Survey and Participants
We conducted a cross-sectional, internet-based survey among English-speaking adults (aged 18 + years) living in the United States who self-identified as living with HIV. We used Amazon Mechanical Turk (MTurk) for recruitment, a commonly used social science research tool for recruiting populations of interest and collecting rigorous data, comparable to popular U.S. online survey panels [28, 29]. To facilitate high-quality responses, the survey was advertised only to U.S.-based MTurk workers who had successfully completed 1,000 previous human intelligence tasks with a 95% approval rating.
The survey took less than 20 min to complete and was administered in Qualtrics (Provo, UT). Participants provided informed consent by clicking a box stating that they agreed to participate in the survey. They then completed a captcha (as a bot control step) to confirm they were human and were given several screening questions to (a) check for English comprehension and attention and (b) verify HIV seropositive status. English comprehension and attention questions included describing a displayed picture, spelling a word backwards, and selecting a multiple-choice response when directed. Five questions were used to verify HIV seropositive status, including: self-reported HIV status, whether the respondent takes pre-exposure prophylaxis (PrEP; since people diagnosed with HIV would not be prescribed PrEP), HIV medications taken, tests taken to confirm HIV diagnosis, and how HIV is transmitted. Upon survey completion, respondents received a code to submit to MTurk and claim $7 compensation for their time. Given online data collection can be vulnerable to fraud, we carefully reviewed the GeoIP addresses of respondents who completed the survey and rejected those who had screened out before making repeated attempts to qualify [30, 31].
The University of California, Riverside IRB reviewed and approved the study on December 10, 2020 (protocol #HS-20-248). All respondents were recruited, provided informed consent, and completed the survey between March 29 and April 28, 2021.
Measures
Attitudes Toward Payment in HIV Research
The survey assessed respondents’ attitudes toward payment in HIV research using the following questions: (1) “Do you consider payment to be a benefit of participating in research?”, (2) “Would payment play a role in your decision to participate in HIV research?”, (3) “Should people receive payment to participate in HIV research?”, (4) “Should there be any standards or policies on participant payment in HIV research?”, and (5) “If an HIV study did not pay you, would you expect to receive another benefit from participation?”. We constructed dummy (no/yes) variables for each of these questions.
Willingness to Participate in HIV Research Without Payment
A categorical variable (coded no [the reference category]/yes/it depends) measured whether respondents stated they would participate in HIV research without any payment.
Risk Perceptions
We used a dummy (no/yes) variable to assess whether respondents could imagine a certain risk level from participating in an HIV intervention study in which no amount of payment could convince them to participate. Two continuous variables (ranging from 0 to 100) assessed the (a) percent chance of harm and (b) percent chance of death that would deter respondents from participating in HIV research. The level of perceived risk associated with different study payment amounts was assessed with the question: “If one early-phase HIV intervention study pays you a total of $20,000, and another early-phase HIV intervention study does not pay you, how would you rate the risk of each study?” Respondents indicated the level of risk they rated each study on a scale of 0 (“no risk”) to 10 (“highest risk”).
Importance of Various Forms of Payment
Respondents rated, on a scale of 0 (“least important”) to 10 (“most important”), how important each of the following types of payment for research participation are: (a) a cash incentive, (b) reimbursement for lost wages, (c) compensation for time, (d) transportation voucher, (e) food, (f) gifts, and (g) post-trial access to the intervention if proven effective.
Sociodemographic Variables
We used binary variables to measure respondents’ gender (man vs. woman), education (no bachelor’s degree vs. bachelor’s degree or higher), and employment status (works full time vs. does not work full time). A categorical variable assessed respondents’ census region of residence: West, Midwest, Northeast, or South (the reference category). Age was measured continuously, in years. Self-rated health was assessed using a visual analogue scale that asked respondents to indicate “how good or bad your own health state is today,” measured continuously from 0 (“worst health you can imagine”) to 100 (“best health you can imagine”).
Race and ethnicity were measured categorically with a single variable, constructed from two questions asking whether the respondent self-identified as (a) Hispanic or Latina/o/x and (b) belonging to one or more of the following groups: White; Black or African American; Asian; Native Hawaiian or other Pacific Islander; American Indian or Alaska Native; or other group. To achieve sufficient power for analyses, race and ethnicity were recoded into a three-category variable as (1) non-Hispanic White (the reference category), (2) non-Hispanic Black, and (3) Hispanic or other/multiple races and ethnicities (hereafter referred to as “Hispanic/other”).
Statistical Analyses
We calculated descriptive statistics (i.e., frequencies, percentages, means, standard deviations, and ranges) for all variables. We used a series of multivariate logistic regression models to test for significant associations between all sociodemographic variables and respondents’ (a) attitudes toward payment in HIV research and (b) ability to imagine a level of risk that would deter research participation. A single multinomial logistic regression model assessed sociodemographic differences in willingness to participate in HIV research without payment. A series of multivariate linear regression models tested for significant associations between all sociodemographic characteristics, and: (a) percent harm or death that would deter HIV research participation, (b) the perceived risk attributed to studies paying $0 versus $20,000, and (c) the importance of various forms of payment.
We conducted all analyses using Stata SE v. 17 (College Station, TX). We report odds ratios (ORs) for logistic regression models, relative risks (RRs) for multinomial logistic regression models, and unstandardized regression coefficients (bs) for linear models. For all models, we report p values < 0.05 as statistically significant.