Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Effects of China’s urban basic health insurance on preventive care service utilization and health behaviors: Evidence from the China Health and Nutrition Survey

  • Wanyue Dong,

    Roles Formal analysis, Writing – original draft

    Affiliation School of Public Health, Health Science Center, Xi’an Jiaotong University, Xi’an, Shaanxi, People’s Republic of China

  • Jianmin Gao ,

    Roles Conceptualization

    Affiliation School of Public Policy and Administration, Xi’an Jiaotong University, Xi’an, Shaanxi, People’s Republic of China

  • Zhongliang Zhou,

    Roles Resources, Writing – review & editing

    Affiliations School of Public Policy and Administration, Xi’an Jiaotong University, Xi’an, Shaanxi, People’s Republic of China, Global Health Institute, Health Science Center, Xi’an Jiaotong University, Xi’an, Shaanxi, People’s Republic of China

  • Ruhai Bai,

    Roles Formal analysis

    Affiliation Global Health Institute, Health Science Center, Xi’an Jiaotong University, Xi’an, Shaanxi, People’s Republic of China

  • Yue Wu,

    Roles Formal analysis

    Affiliation School of Public Policy and Administration, Xi’an Jiaotong University, Xi’an, Shaanxi, People’s Republic of China

  • Min Su,

    Roles Formal analysis

    Affiliation School of Public Policy and Administration, Xi’an Jiaotong University, Xi’an, Shaanxi, People’s Republic of China

  • Chi Shen,

    Roles Writing – review & editing

    Affiliation School of Public Policy and Administration, Xi’an Jiaotong University, Xi’an, Shaanxi, People’s Republic of China

  • Xin Lan,

    Roles Writing – review & editing

    Affiliation School of Public Health, Health Science Center, Xi’an Jiaotong University, Xi’an, Shaanxi, People’s Republic of China

  • Xiao Wang

    Roles Resources, Writing – review & editing

    Affiliation International Business School Suzhou, Xi’an Jiaotong-Liverpool University, Suzhou, Jiangsu, People’s Republic of China



Lifestyle choices are important determinants of individual health. Few studies have investigated changes in health behaviors and preventive activities brought about by the 2007 implementation of Urban Resident Basic Health Insurance (URBMI) in China. This study, therefore, aimed to explore whether URBMI has reduced individuals’ incentives to adopt healthy behaviors and utilize preventive care services.


Data were drawn from two waves of the China Health and Nutrition Survey. Respondents were categorized according to their insurance situation before and after the URBMI reform in 2006 and 2011. Propensity score matching and difference-in-differences methods were used to measure levels of preventive care and behavior changes over time. Estimations were also made based on gender, self-reported health, and income.


We found that URBMI implementation did not change residents’ utilization of preventive care services or their smoking habits, drinking habits, or other risky behaviors overall. However, the likelihood of sedentariness did increase by five percentage points. Females tended to be more sedentary while males were less likely to drink soft drinks. Residents with poor self-reported health exercised less while those who reported good health were more likely to be sedentary. Low- and middle-income residents were likely to be sedentary while middle-income people tended to smoke after becoming insured.


Since URBMI implementation, some unhealthy behaviors like sedentariness have increased among those who were newly insured, and different subgroups have reacted differently. This suggests that the insurance design needs to be optimized and effective measures need to be adopted to help improve people’s lifestyle choices.


To improve the health of residents, China has launched a series of health reforms, including establishing three basic health insurance programs targeting different groups. Among these, the Urban Employee Basic Medical Insurance (UEBMI) and Urban Resident Basic Medical Insurance (URBMI) were implemented for urban residents in 1998 and 2007, respectively[1], while the New Rural Cooperative Medical Scheme (NRCMS) was implemented for rural residents in 2003[2].

A number of studies have investigated the effect of insurance in China on the use of medical services[3], and there is evidence that URBMI has improved self-reported health status to some extent and prompted greater utilization of medical services[46]. However, relatively few studies have investigated the effect of insurance on preventive care service utilization and health behaviors, especially in consideration of the differences between types of insurance and subsamples. This gap exists despite the fact that studies investigating the effects of NRCMS found that insured rural residents tended to smoke, drink, and engage in other risky behaviors more so than uninsured people in rural China [7, 8].

