Are health care scams infectious? Empirical evidence on contagion in health care fraud

Goel, Rajeev K. Article — Published Version Are health care scams infectious? Empirical evidence on contagion in health care fraud Managerial and Decision Economics Provided in Cooperation with: John Wiley & Sons Suggested Citation: Goel, Rajeev K. (2020) : Are health care scams infectious? Empirical evidence on contagion in health care fraud, Managerial and Decision Economics, ISSN 1099-1468, Wiley, Hoboken, NJ, Vol. 42, Iss. 1, pp. 198-208, https://doi.org/10.1002/mde.3224 This Version is available at: https://hdl.handle.net/10419/233680 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/ RESEARCH ARTICLE Are health care scams infectious? Empirical evidence on contagion in health care fraud Rajeev K. Goel 1,2 1 Department of Economics, Illinois State University, Normal, Illinois, USA 2 Knowledge Creation and Growth, Kiel Institute for the World Economy, Kiel, Germany Correspondence Rajeev K. Goel, Department of Economics, Illinois State University, Normal, IL 61790-4200, USA. Email: [email protected] This paper examines the presence of contagion in health care fraud across jurisdictional boundaries. Using state-level data for the United States, we find evidence of contagion in medical fraud. There are also spillovers from border corruption on medical fraud, but no evidence of spillovers from international borders. In other findings, greater urbanization, greater elderly population, and higher hospital occupancy positively contribute to medical fraud, while nursing employment has a mitigating effect. Further, it is economic inequality rather than economic prosperity that seems relevant. The main findings are robust to consideration of simultaneity, but dependent upon the prevalence of fraud across states. JEL CLASSIFICATION I11; K42 1|INTRODUCTION There has been increasing attention to health care provision and costs of providing health services in recent years, as individuals and governments have struggled to maintain and increase medical coverage. With this focus has come the realization that a significant contributor to the increasing health care costs has been the increase in fraudulent claims and other abuses of health services (Beaton, 2017; Kalb, 1999; Mackey, Vian, & Kohler, 2018; Rudman, Eberhardt, Pierce, & HartHester, 2009; Stelfox & Redelmeier, 2003; Sparrow, 2000). Combating health care fraud and conserving government resources devoted to health service gains added importance as governments worldwide try to deal with the COVID-19 crisis. Accordingly, researchers have started devoting attention to the drivers of health fraud, and governments have begun focusing on greater accountability and monitoring of the health sector (see Drabiak & Wolfson, 2020; Kang, Hong, Lee, & Kim, 2010; Krause, 2004; Rashidian, Joudaki, & Vian, 2012). Fraud schemes range from an individual's dishonest activity to broad-based operations by an institution or group. Examples include (a) health care providers knowingly billing for services and/or supplies that were not provided; (b) health care providers intentionally inflating bills with more expensive services than those provided; (c) health care recipients (patients) claiming exemptions for health care costs that they are not entitled to; and (d) someone using another patient's health credentials to access medical care, supplies, or equipment (http://www.ehfcn.org/what-is-fraud/). The wide prevalence of health care fraud has prompted the FBI to launch a website devoted to reports of health fraud news (https://www. fbi.gov/investigate/white-collar-crime/health-care-fraud/health-carefraud-news). The prevalence and costs of health fraud prompted lawmakers in the United States in 1996 to enact the Health Insurance Portability and Accountability Act of 1996 (HIPAA). The Act established a comprehensive program to combat health care fraud (https://oig.hhs.gov/reports-and-publications/hcfac/index.asp). However, the continuing reports of health care fraud losses suggest that more needs to be done. One oversight could be a lack of sufficient attention to crossborder spillovers or contagion effects in health fraud. Almost exclusively, the related formal analyses of health fraud have focused on the Received: 29 January 2020 Revised: 3 August 2020 Accepted: 10 August 2020 DOI: 10.1002/mde.3224 This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2020 The Authors. Managerial and Decision Economics published by John Wiley & Sons Ltd 198 Manage Decis Econ. 2021;42:198–208.wileyonlinelibrary.com/journal/mde factors that affect health care fraud or scams within a given jurisdiction (i.e., a country, city, or state), and little attention is paid to spillovers/influences from neighboring jurisdictions. With greater globalization, triggered in large part by advances in transportation and communications technologies, there is an increasing likelihood of spatial