Psychometric Validation of the East Africa Alcohol Expectancy Scale (AFEXS) for Youth Living with HIV in Southwestern Uganda

Abstract

Reliable and valid assessment tools are crucial for measuring alcohol use expectancies (AUE) among youth living with HIV (YLHIV). In this paper, we assess the psychometric properties of the original 17-item East Africa Alcohol Expectancy Scale (AFEXS) among Ugandan YLHIV—a scale developed and validated by Woolf-King et al. for use among Uganda adults living with HIV. Specifically, we applied the measure to a sample of 100 Ugandan YLHIV enrolled in a substance use intervention. To analyze the psychometric properties, we conducted exploratory factor analysis (EFA), confirmatory factor analysis (CFA), and convergent validity analyses. The resulting 15-item youth AFEXS scale differed from the 11-item adult version. The 15-item Youth AFEXS demonstrated high internal consistency (α = 0.85), aligning with theoretical expectations of a three-factor structure related to sexual expectancies (7 items; α = 0.89), negative expectancies (4 items; α = 0.72), and release of inhibition expectancies (4 items; α = 0.77). The global scale and sub-scales demonstrated satisfactory convergent validity through positive correlations measures of frequency of alcohol consumption in the past 12 months: global AFEXS (r = 0.617; p = 0.000), sexual expectancies (r = 0.583; p = 0.000), negative expectancies (r = 0.618; p = 0.000), and release expectancies (r = 0.596; p = 0.000). The study confirms the youth AFEXS as a reliable and valid tool for assessing AUE among Ugandan YLHIV.

Introduction

The psychometric propeties of measures used to assess alcohol use expectancies (AUE) play a critical role in ensuring the validity and reliability of research findings. AUE, defined as the anticipated positive and negative effects of alcohol consumption, play a significant role in shaping alcohol use (frequency and quantity) [1–5] and alcohol-related risky behaviors [6–8]. Positive expectancies can increase the likelihood of problem drinking in anticipation of positive expected effects such as social facilitation, coping, sexual enhancement, behavioral disinhibition, and conformity, while negative expectancies can reduce the likelihood of problem drinking in anticipation of negative expected effects such as negative affective change, loss of control, and health effects [9–13].

In Uganda, alcohol consumption among youth is a major public health concern due to its associations with HIV [6, 7, 14], high-risk sexual behaviors [15, 16], and poor mental health [17, 18] among others. Alcohol use among youth living with HIV (YLHIV) is particularly concerning as it greatly exacerbates HIV treatment challenges, including non-adherence to antiretroviral therapy [19–22], retention in HIV care [23] and achieving viral suppression [21, 24] thereby increasing the risk of HIV transmission and AIDS-related mortality [25–27]. Despite growing recognition of these issues, there remains a critical gap in targeted interventions that address the unique challenges faced by this population, including culturally relevant prevention and treatment strategies that are specifically tailored to the Ugandan context.

The assessment of AUE has progressed significantly with the development of the East Africa Alcohol Expectancy Scale (AFEXS), a measure that demonstrated validity and reliability in evaluating AUE among adults living with HIV in Uganda [28]. However, challenges persist in the widespread application of psychometric scales, such as AFEXS, in diverse settings such as Uganda due to variations in literacy levels, language barriers, and cultural differences across regions, which could affect the accuracy and reliability of psychometric assessment tools. Moreover, as AUE could vary by age and sex [29–31], it is important to assess the structural variability of these measures across these socio-demographic characteristics. Thus, additional research is needed to validate the AFEXS among Uganda YLHIV, to ensure that the scale adequately captures the full construct of alcohol use expectancies in this population. This paper aims to advance our understanding of the psychometric properties of AUE among YLHIV in southwestern Uganda, with the goal of validating this measure for use in alcohol related research and interventions targeting this population.

Methods

Study Design and Setting

We utilized data from a cross-sectional study designed to examine the prevalence and consequences of alcohol and drug use (ADU) and to identify the multilevel risk and resilience factors associated with substance use among YLHIV (n = 200) in Southwestern Uganda [32]. The study was conducted in six HIV clinics that were randomly selected from fishing communities in the Greater Masaka region of Southwestern Uganda. This area is particularly impacted by HIV, with a prevalence rate of 8.1%, which is higher than the national average of 5.8% [33]. These communities also have significant levels of poverty and alcohol consumption [34, 35]. The selected clinics, located in regions known for high HIV rates, are recognized by the Uganda Ministry of Health as authorized providers of antiretroviral therapy and offer established services for adolescents and youth living with HIV.

