#Dataprivacy #Trust #Youngadults #SexualHealth #Exploratoryfactoranalysis #Sturcturalequationmodeling
This research was conducted as a sub study from a larger UXR case study. Please reach out for the full manuscript
A while back while conducting my initial research examining young adults' sentiments when using sexual and reproductive health chatbots, I decided to concurrently collect some quantitative data out of curiosity!
Having scoured through hours of literature in this research area and conducted qualitative interviews with young adults, the concerns about data privacy and trust seemed to be the main factors determining one's intentions to use health chatbots.
Using the Theory of Planned Behavior by Azjen as a framework for this study, I was determined to understand how factors such as data privacy and trust could have an effect in determining intentions to adopt chatbot usage. Here are the research questions:
H1: Young adults' confidence in chatbot use has a positive effect in contributing to overall chatbot usage to seek sexual health information.
H2: When data privacy is guaranteed, there will be a positive effect on confidence when using a chatbot to seek sexual health information.
H3: When data privacy is guaranteed, there will be a positive effect on the amount of trust when using chatbots to seek sexual health information.
H4: Trust developed when using chatbots will result in a positive effect on one’s willingness to use chatbots when seeking sexual health information.
Data Collection
Participants were young college adults (18-24) and a total of 467 survey responses were collected. Among the participants that took part in the study, the majority were females (245, 52.4%), followed by males (111, 23.8%) and others (111, 23.8%). The majority of the participants in this study were also White or Caucasian Americans (275, 58.9%).
The online survey followed an attitudinal and intention scale developed by Nadarzynski and colleagues (2019). There was a total of 25 items in the survey. Participants were asked questions about their attitudes towards using health chatbots to seek health information and the likelihood they would use these chatbots for various health and healthcare needs. The likelihood and intention questions were assessed using a 5-point Likert scale (from ‘extremely unlikely’ to ‘extremely likely’), and asked participants to rate their likelihood of using chatbots for seeking sexual health information, information about medication, various diseases, potential symptoms, seeking results of medical tests, booking a medical appointment and looking for specialist medical services. The attitudinal questions were assessed using a 5-point Likert scale (from ‘strongly disagree’ to ‘strongly agree’), asking participants to indicate their attitudes towards using chatbots to seek sexual health information, data privacy and information accuracy concerns and overall trust and confidence towards seeking advice from health chatbots.
Analysis and Results
Exploratory Factor Analysis (EFA) was conducted to determine if any underlying structures existed for the measures on the following 9 usage likelihood variables: seeking sexual health information (likelihood 1), seeking information about medication (likelihood 2), seeking information about STIs (likelihood 3), seeking information about symptoms (likelihood 4), seeking information about medical test results (likelihood 5), booking a medical appointment (likelihood 6), looking for medical services (likelihood 7), seeking specialist advice for sexual health (likelihood 8) and seeking health information on chatbots in the next 12 months (likelihood 9). The analysis produced a two-component solution, which was evaluated with the following criteria: eigenvalue, variance, scree plot, and residuals. The criteria indicated a two-component solution was appropriate. After rotation, the first component accounted for 40.5% of the total variance of the original variables, while the second component accounted for 26.9%.
Table 1 presents the loadings for each variable. Component 1 consisted of Usage Likelihood 1, 2, 3, 4, 8, 9. Component 2 consisted of the remaining three variables (Usage Likelihood 5, 6, 7). All variables had positive loadings. As the purpose of this study was to focus on information-seeking behavior instead of service-seeking behavior, in which chatbot Roo cannot help patients book appointments, component 2 was removed from the subsequent PLS-SEM model.