spillovers, both legal and illegal. For instance, doctors and medical professionals can often work in different jurisdictions, large corporations can own/manage medical facilities in multiple states/nations, and the agglomeration of data and patient records via the internet in one location can potentially be related to information gathered from multiple locations. All this has implications for both positive and negative externalities, and the contagion or spatial spillovers of medical fraud is one example of such negative externalities. Fraudsters in neighboring jurisdictions, with different monitoring and punishments for fraud, might defraud health providers in a given state or a large case of fraud in a state might trigger copycat crimes in neighboring states. This contagion in medical fraud forms the focus of this work. With the large, and in some nations almost exclusive, involvement of the government in the provision of health care, bureaucratic corruption and rent-seeking can be quite pervasive in health care and intertwined with health fraud. Fraudsters might bribe corrupt officials (or corrupt officials themselves approach fraudsters) to increase payoffs (in which case corruption would be complementary to fraud) or mitigate punishments (see Mackey et al., 2018). Again, such relations might spill over jurisdictional boundaries. Whereas there is evidence of contagion in other white-collar crimes (e.g., corruption; Goel & Nelson, 2007), contagion in health fraud has not been formally studied. Besides contributing to the literature, this research will contribute to the broader health policy questions regarding policy coordination across jurisdictions to fight fraud in the health system. Key questions addressed in this research are: •What are the significant drivers of health care scams? •Is there contagion or spillovers of health care scams across territorial borders? •Are corrupt acts complementary to health scams? Specifically, this paper contributes to the literature on health care fraud by examining the presence of contagion in fraud across jurisdictional boundaries. Using state-level data for the United States, we find evidence of contagion in fraud across states. 1 There are also spillovers from border corruption on medical fraud, but no evidence of spillovers from international borders. In other findings, greater urbanization rates, greater elderly population, and higher hospital occupancy rates positively contribute to medical fraud, while nursing employment (but not physician employment) has a mitigating effect on fraud. The layout of the rest of the paper includes the theoretical background and the model in the next section, followed by data and estimation, results, and conclusions. 2|THEORETICAL BACKGROUND, LITERATURE, AND MODEL 2.1 |Theoretical background and literature This research ties to and draws on several streams—the economics of white-collar crimes, health economics, and spatial economics. First, with respect to the economics of crime, the main formal theoretical arguments can be seen as rooted in the work of Becker (1968). Becker claimed that criminals and lawbreakers also act rationally by considering the potential benefits and costs of their illegal acts. Taking this view, the fraudsters in health fraud would also be acting rationally, and the benefit–cost of potential abusers in neighboring jurisdictions would likely be different than resident fraudsters, depending, among other things, upon the differences in relative monitoring and punishments across jurisdictions. Second, the health economics literature has been concerned with analyzing the benefits and funding of health services, with an increasing focus on cost-containment over time (Berwick & Hackbarth, 2012; Morris, 2009; Rai, 2001). Within the cost containment focus has been the attention on limiting or eliminating health care fraud (see Drabiak & Wolfson, 2020; Fan, Zhang, & Fan, 2019; Savino & Turvey, 2018). In this overall spectrum of scholarship, there has been limited or no formal attention to spatial spillovers of health care fraud and the present work attempts to make a contribution. A related strand of the literature talks about the information asymmetries that are unique to the medical profession. The accompanying moral hazard issues then lead to greater chances of fraud by parties with superior or proprietary information (Crea, Galizzi, Linnosmaa, & Miraldo, 2019; Hyman, 2001; McGuire, 2000). We account for this aspect by including the employment of medical personnel. Third, the subfield of spatial economics in all different contexts has come to the fore more recently, with researchers focusing on both positive spillovers (e.g., knowledge spillovers) and negative spillovers (e.g., corruption and smuggling of cigarettes) across jurisdictional boundaries (Goel & Nelson, 2007; Goel & Saunoris, 2014). For example, Goel and Nelson (2007) find positive spillovers from neighboring