Eligibility criteria for YLHIV to participate in the study included being male or female aged between 18 and 24 years, having a medical diagnosis of HIV with awareness of their status, and being actively enrolled in care at one of these clinics. Our analyses focus on a subsample of 100 YLHIV who tested positive (through self-report or positive urine test) for ADU and were randomized to receive the intervention. Due to missing data, only 84 participants were included in the final analyses.

Ethical approval for the study was obtained from the institutional ethics and review boards at Washington University in St. Louis (202301145), the University of Michigan (HUM00223732), and from the in-country local institutional review boards in Uganda: the Uganda Virus Research Institute (GC/127/933), and the Uganda National Council of Science and Technology (HS2683ES). A certificate of confidentiality was acquired to safeguard participants’ data, which prevents the sharing of identifiable or sensitive participant information with those not associated with the research, unless explicitly authorized by the participant. All YLHIV involved in the study provided written informed consent before participating in study activities.

Measures

Outcome: Alcohol Use Expectancies (AUE)

The goal of the East Africa Alcohol Expectancy Scale (AFEXS) [28] is to provide a reliable and culturally relevant measure for assessing alcohol expectancies among HIV-positive adults in East Africa, particularly in Uganda. Woolf-King et al. argued that previous tools, largely developed in the Global north, were inadequate for capturing the complex dynamics of alcohol use within this population, lacking both psychometric validation and a focus on negative expectancies associated with alcohol use among those living with HIV/AIDS. Woolf and colleagues [28] drew on the prior works [6, 7, 36] to draft a 17-item draft of the AFEXS (Table 1), which included both positive and negative AUE related to sexual enhancement, social facilitation, release of inhibitions, tension reduction, and emotional and physical consequences of alcohol use. The scale was evaluated in two steps: exploratory factor analysis (EFA) was first conducted with 209 HIV-positive Ugandan adults from the BREATH cohort study, leading to a 13-item, three-factor scale (i.e., sex related, release of inhibition and negative expectancies) that explained 67% of the variance and had strong internal consistency (alpha (α = 0.89). Subsequently, confirmatory factor analysis (CFA) with a second demographically similar sample of 181 participants from the Urban ARCH Uganda cohort confirmed the scale’s three-factor structure and refined the scale to an 11-item measure with excellent internal consistency for the overall scale (α = 0.88) and across the three domains: sex-related (α = 0.94), release of inhibition (α = 0.61) and negative expectancies (α = 0.73). In this paper, we assess the psychometric properties of the original 17-item AFEXS Scale, as developed by Woolf-King et al. Specifically, we apply the measure to a sample of 100 YLHIV enrolled in an ADU reduction intervention [32].

Table 1.

Original 17-item East Africa expectancy scale

	Sexual expectancies	Negative expectancies	Release expectancies
1. It’s easier for me to talk to people			X
2. I get mean			X
3. I am more likely to do things I normally wouldn’t do			X
4. I feel sad		X
5. I feel sick the next day		X
6. I feel sick (at the time I’m drinking)		X
7. I become more extroverted			X
8. I feel ashamed of myself		X
9. I am able to take my mind off my problems			X
10. I become more aggressive			X
11. I feel closer to a sexual partner	X
12. I am less nervous about sex	X
13. I find it harder to say no to sexual advances	X
14. I am a better lover	X
15. I am more sexually responsive	X
16. I am less likely to use a condom	X
17. I enjoy sex more than usual	X

