We conducted a preregistered vignette experiment (or factorial survey experiment; [30]) in which we randomly varied 3 factors based on theoretical considerations. Prior to the data collection process, we estimated for the experiment a minimum sample size of 1280 respondents. Sample size calculations were based on an approximated power analysis using an ANOVA design with repeated measures and within-between interaction, using the software G*Power [31] (input parameters: effect size=0.1, alpha error probability=0.05, power=0.95, number of groups=64, number of measurements=2, nonsphericity correction=1). To meet this minimum sample size requirement for every country, we aimed to achieve 1500 completed interviews per country. The three factors (Table 1) included in the experiment were as follows: (1) topic (HIV, diabetes), (2) financial incentive (none, US $2 cash, US $2 lottery, and US $2 voucher), and (3) sponsor (nongovernmental organization [NGO], statistical office, health ministry, or local university). This resulted in 32 unique vignettes (2×4×4). Individuals were randomized into 1 out of the 32 vignette scenarios. Randomization was done separately for each of the 2 tasks (in-person survey participation vs participation in blood test).

In the following, we provide an example vignette for the blood test task (parameters: HIV, US $2 cash, statistical office):

This study was approved by the Medical Chamber of the state of Hamburg (#2022‐300276-WF) and the German Institute for Global and Area Studies (#5/2023). All participants who took part in the study consented to the primary and secondary analysis of research data and to providing public access to the dataset. The information was collected using UniPark, stored inside a local, secure server in Germany, and was encrypted at all times. All data collected were anonymized and no identifiable information was collected (names, addresses, contact information, and images). All respondents were offered the chance to enter a draw to win 1 of 12 phone vouchers (approximately US $12 per voucher).

Table 3 shows descriptive results for each vignette. Overall, we observe a high willingness to participate in both the survey and the blood test. The average willingness to participate ranged from 5.54 (survey on HIV, conducted by the National Statistical Office, no incentive offered) to 6.09 (survey on HIV, conducted by the Ministry of Health, incentive to have a 1% chance to win US $200 in a lottery). In total, 6 vignettes showed values of 6.00 or higher in terms of the willingness to participate in either survey or blood test. Each of those 6 vignettes included financial incentives (either as a lottery or cash incentive).

Turning toward the multivariate OLS analysis, Figure 1 depicts results that relate to the 2 tasks (survey vs blood test) and all 3 factors. In the following, we discuss results from model specifications without any interaction effects (Model 1) that are related to hypotheses 1 to 5 from the preanalysis plan (Table S4 [contains the underlying regression] in Multimedia Appendix 1). In contrast to our initial expectations, we do not find any evidence that individuals’ willingness differs across task types (Hypothesis #1; survey vs blood test), the topic of the task (Hypothesis #2; diabetes vs HIV), and the sponsor (Hypothesis #5; local university vs Ministry of health vs statistical office vs NGO). A minor exception concerns vignettes related to the Ministry of Health as a sponsor of the survey or blood test. Having the Ministry of Health as a sponsor increases the probability of participating in surveys or blood tests by about 6.3 percentage points versus the reference category (local universities). However, having the Ministry of Health as a sponsor is not advantageous over the 2 other categories (NGOs, National Statistical Office). Further, we find strong evidence in support of the view that financial incentives matter for participation in surveys and blood donation tasks. On average, including financial incentives increases the willingness to participate in surveys or blood tests by about 22 to 26 percentage points. In particular, the findings suggest that participants would be more willing to participate in health-related in-person data collection activities if they are financially rewarded which is in line with Hypothesis #3 and the economic participation theory. Yet, we do not observe statistically significant differences between the 3 financial incentive groups (no support of Hypothesis #4; cash vs lottery vs voucher incentive).

We conducted a number of robustness checks to test for the sensitivity of our main results (Tables S5 to S10 in Multimedia Appendix 1). By and large, we find that our results do not change once (1) using ordered logit regressions compared with OLS (Table S5 in Multimedia Appendix 1), (2) re-estimating our main specification using alternative covariate specifications (Table S6 in Multimedia Appendix 1), (3) using a lasso regression and double machine learning specification (Table S7 in Multimedia Appendix 1), (4) adopting a multilevel random effects model (Table S8 in Multimedia Appendix 1), (5) dropping observations in which individuals fail any of the two included survey attention checks (Table S9 in Multimedia Appendix 1), and (6) dropping observations for which the outcome variable was based on the simplified 5-point rating scale (Table S10 in Multimedia Appendix 1). Additionally, we conducted a further analysis to assess the robustness of our results, in which we found that the results were similar considering individual countries when comparing them to the pooled results presented in the main text (Figures S2 to S4 in Multimedia Appendix 1).