states' corruption on corruption in a given U.S. state, whereas Goel and Saunoris' (2014) international study finds the presence of contagion in corruption and the shadow economy. With increasing economic and cultural integration, spatial influences in many other aspects seem relevant, and this work examines such effects with regard to health care fraud. Spatial spillovers might occur due to formal and informal exchanges of neighboring communities and via demonstration effects through media reports. The presence of spatial spillovers or contagion in medical fraud has not been formally studied in the literature. Finally, this research can be seen as contributing to the effects of corruption, with a focus on health economics. Dimant and Tosato (2018) provide a review of the larger empirical literature on GOEL 199 the causes and effects of corruption, whereas SommersguterReichmann, Wild, Stepan, Reichmann, and Fried (2018) provide a more specific review of the corruption in health care literature. Corruption can reduce the potential costs of health care fraud by reducing punishment (e.g., when corrupt judges can be bribed to hand out reduced sentences/fines or when the probability of conviction is reduced) and apprehension for fraudsters, while at the same time, it might increase the benefits of fraud (by, for instance, increasing the avenues or opportunities for health fraud). Various scholars, for example, Berger (2014) and Mackey et al. (2018), have studied individual dimensions of corruption in health care (also see Transparency International, 2016). Some studies have taken a somewhat broader view by examining the link between overall insurance fraud and corruption (Goel, 2014). However, our treatment of corruption and its spatial effects on health fraud appears somewhat unique. We move next to a discussion of our formal analysis. 2.2 |Model Based on the above discussion, we can formulate two testable hypotheses. Hypothesis 1. There are positive spillovers from cross-border contagion of health care scams. Hypothesis 2. Greater corrupt activities would, ceteris paribus, be associated with more health care scams. The general form of our estimated equation to test the above hypotheses and answer the questions posed is the following (with subscript idenoting a state): HealthSCAMi=fðContagionij,CORRUPTIONi,Medical sector supply characteristicsim,URBANi,POP65plusi, NoINSUREi,GDPi,GINIiÞ ð1Þ i= 1, 2, …,51 j= HealthSCAMbor, CORRUPTIONbor, CAbor, MEXbor m= PHYSICIANS, NURSES, HOSPocc Our dependent variable is the instances of health care scams or fraud in a given state in the year 2015 (HealthSCAM). Details about what constitutes such fraud are provided in the data section. In our sample, there was considerable variation in medical fraud across states, ranging from a high of 49 cases to a low of zero, with an average per state of about six cases. 2 The main focus and novelty of this work pertains to the consideration of the contagion effects from neighboring states' medical fraud. Accordingly, we include the average medical fraud cases in all the states geographically bordering a given state (HealthSCAMbor). 3 A positive and statistically significant coefficient on HealthSCAMbor would support Hypothesis 1. Another dimension of contagion is accounted for including dummy variables, CAdum and MEXdum, for states bordering Canada and Mexico, respectively. For example, MEXdum takes the value of one for California, Arizona, New Mexico, and Texas, and zero otherwise. Border states face disproportionate externalities (positive or negative) from foreign information and exchanges than states elsewhere. The characteristics of the health system could crucially impact fraud and we consider three dimensions: the employment of physicians and nurses and hospital occupancy rates. Medical professionals, having firsthand information and access to medical records, can prove to be effective sentinels against fraud, although there is evidence that some professionals might themselves be perpetrating fraud (e.g., inflating insurance claims or inflating medical procedures performed; Jesilow, Pontell, & Geis, 1993). The separate consideration of physicians and nurses enables us to consider the qualitative differences between the two, for example, differences in access to patient records and in the authority to order medical procedures. Furthermore, hospital occupancy rates capture the competition for favors and, other things being the same, congestion in the provision of medical services would engender fraud. We also consider possible spillovers from other crimes by including the level of corruption as a regressor. The presence of corruption in a state could impact both the costs and benefits of medical fraud. For example, corruption could increase payoffs and lower costs of fraud (by reducing the chances of getting caught or reducing punishment upon being apprehended), both of which would support Hypothesis 2. The