What exactly is the effect of insurance on the utilization of preventive care services and the adoption of healthy lifestyles? The evidence from developed countries is inconclusive. The well-known Rand Health Insurance Experiment found that health insurance had no significant effect on weight, physical activity, smoking, or alcohol consumption [9]. Another analysis based on a nationally representative sample found that insurance was not associated with significant changes in health behaviors but was associated with increases in preventive care [10]. Moreover, research conducted in the US has found that insurance strongly encouraged heavy smoking and sedentariness [11]. Similarly, research on Medicare confirmed a reduction in physical activity just before receiving Medicare [12]. A study based on the Portuguese Health Survey also found that holding voluntary private health insurance decreased the likelihood of engaging in sports [13]. Evidence from the UK, however, indicated that private health insurance did not reduce preventive activities such as exercise or regular checkups [14].

Regarding body mass index (BMI), one study, using data from the Behavioral Risk Factor Surveillance System, found that health insurance coverage reform in Massachusetts was associated with a reduction in BMI [15]. Other studies, however, have found that Medicaid expansion had a positive correlation with BMI among nonelderly adults or diabetics [16, 17].

Research focused on women has found that welfare reforms in the US increased drinking among single mothers while smoking and weight gain increased among pregnant women with the expansion of Medicaid [18, 19]. A study of elderly American males found that those who had recently obtained health insurance reduced preventive behaviors and increased unhealthy behaviors [20]. Among younger people, meanwhile, one study found that health insurance decreased heavy alcohol consumption among young adults while another found increased use of preventive care, such as checkups, for children [21, 22]. One cross-sectional study in Colombia found that insurance did not reduce preventive care among patients with diabetes [23].

Similar studies are fairly rare in the context of developing countries. One study found that Universal Health Coverage in Thailand had a positive effect on annual checkups while another found that the Seguro Popular Experiment in Mexico negatively affected the use of preventive care services [24, 25]. A study in Ghana investigated the relationship between insurance and malaria prevention and found that insured households were less likely to sleep under insecticide-treated bed nets [26].

The World Health Organization (WHO) has declared that lifestyle choices are important determinants of individual health [27]. However, with the expansion of basic health insurance in China, there is insufficient evidence to clarify the effect on urban residents’ health behaviors. The present study, therefore, investigated through longitudinal comparison whether health insurance reduced incentives to pursue preventive activities and healthy behaviors among urban residents in China.



Data were drawn from the China Health and Nutrition Survey (CHNS), which is an ongoing international collaborative project between the University of North Carolina and the Chinese Center for Disease Control and Prevention. The CHNS has had nine waves to date (1989, 1991, 1993, 1997, 2000, 2004, 2006, 2009, 2011), detailed descriptions of which can be found at the official website (

The data used in this study came from urban residents who completed the adult questionnaires in the 2006 and 2011 waves, comprising 3,360 and 5,361 adults, respectively, accounting for approximately one-third of respondents. Those who did not complete both survey waves were excluded, and a sample of 2,020 respondents was thus obtained. After processing for errors and missing data, we obtained 1,934 respondents. Based on the research objective, the following were also excluded: 749 respondents who initially had no insurance but had other basic medical coverage in 2011, 518 who changed insurance schemes or became uninsured during the survey period, and 64 who had other supplementary insurance schemes in both waves. Thus, a total sample of 603 respondents was obtained. According to their insurance situation, 378 respondents who had UEBMI for the entire period 2006–2011 were classified as the control group, while 225 respondents who had no basic medical insurance in 2006 but participated in URBMI before 2011 were classified as the intervention group.


Preventive care service utilization.

All respondents were asked if they had received any kind of preventive care service in the last four weeks. Preventive care services can include check-ups, visual activity tests, blood tests, hypertension examinations, tumor examinations, and so forth. Respondents were categorized into two groups according to their answers (yes or no).

