An Empirical Analysis of the Impact of Legal Sports Betting On Consumer Credit Health


 
 
The Supreme Court’s May 2018 decision in Murphy v. NCAA removed the federal prohibition against sports betting and invited states to regulate the practice for themselves. This has launched a national debate. Advocates in favor of legal sports betting champion increased tax revenues, business for struggling casinos and racetracks, and regulation of a practice that has flourished in the shadows. Detractors warn of the social ills commonly associated with gambling, including crime, addiction, and financial waste. 
This Note provides the first empirical analysis of the impact of legal sports betting on consumer credit health. Making use of the staggered sequencing of state legalization, I find that legal sports betting accounts for a small but statistically significant increase in mortgage delinquency rates. I submit that this finding justifies caution as policymakers explore legal sports betting opportunities. 
 
 


("New Jersey hopes sports betting will boost the state's struggling casino and horse-racing industries, as well as provide the state with new tax revenue."); see also Kelsey Butler, How Casinos Failed Atlantic City and Why They're Still Part of Its Future, THESTREET (Apr. 13, at least since 1949 and was exempt from PASPA. 9 Together, these thirteen states have embraced a variety of regulatory approaches. Some rolled out the red carpet, with legal inperson and online betting options and few barriers to new players. 10 Others opted for "legalization-lite," issuing sports book operator licenses for in-person gambling only. 11 A third group of states allowed consumers to bet at physical locations but required them first to register in person to make an online wager. 12 And most of the country-thirty-seven states and Washington, D.C.-still prohibited sports betting as of September 30, 2019. The different approaches reflect an ongoing national debate. To many, the benefits of legal sports betting are important and obvious. 13 They include new tax revenues (New Jersey earned about $26 million from sports betting in the 9 11 See, e.g., MISS. CODE. ANN. § 75-76-5(v) (2020) (defining "[l]icensed gaming establishment" as "any premises licensed pursuant to the provisions of this chapter" (emphasis added)); Joe Williams, Mississippi Sports Betting: Is Legal Sports Betting Available in Mississippi?, USA TODAY: SPORTSBOOK WIRE (May 2, 2020, 11:00 AM), https://sportsbookwire.usatoday.com/ 2020/05/02/mississippi-sports-betting-is-legal-sports-betting-available-inmississippi/ [https://perma.cc/S9CJ-D82Q]. 12 See, e.g., Operation of Gaming Establishments, NEV. GAMING COMM'N. REG. 5.225 (7)  twelve months following its June 2018 legalization), 14 more business for struggling state race tracks and casinos, 15 and the introduction of some consumer protections to a practice that has flourished in the shadows and been linked to organized crime. 16 But to many others, sports betting-like other forms of gambling-is a dangerous activity accompanied by a range of social ills like addiction, crime, and financial waste. 17 Perhaps because the regulatory changes in the sports betting space are so recent, there has been little empirical investigation into their impact on consumer financial 14 See US Sports Betting Revenue and Handle, LEGAL SPORTS REP. (last updated Dec. 31, 2019, 10:10 AM), https://www.legalsportsreport.com/ sports-betting/revenue/ [https://perma.cc/3AFV-NKHE]. For convenience, the same data appear infra Part VI app. A. 15 See Corasaniti, supra note 13. 16 Cf. Bennett Baumer, Betting the House: The Mob and Sports Gambling, INDYPENDENT (Jan. 21, 2014), https://indypendent.org/2014/01/ betting-the-house-the-mob-and-sports-gambling/ [https://perma.cc/99SQ-8YZN] (discussing the relationship between sports gambling and organized crime). 17  health. 18 This Note takes advantage of the staggered adoption of legalized gambling after Murphy to offer a first attempt. 19 It focuses on policy variables germane to the longstanding public debate around gambling, asking how legal sports betting affects consumer credit health. If sports betting visits a substantial drain on financial resources without providing offsetting benefits, legalizing the practice might contribute to negative consumer credit outcomes. 20 One observable proxy for such outcomes is whether or not consumers keep current on their mortgage payments, which are significant monthly obligations for many homeowners. 21 This Note's empirical analysis includes nearly ten years of monthly mortgage delinquency rates for each state-nearly 6,000 observations in total. 22 In addition to mortgage data, the empirical analysis uses a novel, comprehensive panel dataset that describes each state's approach to regulating sports betting. 23 18 See Håkansson, supra note 17, at 14 (observing a need for more research on the relationship between gambling and consumer credit). 19 The datasets and models used in this Note are available for replication and cross-checking. Matthew Q. Clarida, Models of Legal Sports Gambling and Consumer Credit Health, DROPBOX (last updated Dec. 8,2020), https://www.dropbox.com/s/zarux001j7var7x/Clarida%20-%20Note% 20Data.xlsm?dl=0 (on file with the Columbia Business Law Review). Readers and researchers are free to utilize these data, provided they cite this Note in doing so. 20 For an example of such outcomes, see Håkansson, supra note 17, at 2 (finding that "short-term and intense gambling, rather than gambling itself, may identify risk of payback failure and risk of indebtedness."). 21  I find that legal sports betting is associated with a small but statistically significant increase in mortgage delinquency rates. 24 However, I also find that employment gains from legalization may partially or totally offset this negative effect. 25 I submit that these results-while not definitive evidence of cause and effect-suggest causality due to the quasi-experimental setting they reflect. 26 I therefore recommend that policymakers prioritize employment gains when implementing legal sports betting. 27 This may be done, to give one example, by routing sports betting through existing casinos and racetracks via regulations which require potential online bettors to visit a casino in order to open an internet gaming account.
The rest of this Note unfolds in three parts. In Part II, I provide an overview of sports betting regulation in U.S. states as of September 30, 2019, the date through which the empirical models used in this Note are current. In Part III, I present these empirical models, explaining the data and methodology I used and showcasing three central models. This Part also includes robustness checks of each model and describes the limitations of my empirical analysis. In Part IV, I use Connecticut, a state considering legalization at the time data, see Matthew Q. Clarida, Data on State Regulation of Sports Gambling, Dropbox (last updated Sept. 8, 2020), https://www.dropbox.com/s/ 8mq4e60d5vz6sj3/Sep%208%20Models_v5.xlsm?dl=0 (on file with the Columbia Business Law Review). Readers and researchers are free to utilize these data, provided they cite this Note in doing so. 24 See infra Section III.C. 25  These thirty-seven states are central to this Note's analysis for two reasons. First, they serve as natural comparisons to the states which have legalized, supporting the empirical models in Part III. 37 Second, they offer visibility into the 31 Of course, these states have different tribal and illegal gambling environments. These variables present opportunities for further research, see infra Section III.D, but they are largely outside the scope of this Note. 32 See, e.g., ARK. CONST. amend. 