Preference for supplementary voluntary health insurance and heterogeneity in China: a discrete choice experiment Authors: Lizheng Xu, Lei Si, Shunping Li, Jiaqi Yuan, Xu Si, Dan Zhang, Stephen Jan, Chaofan Li Published: 2026 ABSTRACT AbstractUnderstanding the population’s preferences for health insurance plays an important role in optimizing insurance scheme design and improving enrollment rate. This study aims to quantitatively investigate preference for supplementary voluntary health insurance (SVHI) from a multi-site survey and examine its heterogeneity in China. A discrete choice experiment was conducted in Shandong, Henan, and Sichuan provinces using multi-stage stratified sampling method. Five SVHI attributes were identified: premium, benefit package, deductible, reimbursement rate, and reimbursement for preexisting conditions. Choice sets were generated using a D-efficient design, grouped into two blocks randomly assigned to respondents, with each set comprising two SVHI options and an opt-out. Data were collected via face-to-face computer-assisted interviews. Mixed logit models were used to estimate preference weights, willingness-to-pay (WTP), and attribute importance scores. Preference heterogeneity was analyzed by disease-related financial risk awareness, numeracy, and health insurance knowledge, demographic, socioeconomic, and health characteristics. Of the 1326 respondents who completed the questionnaire, 1254 were included in the analysis. Reimbursement rate was the most important attribute (34.26%), followed by premium (25.06%), benefit package (17.60%), deductible (17.50%), and reimbursement for preexisting conditions (5.58%). Overall, respondents expressed the highest WTP (USD 48.50) for improving the reimbursement rate from 50% to 90%, while they showed lowest WTP (USD 8.60) for decreasing deductible from USD 2777.78 to 1388.89. Heterogeneity analysis revealed stronger enrollment preferences among respondents with risk awareness, higher health insurance knowledge, higher numeracy, higher educational attainment, higher income, and those living in urban areas. In addition, higher levels of risk awareness, insurance knowledge, numeracy, income, and education were associated with increased WTP for SVHI attributes. Preference heterogeneity by risk awareness and insurance knowledge suggests need for targeted risk information communication and education campaign to promote SVHI uptake and diverse insurance design tailored to socioeconomic differences in preferences for attributes.Keywords: supplementary voluntary health insurance, preference, discrete choice experiment, heterogeneity, China Rationale for the research: Supplementary voluntary private health insurance (SVHI) has the potential to serve as a complementary financial mechanism for mitigating catastrophic health expenditure, reducing medical impoverishment, and improving access to high-cost innovative medical treatments among patients with severe illnesses in China. To enhance the role and potential benefits of SVHI in the context of its currently low uptake, understanding individuals’ preferences and their heterogeneity is important for informing SVHI design and improving uptake. Rationale for the research: Supplementary voluntary private health insurance (SVHI) has the potential to serve as a complementary financial mechanism for mitigating catastrophic health expenditure, reducing medical impoverishment, and improving access to high-cost innovative medical treatments among patients with severe illnesses in China. To enhance the role and potential benefits of SVHI in the context of its currently low uptake, understanding individuals’ preferences and their heterogeneity is important for informing SVHI design and improving uptake. Key findings: Reimbursement rate and premium were the most important attributes of SVHI, followed by benefit package, deductible, and coverage for preexisting conditions. Marginal willingness-to-pay (WTP) estimates for different attribute levels provide policy-relevant insights to inform SVHI design. Participants’ disease-related financial risk awareness, higher health insurance knowledge and numeracy, more education attainment, and better economic status were positively associated with their stronger preferences and higher WTP for SVHI. Key findings: Reimbursement rate and premium were the most important attributes of SVHI, followed by benefit package, deductible, and coverage for preexisting conditions. Marginal willingness-to-pay (WTP) estimates for different attribute levels provide policy-relevant insights to inform SVHI design. Participants’ disease-related financial risk awareness, higher health insurance knowledge and numeracy, more education attainment, and better economic status were positively associated with their stronger preferences and higher WTP for SVHI. Lessons for health systems and health policy: Preferences for SVHI varied across individual socioeconomic and cognitive characteristics. These findings may also inform the need for more tailored SVHI schemes, alongside targeted disease-related financial risk communication and education strategies to improve SVHI uptake. These insights may also inform the design of context-appropriate complementary health insurance schemes in other low- and middle-income countries. Lessons for health systems and health policy: Preferences for SVHI varied across individual socioeconomic and cognitive characteristics. These findings may also inform the need for more tailored SVHI schemes, alongside targeted disease-related financial risk communication and education strategies to improve SVHI uptake. These insights may also inform the design of context-appropriate complementary health insurance schemes in other low- and middle-income countries. Introduction China has achieved universal health insurance coverage during past 20 years by establishing two basic public health insurance schemes: Urban Employee Basic Medical Insurance (UEBMI) and Urban–Rural Resident Basic Medical Insurance (URRBMI) (Fang et al. 2019). However, patients with severe disease, such as cancer or rare disease, still faced high out-of-pocket payments and likelihood of incurring catastrophic health expenditure (Xin et al. 2016, Li et al. 2023). To improve affordability for serious illness treatment and access to high-cost innovative drugs or health technologies, supplementary voluntary private health insurance (SVHI), known as “Huiminbao,” has been launched and expanded in China since 2020 (Zhang et al. 2024). The SVHI allows all enrollees of basic medical insurance to participate voluntarily and has minimal restrictions on their age, occupation, or preexisting conditions. The premium of SVHI is low (USD 7–56 per year in 2023). The SVHI schemes are operated and