Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The Effect of Mandatory Insurer Reporting on Settlement Delay

The Effect of Mandatory Insurer Reporting on Settlement Delay Abstract To improve their fiscal position, Medicare and some state Medicaid programs have recently taken steps to mandate reporting of personal injury awards and thus facilitate subrogation against such awards. Participants in the tort system have argued these additional reporting requirements might delay settlement of claims, harming both plaintiffs and defendants. This article examines this problem empirically, using a rich, national data set of closed automobile bodily injury claims. Using a differences-in-differences research design that exploits the introduction of a new Medicare reporting requirement in 2011, it demonstrates that mandated reporting increased time to settlement by 19%, or an average of 58 days. Robustness checks using data from closed malpractice claims reveal a similar delay. Conservative calculations suggest such delays could generate hundreds of millions of dollars in waiting costs each year. Policymakers should be aware of and seek to avoid such costs as they assess whether and how to expand reporting of personal injury awards. 1. Introduction Policymakers have expressed alarm regarding large and growing expenditures on health care by federal and state government. In 2000, federal Medicare outlays accounted for 1.8% of US gross domestic product (GDP), while Medicaid and related state programs consumed 1.3% of GDP (Congressional Budget Office (CBO) 2016). By 2015, these shares had risen to 3.1% for Medicare and 2.2% for Medicaid, a combined increase of 70%. CBO projections suggest that if current policies continue, Medicare and Medicaid expenditures will collectively account for over 8% of GDP by 2040. Recognizing that such cost growth is unsustainable, Democrats and Republicans alike have introduced proposals designed to address the looming fiscal challenges facing both programs. Included among these are various efforts that attempt to offset these programs’ costs from other sources, and in recent years the tort system has become a favored target. When bodily injuries occur due to negligent actions by others, injured parties can seek compensation for their losses from responsible parties through the tort system. This system is overlaid on top of the traditional health insurance system that provides coverage for medical care largely without regard for the underlying cause or fault associated with a particular medical condition. Interactions between the two systems are governed by statutes, regulations, case law, and norms that determine who pays first, what levels of compensation are provided, and what subrogation rights exist. Although government insurers have long been designated as payers of last resort, traditionally they have nonetheless borne the cost of considerable portion of medical care ultimately compensated through the tort system, due to practical barriers in pursuing recovery from tortfeasors or their payees. However, recent legislative and regulatory changes have sought to strengthen the ability of both Medicare and Medicaid to obtain information about tort awards and obtain recoveries from these awards. A key component of reform involves efforts to mandate reporting of bodily injury awards to government insurers. Such reporting would allow government insurers to identify care episodes subject to recovery and issue demands for payment. While such reporting systems clearly promote the fiscal interests of Medicare and Medicaid, they also carry potential for generating other, conceivably more far-reaching impacts on the civil justice system. By requiring additional data collection for all cases, and, for at least some cases, injecting an additional party into settlements, reporting can complicate the negotiation process, delaying the final resolution of cases and increasing administrative cost. Moreover, plaintiff advocates have highlighted the possibility that enhanced recovery by Medicare or Medicaid might render some previously viable claims noneconomical for plaintiffs to pursue, altering the pool of claims in the system, and reducing access to justice. While such impacts are plausible in theory and have been noted by many commentators on this issue,1 to date there has been virtually no empirical evidence demonstrating whether or to what degree such effects occur in actual practice. In this article, I provide one of the first large-scale analyses of the effects of a new reporting requirement on time to claim resolution, one key outcome thought to be affected by such reforms. Time to claim resolution is of particular interest because delay in resolving claims reduces welfare for everyone in the system. Recovery and subrogation primarily involve a transfer from claimants, insurers, and policyholders, and their representatives to Medicare or Medicaid; the appropriate degree of such transfers is largely a normative question about which different stakeholders may disagree. However, other factors being equal, all parties in the system would prefer less delay in resolving cases, as delay increases administrative costs for all parties, increases reserving costs for insurers,2 and reduces the value of settlements for claimants and Medicare and Medicaid due to time preference. Thus, to the extent that introducing a new reporting and recovery process affects delay, these costs must be weighed against the benefits of allocating losses more closely in line with the intended priority rules making Medicare and Medicaid payers of last resort. To assess the impact of new mandatory reporting and recovery requirements, I exploit a Medicare reform enacted at the end of 2007 and implemented in 2011 that for the first time obligated insurers to report personal injury awards involving Medicare beneficiaries to the federal government. Focusing on auto injury claims—a large and economically important subset of the overall tort personal injury space—I measure the impacts of the reform by leveraging a rich, national data set covering thousands of individual personal injury awards. Using a differences-in-differences (DD) research design that contrasts claim resolution times for claimants above and below 65 years of age before and after the reform was implemented, and which also controls for a rich set of claimant and injury characteristics likely to affect claim latency, I demonstrate that the new reporting requirement is associated with a 19% increase in time to claim resolution, or a roughly 2-month delay for the average claim. This estimate is likely conservative and survives across a number of robustness checks. Simple and conservative back-of-the-envelope calculations suggest that mandatory reporting can generate tens or even hundreds of millions of dollars of delay costs each year. While the fiscal and other benefits may in the end justify increased efforts at subrogation, policymakers should be aware of such delay costs as they consider whether to further expand recovery efforts to Medicaid and other government insurance programs. This works generalize contemporaneous work by Helland and Klick (2018) that uses data from a single large auto insurer to analyze the impacts of Medicare reporting. As here, Helland and Klick (2018) find that Medicare reporting increases time to claim resolution, although they identify larger impacts, on the order of 6 months or longer. However, the average claim duration pre-reform in their sample is over 800 days, and nearly half of their postreform claims were censored as of February 2014, suggesting that their sample may have been composed of a comparatively complex set of claims.3 This article provides additional evidence using a more broadly representative sample of auto claims and also demonstrates that the delay effects of the new reporting requirement persist even after controlling for a rich set of underlying claim characteristics. I also extend the analysis to a different personal injury setting—closed malpractice claims—and measure a similar delay in settlement due to Medicare reporting. The article proceeds as follows. Section 2 outlines the difficulties that have historically limited Medicare and Medicaid’s ability to pursue recovery form tort awards and briefly discusses some of the key reforms introduced by Congress and regulatory agencies in an effort to address these problems. Recent developments described in this section suggest that mandatory reporting to facilitate recovery against tort awards may become more widespread in the future, highlighting the need for empirical research on the effects of such requirements. Section 3 describes the data and DD research design and its underlying assumptions. Section 4 presents results from the main analysis relating mandatory reporting to case resolution time along with a series of robustness checks and two falsification tests designed to examine the validity of one of the key assumptions of the research design. I also confirm the basic findings that reporting increases delay in a separate dataset of closed medical malpractice claims. Section 5 discusses the implications of these findings for potential future efforts to mandate reporting. 2. Background and Prior Research By statute, Medicare and Medicaid have long been designated as payers of last resort is many situations,4 bearing responsibility for reimbursing providers for medical care only once other sources of payment beyond the patient have been exhausted. However, a number of practical obstacles have historically limited the ability of these programs to collect payments made for medical care incident to a personal injury. A key difficulty arises due to the timing of tort awards relative to the billing cycle. Following an injury, an injured party will seek medical care, and at the time the care is received, it is not yet clear whether a tort recovery will be available, what care it might cover, and how much will be available from any award to compensate for medical care. Providers, accustomed to fact that Medicaid and Medicare are the presumptive payers in almost all other episodes of medical care and anxious to receive payments for their services, will often bill the health insurers using their usual processes and receive payment. Months or years after payment has been received, when a tort settlement is reached, Medicare and Medicaid in theory could seek recovery5 for their prior payments from either the Medicare beneficiary (injured party) or the tortfeasor, which would allow them to shift the cost of the care from the Medicare trust fund and taxpayers to those directly involved in the injury-causing incident. However, as a practical matter, seeking such recovery was difficult historically for two reasons. First, Medicare and Medicaid had no easy way to ascertain whether a tort award was made in a particular case. Because there is substantial variability from patient to patient in the amount of time it takes to reach a settlement following an injury, there was no set time frame within which Medicare and Medicaid might expect to observe a tort recovery had occurred, and there is little in patient or billing records that might indicate whether a particular care episode is likely to be covered by tort. Moreover, even for patients for whom there is a high likelihood of a tort award (e.g. a patient presenting with an auto crash injury), some fraction will ultimately end up not receiving any award. Thus, Medicare and Medicaid had little ability to identify those cases where a tort award was available from which it might seek recovery. A second obstacle relates to determining which parts of the settlement should be available to Medicare and Medicaid as compensation. While it is not uncommon for injured parties in personal injury cases to itemize various losses (e.g. lost wages, medical care, pain, and suffering) as they present demands for compensation to tortfeasors and their insurers, when a settlement is reached, it is often done so without apportioning its final value into components of economic and noneconomic loss. To take a simple example, in an injury involving a Medicare beneficiary with claimed losses of |${\$}$|50,000 for medical care and |${\$}$|50,000 for pain and suffering, where the ultimate settlement after attorney’s fees is |${\$}$|60,000, absent other information it would appear uncertain whether |${\$}$|50,000, |${\$}$|30,000, or some other amount would be available to Medicare to help it recover its outlays associated with the injury. Because of this ambiguity, Medicare and Medicaid were poorly positioned to know how much recovery might be available to them in a given situation, and therefore which claims might deserve priority and which might be inefficient to pursue. In the Medicaid context, two U.S. Supreme Court decisions related to this second problem served to further dampen incentives to pursue recovery. In Arkansas Dept. of Health and Human Servs. v. Ahlborn 547 U.S. 268 (2006), the Court ruled that Medicaid’s recovery rights were limited to only the portion of the settlement that represents medical care, which in the particular case presented before the Court reduced Medicaid’s entitlement to about 1/6 of the amount asserted by the state. Later, in Wos v E.M.A. 568 US ___ (2013), the Court invalidated a state statute that set the amount of Medicaid recovery at one-third of the tort award, reasoning that such a rule would run afoul of the federal Medicaid law’s prohibition against states filing liens against personal property of Medicaid recipients to recover their costs for the program. Together, the rulings signaled some hostility in the courts to efforts to expand Medicaid’s ability to recovery against personal injury awards. 2.1. The Medicare Reporting System In the early 1980’s, Congress passed legislation designed to compel plaintiffs and defendants to consider Medicare’s interests during settlement negotiations by granting Medicare a right of recovery against either party in the event of a resolution that did not adequately address its statutory position as a secondary payer, with scope for additional damages in the event of noncompliance.6 This legislation and the resultant requirement are commonly referred to as Medicare Secondary Payer (MSP). However, Medicare was poorly positioned to enforce its rights due to its inability to identify cases where personal injury awards had been issued. To address that problem, Congress enacted Public Law 110–173, The Medicare, Medicaid, and SCHIP Extension Act of 2007 (MMSEA), which established a reporting system designed to enabled Medicare to track beneficiaries who had received personal injury award payments. Under the new law, liability insurers were required to report information about bodily injury award payments made to any individual who was a Medicare beneficiary, with substantial penalties for omissions or noncompliance. As a practical matter, this meant that insurers needed to query Medicare to determine beneficiary status of anyone who filed a tort claim, and then, for those who were beneficiaries, ensure in consultation with the claimant’s attorney that Medicare’s interests could be satisfied prior to the finalization of a settlement, so as to foreclose the possibility that, after receiving the mandated report, Medicare might seek additional unforeseen payments, and penalties from either party. The new law thus both initiated a reporting process and triggered a heightened effort to involve Medicare in the settlement of liability claims. Implementation of the new law was not without difficulty. After its passage, insurers and plaintiff attorneys reported difficulties in getting information from the Centers for Medicare and Medicaid Services (CMS) needed to resolve claims, and there were substantial problems with the IT and other systems CMS established to handle queries and reporting.7 As a result of these difficulties, the original implementation date for mandatory reporting by liability insurers was moved back from July 2009 to January 2011 for no-fault insurers and January 2012 for liability insurers, and Congress enacted additional legislation in 2012 to improve the process.8 CMS reported recovering |${\$}$|6 billion through MSP in 2006, which rose to |${\$}$|8 billion by 2012 (Kirchoff 2014), an increase roughly in line with Medicare’s overall growth over the same period. Thus, the extent to which the new reporting requirements helped increase Medicare’s recovery from third-party payers remains somewhat ambiguous. It is also not well understood how these provisions affected settlement times or other legal outcomes of interest, the issue explored in this article. 2.2. Is Medicaid the Next Wave? One reason the Medicare reforms hold relevance for future policy is that the success of the reporting requirements has prompted policymakers to consider expanding reporting to other government insurance programs, most notably Medicaid. In response to the Wos v. E.M.A. decision, Congress included a provision in § 202 of Public Law 113–67, the Bipartisan Budget Act of 2013, that grants Medicaid recovery rights for its outlays up to the full value of the settlement in a personal injury case, removing the Supreme Court’s restriction that Medicaid access only that portion of the settlement designated for medical care. Thus, in situations where most or all of the settlement is allocated for general damages, wage loss, or other nonmedical loss, under the new law Medicaid would enjoy much broader rights of recovery. This provision, which initially went into effect in October 2017 but was then repealed when the Bipartisan Budget Act of 2018 was passed in February 2018,9 would have substantially increased the dollars available to state Medicaid programs through third-party recovery efforts. Whether future Congresses will choose to restore Medicaid’s enhanced recovery rights remains an open question. Because Medicaid is largely administered by states rather than the federal government, a blanket reporting requirement such as that that was introduced for Medicare is infeasible. Nevertheless, some individual states have begun pursuing reporting systems and requirements patterned after the Medicare system. In 2012, for example, Rhode Island mandated liability insurer reporting through a newly developed Medical Assistance Intercept System (MAIS),10 allowing it to subrogate against injury awards made to state Medicaid beneficiaries. As of 2016, the state reported collections of |${\$}$|3.2 million due to the new reporting program. In 2015, the National Conference of Insurance Legislators (NCOIL) also adopted and began promulgating a model state law establishing a mandatory Medicaid reporting,11 and legislation patterned after the model law has been proposed in some states. 2.3. Should We Expect Reporting to Affect Speed of Claim Resolution? The introduction of new reporting requirements, as occurred for Medicare and as seems to be beginning for Medicaid, introduces several new elements to claim process that might affect the time needed to resolve a claim. First, insurers must query the government insurer to determine whether a claimant in their system is a beneficiary and therefore potentially subject to subrogation. While at first glance it may seem that the use of an electronic querying system should allow such queries to occur nearly instantaneously—and therefore not contribute to delay—all querying systems require the ability to uniquely identify and match individuals, which, in the U.S. context, means collecting unique identifying information such as a Social Security Number (SSN) which traditionally would not be required to settle a claim. In other words, the query process itself requires compiling additional information from the claimant which may take some time to collect. For those claimants who are government health insurance beneficiaries, the health insurer must then examine billing records in its possession in an effort to identify care episodes that were related to the bodily injury in question, and then indicate its intention to request reimbursement from the parties for that care. A challenge here is that medical billing systems often do not capture the information necessary to determine whether a particular care episode arose due to a tort or for some other reason, and the government insurer has incentives to be expansive in its requests for reimbursement. Following the initial demand, the governmental insurer and parties to the claim must undertake a reconciliation process to determine which episodes of care are subject to subrogation, and doing so may require multiple additional requests for information to the claimant and her health care providers. Moreover, once the parties have determined which care episodes are subject to subrogation, there is still scope for negotiation over the appropriate reimbursement rates, with the claimant facing incentives to minimize subrogation payments in order maximize her take-home award, and the government insurers facing opposite incentive to maximize their recovery by depleting the settlement. Once the liability insurer and claimant have determined what portion of the settlement will be paid to the government insurer, then they are in a position to negotiate a final settlement. Each of these steps in the process can further delay settlement. Contemporary accounts produced around the time of the passage of the new reporting requirements highlight some of the difficulties that arose once Medicare became more involved in settling claims. Andrews (2013), for example, described a case involving an 80-year-old auto injury victim who had to wait over a year for Medicare to calculate how much it was owed in his case; after the yearlong wait, it submitted a demand letter asking for nearly 10 times what it was actually owed. In another account, a plaintiff’s attorney described a case in which the total settlement amount was less than the attorney’s expenses in pursuing the case; Medicare nonetheless required documentation of the contingency fee contract and proposed settlement before it would relinquish its rights to collection, delaying the final resolution of the case (Hagy 2010). In testimony before Congress, another plaintiff attorney described numerous difficulties in getting clear information from Medicare necessary to settle cases, including in an ongoing case in which a |${\$}$|400,000 personal injury settlement obtained on behalf of an elderly client in failing health was sitting in escrow because of Medicare’s inaction (Matzus 2011). While there are clear reasons to imagine that the introduction of reporting and subrogation requirements could delay the resolution of claims, to date there has been virtually no empirical work documenting the existence of or magnitude of such effects. To my knowledge, there is a single existing study that attempts to measure the effects of Medicare reporting on settlement delay. Using a DD research design that also leverages the introduction of the new reporting requirement, Helland and Klick (2018) find that reporting increases the time to case resolution in auto injury cases by about 6 months. An important limitation of that study is its use of data from only a single large insurer; this study expands upon that research in that it includes larger and more nationally representative set of insurers. Additionally, due to data limitations, the unaffected control group in the Helland and Klick study constitutes less than 1% of the overall sample; in the present study, there are substantial numbers of claimants not subject to the reporting requirement who can provide a clear counterfactual for what might have happened to the Medicare patients post-2011 had no reporting requirement been in place. 