Similarly, EFA was used on the following 16 attitudinal variables: Are you worried about health (chatbot accept 1), Are you worried about privacy using a health chatbot (chatbot accept 2), are you worried about the security of information (chatbot accept 3), are you confident in finding accurate health information using chatbots (chatbot accept 4), are you confident in identifying your own health symptoms (chatbot accept 5), are you comfortable in outlining symptoms to a chatbot (chatbot accept 6), do you prefer to talk face to face with a doctor about sexual health (chatbot accept 7), do you dislike talking to chatbots (chatbot accept 8), do you feel strange talking to chatbots about health (chatbot accept 9), do you believe that health chatbot could help to make better decisions (chatbot accept 10), would you trust sexual health advice from a chatbot (chatbot accept 11), do you feel a health chatbot is a good idea (chatbot accept 12), would you be willing to enter symptoms to an online form (chatbot accept 13), would you be curious how new technologies could improve health (chatbot accept 14), do you feel that reliable and accurate health information is important (chatbot accept 15) and do you only seek a doctor if you have an urgent health problem (chatbot accept 16).
The analysis produced a four-component solution, which was evaluated with the following criteria: eigenvalue, variance, scree plot, and residuals. The criteria indicated a four-component solution was appropriate. After rotation, the first component accounted for 20.0% of the total variance of the original variables, the second component accounted for 16.6%, the third component accounted for 12.5% and the fourth component accounted for 9.87%.
Table 2 presents the loadings for each variable. Component 1 consisted of Chatbot accept 7, 8, 9, 10, 11, 12. Component 2 consisted of Chatbot accept 13,14,15. Component 3 consisted of Chatbot accept 2 and 3. Component 4 consisted of Chabot accept 4 and 5. There was a mix of positive and negative loadings as indicated in Table 2.
After reducing the variables into specific groups, PLS-SEM was used. The results from the structural equation model are found in Figure 3. In keeping with standard practice, theoretical variables are indicated by circles; observed variables are by rectangles. The structural procedure of investigating the loadings and eliminating indicators (with loadings < 0.70) was adapted from Figure 3.
The leading step during the evaluation of a PLS-SEM framework was to investigate the outer model to facilitate the exertion and validation of the model dimension. For this reason inner relationships among paradigms and their indicators were measured. Table 3 shows the composite reliability wide-ranging between 0.798 to 0.958 for 3 out of 4 paradigms, which are far greater than the minimum requirement of 0.7.
Figure 3
The Model and its paths coefficient
Using the bootstrapping method, the path coefficient has been calculated. From Table 4, the confidence in using chatbot resulting in increased intentions to seek health information through that means was not supported and not significant (β = .089, t = 1.615, p=.106). The increase in respect towards data privacy was positively related to confidence in using the chatbot (β = 0.172, t = 3.144, p < 0.005). However, data privacy was negatively related to the overall trust towards chatbot use (β = -.249, t = 3.594, p < 0.001). Trust towards chatbots was positively related to the willingness to use technology such as chatbots (β = 0.444, t = 7.558, p < 0.001) and this pathway connecting willingness to use technology towards seeking health information was also positively related (β = 0.255, t = 6.285, p < 0.001).
Discussion and Future Implications
The results did not support the first hypothesis. The initial hypothesis assumed that based on the theoretical framework, having more confidence in using chatbots would increase one’s intention towards using the chatbot for this purpose. However, the results proved otherwise. Seeking sexual health information through chatbots or online robots can prove to be an intimidating task. While one could be confident in using chatbots from a technological perspective, there could be other factors such as comfort in disclosing sensitive information to a robot that could increase hesitancy when adopting this behavior. Furthermore, some might prefer the traditional face-to-face process of disclosing health information to a medical professional instead.
The other hypotheses were all satisfied. In particular, it was interesting to note that increased reassurance in data privacy did not necessarily increase trust towards chatbots. It is possible that other factors such as the ability to receive accurate health information from chatbots could also contribute to increased trust.
However, data privacy did have a positive relationship with confidence in using chatbots. Anonymity and confidentiality are often emphasized when disclosing vulnerable information to the chatbot. This is a reassuring factor that could increase usability and prevent the under-utilization of this helpful resource. This could reduce the spread of sexual health misinformation which is very harmful to one’s health.
For future implications, more robust quantitative methods such as the comparison of more models using PLS-SEM can be conducted. Overall, this quantitative phase of the study examined how various attitudes towards chatbots could lead to the intentions to adopt chatbot usage when seeking sexual health information. In particular, having the reassurance of data privacy and trust were important factors that could increase the adoption of this behavior.
ความคิดเห็น