managed by commercial insurance companies at the city level, with municipal governments providing institutional and administrative support, including policy endorsement and cross-sector coordination, rather than direct premium subsidies. With increasing support from government, SVHI has been expanded to many cities in China. It covers patient co-payment expenditures, and innovative interventions excluded from public health insurance benefit packages. Serving as a supplementary insurance scheme designed to provide additional reimbursement for catastrophic or high medical expenses beyond public basic medical insurance, SVHI has been regarded as a tool for enhancing financial risk protection against critical diseases at an affordable premium (Yan and Faure 2025, Ding et al. 2026). Despite its rapid expansion, low enrollment rate of SVHI remains a critical challenge in China—the uptake rate was below 20% of the total population in most cities in 2023 (Khanna 2024). A synthesis of economic and social theories has identified both insurance scheme characteristics and individual characteristics as key determinants of health insurance enrollment decisions (Abiiro et al. 2016). As SVHI is purchased voluntarily, understanding potential enrollees’ preferences for key insurance attributes may help inform product design to better align with individuals’ needs and improve enrollment (Nguyen et al. 2017, Karim et al. 2021). Several studies have used discrete choice experiment (DCE) to elicit public preference for different attributes to inform new health insurance design and attract individuals to enroll (Pendzialek et al. 2017, Poteet and Craig 2019). Beyond the attributes of health insurance products, individual characteristics also played critical roles in shaping enrolment decisions. A systematic review identified sociodemographic and health-related factors as important determinants of voluntary health insurance enrollment (Adebayo et al. 2015). Furthermore, behavioral economics theory provides novel insights into how individuals’ psychological and cognitive factors could be used to explain consumer decisions to enroll or not to enroll in health insurance. Behavioral economics is an interdisciplinary field integrating psychology and economics, which argue that psychological and cognitive factors associated with bounded rationality are important determinants of individuals’ choices (Thaler 2016). As discussed in previous studies, the psychological factors influencing health insurance take-up include choice overload and complexity, misperceptions of risk, social comparisons, and reference dependence and framing, while cognitive factors mainly refer to individuals’ limited knowledge and numeracy regrading on insurance costs and benefits, which may influence insurance decision-making and preferences (Slovic 1987, Baicker et al. 2012, van Winssen et al. 2016, Skinner and Volpp 2017, Feher and Menashe 2021). Limited understanding of health insurance may affect decision-making, as individuals may not know which services are covered by SVHI (van Winssen et al. 2016). Moreover, it is difficult for individuals with low numeracy to evaluate the trade-off between premium and potential benefits (Dillingh et al. 2016). In addition, individuals with misperception of risk tend to be overly optimistic and underestimate the likelihood of adverse health events, which may lead them to forgo insurance (Baicker et al. 2012). Enhancing risk awareness is associated with higher preference for insurance (Miesler et al. 2017). Some empirical studies have demonstrated that better health insurance knowledge and numeracy are positively associated with insurance enrollment (McGarry et al. 2016, Park et al. 2022). Understanding whether and to what extent cognitive and psychological factors are associated with health insurance preference may inform the development of effective behavioral interventions to improve enrollment rate. Several DCEs have been conducted to investigate preferences for health insurance in both low-and middle-income countries (LMICs) and high-income countries (Leukert-Becker and Zweifel 2014, Abiiro et al. 2016, Obse et al. 2016, Karim et al. 2021). Previous studies also explored preference heterogeneity across individuals’ demographic (sex and age) and socioeconomic (education and income) characteristics (Abiiro et al. 2016, Karim et al. 2021), which implied that policy makers ought to design differentiated health insurance products. However, findings from studies conducted outside China may not be generalizable to Chinese population because of differences in health insurance system, economic development, and cultural context. As far as we know, no study has elicited preference for the innovative SVHI in China based on a multi-site survey or explored preference heterogeneity across cognitive and psychological factors. To fill these research gaps, this study aims to elicit the preference for SVHI using the DCE method and to examine preference heterogeneity from a behavioral economics perspective. Findings from this study may provide guidance on the optimal design of SVHI products and development of targeted behavioral interventions, thus improving the enrollment rate of SVHI. Methods Patient and public involvement Public participants were involved in the questionnaire design, pilot study and amendments to the questionnaire. Attributes and levels development of discrete choice experiment questionnaire The attributes and their levels were determined based on the following approaches (Supplementary Figure S1): (i) literature review of international health insurance schemes (Supplementary Table S1) (Vroomen and Zweifel 2011, Wang et al. 2012, Leukert-Becker and Zweifel 2014, Abiiro et al. 2014a, 2014b, 2016, Obse et al. 2016, Ozawa et al. 2016, Kazemi Karyani et al. 2018, Kazemi Karyani et al. 2019, Poteet and Craig 2019, Sydavong et al. 2019, Kalyango et al. 2021, Chen et al. 2022); (ii) document analysis of 182 Chinese SVHI schemes in 2023 (Supplementary Table S2); (iii) two-rounds Delphi consultation with 19–21 experts from academic institutions, official health insurance institutions and insurance companies in China (Supplementary Table S3) to evaluate the appropriateness of the DCE design and the scientific validity of the attribute and level selection; (iv) qualitative interviews with 37 residents (Supplementary Table S4); and (v) a pilot study (Supplementary Method and Tables S5–S7). The pilot study was conducted for 160 residents in Zibo City, Shandong Province in June 2024, to verify the comprehension, burden, and timing for the DCE questionnaire. Results showed expected signs and significance in most estimated variable coefficients and standard deviations (SDs). Based on the pilot study, number of