3. Data and Empirical Approach Analyzing the effects of mandatory reporting requires access to data that includes otherwise similar personal injury claims, some of which were and were not subject to a reporting requirement. My data are drawn from the Insurance Research Council’s 2007 and 2012 Closed Claim databases. These databases include abstracted information from a sample of auto liability claims closed in the given year from all 50 states and the District of Columbia; participating insurers comprise more than half of the U.S. personal passenger automobile market, and therefore these data are likely broadly representative of auto tort claims in the U.S. For each claim, we observe the date the claim was filed and the date of final resolution (which can be used to construct the time to resolution), claimant demographic information, information about the nature and severity of the accident, claimed injuries and other financial losses, medical treatment received, and the total claimed loss. We also observe the final claim payment amount.12 For the bulk of the analysis, I focus attention on claimants aged 50 and over filing bodily injury claims. Because this is a closed claim sample, there is no censoring of claim times, moreover, because the payment date rather than the claim initiation date is the operative date for triggering the reporting requirements, we know that claims closed in 2007 were not subject to the Medicare reporting requirements, whereas those in 2012 were. For the 2012 claims, because claims would have been reportable to Medicare under the new requirements, participants were incentivized to query Medicare in advance to determine whether claimants were beneficiaries and, if so, to attempt to establish the amounts of conditional payments that would need to be reimbursed to Medicare prior to settling the claim. In the case of claimants over 65, the vast majority were beneficiaries and would need to undergo the reconciliation process or risk settling a claim but then needing to reimburse Medicare for its conditional payments separately, either in addition to the full claim payment on the insurer side, or by paying out of the settlement for the claimant. To measure the effects of the reporting requirement, I exploit the timing of the introduction of the requirement coupled with the fact that Medicare is targeted primarily to seniors aged 65 and over. I adopt a DD research design that contrasts claim resolution time for individuals above and below age 65 at the time of settlement, before and after the introduction of the new reporting requirement. In particular, I estimate regressions of the following form: $$\begin{align}\label{eq1} &\textit{SettlementTime}_{i}\nonumber\\ &\quad{} = \alpha \cdot (\textit{Age}_{i}>65 \times \textit{Post}_{i}) + \beta_{1} \cdot \textit{Age} + \beta_{2} \cdot \textit{Post}_{i} + \beta_{3} \cdot X_{i}, \end{align} $$(1) where SettlementTime|$_{i}$| represents the time in days required to settle the claim filed by individual |$i$|⁠, Age|$_{i} >65 \times \textit{Post}_{i}$| is an interaction term equal to |$1$| for individuals ages 65 and older with claims closed after 2011, when the requirement was enacted, Age represents a vector of claimant age fixed effects, Post|$_{i}$| is a dummy variable for a claim closed in 2012 (as opposed to 2007), and |$X_{i}$| a vector of additional controls measured at the individual level. In this regression, |$\alpha $| yields the DD estimate of the effect of reporting and subrogation on case resolution time. Given the long right tail for case settlement times, in most specifications I implement this equation using log settlement time as the outcome. Table 1 reports summary statistics capturing average outcomes for the overall sample and by age group and year of claim. For claimants under age 65, there was a reduction in various indicia of claim severity (rate of disability, lost work days, emergency room (ER), and physician use, etc.) at least on average between 2007 and 2012, and a corresponding reduction in the average time required to resolve the claim. Older claimants also saw an apparent decline in the seriousness of injury and complexity of claims, but their average claim resolution time increased between 2007 and 2012. Comparing the young and old, as expected, older claimants are less likely to be employed, but the groups actually have roughly similar patterns of disability and medical care utilization in these data. Table 1. Summary Statistics . . . . Average for claimants: . . Overall . 2007 . 2012 . Characteristic . |$N$| . Mean . SD . Age |$<$| 65 . Age |$ \ge $| 65 . Age |$<$| 65 . Age |$ \ge $| 65 . Time to claim resolution (days) 9,431 422.2 434.0 433.8 420.6 403.1 444.2 Age (years) 10,058 60.8 9.26 55.9 73.5 56.0 72.6 Male 10,017 0.431 0.495 0.424 0.422 0.444 0.423 Employed 7,357 0.587 0.492 0.751 0.238 0.753 0.226 Degree of fault (%) 10,044 3.64 15.1 3.62 4.49 3.35 3.60 Accident location  Large city 10,002 0.355 0.479 0.367 0.309 0.369 0.335  Medium city 10,002 0.349 0.477 0.320 0.320 0.366 0.403  Suburb/town/rural 10,002 0.296 0.457 0.313 0.371 0.264 0.262  Impact severity  None/minor 9,489 0.294 0.456 0.279 0.211 0.334 0.308  Moderate 9,489 0.529 0.499 0.542 0.542 0.520 0.505  Major 9,489 0.178 0.382 0.179 0.247 0.146 0.187 Most serious injury  Sprain/strain 9,748 0.676 0.468 0.704 0.563 0.718 0.617  Knee, disc, or shoulder injury 9,748 0.120 0.325 0.125 0.160 0.105 0.107  Fracture 9,748 0.059 0.236 0.049 0.094 0.048 0.079  Other 9,748 0.144 0.351 0.122 0.183 0.129 0.197 Extent of disability  None 9,846 0.737 0.440 0.726 0.729 0.737 0.770  Temporary 9,846 0.204 0.403 0.205 0.183 0.221 0.180  Permanent 9,846 0.052 0.221 0.059 0.073 0.038 0.045 Medical care received  Went to ER 10,058 0.421 0.494 0.429 0.485 0.388 0.425  Imaging performed 10,058 0.638 0.481 0.681 0.733 0.571 0.607  Visited chiropractor 10,058 0.340 0.474 0.370 0.257 0.365 0.284  Visited physical therapist 10,058 0.227 0.419 0.248 0.254 0.211 0.192 Total doctor visits 10,058 10.5 17.0 11.8 9.75 10.2 8.46 Number of lost work days 10,058 14.8 182.2 27.3 14.61 7.6 2.10 Hired attorney 9,982 0.499 0.500 0.506 0.481 0.504 0.484 Claimed loss amount ( |${\$}$|⁠) 9,007 |${\$}$|14,398 41917 |${\$}$|12,135 |${\$}$|14,639 |${\$}$|15,549 |${\$}$|16,980 Payment amount ( |${\$}$|⁠) 10,058 |${\$}$|14,450 30665 |${\$}$|14,011 |${\$}$|14,922 |${\$}$|14,171 |${\$}$|15,750 . . . . Average for claimants: . . Overall . 2007 . 2012 . Characteristic . |$N$| . Mean . SD . Age |$<$| 65 . Age |$ \ge $| 65 . Age |$<$| 65 . Age |$ \ge $| 65 . Time to claim resolution (days) 9,431 422.2 434.0 433.8 420.6 403.1 444.2 Age (years) 10,058 60.8 9.26 55.9 73.5 56.0 72.6 Male 10,017 0.431 0.495 0.424 0.422 0.444 0.423 Employed 7,357 0.587 0.492 0.751 0.238 0.753 0.226 Degree of fault (%) 10,044 3.64 15.1 3.62 4.49 3.35 3.60 Accident location  Large city 10,002 0.355 0.479 0.367 0.309 0.369 0.335  Medium city 10,002 0.349 0.477 0.320 0.320 0.366 0.403  Suburb/town/rural 10,002 0.296 0.457 0.313 0.371 0.264 0.262  Impact severity  None/minor 9,489 0.294 0.456 0.279 0.211 0.334 0.308  Moderate 9,489 0.529 0.499 0.542 0.542 0.520 0.505  Major 9,489 0.178 0.382 0.179 0.247 0.146 0.187 Most serious injury  Sprain/strain 9,748 0.676 0.468 0.704 0.563 0.718 0.617  Knee, disc, or shoulder injury 9,748 0.120 0.325 0.125 0.160 0.105 0.107  Fracture 9,748 0.059 0.236 0.049 0.094 0.048 0.079  Other 9,748 0.144 0.351 0.122 0.183 0.129 0.197 Extent of disability  None 9,846 0.737 0.440 0.726 0.729 0.737 0.770  Temporary 9,846 0.204 0.403 0.205 0.183 0.221 0.180  Permanent 9,846 0.052 0.221 0.059 0.073 0.038 0.045 Medical care received  Went to ER 10,058 0.421 0.494 0.429 0.485 0.388 0.425  Imaging performed 10,058 0.638 0.481 0.681 0.733 0.571 0.607  Visited chiropractor 10,058 0.340 0.474 0.370 0.257 0.365 0.284  Visited physical therapist 10,058 0.227 0.419 0.248 0.254 0.211 0.192 Total doctor visits 10,058 10.5 17.0 11.8 9.75 10.2 8.46 Number of lost work days 10,058 14.8 182.2 27.3 14.61 7.6 2.10 Hired attorney 9,982 0.499 0.500 0.506 0.481 0.504 0.484 Claimed loss amount ( |${\$}$|⁠) 9,007 |${\$}$|14,398 41917 |${\$}$|12,135 |${\$}$|14,639 |${\$}$|15,549 |${\$}$|16,980 Payment amount ( |${\$}$|⁠) 10,058 |${\$}$|14,450 30665 |${\$}$|14,011 |${\$}$|14,922 |${\$}$|14,171 |${\$}$|15,750 Notes: This table reposts summary statistics for the 2007 and 2012 Insurance Research Council Closed Auto Claim databases. The sample is limited to bodily injury claims involving claimants ages 50 and above at the time of settlement. Open in new tab Table 1. Summary Statistics . . . . Average for claimants: . . Overall . 2007 . 2012 . Characteristic . |$N$| . Mean . SD . Age |$<$| 65 . Age |$ \ge $| 65 . Age |$<$| 65 . Age |$ \ge $| 65 . Time to claim resolution (days) 9,431 422.2 434.0 433.8 420.6 403.1 444.2 Age (years) 10,058 60.8 9.26 55.9 73.5 56.0 72.6 Male 10,017 0.431 0.495 0.424 0.422 0.444 0.423 Employed 7,357 0.587 0.492 0.751 0.238 0.753 0.226 Degree of fault (%) 10,044 3.64 15.1 3.62 4.49 3.35 3.60 Accident location  Large city 10,002 0.355 0.479 0.367 0.309 0.369 0.335  Medium city 10,002 0.349 0.477 0.320 0.320 0.366 0.403  Suburb/town/rural 10,002 0.296 0.457 0.313 0.371 0.264 0.262  Impact severity  None/minor 9,489 0.294 0.456 0.279 0.211 0.334 0.308  Moderate 9,489 0.529 0.499 0.542 0.542 0.520 0.505  Major 9,489 0.178 0.382 0.179 0.247 0.146 0.187 Most serious injury  Sprain/strain 9,748 0.676 0.468 0.704 0.563 0.718 0.617  Knee, disc, or shoulder injury 9,748 0.120 0.325 0.125 0.160 0.105 0.107  Fracture 9,748 0.059 0.236 0.049 0.094 0.048 0.079  Other 9,748 0.144 0.351 0.122 0.183 0.129 0.197 Extent of disability  None 9,846 0.737 0.440 0.726 0.729 0.737 0.770  Temporary 9,846 0.204 0.403 0.205 0.183 0.221 0.180  Permanent 9,846 0.052 0.221 0.059 0.073 0.038 0.045 Medical care received  Went to ER 10,058 0.421 0.494 0.429 0.485 0.388 0.425  Imaging performed 10,058 0.638 0.481 0.681 0.733 0.571 0.607  Visited chiropractor 10,058 0.340 0.474 0.370 0.257 0.365 0.284  Visited physical therapist 10,058 0.227 0.419 0.248 0.254 0.211 0.192 Total doctor visits 10,058 10.5 17.0 11.8 9.75 10.2 8.46 Number of lost work days 10,058 14.8 182.2 27.3 14.61 7.6 2.10 Hired attorney 9,982 0.499 0.500 0.506 0.481 0.504 0.484 Claimed loss amount ( |${\$}$|⁠) 9,007 |${\$}$|14,398 41917 |${\$}$|12,135 |${\$}$|14,639 |${\$}$|15,549 |${\$}$|16,980 Payment amount ( |${\$}$|⁠) 10,058 |${\$}$|14,450 30665 |${\$}$|14,011 |${\$}$|14,922 |${\$}$|14,171 |${\$}$|15,750 . . . . Average for claimants: . . Overall . 2007 . 2012 . Characteristic . |$N$| . Mean . SD . Age |$<$| 65 . Age |$ \ge $| 65 . Age |$<$| 65 . Age |$ \ge $| 65 . Time to claim resolution (days) 9,431 422.2 434.0 433.8 420.6 403.1 444.2 Age (years) 10,058 60.8 9.26 55.9 73.5 56.0 72.6 Male 10,017 0.431 0.495 0.424 0.422 0.444 0.423 Employed 7,357 0.587 0.492 0.751 0.238 0.753 0.226 Degree of fault (%) 10,044 3.64 15.1 3.62 4.49 3.35 3.60 Accident location  Large city 10,002 0.355 0.479 0.367 0.309 0.369 0.335  Medium city 10,002 0.349 0.477 0.320 0.320 0.366 0.403  Suburb/town/rural 10,002 0.296 0.457 0.313 0.371 0.264 0.262  Impact severity  None/minor 9,489 0.294 0.456 0.279 0.211 0.334 0.308  Moderate 9,489 0.529 0.499 0.542 0.542 0.520 0.505  Major 9,489 0.178 0.382 0.179 0.247 0.146 0.187 Most serious injury  Sprain/strain 9,748 0.676 0.468 0.704 0.563 0.718 0.617  Knee, disc, or shoulder injury 9,748 0.120 0.325 0.125 0.160 0.105 0.107  Fracture 9,748 0.059 0.236 0.049 0.094 0.048 0.079  Other 9,748 0.144 0.351 0.122 0.183 0.129 0.197 Extent of disability  None 9,846 0.737 0.440 0.726 0.729 0.737 0.770  Temporary 9,846 0.204 0.403 0.205 0.183 0.221 0.180  Permanent 9,846 0.052 0.221 0.059 0.073 0.038 0.045 Medical care received  Went to ER 10,058 0.421 0.494 0.429 0.485 0.388 0.425  Imaging performed 10,058 0.638 0.481 0.681 0.733 0.571 0.607  Visited chiropractor 10,058 0.340 0.474 0.370 0.257 0.365 0.284  Visited physical therapist 10,058 0.227 0.419 0.248 0.254 0.211 0.192 Total doctor visits 10,058 10.5 17.0 11.8 9.75 10.2 8.46 Number of lost work days 10,058 14.8 182.2 27.3 14.61 7.6 2.10 Hired attorney 9,982 0.499 0.500 0.506 0.481 0.504 0.484 Claimed loss amount ( |${\$}$|⁠) 9,007 |${\$}$|14,398 41917 |${\$}$|12,135 |${\$}$|14,639 |${\$}$|15,549 |${\$}$|16,980 Payment amount ( |${\$}$|⁠) 10,058 |${\$}$|14,450 30665 |${\$}$|14,011 |${\$}$|14,922 |${\$}$|14,171 |${\$}$|15,750 Notes: This table reposts summary statistics for the 2007 and 2012 Insurance Research Council Closed Auto Claim databases. The sample is limited to bodily injury claims involving claimants ages 50 and above at the time of settlement. Open in new tab The rich information available regarding each claim permits us to control for a range of factors shown in Table 1 likely to impact the complexity of the claim, including the demographics of the claimant, accident characteristics (e.g. location, number of vehicles, impact severity), nature and severity of injury (45 variables), types and amounts of medical care received (38 variables), and liability characteristics (e.g. policy limits, degree of fault of insurer, attorney involvement). Conceptually, this means that we are able to compare two claimants involved in similar accidents, with similar injuries and medical treatment, facing legally comparable claims, but who differ in whether or not their claim must undergo the reporting and subrogation process. While my primary outcome of interest is time to claim resolution, I also estimate versions of the regression specification above where I use the payment amount or number of claims as the outcome. Characteristics of state tort law, such as whether the state is a no-fault state, rules governing bad faith claims, and contingency fee rules are also likely to affect the claims resolution process. For example, states vary in their treatment of the collateral source rule, a doctrine governing whether payments from other sources such as health insurance can be considering in determining the amount of damages in a personal injury suit. It seems plausible to imagine that the status of the collateral source rule might affect which claims are filed and how long they take to resolve. Because the identification strategy relies on contrasts across age groups, I can include in the regressions as additional controls a full set of state by year of injury fixed effects. These controls mitigate any potential omitted variable bias from state-specific factors that change over time, including features of tort law such as the collateral source rule along with any other laws and regulations related to general medical care or vehicle safety. The main underlying assumption of the DD analysis is that, after adjusting for underlying claim characteristics, the claims involving individuals ages 50–64 provide an appropriate counterfactual for what claim resolution times for the older group would have been, absent the new reporting requirement. These estimates of the effect of reporting and subrogation on delay are likely to be conservative for two reasons. First, as noted previously, a new reporting requirement affects all future claims in that they must, at a minimum, have claimant information entered into a query system to establish whether further reporting is necessary. Thus, in the analysis, the control group is in fact affected by the new requirement, but to a lesser extent than those who are Medicare eligible. We imagine that, once a sufficient electronic query process is in place, the delay effects of querying alone are small relative to the effects of querying, receiving a hit, and then the subsequent reporting and reconciliation process. The research design is best suited to measure the difference in claim resolution time between otherwise similar claims that required querying and reporting as compared to querying alone. In other words, to the extent that the query process itself generates delay, this analysis is not ideally suited to capture such effects. A second reason these estimates are conservative is that some portion of the under 65 population are in fact Medicare eligible, and therefore would have to complete the entire reporting process. Because Medicare eligibility is unobserved in the data I cannot directly account for this; however, data from the American Community Survey and other sources (e.g. Card et al. 2009) suggest that roughly 10% of adults ages 55–64 are Medicare eligible. This, a modest fraction of the individuals who are labeled as controls in this analysis actually undergo the entire reporting process; this misclassification would tend to attenuate any differences estimated using a DD analysis. Similarly, a small segment of the 65 and older population are noncitizens or otherwise ineligible for Medicare; Barnett and Vornovitsky (2016) estimate this population to comprise about 6% of those ages 65 and older. 4. Results Table 2 reports estimates of the effect of mandatory reporting and subrogation on the amount of time needed to settle a case. Column I reports results from a simple DD specification as described above with age and year of claim fixed effects but no further controls. The estimated coefficient of 0.178, which is statistically significant at the 1% level,13 implies that the new reporting requirement is associated with a 19% increase in the amount of time needed to make final payment on a claim. Relative to the mean, this represents an increase of 75 days to resolve a claim. Table 2. Baseline Estimates of the Effects of Reporting on Claim Resolution Time . I . II . III . IV . V . DD estimate 0.178*** 0.203*** 0.215*** 0.188*** 0.176*** (0.065) (0.060) (0.056) (0.053) (0.053) |$N$| 9,431 9,419 9,419 9,419 9,419 |$R^{2}$| 0.009 0.184 0.328 0.389 0.414 Control for claimant demographics, accident, and injury characteristics? N Y Y Y Y Control for medical care received? N N Y Y Y Control for attorney representation? N N N Y Y Include state/year fixed effects? N N N N Y . I . II . III . IV . V . DD estimate 0.178*** 0.203*** 0.215*** 0.188*** 0.176*** (0.065) (0.060) (0.056) (0.053) (0.053) |$N$| 9,431 9,419 9,419 9,419 9,419 |$R^{2}$| 0.009 0.184 0.328 0.389 0.414 Control for claimant demographics, accident, and injury characteristics? N Y Y Y Y Control for medical care received? N N Y Y Y Control for attorney representation? N N N Y Y Include state/year fixed effects? N N N N Y Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on personal injury claim resolution speed. The unit of observation is a claim, and the outcome variable is the logged number of days between the initial injury report and final claim payment. Claimants 65 and over who had claims resolved in 2012 are the treated group subject to mandatory reporting; differences are taken across age groups and across years. Column I includes as controls a set of age of claimant fixed effects (51 categories) and year of claim fixed effects (2 categories). Column II adds controls for policy limits (9 categories), accident location (urban/rural, 5 categories), number of vehicles involved (3 categories), whether there were passengers in the vehicle, impact severity (5 categories), injury severity at scene (6 categories); claimant role in the accident (driver/passenger/etc., 7 categories), relationship to insured (3 categories), sex, employment status (3 categories), degree of fault (7 categories), presence of injuries (25 categories), most severe injury (22 categories), and disability status (3 categories). Column III adds controls for whether the claimant used various medical services (21 categories) and the number of visits made to each of 16 different categories of providers. Column IV adds controls for whether the claimant hired an attorney. Column V adds a full set of state of accident by claim year interactions (102 categories) as controls. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 2. Baseline Estimates of the Effects of Reporting on Claim Resolution Time . I . II . III . IV . V . DD estimate 0.178*** 0.203*** 0.215*** 0.188*** 0.176*** (0.065) (0.060) (0.056) (0.053) (0.053) |$N$| 9,431 9,419 9,419 9,419 9,419 |$R^{2}$| 0.009 0.184 0.328 0.389 0.414 Control for claimant demographics, accident, and injury characteristics? N Y Y Y Y Control for medical care received? N N Y Y Y Control for attorney representation? N N N Y Y Include state/year fixed effects? N N N N Y . I . II . III . IV . V . DD estimate 0.178*** 0.203*** 0.215*** 0.188*** 0.176*** (0.065) (0.060) (0.056) (0.053) (0.053) |$N$| 9,431 9,419 9,419 9,419 9,419 |$R^{2}$| 0.009 0.184 0.328 0.389 0.414 Control for claimant demographics, accident, and injury characteristics? N Y Y Y Y Control for medical care received? N N Y Y Y Control for attorney representation? N N N Y Y Include state/year fixed effects? N N N N Y Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on personal injury claim resolution speed. The unit of observation is a claim, and the outcome variable is the logged number of days between the initial injury report and final claim payment. Claimants 65 and over who had claims resolved in 2012 are the treated group subject to mandatory reporting; differences are taken across age groups and across years. Column I includes as controls a set of age of claimant fixed effects (51 categories) and year of claim fixed effects (2 categories). Column II adds controls for policy limits (9 categories), accident location (urban/rural, 5 categories), number of vehicles involved (3 categories), whether there were passengers in the vehicle, impact severity (5 categories), injury severity at scene (6 categories); claimant role in the accident (driver/passenger/etc., 7 categories), relationship to insured (3 categories), sex, employment status (3 categories), degree of fault (7 categories), presence of injuries (25 categories), most severe injury (22 categories), and disability status (3 categories). Column III adds controls for whether the claimant used various medical services (21 categories) and the number of visits made to each of 16 different categories of providers. Column IV adds controls for whether the claimant hired an attorney. Column V adds a full set of state of accident by claim year interactions (102 categories) as controls. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Columns II–V of Table 2 progressively introduce additional sets of control variables to the DD specification. Column II adds controls for claimant demographics, accident characteristics (e.g. severity), detailed measures of claimant injury, and some of the legal characteristics of the case (e.g. estimated degree of fault of the claimant). Column III adds numerous further controls for the types of medical care received by the claimant. While it seems unlikely that the volume of treatment received should respond to the reporting requirement because treatment is largely determined prior to the point at which the reconciliation process with Medicare begins, in theory it is possible that reporting could affect treatment, in which case controlling for treatment would be problematic. However, an advantage of controlling for medical care received is that it seems reasonable to expect that, for a given injury, older patients who may be in worse baseline health would likely require more treatment than younger patients, and these controls would account for any such differences. In any case, across both variants, the relationship between the reporting requirement and delay remains statistically significant and of similar magnitude. Column IV adds additional controls for whether the claimant hired an attorney—another potentially endogenous variable, but also one that is likely to have important impacts on the speed of settlement. Although controlling for attorney presence appreciably increases the |$R^{2}$| of the regression, it reassuringly does not affect the estimated impact of the reporting requirement. In my preferred estimate in column V—which adds a full set of state and year fixed effects, and is therefore robust to unobserved state-level legal and other environmental changes over time—mandatory reporting is associated with a 19% increase in the time needed to resolve a claim.14 When calculated as an average marginal effect, this represents a 58 day delay. Despite the fact that this is an individual-level regression, the available controls explain a sizeable portion of the overall variance in claim resolution time, suggesting that we observe many of the factually and legally relevant factors that affect how quickly claims can be resolved. Although the most straightforward interpretation of the results above is that the new reporting requirement increased delay, an alternative interpretation is that the requirement did not directly affect delay, but rather altered the composition of claims within the system so as to reduce the average claim duration. One simple story would be that the possibility of Medicare subrogation made some lower dollar claims no longer economical for a claimant or their attorneys to pursue, because once Medicare removed its portion of the settlement, the remaining amount going to the plaintiff would no longer cover the costs of pursuing the claim. If such lower value claims were also easier to resolve, then after they were removed from the system, we would expect to observe longer average settlement times for the remaining claims. Note that under this scenario as before, the introduction of the requirement does in fact affect claims; however, under the conventional interpretation, its takes a given claim and prolongs it, whereas under the latter interpretation, it drives some claims out of the system. To examine this possibility, in Table 3, I report coefficients from regressions similar to those presented previously but using the log total payment as the outcome variable. In these regressions, we observe a 9% increase in the total payment as a result of the reform. This regression coefficient combines two impacts—any selection effect that might occur as claims drop out of the system, and any change in the final settlement amount that would be induced by the greater possibility of subrogation by Medicare. This latter quantity seems likely to, if anything, be positive—following the reporting change, the claimant would have to obtain a higher settlement in order achieve the same level of take-home compensation because of the subrogation. Thus both effects would move in the same direction. Table 3. Effects of Reporting on Payment Amount and Claim Volume . Log(total payment) . Number of claims . DD estimate 0.089** 0.48 (0.042) (6.89) |$N$| 10,046 82 |$R^{2}$| 0.658 0.991 . Log(total payment) . Number of claims . DD estimate 0.089** 0.48 (0.042) (6.89) |$N$| 10,046 82 |$R^{2}$| 0.658 0.991 Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on additional claim outcomes. Column I reports a regression similar to the baseline where the dependent variable is the log total claim payment in dollars. Here the specification is as in column V of Table 2; see notes for Table 2. Column II reports coefficients from simple DD regressions where the unit of observation is an age/claim year cell and the outcome is the number of claims in the database; this regression includes age and year fixed effects as controls. The mean of the dependent variable for column 2 is 127. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 3. Effects of Reporting on Payment Amount and Claim Volume . Log(total payment) . Number of claims . DD estimate 0.089** 0.48 (0.042) (6.89) |$N$| 10,046 82 |$R^{2}$| 0.658 0.991 . Log(total payment) . Number of claims . DD estimate 0.089** 0.48 (0.042) (6.89) |$N$| 10,046 82 |$R^{2}$| 0.658 0.991 Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on additional claim outcomes. Column I reports a regression similar to the baseline where the dependent variable is the log total claim payment in dollars. Here the specification is as in column V of Table 2; see notes for Table 2. Column II reports coefficients from simple DD regressions where the unit of observation is an age/claim year cell and the outcome is the number of claims in the database; this regression includes age and year fixed effects as controls. The mean of the dependent variable for column 2 is 127. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab To further examine any selection effects, in the second specification, I collapse the data to the claimant age/year level and estimate a DD regression on the collapsed data to assess whether there was a change in the number of claims originating from those subject to the new reporting requirement. This provides a more direct test for whether the number of claims fell for the target population after the reporting requirement was introduced. There is no measurable change in the volume of claims due to the reform, although admittedly the estimate is somewhat imprecise. Together these results suggest that the new requirement may have caused modest changes in the composition of claims, although the data are not fully dispositive.15 Table 4 reports a series of robustness checks designed to test the sensitivity of the results to changes to the sample and specification. For reference, the first row reports the baseline results. Specification 1 adds flexible controls16 for the log total payment amount as additional controls. Payment amounts might be endogenous, which is why I do not control for them in the preferred specification. However, given that more complex claims seem likely to result in higher payments but also take longer to resolve, controlling flexibly for payment amount may provide a useful means to partly account for unobservable age-related differences in the complexity of claims. While including these controls appreciably increases the share of the variance of the outcome that is explained by the model, it yields a similar estimated effect of reporting. Table 4. Robustness Checks . DD Estimate . |$N$| . |$R^{2}$| . 0. Baseline estimate 0.176*** 9,419 0.414 (0.053) 1. Include controls for total payment amount 0.146*** 9,419 0.516 (0.048) 2. Estimate in levels rather than logs 54.3*** 9,419 0.366 (16.3) 3. Matching model 0.086** 9,171 (0.037) 4. Inverse probability weighting model 0.109** 4,853 (0.048) . DD Estimate . |$N$| . |$R^{2}$| . 0. Baseline estimate 0.176*** 9,419 0.414 (0.053) 1. Include controls for total payment amount 0.146*** 9,419 0.516 (0.048) 2. Estimate in levels rather than logs 54.3*** 9,419 0.366 (16.3) 3. Matching model 0.086** 9,171 (0.037) 4. Inverse probability weighting model 0.109** 4,853 (0.048) Notes: This table reports coefficient estimates from alternatives to the baseline specification. Except as otherwise noted, the specification is as in column V of Table 2; see notes for Table 2. Specification 1 includes the log total payment amount interacted with indicators for deciles of the payment amount as additional controls. Specification 2 uses the claim time measured in days as the outcome variable rather than log(claim time). Specification 3 implements a matching model by first estimating the predicted claim time in days as a function of all the covariates in the model using Poisson regression, and then, in a second stage, modeling log claim time as a function of treatment and a full set of fixed effects for predicted claim time (in days) as controls. Specification 4 limits the sample to 2012 claimants; models the probability of treatment as a function of the other covariates in the model, and then estimates treatment effects via inverse probability weighting. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 4. Robustness Checks . DD Estimate . |$N$| . |$R^{2}$| . 0. Baseline estimate 0.176*** 9,419 0.414 (0.053) 1. Include controls for total payment amount 0.146*** 9,419 0.516 (0.048) 2. Estimate in levels rather than logs 54.3*** 9,419 0.366 (16.3) 3. Matching model 0.086** 9,171 (0.037) 4. Inverse probability weighting model 0.109** 4,853 (0.048) . DD Estimate . |$N$| . |$R^{2}$| . 0. Baseline estimate 0.176*** 9,419 0.414 (0.053) 1. Include controls for total payment amount 0.146*** 9,419 0.516 (0.048) 2. Estimate in levels rather than logs 54.3*** 9,419 0.366 (16.3) 3. Matching model 0.086** 9,171 (0.037) 4. Inverse probability weighting model 0.109** 4,853 (0.048) Notes: This table reports coefficient estimates from alternatives to the baseline specification. Except as otherwise noted, the specification is as in column V of Table 2; see notes for Table 2. Specification 1 includes the log total payment amount interacted with indicators for deciles of the payment amount as additional controls. Specification 2 uses the claim time measured in days as the outcome variable rather than log(claim time). Specification 3 implements a matching model by first estimating the predicted claim time in days as a function of all the covariates in the model using Poisson regression, and then, in a second stage, modeling log claim time as a function of treatment and a full set of fixed effects for predicted claim time (in days) as controls. Specification 4 limits the sample to 2012 claimants; models the probability of treatment as a function of the other covariates in the model, and then estimates treatment effects via inverse probability weighting. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Given that some claims take years to resolve, and therefore the claim time distribution is right-skewed, the data are best fit using a logged model; however, in the next robustness check I estimate the reporting/claim time relationship where the dependent variable is measured in levels. The point estimate indicates a 54 day increase in claim resolution time on average, an estimate that is highly statistically significant and in line with the implied effect from the logged model. The final two robustness checks implement alternatives to the simple DD model. Specification 3 employs a matching-type estimator to estimate effects. I first use Poisson regression to model the expected number of days to claim resolution as a function of claimant age, year of claim, and the other covariates in the model and then, for each observation, I construct the expected claim duration based on these observable covariates. Then, in a second stage, I estimate a regression where log claim time is the outcome, the primary explanatory variable is the treatment interaction ( |$\textit{Age}_{i}>65 \times \textit{Post}_{i}$|⁠), and I include a full set of fixed effects for predicted claim time (in single days) as additional controls. This has the effect of measuring effects by matching each treated subject to the set of control subjects with the exact same expected claim complexity based upon the underlying observable characteristics of their claim. This specification yields a smaller estimated effect, but it remains statistically and economically significant. The final robustness check focuses attention on claims settled in 2012 and obtains an average treatment effect estimate by comparing treated and comparison subjects via inverse probability weighting. Here, I construct the probability weights by modeling the likelihood of treatment as a function of the various covariates included in the model ( |$X_{i}$|⁠). This matching approach yields an estimated impact of 0.109 which remains statistically significant. Overall, these robustness checks indicate the baseline finding regarding the magnitude and significance of the reporting effect is robust. How sensitive are the results the choice to focus on claimants ages 50 and over? For example, if we limited the analysis to ages just above and below the reporting cutoff, would we observe similar impacts? To explore this issue, the Appendix plots results from a series of regressions that systematically vary the age bandwidth around age 65, the most common age at which people gain Medicare.17 Outside of very narrow bandwidths, which, as discussed in the Appendix, are likely affected by measurement error, the estimated impacts are remarkably robust and consistent across age bands. As an additional robustness check, I examine whether any indication of delay can be found in another data set, the National Practitioner Data Bank (NPDB).18 The NPDB is a database of paid medical malpractice claims that has been widely used in research on malpractice.19 Relative to the auto claims data used thus far, the NPDB has the advantage of including an appreciably larger number of claims and covering more years. However, the NPDB is not without drawbacks for examining the effects of the new reporting requirement. The NPDB only offers coarse measures of claim resolution time, with the public use data only reporting the year of injury and year of resolution. Moreover, patient age at the time of alleged injury is only reported in 10-year categories (20–29, 30–39, etc.), meaning that treatment status for those ages 60–69 cannot be clearly assigned. Additionally, the NPDB contains much less information about the nature of injury and types of medical care received than the Insurance Research Council (IRC) auto closed claim data, limiting my ability to account for other claim characteristics that might affect treatment. To estimate the effects of the new reporting requirement on medical malpractice claim resolution time, I estimate claim-level differences-in-differences regressions on a sample of 115,622 claims resolved between January 2004 and March 2018 involving claimants aged 40 and up at the time of injury. Here, the outcome is the number of years between injury and claim payment, and, as in equation 1, the primary explanatory variable is an interaction for a claim involving an individual over age 65 where payment occurred in January 2012 or later. Lacking precise information on who was above the Medicare eligibility age, I code treatment at the intermediate value of 0.5 for claims involving 60–69 year-olds that were paid on or after January 1, 2012.20 I also control for state by year fixed effects and indicators for patient 10-year age group, patient gender, patient type (inpatient/outpatient), physician 10-year age group, physician specialty, alleged form of malpractice, injury severity, and type of insurer involved in making payment. The results of the NPDB analysis are reported in Table 5. In the baseline DD specification where the outcome is measured in levels, the new reporting requirement is associated with a statistically significant increase in claim resolution time of 0.07 years, or 26 days. This finding is robust to excluding individuals ages 60–69—for whom Medicare eligibility status cannot be readily assigned—from the analysis (Specification 2) and to log transforming the outcome to account for the right-skewed nature of the malpractice claim time distribution (Specification 3).21 While the magnitude of the estimated delay effect is much smaller in percentage terms for malpractice claims than auto claims, due to the greater length and complexity of the former, in absolute terms we observe a delay of about one month in malpractice cases, which is on par with the estimate obtained from the more detailed auto closed claim data.22 This robustness check using a different data source thus seems to reinforce the conclusion that the new Medicare reporting and subrogation requirements increased liability claim resolution time. Table 5. Estimated Effects for Malpractice Claims . DD . Implied delay . . . . estimate . in days . |$N$| . |$R^{2}$| . 1. Baseline estimate in levels 0.072** 26.3 115,370 0.198 (0.035) 2. Exclude individuals ages 60–69 0.097*** 35.4 89,927 0.198 (0.036) 3. Estimate in logs rather than levels 0.013** 21.9 115,370 0.230 (0.006) . DD . Implied delay . . . . estimate . in days . |$N$| . |$R^{2}$| . 1. Baseline estimate in levels 0.072** 26.3 115,370 0.198 (0.035) 2. Exclude individuals ages 60–69 0.097*** 35.4 89,927 0.198 (0.036) 3. Estimate in logs rather than levels 0.013** 21.9 115,370 0.230 (0.006) Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on malpractice claim resolution speed using data from the National Practitioner Data Bank. The unit of observation is a claim, and the outcome variable is the number of years between the initial injury and claim payment (mean |$= 4.44$|⁠). Differences are taken across age groups and across payment years. Because the NPDB only reports claimant ages in 10-year increments, claimants below age 60 are part of the control group; claimants ages 60–69 are coded as untreated prior to 2012 and 0.5 treated in 2012 and later, and claimants ages 70 and over are coded as untreated prior to 2012 and treated in 2012 and later. Additional controls include a full set of state by payment year interactions and indicators for claimant age and gender; physician age and specialty; injury severity; alleged type of malpractice; and type of insurer. Specification 1 is the baseline. Specification 2 omits individuals aged 60–69 form the analysis in light of the ambiguity regarding whether any particular individual is Medicare eligible. Specification 3 measures the dependent variable in logs rather than levels. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 5. Estimated Effects for Malpractice Claims . DD . Implied delay . . . . estimate . in days . |$N$| . |$R^{2}$| . 1. Baseline estimate in levels 0.072** 26.3 115,370 0.198 (0.035) 2. Exclude individuals ages 60–69 0.097*** 35.4 89,927 0.198 (0.036) 3. Estimate in logs rather than levels 0.013** 21.9 115,370 0.230 (0.006) . DD . Implied delay . . . . estimate . in days . |$N$| . |$R^{2}$| . 1. Baseline estimate in levels 0.072** 26.3 115,370 0.198 (0.035) 2. Exclude individuals ages 60–69 0.097*** 35.4 89,927 0.198 (0.036) 3. Estimate in logs rather than levels 0.013** 21.9 115,370 0.230 (0.006) Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on malpractice claim resolution speed using data from the National Practitioner Data Bank. The unit of observation is a claim, and the outcome variable is the number of years between the initial injury and claim payment (mean |$= 4.44$|⁠). Differences are taken across age groups and across payment years. Because the NPDB only reports claimant ages in 10-year increments, claimants below age 60 are part of the control group; claimants ages 60–69 are coded as untreated prior to 2012 and 0.5 treated in 2012 and later, and claimants ages 70 and over are coded as untreated prior to 2012 and treated in 2012 and later. Additional controls include a full set of state by payment year interactions and indicators for claimant age and gender; physician age and specialty; injury severity; alleged type of malpractice; and type of insurer. Specification 1 is the baseline. Specification 2 omits individuals aged 60–69 form the analysis in light of the ambiguity regarding whether any particular individual is Medicare eligible. Specification 3 measures the dependent variable in logs rather than levels. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab A key assumption needed for the baseline estimates above to properly capture the effects of reporting is that the younger claimants provide an appropriate counterfactual for what would have happened to older claimants absent the policy reform. Because the analysis essentially uses only two time periods, if there were differential time trends in expected injury costs by age, such trends might be confounded with the new policy. While the underlying DD assumption is not directly testable, I can perform a similar analysis but contrast years prior to the policy reform as a falsification exercise. If there are similar time trends by age and if younger claimants are sufficiently comparable to older claimants so as to render them a good control group, we should expect that when I conduct a similar DD analysis across years where no policy intervention occurred, I should measure no effect. The first row of Table 6 reports the results from such an analysis, where I use data from the 2002 edition of the IRC closed claim survey as a preperiod, and the 2007 data as the post period. The point estimate from this DD regression is small, nonsignificant, and fairly precisely measured. This falsification test suggests that the analysis is capturing the effects of the new reporting requirement and does not simply reflect noncomparabilities between older and younger auto injury claimants. Moreover, if older claimants are experiencing different time trends from younger claimants, we would expect to observe some measurable “effect” in these regressions, yet we do not. Table 6. Falsification Tests . DD Estimate . |$N$| . |$R^{2}$| . 1. Compare 2007 to 2002 -0.023 10,105 0.421 (0.051) 2. PIP and Medpay claims rather than bodily injury (BI) claims -0.021 8,104 0.316 (0.046) . DD Estimate . |$N$| . |$R^{2}$| . 1. Compare 2007 to 2002 -0.023 10,105 0.421 (0.051) 2. PIP and Medpay claims rather than bodily injury (BI) claims -0.021 8,104 0.316 (0.046) Notes: This table reports coefficient estimates from regressions designed to serve as placebo test of the baseline specification. In the first specification, the years of data are 2002 and 2007, and 2007 is used as the implementation year, so claimants ages 65 and older in 2007 are the treated group. In the second specification, the sample is comprised of PIP and Medpay claims from the 2012 and 2007 IRC database rather than BI claims. Except as noted below, the specification is as in column V of Table 2; see notes for Table 2. The second specification omits driver degree of fault as a control as that does not apply to PIP and Medpay claims. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 6. Falsification Tests . DD Estimate . |$N$| . |$R^{2}$| . 1. Compare 2007 to 2002 -0.023 10,105 0.421 (0.051) 2. PIP and Medpay claims rather than bodily injury (BI) claims -0.021 8,104 0.316 (0.046) . DD Estimate . |$N$| . |$R^{2}$| . 1. Compare 2007 to 2002 -0.023 10,105 0.421 (0.051) 2. PIP and Medpay claims rather than bodily injury (BI) claims -0.021 8,104 0.316 (0.046) Notes: This table reports coefficient estimates from regressions designed to serve as placebo test of the baseline specification. In the first specification, the years of data are 2002 and 2007, and 2007 is used as the implementation year, so claimants ages 65 and older in 2007 are the treated group. In the second specification, the sample is comprised of PIP and Medpay claims from the 2012 and 2007 IRC database rather than BI claims. Except as noted below, the specification is as in column V of Table 2; see notes for Table 2. The second specification omits driver degree of fault as a control as that does not apply to PIP and Medpay claims. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab As a second falsification exercise, I re-estimate the main specification on 2007 and 2012 data, but construct the sample using personal injury protection (PIP) and Medpay claims rather than bodily injury claims. PIP and Medpay are first-party coverages that pay for medical care regardless of the fault of the driver; these components of auto insurance are primary to health insurance, so providers will typical bill them before billing health insurers.23 Because PIP and MedPay are available from the moment of an accident and payments do not involve an adversarial settlement negotiation, the same forces that would generate delay in the bodily injury context under enhanced Medicare reporting and recovery would be expected to operate to a much lesser extent, if at all, for these coverages. Thus, if my preferred specification is appropriately capturing the effects of reporting and recovery, in the replication using PIP and Medpay claims, we should observe no effects. The second column of Table 6 reveals that this is indeed the case. The point estimate is close to zero and sufficiently precise so as the exclude impacts of the magnitude observed on the BI sample. The preferred specification thus finds measurable effects for contexts where reporting might plausibly affect case resolution times, while finding no relationship where none should be present. Are certain types of claims more extensively impacted by new reporting requirements? Figure 1 plots coefficients from a series of quantile DD regressions that assess the effects of the new reporting requirements on different points within the claim time distribution. Because these regressions include a range of covariates capturing the nature and severity of the claim, these estimates are best conceptualized as indicating the effects of reporting and recovery on claims that are shorter or longer than is typical given their underlying complexity, rather than as estimates of the effects of reporting on longer or shorter claims per se.24 While there are measurable increases in claim time across most of the distribution, claims in the lowest deciles, that are therefore shorter than expected, appear to be more affected than claims in the highest deciles. This would be consistent with an environment in which reporting engenders a fixed time cost of involving Medicare in negotiations—for simpler than expected claims, this fixed cost represents a higher fraction of the overall time to resolution than for longer than expected claims. This fixed cost interpretation also seems consistent with the results above for malpractice claims, which tend to take much longer to resolve on average than auto claims, but for which I obtained effects estimates of roughly similar magnitude in absolute terms. If involving Medicare in the settlement does entail a fixed cost, policies such as the federal exemption for low-value claims enacted in 2012 seem warranted, as such a fixed cost would be more burdensome for smaller claims. Figure 1 Open in new tabDownload slide Quantile Estimates of the Effect of Mandatory Reporting. Notes: This figure reports coefficient estimates from quantile DD regression estimates of the effect of mandatory reporting on log(claim time), with controls as in specification 4 of Table 2. The quantile regressions estimate impacts of the reporting requirement for claims that are longer or shorter than expected based upon the observable covariates; claims in the lower quantiles are simpler, shorter claims and those in the upper quantiles are longer, more complex claims. See notes for Table 2. Dotted lines represent 95% confidence bands. Figure 1 Open in new tabDownload slide Quantile Estimates of the Effect of Mandatory Reporting. Notes: This figure reports coefficient estimates from quantile DD regression estimates of the effect of mandatory reporting on log(claim time), with controls as in specification 4 of Table 2. The quantile regressions estimate impacts of the reporting requirement for claims that are longer or shorter than expected based upon the observable covariates; claims in the lower quantiles are simpler, shorter claims and those in the upper quantiles are longer, more complex claims. See notes for Table 2. Dotted lines represent 95% confidence bands. Another way to examine the fixed cost hypothesis is to ask whether most claimants appear equally affected by the new requirements, or if certain subgroups are more or less affected. To assess this, in Table 7, I report results from regressions that include interaction terms between the new requirement and group membership, focusing on groups for which there are plausible reasons to imagine that there could be differences in the settlement negotiation process. I first examine whether the impacts of reporting vary according to whether the claimed injury was a sprain or strain; because such injuries are harder to objectively verify, they are often considered a potential marker for claim buildup or fraud (Derrig et al. 1994), and are therefore settled differently than other claims (Crocker and Tennyson 2002; Loughran 2005). For both strain and nonstrain injuries we see evidence of a delay effect of similar magnitude. Table 7. Estimated Effects by Subgroup . DD . |$P$|-value H |$_{0}$|⁠: . . estimate . Groups equal . 1. Type of injury  Sprain/strain ( |$N=$| 6,156) 0.169*** 0.943 (0.060)  Other ( |$N =$| 3,199) 0.174** (0.069) 2. Claimed loss amount  Below median ( |$N =$| 4,209) 0.200*** 0.290 (0.067)  Above median ( |$N=$| 4,258) 0.132** (0.052) 3. Legal regime  No-fault ( |$N =$| 1,607) 0.164* 0.870 (0.087)  Tort ( |$N =$| 7,812) 0.178*** (0.057) 4. Claimant shared fault for crash  Yes ( |$N =$| 729) 0.316*** 0.165 (0.111)  No ( |$N =$| 8,689) 0.164*** (0.054) 5. Reported economic losses other than medical  Yes ( |$N =$| 2,006) 0.285*** 0.087 (0.085)  No ( |$N =$| 6,294) 0.143*** (0.051) . DD . |$P$|-value H |$_{0}$|⁠: . . estimate . Groups equal . 1. Type of injury  Sprain/strain ( |$N=$| 6,156) 0.169*** 0.943 (0.060)  Other ( |$N =$| 3,199) 0.174** (0.069) 2. Claimed loss amount  Below median ( |$N =$| 4,209) 0.200*** 0.290 (0.067)  Above median ( |$N=$| 4,258) 0.132** (0.052) 3. Legal regime  No-fault ( |$N =$| 1,607) 0.164* 0.870 (0.087)  Tort ( |$N =$| 7,812) 0.178*** (0.057) 4. Claimant shared fault for crash  Yes ( |$N =$| 729) 0.316*** 0.165 (0.111)  No ( |$N =$| 8,689) 0.164*** (0.054) 5. Reported economic losses other than medical  Yes ( |$N =$| 2,006) 0.285*** 0.087 (0.085)  No ( |$N =$| 6,294) 0.143*** (0.051) Notes: This table reports coefficient estimates from alternatives to the baseline specification that include interactions between required MSP reporting and membership in the listed group. Each numbered specification presents results from a unique regression. The specification is as in column V of Table 2; see notes for Table 2. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 7. Estimated Effects by Subgroup . DD . |$P$|-value H |$_{0}$|⁠: . . estimate . Groups equal . 1. Type of injury  Sprain/strain ( |$N=$| 6,156) 0.169*** 0.943 (0.060)  Other ( |$N =$| 3,199) 0.174** (0.069) 2. Claimed loss amount  Below median ( |$N =$| 4,209) 0.200*** 0.290 (0.067)  Above median ( |$N=$| 4,258) 0.132** (0.052) 3. Legal regime  No-fault ( |$N =$| 1,607) 0.164* 0.870 (0.087)  Tort ( |$N =$| 7,812) 0.178*** (0.057) 4. Claimant shared fault for crash  Yes ( |$N =$| 729) 0.316*** 0.165 (0.111)  No ( |$N =$| 8,689) 0.164*** (0.054) 5. Reported economic losses other than medical  Yes ( |$N =$| 2,006) 0.285*** 0.087 (0.085)  No ( |$N =$| 6,294) 0.143*** (0.051) . DD . |$P$|-value H |$_{0}$|⁠: . . estimate . Groups equal . 1. Type of injury  Sprain/strain ( |$N=$| 6,156) 0.169*** 0.943 (0.060)  Other ( |$N =$| 3,199) 0.174** (0.069) 2. Claimed loss amount  Below median ( |$N =$| 4,209) 0.200*** 0.290 (0.067)  Above median ( |$N=$| 4,258) 0.132** (0.052) 3. Legal regime  No-fault ( |$N =$| 1,607) 0.164* 0.870 (0.087)  Tort ( |$N =$| 7,812) 0.178*** (0.057) 4. Claimant shared fault for crash  Yes ( |$N =$| 729) 0.316*** 0.165 (0.111)  No ( |$N =$| 8,689) 0.164*** (0.054) 5. Reported economic losses other than medical  Yes ( |$N =$| 2,006) 0.285*** 0.087 (0.085)  No ( |$N =$| 6,294) 0.143*** (0.051) Notes: This table reports coefficient estimates from alternatives to the baseline specification that include interactions between required MSP reporting and membership in the listed group. Each numbered specification presents results from a unique regression. The specification is as in column V of Table 2; see notes for Table 2. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Specification 2 examines whether delay varies for smaller versus larger claims, where the size of claim is measured based upon the reported loss in dollars. Again, we observe similar delay effects for both categories of claims, and fail to reject the null of no group differences. Similarly, allowing the effects of reporting to vary across no-fault states—which see different types of bodily injury claims due to statutory restrictions on the types of claims that can be brought within the tort system (Anderson et al. 2010)—and tort states yields little evidence of heterogeneous effects. Specification 4 separately considers claims in which the claimant was estimated to have some level of fault for the accident from claims where the fault lay solely with the defendant. These former claims may be more difficult to resolve due to disagreements over culpability. However, for both types of claims there is evidence of a delay effect of mandatory Medicare reporting. Finally, in Specification 5, I consider whether reporting differentially affects cases involving claimed economic losses other than medical care (e.g. lost wages) versus purely medical losses. When medical care is the sole source of losses and therefore highly influences the amount of general damages available to the plaintiff, one might expect plaintiffs and their attorneys to be particularly protective against Medicare encroaching on the settlement. Although the point estimates suggest, contrarily, that there may be somewhat higher delays for combined medical/nonmedical claims, the data do not clearly reject equal impacts across the two categories of claims. Overall, the patterns in Table 6 indicate that the unintended delay effects of reporting are experienced across a wide range of different types of claimants and across claims of varying complexity, with little concrete evidence of significant heterogeneity. 5. Conclusions The analysis indicates that the introduction of a new reporting requirement for Medicare slowed the resolution of auto injury claims by 19%, or about 2 months. Importantly, this is the average effect across all individuals subject to the new reporting requirement, including claimants for whom there ultimately was no recovery for Medicare. While reporting and subrogation may benefit Medicare fiscally by allowing it to recoup otherwise unreimbursed outlays for medical care that by law are assignable to other parties, it comes at the cost of delaying the resolution of injury settlements for Medicare beneficiaries. It is instructive to compare the estimates here to those of Helland and Klick (2018). Helland and Klick (2018) employ a similar DD research design, but apply it to a somewhat idiosyncratic sample of claims from a single large insurer. Their dataset includes accidents that occurred at varying points in time, some prior the reform and some after, but all of which triggered queries by the insurer to Medicare to determine if the claimant was a Medicare beneficiary. The average claim duration for claims involving preperiod accidents in the Helland and Klick (2018) sample is over 800 days, nearly double that of my sample, and nearly half of their claims had not been resolved as of 2014. This likely reflects a different composition of their sample—during the pre-reform period, when reporting was not yet required, it seems plausible to imagine that the insurer would have queried Medicare mostly in cases where claim adjusters anticipated the possibility that resolving the claim might take sufficiently long that the new reporting requirement would come into effect, leaving comparatively complex claims in their sampling frame. Although coming to similar qualitative conclusions, the new results here thus generalize the finding of Helland and Klick (2018) to a broader and arguably more representative set of claims. In documenting a similar delay effect in data covering closed medical malpractice claims, I also demonstrate that the impacts of Medicare reporting appear to extend to a broader set of personal injury cases. While on the surface a 19% increase in case resolution time may seem relatively benign, simple calculations suggest that such delay could be costly. One simple way to assess the magnitude of the time costs of delay is to monetize the delays using traditional time value of money calculations. While to my knowledge there is no data source that measures the value of auto claims payments directed specifically to Medicare beneficiaries, applying the patterns from the closed claim data to data on aggregate auto liability payments from the Insurance Information Institute (2017) suggests that there were roughly |${\$}$|6 billion in auto liability payments in 2014 made to claimants ages 65 and older. While different stakeholders may have different views regarding the appropriate rate of time preference, at a lower bound we might apply a risk-free market interest rate of 3% per year to this sum, in which case the delays documented here would account for roughly |${\$}$|30 million in aggregate losses due to delay. If one wished to adopt a more plaintiff-centric view, an alternative logical time preference benchmark would be to apply rates derived from real market transactions in the structured settlement market, a market in which individuals with claims on future payment streams (including tort awards) can transform these into lump sum payments. Hindert and Ulman (2005) document an average implied annual interest rate in structured settlement transactions of 20%, which would equate to |${\$}$|200 million in annual delay costs. To put these amounts in perspective, in its CMS Statistics publication, CMS (2009, 2015) reports that it obtained a total of |${\$}$|465 million in secondary payer recoveries from auto and liability insurers in FY2007, and this amount had risen to |${\$}$|770 million by FY2015. Assuming generously that this entire increase was attributable to the new reporting provisions, Medicare gained about |${\$}$|300 million per year in additional revenues from reporting. The most conservative estimate above of |${\$}$|30 million puts the waiting costs at roughly 10% of Medicare’s total recovery. Although many stakeholders might view a 10% as an acceptable cost to enable Medicare to collect funds to which it is legally entitled, it is important to note that these costs are borne by injured parties, insurers, and their attorneys rather than Medicare. Moreover, under the upper bound estimates discussed above, waiting costs represent a much larger share of Medicare’s total recovery. Additionally, these cost estimates are conservative for several reasons. First, the 19% estimate used above is itself likely an understatement of the true impacts of reporting for the reasons discussed in Section II. Second, delays from reporting apply not only to those aged 65 and over, but all Medicare recipients. Individuals under age 65 constitute about 16% of the overall Medicare population, and there are logical reasons to imagine that reconciliation and negotiations during the subrogation process for this group might even be more difficult than for those aged 65 and older, since these beneficiaries by definition have chronic health conditions that may be difficult to disentangle from the conditions brought about by the tort in question. Third, the above calculations apply only to private passenger autorelated injuries, but reporting requirements generally apply to the universe of bodily injury claims within the tort system—for example, commercial auto payments or bodily injury payments under homeowners or commercial liability policies. Because the ICD coding systems commonly used for capturing medical diagnoses for billing purposes include codes that make it comparatively easy to identify care associated with an auto injury than would be possible for these other types of claims, if anything it seems likely that effects of new reporting requirements would be more pronounced for nonauto tort claims. Finally, the calculations above assume that the only costs of delay are the time costs of money, when in actuality there are administrative costs associated with querying the system, reconciling bills, and continuing to monitor open claims. Thus, while delay costs of |${\$}$|30– |${\$}$|200 million are not insubstantial, these numbers may be appreciably below the true systemwide cost of the new reporting and subrogation requirements. While understanding delay and its costs are important, delay is but one among a complex set of policy considerations relevant to decisions regarding the most appropriate rules for subrogation. In addition to improving Medicare’s fiscal health, the reporting rule also promoted greater parity between the government and private health insurers such as Medigap plans that may already pursue subrogation. Changing subrogation practices also can have tax implications, as the tax treatment of benefits payments can vary depending on whether these costs are borne by Medicare, injured parties, or private liability insurers. Because the analysis here considers only one effect of more aggressive subrogation, it does not definitely