Next, we examine interaction effects (Model 2 in Figure 1) that relate to hypotheses #6 to #8. In contrast to Hypothesis #7, we find that individuals with elevated privacy concerns seem to be more willing to participate in health-related data collection efforts (b=.034). Moreover, and related to Hypothesis #8, we observe that opposite to our hypothesis, individuals with their own health concerns (being HIV positive, being diabetic) are substantially more likely to state their willingness to participate in health-related data collection efforts (b=.296 and b=.135). Though not preregistered, we believe that this finding can be reconciled with both the economic participation and the social exchange theory. First, individuals who are diagnosed with severe medical conditions (HIV or diabetes) are likely to routinely follow medical check-ups and perceive participation in health surveys and health data collection activities as beneficial for themselves (and others). Second, if they are regularly in need of medical care by nurses or medical doctors, they might be inclined to positively reciprocate by participating in health research.

Last, and in line with Hypothesis #6, trust relationships appear to play an important role, albeit in a very context-dependent manner. For 4 trust indicators (NGOs, Ministry of Health, trust in science, social trust), we observe a clear positive relationship between higher levels of trust and the willingness to participate in surveys or blood tests (Model 2 in Figure 1). In contrast, for 2 trust indicators (National Statistical Office, local universities), we find that these trust indicators are not related to a higher willingness to participate in surveys or blood tests. To better understand the role of trust in our context, we present in Figure 2 results from additional regression specifications that include an interaction effect between trust and incentives. These interaction effects help explore whether these factors are substitutes or complements (eg, re-enforcing each other). Once controlling for the interaction between incentives and trust, we find that higher levels of trust are positively related to the willingness to participate in surveys across all 6 trust measures. In line with this finding, we observe for the 2 institutions for which the trust variable was statistically nonsignificant without adding the interaction (Figure 1), a crowding-out pattern between financial incentives and trust. The related coefficient on the interaction effect is always negative and statistically significant at the 5 percent (local universities) or 10 percent level (National Statistical Office). In general, however, we do not find strong evidence for a crowding-out or crowding-in among trust and incentives. Specifically, the figure suggests that incentives mostly cannot replace trust in the sponsor and vice versa.

Our analysis concluded by examining determinants of the willingness to participate in health surveys and blood tests that are not directly related to factors varied in the vignette. Specifically, we explore in Figure 3 (Table S12 in the Multimedia Appendix 1) the role of demographic and socioeconomic factors and other background variables. Regarding demographics, we found a positive and statistically significant effect of having children on the willingness to participate in health-related, in-person data collection efforts (b=.229), whereas we did not find significant effects related to age, education, gender, marital status, and wealth of the study subject. We speculate that the positive and statistically significant coefficient of having children might be reflective of economic and social exchange motives. Bearing in mind that our sample population is relatively young and healthy, many individuals might have their first important contact with nurses and medical doctors when they become parents, which, ceteris paribus, might positively affect individuals’ perceptions of their own benefit from health care and reliable information for health-policy decision-making (economic exchange theory). Likewise, exposure to nurses and doctors at the time of birth might positively influence trust in these actors and lead to a higher willingness to participate in health research (social exchange theory). With respect to the other background variables, we show that the willingness to take risks, generally better health, possessing higher cognitive skills, and positive overall survey enjoyment and value increased the willingness to participate.

This article presents results from a preregistered vignette experiment on health research participation within a web survey implemented in 6 African countries. The design was developed drawing from the theory of contextual integrity [24], the economic participation theory [25-27], and the social exchange theory [28,29]. Regarding the 8 hypotheses, Hypothesis 3, which suggested a higher willingness when an incentive was promised, and Hypothesis 6, which suggested a positive effect of trust in the respective sponsor, were supported. In contrast, we did not find the expected effects regarding H1 (task), H2 (topic), H4 (difference across incentives), H5 (difference across sponsors), H7 (privacy concerns), and H8 (HIV or diabetes diagnosis). Notably, for H7 and H8, we found effects opposite to the expected directions, as respondents who had more privacy concerns or were diagnosed with the respective diseases were more likely to participate, possibly due to their personal involvement due to the diagnosis and because privacy concerns might be less relevant for participation than other reasons [32].