deductible levels was reduced from four to three, and the descriptions of the attributes were revised to avoid potential misunderstanding. Data for pilot study was not included in the final analysis. The identified attributes included premium (four levels), benefit package (four levels), deductible (three levels), reimbursement rate (three levels), and reimbursement policy for preexisting conditions (three levels). Detailed attributes and levels are provided in Table 1. Table 1Attributes and levels of supplementary voluntary health insurance.AttributesDescriptionLevelsPremiumIndividual premium payment per yearUSD 27.78aUSD 20.83USD 13.89USD 6.94Benefit packageA set of services, drugs, or medical devices covered by supplementary voluntary health insurance schemeBasic benefit package: out-of-pocket expenses of services covered by basic medical insuranceExpanded 1: Basic benefit package + out-of-pocket expenses for services excluded from the basic health insurance schemeExpanded 2: Basic benefit package + out-of-pocket expenses for services excluded from the basic health insurance scheme + expenses of specific innovative high-price drugs (e.g. anticancer drugs or orphan drugs)Expanded 3: Basic benefit package + out-of-pocket expenses for services excluded from the basic health insurance scheme + expenses of specific innovative high-price drugs + value-added services (e.g. medication delivery services, health education and promotion, teeth cleaning, and so on)DeductibleFixed amounts of money that an insured individual needs to pay before the supplementary insurance covers any expensesUSD 2777.78USD 1388.89USD 694.44Reimbursement ratePercentage of eligible medical expenses between the deductible and ceiling that the insurance company will pay50%70%90%Reimbursement for preexisting conditionsHow does supplementary voluntary health insurance cover preexisting conditions?Exclusions for preexisting conditionsThe reimbursement rate for patients with preexisting conditions is lower than that for the healthy populationThe reimbursement rate for patients with preexisting conditions is as same as that for the healthy population aAll amounts in Chinese Renminbi were converted to 2024 US dollar (USD). Attributes and levels of supplementary voluntary health insurance. aAll amounts in Chinese Renminbi were converted to 2024 US dollar (USD). Experimental design The experimental design of DCE adhered to good research practice for conjoint analysis published by the International Society for Pharmacoeconomics and Outcomes Research (Bridges et al. 2011, Reed Johnson et al. 2013). We generated a D-efficient experimental design using Ngene software, resulting in 14 choice sets that were randomly divided into two blocks to reduce respondent burden. The final design was further evaluated using commonly reported design quality criteria, including level balance, minimal overlap, and orthogonality, to ensure acceptable statistical properties and respondent efficiency. To test the internal validity of the DCE, one consistency test assessing preference stability and one dominance test assessing respondents’ rationality were included in the questionnaire (Johnson et al. 2019). Consistency and dominance checks are two complementary internal validity tests commonly used in DCE studies, and they serve different purposes. The consistency check evaluates whether respondents show stable preference when presented with repeated choice tasks, thereby assessing response reliability. The dominance check assesses whether respondents can correctly identify and choose the clearly superior alternative when one option strictly dominates another across all attributes, thereby evaluating task understanding and attention. Using both checks allows us to capture different dimensions of response quality—consistency in preferences and rationality and attention in decision-making. Finally, nine choice sets were included in each questionnaire. Given that SVHI participation in China is voluntary, we used an unlabeled design with three alternatives per choice set, including two SVHI schemes and one “opt-out” option. Supplementary Figure S2 shows an example of the choice sets. Each participant was assigned to one block containing nine choice sets. The questionnaire included informed consent, introduction to the research, basic demographic information of residents, nine DCE choice sets, and three cognitive and psychological factors. First, demographic information included sex, age, educational level, economic status, and region of residence. Second, we assessed three cognitive and psychological factors: health insurance knowledge was measured by an eight-item true/false quiz developed using literature review (Levitt 2015, James et al. 2021, Park et al. 2022, Gutiérrez 2023), Delphi, and qualitative interview method. The health insurance knowledge questionnaire has been published in Chinese (Si 2025). Numeracy—defined as respondents’ proficiency with basic probability and mathematical concepts—was measured by three probability questions (Schwartz et al. 1997, Lipkus et al. 2001). Awareness of health-related financial risk (abbreviated as “risk awareness”) refers to the recognition of the potential for hazards and adverse incidents (Woodward 2011). Following existing study on disaster insurance (Robinson et al. 2021, Li et al. 2023), risk awareness was assessed with a single question: “Do you know the proportion of medical impoverishment among the total poor population?” Responses were recorded as a dichotomous variable, with 0 indicating “No” and 1 indicating “Yes.” The full questionnaire is shown in Supplementary Table S8. Survey design We employed a multi-stage stratified sampling method. First, considering geographic distribution and economic development, three provinces (Shandong Province in eastern, Henan Province in central region, and Sichuan Province in western China) were selected. Second, we selected one provincial capital city (developed region) and one non-provincial capital city (underdeveloped region) in each province. Finally, one urban district and one rural county were randomly selected in each city. The rule of thumb proposed by Pearmain suggests that a sample size exceeding 100 could provide a basis for estimating preference data regardless of experiment design (Pearmain 1991, de Bekker-Grob et al. 2015). Considering 92% (147/160) valid response rate in pilot survey, no less than 110 residents were surveyed in each urban district or rural county. To conduct preference heterogeneity analyses, the sample size in each city was set as 220 and the final sample size was 1320. According to the ISPOR good research practices, sample sizes in the range of 1000–2000 respondents are expected to produce relatively small confidence