Health behaviors.

Multiple indicators were selected to measure an individual’s health behaviors, including whether they currently smoked or drank alcohol or soft drinks. Physical activity was measured by a binary variable based on the respondent’s answer to the question of whether “he or she engages in activities such as Kong Fu, dancing, running, swimming, and playing ball games.” Sedentariness was classified according to the level of engagement in activities such as watching television, recreational computer use, online chatting, playing board games, or reading newspapers or magazines [7]. Meanwhile, a dummy variable indicating whether an individual’s BMI exceeded the WHO recommended threshold (BMI>25) was also created, based on the results of physical examinations conducted by doctors, nurses, or other health workers.

Other variables.

Age, gender, marital status, educational level, work status, annual household income, number of people in the household, and region were selected to control for natural and social characteristics. Whether the participants had been diagnosed with a chronic disease and their self-reported health status were selected to control for health status. Answers to the survey questions on chronic diseases relied on doctors’ diagnoses of hypertension, diabetes, myocardial infarction, or stroke. A value of 0 was assigned if none of those diseases was reported while 1 was assigned if at least one was reported. The respondents’ self-reported health was grouped into two categories according to their answers (good or poor).

Statistical analyses

Propensity score matching.

To estimate the effects of insurance on preventive care service utilization and health behaviors, it was necessary to distinguish between an intervention group (D = 1) that had experienced insurance transition and a control group (D = 0) that had not. Although the intervention group had been exposed to URBMI, the comparison sample remained heterogeneous given that insurance uptake is nonrandom. To solve this problem, propensity score matching (PSM)—proposed by Rosenbaum and Rubin as the probability of receiving a particular intervention given the site of the correlate Xi [28]—was used to balance the characteristics of insurance participation between the two groups [29].

Following this strategy, a probit model was used to regress the intervention status on all baseline correlates and indicate the probability of being insured with the propensity score [30, 31]. Various PSM techniques have been adopted in the literature [32]. Following the approaches described above, we adopted a 5-nearest neighbor (NN) matching method by matching each intervened individual with five individuals in the control group who were close to him or her in the propensity score, as this indicated the best balancing properties among the correlated variables [33]. To assess the quality of the matching, correlate balance was tested using two-sample t-tests for both groups before and after the match. S2 Table shows the results of the correlate balance check. The correlates were balanced in both groups and became statistically indistinguishable after matching, as suggested by Rosenbaum and Rubin [28].

Methods of estimating intervention effects.

The difference-in-differences (DID) method combined with PSM was used to estimate the effects of insurance on preventive activities and health behavior changes. Compared to the matching approach alone, the DID matching estimator relaxes the independence assumption between outcome and program participation, which means any bias caused by time-invariant unobserved systematic differences common to URBMI and UEBMI can be implicitly controlled [34].

For each outcome (Yi), the change in the intervention group and the change in the control group were compared during the study period to estimate the average treatment effect (ATT). The DID propensity score matching estimator was based on the following identifying assumption [35, 36]: (1) where D1(D0) represents the intervention(control) group, t(t+1) denotes the pre-(post-) intervention period, wij identifies the NN matching weight, and S stands for the area of common support.

The model above was estimated separately for each outcome to obtain the insurance effects. Standard errors were obtained by bootstrapping, and heterogeneity analysis was also performed by splitting the sample into subsamples. All PSM analyses were performed using Stata ado psmatch2 [37].


Descriptive analysis

Table 1 describes the characteristics and outcomes of the control and intervention groups in the two waves. In the 2006 wave, age was concentrated in the 40–60 range in the control group and >60 in the intervention group. Compared to participants in the control group, those in the intervention group were significantly less likely to be married, well educated, or employed with a high income but were more likely to have more family members (all Ps<0.01). However, the comparison of self-reported health and chronic disease presented no statistical difference on the baseline. In terms of behaviors, from 2006 to 2011, the probability of preventive care service utilization decreased in the intervention group compared to an increase in the control group, showing a significant difference in the 2011 wave. Meanwhile, individuals in the intervention group tended to increase their probability of drinking soft drinks and engaging in physical activity more than the control group after the insurance reform (P<0.05).

