Examining the Psychometric Properties of a Revised 40-Item Berger HIV Stigma Scale

Abstract

Over the past decades, as social contexts and knowledge about HIV have evolved, the conceptualization and understanding of HIV stigma, as well as the measures used to assess it, may have also shifted. This study aimed to examine the psychometric properties of a revised version of the Berger scale which updated language to better capture HIV stigma in the Southern U.S. The revisions were informed by focus group discussions with people with HIV (PWH) in Florida. Following three focus groups, the updated scale was tested among 461 PWH. The internal structure was evaluated using exploratory factor analysis (EFA) and confirmatory factor analysis (CFA). The EFA analysis supported a four-factor structure. These four latent factors corresponded to the original dimensions of the Berger scale. All evaluated models demonstrated good model fit indices. The revised scale also showed high reliability, with composite reliability (omega) values for each subscale exceeding 0.89. Measurement invariance testing across race and gender groups further indicated that invariance was upheld. The revised scale also demonstrated high reliability, and composite reliability omega for each sub-scale was over 0.89. We further conducted measurement invariance across race and gender groups, and the measurement invariance was hold. We found that the revised scale is highly reliable, encouraging wider testing of this scale in a variety of populations.

Introduction

There is a well-established connection between diseases and symbolic meanings [1]. HIV, like many other conditions, has been moralized as “punishment” (p.133) and “contamination” (p.115) since the beginning of the epidemic [1]. These HIV metaphors result in stigma, which Erving Goffman, defined as the structural force acted upon groups deemed to have characteristics that were societally discredited [2]. HIV-related stigma not only marginalizes [3] and isolates people with HIV (PWH) but also associates with delayed HIV diagnosis [4], increased symptoms of mental health conditions (i.e., depression) [4], decreased engagement in HIV care, and reduced adherence to antiretroviral therapy (ART) [5]. This, in turn, leads to PWH living with unsuppressed viral loads for longer periods of time, leading to more negative individual health outcomes and greater risk of HIV transmission. As a result, understanding and combating HIV stigma have been recurring and essential research foci throughout the epidemic and has remained a recalcitrant issue into the present day.

To fully understand and address the impacts of HIV-related stigma, it is crucial to use appropriate measures. The 40-item HIV Stigma Scale developed by Berger and colleagues has become one of the most widely used tools for measuring stigma [6]. The scale consists of four dimensions: personalized stigma, which evaluates the perceived consequences of disclosing one’s HIV status; negative self-image, which reflects internalized negative perceptions; disclosure concerns, which focus on intentions and behaviors related to disclosure; and concern with public attitudes, which assesses perceptions of societal views toward HIV. Originally, the scale was developed based on the experiences of a predominantly male population in the United States. Later, this scale was tested and validated across diverse populations domestically [7, 8] and internationally [9–11].

Despite its popularity, the Berger scale has two key limitations. First, the original wording of the Berger scale may be outdated, given the changed social context and theoretical understanding of stigma. Developed several decades ago, the scale reflects a period when stigma was primarily conceptualized as both a personal attribute and an outcome [1]. At that time, stigma was framed as a culturally disapproved characteristic of an individual or as a consequence of attaching negative meaning to that individual, a perspective that treats stigma as static. More recent stigma theorists have emphasized an aspect of Goffman’s stigma definition that considers stigma as a dynamic process labeling, stereotyping, separation, and the deprivation of power, all of which occur during social interactions [12, 13]. This conceptual shift underscores that stigma is deeply embedded in, and shaped by, social contexts. Since the Berger scale’s development, the social landscape surrounding HIV has changed substantially due to advances in ART and prevention strategies such as pre-exposure prophylaxis [14], increased public awareness and targeted stigma-reduction interventions [15], and policy reforms that influence the manifestation and experience of HIV stigma [16]. The mortality rate of HIV/AIDS was five times higher at the time the Berger scale was developed than it is today [17]. These changes may alter not only the prevalence but also the forms of stigma experienced by PWH. In addition to considering the forms of stigma, more inclusive language is needed in stigma scales, as evidence suggests that reading judgmental or stereotypical items may have adverse effects by increasing the accessibility of stereotypes [18].

Sample population and contextual characteristics play an important role in recent stigma frameworks. Earnshaw and Chaudoir [19] advanced the theoretical framework by complementing societal-level stigma with the individual-level processes through which stigma is experienced and internalized. They specified the HIV stigma mechanism and argued that PWH devalued themselves through enacted (i.e., experienced prejudgment and discrimination from one’s community), anticipated (i.e., belief in future prejudgment and discrimination), and internalized stigma (i.e., acceptance of negative attitudes and feelings toward their HIV status). From this perspective, the composition of study samples is critical, as multiple social identities may intersect to produce intersectional stigma defined as the joint effect of the convergence of stigmatized identities [20]. Recent surveillance data indicates that there are disparities in rates of HIV, with Hispanic and Black men who have sex with men, as well as Black women in the South, being disproportionately affected [21]. Yet, in the original Berger scale sample, only 18 participants self-identified as Black women (5.9% of the total sample) and 15 as Hispanic men (4.9% of the total sample), raising the question of whether the original wording adequately captures the forms of stigma experienced by these minority groups [6].