intervals (Reed Johnson et al. 2013). Inclusion criteria for participants were (i) enrolled in basic medical insurance; (ii) aged 18 years or older; (iii) having primary school education and above; and (iv) provided informed consent. Exclusion criteria included refusal to participate and having cognitive impairments or other mental conditions. The formal survey was conducted through face-to-face, computer-assisted interviews by trained interviewers on weekdays in July–August 2024. To improve participation and completion rate, each participant received a gift valued at approximately USD 2 after the survey. All participants were informed at the beginning of the survey that the same gift would be provided upon completion of all questionnaire items, regardless of their specific responses or choices. All responses were considered equally acceptable. Quality control methods we implemented are reported in the Supplementary material. Statistical analysis Data analysis from DCE was align to “Statistical Methods for the Analysis of Discrete Choice Experiments: A Report of the ISPOR Conjoint Analysis Good Research Practices Task Force” (Hauber et al. 2016). The utility function with only main-effects was specified as follows: Consistent with previous studies on preference for health insurance (Abiiro et al. 2014b, Chen et al. 2022), the premium attribute was specified as a continuous variable, assuming linear utility changes associated with a one-unit increase in premium. The levels of all other attributes were dummy coded as categorical variables, allowing for non-linear utility effects across attribute levels. An alternative-specific constant (ASC) was included in the mixed logit model to indicate utility generated by the opt-out option (coded as 1) relative to the non–opt-out option (coded as 0). The sign of the coefficient for each attribute’ level (β) indicated whether individuals had higher or lower preferences for the corresponding attribute level compared to the reference level, and the magnitude of coefficient indicated the strength of preference. The statistical significance of SD of β indicated the presence of preference heterogeneity. A positive ASC implied preference toward enrollment in SVHI, whereas a negative ASC implied preference for uninsured status. The parameters were estimated using maximum simulated likelihood with robust standard errors, and Sobol draws were employed to improve simulation efficiency and estimation stability. All attribute parameters were specified as random to capture preference heterogeneity across individuals within the mixed logit model. Based on model fit indicators (pseudo-R-squared, log-likelihood, AIC, and BIC), the mixed logit model outperformed the conditional logit model and was selected for the main analysis. To interpret the results intuitively, the marginal WTP was estimated in both WTP space and preference space approach (Hole and Kolstad 2012, Dolea 2012) and importance of each attribute was estimated using utility range approach (Hauber et al. 2016). Furthermore, in order to provide practical implications for optimal design of SVHI, we predict the probability of individuals enrolling in defined SVHI scenarios with specified attributes and levels. Sensitivity analyses were conducted including the 72 respondents that failed the consistency or dominance checks. To explore the observed preference heterogeneity, interaction terms between respondent characteristics and attribute levels with significant SDs in the main-effects mixed logit model were incorporated, consistent with approaches adopted in previous studies (Karim et al. 2022). The utility function with both main-effects and interaction terms was specified as follows: where Zn represents respondent characteristics, including sex, age group, educational attainment, region of residence, household income per capita, presence of chronic disease, type of basic medical insurance, previous SVHI enrollment experience, risk awareness, numeracy, and health insurance knowledge. Household income per capita was evenly divided into three groups, while numeracy and health insurance knowledge were categorized into two groups based on their mean scores prior to analysis to facilitate the interpretation of interaction effects. The interaction coefficients captured preference heterogeneity in SVHI attribute levels according to respondents’ sociodemographic, health-related, and cognitive characteristics. The sign and statistical significance of the interaction coefficients (β11n−β16n) indicate whether and to what extent preferences for specific SVHI attribute levels varied across respondent subgroups. Data cleaning and descriptive analysis were conducted using STATA version 17.0 (StataCorp 2021) while conditional and mixed logit analyses were performed using “logitr” package in R software (version 4.4.1) (Helveston 2023). The ethical approval was obtained from the Ethics Committee of the Center for Health Management and Policy Research, at the author’s institution (ECSHCMSDU20240701). Results Respondents’ characteristics Among all the 1359 invited residents, 1326 participated in the formal survey (response rate 97.57%). After excluding 72 cases that failed the consistency or dominance check, 1254 were included in the analysis. The respondents were predominately female (71.69%), with an average age of 45.38 (SD = 14.43) years old and an educational level of high school or above (63.31%) (Table 2). A quarter of the respondents had a history of SVHI enrollment prior to the survey. Most residents (63.48%) lacked risk awareness, with an average numeracy score of 1.06 (SD = 1.12) and a health insurance knowledge score of 5.12 (SD = 2.21). Table 2Basic characteristics of the participants (n = 1254).Variables n (%)Sex Male355 (28.31) Female899 (71.69)Age, mean (SD)45.38 (14.43) 18–39458 (36.52) 40–49278 (22.17) 50–59281 (22.41) ≥60237 (18.90)Educational level Primary school177 (14.11) Middle school283 (22.57) High school230 (18.34) Junior college257 (20.49) Bachelor and above307 (24.48)Region of residence Rural615 (49.04) Urban639 (50.96)Monthly household income per capita, mean (SD)a318.10 (332.97) Low404 (32.22) Medium424 (33.81) High426 (33.97)Presence of chronic disease No957 (76.32) Yes297 (23.68)Basic health insurance Urban Employee Basic Medical Insurance709 (56.54) Urban-Rural Resident Basic Medical Insurance545 (43.46)History of supplementary voluntary health insurance enrollment No939 (74.88) Yes315 (25.12)Risk awareness No796 (63.48) Yes458 (36.52)Numeracy, mean (SD)1.06 (1.12) ≤1838 (66.83) ≥2416 (33.17)Health insurance knowledge, mean (SD)5.12 (2.21) <6602 (48.01) ≥6652 (51.99)SD, standard deviation. aAll amounts in Chinese Renminbi were converted to 2024 US dollar (USD). Basic characteristics of