chances of corrupt government officials facilitating health fraud are greater when certain health programs, such as Medicaid and Medicare, are administered by the government, as in the United States (see Jesilow et al., 1993; Savino & Turvey, 2018; Sommersguter-Reichmann et al., 2018). In addition to possible impacts from own corruption, we also consider spatial spillovers from corruption by including CORRUPTIONbor. Again, corrupt acts in a state might be triggered by learning from border states or a single corruption scandal might involve multiple states (see Goel & Nelson, 2007). The elderly (POP65plus) and those without health insurance might turn out to be attractive targets for fraudsters, with the possibility that some of those without health insurance might themselves commit fraud (e.g., trying to file insurance claims in fraudulent names). Greater urbanization rates are associated with greater information flows and the relative ease with which fraudsters could hide their identities. We take account of both the level of economic prosperity (GDP) and the level of economic disparity (GINI). Greater economic prosperity would be associated with heightened checks and balances and increasing the opportunity costs of crime on the one hand and with making crime more lucrative on the other hand. 4 Greater economic disparity might increase the search costs for criminals in identifying economically lucrative targets while at the same time increasing the incentives of those less well-off to commit crimes. 200 GOEL 3|DATA AND ESTIMATION 3.1 |Data All the data for this study come from reputed sources that are available in the public domain. The main novelty of the paper stems from the availability of the health scams data by U.S. states. According to the data source, U.S. Department of Justice, 2015 Internet Crime Report, https://pdf.ic3.gov/2015_IC3Report.pdf, health scams entail: “A scheme attempting to defraud private or government health care programs, which usually involve health care providers, companies, or individuals. Schemes may include offers for (fake) insurance cards, health insurance market place assistance, stolen health information, or various other scams and/or any scheme involving medications, supplements, weight loss products, or diversion/pill mill practices. These scams are often initiated through spam email, Internet advertisements, links in forums/social media, and fraudulent websites”(p. 228). The above detail shows the multidimensional nature of medical scams, which may prevail differently across states. Unfortunately, this rich state-level detail on health scams comes with the limitation that it is available for a single year, 2015. This then limits the time-series dimension of this study. Yet, the variation across states and their number provide sufficient explanatory power to derive meaningful empirical results. Instances of medical fraud caught by the FBI cover a range of fraudulent activities, including false claims and false billing. 5 Furthermore, a given fraud identified/caught in a year might have been going on for a number of years. The medical fraud victims in a state in 2015 ranged from 0 to 49, with California having the maximum number of cases, and more than half a dozen states with no victims in that year. 6 On average, there were about six victims of medical fraud per state, with three states (Maryland, Oregon, and South Carolina) sitting at the mean. This shows that medical scams were not confined to a specific geographic region in the nation. 7 On average, about 9% of the state population was without health insurance, whereas the average elderly percent of a state's population was about 15 and the average urbanization rate was 72%. Further, the average convictions from corruption were about 17 per state. This measure of corruption across U.S. states has been widely used in the literature (see, for example, Goel & Nelson, 2011). Table 1a provides details about the variables and related summary statistics, whereas Table 1b includes a correlation matrix of key variables of interest. The correlation between HealthSCAM and CORRUPTION in our sample is −0.3. The following subsection discusses the estimation techniques employed to the data to estimate the equation above. 