Second, the original Berger scale contains several cross-loading items, which complicates the interpretation of its dimensions. Specifically, five items load onto three subscales and 11 items load onto two subscales. Moreover, factor analysis studies assessing the validity of the original scale or its shortened versions have yielded mixed findings [22]. According to a recent systematic review [22], only two studies have conducted factor analyses on the original 40-item version. One study, using confirmatory factor analysis (CFA), reported poor to sufficient model fit indices, suggesting an unstable factor structure [23]. Another study, employing exploratory factor analysis (EFA), identified five subscales, four of which aligned with the original study’s structure [24]. Notably, these two studies were conducted on populations significantly different from the original, suggesting that HIV-related stigma may vary across social and cultural contexts [23, 24].

Berger’ s scale, however, remains valuable even as stigma research has evolved and social context has changed. First, the four dimensions of the Berger scale overlap with recent conceptualizations of HIV stigma, including enacted, anticipated, and internalized stigma. Personalized stigma and negative self-image reflect the acceptance of negative attributes, which overlap with internalized stigma in the HIV stigma framework. Individuals’ concerns about disclosure are significantly shaped by enacted stigma, while their perceptions of societal attitudes toward HIV reflect both enacted and anticipated stigma. Second, a systematic review identified 166 studies reporting psychometric properties, in which the majority of studies found an acceptable to excellent reliability with Cronbach’s alpha greater than 0.7 [22]. There are two possible reasons for the poor model fit indices in previous studies. As discussed above, the first reason is the change in social contexts. More specifically, original questionnaire’s wording may be outdated. Second, the cutoff criteria for CFA assumes that no cross-loading items are present, which was against the original Berger scale factor structure.

To address the challenges mentioned above, our study aimed to examine the psychometric properties of the revised version of the Berger scale that updated new language and questions to capture HIV stigma in southern U.S. based on focus group discussions with PWH in Florida.

Methods

Development of Revised HIV Stigma Scale

Between 2020 and 2021, our study team conducted three structured focus groups with PWH to better understand the experience of HIV-related stigma and elicit feedback on two widely used HIV-related stigma measures which included Berger’s scale. Participants were recruited from Southern HIV and Alcohol Research Consortium [NIH U24AA020003] contact registry of PWH, HIV community-based organizational leaders, and public health experts. Participants were eligible if they were 18 years of age or older and could speak English. In total, eleven participants were recruited, including two groups of community-dwelling stakeholders (n = 8) and one group of community-based organization stakeholders (n = 3). Among the eleven participants, six self-identified as male and five as female. In terms of race and ethnicity, three identified as Asian, four as White, two as Hispanic, one as Black, and one as Native American.

During the focus groups, participants reviewed the Berger’s 40-item HIV-related Stigma Scale [6] and Sowell’s 13-item Perceptions of Stigma scale [25], which was originally developed to assess perceived stigma among women living in rural communities [25]. Focus group participants were asked: (1) What are some of the items in these questionnaires that you find confusing and how would you rewrite the items in your words?, (2) What are some of the items in these questionnaires that might not trigger an experience/memory about HIV-related stigma?, (3) What are some of the items that you find difficult to generate a response?, (4) What are some items not asked in the questionnaire that you would like to add?, and (5) In general, tell us in detail which questionnaire you think better represents the experience of HIV-related stigma? Based on the focus group participants’ responses to these questions, Berger’s HIV-related Stigma Scale was determined to be better representation of HIV-related stigma.

After a qualitative content analysis of the focus group data, modifications were made to 17 of the 40 items in Berger’s HIV-related Stigma Scale [26] (Table 1). Triangulation was used to validate the modifications [27]. To evaluate consistent meaning of the items that were modified, the 17 items were reviewed by four experts in measurement, three HIV clinical experts, and three PWH. The 17 modified items were endorsed unanimously by these reviewers.

Table 1.

List of revised items

Original	Updated
People’s attitudes about HIV make me feel worse about myself	People or society’s attitudes about HIV make me feel worse about myself
Telling someone I have HIV is risky	I feel unsafe telling someone that I have HIV
People with HIV lose jobs when employers find out	People with HIV might fear negative reactions from coworkers if they find out
I feel I am not as good a person as others because I have HIV	I feel inferior to others because I have HIV
It is easier to avoid new friendships than worry about telling someone that I have HIV	I tend to avoid new relationships because I have HIV
Having HIV makes me feel I am a bad person	Having HIV makes me feel I am not good enough
Some people who know I have HIV have grown more distant	I became distant from others who know my HIV status
Most people are uncomfortable around someone with HIV	People might feel uncomfortable around me because I have HIV
I never feel the need to hide the fact I have HIV (Reverse)	At times, I feel the need to hide the fact I have HIV
Having HIV in my body is disgusting to me	I feel disgusted when I think about having HIV
As a rule, telling others that I have HIV has been a mistake	Telling others that I have HIV has been a mistake
People I care about stopped calling after learning I have HIV	People I care about have become distant from me after learning I have HIV
People have told me that getting HIV is what I deserve for how I lived my life	People blame me for getting HIV because of my lifestyle
People don’t want me around their children once they know I have HIV	people don’t want me around their family or friends once they know I have HIV
I have lost friends by telling them I have HIV	I have lost relationships to other people by telling them I have HIV
People who know I have HIV tend to ignore my good points	People who know I have HIV tend to ignore good things about me
When people learn you have HIV, they look for flaws in your character	People focus on mistakes in you when they find out you have HIV