the participants (n = 1254). SD, standard deviation. aAll amounts in Chinese Renminbi were converted to 2024 US dollar (USD). Participants’ preference for supplementary voluntary health insurance and willingness-to-pay Table 3 reports the preference weights and WTP across attributes and levels of SVHI using mixed logit model. Overall, respondents were more likely to enroll in SVHI than to opt out, as indicated by the significantly negative coefficients of “opt-out” terms (β = −9.742, 95% CI −11.458 to −8.019, P < 0.001). The respondents highly preferred SVHI with low premium (β = −0.003, 95% CI −0.004 to −0.002, P < 0.001). As expected, coefficients for all levels of the other four attributes were significantly positive, suggesting that all attributes played important roles in respondents’ choices of the SVHI scheme. On average, respondents were willing to pay an additional premium of USD 8.60 (95% CI 3.51–13.71) and USD 21.62 (95% CI 14.03–29.18) to decrease the deductible from USD 2777.78 to 1388.89 and 694.44, respectively. They were also willing to pay an additional premium of USD 25.69–41.21 to include high-cost innovative drugs and value-added health services in the benefit package. Increasing reimbursement rate from 50% to 90% was valued at USD 48.50 (95% CI 33.47–63.85). Compared to excluding preexisting conditions, respondents were willing to pay additional USD 22.27 (95% CI 15.01–29.53) for a lowered the reimbursement for preexisting conditions, or USD 28.30 (95% CI 19.40–37.14) to have the same reimbursement rate as for healthy individuals. Among all included attributes, the respondents assigned the highest importance to reimbursement rate (34.26%) and premium (25.06%), followed by benefit package (17.60%) and deductible (17.50%), with the lowest importance to reimbursement for preexisting conditions (5.58%). Table 3Participants’ preference for supplementary voluntary health insurance and willingness-to-pay based on mixed logit model (n = 1254).Attribute and attribute levelsPreference estimatesWillingness-to-payRIS (%) β 95% CISD95% CI β 95% CIPremium−0.003***−0.004 to −0.0020.011***0.009–0.012––25.06Deductible (reference: USD 2777.78a)17.50 USD 1388.890.147**0.053–0.2380.143−0.259–0.5458.60***3.51–13.71 USD 694.440.473***0.373–0.572−0.576***−0.403 to −0.74821.62***14.03–29.18Benefit package (reference: basic)17.60 Expanded 10.666***0.534–0.794−0.653***−0.859 to −0.43727.85***17.26–38.57 Expanded 20.567***0.466–0.669−0.018−0.087–0.04925.69***16.77–34.76 Expanded 30.895***0.783–1.006−0.383***−0.509 to −0.25941.21***28.66–54.03Reimbursement rate (reference: 50%)34.26 70%0.380***0.291–0.467−0.086−0.185–0.01020.72***13.60–28.09 90%1.018***0.900–1.1370.986***0.857–1.11248.50***33.47–63.85Reimbursement for preexisting conditions (reference: exclusion)5.58 Lower than healthy population0.457***0.357–0.557−0.058−0.174–0.05822.27***15.01–29.53 Same with healthy population0.561***0.456–0.662−0.128−0.277–0.02228.30***19.40–37.14Opt-out−9.742***−11.458 to −8.0196.952***5.961–7.949−672.59***−956.64 to −392.08Number of observations26 33426 334Number of respondents12541254Pseudo-R2 (McFadden)0.4090.400Log pseudo-likelihood−5699.192−5785.059AIC11 442.38411 612.118BIC11 598.14411 760.798SD, standard deviation; CI, confidence interval; RIS, relative importance score; OOP, out-of-pocket expenses; AIC, Akaike information criterion; BIC, Bayesian information criterion.*P < 0.05; **P < 0.01; ***P < 0.001. aAll amounts in Chinese Renminbi were converted to 2024 US dollar (USD). The willingness to pay was estimated using WTP space approach. Participants’ preference for supplementary voluntary health insurance and willingness-to-pay based on mixed logit model (n = 1254). SD, standard deviation; CI, confidence interval; RIS, relative importance score; OOP, out-of-pocket expenses; AIC, Akaike information criterion; BIC, Bayesian information criterion. *P < 0.05; **P < 0.01; ***P < 0.001. aAll amounts in Chinese Renminbi were converted to 2024 US dollar (USD). The willingness to pay was estimated using WTP space approach. The predicted probabilities of SVHI uptake across different scenarios were also estimated (Supplementary Table S9 and Figure S3). The opt-out probability (Scenario 14) of 0.01% indicated the high willingness to enroll among the participants. When adjusting each attribute to a better level, the predicted probabilities changed from 53.67% (Scenario 9) to 92.43% (Scenario 13). Improving the reimbursement rate led to the greatest increase in uptake rates, whereas lowering the deductible resulted in the smallest increase in enrollment probability. The results of conditional logit model were similar with those of mixed logit model (Supplementary Tables S10 and S11 and Figure S4). In addition, sensitivity analyses including the 72 participants who failed the consistency or dominance checks (total n = 1326) were conducted, and the results are reported in Supplementary Tables S12–S14. The findings were largely consistent with the main analysis based on the 1254 valid respondents. Heterogeneity of preferences and willingness-to-pay across respondent characteristics Besides of the heterogeneity in preferences across attribute levels and “opt-out” terms shown in Table 3. Further analysis was conducted to examine the preference heterogeneity across respondents’ characteristics (Table 4). The interaction terms between “opt-out” and several respondent characteristics were statistically significant, indicating the presence of preference heterogeneity in decisions to enroll or not enroll in SVHI within the sample. Specifically, the interactions between the opt-out alternative and risk awareness (β = −0.949, P = 0.015), numeracy (β = −1.883, P = 0.036), health insurance knowledge (β = −2.392, P < 0.001) were significantly negative. In contrast, the interaction between respondent characteristics and several SVHI attributes levels were significantly positive (βRisk awareness * Deductible USD 694.44 = 0.255, P = 0.001; βNumeracy * Deductible USD 694.44 = 0.220, P = 0.004; βHealth insurance knowledge * Premium = 0.003, P = 0.003; βHealth insurance knowledge * Benefit package expanded1 = 0.190, P = 0.041). These findings indicate that respondents with risk awareness and high numeracy showed stronger preferences for lower deductibles, while respondents with greater health insurance knowledge were somewhat less sensitive to premiums and showed stronger preferences for expanded benefit packages. Table 4Preference heterogeneity analysis by psychological and cognitive factors using a mixed logit model with interaction terms (n = 1254).Attributes levelsBasic characteristicsRisk awarenessNumeracyHealth insurance knowledgeCoefficient P-valueCoefficient P-valueCoefficient P-valuePremium−0.00040.6730.0020.0810.0030.003Deductible (USD 694.44)0.2550.0010.2200.0040.1240.089Benefit package expanded 10.0200.8450.0530.5930.1900.041Benefit package expanded 30.1190.1200.0230.7660.1370.063Reimbursement rate (90%)0.0940.2910.0010.9870.1020.235Opt-out−0.9490.015−1.8830.036−2.392<0.001Number of observations26 33426 33426 334Number of respondents125412541254Pseudo-R2 (McFadden)0.4090.4080.410Log pseudo-likelihood−5701.118−5704.503−5691.649AIC11 448.23711 455.00611 429.299BIC11 611.07711 617.84611 592.139For clarity, only interaction terms are reported, and other terms have been omitted.AIC, Akaike information criterion; BIC, Bayesian information criterion. Preference heterogeneity analysis by psychological and cognitive factors using a mixed logit model with interaction terms (n = 1254). For clarity, only interaction terms are reported, and other terms have been omitted. AIC, Akaike information criterion; BIC, Bayesian information criterion. The result of preference heterogeneity across other respondent characteristics was reported in Supplementary Tables S15–S25. The significant interaction terms indicated that respondents with higher education levels, higher income, and those residing in urban areas were less likely to opt out, while those enrolled in the URRBMI were more prone to opt out. Furthermore, respondents with high education level and income demonstrated stronger preferences for insurance schemes featuring expanded benefit packages and higher reimbursement rates compared to their respective reference groups. Respondents aged 60 and above, as well as those enrolled in the URRBMI, exhibited higher sensitivity to premium increases. No significant preference differences were observed based on chronic disease status or prior SVHI enrollment. While gender also showed no significant heterogeneous preference for most attributes, female respondents demonstrated higher preference for expanded benefit package. Moreover, there was substantial heterogeneity in WTP for SVHI attributes. As shown in Fig. 1a, respondents with risk awareness were willing to pay USD 27.12 and 55.86 for increases in reimbursement rates from 50% to 70% and 90%, respectively, compared to USD 17.26 and 44.93 among those without risk awareness. Participants with risk awareness were also willing to pay USD 29.95 to reduce deductibles from USD 2777.78 to 694.44 and USD 33.11 to expand the basic benefit package to include high-cost drug coverage—both significantly higher than the WTP among those without risk awareness (USD 17.35 and 21.71, respectively). Figure 1b and c show that respondents with higher health insurance knowledge scores (≥6) and greater numeracy (≥2) were more willing to pay for coverage of preexisting conditions expenses compared to those with lower scores (<6) and numeracy (≤2). Willingness-to-pay for attributes by psychological and cognitive characteristics. Supplementary Tables S26–S34 showed substantial heterogeneity in WTP for SVHI attributes across respondent demographic and socioeconomic characteristics. The results indicated that respondents with higher educational attainment, higher income, urban residence, and UEBMI exhibited relatively higher WTP for favorable SVHI attributes, including lower deductibles, expanded benefit packages, higher reimbursement rates, and reimbursement for preexisting conditions. Notably, no significant differences in WTP for reimbursement for preexisting conditions were observed between respondents with and without chronic diseases (Supplementary Table S31), suggesting relatively similar preferences for preexisting condition coverage across health-status groups. The WTP estimates calculated using the preference-space approach are reported in Supplementary Table S34. Minor differences were observed between the WTP estimates derived from the two estimation approaches, with the WTP-space approach generally producing higher estimates. WTP results expressed in Chinese Yuan (RMB) are also presented in Supplementary Table S34. Discussion To the best of our knowledge, this is the first study to elicit residents’ preference and its heterogeneity for SVHI in China from a multi-site survey, based on cognitive and psychological factors from a behavioral economics perspective. This study found reimbursement rate and premium were the most important attributes for potential enrollees. Moreover, respondents with risk awareness, higher numeracy, higher health insurance knowledge, higher educational attainment, higher income, urban residence, and UEBMI coverage exhibited stronger preferences for SVHI enrollment. Meanwhile, respondents with risk awareness, higher educational attainment, higher income, urban residence, and UEBMI coverage generally reported higher WTP for favorable SVHI attributes, including lower deductibles, expanded benefit packages, and higher reimbursement rates. In addition, individuals with higher numeracy and health insurance knowledge tend to report higher WTP for reimbursement of preexisting conditions and lower deductibles compared with their counterparts. These findings may inform optimal design of SVHI schemes and the development of targeted behavioral interventions associated with higher preference for SVHI. Future experimental studies are needed to evaluate of the effect behavioral interventions on SVHI enrollment behavior. Our findings indicated that most respondents were willing to join SVHI rather than opt out, suggesting a relatively high potential demand for SVHI in China for protection against serious disease-related financial risks. Although basic social health insurance in China covers about 60% of the medical expenditure (Zhang et al. 2024), there remains a high likelihood of catastrophic health expenditures and medical impoverishment in serious disease conditions such as rare diseases and life-threatening diseases, particularly among financially vulnerable populations (Yip et al. 2023). Under the rising healthcare demands in aging population and high incidence of catastrophic health expenditure especially among individuals with cancer or rare disease, enrolling in both basic medical insurance and SVHI would prevent them from financial risks via reducing OOP payments and improve their affordability for high-cost treatments. The elicited attributes importance from this study aligned with previous research on private health insurance (Chen et al. 2022). The highest importance of reimbursement rate reflected participants’ strong desire to reduce OOP payment of severe disease. Although having been enrolled in basic medical insurance, participants still need pay around 40% of the medical expenses, leading to high risks of catastrophic expenditure when facing severe conditions (Zhang et al. 2024). The significance of premium and reimbursement rate also highlighted