100, § 3 ("Casino licensees may accept wagers on sporting events[.]" (emphasis added)). 33 See, e.g., Operation of Gaming Establishments, NEV. GAMING. COMM'N REG. 5.225 (7) (2018). 34 See, e.g., N.J. STAT. ANN. § 5:12A-11(a)-(b) (West 2020) (establishing licensing scheme without an in-person registration requirement for "sports wagering lounge[s]" offering online sports betting). 35 Gaming laws in U.S. territories are outside the scope of this paper. 36  various U.S. laws that regulate gaming activity and remain in place after the Murphy decision. 38 The Indian Gaming Regulatory Act (IGRA) governs the relationship among tribes, states, and the federal government on gambling issues. 39 Under the IGRA, tribes may offer casino-style gaming-including sports betting-only after agreeing to a detailed regulatory and revenue sharing relationship with the state where they are located. 40 These arrangements may operate as hurdles to legal sports betting today. 41 In Arizona, for example, the state government cannot legalize sports wagering unless it agrees to a significant reduction in its revenue share with the Pascua Yaqui Tribe or convinces the tribe to modify the compact's terms. 42 Three acts of Congress apply to the transmission of sports wagers or related material between one state where sports betting is legal and another where it is not. The Wire Act forbids the use of "a wire communication facility" to transmit betting information across state or national lines if sports wagering is illegal either at the origin or the terminus of the 38 For a discussion of federal laws providing basline regulation of sports gambling after Murphy even in these thirty-seven states, see Matthew A. Melone, New Jersey Beat the Spread: Murphy v. National Collegiate Athletic Association and the Demise of PASPA Allows for States To Experiment in Regulating the Rapidly Evolving Sports Gambling Industry, 80 U. PITT. L. REV. 315, 318-24 (2018). 39  transmission. 43 Similarly, the Travel Act prohibits a person from traveling across state lines in order to further gambling businesses that are illegal in the destination state or under federal law. 44 The Interstate Transportation of Wagering Paraphernalia Act prohibits the transport of sports betting materials across state lines, except when the destination state allows sports betting. 45 Two additional acts regulate those who attempt to run sports betting businesses that are not permitted under state law. The Unlawful Internet Gambling Enforcement Act forbids operators from accepting online payments from players attempting to place bets from states that do not allow internet wagering. 46 The Illegal Gambling Business Act levies additional penalties against certain businesses engaged in illegal gaming. 47 These regulations provide the backdrop to legal sports wagering in the United States. Critically, their severe penalties should ensure that legalization efforts in one state are relatively contained to that state's borders-though not necessarily to that state's citizens. 48  Class One states only allow sports betting in person. 50 This policy choice has a major impact on the revenue-generating potential of sports betting: consider that New Jersey now earns more tax revenue from internet betting than from inperson wagers. 51 As of September 2019, the Class One states 49 For the first wager dates, see Rodenberg, supra note 8. For the other data underlying this table, see supra note 14. "Tax Rate" refers to the quotient of tax revenue and operator revenue. 50 See supra note 11 and accompanying text. 51 See Katherine Sayre, Mobile Sports Betting Is the Moneymaker as More States Legalize, WALL ST. J. (Sept. 2, 2019, 7:03 PM), https:// www.wsj.com/articles/mobile-sports-betting-is-the-moneymaker-as-morestates-legalize-11567445689 (on file with the Columbia Business Law Review) ("Online gamblers now account for about 80% of all legal wagers on were Delaware,Mississippi,New Mexico,Arkansas,New York,Oregon,and Indiana. 52 Officials in Class One states have cited two primary reasons to limit legal sports betting to brick-and-mortar operations. The first is a desire to drive business to existing casinos and horseracing venues. 53  (describing initial New York sports betting in a casino); Chinook Winds Casino Opens First Sportsbook Lounge in Oregon, YOGONET (Aug. 28, 2019), https://www.yogonet.com/international/noticias/2019/08/28/50783-chinookwinds-casino-opens-first-sportsbook-lounge-in-oregon [https://perma.cc/CF4X-2MHC] (describing initial Oregon sports betting in a casino, as well as the possibility for public operation of an online sports betting service); Jabari Young, Oregon Lottery to Launch Sports Betting App Scoreboard with an Expected $300 Million in Wagers, CNBC (last updated Oct. 16, 2019, 10:29 AM), https://www.cnbc.com/2019/10/15/oregon-lot tery-to-launch-sports-betting-app-scoreboard.html [https://perma.cc/4RDT-LSRZ] (describing Oregon's publicly-run sports betting service); Indiana Sports Betting, LEGAL SPORTS REP. (May 5, 2020, 7:14 PM), www.legalsportsreport.com/indiana/ [https://perma.cc/C7VJ-LLP6] (noting that Indiana betting apps must associate with "land-based entities" and that no apps launched before September 30, 2019). 53 See, e.g., Johnson, supra note 1 (discussing New Jersey's interest in supporting the "casino and horse-racing industries"); CNBC, Delaware Class One state on June 5, 2019, and Governor John Carney placed a bet that day. 54 He explained that the state's gambling "legislation . . . was designed to reinvigorate the horseracing industry, so I don't expect that we'll take sports betting outside of those three racinos." 55 Mississippi has advanced similar justifications, and officials credited sports betting with increasing business at casinos in the state. 56 Advocates and officials in Class One states also have voiced concerns about the collateral consequences associated with gambling, including addiction and financial waste. 57  [their] hard-earned money." 58 Others raise the related concern that increased access will contribute to addiction. 59 "What we know to be true in any vice exposure-whether it be substance abuse or gambling-is that increased availability leads to increased participation, which leads to the inevitable increase in problems and consequences," New York advocate James Maney testified before a state committee in May 2019. 60 Background law may shape permissiveness as well. New Mexico's government has not acted on sports betting. Tribes within the state, however, have argued that they may offer sports betting under the IGRA because New Mexico does not specifically prohibit it. 61 On October 16, 2018, the Pueblo tribe in Santa Ana accepted its first sports bets at its casino outside Albuquerque. 62 And in Arkansas, sports betting became legal at the state's casinos only after voters approved an amendment to the state constitution. 63 Class Two states have a ready model. Nevada historically has been known as the gambling capital of the United States. 65 Its regulators claim that it accounts for more than half of commercial casino employment in the United States. 66 Employers include the state's 172 sports-wagering licensees. 67 Despite these impressive figures, Nevada strikes a compromise in its regulatory scheme by requiring that new customers visit a casino to register internet sports betting accounts. 68 This policy choice means that it is easier to place sports bets in New Jersey, West Virginia, and Pennsylvania than in Nevada. 69 But this approach also preserves the central role of casinos-and casino employment-in Nevada sports betting.