Correlates

The following correlates were selected based on prior studies assessing correlates of alcohol use. Self-reported alcohol use in the past 12 months and past 3 months were assessed using the NIDA-Modified Alcohol Smoking and Substance Involvement Screening Test (NM-ASSIST) (National Institute on Drug Abuse). Depressive symptoms were assessed using the Center for Epidemiologic Studies Depression Scale (CES-D), a 20-item item assessment of depressive symptomology in the past 2 weeks [37]. Positive items were reverse coded where necessary and items ranged from 0 to 3, with higher scores representing more depressive symptoms (α = 0.83). Loneliness was measured using the UCLA loneliness scale [38], an 18-item scale that measured participants’ subjective feelings of loneliness and social isolation. Higher scores were indicative of higher feelings of loneliness (α = 0.79). Perceived stress was evaluated using the Perceived Stress Scale, which assesses how often participants perceive ten specific situations as stressful over the past month [39]. Positive responses were reverse scored, meaning higher scores on the scale reflected higher levels of perceived stress (α = 0.66).

Data Analysis

Descriptive Analyses

Analyses were conducted using STATA version 18.0. Univariate analyses were conducted for all scale items and correlates included in these analyses. Categorical variables were presented as counts and percentages, while continuous variables were described using median and interquartile range (IQR). We assessed the degree of missing data and suitability of the data for factor analysis was evaluated using the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity of the correlation matrix. The KMO assesses sampling adequacy by comparing the magnitude of the observed correlation coefficients in relation to the magnitude of the partial correlation coefficient: the recommended cut off for sampling adequacy is 0.6. Bartlett’s test of sphericity of the correlation matrix is used to test the hypothesis that the correlation matrix is an identity matrix (all the diagonal terms are one and all the off-diagonal terms are zero); a p value less than 0.05 is required to reject the null hypothesis. Lastly, we inspected the correlation matrix for presence of coefficients greater than 0.4, which may have indicated potential collinearity among scale items.

Exploratory Factor Analysis (EFA)

Following best practice guidelines for EFA [40, 41], we utilized principal axis factoring and oblique rotation to identify the underlying factors (subscales) due to skewed distribution of the survey responses on the scale items. The oblique rotation technique was then applied to accommodate the intercorrelation between scale items and underlying psychological factors. The decision on the number of factors to retain for rotation was guided by a scree plot (Fig. 1), based on eigenvalues exceeding 1.00 and a minimum item loading score was 0.4. Post-rotation, we examined the factor loading to ensure all items loading were greater than 0.4 and checked for any significant cross-loadings greater than 75%. Two scale items (Q9. ‘I am able to take my mind off my problems’, and Q10. ‘I become more aggressive’) had factor loadings less than 0.4 so they were removed from the analyses. EFA was repeated with the remaining 15 items. All the scale items had factor loadings greater than 0.4 and none of the items had any significant cross loadings. We also assessed the proportion of variance explained by the retained factors to ensure that it was greater than 50%. Item-total correlations of ≥0.3 imply that the items discriminate between different constructs, while inter-item correlations > 0.7 indicate that the items measure almost the same and one of them might be deleted [42–44].

Fig. 1.

Scree plot of Eigen values

Confirmatory Factor Analysis (CFA)

CFA was conducted using the 17-item, three-factor model derived from the EFA. Utilizing Hu & Bentler’s criteria, we evaluated the model’s adequacy using several fit indices: chi-square test statistic aiming for non-significance, Root Mean Square Error of Approximation (RMSEA) (recommended cut-off of < 0.06), and comparative fit index (CFI: recommended cut-off > 0.95) [45].

Scale Validity Testing

We assessed convergent validity of the scale by correlating the total scale score and each factor score from the AFEXS with the total NIDA ASSIST score, frequency of drug use in the past 12 months and past 3 months, depression, loneliness and perceived stress, using Spearman’s rank correlation coefficients. Further, we computed subscale scores (sum of sub-scale items) and a composite scale score (sum of all scale items), and conducted correlations between subscales, and between each sub-scale and composite score. Lastly, EFA with principal axis factoring and oblique rotation using the subscale scores as items was conducted to assess the viability of the composite of scale score.

Results

Descriptive Statistics

A total of 84 participants were included in the final analysis. As shown in Table 2, the majority were female (71.4%; n = 60), with an average age of 22.1 years (SD = 2.3). Only 3.6% were currently enrolled in school and the majority had more than a primary school education (70.2%, n = 59), were married or co-habiting (53.6%; n = 55) and were currently employed (52.4%; n = 44). The inter-item correlations of scale items ranged from 0.014 to 0.745, with a mean inter-item correlation of r = 0.265. The KMO measure of sampling adequacy was 0.78 and Bartlett’s test of sphericity of the correlation matrix was significant (χ² = 537.77 (df = 136), p = 0.000), indicating the sample size was adequate for factor analysis.