Study Sample

To assess the psychometric properties of the revised Berger HIV scale, it was incorporated into the Florida Cohort study Wave 3, a prospective longitudinal cohort study to examine healthcare utilization and outcomes among PWH living in the state. Between October 2020 to December 2023, participants were recruited from a collaborative network of county health departments, community-based organizations, and Ryan White and academic clinics in Alachua, Marion, Columbia, Duval, Brevard, Miami-Dade, Hillsborough, and Polk Counties. Inclusion criteria were confirmed HIV infection (verified by linkage to the state surveillance system), aged 18 or older, and receiving HIV care in Florida. The recruited participants were asked to complete baseline questionnaire, and their questionnaire data were further linked to statewide HIV surveillance data. A total of 461 PWH completed the revised HIV stigma scale and other measures as part of the Florida Cohort study and was used as the study sample for the current analysis.

Measures

To examine the convergent and discriminant validity of the revised Berger’s scale, we assessed respondents’ mental health status, alcohol and substance use, HIV discrimination experience, and HIV disclosure.

Mental Health Status

Symptoms of depression and anxiety were assessed via Patient Health Questionnaire depression scale (PHQ-8) [28] and the Generalized Anxiety Disorder scale (GAD-7) [29], respectively. These scales include items with response options of four-point Likert scales ranging from “not at all” to “nearly every day.” The depression scale includes eight items designed to assess whether participants experienced: little interest in doing things, feeling depressed or down, tiredness, poor appetite or overeating, feeling bad about themselves, difficulties in focusing, and moving or speaking slowly (Cronbach’s alpha = 0.89 in this study). As suggested by a previous study [26], this study used a 10-point cutoff score to identify depressive disorder. Anxiety disorder was measured using six items, including feelings of nervousness, inability to stop worrying, excessive worrying, difficulties in relaxing, restlessness, easy irritability, and fear of something awful happening (Cronbach’s alpha = 0.89 in this study). The cutoff score used in this study to identify anxiety disorder is 8 [29].

Alcohol Use

Alcohol use was measured using the Alcohol Use Disorders Identification Test – Consumption (AUDIT-C) [30]. The score ranges from 0 to 12, with higher scores indicating a greater risk of alcohol use disorder. For participants who had not consumed alcohol in the past 12 months or ever, their AUDIT-C score was recorded as 0. All other participants were then asked to complete the AUDIT-C questions. The first two items aim to evaluate respondents’ alcohol consumption volume by asking about the frequency of alcohol consumption and the amount of alcohol consumed per occasion. The last item assesses excessive drinking experiences in the past year by asking the frequency of having more than six drinks. As suggested, a cutoff score of 4 was used to identify participants with greater risk of alcohol abuse [30]. The Cronbach’s alpha of this scale in our sample was 0.77.

Substance Use

Substance use in the past 12 months was self-reported and included marijuana or hashish, cocaine or crack cocaine, heroin, stimulants, non-prescription use of opioids, ecstasy/MDMA, hallucinogens, poppers, and other drugs. Responses were coded into two variables: one binary variable representing any or no substance use, and another continuous variable representing the total number of types of substances used with a range from 0 to 9.

HIV-Related Discrimination

Deacon [31] suggested separating discrimination from stigma because discrimination emphasizes the behaviors and outcomes of the process. Experiences of HIV-related discrimination were assessed by asking participants if they had experienced any of nine specific types of discrimination in the past. These nine types of discrimination included being hurt by people’s reactions upon learning about their HIV status, being avoided by someone who knew they had HIV (such as being touched, having others stay distant, being physically backed away from, having people stop socializing, and being rejected in a relationship), and seeing posts stigmatizing HIV on social media. Based on the responses, we calculated the total number of types of discrimination they experienced in their lifetime.

HIV Disclosure

We measured participants’ HIV disclosure status by comparing the size of their social networks (e.g., best friends and family) with the number of people to whom they disclosed their status. Participants were grouped into three categories [32]: (1) individuals without a social network, (2) those with incomplete disclosure, defined as disclosing to none or some people in their close network, and (3) those with complete disclosure, defined as disclosing to all people in their close social network.

Demographic Information

Last, we also assess respondents’ socio-demographic information, including age, race, ethnicity, gender identity, sex at birth, highest level of education completed, current marital status, employment, household income last year, and years since they were diagnosed with HIV.

Statistical Analysis

Less than 1% of responses contained missing values. We conducted Little’s test to assess whether the missing data in our dataset was missing completely at random (MCAR). The results of the test indicated that the missing data were missing completely at random (χ 2 (487, N = 462) = 487, p =.203). Hence, we handled missing data by using listwise deletion. To assess the construct validity of the scale, we first randomly split total sample into two datasets: one for EFA (n = 260) and the other for CFA (n = 261).