C. Regulatory Compromise: Class Two States
Nevada's regulatory apparatus is robust. The Nevada Gaming Commission (NGC) and Nevada Gaming Control Board (NGCB) regulate gambling, including sports betting, within the state. 70 Their regulations cover everything from the amount of required reserves an operator must hold ($25,000 at minimum) to which types of bets may be accepted (most professional and amateur sports bets) to whether bettors may pay by credit card (generally no). 71 Nevada's deep history with sports betting has made it an attractive template for new entrants. 72 "Nevada has had legal, regulated sports wagering for several decades and the lessons learned from this experience can help guide states or tribes looking to authorize sports wagering," advise two practitionerprofessors. 73 Recently, Iowa has accepted this advice. The state legalized sports betting in August 2019, requiring inperson registration before customers could use the internet to wager. 74 In early September 2019, Rhode Island also decided to follow Nevada's lead 75 and moved from Class One legalization to Class Two legalization after becoming the first state to lose 70  (discussing in-person registration requirement for mobile wagering and noting that Iowa's tax rate on wagers is the lowest in the country, equaling Nevada's), https://www.desmoinesregister.com/story/sports/2019/07/30/ iowa-sport-betting-start-date-legal-sports-gambling-app-ncaa-collegefootball-spread-rule-how-to-bet/1857134001/ [https://perma.cc/R6L3-2CT2]; IOWA CODE § 99F.9 (2020) (allowing online sports betting after in-person registration). 75  money in a month of accepting sports bets. 76 In February 2019, the state incurred $450,000 in losses after its operating partner lost more than $2,000,000 in Super Bowl wagers. 77 The local New England Patriots won the game. 78 D. All In: Class Three States Pennsylvania allow not only brick-and-mortar sports betting but also internet sports betting without in-person registration. 80 A prospective mobile bettor needs only a social security number, checking account, physical presence within a Class Three state, and about ten minutes to place a legal wager. 81 This ease makes sports betting more available to state residents and short-term travelers across state lines. 82 New Jersey became the first Class Three state in July 2018. The state requires internet sports betting operators to affiliate with brick-and-mortar locations, evidencing some desire to use sports betting to support racetracks and casinos. 83 By November 2019, it had ten licensed and operating "Sports Wagering Lounges," each anchored at a casino or racetrack. 84 However, it is unclear if New Jersey's affiliation rules drive casino employment, especially since mobile gaming made up the majority of 2019 sports wagers in the state and fueled rapid revenue growth.  internet sports betting without an in-person registration requirement, 90 although operators were still required to associate with brick-and-mortar casinos. 91 The result has been a rapid increase in the total amount of sports wagers, as Figure 2 shows.