Influencing factors for residents participating in URBMI

Table 2 presents the probit estimates for the influencing factors among residents participating in URBMI. Among all sociodemographic characteristics, age, education, job status, and income were all significantly associated with being URBMI insured. As the table shows, there was a negative correlation between age and insurance, implying that younger residents were more likely to participate in URBMI (P<0.01). Compared to the control individuals, those with a higher educational level were less likely to participate in URBMI (P<0.001). In addition, urban residents without a job or with lower incomes were more inclined to participate in URBMI (P<0.001).

Table 2. Probit estimates of the probability of being URBMI insured.

Effect of insurance on preventive activities and health behaviors

After restricting the analyses to individuals in the common support range (0.017, 0.995), matching analyses were imposed restrictively on 550 respondents. Table 3 reports the main estimates of the effect of URBMI on preventive care service utilization and health behavior. For intervened individuals, the probability of sedentariness increased significantly by 5.1%, which is in line with the results for unmatched individuals. Being URBMI insured increased the likelihood of smoking by 1.2%, but the result was not statistically significant. The probability of using preventive care services, drinking alcohol, consuming soft drinks, performing physical activity, and being overweight did not change significantly after urban residents obtained URBMI.

Table 3. Average treatment effects on the intervention (ATT).

Effect of insurance on different subgroups

Table 4 presents the results for average treatment effects among different subgroups. Similar to the above analyses, the probability of sedentariness increased among those reporting both poor and good health (P<0.10). Meanwhile, residents with poor self-reported health reduced the probability of exercise by 27.5% (P<0.05). Females tended to be more sedentary (P<0.05), as did low- and middle-income individuals (P<0.10). With respect to soft drinks, male enrollees tended to be less likely to drink soft drinks by 22.6% (P<0.10). Regarding the probability of smoking, middle-income enrollees tended to be more likely to smoke (P<0.10), but no significant results were observed for the poor or rich.

Table 4. Average treatment effects in different subgroups.


This study aimed to reveal the effects of China’s urban basic health insurance on individuals’ behaviors. Using the CHNS and combining two approaches—difference-in-differences and propensity score matching—we assessed the effects of a basic insurance scheme through comparison with a previously uninsured group. The sample was composed of adults over 18 years of age. In addition, owing to the implementation of URBMI in 2007, we were able to focus on the completely uninsured group rather than beneficiaries of preexisting insurance schemes and could then distinguish between the uninsured group and those enrolled in insurance schemes (mainly UEBMI) [24].

The empirical results indicated that URBMI did not significantly change preventive care utilization. This does not support the findings of Baicker, who found a positive effect of Medicaid on the utilization of preventive services such as cholesterol screening, mammography, and prostate cancer screening in the Oregon Health Experiment [38]. Although such preventive measures are financed by the insurance scheme—which may incentivize prevention as it makes preventive care less costly [12, 39]—some studies have found that the demand for preventive care may be relatively inelastic, perhaps due to long waiting times or uncomfortable experiences [4042]. URBMI is a government-run voluntary insurance program targeting the prevention of serious illnesses by providing a service package for basic health and against catastrophic diseases. The effects of URBMI on preventive care use are probably negligible since the health insurance offers incomplete coverage [43].