Because of the positive effects of the incentives and levels of trust regarding study participation, we tested whether those effects are independent or whether they multiply with one another. Results showed no interaction but independent effects, suggesting that the study’s extrinsic (incentives) and intrinsic motives (trust levels) individually impact survey participation. Notably, not only trust in the respective research organization but also the levels of social trust and trust in science affected willingness to participate.

Of the other variables, having children, cognitive skills, risk-taking, general health, and positive survey attitudes affected the willingness to participate, whereas sociodemographic variables did not show a statistically significant effect on participation levels. Those results suggest that mainly attitudinal and topic-related (ie, health) impact the participation likelihood, while demographic variables seem to be less relevant. Similar to the trust variables, respondents with positive survey attitudes and positive associations with respect to the survey topic are more likely to participate. Specifically, the positive effect of general health is notable, as we found also that respondents diagnosed with HIV or diabetes were more likely to participate. While this seems perhaps to be contradictory, both effects may suggest that an individual, who perceives a survey as more relevant, is more likely to participate, either because the survey is exactly about the disease the person has or because someone with good health is more interested in answering questions about this topic. In contrast, not being diagnosed with HIV or diabetes does not mean that the person is necessarily in good general health.

When comparing our main results to previous work from Western countries, our study confirms that monetary incentives are a strong driver for participation in surveys [33-35] and other data collections [21,36,37]. In contrast to most previous work [33,34,38], we found that a lottery incentive increased the participation likelihood. With respect to topic interest, our study adds to prior experimental evidence about topic interest [39,40] and the survey sponsor [41], showing mixed results regarding their direct effects.

When comparing our findings with respect to interaction effects reported in prior research, we replicated the effects of trust in the survey sponsor [22,42], but did not find the previously identified negative effect of higher privacy concern on participation rates [22,43-45]. With respect to participants’ own health concerns, our finding of higher participation willingness is in line with a recent experimental study which showed that individuals diagnosed with more medical conditions were more likely to share their health data [46]. This finding could mean that individuals with higher health concerns care more about the topic and are therefore more likely to engage in related research. While our survey did not show previously identified demographic effects, such as higher participation of women [47-49] and individuals with higher levels of education [37,50,51], we replicated higher participation rates of individuals with more positive attitudes toward surveys [21,52].

From a theoretical perspective, the study’s results underline the relevance of all 3 theories. The economic participation theory was supported, as incentives increased the likelihood of participation. The social exchange theory was supported because trust levels were significantly related to the willingness to participate. While our hypotheses were not supported, the theory of contextual integrity was relevant as both privacy concerns and being diagnosed with HIV or diabetes affected respondents’ willingness to participate. Those results align with the perspective that survey participation is a complex social activity of which the economic aspect is a part, but it goes beyond that by constituting a social situation in which trust and other context-specific parameters affect whether an individual is willing to participate.

From a practical point of view, the study showed that several results from other countries are relevant to the African context. As previous research suggested, all 3 theories and their related constructs explain participation willingness. However, there also seem to be specifics regarding the African context. We did not find the expected variations across different incentives and between different sponsors. While this result could have been due to the hypothetical situation, it could also hint at the specifics of the region, in which requests to participate in research are less common than in Western countries so that other incentives (eg, lottery and voucher) are similarly successful as direct cash incentives. More generally, the study shows that individuals from various demographic groups are willing to participate via web-based environments, making this way of collecting data a real alternative to more traditional methods. However, the effects of trust levels, attitudes, and topical variables illustrate, at the same time, that there are certain groups of respondents that are less likely to participate. Including a sufficiently large number of respondents from those groups will be a crucial task for study designers who would like to use web-based environments for health research.

While it is not our primary research question, the study design and particularly the test of H8 allows us to draw initial conclusions on the intent to participate of respondents diagnosed with HIV or diabetes compared with respondents who are not diagnosed with those diseases. For both outcome variables, the survey participation and the blood test, respondents who were diagnosed with at least one of the diseases reported a higher participation willingness than respondents who were not positively diagnosed (Model 1 in Figure 1 and Figure S4 in Multimedia Appendix 1). These results are promising for research that specifically focuses on those health issues since the relevant groups did not indicate high hesitancy to take part, which will be crucial for generalizable conclusions on those vulnerable groups. Further research with a specific focus on those groups is needed to develop an optimal recruitment design to conduct health research among individuals diagnosed with the respective diseases via web-based platforms and to compare those results to alternative data collection strategies and other available data sources.