Figure 2 92
Other states are increasingly exploring the Class Three approach. After experiencing technical issues for weeks with its online betting platform, West Virginia welcomed back internet sports betting by August 2019. 93  from January 2010 through September 2019. 98 I drew this data from the National Mortgage Database (NMDB), a "nationally representative, 5 percent sample of all outstanding, closed-end, first-lien, 1-4 family residential mortgages." 99 Within this sample, I focus on the rate of mortgages by state that were delinquent from thirty to eightynine days, analyzing 5,967 observations dating back to January 2010.
Dependent variable selection is paramount to any regression analysis because regressions simply measure how independent variables influence the dependent variable. 100 I focus on mortgage delinquency for four reasons. First, mortgages represent "the single largest market for consumer finance." 101 Second, mortgage payments traditionally are due each month, representing an important financial decision-to pay or not to pay-that millions of consumers make at regular intervals. 102 Third, the NMDB's monthly data allow for robust empirical analysis of policy changes following Murphy that would be more difficult with data released at less frequent intervals. 103 Fourth, while many national databases update 98 "A mortgage is considered delinquent or late when a scheduled payment is not made on or before the due date." U.S. DEP'T OF HOUS. & URB. DEV., supra note 21, at 8 tbl.Helpful Terms. 99 About the Data, supra note 23.
with considerable lag, the NMDB updates quarterly, providing a fresh look at consumer credit health. 104 Independent variables should help explain the variation observed in the dependent variable. 105 This Note's first model employs a fifty-state dataset that indicates on a binary scale whether a state had legal sports betting in each month since January 2010. The second model uses a comprehensive panel dataset that reflects the degree of legalization in each state by incorporating the regulatory classifications proposed in Part II and summarized in Table 5. The third model uses state-by-state betting data 107 and analyzes the consumer credit impact of each additional dollar wagered on sports.
The final elements of the models are control variables: inputs uncorrelated with the treatments but substantially explaining changes in the dependent variable and thus helping to isolate the impact of legal sports betting. 108 I include four such controls. The first is the unemployment rate given in the Local Area Unemployment Statistics (LAUS) dataset released each month by the Bureau of Labor Statistics. 109 The second is the gross domestic product (GDP), released quarterly by the Bureau of Economic Analysis in annualized form for each state. 110 Finally, I assign each month a season (e.g., winter) and each state a region (e.g., Southeast) in order to account for variation driven by regionality and seasonality. 111

B. Methodology
To estimate the connection between legal sports betting and mortgage delinquency rates within a state, I use the difference-in-differences (DiD) approach. This method is common in studies of regulatory interventions and public 107 For a compilation of these data, see infra Part VI app. A. 108  After New Jersey raised its minimum wage in 1992, Card and Krueger used DiD to analyze New Jersey's divergence from Pennsylvania, which kept its minimum wage constant. 115 They found that the higher minimum wage did not contribute to job losses, despite the predictions of traditional microeconomic models. 116 Similarly, I use the 37 states that have kept sports betting prohibitions in place as control populations to isolate the impact of legalization. The DiD method assumes that the control and treatment groups exhibited similar trends before the policy change took place. 117 As seen in Figure 3, the pre-Murphy mortgage delinquency trends of treatment (legalizing) and control (non-legalizing) states conform to this assumption relatively well.