Exploratory Factor Analysis (EFA)

The Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity were conducted to assess the suitability of the data for EFA. To determine the number of factors to retain, both the scree plot and parallel analysis were utilized. The scree plot and parallel analysis both suggested a four-factor solution, which is consistent with the original Berger scale. We employed principal axis factoring, which is appropriate for identifying the underlying structure of theoretically derived items, along with oblique rotation, which permits correlations among factors [33]. Given that the original Berger scale includes cross-loading items, this study used a factor loading threshold of 0.40 as the criterion for identifying cross-loadings [34].

Confirmatory Factor Analysis (CFA)

We then employed a CFA to verify the structure of the revised Berge’s scale. Specifically, we analyzed and compared the goodness of fit of two estimated models, the four-factor model and the hierarchical model [35]. In the four-factor structure, 40 items were loaded onto one of the four latent factors, as identified through the EFA. For the hierarchical model, we later introduced a second-order latent factor to the model, overlaying the four latent factors onto one higher-order latent factor – stigma. We treated the Likert scales as ordinal variables, and hence, diagonally weighted least squares was applied. The following criteria were applied to assess the model fit [36–39]: (a) χ 2/df = [2, 5]; (b) RMSEA = [0.05, 0.10]; (c) SRMR < 0.08; (d) Goodness of Fit Index (GFI) >0.90; (e) Bentler-Bonett Normed Fit Index (NFI) >0.90; (f) Tucker-Lewis Index (TLI) >0.90; and (g) Comparative Fit Index (CFI) >0.95.

Reliability, Convergent and Discriminant Validity

To assess the reliability of the revised scale, we examined its internal consistency using composite reliability (omega). In addition, we evaluated the construct validity of the revised scale across its four subscales, focusing on both convergent and discriminant validity between measures and within scale. We computed Spearman’s rank correlations for the ordinal or binary variables and Pearson correlations for continuous variables between the revised scale and other established scales. We hypothesized that HIV stigma would be positively correlated with HIV discrimination, with the personalized stigma subscale showing the strongest association due to its conceptual overlap. We also expected a strong negative correlation with HIV disclosure, as higher stigma is theorized to reduce individuals’ willingness to disclose their status. These findings would support convergent validity. We anticipated weaker correlations with mental health conditions (anxiety and depression), as stigma contributes to psychological distress but remains conceptually distinct. Finally, we expected small or no associations with alcohol and substance use, which are not core components of HIV stigma, supporting discriminant validity. In addition to between construct evidence, within-scale convergent validity was assessed using the following criteria [40]: (a) standardized factor loadings greater than 0.50, (b) average variance extracted (AVE) values above 0.50, and (c) composite reliability (omega) above 0.70. These indicators were used in combination to determine whether each subscale adequately captured the same concept, stigma. Within-scale discriminant validity, was assessed using the following criteria: (a) the AVE for each construct exceeding the squared correlations between constructs (Fornell-Larcker criterion), and (b) Heterotrait-Monotrait (HTMT) ratios below 0.85.

Multiple Group Invariance

We conducted a measurement invariance analysis of the revised scale across different subgroups within the total sample. First, we examined invariance across racial groups by categorizing participants into two groups: racial minorities and White individuals. This classification was chosen to ensure sufficient sample sizes for each group. Next, we assessed measurement invariance across sex assigned at birth, comparing participants identified as female and male. The following criteria were used to assess measurement invariance. For evaluating loading invariance, we considered: (a) a change in CFI less than 0.010, (b) a change in RMSEA less than 0.015, or (c) a change in SRMR less than 0.030. For testing intercept or residual invariance, we applied the same thresholds for CFI and RMSEA, but used a stricter criterion for SRMR, with (c) a change less than 0.010 [41].

Results

Table 2 summarizes the socio-demographic characteristics of 461 PWH included in the analysis. The age range of the final sample was between 20 and 80 years, with a mean age of 51 years (SD = 12.6). On average, participants had been living with an HIV diagnosis for 20 years (SD = 10.9). Half identified as Black (50.5%), followed by White (38.2%) and multiracial (6.9%). More than half self-identified as male (53.5%) and 56% were assigned male at birth. Nearly half (48.7%) had some college or higher education, 28.7% completed grade 12 or GED. Half were single or never married, 28.9% were divorced, separated, or widowed, and 14.1% were married. One-third were employed, while 42.8% were unable to work or disabled, and 22.4% were unemployed. 20% had a household income below $5,000, and 23.1% had an income between $10,000 and $19,999. Among all participants, about 46% reported substance use in the past year, while the majority (83.5%) were classified as having a lower risk of alcohol use disorder. However, based on the cutoff scores, most participants met the criteria for both a depressive disorder (75.9%) and an anxiety disorder (76.1%).

Table 2.