3.2 |Estimation Given the cross-sectional nature of our data with observations for the 50 states and the District of Columbia, some outlier state(s) (e.g., a state with no health scams or ones with unusually high scams) could TABLE 1a Variable definitions, summary statistics, and data sources Variable Definition (mean; std. dev.) Source HealthSCAM Health care scam victims by state, per capita (8.29e−07; 6.55e−07); or on average 0.08 victims per 100,000 population in a state. a HealthSCAMbor Average HealthSCAM in states bordering a given state (1.86e−06; 2.48e−06); or on average 0.19 per 100,000 population CORRUPTION Corruption convictions per capita (2.85e−06; 2.32e−06) b CORRUPTIONbor Average CORRUPTION in states bordering a given state (6.80e−06; 0.00001) PHYSICIANS Number of physicians per 100,000 state population (252.51; 77.86) c NURSES Number of nurses per 100,000 state population (785.64; 170.39) NoINSURE Percent of state population without health insurance (8.72; 3.17) d HOSPocc Number of nursing home residents per 100 nursing home beds, (80.56; 8.76) e GDP Real state GDP per capita, in millions of 2009 $, 2014 (0.05; 0.02) f GINI Gini ratio, measuring income inequality (0.61; 0.04) g URBAN State urbanization rate (%), (72.25; 15.28) h POP65plus Percent of state population over the age of 65 (0.15; 0.02) POLICE Number of police personnel (state and local), per capita (0.003; 0.0005) h, i CAbor Dummy variable identifying states bordering Canada (0.20; 0.40) MEXbor Dummy variable identifying states bordering Mexico (0.08; 0.27) Note. All data are annual, by state, for the year 2015 (unless otherwise specified). a U.S. Department of Justice, 2015 Internet Crime Report, https://pdf.ic3. gov/2015_IC3Report.pdf b U.S. Department of Justice, Public Integrity section, https://www.justice. gov/criminal/pin c American Medical Association, Chicago, IL, Physician Characteristics and Distribution in the U.S., 2015 Edition, annual d U.S. Census Bureau, American Community Survey, https://data.census. gov/cedsci/ e U.S. National Center for Health Statistics, Health, United States, 2016, May 2017, http://www.cdc.gov/nchs/hus.htm f U.S. Department of Commerce, Bureau of Economic Analysis g http://www.shsu.edu/eco_mwf/inequality.html h Statistical Abstract of the United States i U.S. Census Bureau, American Community Survey, https://data.census. gov/cedsci/ GOEL 201 impact the results. 8 To address this possibility, we employ robust regression that is less sensitive to outlying values. Later, we also employ 2SLS regression and quantile regression to address different econometric issues and answer the questions posed above. The results section follows. 4|RESULTS 4.1 |Drivers of health care scams: Baseline models Our baseline models in Table 2 examine the determinants of health care scams, without considering the border or contagion effects. The four models presented consider different combinations of possible drivers of health scams, borrowing from the literature (Goel, 2020) and the related discussion above. The overall fit of the three models is decent as shown by the R 2 s and the statistically significant F-values. Further, the variance inflation factors (VIFs) are well below the usual cutoff of 10, which alleviates concerns about multicollinearity. Relatively speaking, Model 2.4 has the best fit, although main findings across the four models are consistent. Greater urbanization rates, greater elderly populations or populations without insurance, and higher hospital occupancy rates lead to health care fraud, whereas more nurses and greater income inequality have reverse effects. The effects of urbanization rates and hospital occupancy can be seen as greater demand or competition for services that lead to fraud, while the elderly and those lacking health insurance are potentially vulnerable population subgroups that could be relatively easy targets of fraudsters. The congestion associated with increased demand for hospital beds tends to facilitate medical scams. To better nest the analysis with Becker's (1968) crime and punishment framework, Model 2.2 adds police employment per capita (POLICE) as a regressor to the setup of Model 2.1. The resulting coefficient on POLICE ends up being statistically insignificant. A plausible explanation is that controlling health care scams is unlikely to rank highly on the enforcement mandates of most police personnel. 9 Greater income inequalities might lower fraud due to the relatively wealthy have greater vigilance (due to greater education and/or better monitoring) on the one hand and greater costs for potential fraudsters in identifying potential victims in an economically unevenly distributed population. Both these effects would tend to lower health fraud. Furthermore, greater employment of nurses tends to lower health fraud, consistent with nurses providing effective vigilance given their TABLE 1b Correlation matrix of key variables HealthSCAM HealthSCAMbor CORRUPTION CORRUPTIONbor HealthSCAM 1.00 HealthSCAMbor −0.02 1.00 CORRUPTION −0.27 0.21 1.00 CORRUPTIONbor −0.09 0.94 0.36 1.00 Note. SeeTable 1a for variable details. Observations: 49. TABLE 2 Drivers of health care scams: Baseline models (dependent variable: HealthSCAM) 2.1 2.2 2.3 2.4 GDP 2.33e−06 (0.3) −0.00001 * (1.8) 5.49e−06 (0.6) −8.75e−06 (1.2) URBAN 1.61e−08 ** (2.7) 1.87e−08 ** (4.0) 2.56e−08 ** (3.5) 2.86e−08 ** (4.3) NURSES −1.58e−09 ** (2.6) −1.61e−09 ** (3.4) −1.57e−09 ** (2.3) −1.20e−09 ** (2.0) PHYSICIANS −7.73e−10 (0.5) −1.23e−09 (0.9) −1.51e−09 (0.8) 2.03e−09 (1.1) POP65plus 0.00001 ** (2.2) 9.71e−06 ** (2.6) 0.00002 ** (2.9) 0.00001 ** (2.1) POLICE 0.0001 (0.5) GINI −2.91e−06 (1.1) −6.27e−06 ** (2.4) NoINSURE 6.26e−08 * (1.9) HOSPocc 1.45e−08 * (1.7) N51 50 51 51 R 2 0.33 0.36 0.36 0.38 F-value 5.93 ** 9.83 ** 5.61 ** 5.10 ** VIF 2.7 1.6 2.6 2.7 Note. See Table 1a for variable definitions. Constant included but not reported in these robust regressions. The numbers in parentheses are absolute t statistics. * Statistical significance at the 10% level. ** Statistical significance at the 5% (or better) level. 