their role as key barriers to health insurance enrollment, supported by previous study in China (Wang et al. 2021). However, the reimbursement for preexisting conditions was the least important attribute. The possible reason maybe that majority of participants (76.32%) had no chronic disease; thus, they paid little attention to preexisting conditions. The preference heterogeneity across psychological and cognitive factors may provide insights into design targeted interventions to promote SVHI enrollment from a behavioral economics perspective. First, awareness of disease-related financial risk was significantly related to participants’ willingness to enroll in SVHI and WTP value, implying that interventions aiming at increasing financial risk awareness were associated with higher preference for SVHI. Previous studies on disaster insurance have also proved that risk communication, such as displaying probability of adverse incidents using visual aids and framing cumulative probability of flood across lifetime, could raise risk awareness and promote enrollment in disaster insurance (Robinson et al. 2021, Li and Liu, 2023). Taken together, the finding of heterogeneous preference for SVHI across risk awareness may imply that disease financial risk communication intervention may shape public preferences toward SVHI enrollment. Second, according to bounded rationality theory and information processing theory, decision-making on complex insurance products is constrained by cognitive limitations, such as limited numeracy and information/knowledge insufficiency (Szrek and Bundorf 2014, McGarry et al. 2016, Holst et al. 2023). Our study also confirmed that higher level of health insurance knowledge was associated with stronger willingness to enroll and higher WTP for SVHI. Future studies are encouraged to examine the effectiveness of interventions aimed at improving health insurance knowledge and numeracy, such as educational seminars, informational booklets, and interactive digital tools, on health insurance enrollment decisions. Another main finding of this study is the socioeconomic heterogeneity of preference for SVHI. Wealthier and educated individuals having greater preference for enrolling in SVHI is potentially due to lower price sensitivity to premiums and higher perceived medical need (Yip et al. 2023). Nevertheless, there was litter difference in WTP for SVHI by region of residence, which could be resulted from the low income levels among rural residents. The heterogeneity of WTP for attributes across socioeconomic characteristics implied diversified insurance demands and market segmentation. This study contributes to quantitative insights into the relative importance of attributes of SVHI and preferences heterogeneity from a behavioral economics perspective. These findings provide important policy implications for optimizing health insurance design to meet population demands and enhance SVHI uptake rate. First, given that SVHI in China is designed at the city level based on local population characteristics and economic conditions, the SVHI attributes importance ranking, participant’s trade-offs, and quantitative WTP evidence from this study may inform SVHI design and pricing decisions across cities, even when contextual settings differ. Moreover, it contributes to predicting the uptake rate of SVHI scenarios with different attributes’ levels and informs strategies to enhance the attraction of SVHI schemes to the general public. Second, this study suggests that psychological and cognitive factors can influence individuals’ choices of SVHI. These findings may provide reference for the development of behavioral interventions to promote SVHI uptake, such as designing risk communication strategies, improving public health insurance literacy and numeracy. A previous study conducted in the USA also demonstrated that behavioral nudge interventions can substantially increase health insurance take-up rates with low cost (Feher and Menashe 2021, Yokum et al. 2022). Together with findings from previous studies, out research also provide SVHI enrollment preference and socioeconomic heterogeneity evidence for other LMICs having not yet achieved universal health insurance coverage, contributing to developing tailored behavioral strategies to promote health insurance enrollment. This study had several limitations. First, more women than men were recruited in the survey. The predominance of female respondents (71.69%) is primarily attributable to the timing of data collection on weekdays. Male residents were away at work, making them less accessible for community-based surveys. Although a previous survey for long-term care insurance study also reported similar sample characteristics (Wang et al. 2021), the sample with a sex proportion failing in matching the actual population may compromise its representativeness of the general population. Importantly, based on our study results of heterogeneity analysis, sex differences in preferences for health insurance attributes were not statistically significant. Therefore, we believe that the higher proportion of women in this study may not seriously bias the findings. Second, the benefit package attribute was simplified in the DCE questions, to reduce respondents’ cognitive burden. In practice, SVHI schemes in different cities include various numbers and types of high-cost drugs. This necessary simplification may influence the results. Third, the DCE design required the participants to make hypothetical choices, which may not fully reflect financial constraints encountered in real-world decision-making. Nevertheless, we sought to minimize the discrepancy between hypothetical and actual choice environments by identifying attributes and levels based on existing SVHI products currently in operation. Furthermore, we followed the DCE design and data analysis guidelines (Bridges et al. 2011, Reed Johnson et al. 2013). Forth, to improve participation and completion rate, we gave each participant a gift valued at approximately USD 2 after the survey. Participants were explicitly informed that all responses were equally acceptable and that the same gift would be provided regardless of their choices. All incentives were provided only after participants had completed the entire questionnaire, minimizing any potential impact on response behavior during the survey process. Previous evidence shows that incentives effectively improve participation and completion rates without necessarily compromising data quality (Chen et al. 2015, Stanley et al. 2020). Fifth, this study used a cross-sectional DCE approach to examine associations between respondent characteristics, cognitive and psychological factors, and stated preferences for health insurance in a hypothetical