Socio-demographic characteristics

Variable	N=4611, n (%)/mean (SD)
Age (Mean, SD)	51.0 (12.6)
Years since diagnosis (Mean, SD)	20.3 (10.9)
Race and ethnicity
Non-hispanic black	215 (46.7%)
Non-hispanic white	146 (31.7%)
Non-hispanic multiracial	26 (5.7%)
Hispanic	69(15.0%)
Others and missing	5 (1.1%)
Gender
Male	246 (53.5%)
Female	203 (44.1%)
Transgender male (female at birth)	2 (0.4%)
Transgender female (male at birth)	5 (1.1%)
Other	4 (0.9%)
Sex at birth
Male	258 (56.0%)
Female	203 (44.0%)
Education
Did not finish high school (grade 12)	104 (22.6%)
Grade 12 or GED	132 (28.7%)
Some college or more	224 (48.7%)
Marital status
Single/never married	233 (50.5%)
Living with long term partner	30 (6.5%)
Married	65 (14.1%)
Divorced, separated, widowed	133 (28.9%)
Employment
Employed	160 (34.8%)
Unemployed	103 (22.4%)
Unable to work or disabled	197 (42.8%)
Household Income last year
Less than $5,000	95 (20.7%)
$5,000 to 9,999	75 (16.4%)
$10,000 to 19,999	106 (23.1%)
$20,000 to 29,999	59 (12.9%)
$30,000 to 39,999	49 (10.7%)
$40,000 to 49,999	23 (5.0%)
Move than $50,000	51 (11.1%)
Have ever used substances in past 12 months
Yes	212 (46.0%)
No	249 (54.0%)
AUDIT-C (Mean, SD)	2.2 (2.6)
Greater risk of alcohol abuse	76 (16.5%)
Lower risk of alcohol abuse	384 (83.5%)
Depression (Mean, SD)	14.1 (5.7)
With depressive disorder	344 (75.9%)
Without depressive disorder	109 (24.1%)
Anxiety (Mean, SD)	12.5 (5.7)
With anxiety disorder	348 (76.1%)
Without anxiety disorder	109 (23.9%)

¹Mean (SD); n (%)

^*Not all numbers sum up to 461 due to missing (< 2.5%)

Exploratory Factor Analysis

We first conducted EFA on all 40 items from half of the randomly selected sample (n = 230) to examine the latent structure of the revised scale. The Kaiser–Meyer–Olkin (KMO) sampling adequacy test (KMO = 0.95) and the Bartlett’s sphericity test (p <.001) indicated that the sample was suitable to conduct factor analysis.

Results from the EFA with this four-factor structure indicated a relatively good fit (see Table 3), and the factor loadings are presented in Appendix. Fifteen items (items 33, 39, 36, 35, 29, 32, 38, 28, 40, 31, 26, 34, 30, 27, 24) loaded onto Factor One, identified as “personalized stigma.” All 15 items had loadings greater than 0.4, with no cross-loadings onto other factors. Factor Two, identified as “disclosure concern,” comprised 11 items (items 21, 4, 22, 17, 6, 25, 37, 5, 19, 11, 1), all of which also loaded above 0.4 without double-loadings. Factor Three, labeled “negative self-image,” included nine items (items 15, 12, 23, 7, 13, 2, 3, 18, 8). Except for item 8, a reverse-coded item with a loading of 0.31, all other items loaded above 0.4 and did not cross-load. Finally, Factor Four, identified as “public attitude concern,” consisted of five items (items 10, 9, 14, 20, 16), all of which had loadings greater than 0.4 and showed no double-loadings.

Table 3.

Model fit indices

Indices	EFA	Four-factor	Hierarchical
X2/df		2.12	2.15
RMSEA	0.055; CI [0.049, 0.061]	0.071; CI [0.066, 0.075]	0.071; CI [0.067, 0.076]
RMR-SRMR	0.035	0.081	0.081
GFI		0.983	0.983
NFI		0.982	0.982
TLI	0.905	0.990	0.990
CFI		0.990	0.990

Confirmatory Factor Analysis

CFA was further conducted on the remaining sample (n = 231) to examine the construct validity. The CFI, NFI, TLI, and GFI for both models exceeded 0.96, suggesting a good fit. Moreover, the four-factor model and hierarchical model have the same RMR-SRMR and RMSEA. Regarding the χ ²/df value, the four-factor model has a slightly lower value than the hierarchical model. The chi-square difference test indicated that there was no significant difference between the four-factor model and the hierarchical model; Δχ 2(2) = 4.30; p =.116.

For the four-factor model, in the personalized stigma and public attitude concern dimensions, item loadings were all above 0.66 (see Fig. 1); for the disclosure concern dimension, one item had a factor loading of 0.35, with others above 0.69; in the negative self-image dimension, one item had a factor loading of 0.12, with others above 0.68. Lastly, the hierarchical model indicated high factor loadings of all four subdimensions (personalized stigma = 0.77; disclosure concerns = 0.87; negative self-image = 0.93; and public attitude concerns = 0.87) onto the higher latent factor (see Fig. 2).

Fig. 1.

Factor loadings of CFA with a four-factor structure

Fig. 2.

Factor loadings of CFA with a 4-factor structure and a second-order latent factor

The mean score of the overall stigma scale was 96.45 (SD = 28.59), with a maximum score of 160 and a minimum score of 40. The mean scores for the four subdimensions were as follows: personalized stigma, 32.5 (SD = 13.17, max = 60, min = 15); disclosure concerns, 31.34 (SD = 9.08, max = 44, min = 11); negative self-image, 18.4 (SD = 7.25, max = 36, min = 9); and public attitude concern, 14.04 (SD = 4.12, max = 20, min = 5). We also examined the correlations among 40 scale items, and no negative association was observed.