202 GOEL greater time (relative to time of physicians) with patients and related medical records. Physicians' focus is primarily on diagnosis, while nurses have a greater attention to medical records. This result supports earlier findings in the literature (Goel, 2020). Finally, the impacts of economic prosperity (GDP) and physician employment are statistically insignificant. Next, we consider the contagion or cross-border spillovers. 4.2 |Contagion in health care scams The contagion effects in health scams, and the main novelty of this research is considered in Table 3. Are there significant spillovers across states from the prevalence of health care fraud? Accordingly, the baseline Model 2.4 is enhanced by introducing the HealthSCAMbor variable that captures the average scams in all the states bordering a given state (e.g., for Florida, HealthSCAMbor would be average of HealthSCAM in Georgia and Alabama). From Table 1b, the correlation between HealthSCAM and HealthSCAMbor is close to zero and the formal analysis in Table 3 tests the strength of this relation when other relevant factors have been accounted for. Results show a statistically strong contagion effect of border health scams—the coefficient on HealthSCAMbor is statistically significant in both the models reported in Table 3. This result supports Hypothesis 1 outlined above. Border scams might be due to learning in other states, or due to multistate holdings of medical facilities, and so forth. Numerically, the elasticity of HealthSCAM with respect to HealthSCAMbor (evaluated at respective means) is 0.18. Whereas this elasticity of health scams in a state with respect to border health scams is somewhat modest, the monetary impact or loss could be substantial. Although firm data on the magnitude of health care fraud losses are hard to come by, according to the National Health Care Anti-Fraud Association (https://www.nhcaa.org/ resources/health-care-anti-fraud-resources/the-challenge-of-healthcare-fraud/), in 2018 $3.6 trillion was spent on health care in the United States and the estimated losses due to fraud range from 3% to 10% of total spending. There would, of course, be significant variations across individual states. Not only the presence of contagion effects is a novel contribution to the health economics literature, it has the important policy recommendation in suggesting a need for regional health policy coordination to combat health care fraud. The other findings are quite consistent with Model 2.4—greater urbanization, the proportion of the elderly, and higher hospital occupancy rates contribute to health care fraud, while more nursing employment checks it. Again, the robustness of the mitigating effect of nurses is a noteworthy finding. Further, health care fraud in more prosperous states, states with more physicians, states with greater income inequality, and those with a larger share of the population with health insurance was no different from other states. As another aspect of contagion, Model 3.2 also includes dummy variables for the states bordering Canada (CAdum) and Mexico (MEXdum). The consideration of international-bordering states enables us to control for contagion or spillovers of health scams from Canada and Mexico. Are health scams in states with international borders different from other states? The international border effects are found to be insignificant. 4.2.1 |Considering possible endogeneity of border scams There could be bidirectional causality between health scams in a state and scams in bordering states. To account for these influences, we reestimated Model 3.1 using 2SLS and instrumenting HealthSCAMbor with state size (population) and the share of the disabled population. The logic being that larger states would, ceteris paribus, have more interactions/information flows with neighbors. 10 The tests of exogeneity showed HealthSCAMbor to be exogenous—both the Wu– Hausman Ftest of exogeneity and the Durbin–Wu–Hausman chisquare test of exogeneity could not reject the exogeneity of HealthSCAMbor. 