demonstrate whether the move to mandatory reporting was beneficial or harmful overall. It does, however, supply an important ingredient required to assess the utility of this policy change. To what extent are the findings of this paper likely to translate to Medicaid reporting? At a conceptual level, a reporting process for Medicaid beneficiaries is likely to contain similar elements to the MSP reporting process studied here—in particular, an initial query process to determine whether a claimant is a Medicaid beneficiary followed by, for those who are, a demand notice in which Medicaid indicates treatment it has reimbursed that it believes to be subject to subrogation, followed by a reconciliation process with the liability insurer and claimant. Under both systems claimants, liability insurers, and the government will face similar negotiation and information-sharing incentives. Thus, the forces that produced delay in the Medicare context seem likely to exist in any states that elect to develop such reporting requirements for Medicaid. However, there are also some important differences between Medicaid and Medicare that could escalate costs in the Medicaid context. First, because Medicaid is operated at the state level, each individual state will have to develop a reporting system for its own claimants; states will likely differ in the resources and expertise available in developing such systems, and it seems plausible to think that some states may develop systems inferior to the current federal system. Multistate liability insurers would also potentially need to develop technology to interact with multiple different reporting systems in different states, which could prove costly. A second reason why delay cost might be higher in the Medicaid context relates to the nature of the Medicaid program itself. Medicaid is targeted towards low-income individuals and families, and Medicaid recipients have fewer assets than the typical claimant. The costs of delayed settlements may be particularly acute for low-income individuals with few resources; indeed, empirical evidence suggests that poor individuals have substantially higher rates of time preference (Lawrence 1991; Paserman 2008). Moreover, policies which delay payments to poor claimants who have recently suffered adverse health shocks may be viewed as particularly problematic by policymakers and the public. It remains to be seen whether states will follow the federal lead and more aggressively pursue reporting and subrogation for their Medicaid programs. This article identifies a drawback of such subrogation efforts—a delay in settling cases induced by the additional reporting and reconciliation requirements. The fact that such delays occur does not mean that increased subrogation is unwarranted—the fiscal pressures facing Medicaid may very well offer sufficient justification to policymakers and the public for new efforts to appropriately pursue reporting and subrogation. However, such options should be pursued only with a clear understanding that reporting carries with it an unintended consequence of delay. For those jurisdictions that do more aggressively pursue recovery from personal injury awards, authorities should seek to develop reporting systems that minimize the adverse effects of delay, by, for example, applying a safe harbor for low-value claims as currently exists in the federal system. More broadly, given the clear costs associated with reporting identified in this article, researchers would do well to develop additional empirical evidence on the potential recovery for Medicaid through additional subrogation, so as to better inform policymakers regarding how the fiscal benefits of additional recovery efforts might compare to the costs. Appendix: Analysis of Choice of Age Bandwidth The baseline specification limits the analysis to those ages 50 and over. In this Appendix, I explore the sensitivity of the results to using different samples by age. In determining the sample, we face a tradeoff—as we narrow the age window, the youngest and oldest individuals grow closer in age, so it becomes more plausible to think that the claim/injury experience of the youngest claimants provides a reasonable counterfactual for the claim/injury experience of the older claimants absent a reporting and subrogation requirement. However, as we narrow the age window, we also reduce the sample size—which may decrease the precision of the estimates—and potentially increase the influence of measurement error on our estimates. The measurement error problem arises because we do not actually observe the Medicare status of any particular claimant, we only observe their age, which we use as a proxy for Medicare status. This data limitation introduces several forms of measurement error. First, as discussed in Section 3 of the article, although the majority of individuals who gain Medicare do so upon turning 65, certain classes of individuals, such as disabled, low-income individuals, can be eligible for Medicare at earlier ages. Card et al. (2009) suggest that roughly 10% of adults ages 55–64 are Medicare eligible; this means that there are some younger individuals in the sample labeled as not being subject to reporting and subrogation who in fact are. Second, rather than knowing a claimant’s exact age, we only know their year of birth. This will lead us to label some individuals as treated in 2012 who actually would not have been subject to Medicare reporting and subrogation. As an example, consider a claimant from 2012 born in 1947 who therefore was 65 in 2012 when their claim was closed. For the purposes of the analysis, the individual will be coded as subject to the Medicare reporting and subrogation requirement. However, if their birthdate was actually in December 1947, their claim was closed in February 2012, and they follow the usual pattern of gaining Medicare eligibility at the time they turn 65, their claim would in actuality have been resolved when they were still age 64, and therefore not subject to Medicare reporting or subrogation. If birthdays and claim resolution dates are uniformly distributed across the calendar, we had expect 50% of the individuals who turned 65 in 2012 to in fact not have been subject to the reporting and subrogation requirement. This form of measurement error only affects individuals who turn 65 in the year their claim is closed. Finally, there are lags between the time that treatment is received and the time that claims are settled that may introduce measurement error. Consider, for example, an individual who is involved in a crash in early 2009 when they are 63 years old who receives medical care for their injuries in 2009 and 2010, but negotiations over the settlement amount, lost wages, etc. delay the resolution of the claim until 2012. This person will be 66 in 2012, and thus treated as subject to Medicare reporting and subrogation for the purposes of this analysis. However, because their medical treatment all occurred when they were 63 and 64—before they became eligible for Medicare—they in fact may not be subject to meaningful Medicare subrogation in actual practice. This form of measurement error seems likely to be most prevalent for individuals close to age 65 in 2012—because most claims are resolved within a few years of the initial accident, as individuals get older, it seems less plausible to imagine that their period of medical care for the auto injury would be completely nonoverlapping with the period during which they were Medicare beneficiaries. To summarize, there are reasons to expect that there may be measurement error with regard to actual reporting and subrogation status at and just above age 65, but as the age window expands, the fraction of the treated sample that was not in fact treated diminishes. Appendix Figure A.1 plots coefficients from a series of DD regressions that systematically vary the age bandwidth around 65. These estimates are analogous to those presented in column V of Table 2, but considering samples. Each point on the chart depicts a separate estimate, and the dotted lines depict 95% confidence intervals for each estimate based on clustered standard errors. The point estimate depicted where the horizontal axis equals 5, for example, corresponds to the DD estimate obtained when the age window includes 5 years before normal Medicare eligibility and 5 year after, or ages 60–69. Figure A.1 Open in new tabDownload slide DD Estimates Varying the Age Bandwidth. Figure A.1 Open in new tabDownload slide DD Estimates Varying the Age Bandwidth. Several patterns are apparent from the figure. First, the point estimates are fairly stable, similar to the baseline reported in Table 2, and economically meaningful for bandwidths above 2 years. Second, the estimates are attenuated and much less precise when the bandwidth is only 1 or 2 years, consistent with the discussion regarding measurement error above. Overall these results suggest that the paper’s main conclusions are not particularly sensitive to the choice of age bandwidth. Acknowledgement This work was supported by the RAND Institute for Civil Justice (ICJ) core funding program. Special thanks to Eric Helland, Allison Hoffman, and Jon Klick who provided comments on the research and David Corum who assisted with Insurance Research Council (IRC) data access. The editor J.J. Prescott and two anonymous referees provided very helpful comments and suggestions. All content is the responsibility of the author and may not reflect the views of RAND or the ICJ. Footnotes 1. See, for example, American Association for Justice (2011), Jordan (2017), and Medicare Advocacy Recovery Coalition (2017). 2. While it may seem at first glance as though insurers should benefit from holding money longer before transferring it to injured parties, such benefits are illusory in practice. Insurers are legally required to hold risk-free reserves against future claim payments, and are therefore constrained from deploying any capital tied up in the claims process towards more productive investments. A long claims process thus impedes them from maximizing returns on their capital. 3. As a benchmark, in my data the average claim resolution time is 422 days, and there is no censoring. 4. See, for example, § 116 of Public Law 97-248, the Tax Equity and Fiscal Responsibility Act of 1982 (Medicare) and § 1902(a)(25) of Public Law 74–271, The Social Security Act (Medicaid). 5. Such recoveries are often referred to colloquially as “Medicare liens,” although they are not liens in the traditional sense. 6. 42 U.S.C. § 1395y. 7. See, for example, Government Accountability Office (2012) and Kirchoff (2014). 8. For example, in response to empirical evidence demonstrating that there were likely to be some claims for which the costs of reporting exceeded Medicare’s recovery (Helland and Kipperman 2012), Congress and CMS ultimately established a de minimis threshold below which reporting was not required. 9. See Section 53102 of the Bipartisan Budget Act of 2018. 10. https://ri-mais.com/index.html. (accessed November 29, 2020) 11. See http://ncoil.org/wp-content/uploads/2016/04/MedicaidInterceptionModelamended.pdf. (accessed November 29, 2020) 12. In this article’s discussion I refer to claim resolution generically as “settlement,” as the vast majority of claims settle, but the data and my analysis also include claims that go to trial. 13. The tables report robust standard errors. Conceptually it seems most logical to think of each claim as representing a fresh realization of the outcome, particularly given that factors such as the nature of the accident and injury have by far the largest influence on claim complexity. Clustering on age/year—the level of treatment—leads to smaller standard errors, and other forms of clustering (e.g. by state) yield standard errors of similar magnitude to the robust standard errors. 14. This estimate arises directly from estimating equation 1, but one might also want to adjust the estimate to account for the fact that we do not directly observe Medicare eligibility in the data, and age is an imperfect proxy for eligibility. In particular, because roughly 10% of older adults below age 65 are Medicare eligible, and 6% of those over age 65 are not eligible, the actual increase in the fraction of claimants subject to the reconciliation process is below 100%. Adjusting for these patterns slightly increases the point estimate to 23%. 15. Table 1 reveals that many claims take a long time to resolve; if a claim is going to take several years to resolve in any case, an increase in resolution time of a couple of months on the margin may not have a big effect on incentives to file. 16. In particular, I add indicators for the deciles of this variable interacted with log payment amount to allow for a non-constant elasticity of claim time with respect to claim amount. 17. I express appreciation to an anonymous referee for suggesting this analysis. 18. Mullan et al. (1992) provide background on the origin and characteristics of the NPDB. Mello and Studdert (2016) discuss some of the NPDB’s limitations. 19. Examples of recent research using the NPDB include Studdert et al. (2016), Schaffer et al. (2017), and Gupta et al. (2018). 20. In addition to the measurement error arising from the lack of exact age in the data, there is also measurement error arising from the fact that some unknown group of claimants likely gained Medicare coverage while the claim was pending, both because the claim latency for malpractice claims is long enough that many claimants could age in while the claim was being resolved, and because adults under age 65 can gain Medicare eligibility if they become permanently disabled. 21. In an unreported analysis, I also estimated an event-study version of the DD regressions for the NPDB data, and found no evidence of a meaningful pre-implementation trend, and a pattern suggestive of an increase in case resolution time coinciding with the new reporting requirement. 22. There are several reasons to expect Medicare to have greater involvement in the settlement process for medmal claims than auto claims, even absent a reporting requirement. Because the source of injury in a medmal case is medical care, Medicare may be able to infer the existence of a malpractice claim involving one of its beneficiaries (and therefore seek recovery) based upon information available from its medical records. Moreover, when there will be ongoing medical payments after the claim is settled—a not uncommon occurrence in medmal cases—it has been longstanding practice for Medicare to be involved in negotiations to ensure its interests are satisfied through so-called Medicare Set-Asides. This means that the “dosage” of treatment associated with the new reporting requirement was likely lower for medical malpractice claims than for auto claims, which could also explain the diminished impact. 23. One reason for this is that PIP and Medpay policies generally do not receive the large discounts from medical provider list prices that are available to large health insurers, so providers get higher reimbursements from auto insurers than they would from traditional health insurers. 24. See Powell (2019) for a more detailed discussion of the interpretation of quantile regression coefficients in the presence of covariates. References American Association for Justice. 2011 . Medicare Secondary Payer: How Streamlining a Broken Bureaucracy Will Protect Seniors and Taxpayers . Washington DC . Andrews, Michele . “ Long Waits For Consumers When Medicare Is ‘Secondary Payer’, ” Kaiser Health News , Available at https://khn.org/news/012913-michelle-andrews-medicare-secondary-payer/ (accessed November 29, 2020). Anderson, James M. , Heaton Paul, and Carroll Stephen J.. 2010 . The US Experience with No-Fault Automobile Insurance: A Retrospective. Santa Monica, CA : RAND Corporation . Barnett, Jessica C. , and Vornovitsky Marina S.. 2016 . Health Insurance Coverage in the United States: 2015. US Census Bureau, Current Population Reports , Report P60-257 . Card, David , Dobkin Carlos, and Maestas Nicole. 2009 . “ Does Medicare Save Lives?, ” 124 Quarterly Journal of Economics 597 - 636 . Google Scholar OpenURL Placeholder Text WorldCat Centers for Medicare and Medicaid Services. 2009 . 2009 CMS Statistics . Available at https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/CMS-Statistics-Reference-Booklet/Downloads/CMS_Stats_2009.pdf (accessed November 29, 2020). Centers for Medicare and Medicaid Services. 2015 . 2015 CMS Statistics . Available at https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/CMS-Statistics-Reference-Booklet/Downloads/2015CMSStatistics.pdf (accessed November 29, 2020). Congressional Budget Office. 2016 . The 2016 Long-Term Budget Outlook . Available at https://www.cbo.gov/publication/51580 (accessed November 29, 2020). Crocker, Keith J. , and Tennyson Sharon. 2002 . “ Insurance Fraud and Optimal Claims Settlement Strategies, ” 45 Journal of Law and Economics 469 – 508 . Google Scholar OpenURL Placeholder Text WorldCat Derrig, Richard A. , Weisberg Herbert I., and Chen Xiu. 1994 . “ Behavioral Factors and Lotteries Under No-Fault with a Monetary Threshold: A Study of Massachusetts Automobile Claims, ” 61 Journal of Risk and Insurance 245 - 275 . Google Scholar Crossref Search ADS WorldCat Government Accountability Office. 2012 . Medicare Secondary Payer: Additional Steps are Needed to Improve Program Effectiveness for Non-Group Health Plans . GAO-12-333 , Washington DC . Gupta, Ashwin , Snyder Ashley, Kachalia Allen, Flanders Scott, Saint Sanjay, and Chopra Vineet. 2018 . “ Malpractice Claims Related to Diagnostic Errors in the Hospital, ” 27 BMJ Quality & Safety 53 – 60 . Google Scholar Crossref Search ADS WorldCat Hagy, Tom . 2010 . Waters & Kraus Attorney on Negotiating Medicare Liens. Available at: https://litigationconferences.com/waters-kraus-attorney-on-negotiating-medicare-liens/ (accessed November 29, 2020). Helland, Eric and Kipperman Fred. 2012 . “ Recovery Under the Medicare Secondary Payer Act Impact of Reporting Thresholds, ” RAND Occasional Paper OP-332-ICJ , Santa Monica, CA . Google Scholar OpenURL Placeholder Text WorldCat Helland, Eric and Klick Jonathan. 2018 . “ Medicare Secondary Payer and Settlement Delay, ” 15 Journal of Empirical Legal Studies 356 – 377 . Google Scholar Crossref Search ADS WorldCat Hindert, Daniel W. , and Ulman Craig H.. 2005 . “ Transfers of Structured Settlement Payment Rights: What Judges Should Know About Structured Settlement Protection Acts, ” 44 Judges Journal 19-31. Google Scholar OpenURL Placeholder Text WorldCat Insurance Information Institute. 2017 . “ Incurred Losses For Auto Insurance, 2011-2015, ” Available at: http://www.iii.org/fact-statistic/auto-insurance (accessed April 27, 2017). Jordan, Jennifer . 2017 . “ Medicare Secondary Payer Concerns in 2017: A Perfect Storm Could Be Brewing, ” LexisNexis Legal Newsroom , Available at: https://www.lexisnexis.com/legalnewsroom/workers-compensation/b/recent-cases-news-trends-developments/archive/2017/05/31/medicare-secondary-payer-concerns-in-2017-a-perfect-storm-could-be-brewing.aspx (accessed November 29, 2020). Kirchoff, Suzanne . 2014 . Medicare Secondary Payer: Coordination of Benefits . Congressional Research Service Report #7-5700 , Washington DC . Lawrence, Emily . 1991 . “ Poverty and the Rate of Time Preference: Evidence from Panel Data, ” 99 Journal of Political Economy 54 - 77 . Google Scholar Crossref Search ADS WorldCat Loughran, David . 2005 . “ Deterring Fraud: The Role of General Damage Awards in Automobile Insurance Settlements, ” 72 Journal of Risk and Insurance 551 - 575 . Google Scholar Crossref Search ADS WorldCat Matzus, Jason . 2011 . Testimony on “Protecting Medicare with Improvements to the Medicare Secondary Payer Regime” before the Subcommittee on Oversight and Investigations, Committee on Energy and Commerce United States House of Representatives. Available at https://www.govinfo.gov/content/pkg/CHRG-112hhrg74294/pdf/CHRG-112hhrg74294.pdf (accessed November 29, 2020). Medicare Advocacy Recovery Coalition. 2017 . “ SMART Act Implementation, ” Available at: http://www.marccoalition.com/other-smart-act-implementation.html (accessed November 29, 2020). Mello, Michelle M. , and Studdert David M.. 2016 . “ Building a National Surveillance System for Malpractice Claims, ” 51 Health Services Research 2642 – 2648 . Google Scholar Crossref Search ADS PubMed WorldCat Mullan, Fitzhugh , Politzer R. M., Lewis C. T., Bastacky S., Rodak J. Jr, and Harmon R. G.. 1992 . “ The National Practitioner Data Bank, ” 268 JAMA 73 – 79 . Google Scholar Crossref Search ADS PubMed WorldCat Paserman, M. Daniele . 2008 . “ Job Search and Hyperbolic Discounting: Structural Estimation and Policy Evaluation, ” 118 Economic Journal 1418 – 1452 . Google Scholar Crossref Search ADS WorldCat Powell, David . 2019 . “ Quantile Treatment Effects in the Presence of Covariates, ” Review of Economics and Statistics . 1 – 12 . Available at: https://www.mitpressjournals.org/doi/abs/10.1162/rest_a_00858. Schaffer, Adam C. , Jena Anupam B., Seabury Seth A., Singh Harnam, Chalasani Venkat, and Kachalia Allen. 2017 . “ Rates and Characteristics of Paid Malpractice Claims Among US physicians by Specialty, 1992–2014, ” 177 JAMA Internal Medicine 710 – 718 . Google Scholar Crossref Search ADS PubMed WorldCat Studdert, David M. , Bismark Marie M., Mello Michelle M., Singh Harnam, and Spittal Matthew J.. 2016 . “ Prevalence and Characteristics of Physicians Prone to Malpractice Claims, ” 374 New England Journal of Medicine 354 – 362 . Google Scholar Crossref Search ADS WorldCat © The Author 2021. Published by Oxford University Press on behalf of the American Law and Economics Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png American Law and Economics Review Oxford University Press

The Effect of Mandatory Insurer Reporting on Settlement Delay

American Law and Economics Review , Volume 22 (2) – Dec 1, 2020

Loading next page...