Reliability, Convergent and Discriminant Validity

Composite reliability (CR; omega) indicated that all four subscales demonstrated strong reliability: personalized stigma = 0.982, disclosure concern = 0.924, negative self-image = 0.894, and public attitude concern = 0.900.

Table 4 summarizes the correlations among the total stigma scale score, sub-scale scores, and measures for other constructs. The total score and sub-scale scores were significantly and positively correlated with HIV discrimination, depression, and anxiety (r >.22) and negatively correlated with their HIV disclosure status (r = −.19), supporting the between-construct convergent validity. We did not observe significant correlations between the revised stigma scale and the alcohol AUDIT score or substance use, supporting the between-construct discriminant validity.

Table 4.

Correlation of HIV stigma subscales and overall stigma score with other measures for assessing convergent and discriminant validity

Variables	Scale-T	Personalized	Disclosure	Negative	Public	Depression	Anxiety	HIV-DS	Sub-T	Sub-U	R-AD
Scale-T
Personalize	0.88 ***
Disclosure	0.84 ***	0.57 ***
Negative	0.83 ***	0.62 ***	0.67 ***
Public	0.76 ***	0.58 ***	0.64 ***	0.59***
Depression	0.32***	0.28 ***	0.22 ***	0.34***	0.23***
Anxiety	0.34***	0.32 ***	0.22 ***	0.35***	0.26***	0.83***
HIV-DS	0.63***	0.7***	0.39***	0.46***	0.46***	0.28***	0.35***
Sub-T	0.07	0.05	0.05	0.06	0.11*	0.2***	0.25***	0.08
Sub-U	0.07	0.05	0.05	0.07	0.11*	0.17***	0.21***	0.07	0.96***
AD	0.02	−0.04	0.03	0.06	0.05	0.13**	0.16***	−0.02	0.4***	0.37***
HIV_SD	−0.19***	−0.1*	−0.3***	−0.18***	−0.05	−0.05	0.02	0.04	0.02	0.04	−0.03

Scale-T stigma scale total; Personalized personalized stigma; Disclosure disclosure concern; Negative negative self image; Public public attitude concern; HIV-DS HIV discrimination; Sub-T total types of substance use; Sub-U substance use status; R-AD risk of alcohol disorder; HIV_SD HIV social network disclosure

^*p<.05

^**p<.01

^***p<.001

We further assessed convergent and discriminant validity within-construct. Table 5 presents the descriptive statistics for each subscale. The within-construct correlations were higher than between-construct correlations. All CR values exceeded the recommended threshold of 0.70, and the average variance extracted (AVE) values for all four subscales were above 0.50. Except for two items, all standardized factor loadings were greater than 0.50, indicating acceptable convergent validity. For discriminant validity, the AVE values for any pair of subscales were greater than their corresponding squared inter-factor correlations. Additionally, the HTMT ratios between all pairs of subscales were below 0.85, supporting strong discriminant validity.

Table 5.

Composite Reliability, average variance Extracted, and Heterotrait-Monotrait (HTMT)

	CR (ω)	AVE	Personalized - HTMT; SQ-C	Disclosure - HTMT; SQ-C	Negative-HTMT; SQ-C
Personalized	0.982	0.677
Disclosure	0.923	0.615	0.600, CI [0.493, 0.692]; 0.313
Negative	0.894	0.594	0.670, CI [0.565, 0.762]; 0.386	0.751, CI [0.676, 0.818]; 0.466
Public	0.900	0.697	0.639, CI [0.543, 0.723]; 0.335	0.742, CI [0.649, 0.815]; 0.424	0.696, CI [0.601, 0.773]; 0.356

Personalized personalized stigma; Disclosure disclosure concern; Negative negative self image; Public public attitude concern; AVE average variance extracted; HTMT heterotrait-monotrait ratio; SQ-C squared correlation

Multiple Group Analysis

The goodness-of-fit statistics for the measurement invariance of the revised Berger scale across racial groups and sex assigned at birth are presented in Table 6. Tests of invariance across racial groups supported metric (weak), scaler (strong), and strict invariance, as did the tests conducted across sex assigned at birth groups.

Table 6.

Measurement invariance

	CFI	Δ	TLI	Δ	RMSEA	Δ	SRMR	Δ
Minority versus non-minority
Configural	0.993		0.992		0.062		0.074
Loading	0.990	0.003	0.989	0.003	0.072	0.01	0.082	0.008
Intercepts	0.992	0.002	0.992	0.003	0.062	0.01	0.075	0.007
Residuals	0.992	0	0.992	0	0.062	0	0.075	0
Sex at birth
Configural	0.993		0.992		0.062		0.074
Loading	0.991	0.002	0.991	0.001	0.066	0.004	0.078	0.004
Intercepts	0.993	0.002	0.993	0.002	0.060	0.006	0.075	0.003
Residuals	0.993	0	0.993	0	0.060	0	0.075	0

Discussion

We sought to determine the ability of the Berger scale to measure stigma experienced by PWH currently given that the social, political, and medical contexts surrounding HIV have changed since the scale was created in 2001. Using the results from our focus groups of PWH in Florida in which we determined their reactions to and feedback on the original Berger scale, we updated 17 items in the original scale. Then, the psychometric properties of the updated version of the Berger scale were assessed among 461 PWH residents in Florida.