Regarding the effects on unhealthy or risky behaviors, our analyses found no evidence that health insurance coverage increased soft drink consumption, physical activity, or obesity between the focus and control groups. Our results suggested a small but measurable problem that URBMI participation, on average, increased the probability of tobacco use by 1.2%. Although this result is not statistically significant, it should be a cause for concern since smoking is universally recognized as an unhealthy behavior. Our results also suggested an approximately 5% increase in the probability of sedentariness after participating in URBMI, which aligns with prior studies [1113]. Consistent with the literature, little association was found between sedentary behavior and physical activity in our study, most likely because of the different measurements [4446]. Our definition and operationalization of physical activity is consistent with most previous studies, comprising moderate to vigorous physical activity, while sedentary behavior was measured in accordance with the suggestion of the “Sedentary Behavior Research Network,” where sedentariness is distinguished from “inactivity” and describes waking behavior characterized by low-energy expenditure in a sitting or reclining posture [47]. Overall, the effect of health insurance on health behaviors is ambiguous. On the one hand, health insurance does not directly insure against health risks but only the financial consequences of illness; thus, it might not sufficiently incentivize engagement in healthy behaviors, which in turn incurs negative effects on health with cross-price effects [14, 17]. On the other hand, the insurance only reimburses a portion of treatment costs, and physical pain and opportunity cost continue to exist in the process of treatment and recovery. Meanwhile, health insurance can increase the likelihood of contact with physicians, who may advise patients to adopt healthy behaviors, which is likely to positively influence the health behaviors of insured persons [20, 48, 49].

Our main models included all adults, but we also estimated models separately by gender, self-reported health, and income because behavioral responses may vary by subpopulation. Prior studies have suggested that men and women may respond differently to insurance coverage [50]. According to our results, an increase of 7.1% for sedentariness was seen in the female group, which is in line with Qin’s results [7]. Among males, URBMI was associated with a negative effect on drinking soft drinks. In addition, residents with self-reported poor health tended to exercise less while those who reported good health were more likely to be sedentary. The results from our subgroup analysis by household income status are consistent with the theory that decreased work-related income can have a negative effect on health [51]. Our results suggest an elevated frequency of sedentariness among individuals who are not rich. Middle-income enrollees tended to be more likely to smoke, but no significant results were found for the poor or rich.

An important issue that should be noted in the interpretation of the results is that lost panel data could be a source of bias if the loss of follow-up was not random [52]. To address this issue, we compared the characteristics of follow-up status in the 2006 base wave. The results are presented in S1 Table. No significant differences were found in the outcomes—except for soft drink consumption—between follow-up individuals and those who were not followed up. Another issue pertains to the study’s methodology. Our analysis was based on panel data rather than cross-sectional data, which differs from prior studies but allows for the control of unobservable time-invariant factors. The combination of PSM and DID provides more robust results for estimating treatment effect due to removing biases caused by covariance [35]. The matching results are presented in S2 Table. Before the match, there were significant differences between the intervention and control groups in terms of gender, marriage, educational level, job status, and household income. After the propensity score matching, the differences between all variables in the intervention and control groups were no longer statistically significant. The results indicate that the PSM method can reduce the difference in observed characteristics before insurance participation. Restrictions on the range of common support substantially reduce differences in the observed variables and control for residual differences. Additionally, the DID model relaxes the PSM restrictions, making model-based adjustments less sensitive to other unobservable variables. This reduced sensitivity again facilitates the estimation of parametric approximations of ATT [53].

The policy implication of our findings is that measures should be adopted to encourage preventive activities and healthy behaviors. For this purpose, a health insurance benefit package that attaches preventive health care programs could be effective. Specifically, this could involve financial incentives such as removing cost sharing for preventive care and providing cash rewards or penalties, respectively, for decreases or increases in unhealthy behavior. Another possible alternative could be built on the peer effect. One study found that employees in the same company tended to join the same insurance schemes, implying that insurance companies may contribute to health choices by promoting healthy lunch menus or other related options [13]. Lastly, increasing access to doctors and reducing barriers to information could play a role in promoting healthy lifestyles. Not only are doctors expected to advise their patients in precautionary behaviors, but health promotion and education can also be used to promote healthy lifestyles.

This study has a number of limitations. First, the results have limited generalizability given the restrictions we placed on our sample. In particular, the restrictions we imposed on the types of health insurance led to a limited sample, which was mainly affected by the integration process in medical insurance reform during the survey period. Regarding the data, it should be noted that incomplete data resulted in follow-up loss, and objective measurements of health behaviors could have led to misreporting. Lastly, the long-term effects of URBMI remain unknown due to unavailable data. Despite these limitations, this study provides some empirical evidence regarding the effects of China’s urban basic health insurance on preventive care service utilization and changes in health behavior.