Figure 3
Turning to the mechanics of the study, I primarily implemented the DiD approach using traditional ordinary least squares linear regression. 118 Linear regression is a common predictive technique in the field of statistics. 119 At bottom, it compares movements in one or more independent variables to the observed fluctuations in a dependent variable. 120 It then analyzes whether each independent variable had a relationship with the dependent variable not 118  caused by random chance and estimates the strength of that relationship (summarized by a coefficient "β"). 121 A regression can be reduced to a simple equation that describes the independent and dependent variables and their relations. 122 This Note's empirical analysis takes the following general form, where i indexes the state jurisdiction and t indexes time 123 : In the above equation, the dependent variable is the percentage of mortgages that were between thirty and eightynine days delinquent in each month in each state. 124 Because time of year and location and influenced delinquency rates as seen in Figure 3, the equation includes "fixed effects" controls for regional variation (β6) and seasonal variation (β7). 125 Additionally, because of the relationship between macroeconomic health and mortgage performance, 126 I included state gross domestic product (β4) and unemployment rate (β5) in the model as control variables. These four inputsregion, season, GDP, and unemployment-improve the 121 See Daniel L. Rubinfeld, Econometrics in the Courtroom, 85 COLUM. L. REV. 1048REV. , 1054REV. , 1065REV. -68 (1985. 122  model's explanatory power by accounting for variation in mortgage delinquency rates that is not explained by legal sports betting. 127 Figure 4 β0 is the model's intercept. This figure represents the baseline average of the mortgage delinquency rate that the independent variables do not explain. 128 Timei,t indicates, by taking a value of zero or one, whether a given observation of a state occurred before or after Murphy. This is critical, because it allows the model to measure-through the coefficient β1how the situation in each state changed after the decision. 129 The next variable, Legali,t, represents the sports betting regulations in each state in each month on a binary basis (Model One), on a degree of legalization basis (Model Two), or by indicating the total amount monthly amount wagered on sports (Model Three).
The crux of the DiD regression is β3, the coefficient of the product of the timing variable (Timei,t) and the treatment variable (Legali,t). This interaction term measures how states that legalized sports betting deviated in mortgage 127 See supra note 111 and accompanying text. 128 See ATA Airlines, Inc. v. Fed. Express Corp., 665 F.3d 882, 890 (7th Cir. 2011) (Posner, J.) (describing the function of the intercept in a regression model). 129 See id. (describing the function of a coefficient in a regression model).
delinquency rates from both their pre-legalization trend and the trends of states that did not legalize. 130 C. Results I investigate the connection between legal sports betting and consumer credit health from three vantage points. Model One examines whether the presence of sports betting in a state-represented by a one-zero binary for each state in each month-affects that state's mortgage delinquency rates. Model Two asks whether the degree of legalizationrepresented by regulatory classifications constructed in Part II-is significant. Model Three charts the relationship between each additional dollar wagered on sports and mortgage delinquency rates. Supporting technical material appears in the appendices. 131 The three models suggest causal-not just correlationalrelationships because the identification strategy capitalizes on a natural experiment: the staggered adoption of unique sports betting regulations. 132 While I could not control for all unobservable factors as one could in a randomized study, this Note's empirical design marks a helpful first step in evaluating the role of legalized sports betting in society and justifying some degree of caution among policymakers.

The Presence of Legal Sports Betting
Model One addresses a threshold question: does the presence of legal sports betting alone -with no consideration of the scope of legalization or the amount of money wagered-130 See Difference-in-Difference Estimation, supra note 105. 131 See infra Part VI app. B (containing summary data as well as tests of linearity and normality). 132 See Craig et al., supra note 37, at 1832-33 (suggesting that welldesigned natural experiments, minimally defined as precluding "exposure to the event or intervention of interest . . . manipulated by the researcher," may support causal inferences). But see ANGRIST & PISCHKE, supra note 26, at 18-22 (endorsing causal inference from natural experiments but implicitly limiting the term to occurrences close to random in their mitigation of selection bias).
contribute to a statistically significant change in mortgage delinquency rates, controlling for GDP, unemployment, regionality, and seasonality? This is a DiD regression of the following form, with the interaction term bolded 133 : I implemented this design through three regressions. The first uses traditional standard errors, a common starting point for regression analyses. 134 The second uses Eicker-Huber-White "robust" standard errors to address heteroskedasticity in the data. 135 The third uses the Cochrane-Orcutt estimation procedure to correct for serial autocorrelation common in time series studies. 136 These alternate approaches serve as robustness checks and support the estimations of the initial model. 137 In the primary model, the "Time*Legal" coefficient, estimated at 0.304 percent, indicates the impact of sports betting legalization. 138 This coefficient is statistically significant at the one-percent level, meaning that there is only a remote chance that randomness explains the result. 139 Moreover, the coefficient appears substantively significant: the treatment effect of legalization-0.304 percent-explains more than a quarter of the observed sample standard deviation in mortgage delinquency rates (0.96) 140 and is slightly greater than the estimated impact of a one-percent change in unemployment. Additionally, the model's R-Squared figure indicates that the independent variables explain 73.57 percent of the observed variability. 141 The relationships between the control and dependent variables further support the model's internal validity. Rising GDP correlates positively with consumption 142 and should correlate negatively with mortgage delinquency rates. Thus, these rates should fall as GDP rises, and Model One estimates that they do. Rising unemployment should have the opposite effect, placing downward pressure on consumption. Mortgage delinquency rates should therefore rise as unemployment rises, 143 and Model One reflects this. Nevertheless, this model provides a rather blunt view of the policy choice at hand. Legalizing sports betting is not a binary choice: states may craft diverse and detailed regulatory 137 A further discussion follows infra Section III.D. 138 Note that the estimate is 0.304 / 1, not 0.304 / 100, because the NMDB represents five percent as "5.0" rather than ".05." 139 fig.1 (2015). 143 See Campbell & Cocco, supra note 126, at 1499.
schemes. While Class Three states, like New Jersey, have permissive approaches to sports betting with many in-person and online options, 144 Class One states like Delaware only offer sports betting at casinos and retailers. 145 A shortcoming of Model One is that it treats these states identically.