11 4.3 |Additional consideration 1: Corruption, corruption contagion, and health care scams It is quite possible that the degree of corruption in state is linked to health care scams, where the presence of corruption promotes health care fraud both by potentially increasing the fraud payoff (making fraud potentially more lucrative) and by reducing the expected costs (punishment). Broadly speaking, the consideration of corruption can be seen as an indicator of the weakness of institutions. Some studies in the literature have recognized the linkage between corruption and the health sector (Berger, 2014; Vian, 2008). This consideration also TABLE 3 Contagion in health care scams (dependent variable: HealthSCAM) 3.1 3.2 HealthSCAMbor 0.08 ** (2.4) 0.08 ** (2.4) GDP −3.15e−06 (0.4) −2.92e−06 (0.4) URBAN 3.64e−08 ** (5.4) 3.69e−08 ** (4.5) NURSES −1.17e−09 * (1.9) −1.28e−09 * (1.9) PHYSICIANS −2.31e−09 (1.3) −2.53e−09 (1.4) POP65plus 0.00001 * (2.0) 0.00001 * (1.8) GINI −5.22e−06 * (1.9) −4.69e−06 (1.6) NoINSURE 5.33e−08 (1.6) 5.36e−08 (1.4) HOSPocc 2.58e−08 ** (2.8) 2.59e−08 ** (2.7) CAbor −1.02e−07 (0.5) MEXbor −3.42e−07 (1.2) N49 49 R 2 0.49 0.49 F-value 6.12 ** 4.76 ** Note. SeeTable 2. GOEL 203 adds to the broader literature on the effects of corruption (Dimant & Tosato, 2018; Prasad, da Silva, & Nickow, 2019). Our estimation results in Table 4, alternatively including CORRUPTION and CORRUPTIONbor, as additional regressors to the format of the models in Table 3, consider both the own and border effects of corruption. Contagion in corrupt acts in other contexts has been shown to be significant (Goel & Nelson, 2007). We find that while own corruption is statistically insignificant, border corruption (CORRUPTIONbor) is statistically significant in Model 4.2 (at the 10% level). In other words, greater corruption in states bordering a given state tends to promote health fraud. The relative significance of border corruption makes sense when one thinks that participants in corrupt exchanges in border states (i.e., bribe takers or bribe givers) are less susceptible to apprehension/prosecution in a given state. In other words, corrupt acts in a border state are not directly under the purview of a state's law enforcement, and yet border corruption can encourage health care crimes in a given state (by, for example, making it easier to stash fraudulent earnings across state borders). 12 In terms of the magnitude of such an impact, a 10% increase in corruption in bordering states (i.e., per capita convictions for the abuse of public office per Table 1a), would increase health care fraud by about 0.8%. Thus, there is contagion from corruption, although the magnitude of such an impact is not too substantial. With regard to corruption, there appears mixed support for Hypothesis 2—it holds with regard to border effects, but not with regard to own corruption. A policy consequence of these findings is that corruption control, especially regional corruption control, would have payoffs in terms of reducing health care fraud. The results for the other drivers of health scams are consistent with what was reported earlier, except that greater income inequality (GINI) is now associated with less health care scams (consistent with Model 2.4). Thus, while economic prosperity (GDP) has an insignificant effect, income disparity tends to matter. 4.3.1 |Considering possible endogeneity of corruption It is possible that the nexus between corruption and health scams could flow in both directions—health care fraud could impact the level of corrupt activity, whereby fraudsters bribe government officials to avoid apprehension or punishment (see Goel & Rich, 1989). To address this possibility, we re-estimated Models 4.1 and 4.2, alternately taking CORRUPTION and CORRUPTIONbor to be endogenous in 2SLS regressions. The instruments employed for the corruption variables were police employment per capita and the degree of economic freedom in a state. 13 Greater economic freedom is associated with fewer regulations that limit rent-seeking opportunities for corrupt bureaucrats. In both cases, the tests of exogeneity of CORRUPTION and CORRUPTIONbor, respectively, could not reject exogeneity. 14 As another consideration of simultaneity, the CORRUPTION variable was included with a two-year lag (i.e., for 2013). The coefficient on lagged corruption was again statistically insignificant. 15 4.4 |Additional consideration 2: Drivers of health care scams across the prevalence of scams: Quantile regression To gain further insights into the drivers of health care scams and to determine if the efficacy of certain influences varies across the prevalence of scams, we estimate a quantile regression to the baseline Model 2.4. 16 The corresponding results are presented in Table 5a, with q20, q50, and q80, respectively, denoting 20th percentile, the median, and 80th percentile. 