Using the data drawn from the CHNS during the period 2006–2011, we found that the utilization of preventive care did not change significantly after the URBMI reform. We also found that while sedentariness increased among urban residents, other unhealthy behaviors such as smoking and drinking did not, nor were there increases in obesity. In addition, different subsamples reacted differently. Female and low- or middle-income enrollees were more likely to be sedentary, people with poor health tended to exercise less, and middle-income enrollees were more likely to smoke. It is essential to increase awareness regarding the importance of preventive activities and healthy behaviors, especially among newly insured individuals.

Supporting information

S1 Table. Differences between follow-up and loss of follow-up samples.


S2 Table. Covariate balance results after propensity score matching.



The authors would like to thank the China Health and Nutrition Survey (CHNS) team for providing data.


  1. 1. Sun YL, Gregersen H, Yuan W. Chinese health care system and clinical epidemiology. Clin Epidemiol. 2017;9:167–78. Epub 2017/03/31. pmid:28356772
  2. 2. Wagstaff A, Lindelow M, Jun G, Ling X, Juncheng Q. Extending health insurance to the rural population: an impact evaluation of China's new cooperative medical scheme. J Health Econ. 2009;28(1):1–19. pmid:19058865.
  3. 3. Zhang C, Lei X, Strauss J, Zhao Y. Health insurance and health care among the mid-aged and older Chinese: evidence from the National Baseline Survey of CHARLS. Health Econ. 2017;26(4):431–49. pmid:26856894; PubMed Central PMCID: PMCPMC4980285.
  4. 4. Chen G, Liu GG, Xu F. The impact of the Urban Resident Basic Medical Insurance on health services utilisation in China. Pharmacoeconomics. 2014;32(3):277–92. pmid:24178373
  5. 5. Pan J, Lei XY, Liu GG. Health insurance and health status: exploring the causal effect from a policy intervention. Health Econ. 2016;25(11):1389–402. WOS:000386142000004. pmid:26350053
  6. 6. ZHOU JJ, ZHANG X, CAO Q. The impacts of urban residents' basic medical insurance on the medical treatment and preventive care service utilization. Chin J Health Pol. 2015;8(12):36–40.
  7. 7. Qin XZ, Lu TY. Does health insurance lead to ex ante moral hazard? Evidence from China’s New Rural Cooperative Medical Scheme. Geneva Papers Risk Insur Issues Pract. 2014;39(4):625–50.
  8. 8. Fu HQ, Yuan D, Lei XY. Health status and ex ante moral hazard of health insurance: an empirical investigation on China’s New Rural Cooperative Medical Scheme. Chin Econ Q. 2017;16(2):599–620.
  9. 9. Newhouse JP. Free for all?: Lessons from the Rand Health Insurance Experiment: Cambridge, MA: Harvard University Press; 1993.
  10. 10. Jerant A, Fiscella K, Tancredi DJ, Franks P. Health insurance is associated with preventive care but not personal health behaviors. J Am Board Fam Med. 2013;26(6):759–67. pmid:24204073.
  11. 11. Stanciole AE. Health insurance and lifestyle choices: identifying ex ante moral hazard in the US market. Geneva Papers Risk Insur Issues Pract. 2008;33(4):627–44.
  12. 12. de Preux LB. Anticipatory ex ante moral hazard and the effect of Medicare on prevention. Health Econ. 2011;20(9):1056–72. pmid:21830252.
  13. 13. Tavares AI. Health insurance and lifestyles. Appl Econ. 2014;46(16):1910–23.
  14. 14. Courbage C, Coulon Ad. Prevention and private health insurance in the UK. Geneva Papers Risk Insur Issues Pract. 2004;29(4):719–27.