Our research has provided strong evidence supporting the reliability and validity of what we propose as the updated Berger scale, suggesting its readiness for future application and testing. Our updated version of the Berger scale demonstrated excellent reliability with composite reliability omega over 0.89. Whereas the original Berger scale has been criticized for its unstable factor structure, our EFA results indicated that only one item loaded onto a single latent factor. The EFA findings were generally consistent with the original Berger scale, revealing four subscales. However, we observed changes in the subscale assignment of six items. Four of these were updated items: “People focus on mistakes in you when they find out you have HIV” (item 40); “People with HIV might fear negative reactions from coworkers if they find out” (item 5); “I tend to avoid new relationships because I have HIV” (item 11); and “I became distant from others who know my HIV status” (item 18). Except for item 5, the other three items were originally cross-loading onto multiple subscales, one of which corresponds to the subscale assignment identified in our study. In addition, item 13 “Since learning I have HIV, I feel set apart and isolated from the rest of the world” and item 19 “Since learning I have HIV, I worry about people discriminating against me” were only loaded onto one sub-scale in our study but they were cross-loading to more than 2 sub-scale in the original study. We used the same rotation and factoring method as the original study. Variations in subscale assignment may reflect differences in the demographic and socioeconomic characteristics of the two study populations, as well as temporal changes in how individuals interpret and respond to the items over the two decades between studies.

The CFA and the difference test suggested that the four-factor model of the original Berger scale has superior model performance, suggesting that the revised Berger scale is valuable in understanding and measuring the dimensions of HIV stigma. We also noticed that two dimensions, disclosure concern and negative self-image, each contained one item with a low factor loading. The lowest factor loading in the Disclosure Concerns subscale was observed for item 1 (“In many areas of my life, no one knows I have HIV”), with a loading of 0.35. The lowest factor loading in the Negative Self-Image subscale was for the reverse-coded item “I never feel ashamed of having HIV,” with a loading of 0.12. Low factor loadings indicate that an item shares little to no variance with other items that are theoretically within the same latent factor. In our study, item 1 loaded onto the same dimension as in the original Berger scale; however, its low shared variance may reflect changes in the social context. Specifically, PWH may still have disclosure concerns, yet those around them may already be aware of their status, as the average duration since diagnosis in our sample (20 years) was substantially longer than in the original study (4 years). The second lowest loading was for the reverse item. While reverse items can serve useful purposes, such as mitigating agreement bias and expanding the range of beliefs captured in respondents’ answers, they can also produce undesirable consequences, including low factor loadings, low item–total correlations, and increased susceptibility to misinterpretation by participants [42]. Within this subscale, we also revised another item, “having HIV makes me feel shameful,” which is conceptually similar to the reverse-coded item. Therefore, future studies may consider either removing or retaining the reverse item while ensuring that conceptually similar items are dispersed within the scale. The multigroup invariance analysis further indicated that the revised scale demonstrated stable performance across different groups.

Convergent and discriminant validity were also supported. Our results indicated that the four dimensions of the updated Berger scale were highly positively correlated with each other and the total scale score. In addition, our research also found that the updated version was significantly associated with HIV discrimination, depression, and anxiety, which was consistent with previous studies that found positive association between HIV stigma and mental health [38–40]. A weak but statistically non-significant positive correlation was observed between the stigma score and alcohol AUDIT score or substance use. Furthermore, the results also supported the convergent and discriminant validity among four sub-scales.

HIV stigma remains a significant barrier to ending the HIV epidemic due to its impacts on physical and mental health of PWH as well as its negative effects on the use of HIV treatment and prevention services [43, 44]. Though stigma exists and is enacted interpersonally and in the community, most interventions to reduce HIV transmission and improve outcomes are focused on addressing individual attitudes, beliefs, and actions [45, 46]. Thus, understanding how HIV operates at the individual level is essential to make better interventions that improve health outcomes of PWH and prevent HIV transmission. Gaining this understanding will require improved scales to measure HIV stigma that captures its true impact as lived and experienced by PWH. This study aimed to meet this need by updating the existing Berger scale to measure HIV stigma as experienced by current PWH, with the goal of creating a measurement instrument that would more accurately and completely capture the stigma experienced by PWH given the changed landscape of HIV.

Since its initial development two decades ago, the Berger scale and its abridged versions have been validated in diverse populations. One line of research has focused on culturally adapting the scale, with an emphasis on translation and linguistic accuracy. For example, a study conducted in Myanmar identified three items whose meanings changed during translation and addressed this issue through retranslation [47]. Similarly, a study involving Tanzania PWH who use drugs prioritized translation accuracy and incorporated new items to capture intersectional stigma [48]. Kamitani and colleagues, working with Asian American participants who spoke English as a second language, adapted certain items in the Berger scale to better reflect collectivist cultural orientations [49]. To our knowledge, our study is the first to examine the appropriateness of item wording for South American populations, rather than focusing solely on translation accuracy or cultural adaptation. Our work contributes to the literature by providing a revised version of the Berger scale designed for a sample comprising multiple racial groups. However, it should be noted that our sample included a limited number of youth and transgender participants. It is possible that some items may require further revision for these groups, as intersecting identities can merge to produce unique forms of intersectional stigma.