choice experiment rather than actual enrollment behavior. To test causal relationships, future experimental or longitudinal studies are needed to assess whether interventions targeting these factors can influence actual insurance enrollment behavior. Finally, there is no standardized or widely accepted approach for measuring risk awareness. In this study, risk awareness was assessed using a single survey question, which may not fully capture the multidimensional nature of individual risk awareness or perceived susceptibility to disease-related financial risk. Future research should develop and validate more comprehensive measurement tools with established reliability and validity to better assess risk awareness. Conclusion This study suggests that reimbursement rate and premium are two most important attributes of SVHI in China, and risk awareness and higher levels of health insurance knowledge and numeracy were associated with stronger preferences for joining SVHI. Moreover, there exists preference heterogeneity by respondent socioeconomic characteristics. Optimizing insurance design in the future should account for preference heterogeneity based on the characteristics of potential enrollees and design various SVHI schemes to meet diverse demands. These results imply that efforts to improve risk communication and insurance knowledge maybe helpful in overcoming certain cognitive barriers, thereby contributing to higher uptake rates. Supplementary Material Contributor Information Lizheng Xu, School of Public Health, Cheeloo College of Medicine, Shandong University, 44 Wenhuaxi Road, Jinan, Shandong 250012, China; National Institute of Health Data Science of China, Shandong University, 44 Wenhuaxi Road, Jinan, Shandong 250012, China; The George Institute for Global Health, Faculty of Medicine and Health, University of New South Wales, 55 Botany Street Randwick, Sydney, New South Wales 2031, Australia. Lei Si, School of Health Sciences, Faculty of Health, Western Sydney University, Locked Bag 1797, Penrith, New South Wales 2751, Australia; Translational Health Research Institute, Western Sydney University, Locked Bag 1797, Penrith, New South Wales 2751, Australia; School of Health Management, Anhui Medical University, 81 Meishan Road, Hefei 230032, China. Shunping Li, Department of Social Medicine and Health Management, School of Public Health, Cheeloo College of Medicine, Shandong University, 44 Wenhuaxi Road, Jinan, Shandong 250012, China; NHC Key Laboratory of Health Economics and Policy Research (Shandong University), 44 Wenhuaxi Road, Jinan, Shandong 250012, China; Center for Health Management and Policy Research, Shandong University (Shandong Provincial Key New Think Tank), 44 Wenhuaxi Road, Jinan, Shandong 250012, China; Center for Health Preference Research, Shandong University, 44 Wenhuaxi Road, Jinan, Shandong 250012, China. Jiaqi Yuan, Department of Social Medicine and Health Management, School of Public Health, Cheeloo College of Medicine, Shandong University, 44 Wenhuaxi Road, Jinan, Shandong 250012, China; NHC Key Laboratory of Health Economics and Policy Research (Shandong University), 44 Wenhuaxi Road, Jinan, Shandong 250012, China; Center for Health Management and Policy Research, Shandong University (Shandong Provincial Key New Think Tank), 44 Wenhuaxi Road, Jinan, Shandong 250012, China; Center for Health Preference Research, Shandong University, 44 Wenhuaxi Road, Jinan, Shandong 250012, China. Xu Si, Department of Social Medicine and Health Management, School of Public Health, Cheeloo College of Medicine, Shandong University, 44 Wenhuaxi Road, Jinan, Shandong 250012, China; NHC Key Laboratory of Health Economics and Policy Research (Shandong University), 44 Wenhuaxi Road, Jinan, Shandong 250012, China; Center for Health Management and Policy Research, Shandong University (Shandong Provincial Key New Think Tank), 44 Wenhuaxi Road, Jinan, Shandong 250012, China. Dan Zhang, Institute for Hospital Management, Tsinghua Shenzhen International Graduate School, Tsinghua University, 2279 Lishui Road, Shenzhen, Guangdong 518055, China. Stephen Jan, The George Institute for Global Health, Faculty of Medicine and Health, University of New South Wales, 55 Botany Street Randwick, Sydney, New South Wales 2031, Australia; The George Institute of Global Health UK, Faculty of Medicine, Imperial College London, White City Campus, London W12 0SN, United Kingdom. Chaofan Li, Department of Social Medicine and Health Management, School of Public Health, Cheeloo College of Medicine, Shandong University, 44 Wenhuaxi Road, Jinan, Shandong 250012, China; NHC Key Laboratory of Health Economics and Policy Research (Shandong University), 44 Wenhuaxi Road, Jinan, Shandong 250012, China; Center for Health Management and Policy Research, Shandong University (Shandong Provincial Key New Think Tank), 44 Wenhuaxi Road, Jinan, Shandong 250012, China; Center for Health Preference Research, Shandong University, 44 Wenhuaxi Road, Jinan, Shandong 250012, China. Author contributions Concept and design: C.L. and L.X. Data collection: C.L., L.X., J.Y. and X.S. Data analysis and interpretation: C.L., L.X., S.J., L.S., J.Y., X.S. and S.J. Drafting of the manuscript: L.X. and C.L. Critical revision of the manuscript: L.X., C.L., L.S., S.J., S.L. and D.Z. Final approval of the version to be submitted: All authors. Supplementary material Supplementary material is available at Health Policy and Planning online. Ethical approval This study obtained ethics approval from the Ethics Committee of the Center for Health Management and Policy Research, Shandong University (ECSHCMSDU20240701). Funding This study was supported by the National Natural Science Foundation of China (no. 72204146), Shandong Provincial Natural Science Foundation (no. ZR2022QG018), and MOE (Ministry of Education in China) Liberal arts and Social Sciences Foundation (grant no. 22YJCZH076). The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. Data availability The code used in this study is available from the corresponding author upon reasonable request. Reflexivity statement The authors include three female and five male and span multiple levels of seniority and disciplines. The authors include six professors and experienced researchers, and two masters. The authors have expertise in several areas including health insurance, discrete choice experiment, and health system science. Three authors have extensive public health experience in global settings. Five authors are from low- and middle-income countries and three authors are currently based in high-income countries. References Associated Data Supplementary Materials Data Availability Statement The code used in this study is available from the corresponding author upon reasonable request.