  15. 15. Courtemanche CJ, Zapata D. Does universal coverage improve health? The Massachusetts experience. J Policy Anal Manag. 2014;33(1):36–69.
  16. 16. Kelly IR, Markowitz S. Incentives in obesity and health insurance. Inquiry. 2009;46(4):418–32. pmid:20184168
  17. 17. Klick J, Stratmann T. Diabetes treatments and moral hazard. J Law Econ. 2007;50(3):519–38.
  18. 18. Basu S, Rehkopf DH, Siddiqi A, Glymour MM, Kawachi I. Health behaviors, mental health, and health care utilization among single mothers after welfare reforms in the 1990s. Am J Epidemiol. 2016;183(6):531–8. pmid:26946395; PubMed Central PMCID: PMCPMC5013929.
  19. 19. Dave DM, Kaestner R, Wehby GL. Does medicaid coverage for pregnant women affect prenatal health behaviors? National Bureau of Econ Res, 2015.
  20. 20. Dave D, Kaestner R. Health insurance and ex ante moral hazard: evidence from Medicare. Int J Health Care Financ Econ. 2009;9(4):367–90. WOS:000271808300003. pmid:19277859
  21. 21. Yoruk BK. Health insurance coverage and risky health behaviors among young adults. B E J Econ Anal Policy. 2017;17(3):21. WOS:000406652200004.
  22. 22. Miller S. The impact of the Massachusetts health care reform on health care use among children. Am Econ Rev. 2012;102(3):502–7. WOS:000304262000088. pmid:29521491
  23. 23. Trujillo AJ, Vecino Ortiz AI, Ruiz Gomez F, Steinhardt LC. Health insurance doesn't seem to discourage prevention among diabetes patients in Colombia. Health Aff (Millwood). 2010;29(12):2180–8. pmid:21134918.
  24. 24. Ghislandi S, Manachotphong W, Perego VM. The impact of Universal Health Coverage on health care consumption and risky behaviours: evidence from Thailand. Health Econ Policy Law. 2015;10(3):251–66. pmid:25116081.
  25. 25. Spenkuch JL. Moral hazard and selection among the poor: evidence from a randomized experiment. J Health Econ. 2012;31(1):72–85. WOS:000302972200007. pmid:22307034
  26. 26. Yilma Z, van Kempen L, de Hoop T. A perverse "net" effect? Health insurance and ex-ante moral hazard in Ghana. Soc Sci Med. 2012;75(1):138–47. pmid:22507951
  27. 27. WHO. Health and development through physical activity and sport: Geneva: World Health Organization; 2003.
  28. 28. Rosenbaum PR, Rubin D, B. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70(1):41–55.
  29. 29. Zeng F, An JJ, Scully R, Barrington C, Patel BV, Nichol MB. The impact of value-based benefit design on adherence to diabetes medications: a propensity score-weighted difference in difference evaluation. Value in Health. 2010;13(6):846–52. pmid:20561344
  30. 30. Fu AZ, Dow WH, Liu GG. Propensity score and difference-in-difference methods: a study of second-generation antidepressant use in patients with bipolar disorder. Health Serv Outcomes Res Method. 2007;7(1–2):23–38.
  31. 31. Sari N, Osman M. The effects of patient education programs on medication use among asthma and COPD patients: a propensity score matching with a difference-in-difference regression approach. BMC Health Serv Res. 2015;15:332. pmid:26277920; PubMed Central PMCID: PMCPMC4537780.
  32. 32. Austin PC. Propensity-score matching in the cardiovascular surgery literature from 2004 to 2006: a systematic review and suggestions for improvement. J Thorac Cardiovas Surg. 2007;134(5):1128–35.
  33. 33. Caliendo M, Kopeinig S. Some practical guidance for the implementation of propensity score matching. Disc Papers Diw Berlin. 2008;22(1):31–72.
  34. 34. Lu JH. The performance of performance-based contracting in human services: a quasi-experiment. J Public Admin Res Theory. 2016;26(2):277–93.
