scale by Berger et al. [16
]. Although developed and originally found valid and reliable in a US context [16
], the instrument has been translated and found to be relevant and valid across various countries and cultures [17
]. These findings suggest that HIV-related stigma as assessed by the Berger’s scale is, to some extent, universal. For example, the HIV Stigma Scale has been used to compare levels of perceived HIV-related stigma among persons living with HIV in Kenyan, Puerto Rican, and the United States contexts [3
], where Hispanics reported significantly higher levels of stigma than persons from the African continent. When used in Kenya, Puerto Rico, and the United States, the HIV Stigma Scale showed good internal consistency measured by Cronbach’s α for the combined data. The authors, however, did not explore differential item functioning for items in the HIV Stigma Scale, and it is possible that participants from different countries interpreted the items differently, resulting in bias.
Participants’ interpretation of scales can be assessed with methods based on Item Response Theory (IRT) and analysis of Differential Item Functioning (DIF). As items in the HIV Stigma Scale are statements that participants are requested to agree or disagree with on a four-point Likert scale, the probability that an individual will agree with the statement on a certain item can, according to IRT, be seen as a mathematical function of how stigmatized the person is and how severe the stigma is that the item captures. Items in the HIV Stigma Scale would be considered to have DIF if participants with different sociodemographic backgrounds have unequal probabilities of agreeing with statements in the items, while experiencing the same level of stigma [19
]. For example, an analysis of DIF in an American sample, showed that black, non-Hispanic persons and white, non-Hispanic persons with the same level of stigma, had different probabilities of agreeing with items in the HIV Stigma Scale, indicating that persons with different backgrounds may experience HIV-related stigma differently [20
In addition to its use in several studies in the United States [21
], the HIV Stigma Scale has been used, for example, in Sweden [18
] and South India [17
], providing an ideal opportunity to evaluate whether DIF occurs. The HIV Stigma Scale has been translated into the respective native languages, back translated into English, and checked for comparability with the original English questionnaire [17
]. The English, Tamil, and Swedish versions of the instrument could therefore be considered consistent regarding content and all items were found relevant for both a Swedish and a South Indian context by both experts and people living with HIV in each country [17
]. There are, however, results from the psychometric validation from both the Swedish and the Indian contexts that may indicate that at least parts of the concept of HIV-related stigma are culturally embedded. For the Swedish version, a high rate of missing responses were found for an item regarding the risk of losing employment if one’s HIV status is disclosed, where written comments in the margin of the questionnaire (“Does this happen in Sweden?”) indicated that some Swedish respondents found this item irrelevant [18
]. In validation of the Indian version of the HIV Stigma Scale, respondents had difficulties understanding the four-point Likert scale and the word “unclean” in the item “Having HIV makes me feel unclean” was often misinterpreted to mean “personal hygiene” [17
Although the HIV Stigma Scale has been used for measuring stigma in a wide range of different contexts, it is not clear whether items in the HIV Stigma Scale are interpreted differently by persons with different backgrounds. The aim of the present study was, therefore, to examine whether items in the HIV Stigma Scale function differently with regard to gender and cultural background.
The HIV Stigma Scale was developed for quantitative measurement of HIV-related stigma, as perceived by persons living with HIV. Though adapted and used in diverse contexts, this is the first time that the instrument has been assessed for gender-related DIF and DIF across different cultural contexts. The results of the present study indicate that the items in the HIV stigma scale were not especially prone to present culture-related DIF for the subscales Personalized stigma, Concerns about public attitudes,
and Negative self-image
. These subscales seem to cover aspects of stigma that in general are equally interpreted regarding content across the different groups investigated. These results indicate that levels of Personalized stigma, Concerns about public attitudes,
and Negative self-image
may be compared between the cohorts without the risk of results being culturally biased. Salient DIF was, however, found between the South Indian and Swedish cohorts for the subscale Disclosure concerns
. The subscale Disclosure concerns
measures concerns that one can have over disclosing one’s HIV status to others. The items that present DIF mainly covers the aspect that one’s HIV is a secret and we can only speculate about potential reasons for this DIF. Since India is more densely populated than Sweden, it may be likely that participants in the South Indian cohort live physically closer to family and neighbors than the participants in the Swedish cohort, making it harder for the South Indian participants to keep their HIV a secret. India has been characterized as a ‘collectivist’ society [37
] and Sweden more of an individualistic society [38
], particularly around interpersonal issues. According to Chadda and Deb [37
] family is far more involved in the care of its members in the Indian society, compared to western societies. This might also make it more difficult for the South Indian participants to keep their HIV a secret from their family, compared to the Swedish participants. If keeping one’s HIV a secret is difficult, on the edge of being impossible, in the South Indian context, items like “I work hard to keep my HIV a secret” might have been perceived as irrelevant for South Indian participant, thus generating the detected DIF.
When Rao et al. [20
] examined DIF between Black, non-Hispanic persons and White, non-Hispanic persons, nine items of the scale demonstrated DIF. Our present results did not replicate these findings, as no items were flagged for DIF between Black, non-Hispanic persons and White, non-Hispanic persons. A possible explanation for the differences in results could be that Rao et al. [20
] examined the 40-item version of the HIV Stigma Scale, while the 32-item version was examined in the present study. Seven of the items, from the 40-item full version of the instrument, that demonstrated DIF in the earlier study by Rao et al. were items that cross-loaded in an exploratory factor analysis and therefore were excluded from the 32-item version of the instrument [29
A common reason for DIF is a lack of translation equivalence [19
]. The item “Having HIV makes me feel unclean” (Item 12), for example, could not be adequately translated to Tamil, since no words in Tamil captured the intended meaning of “unclean” [17
]. This was adjusted for by letting the questionnaire be administered by professional raters in the South Indian cohort, who assured that item content was understood as intended [17
]. A similar procedure was used for the Swedish cohort where members of the research team also were present to answer questions and assist respondents with explanations of items if needed [18
]. The item “Having HIV makes me feel unclean” (Item 12) did not demonstrate DIF in the present analysis, but if the HIV Stigma Scale is used as a self-administered instrument as originally intended, the detected DIF may possibly be even more pronounced.