 
/lp/oxford-university-press/the-effect-of-mandatory-insurer-reporting-on-settlement-delay-ahNoJiejxn
Publisher
Oxford University Press
Copyright
© The Author 2021. Published by Oxford University Press on behalf of the American Law and Economics Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
1465-7252
eISSN
1465-7260
DOI
10.1093/aler/ahaa010
Publisher site
See Article on Publisher Site

Abstract

Abstract To improve their fiscal position, Medicare and some state Medicaid programs have recently taken steps to mandate reporting of personal injury awards and thus facilitate subrogation against such awards. Participants in the tort system have argued these additional reporting requirements might delay settlement of claims, harming both plaintiffs and defendants. This article examines this problem empirically, using a rich, national data set of closed automobile bodily injury claims. Using a differences-in-differences research design that exploits the introduction of a new Medicare reporting requirement in 2011, it demonstrates that mandated reporting increased time to settlement by 19%, or an average of 58 days. Robustness checks using data from closed malpractice claims reveal a similar delay. Conservative calculations suggest such delays could generate hundreds of millions of dollars in waiting costs each year. Policymakers should be aware of and seek to avoid such costs as they assess whether and how to expand reporting of personal injury awards. 1. Introduction Policymakers have expressed alarm regarding large and growing expenditures on health care by federal and state government. In 2000, federal Medicare outlays accounted for 1.8% of US gross domestic product (GDP), while Medicaid and related state programs consumed 1.3% of GDP (Congressional Budget Office (CBO) 2016). By 2015, these shares had risen to 3.1% for Medicare and 2.2% for Medicaid, a combined increase of 70%. CBO projections suggest that if current policies continue, Medicare and Medicaid expenditures will collectively account for over 8% of GDP by 2040. Recognizing that such cost growth is unsustainable, Democrats and Republicans alike have introduced proposals designed to address the looming fiscal challenges facing both programs. Included among these are various efforts that attempt to offset these programs’ costs from other sources, and in recent years the tort system has become a favored target. When bodily injuries occur due to negligent actions by others, injured parties can seek compensation for their losses from responsible parties through the tort system. This system is overlaid on top of the traditional health insurance system that provides coverage for medical care largely without regard for the underlying cause or fault associated with a particular medical condition. Interactions between the two systems are governed by statutes, regulations, case law, and norms that determine who pays first, what levels of compensation are provided, and what subrogation rights exist. Although government insurers have long been designated as payers of last resort, traditionally they have nonetheless borne the cost of considerable portion of medical care ultimately compensated through the tort system, due to practical barriers in pursuing recovery from tortfeasors or their payees. However, recent legislative and regulatory changes have sought to strengthen the ability of both Medicare and Medicaid to obtain information about tort awards and obtain recoveries from these awards. A key component of reform involves efforts to mandate reporting of bodily injury awards to government insurers. Such reporting would allow government insurers to identify care episodes subject to recovery and issue demands for payment. While such reporting systems clearly promote the fiscal interests of Medicare and Medicaid, they also carry potential for generating other, conceivably more far-reaching impacts on the civil justice system. By requiring additional data collection for all cases, and, for at least some cases, injecting an additional party into settlements, reporting can complicate the negotiation process, delaying the final resolution of cases and increasing administrative cost. Moreover, plaintiff advocates have highlighted the possibility that enhanced recovery by Medicare or Medicaid might render some previously viable claims noneconomical for plaintiffs to pursue, altering the pool of claims in the system, and reducing access to justice. While such impacts are plausible in theory and have been noted by many commentators on this issue,1 to date there has been virtually no empirical evidence demonstrating whether or to what degree such effects occur in actual practice. In this article, I provide one of the first large-scale analyses of the effects of a new reporting requirement on time to claim resolution, one key outcome thought to be affected by such reforms. Time to claim resolution is of particular interest because delay in resolving claims reduces welfare for everyone in the system. Recovery and subrogation primarily involve a transfer from claimants, insurers, and policyholders, and their representatives to Medicare or Medicaid; the appropriate degree of such transfers is largely a normative question about which different stakeholders may disagree. However, other factors being equal, all parties in the system would prefer less delay in resolving cases, as delay increases administrative costs for all parties, increases reserving costs for insurers,2 and reduces the value of settlements for claimants and Medicare and Medicaid due to time preference. Thus, to the extent that introducing a new reporting and recovery process affects delay, these costs must be weighed against the benefits of allocating losses more closely in line with the intended priority rules making Medicare and Medicaid payers of last resort. To assess the impact of new mandatory reporting and recovery requirements, I exploit a Medicare reform enacted at the end of 2007 and implemented in 2011 that for the first time obligated insurers to report personal injury awards involving Medicare beneficiaries to the federal government. Focusing on auto injury claims—a large and economically important subset of the overall tort personal injury space—I measure the impacts of the reform by leveraging a rich, national data set covering thousands of individual personal injury awards. Using a differences-in-differences (DD) research design that contrasts claim resolution times for claimants above and below 65 years of age before and after the reform was implemented, and which also controls for a rich set of claimant and injury characteristics likely to affect claim latency, I demonstrate that the new reporting requirement is associated with a 19% increase in time to claim resolution, or a roughly 2-month delay for the average claim. This estimate is likely conservative and survives across a number of robustness checks. Simple and conservative back-of-the-envelope calculations suggest that mandatory reporting can generate tens or even hundreds of millions of dollars of delay costs each year. While the fiscal and other benefits may in the end justify increased efforts at subrogation, policymakers should be aware of such delay costs as they consider whether to further expand recovery efforts to Medicaid and other government insurance programs. This works generalize contemporaneous work by Helland and Klick (2018) that uses data from a single large auto insurer to analyze the impacts of Medicare reporting. As here, Helland and Klick (2018) find that Medicare reporting increases time to claim resolution, although they identify larger impacts, on the order of 6 months or longer. However, the average claim duration pre-reform in their sample is over 800 days, and nearly half of their postreform claims were censored as of February 2014, suggesting that their sample may have been composed of a comparatively complex set of claims.3 This article provides additional evidence using a more broadly representative sample of auto claims and also demonstrates that the delay effects of the new reporting requirement persist even after controlling for a rich set of underlying claim characteristics. I also extend the analysis to a different personal injury setting—closed malpractice claims—and measure a similar delay in settlement due to Medicare reporting. The article proceeds as follows. Section 2 outlines the difficulties that have historically limited Medicare and Medicaid’s ability to pursue recovery form tort awards and briefly discusses some of the key reforms introduced by Congress and regulatory agencies in an effort to address these problems. Recent developments described in this section suggest that mandatory reporting to facilitate recovery against tort awards may become more widespread in the future, highlighting the need for empirical research on the effects of such requirements. Section 3 describes the data and DD research design and its underlying assumptions. Section 4 presents results from the main analysis relating mandatory reporting to case resolution time along with a series of robustness checks and two falsification tests designed to examine the validity of one of the key assumptions of the research design. I also confirm the basic findings that reporting increases delay in a separate dataset of closed medical malpractice claims. Section 5 discusses the implications of these findings for potential future efforts to mandate reporting. 2. Background and Prior Research By statute, Medicare and Medicaid have long been designated as payers of last resort is many situations,4 bearing responsibility for reimbursing providers for medical care only once other sources of payment beyond the patient have been exhausted. However, a number of practical obstacles have historically limited the ability of these programs to collect payments made for medical care incident to a personal injury. A key difficulty arises due to the timing of tort awards relative to the billing cycle. Following an injury, an injured party will seek medical care, and at the time the care is received, it is not yet clear whether a tort recovery will be available, what care it might cover, and how much will be available from any award to compensate for medical care. Providers, accustomed to fact that Medicaid and Medicare are the presumptive payers in almost all other episodes of medical care and anxious to receive payments for their services, will often bill the health insurers using their usual processes and receive payment. Months or years after payment has been received, when a tort settlement is reached, Medicare and Medicaid in theory could seek recovery5 for their prior payments from either the Medicare beneficiary (injured party) or the tortfeasor, which would allow them to shift the cost of the care from the Medicare trust fund and taxpayers to those directly involved in the injury-causing incident. However, as a practical matter, seeking such recovery was difficult historically for two reasons. First, Medicare and Medicaid had no easy way to ascertain whether a tort award was made in a particular case. Because there is substantial variability from patient to patient in the amount of time it takes to reach a settlement following an injury, there was no set time frame within which Medicare and Medicaid might expect to observe a tort recovery had occurred, and there is little in patient or billing records that might indicate whether a particular care episode is likely to be covered by tort. Moreover, even for patients for whom there is a high likelihood of a tort award (e.g. a patient presenting with an auto crash injury), some fraction will ultimately end up not receiving any award. Thus, Medicare and Medicaid had little ability to identify those cases where a tort award was available from which it might seek recovery. A second obstacle relates to determining which parts of the settlement should be available to Medicare and Medicaid as compensation. While it is not uncommon for injured parties in personal injury cases to itemize various losses (e.g. lost wages, medical care, pain, and suffering) as they present demands for compensation to tortfeasors and their insurers, when a settlement is reached, it is often done so without apportioning its final value into components of economic and noneconomic loss. To take a simple example, in an injury involving a Medicare beneficiary with claimed losses of |${\$}$|50,000 for medical care and |${\$}$|50,000 for pain and suffering, where the ultimate settlement after attorney’s fees is |${\$}$|60,000, absent other information it would appear uncertain whether |${\$}$|50,000, |${\$}$|30,000, or some other amount would be available to Medicare to help it recover its outlays associated with the injury. Because of this ambiguity, Medicare and Medicaid were poorly positioned to know how much recovery might be available to them in a given situation, and therefore which claims might deserve priority and which might be inefficient to pursue. In the Medicaid context, two U.S. Supreme Court decisions related to this second problem served to further dampen incentives to pursue recovery. In Arkansas Dept. of Health and Human Servs. v. Ahlborn 547 U.S. 268 (2006), the Court ruled that Medicaid’s recovery rights were limited to only the portion of the settlement that represents medical care, which in the particular case presented before the Court reduced Medicaid’s entitlement to about 1/6 of the amount asserted by the state. Later, in Wos v E.M.A. 568 US ___ (2013), the Court invalidated a state statute that set the amount of Medicaid recovery at one-third of the tort award, reasoning that such a rule would run afoul of the federal Medicaid law’s prohibition against states filing liens against personal property of Medicaid recipients to recover their costs for the program. Together, the rulings signaled some hostility in the courts to efforts to expand Medicaid’s ability to recovery against personal injury awards. 2.1. The Medicare Reporting System In the early 1980’s, Congress passed legislation designed to compel plaintiffs and defendants to consider Medicare’s interests during settlement negotiations by granting Medicare a right of recovery against either party in the event of a resolution that did not adequately address its statutory position as a secondary payer, with scope for additional damages in the event of noncompliance.6 This legislation and the resultant requirement are commonly referred to as Medicare Secondary Payer (MSP). However, Medicare was poorly positioned to enforce its rights due to its inability to identify cases where personal injury awards had been issued. To address that problem, Congress enacted Public Law 110–173, The Medicare, Medicaid, and SCHIP Extension Act of 2007 (MMSEA), which established a reporting system designed to enabled Medicare to track beneficiaries who had received personal injury award payments. Under the new law, liability insurers were required to report information about bodily injury award payments made to any individual who was a Medicare beneficiary, with substantial penalties for omissions or noncompliance. As a practical matter, this meant that insurers needed to query Medicare to determine beneficiary status of anyone who filed a tort claim, and then, for those who were beneficiaries, ensure in consultation with the claimant’s attorney that Medicare’s interests could be satisfied prior to the finalization of a settlement, so as to foreclose the possibility that, after receiving the mandated report, Medicare might seek additional unforeseen payments, and penalties from either party. The new law thus both initiated a reporting process and triggered a heightened effort to involve Medicare in the settlement of liability claims. Implementation of the new law was not without difficulty. After its passage, insurers and plaintiff attorneys reported difficulties in getting information from the Centers for Medicare and Medicaid Services (CMS) needed to resolve claims, and there were substantial problems with the IT and other systems CMS established to handle queries and reporting.7 As a result of these difficulties, the original implementation date for mandatory reporting by liability insurers was moved back from July 2009 to January 2011 for no-fault insurers and January 2012 for liability insurers, and Congress enacted additional legislation in 2012 to improve the process.8 CMS reported recovering |${\$}$|6 billion through MSP in 2006, which rose to |${\$}$|8 billion by 2012 (Kirchoff 2014), an increase roughly in line with Medicare’s overall growth over the same period. Thus, the extent to which the new reporting requirements helped increase Medicare’s recovery from third-party payers remains somewhat ambiguous. It is also not well understood how these provisions affected settlement times or other legal outcomes of interest, the issue explored in this article. 2.2. Is Medicaid the Next Wave? One reason the Medicare reforms hold relevance for future policy is that the success of the reporting requirements has prompted policymakers to consider expanding reporting to other government insurance programs, most notably Medicaid. In response to the Wos v. E.M.A. decision, Congress included a provision in § 202 of Public Law 113–67, the Bipartisan Budget Act of 2013, that grants Medicaid recovery rights for its outlays up to the full value of the settlement in a personal injury case, removing the Supreme Court’s restriction that Medicaid access only that portion of the settlement designated for medical care. Thus, in situations where most or all of the settlement is allocated for general damages, wage loss, or other nonmedical loss, under the new law Medicaid would enjoy much broader rights of recovery. This provision, which initially went into effect in October 2017 but was then repealed when the Bipartisan Budget Act of 2018 was passed in February 2018,9 would have substantially increased the dollars available to state Medicaid programs through third-party recovery efforts. Whether future Congresses will choose to restore Medicaid’s enhanced recovery rights remains an open question. Because Medicaid is largely administered by states rather than the federal government, a blanket reporting requirement such as that that was introduced for Medicare is infeasible. Nevertheless, some individual states have begun pursuing reporting systems and requirements patterned after the Medicare system. In 2012, for example, Rhode Island mandated liability insurer reporting through a newly developed Medical Assistance Intercept System (MAIS),10 allowing it to subrogate against injury awards made to state Medicaid beneficiaries. As of 2016, the state reported collections of |${\$}$|3.2 million due to the new reporting program. In 2015, the National Conference of Insurance Legislators (NCOIL) also adopted and began promulgating a model state law establishing a mandatory Medicaid reporting,11 and legislation patterned after the model law has been proposed in some states. 2.3. Should We Expect Reporting to Affect Speed of Claim Resolution? The introduction of new reporting requirements, as occurred for Medicare and as seems to be beginning for Medicaid, introduces several new elements to claim process that might affect the time needed to resolve a claim. First, insurers must query the government insurer to determine whether a claimant in their system is a beneficiary and therefore potentially subject to subrogation. While at first glance it may seem that the use of an electronic querying system should allow such queries to occur nearly instantaneously—and therefore not contribute to delay—all querying systems require the ability to uniquely identify and match individuals, which, in the U.S. context, means collecting unique identifying information such as a Social Security Number (SSN) which traditionally would not be required to settle a claim. In other words, the query process itself requires compiling additional information from the claimant which may take some time to collect. For those claimants who are government health insurance beneficiaries, the health insurer must then examine billing records in its possession in an effort to identify care episodes that were related to the bodily injury in question, and then indicate its intention to request reimbursement from the parties for that care. A challenge here is that medical billing systems often do not capture the information necessary to determine whether a particular care episode arose due to a tort or for some other reason, and the government insurer has incentives to be expansive in its requests for reimbursement. Following the initial demand, the governmental insurer and parties to the claim must undertake a reconciliation process to determine which episodes of care are subject to subrogation, and doing so may require multiple additional requests for information to the claimant and her health care providers. Moreover, once the parties have determined which care episodes are subject to subrogation, there is still scope for negotiation over the appropriate reimbursement rates, with the claimant facing incentives to minimize subrogation payments in order maximize her take-home award, and the government insurers facing opposite incentive to maximize their recovery by depleting the settlement. Once the liability insurer and claimant have determined what portion of the settlement will be paid to the government insurer, then they are in a position to negotiate a final settlement. Each of these steps in the process can further delay settlement. Contemporary accounts produced around the time of the passage of the new reporting requirements highlight some of the difficulties that arose once Medicare became more involved in settling claims. Andrews (2013), for example, described a case involving an 80-year-old auto injury victim who had to wait over a year for Medicare to calculate how much it was owed in his case; after the yearlong wait, it submitted a demand letter asking for nearly 10 times what it was actually owed. In another account, a plaintiff’s attorney described a case in which the total settlement amount was less than the attorney’s expenses in pursuing the case; Medicare nonetheless required documentation of the contingency fee contract and proposed settlement before it would relinquish its rights to collection, delaying the final resolution of the case (Hagy 2010). In testimony before Congress, another plaintiff attorney described numerous difficulties in getting clear information from Medicare necessary to settle cases, including in an ongoing case in which a |${\$}$|400,000 personal injury settlement obtained on behalf of an elderly client in failing health was sitting in escrow because of Medicare’s inaction (Matzus 2011). While there are clear reasons to imagine that the introduction of reporting and subrogation requirements could delay the resolution of claims, to date there has been virtually no empirical work documenting the existence of or magnitude of such effects. To my knowledge, there is a single existing study that attempts to measure the effects of Medicare reporting on settlement delay. Using a DD research design that also leverages the introduction of the new reporting requirement, Helland and Klick (2018) find that reporting increases the time to case resolution in auto injury cases by about 6 months. An important limitation of that study is its use of data from only a single large insurer; this study expands upon that research in that it includes larger and more nationally representative set of insurers. Additionally, due to data limitations, the unaffected control group in the Helland and Klick study constitutes less than 1% of the overall sample; in the present study, there are substantial numbers of claimants not subject to the reporting requirement who can provide a clear counterfactual for what might have happened to the Medicare patients post-2011 had no reporting requirement been in place. 