Table 2.

Participant socio-demographic characteristics

	Freq.	Percent	Cumulative frequency
Age
18	8	9.52	9.52
19	5	5.95	15.48
20	8	9.52	25
21	9	10.71	35.71
22	8	9.52	45.24
23	18	21.43	66.67
24	27	32.14	98.81
32	1	1.19	100
Gender
Female	60	71.43	71.43
Male	24	28.57	100
Currently in school
No	81	96.43	96.43
Yes	3	3.57	100
Highest level of education attainment
Primary	25	29.76	29.76
Lower secondary (S1-S4)	22	26.19	55.95
Upper secondary (S5-S6)	23	27.38	83.33
Some college/certificate	3	3.57	86.9
Vocational/trade school	11	13.1	100
Marital status
Single	25	29.76	29.76
Married	22	26.19	55.95
Cohabiting	23	27.38	83.33
Divorced	3	3.57	86.9
Separated	11	13.1	100
Orphanhood
Not orphaned	40	47.62	47.62
Single orphaned	34	40.48	88.1
Double orphaned	10	11.9	100

Exploratory Factor Analysis

The first EFA without forcing any solution, revealed 3 factors based on eigenvalues greater than 1.0 and examination of the scree plot (Fig. 1). The EFA was re-run with a three-factor solution and oblique rotation (“Appendix”).

None of the scale items had any significant overlap across the factors but two scale items (9 and 10) had factor loadings less than 0.4, so they were removed. Another EFA with the remaining 15 items did not reveal any significant cross-item loadings and all items had factor loadings greater than 0.4 (Table 3). This 15-item, three factor solution explained 93.9% of the variance and excellent internal reliability (α = 0.85). The three factors (sub-scales) explained 58.7% (sexual expectancies), 21.3% (negative expectancies), and 13.9% (release of inhibition expectancies) of the total variance. Inter-item correlations of the 15-item AFEXS ranged from 0.252 to 0.283 with a mean inter-item correlation of r = 0.268. The average and range inter-item correlations of the subscales were as follows: sexual expectancies, r = 0.558 (0.525–0.594); negative expectancies, r = 0.389 (0.297–0.458); and release expectancies, r = 0.369 (0.318–0.394), all within normal range.

Table 3.

Results of exploratory Factor analysis on the AFEXS for youth living with HIV in Uganda

Scale item	Factor 1	Factor 2	Factor 3	Uniqueness
1. Easier for me to talk to people		0.6267		0.6121
2. I get mean		0.7056		0.5081
3. More likely to do things		0.5079		0.6875
4. I become extroverted		0.5346		0.5893
5. I feel sad			0.4329	0.7498
6. I feel sick the next day			0.8371	0.3037
7. I feel sick at time of drinking			0.7877	0.3453
8. I feel ashamed of myself			0.4349	0.7442
11. I feel closer to my sexual partner	0.5863			0.6328
12. I am less nervous about sex	0.7927			0.3824
13. I find it harder to say no to sexual advances	0.7027			0.4822
14. I am a better lover	0.7537			0.3891
15. I am more sexually responsive	0.8153			0.3788
16. I am less likely to use a condom	0.5384			0.5082
17. I enjoy sex more than usual	0.8475			0.2688

Confirmatory Factor Analysis

The CFA was first conducted on the 15-item AFEXS resulting from the EFA (Table 4). The resulting model fit was moderate: χ²(62) = 128.93, p = 0.002, CFI = 0.91, TLI = 0.89, RMSEA = 0.80, 95% CI [0.05–0.11], p = 0.058. Item 4 (0.419, p = 0.000) and item 8 (0.336, p = 0.000) had fairly low but statistically significant loadings on the negative expectancies scales and were left in the analyses because the model could not converge without them. All other items had moderate to high and statistically significant loadings, ranging from 0.557 to 0.955. As shown in Table 5, the overall scale (15 items; α = 0.85) and each sub-scale had good to excellent internal reliability: sexual expectancies (7 items; α = 0.89), release of inhibition expectancies (4 items; α = 0.77), and negative expectancies (4 items; α = 0.72). The global AFEXS and its sub-scales correlated as follows: sexual expectancies (r = 0.965; p = 0.000), negative expectancies (r = 0.920; p = 0.000), and release of inhibition expectancies (r = 0.925; p = 0.000). EFA with principal axis factoring and oblique rotation using the subscale scores as items confirmed the one factor structure explained 91.9% of the variance, thereby validating the viability of the global AFEXS score.