The Degree of Legalization
Model Two retains the DiD approach but replaces the Legali,t binary term with a score between zero and three for each state in each month from January 2010 through September 2019. These scores, presented again in Table 7, reflect the classificatory scheme in Part II. The regression equation, with the interaction term bolded, is as follows: As with Model One, I evaluate Model Two using traditional standard errors, Eicker-Huber-White robust standard errors, and the Cochrane-Orcutt method. 147 The three approaches each find a statistically significant positive association between more permissive legal sports betting and mortgage delinquency. The estimated effect of "increasing" a state's regulatory class-0.142 percent in the primary model 148 -is the "Time*RegClassification" coefficient, which is statistically significant at the one-percent level. 149 This estimate is also substantively significant: a one-class change accounts for nearly fifteen percent of the observed standard deviation in delinquency rates (0.96). 150 Compared to Model One, the R-Squared figure falls slightly from 73.57 151 percent to 73.53 percent of delinquency variability explained, while the GDP and unemployment controls have the expected relationship with the outcome variable. 152 While Model Two's treatment coefficient is smaller than Model One's, it is important to note that the corresponding variable takes values from zero to three, not just the values zero and one. Thus, Class Three states like New Jersey would expect a larger impact from sports gambling-three multiplied by 0.142, or 0.426-than Class One states. The model therefore estimates higher delinquency rates in states that have legalized more aggressively, holding the control variables equal. Model Two is thus fairly consistent with 147 See supra notes 134-37 and accompanying text. 148 For a clarification of this interpretation, see supra note 138. 149 For a brief discussion of statistical significance, see supra note 139 and accompanying text. 150 For the standard deviation and other summary data, see infra Part VI app. B tbl.Summary Statistics. 151 See supra Section III.C.1. 152 See supra notes 142-43 and accompanying text.
Model One but provides more insight into the impact of the scope of legalization offered in each state.

The Amount Wagered
The regulatory classifications used in Model Two approximate the level of legalization in each state that permits sports betting. A potentially more precise way to analyze a state's degree of legalization is to examine the amount of money bet on sports in each state, which is generally referred to as the handle. Unfortunately, a major challenge with this method is that, at the time of this study, four states-New Mexico, New York, Arkansas, and Oregoneither had released no data or incomplete data. 153 However, these are states where there has been relatively little legal sports betting. New Mexico offers sports betting only at a few tribal casinos, 154 and New York and Arkansas also have few betting options. 155 Oregon's tribal sports betting did not begin until August 27, 2019. 156 Therefore, while this model has important data limitations, it still may provide a more nuanced look than the research classifications used in Models One and Two. The regression equation is as follows: (Mortgage Delinquency Rate)i,t = β0 + β1Handlei,t + β2GDPi,t + β3Unemploymenti,t + β4RegionFEi + β5SeasonalFEi,t + εi,t.  Model Three's output is largely consistent with Models One and Two. 157 It indicates that each additional dollar bet on sports has a small, upward, and statistically significant relationship with mortgage delinquency rates. The coefficient for the "Handle" variable reflects this and is significant at the one-percent level. 158 As in Model One and Model Two, the GDP and unemployment variables in Model Three have the expected relationships with mortgage delinquency. 159

D. Objections and Limitations
While regressions alone do not provide definitive evidence of causal links, I submit that the time-series methodology and associated results above do report a form of correlation suggestive of causality. 160 Nevertheless, my research is only a first step toward analyzing a relatively recent policy problem, and I acknowledge its limitations. First, the models presented above each display evidence of heteroskedasticity and autocorrelation. Second, primarily due to data availability, I used a relatively small number of control variables. Finally, Model Two relied on the regulatory classifications I described in Part II, and these state-by-state classifications involved research judgments that may be revisited. 157 For the results of these models, see supra Sections III.C.1-.2. 158 Models 1 and 2 also were significant at the one-percent level. However, the statistics academy generally accepts significance at the five percent level. See Rubinfeld, supra note 142, § 6:13. 159 See supra notes 142-43 and accompanying text. 160 See supra note 132 and accompanying text.