Unidimensionality was assessed with both confirmatory and exploratory methods. Since confirmatory methods indicated that some subscales might not be unidimensional, dimensionality was examined further with both parallel analysis and the Empirical Kaiser Criterion. Parallel analysis [39
] supported unidimensionality for all subscales across all cohorts, while the Empirical Kaiser Criterion suggested two-factor solutions for the South Indian and Swedish cohort. Parallel analysis is an often recommended approach for dimensionality assessment [39
]. The HIV stigma scale is, however, an instrument constituted of oblique, highly correlated factors [16
], and for this specific case the Empirical Kaiser Criterion has been shown to outperform parallel analysis [33
]. We therefor conclude that the subscales Disclosure concerns
and Negative self-image
may be, at least, bi-factorial scales when used in the Swedish or South Indian context. These findings may have implications for the interpretation of the detected DIF, as multidimensionality can be mistaken for DIF [40
There are several techniques available for DIF detection, and there is a lack of consensus regarding thresholds for detection of DIF [28
]. As we used a hybrid ordinal logistic regression—IRT approach, other techniques may have produced different results. Since statistical significance not necessarily implies practical significance, we used a measure of effect size (changes in pseudo R2
) as a criterion for when DIF should be detected [19
]. Different sets of rules have been presented for when pseudo R2
should be considered to represent DIF [19
], where Zumbo [41
] suggested that cut-offs indicating moderate and large DIF should be 0.13 and 0.26, respectively, while the Jodoin and Gierl approach suggests cut-offs of 0.035 and 0.070 [42
]. A cut-off level of 0.02 has also been commonly used and these different cut-offs can, unsurprisingly, produce very different numbers of items flagged for DIF [19
]. In the present work we used Monte Carlo simulations, as implemented in the lordif-package [26
], to generate empirical cut-offs. This rendered cut-offs as low as 0.01 for some analyses, which also resulted in detection of DIF that did not have a salient impact on individual scores. This low threshold for DIF detection was, however, set knowing that cultural DIF probably was adjusted for already in data collection and we sought to find patterns in present DIF that perhaps would have been more pronounced if questionnaires had been exclusively self-administered.
Limitations in the present work are that data were collected at different time points with over 5 years apart and that data from the United States and India were not collected with an intention to be representative for people living with HIV in the respective countries [17
]. In those settings, the study populations represented a sub-set of all people living with HIV. The results, therefore, cannot be generalized to represent differences between countries. Furthermore, we do not know if the results would be replicated if data were collected at the same locations today. A recommended minimum sample size for ordinal logistic regression is 200 participants per group [19
]. Thus, a limitation in the present study is the Swedish sample size of 157–180 participants depending on subscale analyzed. In a simulation study, sample sizes as low as 100 per group were sufficient to detect large DIF but not moderate DIF [43
]. It is therefore possible that the present study failed to detect moderate but practically important DIF. Further limitations were that the datasets differed regarding gender and age, and we did not have access to other sociodemographic data. It is generally known that HIV-related stigma is linked to stigma related to other attributes, which can potentiate the power of stigmatization [1
]; groups of people who are already exposed of racism, homophobia, sexism, or poverty are predisposed to greater HIV-related stigma [1
]. As we did not have access to data other than ratings to the HIV stigma scale and gender and, to some extent, age, we were not able to control for potentially confounding variables that may have been a true source of DIF. For the HIV Stigma Scale it would also be interesting to analyze if DIF occurs between groups of different health statuses, e.g., persons who are virally suppressed or not, persons who have been living with HIV for a longer or shorter period of time.
Aside from these limitations, we propose that the results in the present work indicate that the items in the subscales Personalized stigma, Concerns about public attitudes, and Negative self-image in the HIV Stigma Scale are not especially prone to present salient DIF. However, the detected DIF between the Indian and Swedish cohort for the subscale Disclosure concerns did have a cumulative, albeit small, salient effect on individual IRT scores, which could result in both type I and type II errors if levels of Disclosure concerns should be compared between the Swedish and Indian cohorts. The detected DIF in the subscale Disclosure concerns could be a “true” differential item functioning, i.e., that persons from the Swedish and South Indian cohort respond differently to items regarding “keeping one’s HIV a secret” and “hiding one’s HIV,” respectively, even after accounting for their overall level of disclosure concerns. One possible explanation for this could be differences in the actual possibility to keep private things a secret in the Indian society, compared to the Swedish. An additional possible explanation could be that the unidimensional nature that the subscale Disclosure concerns seems to have in the Swedish and South Indian cohort may have caused a DIF due to multidimensionality. Apart from the cause of the detected DIF, we cannot recommend the subscales Disclosure concerns for comparisons between Sweden and South India.
The results in the present study, however, support the use of the subscales Personalized stigma, Concerns about public attitudes, and Negative self-image for comparisons of levels of stigma between the cohorts investigated. As the HIV Stigma Scale is being used to assess stigma in a wide range of different countries, we encourage researchers using the HIV Stigma Scale to cooperate across country borders and examine the cross-cultural validity of the instrument further. This would broaden the understanding of the extent and forms of stigma faced by people living with HIV in different countries.