  35. 35. Heckman JJ, Ichimura H, Todd PE. Matching as an econometric evaluation estimator: evidence from evaluating a job training programme. Rev Econ Stud. 1997;64(4):605–54.
  36. 36. Heckman J, Ichimura H, Smith J, Todd P. Characterizing selection bias using experimental data. Econometrica. 1998;66(5):1017–98.
  37. 37. Leuven E, Sianesi B. PSMATCH2: Stata module to perform full Mahalanobis and propensity score matching, common support graphing, and covariate imbalance testing: Statistical Software Components S432001, Boston College Department of Economics; 2017.
  38. 38. Baicker K, Taubman SL, Allen HL, Bernstein M, Gruber JH, Newhouse JP, et al. The Oregon experiment: effects of Medicaid on clinical outcomes. N Engl J Med. 2013;368(18):1713–22. pmid:23635051; PubMed Central PMCID: PMCPMC3701298.
  39. 39. Pagán JA, Puig A, Soldo BJ. Health insurance coverage and the use of preventive services by Mexican adults. Health Econ. 2007;16(12):1359–69. pmid:17334977
  40. 40. Aron-Dine A, Einav L, Finkelstein A. The RAND health insurance experiment: three decades later. J Econ Perspect. 2013;27(1):197–222. WOS:000314799800010. pmid:24610973
  41. 41. Ringel J, Hosek SD, Vollaard BA, Mahnovski S. The elasticity of demand for health care: Santa Monica: Rand Corporation; 2002.
  42. 42. Anderson RT, Camacho FT, Balkrishnan R. Willing to wait? The influence of patient wait time on satisfaction with primary care. BMC Health Serv Res. 2007;7(1):31.
  43. 43. Zweifel P, Manning WG. Moral hazard and consumer incentives in health care. In: Chapter 8: Amsterdam: Elsevier; 2000. 409–59 p.
  44. 44. Van DHK, Paw MJ, Twisk JW, Van MW. A brief review on correlates of physical activity and sedentariness in youth. Med Sci Sports Exerc. 2007;39(8):1241–50. pmid:17762356
  45. 45. Owen N, Healy GN, Matthews CE, Dunstan DW. Too much sitting: The population-health science of sedentary behavior. Exerc Sport Sci Rev. 2010;38(3):105–13. pmid:20577058
  46. 46. Ekelund U, Brage S, Froberg K, Harro M, Anderssen SA, Sardinha LB, et al. TV viewing and physical activity are independently associated with metabolic risk in children: the European youth heart study. PloS Med. 2006;3(12):e488. pmid:17194189
  47. 47. Network SBR. Letter to the editor: standardized use of the terms “sedentary” and “sedentary behaviours.”. Ment Health Phys Act. 2013;6(1):55–6.
  48. 48. Ehrlich I, Becker GS. Markt insurance, self-insurance, and self-protection. J Polit Econ. 1972;80(4):623–48. WOS:A1972N040700001.
  49. 49. Simon K, Soni A, Cawley J. The impact of health insurance on preventive care and health behaviors: evidence from the first two years of the ACA Medicaid expansions. J Policy Anal Manag. 2017;36(2):390–417.
  50. 50. Barbaresco S, Courtemanche CJ, Qi Y. Impacts of the affordable care act dependent coverage provision on health-related outcomes of young adults. J Health Econ. 2015;40:54–68. pmid:25594956
  51. 51. Riumalloherl C, Basu S, Stuckler D, Courtin E, Avendano M. Job loss, wealth and depression during the Great Recession in the USA and Europe. Int J Epidemiol. 2014;43(5):1508–17. pmid:24942142
  52. 52. Cheng L, Liu H, Zhang Y, Shen K, Zeng Y. The impact of health insurance on health outcomes and spending of the elderly: evidence from China's New Cooperative Medical Scheme. Health Econ. 2015;24(6):672–91. pmid:24777657; PubMed Central PMCID: PMCPMC4790431.
  53. 53. Ho DE, Kosuke I, Gary K, Stuart EA. Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference. Polit Anal. 2007;15(3):199–236.