Table 4.

Results of confirmatory factor analysis on the AFEXS for youth living with HIV in Uganda

Scale item	Factor 1	Factor 2	Factor 3
Release of inhibition
1. Easier for me to talk to people		0.589
2. I get mean		0.693
3. I am more likely to do things I normally wouldn’t do		0.565
7. I become more extroverted		0.659
Negative expectancies
4. I feel sad			0.419
5. I feel sick the next day			0.955
6. I feel sick at time of drinking			0.780
8. I feel ashamed of myself			0.336
Sexual expectancies
11. I feel closer to my sexual partner	0.557
12. I am less nervous about sex	0.755
13. I find it harder to say no to sexual advances	0.692
14. I am a better lover	0.801
15. I am more sexually responsive	0.778
16. I am less likely to use a condom	0.644
17. I enjoy sex more than usual	0.870

Table 5.

Descriptive summary of the AFEXS and sub-scales

Variable	Number of items	Mean	SD	Min	Max	Reliability
Global alcohol use expectancies	15	18.06	23.72066	0	74	0.85
Sexual expectancies	7	8.91	12.53786	0	42	0.89
Negative expectancies	4	5.25	7.201165	0	22	0.72
Release of inhibition expectancies	4	5.415	7.384812	0	24	0.77

Scale Validity Testing

As anticipated (Table 6), results of the convergent validity analyses showed positive correlations between the global AFEXS Scale and sub-scale with frequency of alcohol consumption in the past 12 months: global AFEXS (r = 0.618; p = 0.000), sexual expectancies (r = 0.583; p = 0.000), negative expectancies (r = 0.618; p = 0.000), and release expectancies (r = 0.601; p = 0.000). Similar associations were also found with frequency of alcohol consumption in the past 3 months: global AFEXS (r = 0. 569; p = 0.000), sexual expectancies (r = 0.543; p = 0.000), negative expectancies (r = 0.566; p = 0.000), and release expectancies (r = 0.538; p = 0.000). The global scale and sub-scale scores were significantly associated with depressive symptoms: global AFEXS (r = 0.322; p = 0.000), sexual expectancies (r = 0.290; p = 0.000), negative expectancies (r = 0.338; p = 0.000), and release of inhibition expectancies (r = 0.295; p = 0.000). Statistically significant correlations were found with loneliness: global AFEXS (r = 0.216; p = 0.002), sexual expectancies (r = 0.190; p = 0.007), negative expectancies (r = 0.207; p = 0.003), and release of inhibition expectancies (r = 0.192; p = 0.007). However, the associations between the perceived stress were only marginally significant, for both the global AFEXS and sub-scales.

Table 6.

Validity coefficients for the AFEXS among youth living with HIV in Uganda

	Global AFEXS score	Sexual expectancies	Negative expectancies	Release expectancies
Frequency of alcohol consumption in the past 12 months	0.618***	0.583***	0.618***	0.601***
Frequency of alcohol consumption in the past 3 months	0.569***	0.543***	0.566***	0.538***
Depression	0.322***	0.290***	0.338***	0.295***
Loneliness	0.205**	0.190**	0.207**	0.182**
Perceived stress	0.135*	0.119*	0.137*	0.113

^*p<0.10;

^**p<0.05;

^***p<0.001

Discussion

We examined the psychometric properties of the AFEXS among YLHIV in southwestern Uganda. The AFEXS was developed to provide a reliable and culturally relevant measure for assessing AUE among HIV-positive adults in Uganda [28]. Given the variations in age [29–31], literacy levels, language barriers, and cultural differences across Uganda, it is important to verify the accuracy and reliability of psychometric measures used for assessing AUE in this unique population, as these differences could impact the validity of the results. Our analysis confirmed the three-factor structure of the AFEXS—sex-related, release of inhibition and negative expectancies—identified in previous studies within different populations [28, 36]. The scale also demonstrated excellent internal consistency and reliability, affirming its utility in both clinical and research settings to understand and predict alcohol use behaviors among YLHIV.