We deviated from the preregistration plan regarding the number of participants per country (ie, 1500). This was done due to an unexpectedly high variation in the number of participants per US $1 spent. During the first 3 weeks of data collection, Kenya outperformed all other countries, while Nigeria and Uganda also reached relatively high participation numbers. In contrast, the data collection was less successful in Ghana, South Africa, and Tanzania in which we would not have reached the 1500 respondents target even if we had continued as originally planned. Therefore, after 3 weeks, we focused the remaining budget on countries that were more cost-effective, regarding the number of participants. While we do not know exactly why Kenya performed so well, this seems to be a common finding, as other research conducted via Facebook advertisement in sub-Saharan Africa also focuses on this country [53-55].

A second deviation from the preregistration plan concerns our experimental design and related analyses. To simplify the experimental design, we decided against randomization by task after the preregistration, leading to a 2×4×4 design instead of a 2×2×4×4 design of the vignette experiment. In line with that adjustment and because the results of separate models for each of the 2 tasks are similar to each other and the pooled analyses (Figure 1 and Figure S5 in Multimedia Appendix 1), we decided to report the pooled results in the manuscript. However, a notable difference between the results of the 2 tasks is that the effect sizes for financial incentives on participation in blood tests are about half as large (and occasionally lose significance) compared with the effects on participation in a survey, which are always highly significant (Figure S5 in Multimedia Appendix 1).

Our study has several limitations and opportunities for further research. First, we only included 6 African countries, and future research has to show whether the results apply to other countries as well. Second, recruitment strategies via social media sites have their caveats. For example, related research [56] shows that the respondents of our vignette experiment are comparable to the general population with respect to a number of sociodemographics such as respondent’s gender, age, and region of living. However, this research also shows that respondents are more likely to be single (not married), have higher daily internet use, and possess higher levels of education compared with the general population. Those results should especially be taken into consideration regarding the generalizability of our findings with respect to the association between the willingness to participate and sociodemographic variables (see Figure 3). It was also found that bivariate and multivariate estimates, which we focus on in this study, were overall less biased compared with the general population than univariate estimates [56]. To what extent a more representative pool of respondents can be generated by optimizing the ads, especially the images that are shown to social media users, remains to be seen and appears to be a fruitful avenue for future research. Third, and regarding the experiment, we have to note that we only asked for hypothetical willingness to participate. While a large body of prior research (eg, [57-59]) has documented that intended behavior and actual behavior are strongly connected, it would be worthwhile to repeat the study using a setting in which individuals are recruited via the web for an actual health survey that involves the real-world collection of biomarkers. Fourth, our research presents only a first step toward a better understanding of how web-based recruitment for in-person health surveys can be optimally designed in sub-Saharan Africa. We believe that important follow-up research could, among others, further explore whether specific target groups (eg, people with HIV infections) can be effectively recruited via social media outreach activities. Previous research has shown that Facebook surveys might be feasible to recruit special populations in Western countries, for example, HIV-positive individuals [60], kidney transplant recipients [61], or cannabis users [62,63], but further research is needed particularly for special populations in other regions like sub-Saharan Africa.

The results suggest that using web-based environments to recruit respondents for health-related data collections, such as for surveys and blood tests, can be a viable option in the sub-Saharan Africa region. Furthermore, our results provide first insights into important design parameters that seem to be critical in affecting the success of web-based recruitment strategies via Facebook ads and could serve as an initial blueprint. First, financial incentives matter, as promising rewards were related to a substantially higher willingness to participate in in-person health data collection activities. Second, trust in the sponsor behind the in-person health data efforts plays an important role with higher levels of trust leading to (1) a higher absolute number of persons who are willing to participate, and (2) selection effects with persons who distrust a specific sponsor being less likely to be willing to participate. Since trust levels differ substantially for different types of institutions and sponsors, the choice of sponsor should be considered carefully for the respective context. Beyond that, the study’s findings also indicated that several attitudinal but almost none of the sociodemographic variables are systematically related to the willingness to participate in health-related in-person data collection activities. Given that the Demographic and Health Surveys program, an important data source in the sub-Saharan Africa region, was recently discontinued, new data collection methods and recruitment strategies are likely to become more important in the near future.