A strength of this study was that it was informed by and tested directly with multiple groups of PWH and field experts. This ensured that people were able to provide feedback based on their real lived experiences, which allowed us to be confident that we were asking the correct questions in order to accurately measure HIV stigma. This study is subject to several limitations. First, the focus groups conducted for scale revision and study sample used in the current analysis were both conducted among PWH in Florida. Besides, as the majority of recruitment was at department of health or non-profit clinics, the majority of the participants would fall in the under 400% poverty line and only 11.1% reported an annual income over $50,000, which could limit the generalizability of the results. The scale should be tested in other settings and among diverse groups to address this. Moreover, we observed a high prevalence of elevated psychological distress in our sample. Elevated psychological distress is closely associated with perceived HIV stigma, which may introduce bias. Such distress may be attributable to the low socioeconomic status of participants in our sample; however, it is also possible that perceived HIV stigma contributes to elevated psychological distress. Future research should examine the influence of psychological distress on perceived HIV stigma in other populations. Finally, as this scale makes updates upon the original Berger scale to better measure HIV stigma in today’s world, this scale will likely need to be updated in the future as the HIV world changes.

Conclusion

HIV has dramatically evolved in the four decades since the epidemic began, as has our collective knowledge and attitudes about HIV. With these changes comes a need for researchers, healthcare providers, and public health professionals to have a more comprehensive understanding of HIV stigma in order to address its effects. Using information collected from PWH in focus group interviews, we updated one of the most widely used measures of HIV stigma, the 2001 Berger scale, and tested it among a sample of 461 PWH in Florida. We found that the revised scale demonstrated high reliability and validity among Florida residents. Given the similarities between Florida’s population and those of other states, this revised scale has potential applicability beyond Florida. However, our sample primarily comprised low-income residents and exhibited elevated psychological distress. Future studies should test this scale with more diverse populations to further establish its generalizability. We hope this scale will enable improved research on HIV stigma and enhance HIV treatment and prevention efforts.

Appendix

Pattern Matrix for Exploratory Factor Analysis (EFA)

See Table 7.

Table 7.

Results of EFA (Principal axis factoring, Promax rotation, N = 260)

Item	Factor 1	Factor 2	Factor 3	Factor 4	h2
S33	0.84	−0.06	0.01	0.04	0.70
S39	0.82	−0.09	0.03	0.11	0.72
S36	0.79	0.04	0.03	−0.09	0.64
S35	0.76	0.02	0.00	0.10	0.67
S29	0.76	−0.12	0.10	−0.02	0.58
S32	0.76	0.01	0.09	0.02	0.68
S38	0.76	−0.01	0.11	0.03	0.70
S28	0.74	0.10	0.02	−0.06	0.62
S40	0.70	−0.04	0.04	0.24	0.68
S31	0.69	0.10	−0.05	0.05	0.55
S26	0.64	0.30	−0.02	−0.22	0.57
S34	0.64	0.05	0.12	0.10	0.61
S30	0.56	0.19	−0.09	0.06	0.43
S27	0.54	0.37	−0.01	−0.14	0.54
S24	0.53	0.14	0.01	0.11	0.45
S21	0.08	0.79	−0.01	0.01	0.68
S4	−0.08	0.68	0.16	−0.03	0.53
S22	0.05	0.68	0.06	0.17	0.66
S17	−0.06	0.66	−0.02	0.14	0.46
S6	0.04	0.63	0.18	−0.07	0.55
S25	0.29	0.61	0.04	−0.02	0.66
S37	0.14	0.59	0.08	0.07	0.56
S5	0.03	0.51	0.00	0.20	0.38
S19	0.14	0.50	0.09	0.22	0.57
S11	−0.04	0.47	0.34	0.09	0.53
S1	−0.35	0.47	0.13	0.02	0.22
S15	−0.07	0.04	0.82	0.08	0.69
S12	0.08	−0.11	0.81	0.00	0.63
S23	0.13	0.03	0.74	−0.08	0.65
S7	−0.03	0.10	0.72	−0.04	0.56
S13	0.16	0.06	0.65	0.10	0.70
S2	−0.06	0.13	0.64	0.07	0.51
S3	0.10	0.17	0.45	0.19	0.52
S18	0.35	0.08	0.43	−0.07	0.49
S8	−0.01	−0.05	0.31	−0.11	0.07
S10	0.07	0.11	−0.05	0.70	0.57
S9	0.14	0.10	0.05	0.63	0.59
S14	0.08	−0.03	0.25	0.60	0.59
S20	0.14	0.26	0.10	0.44	0.52
S16	0.07	0.34	0.09	0.40	0.48

Factor loadings ≥ .30 are bolded. h² = communality. PA1–PA4 = principal axis factors.