There is a notable difference between our derived youth version and the adult AFEXS scale. The youth AFEXS contains 15 items, compared to the 11 items of the adult version. The additional items in the youth AFEXS scale include four items related to negative expectancies (e.g., “I feel sad” and “I feel ashamed of myself”) and release expectancies (i.e., “I get mean” and “It is easier for me to talk to people”), which may reflect the unique cognitive and behavioral patterns of alcohol use among this demographic. Thus, the inclusion of these items could be crucial for understanding alcohol use behaviors in this population. Using the youth AFEXS, our results demonstrated statistically significant correlations between AUE and frequency of alcohol consumption. This finding is consistent with the expectancy theory, which posits that AUE can significantly influence individuals’ drinking behaviors and has been documented in other studies [1–5]. By validating a tool that can identify specific AUE that are predictive of alcohol consumption among YLHIV, we can effectively identify and fine-tune interventions to address these cognitive precursors of alcohol use.

Limitations

Our study has several limitations that must be considered when interpreting the findings. First, our small sample size may reduce the statistical power to identify specific effects or extrapolate results to a larger population. To enhance the robustness and generalizability of our findings, future research with larger and more representative samples is needed. Second, the cross-sectional design of the study makes it impossible to establish a causal relationship between AUE and alcohol use behaviors. To determine causal links between AUE and alcohol consumption among YLHIV, longitudinal studies are necessary. Lastly, the context-specific nature of the study’s findings in Southwest Uganda may restrict its applicability to other areas or groups with distinct cultural or socioeconomic backgrounds. Further studies are needed to validate the AFEXS in diverse Ugandan communities and its suitability in other sub-Saharan African contexts, where cultural distinctions may have unique effects on alcohol consumption patterns and expectations.

Conclusions

Nonetheless, our 15-item youth AFEXS is a reliable and culturally sensitive tool that captures the nuances of AUE among YLHIV in southwestern Uganda. The scale’s utility in measuring factors associated with alcohol-related behaviors is an asset for researchers and healthcare providers aiming to mitigate the risks associated with alcohol use in this vulnerable population. By integrating the youth AFEXS into regular clinical assessments and intervention programs, we can better tailor strategies to combat the detrimental effects of alcohol use on the health and well-being of YLHIV. Leveraging the support of allied health professionals (such as counselors) and lay health providers (such as community health workers and adolescent peer supporters) already embedded within Uganda’s HIV clinical services can facilitate the routine use of the AFEXS, ensuring comprehensive and sustainable care without overburdening existing clinical staff. The scale’s robust psychometric properties ensure that it is a reliable tool for both screening and monitoring the effectiveness of interventions over time.

Appendix

See Table 7.

Table 7.

Results from initial exploratory factor analysis on AFEXS among YLHIV

Scale item	Factor 1	Factor 2	Factor 3	Uniqueness
1. Easier for me to talk to people		0.6117		0.6189
2. I get mean		0.7145		0.4934
3. More likely to do things		0.4867		0.7039
4. I feel sad			0.4505	0.7432
5. I feel sick the next day			0.8391	0.2991
6. I feel sick at time of drinking			0.770	0.3648
7. I become extroverted		0.4917		0.6319
8. I feel ashamed of myself			0.4538	0.7331
9. Able to take my mind off my problems		0.3623		0.7537
10. I become more aggressive				0.8236
11. I feel closer to my sexual partner	0.6103			0.6276
12. I am less nervous about sex	0.8188			0.3418
13. I find it harder to say no to sexual advances	0.7143			0.4769
14. I am a better lover	0.7557			0.3849
15. I am more sexually responsive	0.7977			0.3906
16. I am less likely to use a condom	0.5506			0.521
17. I enjoy sex more than usual	0.8453			0.2765