Study design and setting
We conducted a cross-sectional study over a period of eight months post COVID-19 lockdown in Singapore, from July 6, 2020, through February 26, 2021. Up to 160 respondents (patients or their caregivers) were purposively sampled weekly to complete an interviewer-administered questionnaire during their visit to the two busiest ambulatory clinics at the second largest public hospital in Singapore. The respondents were proportionately stratified by gender and the following age categories (in years): 21–35; 36–50; 51–65; 65–80. We included only citizens and permanent residents of Singapore between ages 21–80 as this population was the most probable group of people who would fit the context of our study.
Discrete choice experiments
We conducted a discrete choice experiment (DCE) to elicit respondents’ preferences in the uptake of DCT(18). The DCE approach is anchored on the utility theory which postulates that when presented with alternatives, a rational individual (who is somewhat self-centred and who does not subscribe to other philosophical thoughts such as virtue ethics) would select the most preferred alternative that maximizes his/her utility (satisfaction or benefit)(19).
The utility function is defined as:
McFadden (1973) proposed modelling the expected utilities in terms of characteristics of the alternatives rather than attributes of the individuals(20). In the equation above, Ui represents the total utility derived from the ith alternative, β and Xi are a vector of estimated coefficients and attribute levels defining the alternative i. Each estimated coefficient is a preference weight and represents the relative contribution of the attribute level to the utility that respondents assign to an alternative. The probability of choosing the alternative i is equivalent to one alternative i among the choice of jth alternatives(21).
The logit function of a three-attribute study can be simplified as a linear function.
Marginal effects can be obtained from the partial derivatives of the attributes. The ratio of coefficients (-β1/ β2) represents the trade-off between two attributes (trading x1 for x2) when x3 is set to zero.
Questionnaire design
We designed a DCE questionnaire with three attributes (1. Social interactions, 2. Traced by a DCT tool, 3. Incentives to use a DCT tool) and two levels (presence or absence of the attribute) in the context of the COVID-19 situation in Singapore in June 2020. The country had just exited a 2-month partial lockdown and the use of “TraceTogether” was widely promoted during the months post lockdown. All eight combinations of the attribute levels were considered and combinations that mirrored each other were paired as a choice set (Table 1).
In this context, “Social interactions” refers to the ability of the respondent to engage in social activities when a lockdown was not in force; “Traced by a DCT tool” refers to the capturing of signals of 2 “TraceTogether” devices (app or token) within 2 meters of each other due to the respondent’s carrying of a DCT tool; and “Incentives” refers to any incentive (e.g., monetary, virtual rewards, lucky draw) which the respondent thought was reasonable to spur him/her to carry a DCT tool and/or to reduce his/her social activities. The questionnaire was piloted over two weeks with 156 respondents to refine the expression based on feedback from respondents. The questionnaire was also translated into Mandarin Chinese to cater to respondents who could not comprehend the English language.
Table 1
Discrete choice experiment choice sets
Q set
|
Choice
|
Attributes
|
|
Social interaction a
|
Traced by a DCT tool b
|
Incentive c
|
1
|
A
|
Yes
|
Yes
|
Yes
|
Respondents who chose option B do not place a high value on incentives and social interactions and may have concerns on being traced by a DCT tool.
|
B
|
No
|
No
|
No
|
2
|
A
|
Yes
|
No
|
No
|
Respondents who chose option B place a high value on incentives.
|
B
|
No
|
Yes
|
Yes
|
3
|
A
|
No
|
No
|
Yes
|
Respondents who chose option B place a high value on social interactions.
|
B
|
Yes
|
Yes
|
No
|
4
|
A
|
Yes
|
No
|
Yes
|
This choice set is a test of rationality. Respondents who chose option B were asked for the reason(s) for their choice.
|
B
|
No
|
Yes
|
No
|
a Social interaction: Ability of the respondent to engage in social activities when a lockdown was not in force. |
b Traced by a DCT tool: Whether close contact within 2 metres had occurred between 2 devices were captured due to the carrying of a DCT tool. Negative attribute. |
c Incentive: Any incentive (e.g., monetary, virtual rewards, lucky draw) which the respondent thought was reasonable to spur him/her to carry a DCT tool and/or to reduce his/her social activities. |
DCE indicates discrete choice experiment |
Data collection
All data collectors were trained to administer the questionnaire in a standardized manner guided by infographics, to minimize misinterpretation of the survey questions. Respondents were first asked if they were using the “TraceTogether” app or token, their willingness to use the “TraceTogether” tool, and whether they believed the data collected by “TraceTogether” was secured. They were then presented with two hypothetical scenarios for each DCE choice set and asked to choose their preferred option. After the DCE choice sets, respondents were asked the type of incentives that they thought would most likely motivate the population to use a DCT tool and who could persuade themselves to use a DCT tool during a pandemic. Demographic information was also collected to perform segmented analyses. We screened 5973 potential respondents, of which, 689 (11.5%) were not eligible for the study, 1341 (22.5%) refused to participate, and 3943 (66.0%) were interviewed. The dataset comprises 3892 respondents after dropping 51 (1.3%) respondents who failed to complete all the DCE choice sets.
Analysis
We further removed 53 respondents who did not provide a valid reason for choosing the “irrational” choice (Table 1, Q4, Choice B) and dropped Q4 from the analysis. The remaining 3839 responses (Q1 – Q3) were analyzed using the panel fixed conditional logit model with the robust variance estimator to correct for heterogeneity of variance. Akaike’s information criterion (AIC) and Bayesian information criterion (BIC) were computed for the selection of the best model, with a preference for lower AIC and BIC values (Table S2). The variables included in the final model are gender, age, tertiary education, willingness to use “TraceTogether”, using “TraceTogether”, and whether the respondent thought the data collected by “TraceTogether” will be secured.
Segmented analyses were performed to assess attribute trade-offs (by dividing the coefficients of the final model) between sociodemographic groups. We computed the total satisfaction scores of various profiles by adding up the coefficients of individual characteristics to illustrate the preferences between profile groups.
Descriptive analyses were conducted to assess the type of incentives participants thought could most likely motivate the population to use a DCT tool. Lucky draw or intangible items (e.g., points to claim vouchers) were converted to a monetary value based on the average cost of the item in the year 2020 to assess the monetary value of the incentives (Table S3). All analyses were performed with STATA version 15.0(22) and RStudio version 1.2.5033(23).
Sample size
We used the method proposed by de Bekker-Grob et al. to compute the minimum sample size required for this DCE analysis(24). Initial estimates based on the pilot dataset suggested a sample size of 1481 to detect differences in the main effects at a 0.05 statistical significance and 80% statistical power. Our post hoc analysis revealed that our 3839 responses were sufficient to detect differences in the main effects at 0.01 significance level with a power of 90% (Table S1).