17 The overall fit the three models in Table 5a, denoted by the pseudo-R 2 s is at least 0.32 and in line with the R 2 s reported in other estimations. Are states with a low (q20) or a high prevalence (q80) of health scams differently affected by the factors that influence them? The overall findings are supportive of Model 2.4 in terms of factors that impact health scams. However, the efficacy of the influences varies across states with varying prevalence. In fact, no factor significantly drives health scams at the low end of the distribution (q20), while most are significant at the high end of the distribution (q80). Specifically, states with a greater share of the elderly and states with higher hospital occupancy rates experience greater health scams when the prevalence of such scams is already high (Model 5.3). Similarly, greater employment of nurses exerts a countervailing power only in such states. The degree of urbanization, on the other hand, increases health scams in the median and higher prevalence states. One implication of these findings is that states with a low prevalence TABLE 4 Corruption, corruption contagion, and health care scams (dependent variable: HealthSCAM) 4.1 4.2 CORRUPTION −0.05 (1.1) CORRUPTIONbor 0.01 * (2.0) GDP 6.31e−06 (0.7) −6.23e−06 (0.9) URBAN 2.97e−08 ** (3.9) 3.76e−08 ** (5.5) NURSES −1.54e−09 ** (2.2) −1.29e−09 ** (2.4) PHYSICIANS −2.50e−10 (0.1) −9.13e−10 (0.6) POP65plus 0.00002 ** (3.0) 0.00001 ** (3.0) GINI −5.79e−06 * (1.9) −7.92e−06 ** (3.2) NoINSURE 6.42e−08 (1.4) 5.23e−08 * (1.7) HOSPocc 1.81e−08 * (1.8) 2.08e−08 ** (2.7) CAbor −1.68e−07 (1.0) MEXbor −5.32e−08 (0.2) N51 49 R 2 0.42 0.45 F-value 4.31 ** 7.02 ** Note. SeeTable 2. 204 GOEL of health scams may have few policy options in case they are looking to reduce health fraud further. As an alternative consideration, and to focus on the role of border health scams, Table 5b adds HealthSCAMbor to Table 5a to see whether the impact of border health scams varies across states with different prevalence of health scams. The resulting coefficient on HealthSCAMbor was positive in all cases, but statistically insignificant. This insignificance of border health scams makes sense when one thinks about states with low (or high) health scams do not necessarily have to have border states that have high health scams (or the clusters of health scams in more populous areas might be away from state borders). 18 The other results are quite similar toTable 5a. The upshot of all this is that policies to control health care fraud should consider the existing prevalence of such scams in a state. For instance, increasing the share of nurses would provide extra vigilance in states with a high prevalence of scams and not otherwise. The concluding section follows. 5|CONCLUDING REMARKS This paper contributes to the literature on health care fraud by examining the presence of contagion in fraud across jurisdictional boundaries. Whereas health care fraud has drawn the attention of academics and policymakers in recent years (https://www.fbi.gov/ investigate/white-collar-crime/health-care-fraud), the possible impacts or spillovers from bordering jurisdictions have not been considered. Using state-level data for the United States, we find evidence of contagion in medical fraud across states. Health care fraud in a given state is significantly and positively impacted by such fraud in bordering states. These are also cross-border spillovers from corruption on medical fraud, but no evidence of such spillovers from international borders (i.e., states bordering Canada and Mexico). Consistent with intuition, the contagion from border health scams is more than twice that from corruption (respective elasticities are 0.18 [Model 3.1] and TABLE 5a Drivers of health care scams across prevalence of scams: Quantile regression (dependent variable: HealthSCAM) 5a.1 (q20) 5a.2 (q50) 5a.3 (q80) GDP −0.00001 (0.6) 3.80e−06 (0.3) 5.38e−06 (0.4) URBAN 1.85e−08 (1.4) 2.59e−08 ** (2.0) 3.82e−08 ** (3.6) NURSES −7.09e−10 (0.6) −1.47e−09 ** (2.4) −2.24e−09 ** (2.8) PHYSICIANS 2.63e−09 (0.6) −9.39e−10 (0.3) −1.85e−09 (0.7) POP65plus 1.73e−06 (0.2) 0.00001 (1.5) 0.00002 * (1.8) GINI −7.32e−07 (0.1) −5.53e−06 (1.1) −7.29e−06 (1.3) NoINSURE 5.79e−08 (0.9) 3.43e−08 (0.8) 2.29e−08 (0.4) HOSPocc 2.13e−08 (1.5) 1.39e−08 (1.0) 3.19e−08 ** (2.3) N51 51 51 Pseudo-R 2 0.32 0.37 0.45 Note. See Table 1a for variable definitions. Constant included but not reported in these quantile regressions. q50 is median regression. The numbers in parentheses are absolute tstatistics based on 200 replications of standard errors. * Statistical significance at the 10% level. ** Statistical significance at the 5% (or better) level. TABLE 5b Drivers of health care scams across prevalence of scams, including contagion effects: Quantile regression (dependent variable: HealthSCAM) 5b.1 (q20) 5b.2 (q50) 5b.3 (q80) HealthSCAMbor 0.03 (0.4) 0.08 (1.2) 0.12 (1.6) GDP −0.00001 (0.7) −2.88e−06 (0.2) −7.13e−06 (0.4) URBAN 1.59e−08 (1.1) 2.77e−08 ** (2.0) 4.19e−08 ** (3.1) NURSES −9.34e−10 (0.8) −1.28e−09 * (1.7) −1.50e−09 * (1.7) PHYSICIANS 1.79e−09 (0.4) −2.06e−09 (0.7) −2.52e−09 (0.9) POP65plus −4.10e−06 (0.5) 0.00001 (1.1) 0.00002 * (1.8) GINI −6.97e−08 (0.01) −3.50e−06 (0.7) −5.44e−06 (1.2) NoINSURE 5.11e−08 (1.1) 1.94e−08 (0.4) 4.42e−08 (0.9) HOSPocc 2.51e−08 (1.6) 1.37e−08 (0.9) 2.96e−08 ** (2.3) N49 49 49 Pseudo-R 2 0.40 0.40 0.50 Note. SeeTable 5a. GOEL 205