Technical Issues
Heteroskedasticity is relatively common in economic studies. 161 It occurs when the variances of the errors in a regression model are not identical. 162 This is a violation of standard regression assumptions that often occurs when prediction errors correlate with model variables. 163 This Note's three models exhibit heteroskedasticity: plots of residuals against fitted values for each model show trademark "cone" shapes, with each model becoming less reliable as predicted values rise. 164 The presence of heteroskedasticity does not mean that a model's coefficients are incorrect. 166 Rather, heteroskedasticity indicates that the significance testswhich report the probability of the coefficients being caused by random chance-may be inaccurate. 167 To correct this issue, I ran a second iteration of each model using the heteroskedasticity-consistent covariance estimator approach developed by White,Huber,and Eicker. 168 This technique produces "robust" standard errors not influenced by the unfulfilled regression assumptions and therefore supports more reliable significance tests. 169 These tests confirmed my original results. 170 Serial correlation also is a common issue in time series analyses, and it is present in this Note's models. 171 I confirmed this by performing the Durbin-Watson test, which reported statistically significant serial correlation in the error 166 ROBERT L. KAUFMAN, HETEROSKEDASTICITY IN REGRESSION: DETECTION AND CORRELATION 3 (2013) ("If there is heteroskedasticity, the good news is that using [ordinary least squares regression] to estimate [the model] provides unbiased estimates of the coefficients."). 167 See id. at 3; White, supra note 135, at 817 ("It is well known that the presence of heteroskedasticity in the disturbances of an otherwise properly specified linear model leads to consistent but inefficient parameter estimates and inconsistent covariance matrix estimates. As a result, faulty inferences will be drawn when testing statistical hypotheses in the presence of heteroskedasticity."). 168 See supra note 135 and accompanying text. But see Gary King & Margaret E. Roberts, How Robust Standard Errors Expose Methodological Problems They Do Not Fix, and What To Do About It, 23 POL. ANALYSIS 159, 159-60 (2015) (criticizing widespread improper use of robust standard errors). Calculated robust standard errors appear supra Sections III.C.1 tbl.6, III.C.2 tbl. 8 & III.C.3 tbl.9. 169 See WILLIAMS, supra note 162, at 6-7. 170 Note that the coefficient of each variable stayed the same. This is because calculating standard error in a different way is unrelated to the value of the coefficient. See id. at 7. In fact, the heteroskedasticityconsistent errors generally were lower than traditional standard errors. Professors Angrist and Pischke suggest "taking the maximum of the conventional standard error and a robust standard error as your best measure of precision." ANGRIST & PISCHKE, supra note 26, at 230. 171 See Difference-in-Difference Estimation, supra note 105 (advising corrections for autocorrelation in time series DiD regressions). terms of each model. 172 This correlation may impact the reliability of significance tests. 173 While robust standard errors may mitigate this issue, I also used the Cochrane-Orcutt procedure to address it. 174 The correction did modify the coefficients of the treatment variables, but the changes were minor, and the results retained statistical significance at the one-percent level. 175 2. Variable Selection I turn next to my selection of variables. At the outset of this study, I considered a number of options for the dependent variable, including a variety of data related to consumer credit health. I selected mortgage delinquency in part because monthly readings are available with a short lag, making this data fresher than other options. 176 This freshness is critical because the policy changes at issue in this Note are less than three years old. As time passes, however, a wider variety of datasets will become suitable for analysis. Future researchers may find, for example, that other datasets are more responsive to the policy changes at issue than mortgage delinquency rates.
My choice of independent variables also presents an opportunity for extension. From the outset, I chose to focus on simple, macroeconomic relationships-those involving GDP 172  and unemployment-that are familiar to most readers and would allow for relatively straightforward statistical analysis. I also limited independent variables to ensure that the variance inflation factors remained within acceptable ranges. 177 Nevertheless, more comprehensive controls may be beneficial. For example, future researchers may attempt to control for the availability of traditional gambling in each state under the assumption that sports betting activity may be greater or smaller in states where significant alternatives exist. Relatedly, researchers may attempt to control for spillover effects, such as New Yorkers' access to New Jersey internet sports betting upon crossing state borders. 178 To the extent possible, researchers may also control for the relative availability of illegal sports betting in each state, especially if the state offers no other sports betting or if these illegal options attract players with better user experiences or more favorable wagering odds. 179 Another potential set of controls would try to capture how income distribution impacts both who decides to bet on sports and how that choice affects mortgage delinquency, especially because gambling in general appears to be regressive. 180 Finally, as with many policy impact studies, it remains 177  possible that state legislatures have addressed sports betting in a way that reflects unobserved qualities shared by their citizens. 181 While the DiD methodology aims to account for unobserved pre-treatment trends, I still acknowledge that the models in this Note do not by themselves support a definitive causal link. 182

Regulatory Classification
A third shortcoming of my empirical design is the regulatory panel data elaborated in Part II and used in Model Two. The coding of this data demanded some exercise of judgment. The weightiest judgment call was the decision to score Nevada-the epicenter of gambling in the United States-as a Class Two state. I did this because Nevada's regulatory scheme requires potential bettors to register their online wagering accounts in person, supporting casino employment, but also posing an obstacle to betting that other states-most notably New Jersey-have removed. 183 Another example of research judgment is the coding of Delaware and Oregon as "zeroes" before May 2018, even though each of those states legalized sports lottery games during some period beforehand. 184 I did this because neither allowed single-game sports wagering, unlike Nevada. 185 Finally, I acknowledge that my coding procedure lumped states together only on the basis of their sports betting regulations without taking into account other factors that may be significant, like the availability of non-sports gambling. Nevertheless, this regulatory panel data was an input only in Model Two, which returned estimates that were consistent with Models One and Three, lending some confidence to the classifications. 181 Cf. Difference-in-Difference Estimation, supra note 105 (discussing the selection problem). 182 See id. (discussing the conditions necessary to permit reliable inference from DiD studies). 183 See supra Sections II.C-.D. 184 See S. REP. NO. 102-248, at 10 (1991). 185 See id. (comparing the two states' sports betting systems with Nevada's).

IV. Implications for Policymakers and Researchers
In Part III, I used DiD regression to analyze the relationship between legal sports gambling and consumer credit health, for which I selected mortgage delinquency rates as a proxy. I constructed three models, each of which suggested a small, positive, and statistically significant relationship between legal sports betting and mortgage delinquency rates. I noted that while my findings do not justify a definitive causal link, the time series design and associated results do report a form of correlation suggestive of causality. 186 I then highlighted various opportunities to improve the empirical design. I now propose a preliminary application of these findings to pending policy decisions.
In August of 2019, Connecticut continued to debate legal sports betting. 187 State legislative analysts had estimated in 2018 that a "limited availability" legalization would result in $2.2 billion of annual handle in the state, or an average of approximately $180 million bet legally on sports each month. 188 The models I showcase above can help analyze the impact of this potential policy choice. The estimates returned by Models One and Two predict that a small, upward, and statistically significant pressure on mortgage delinquency rates would accompany legalization. 189 Model Three's estimates sharpen this insight by indicating the impact of each dollar legally wagered on sports. Thus, Model Three allows for a comparison between Connecticut's eventual decision not to legalize (yet) and a counterfactual situation in 186 190 To analyze the no-legalization status quo, Model Three ingests Connecticut's actual GDP and unemployment figures for October 2019. We can then safely assume that Connecticut residents legally wagered $0 on sports in the state, because sports betting is illegal in the status quo. From these inputs, Model Three estimates that Connecticut's rate of delinquent mortgages would have declined slightly, from 2.1 percent to 1.96 percent. The prediction column records the product of the Model Three coefficient and the corresponding assumed input. In the counterfactual analysis, Model Three again incorporates Connecticut's actual GDP and unemployment figures for October 2019. We then assume that Connecticut approved sports betting in September 2019 and launched it on October 1, 2019, seeing $180 million in legal bets that month. 191 The model estimates that, in this scenario, Connecticut's rate of delinquent mortgages at the end of October reaches 2.08 percent, versus 1.96 percent in the status quo analysis. Of course, this counterfactual analysis depends on its assumptions and may be refined. For example, Connecticut's 2018 analysis projected that legalization would create new jobs. 192 According to Model Three, if the state had experienced a modest decline in unemployment from 3.6 percent to 3.3 percent in addition to seeing $180 million in legal sports bets, nearly all of the mortgage delinquency effects of the change would have disappeared. This case study suggests an important policy insight: the negative consumer credit consequences of legal sports betting may be offset by gains in employment. This proposition may support the regulatory strategies of states which have decided to route legal sports betting through existing casinos and racetracks. States have accomplished this in a number of ways. Class One states center all sports betting at existing brick-and-mortar facilities since they prohibit internet sports wagers. 193 While this may encourage new employment, these states have also seen relatively small amounts of legal wagering compared to states that allow internet sports betting. 194 Class Two states employ a compromise: all new internet sports bettors must register at existing casinos or racetracks, so the regulations encourage patronization of these facilities while tapping into the lucrative online market. 195 Even Class Three states generally require internet operators to affiliate with existing casinos, although it is unclear if this arrangement encourages employment. 196 This Note's results suggest that protecting and even incentivizing casino employment, as Class One and Class Two states do, may be a viable way to undercut the negative consumer credit impacts of legal sports betting.
As Justice Alito observed in Murphy, "Congress can regulate sports gambling directly, but if it elects not to do so, each State is free to act on its own." 197 So far, states have done all of the work, crafting their own regulatory schemes. If the federal government decides to get involved, it will have to make the same policy decisions this Note has analyzed at the state level. Notably, the recent Hatch-Schumer Act, a regulatory proposal which failed to advance in Congress, sought to allow internet sports betting without an in-person registration requirement. 198 This Note's models indicate that 193 See supra Section II.B. 194 See infra Part VI app. A. 195 See supra Section II.C. 196 See supra text accompanying notes 83-85. 197 Murphy v. Nat'l Collegiate Athletic Ass 'n, 138 S. Ct. 1461'n, 138 S. Ct. , 1484'n, 138 S. Ct. -85 (2018. 198 See Hatch-Schumer Act, S. 3793, 115th Cong. § 103 (2018).
such a requirement may be beneficial if it allows enough casino employees to share in the benefits of legal sports betting and therefore mitigates the consumer credit impact of legalization.
Finally, I stress that I have made a preliminary attempt at addressing an urgent question. If nothing else, my empirical methods indicate that there is enough data available to begin rigorous quantitative study of the policy choices associated with legal sports betting. The research opportunity is ripe and will improve over time as more regulatory approaches emerge and more data become available.

V. CONCLUSION
As more states throw their hats into the ring of legalized sports betting, it is incumbent on policymakers to develop a more complete understanding of the impacts of legalization. This is not an easy mandate. Isolating the effects of the available policy choices requires robust data and creative methodologies. This Note makes the first attempt at an empirical analysis of these policy options and their potential impacts on consumer credit health. It finds a small, upward, and statistically significant relationship between mortgage delinquency rates and both legal sports betting and unemployment. This result suggests that the optimal legal sports betting framework incentivizes employment in order to mitigate the negative consumer health consequences of increased gambling. It is also consistent with the observation that betting on sports is not a good investment. At best, it is a leisure activity that is a drain on our available resources. Policymakers should consider it accordingly.

VI. APPENDICES Appendix A: State Betting Totals and Panel Data
This data is released periodically by state authorities and compiled by Legal Sports Report, an industry website. 199