3. Data and Empirical Approach Analyzing the effects of mandatory reporting requires access to data that includes otherwise similar personal injury claims, some of which were and were not subject to a reporting requirement. My data are drawn from the Insurance Research Council’s 2007 and 2012 Closed Claim databases. These databases include abstracted information from a sample of auto liability claims closed in the given year from all 50 states and the District of Columbia; participating insurers comprise more than half of the U.S. personal passenger automobile market, and therefore these data are likely broadly representative of auto tort claims in the U.S. For each claim, we observe the date the claim was filed and the date of final resolution (which can be used to construct the time to resolution), claimant demographic information, information about the nature and severity of the accident, claimed injuries and other financial losses, medical treatment received, and the total claimed loss. We also observe the final claim payment amount.12 For the bulk of the analysis, I focus attention on claimants aged 50 and over filing bodily injury claims. Because this is a closed claim sample, there is no censoring of claim times, moreover, because the payment date rather than the claim initiation date is the operative date for triggering the reporting requirements, we know that claims closed in 2007 were not subject to the Medicare reporting requirements, whereas those in 2012 were. For the 2012 claims, because claims would have been reportable to Medicare under the new requirements, participants were incentivized to query Medicare in advance to determine whether claimants were beneficiaries and, if so, to attempt to establish the amounts of conditional payments that would need to be reimbursed to Medicare prior to settling the claim. In the case of claimants over 65, the vast majority were beneficiaries and would need to undergo the reconciliation process or risk settling a claim but then needing to reimburse Medicare for its conditional payments separately, either in addition to the full claim payment on the insurer side, or by paying out of the settlement for the claimant. To measure the effects of the reporting requirement, I exploit the timing of the introduction of the requirement coupled with the fact that Medicare is targeted primarily to seniors aged 65 and over. I adopt a DD research design that contrasts claim resolution time for individuals above and below age 65 at the time of settlement, before and after the introduction of the new reporting requirement. In particular, I estimate regressions of the following form: $$\begin{align}\label{eq1} &\textit{SettlementTime}_{i}\nonumber\\ &\quad{} = \alpha \cdot (\textit{Age}_{i}>65 \times \textit{Post}_{i}) + \beta_{1} \cdot \textit{Age} + \beta_{2} \cdot \textit{Post}_{i} + \beta_{3} \cdot X_{i}, \end{align} $$(1) where SettlementTime|$_{i}$| represents the time in days required to settle the claim filed by individual |$i$|⁠, Age|$_{i} >65 \times \textit{Post}_{i}$| is an interaction term equal to |$1$| for individuals ages 65 and older with claims closed after 2011, when the requirement was enacted, Age represents a vector of claimant age fixed effects, Post|$_{i}$| is a dummy variable for a claim closed in 2012 (as opposed to 2007), and |$X_{i}$| a vector of additional controls measured at the individual level. In this regression, |$\alpha $| yields the DD estimate of the effect of reporting and subrogation on case resolution time. Given the long right tail for case settlement times, in most specifications I implement this equation using log settlement time as the outcome. Table 1 reports summary statistics capturing average outcomes for the overall sample and by age group and year of claim. For claimants under age 65, there was a reduction in various indicia of claim severity (rate of disability, lost work days, emergency room (ER), and physician use, etc.) at least on average between 2007 and 2012, and a corresponding reduction in the average time required to resolve the claim. Older claimants also saw an apparent decline in the seriousness of injury and complexity of claims, but their average claim resolution time increased between 2007 and 2012. Comparing the young and old, as expected, older claimants are less likely to be employed, but the groups actually have roughly similar patterns of disability and medical care utilization in these data. Table 1. Summary Statistics . . . . Average for claimants: . . Overall . 2007 . 2012 . Characteristic . |$N$| . Mean . SD . Age |$<$| 65 . Age |$ \ge $| 65 . Age |$<$| 65 . Age |$ \ge $| 65 . Time to claim resolution (days) 9,431 422.2 434.0 433.8 420.6 403.1 444.2 Age (years) 10,058 60.8 9.26 55.9 73.5 56.0 72.6 Male 10,017 0.431 0.495 0.424 0.422 0.444 0.423 Employed 7,357 0.587 0.492 0.751 0.238 0.753 0.226 Degree of fault (%) 10,044 3.64 15.1 3.62 4.49 3.35 3.60 Accident location  Large city 10,002 0.355 0.479 0.367 0.309 0.369 0.335  Medium city 10,002 0.349 0.477 0.320 0.320 0.366 0.403  Suburb/town/rural 10,002 0.296 0.457 0.313 0.371 0.264 0.262  Impact severity  None/minor 9,489 0.294 0.456 0.279 0.211 0.334 0.308  Moderate 9,489 0.529 0.499 0.542 0.542 0.520 0.505  Major 9,489 0.178 0.382 0.179 0.247 0.146 0.187 Most serious injury  Sprain/strain 9,748 0.676 0.468 0.704 0.563 0.718 0.617  Knee, disc, or shoulder injury 9,748 0.120 0.325 0.125 0.160 0.105 0.107  Fracture 9,748 0.059 0.236 0.049 0.094 0.048 0.079  Other 9,748 0.144 0.351 0.122 0.183 0.129 0.197 Extent of disability  None 9,846 0.737 0.440 0.726 0.729 0.737 0.770  Temporary 9,846 0.204 0.403 0.205 0.183 0.221 0.180  Permanent 9,846 0.052 0.221 0.059 0.073 0.038 0.045 Medical care received  Went to ER 10,058 0.421 0.494 0.429 0.485 0.388 0.425  Imaging performed 10,058 0.638 0.481 0.681 0.733 0.571 0.607  Visited chiropractor 10,058 0.340 0.474 0.370 0.257 0.365 0.284  Visited physical therapist 10,058 0.227 0.419 0.248 0.254 0.211 0.192 Total doctor visits 10,058 10.5 17.0 11.8 9.75 10.2 8.46 Number of lost work days 10,058 14.8 182.2 27.3 14.61 7.6 2.10 Hired attorney 9,982 0.499 0.500 0.506 0.481 0.504 0.484 Claimed loss amount ( |${\$}$|⁠) 9,007 |${\$}$|14,398 41917 |${\$}$|12,135 |${\$}$|14,639 |${\$}$|15,549 |${\$}$|16,980 Payment amount ( |${\$}$|⁠) 10,058 |${\$}$|14,450 30665 |${\$}$|14,011 |${\$}$|14,922 |${\$}$|14,171 |${\$}$|15,750 . . . . Average for claimants: . . Overall . 2007 . 2012 . Characteristic . |$N$| . Mean . SD . Age |$<$| 65 . Age |$ \ge $| 65 . Age |$<$| 65 . Age |$ \ge $| 65 . Time to claim resolution (days) 9,431 422.2 434.0 433.8 420.6 403.1 444.2 Age (years) 10,058 60.8 9.26 55.9 73.5 56.0 72.6 Male 10,017 0.431 0.495 0.424 0.422 0.444 0.423 Employed 7,357 0.587 0.492 0.751 0.238 0.753 0.226 Degree of fault (%) 10,044 3.64 15.1 3.62 4.49 3.35 3.60 Accident location  Large city 10,002 0.355 0.479 0.367 0.309 0.369 0.335  Medium city 10,002 0.349 0.477 0.320 0.320 0.366 0.403  Suburb/town/rural 10,002 0.296 0.457 0.313 0.371 0.264 0.262  Impact severity  None/minor 9,489 0.294 0.456 0.279 0.211 0.334 0.308  Moderate 9,489 0.529 0.499 0.542 0.542 0.520 0.505  Major 9,489 0.178 0.382 0.179 0.247 0.146 0.187 Most serious injury  Sprain/strain 9,748 0.676 0.468 0.704 0.563 0.718 0.617  Knee, disc, or shoulder injury 9,748 0.120 0.325 0.125 0.160 0.105 0.107  Fracture 9,748 0.059 0.236 0.049 0.094 0.048 0.079  Other 9,748 0.144 0.351 0.122 0.183 0.129 0.197 Extent of disability  None 9,846 0.737 0.440 0.726 0.729 0.737 0.770  Temporary 9,846 0.204 0.403 0.205 0.183 0.221 0.180  Permanent 9,846 0.052 0.221 0.059 0.073 0.038 0.045 Medical care received  Went to ER 10,058 0.421 0.494 0.429 0.485 0.388 0.425  Imaging performed 10,058 0.638 0.481 0.681 0.733 0.571 0.607  Visited chiropractor 10,058 0.340 0.474 0.370 0.257 0.365 0.284  Visited physical therapist 10,058 0.227 0.419 0.248 0.254 0.211 0.192 Total doctor visits 10,058 10.5 17.0 11.8 9.75 10.2 8.46 Number of lost work days 10,058 14.8 182.2 27.3 14.61 7.6 2.10 Hired attorney 9,982 0.499 0.500 0.506 0.481 0.504 0.484 Claimed loss amount ( |${\$}$|⁠) 9,007 |${\$}$|14,398 41917 |${\$}$|12,135 |${\$}$|14,639 |${\$}$|15,549 |${\$}$|16,980 Payment amount ( |${\$}$|⁠) 10,058 |${\$}$|14,450 30665 |${\$}$|14,011 |${\$}$|14,922 |${\$}$|14,171 |${\$}$|15,750 Notes: This table reposts summary statistics for the 2007 and 2012 Insurance Research Council Closed Auto Claim databases. The sample is limited to bodily injury claims involving claimants ages 50 and above at the time of settlement. Open in new tab Table 1. Summary Statistics . . . . Average for claimants: . . Overall . 2007 . 2012 . Characteristic . |$N$| . Mean . SD . Age |$<$| 65 . Age |$ \ge $| 65 . Age |$<$| 65 . Age |$ \ge $| 65 . Time to claim resolution (days) 9,431 422.2 434.0 433.8 420.6 403.1 444.2 Age (years) 10,058 60.8 9.26 55.9 73.5 56.0 72.6 Male 10,017 0.431 0.495 0.424 0.422 0.444 0.423 Employed 7,357 0.587 0.492 0.751 0.238 0.753 0.226 Degree of fault (%) 10,044 3.64 15.1 3.62 4.49 3.35 3.60 Accident location  Large city 10,002 0.355 0.479 0.367 0.309 0.369 0.335  Medium city 10,002 0.349 0.477 0.320 0.320 0.366 0.403  Suburb/town/rural 10,002 0.296 0.457 0.313 0.371 0.264 0.262  Impact severity  None/minor 9,489 0.294 0.456 0.279 0.211 0.334 0.308  Moderate 9,489 0.529 0.499 0.542 0.542 0.520 0.505  Major 9,489 0.178 0.382 0.179 0.247 0.146 0.187 Most serious injury  Sprain/strain 9,748 0.676 0.468 0.704 0.563 0.718 0.617  Knee, disc, or shoulder injury 9,748 0.120 0.325 0.125 0.160 0.105 0.107  Fracture 9,748 0.059 0.236 0.049 0.094 0.048 0.079  Other 9,748 0.144 0.351 0.122 0.183 0.129 0.197 Extent of disability  None 9,846 0.737 0.440 0.726 0.729 0.737 0.770  Temporary 9,846 0.204 0.403 0.205 0.183 0.221 0.180  Permanent 9,846 0.052 0.221 0.059 0.073 0.038 0.045 Medical care received  Went to ER 10,058 0.421 0.494 0.429 0.485 0.388 0.425  Imaging performed 10,058 0.638 0.481 0.681 0.733 0.571 0.607  Visited chiropractor 10,058 0.340 0.474 0.370 0.257 0.365 0.284  Visited physical therapist 10,058 0.227 0.419 0.248 0.254 0.211 0.192 Total doctor visits 10,058 10.5 17.0 11.8 9.75 10.2 8.46 Number of lost work days 10,058 14.8 182.2 27.3 14.61 7.6 2.10 Hired attorney 9,982 0.499 0.500 0.506 0.481 0.504 0.484 Claimed loss amount ( |${\$}$|⁠) 9,007 |${\$}$|14,398 41917 |${\$}$|12,135 |${\$}$|14,639 |${\$}$|15,549 |${\$}$|16,980 Payment amount ( |${\$}$|⁠) 10,058 |${\$}$|14,450 30665 |${\$}$|14,011 |${\$}$|14,922 |${\$}$|14,171 |${\$}$|15,750 . . . . Average for claimants: . . Overall . 2007 . 2012 . Characteristic . |$N$| . Mean . SD . Age |$<$| 65 . Age |$ \ge $| 65 . Age |$<$| 65 . Age |$ \ge $| 65 . Time to claim resolution (days) 9,431 422.2 434.0 433.8 420.6 403.1 444.2 Age (years) 10,058 60.8 9.26 55.9 73.5 56.0 72.6 Male 10,017 0.431 0.495 0.424 0.422 0.444 0.423 Employed 7,357 0.587 0.492 0.751 0.238 0.753 0.226 Degree of fault (%) 10,044 3.64 15.1 3.62 4.49 3.35 3.60 Accident location  Large city 10,002 0.355 0.479 0.367 0.309 0.369 0.335  Medium city 10,002 0.349 0.477 0.320 0.320 0.366 0.403  Suburb/town/rural 10,002 0.296 0.457 0.313 0.371 0.264 0.262  Impact severity  None/minor 9,489 0.294 0.456 0.279 0.211 0.334 0.308  Moderate 9,489 0.529 0.499 0.542 0.542 0.520 0.505  Major 9,489 0.178 0.382 0.179 0.247 0.146 0.187 Most serious injury  Sprain/strain 9,748 0.676 0.468 0.704 0.563 0.718 0.617  Knee, disc, or shoulder injury 9,748 0.120 0.325 0.125 0.160 0.105 0.107  Fracture 9,748 0.059 0.236 0.049 0.094 0.048 0.079  Other 9,748 0.144 0.351 0.122 0.183 0.129 0.197 Extent of disability  None 9,846 0.737 0.440 0.726 0.729 0.737 0.770  Temporary 9,846 0.204 0.403 0.205 0.183 0.221 0.180  Permanent 9,846 0.052 0.221 0.059 0.073 0.038 0.045 Medical care received  Went to ER 10,058 0.421 0.494 0.429 0.485 0.388 0.425  Imaging performed 10,058 0.638 0.481 0.681 0.733 0.571 0.607  Visited chiropractor 10,058 0.340 0.474 0.370 0.257 0.365 0.284  Visited physical therapist 10,058 0.227 0.419 0.248 0.254 0.211 0.192 Total doctor visits 10,058 10.5 17.0 11.8 9.75 10.2 8.46 Number of lost work days 10,058 14.8 182.2 27.3 14.61 7.6 2.10 Hired attorney 9,982 0.499 0.500 0.506 0.481 0.504 0.484 Claimed loss amount ( |${\$}$|⁠) 9,007 |${\$}$|14,398 41917 |${\$}$|12,135 |${\$}$|14,639 |${\$}$|15,549 |${\$}$|16,980 Payment amount ( |${\$}$|⁠) 10,058 |${\$}$|14,450 30665 |${\$}$|14,011 |${\$}$|14,922 |${\$}$|14,171 |${\$}$|15,750 Notes: This table reposts summary statistics for the 2007 and 2012 Insurance Research Council Closed Auto Claim databases. The sample is limited to bodily injury claims involving claimants ages 50 and above at the time of settlement. Open in new tab The rich information available regarding each claim permits us to control for a range of factors shown in Table 1 likely to impact the complexity of the claim, including the demographics of the claimant, accident characteristics (e.g. location, number of vehicles, impact severity), nature and severity of injury (45 variables), types and amounts of medical care received (38 variables), and liability characteristics (e.g. policy limits, degree of fault of insurer, attorney involvement). Conceptually, this means that we are able to compare two claimants involved in similar accidents, with similar injuries and medical treatment, facing legally comparable claims, but who differ in whether or not their claim must undergo the reporting and subrogation process. While my primary outcome of interest is time to claim resolution, I also estimate versions of the regression specification above where I use the payment amount or number of claims as the outcome. Characteristics of state tort law, such as whether the state is a no-fault state, rules governing bad faith claims, and contingency fee rules are also likely to affect the claims resolution process. For example, states vary in their treatment of the collateral source rule, a doctrine governing whether payments from other sources such as health insurance can be considering in determining the amount of damages in a personal injury suit. It seems plausible to imagine that the status of the collateral source rule might affect which claims are filed and how long they take to resolve. Because the identification strategy relies on contrasts across age groups, I can include in the regressions as additional controls a full set of state by year of injury fixed effects. These controls mitigate any potential omitted variable bias from state-specific factors that change over time, including features of tort law such as the collateral source rule along with any other laws and regulations related to general medical care or vehicle safety. The main underlying assumption of the DD analysis is that, after adjusting for underlying claim characteristics, the claims involving individuals ages 50–64 provide an appropriate counterfactual for what claim resolution times for the older group would have been, absent the new reporting requirement. These estimates of the effect of reporting and subrogation on delay are likely to be conservative for two reasons. First, as noted previously, a new reporting requirement affects all future claims in that they must, at a minimum, have claimant information entered into a query system to establish whether further reporting is necessary. Thus, in the analysis, the control group is in fact affected by the new requirement, but to a lesser extent than those who are Medicare eligible. We imagine that, once a sufficient electronic query process is in place, the delay effects of querying alone are small relative to the effects of querying, receiving a hit, and then the subsequent reporting and reconciliation process. The research design is best suited to measure the difference in claim resolution time between otherwise similar claims that required querying and reporting as compared to querying alone. In other words, to the extent that the query process itself generates delay, this analysis is not ideally suited to capture such effects. A second reason these estimates are conservative is that some portion of the under 65 population are in fact Medicare eligible, and therefore would have to complete the entire reporting process. Because Medicare eligibility is unobserved in the data I cannot directly account for this; however, data from the American Community Survey and other sources (e.g. Card et al. 2009) suggest that roughly 10% of adults ages 55–64 are Medicare eligible. This, a modest fraction of the individuals who are labeled as controls in this analysis actually undergo the entire reporting process; this misclassification would tend to attenuate any differences estimated using a DD analysis. Similarly, a small segment of the 65 and older population are noncitizens or otherwise ineligible for Medicare; Barnett and Vornovitsky (2016) estimate this population to comprise about 6% of those ages 65 and older. 4. Results Table 2 reports estimates of the effect of mandatory reporting and subrogation on the amount of time needed to settle a case. Column I reports results from a simple DD specification as described above with age and year of claim fixed effects but no further controls. The estimated coefficient of 0.178, which is statistically significant at the 1% level,13 implies that the new reporting requirement is associated with a 19% increase in the amount of time needed to make final payment on a claim. Relative to the mean, this represents an increase of 75 days to resolve a claim. Table 2. Baseline Estimates of the Effects of Reporting on Claim Resolution Time . I . II . III . IV . V . DD estimate 0.178*** 0.203*** 0.215*** 0.188*** 0.176*** (0.065) (0.060) (0.056) (0.053) (0.053) |$N$| 9,431 9,419 9,419 9,419 9,419 |$R^{2}$| 0.009 0.184 0.328 0.389 0.414 Control for claimant demographics, accident, and injury characteristics? N Y Y Y Y Control for medical care received? N N Y Y Y Control for attorney representation? N N N Y Y Include state/year fixed effects? N N N N Y . I . II . III . IV . V . DD estimate 0.178*** 0.203*** 0.215*** 0.188*** 0.176*** (0.065) (0.060) (0.056) (0.053) (0.053) |$N$| 9,431 9,419 9,419 9,419 9,419 |$R^{2}$| 0.009 0.184 0.328 0.389 0.414 Control for claimant demographics, accident, and injury characteristics? N Y Y Y Y Control for medical care received? N N Y Y Y Control for attorney representation? N N N Y Y Include state/year fixed effects? N N N N Y Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on personal injury claim resolution speed. The unit of observation is a claim, and the outcome variable is the logged number of days between the initial injury report and final claim payment. Claimants 65 and over who had claims resolved in 2012 are the treated group subject to mandatory reporting; differences are taken across age groups and across years. Column I includes as controls a set of age of claimant fixed effects (51 categories) and year of claim fixed effects (2 categories). Column II adds controls for policy limits (9 categories), accident location (urban/rural, 5 categories), number of vehicles involved (3 categories), whether there were passengers in the vehicle, impact severity (5 categories), injury severity at scene (6 categories); claimant role in the accident (driver/passenger/etc., 7 categories), relationship to insured (3 categories), sex, employment status (3 categories), degree of fault (7 categories), presence of injuries (25 categories), most severe injury (22 categories), and disability status (3 categories). Column III adds controls for whether the claimant used various medical services (21 categories) and the number of visits made to each of 16 different categories of providers. Column IV adds controls for whether the claimant hired an attorney. Column V adds a full set of state of accident by claim year interactions (102 categories) as controls. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 2. Baseline Estimates of the Effects of Reporting on Claim Resolution Time . I . II . III . IV . V . DD estimate 0.178*** 0.203*** 0.215*** 0.188*** 0.176*** (0.065) (0.060) (0.056) (0.053) (0.053) |$N$| 9,431 9,419 9,419 9,419 9,419 |$R^{2}$| 0.009 0.184 0.328 0.389 0.414 Control for claimant demographics, accident, and injury characteristics? N Y Y Y Y Control for medical care received? N N Y Y Y Control for attorney representation? N N N Y Y Include state/year fixed effects? N N N N Y . I . II . III . IV . V . DD estimate 0.178*** 0.203*** 0.215*** 0.188*** 0.176*** (0.065) (0.060) (0.056) (0.053) (0.053) |$N$| 9,431 9,419 9,419 9,419 9,419 |$R^{2}$| 0.009 0.184 0.328 0.389 0.414 Control for claimant demographics, accident, and injury characteristics? N Y Y Y Y Control for medical care received? N N Y Y Y Control for attorney representation? N N N Y Y Include state/year fixed effects? N N N N Y Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on personal injury claim resolution speed. The unit of observation is a claim, and the outcome variable is the logged number of days between the initial injury report and final claim payment. Claimants 65 and over who had claims resolved in 2012 are the treated group subject to mandatory reporting; differences are taken across age groups and across years. Column I includes as controls a set of age of claimant fixed effects (51 categories) and year of claim fixed effects (2 categories). Column II adds controls for policy limits (9 categories), accident location (urban/rural, 5 categories), number of vehicles involved (3 categories), whether there were passengers in the vehicle, impact severity (5 categories), injury severity at scene (6 categories); claimant role in the accident (driver/passenger/etc., 7 categories), relationship to insured (3 categories), sex, employment status (3 categories), degree of fault (7 categories), presence of injuries (25 categories), most severe injury (22 categories), and disability status (3 categories). Column III adds controls for whether the claimant used various medical services (21 categories) and the number of visits made to each of 16 different categories of providers. Column IV adds controls for whether the claimant hired an attorney. Column V adds a full set of state of accident by claim year interactions (102 categories) as controls. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Columns II–V of Table 2 progressively introduce additional sets of control variables to the DD specification. Column II adds controls for claimant demographics, accident characteristics (e.g. severity), detailed measures of claimant injury, and some of the legal characteristics of the case (e.g. estimated degree of fault of the claimant). Column III adds numerous further controls for the types of medical care received by the claimant. While it seems unlikely that the volume of treatment received should respond to the reporting requirement because treatment is largely determined prior to the point at which the reconciliation process with Medicare begins, in theory it is possible that reporting could affect treatment, in which case controlling for treatment would be problematic. However, an advantage of controlling for medical care received is that it seems reasonable to expect that, for a given injury, older patients who may be in worse baseline health would likely require more treatment than younger patients, and these controls would account for any such differences. In any case, across both variants, the relationship between the reporting requirement and delay remains statistically significant and of similar magnitude. Column IV adds additional controls for whether the claimant hired an attorney—another potentially endogenous variable, but also one that is likely to have important impacts on the speed of settlement. Although controlling for attorney presence appreciably increases the |$R^{2}$| of the regression, it reassuringly does not affect the estimated impact of the reporting requirement. In my preferred estimate in column V—which adds a full set of state and year fixed effects, and is therefore robust to unobserved state-level legal and other environmental changes over time—mandatory reporting is associated with a 19% increase in the time needed to resolve a claim.14 When calculated as an average marginal effect, this represents a 58 day delay. Despite the fact that this is an individual-level regression, the available controls explain a sizeable portion of the overall variance in claim resolution time, suggesting that we observe many of the factually and legally relevant factors that affect how quickly claims can be resolved. Although the most straightforward interpretation of the results above is that the new reporting requirement increased delay, an alternative interpretation is that the requirement did not directly affect delay, but rather altered the composition of claims within the system so as to reduce the average claim duration. One simple story would be that the possibility of Medicare subrogation made some lower dollar claims no longer economical for a claimant or their attorneys to pursue, because once Medicare removed its portion of the settlement, the remaining amount going to the plaintiff would no longer cover the costs of pursuing the claim. If such lower value claims were also easier to resolve, then after they were removed from the system, we would expect to observe longer average settlement times for the remaining claims. Note that under this scenario as before, the introduction of the requirement does in fact affect claims; however, under the conventional interpretation, its takes a given claim and prolongs it, whereas under the latter interpretation, it drives some claims out of the system. To examine this possibility, in Table 3, I report coefficients from regressions similar to those presented previously but using the log total payment as the outcome variable. In these regressions, we observe a 9% increase in the total payment as a result of the reform. This regression coefficient combines two impacts—any selection effect that might occur as claims drop out of the system, and any change in the final settlement amount that would be induced by the greater possibility of subrogation by Medicare. This latter quantity seems likely to, if anything, be positive—following the reporting change, the claimant would have to obtain a higher settlement in order achieve the same level of take-home compensation because of the subrogation. Thus both effects would move in the same direction. Table 3. Effects of Reporting on Payment Amount and Claim Volume . Log(total payment) . Number of claims . DD estimate 0.089** 0.48 (0.042) (6.89) |$N$| 10,046 82 |$R^{2}$| 0.658 0.991 . Log(total payment) . Number of claims . DD estimate 0.089** 0.48 (0.042) (6.89) |$N$| 10,046 82 |$R^{2}$| 0.658 0.991 Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on additional claim outcomes. Column I reports a regression similar to the baseline where the dependent variable is the log total claim payment in dollars. Here the specification is as in column V of Table 2; see notes for Table 2. Column II reports coefficients from simple DD regressions where the unit of observation is an age/claim year cell and the outcome is the number of claims in the database; this regression includes age and year fixed effects as controls. The mean of the dependent variable for column 2 is 127. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 3. Effects of Reporting on Payment Amount and Claim Volume . Log(total payment) . Number of claims . DD estimate 0.089** 0.48 (0.042) (6.89) |$N$| 10,046 82 |$R^{2}$| 0.658 0.991 . Log(total payment) . Number of claims . DD estimate 0.089** 0.48 (0.042) (6.89) |$N$| 10,046 82 |$R^{2}$| 0.658 0.991 Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on additional claim outcomes. Column I reports a regression similar to the baseline where the dependent variable is the log total claim payment in dollars. Here the specification is as in column V of Table 2; see notes for Table 2. Column II reports coefficients from simple DD regressions where the unit of observation is an age/claim year cell and the outcome is the number of claims in the database; this regression includes age and year fixed effects as controls. The mean of the dependent variable for column 2 is 127. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab To further examine any selection effects, in the second specification, I collapse the data to the claimant age/year level and estimate a DD regression on the collapsed data to assess whether there was a change in the number of claims originating from those subject to the new reporting requirement. This provides a more direct test for whether the number of claims fell for the target population after the reporting requirement was introduced. There is no measurable change in the volume of claims due to the reform, although admittedly the estimate is somewhat imprecise. Together these results suggest that the new requirement may have caused modest changes in the composition of claims, although the data are not fully dispositive.15 Table 4 reports a series of robustness checks designed to test the sensitivity of the results to changes to the sample and specification. For reference, the first row reports the baseline results. Specification 1 adds flexible controls16 for the log total payment amount as additional controls. Payment amounts might be endogenous, which is why I do not control for them in the preferred specification. However, given that more complex claims seem likely to result in higher payments but also take longer to resolve, controlling flexibly for payment amount may provide a useful means to partly account for unobservable age-related differences in the complexity of claims. While including these controls appreciably increases the share of the variance of the outcome that is explained by the model, it yields a similar estimated effect of reporting. Table 4. Robustness Checks . DD Estimate . |$N$| . |$R^{2}$| . 0. Baseline estimate 0.176*** 9,419 0.414 (0.053) 1. Include controls for total payment amount 0.146*** 9,419 0.516 (0.048) 2. Estimate in levels rather than logs 54.3*** 9,419 0.366 (16.3) 3. Matching model 0.086** 9,171 (0.037) 4. Inverse probability weighting model 0.109** 4,853 (0.048) . DD Estimate . |$N$| . |$R^{2}$| . 0. Baseline estimate 0.176*** 9,419 0.414 (0.053) 1. Include controls for total payment amount 0.146*** 9,419 0.516 (0.048) 2. Estimate in levels rather than logs 54.3*** 9,419 0.366 (16.3) 3. Matching model 0.086** 9,171 (0.037) 4. Inverse probability weighting model 0.109** 4,853 (0.048) Notes: This table reports coefficient estimates from alternatives to the baseline specification. Except as otherwise noted, the specification is as in column V of Table 2; see notes for Table 2. Specification 1 includes the log total payment amount interacted with indicators for deciles of the payment amount as additional controls. Specification 2 uses the claim time measured in days as the outcome variable rather than log(claim time). Specification 3 implements a matching model by first estimating the predicted claim time in days as a function of all the covariates in the model using Poisson regression, and then, in a second stage, modeling log claim time as a function of treatment and a full set of fixed effects for predicted claim time (in days) as controls. Specification 4 limits the sample to 2012 claimants; models the probability of treatment as a function of the other covariates in the model, and then estimates treatment effects via inverse probability weighting. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 4. Robustness Checks . DD Estimate . |$N$| . |$R^{2}$| . 0. Baseline estimate 0.176*** 9,419 0.414 (0.053) 1. Include controls for total payment amount 0.146*** 9,419 0.516 (0.048) 2. Estimate in levels rather than logs 54.3*** 9,419 0.366 (16.3) 3. Matching model 0.086** 9,171 (0.037) 4. Inverse probability weighting model 0.109** 4,853 (0.048) . DD Estimate . |$N$| . |$R^{2}$| . 0. Baseline estimate 0.176*** 9,419 0.414 (0.053) 1. Include controls for total payment amount 0.146*** 9,419 0.516 (0.048) 2. Estimate in levels rather than logs 54.3*** 9,419 0.366 (16.3) 3. Matching model 0.086** 9,171 (0.037) 4. Inverse probability weighting model 0.109** 4,853 (0.048) Notes: This table reports coefficient estimates from alternatives to the baseline specification. Except as otherwise noted, the specification is as in column V of Table 2; see notes for Table 2. Specification 1 includes the log total payment amount interacted with indicators for deciles of the payment amount as additional controls. Specification 2 uses the claim time measured in days as the outcome variable rather than log(claim time). Specification 3 implements a matching model by first estimating the predicted claim time in days as a function of all the covariates in the model using Poisson regression, and then, in a second stage, modeling log claim time as a function of treatment and a full set of fixed effects for predicted claim time (in days) as controls. Specification 4 limits the sample to 2012 claimants; models the probability of treatment as a function of the other covariates in the model, and then estimates treatment effects via inverse probability weighting. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Given that some claims take years to resolve, and therefore the claim time distribution is right-skewed, the data are best fit using a logged model; however, in the next robustness check I estimate the reporting/claim time relationship where the dependent variable is measured in levels. The point estimate indicates a 54 day increase in claim resolution time on average, an estimate that is highly statistically significant and in line with the implied effect from the logged model. The final two robustness checks implement alternatives to the simple DD model. Specification 3 employs a matching-type estimator to estimate effects. I first use Poisson regression to model the expected number of days to claim resolution as a function of claimant age, year of claim, and the other covariates in the model and then, for each observation, I construct the expected claim duration based on these observable covariates. Then, in a second stage, I estimate a regression where log claim time is the outcome, the primary explanatory variable is the treatment interaction ( |$\textit{Age}_{i}>65 \times \textit{Post}_{i}$|⁠), and I include a full set of fixed effects for predicted claim time (in single days) as additional controls. This has the effect of measuring effects by matching each treated subject to the set of control subjects with the exact same expected claim complexity based upon the underlying observable characteristics of their claim. This specification yields a smaller estimated effect, but it remains statistically and economically significant. The final robustness check focuses attention on claims settled in 2012 and obtains an average treatment effect estimate by comparing treated and comparison subjects via inverse probability weighting. Here, I construct the probability weights by modeling the likelihood of treatment as a function of the various covariates included in the model ( |$X_{i}$|⁠). This matching approach yields an estimated impact of 0.109 which remains statistically significant. Overall, these robustness checks indicate the baseline finding regarding the magnitude and significance of the reporting effect is robust. How sensitive are the results the choice to focus on claimants ages 50 and over? For example, if we limited the analysis to ages just above and below the reporting cutoff, would we observe similar impacts? To explore this issue, the Appendix plots results from a series of regressions that systematically vary the age bandwidth around age 65, the most common age at which people gain Medicare.17 Outside of very narrow bandwidths, which, as discussed in the Appendix, are likely affected by measurement error, the estimated impacts are remarkably robust and consistent across age bands. As an additional robustness check, I examine whether any indication of delay can be found in another data set, the National Practitioner Data Bank (NPDB).18 The NPDB is a database of paid medical malpractice claims that has been widely used in research on malpractice.19 Relative to the auto claims data used thus far, the NPDB has the advantage of including an appreciably larger number of claims and covering more years. However, the NPDB is not without drawbacks for examining the effects of the new reporting requirement. The NPDB only offers coarse measures of claim resolution time, with the public use data only reporting the year of injury and year of resolution. Moreover, patient age at the time of alleged injury is only reported in 10-year categories (20–29, 30–39, etc.), meaning that treatment status for those ages 60–69 cannot be clearly assigned. Additionally, the NPDB contains much less information about the nature of injury and types of medical care received than the Insurance Research Council (IRC) auto closed claim data, limiting my ability to account for other claim characteristics that might affect treatment. To estimate the effects of the new reporting requirement on medical malpractice claim resolution time, I estimate claim-level differences-in-differences regressions on a sample of 115,622 claims resolved between January 2004 and March 2018 involving claimants aged 40 and up at the time of injury. Here, the outcome is the number of years between injury and claim payment, and, as in equation 1, the primary explanatory variable is an interaction for a claim involving an individual over age 65 where payment occurred in January 2012 or later. Lacking precise information on who was above the Medicare eligibility age, I code treatment at the intermediate value of 0.5 for claims involving 60–69 year-olds that were paid on or after January 1, 2012.20 I also control for state by year fixed effects and indicators for patient 10-year age group, patient gender, patient type (inpatient/outpatient), physician 10-year age group, physician specialty, alleged form of malpractice, injury severity, and type of insurer involved in making payment. The results of the NPDB analysis are reported in Table 5. In the baseline DD specification where the outcome is measured in levels, the new reporting requirement is associated with a statistically significant increase in claim resolution time of 0.07 years, or 26 days. This finding is robust to excluding individuals ages 60–69—for whom Medicare eligibility status cannot be readily assigned—from the analysis (Specification 2) and to log transforming the outcome to account for the right-skewed nature of the malpractice claim time distribution (Specification 3).21 While the magnitude of the estimated delay effect is much smaller in percentage terms for malpractice claims than auto claims, due to the greater length and complexity of the former, in absolute terms we observe a delay of about one month in malpractice cases, which is on par with the estimate obtained from the more detailed auto closed claim data.22 This robustness check using a different data source thus seems to reinforce the conclusion that the new Medicare reporting and subrogation requirements increased liability claim resolution time. Table 5. Estimated Effects for Malpractice Claims . DD . Implied delay . . . . estimate . in days . |$N$| . |$R^{2}$| . 1. Baseline estimate in levels 0.072** 26.3 115,370 0.198 (0.035) 2. Exclude individuals ages 60–69 0.097*** 35.4 89,927 0.198 (0.036) 3. Estimate in logs rather than levels 0.013** 21.9 115,370 0.230 (0.006) . DD . Implied delay . . . . estimate . in days . |$N$| . |$R^{2}$| . 1. Baseline estimate in levels 0.072** 26.3 115,370 0.198 (0.035) 2. Exclude individuals ages 60–69 0.097*** 35.4 89,927 0.198 (0.036) 3. Estimate in logs rather than levels 0.013** 21.9 115,370 0.230 (0.006) Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on malpractice claim resolution speed using data from the National Practitioner Data Bank. The unit of observation is a claim, and the outcome variable is the number of years between the initial injury and claim payment (mean |$= 4.44$|⁠). Differences are taken across age groups and across payment years. Because the NPDB only reports claimant ages in 10-year increments, claimants below age 60 are part of the control group; claimants ages 60–69 are coded as untreated prior to 2012 and 0.5 treated in 2012 and later, and claimants ages 70 and over are coded as untreated prior to 2012 and treated in 2012 and later. Additional controls include a full set of state by payment year interactions and indicators for claimant age and gender; physician age and specialty; injury severity; alleged type of malpractice; and type of insurer. Specification 1 is the baseline. Specification 2 omits individuals aged 60–69 form the analysis in light of the ambiguity regarding whether any particular individual is Medicare eligible. Specification 3 measures the dependent variable in logs rather than levels. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 5. Estimated Effects for Malpractice Claims . DD . Implied delay . . . . estimate . in days . |$N$| . |$R^{2}$| . 1. Baseline estimate in levels 0.072** 26.3 115,370 0.198 (0.035) 2. Exclude individuals ages 60–69 0.097*** 35.4 89,927 0.198 (0.036) 3. Estimate in logs rather than levels 0.013** 21.9 115,370 0.230 (0.006) . DD . Implied delay . . . . estimate . in days . |$N$| . |$R^{2}$| . 1. Baseline estimate in levels 0.072** 26.3 115,370 0.198 (0.035) 2. Exclude individuals ages 60–69 0.097*** 35.4 89,927 0.198 (0.036) 3. Estimate in logs rather than levels 0.013** 21.9 115,370 0.230 (0.006) Notes: This table reports coefficient estimates from a differences-in-differences regression designed to measure the effects of mandatory reporting on malpractice claim resolution speed using data from the National Practitioner Data Bank. The unit of observation is a claim, and the outcome variable is the number of years between the initial injury and claim payment (mean |$= 4.44$|⁠). Differences are taken across age groups and across payment years. Because the NPDB only reports claimant ages in 10-year increments, claimants below age 60 are part of the control group; claimants ages 60–69 are coded as untreated prior to 2012 and 0.5 treated in 2012 and later, and claimants ages 70 and over are coded as untreated prior to 2012 and treated in 2012 and later. Additional controls include a full set of state by payment year interactions and indicators for claimant age and gender; physician age and specialty; injury severity; alleged type of malpractice; and type of insurer. Specification 1 is the baseline. Specification 2 omits individuals aged 60–69 form the analysis in light of the ambiguity regarding whether any particular individual is Medicare eligible. Specification 3 measures the dependent variable in logs rather than levels. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab A key assumption needed for the baseline estimates above to properly capture the effects of reporting is that the younger claimants provide an appropriate counterfactual for what would have happened to older claimants absent the policy reform. Because the analysis essentially uses only two time periods, if there were differential time trends in expected injury costs by age, such trends might be confounded with the new policy. While the underlying DD assumption is not directly testable, I can perform a similar analysis but contrast years prior to the policy reform as a falsification exercise. If there are similar time trends by age and if younger claimants are sufficiently comparable to older claimants so as to render them a good control group, we should expect that when I conduct a similar DD analysis across years where no policy intervention occurred, I should measure no effect. The first row of Table 6 reports the results from such an analysis, where I use data from the 2002 edition of the IRC closed claim survey as a preperiod, and the 2007 data as the post period. The point estimate from this DD regression is small, nonsignificant, and fairly precisely measured. This falsification test suggests that the analysis is capturing the effects of the new reporting requirement and does not simply reflect noncomparabilities between older and younger auto injury claimants. Moreover, if older claimants are experiencing different time trends from younger claimants, we would expect to observe some measurable “effect” in these regressions, yet we do not. Table 6. Falsification Tests . DD Estimate . |$N$| . |$R^{2}$| . 1. Compare 2007 to 2002 -0.023 10,105 0.421 (0.051) 2. PIP and Medpay claims rather than bodily injury (BI) claims -0.021 8,104 0.316 (0.046) . DD Estimate . |$N$| . |$R^{2}$| . 1. Compare 2007 to 2002 -0.023 10,105 0.421 (0.051) 2. PIP and Medpay claims rather than bodily injury (BI) claims -0.021 8,104 0.316 (0.046) Notes: This table reports coefficient estimates from regressions designed to serve as placebo test of the baseline specification. In the first specification, the years of data are 2002 and 2007, and 2007 is used as the implementation year, so claimants ages 65 and older in 2007 are the treated group. In the second specification, the sample is comprised of PIP and Medpay claims from the 2012 and 2007 IRC database rather than BI claims. Except as noted below, the specification is as in column V of Table 2; see notes for Table 2. The second specification omits driver degree of fault as a control as that does not apply to PIP and Medpay claims. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 6. Falsification Tests . DD Estimate . |$N$| . |$R^{2}$| . 1. Compare 2007 to 2002 -0.023 10,105 0.421 (0.051) 2. PIP and Medpay claims rather than bodily injury (BI) claims -0.021 8,104 0.316 (0.046) . DD Estimate . |$N$| . |$R^{2}$| . 1. Compare 2007 to 2002 -0.023 10,105 0.421 (0.051) 2. PIP and Medpay claims rather than bodily injury (BI) claims -0.021 8,104 0.316 (0.046) Notes: This table reports coefficient estimates from regressions designed to serve as placebo test of the baseline specification. In the first specification, the years of data are 2002 and 2007, and 2007 is used as the implementation year, so claimants ages 65 and older in 2007 are the treated group. In the second specification, the sample is comprised of PIP and Medpay claims from the 2012 and 2007 IRC database rather than BI claims. Except as noted below, the specification is as in column V of Table 2; see notes for Table 2. The second specification omits driver degree of fault as a control as that does not apply to PIP and Medpay claims. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab As a second falsification exercise, I re-estimate the main specification on 2007 and 2012 data, but construct the sample using personal injury protection (PIP) and Medpay claims rather than bodily injury claims. PIP and Medpay are first-party coverages that pay for medical care regardless of the fault of the driver; these components of auto insurance are primary to health insurance, so providers will typical bill them before billing health insurers.23 Because PIP and MedPay are available from the moment of an accident and payments do not involve an adversarial settlement negotiation, the same forces that would generate delay in the bodily injury context under enhanced Medicare reporting and recovery would be expected to operate to a much lesser extent, if at all, for these coverages. Thus, if my preferred specification is appropriately capturing the effects of reporting and recovery, in the replication using PIP and Medpay claims, we should observe no effects. The second column of Table 6 reveals that this is indeed the case. The point estimate is close to zero and sufficiently precise so as the exclude impacts of the magnitude observed on the BI sample. The preferred specification thus finds measurable effects for contexts where reporting might plausibly affect case resolution times, while finding no relationship where none should be present. Are certain types of claims more extensively impacted by new reporting requirements? Figure 1 plots coefficients from a series of quantile DD regressions that assess the effects of the new reporting requirements on different points within the claim time distribution. Because these regressions include a range of covariates capturing the nature and severity of the claim, these estimates are best conceptualized as indicating the effects of reporting and recovery on claims that are shorter or longer than is typical given their underlying complexity, rather than as estimates of the effects of reporting on longer or shorter claims per se.24 While there are measurable increases in claim time across most of the distribution, claims in the lowest deciles, that are therefore shorter than expected, appear to be more affected than claims in the highest deciles. This would be consistent with an environment in which reporting engenders a fixed time cost of involving Medicare in negotiations—for simpler than expected claims, this fixed cost represents a higher fraction of the overall time to resolution than for longer than expected claims. This fixed cost interpretation also seems consistent with the results above for malpractice claims, which tend to take much longer to resolve on average than auto claims, but for which I obtained effects estimates of roughly similar magnitude in absolute terms. If involving Medicare in the settlement does entail a fixed cost, policies such as the federal exemption for low-value claims enacted in 2012 seem warranted, as such a fixed cost would be more burdensome for smaller claims. Figure 1 Open in new tabDownload slide Quantile Estimates of the Effect of Mandatory Reporting. Notes: This figure reports coefficient estimates from quantile DD regression estimates of the effect of mandatory reporting on log(claim time), with controls as in specification 4 of Table 2. The quantile regressions estimate impacts of the reporting requirement for claims that are longer or shorter than expected based upon the observable covariates; claims in the lower quantiles are simpler, shorter claims and those in the upper quantiles are longer, more complex claims. See notes for Table 2. Dotted lines represent 95% confidence bands. Figure 1 Open in new tabDownload slide Quantile Estimates of the Effect of Mandatory Reporting. Notes: This figure reports coefficient estimates from quantile DD regression estimates of the effect of mandatory reporting on log(claim time), with controls as in specification 4 of Table 2. The quantile regressions estimate impacts of the reporting requirement for claims that are longer or shorter than expected based upon the observable covariates; claims in the lower quantiles are simpler, shorter claims and those in the upper quantiles are longer, more complex claims. See notes for Table 2. Dotted lines represent 95% confidence bands. Another way to examine the fixed cost hypothesis is to ask whether most claimants appear equally affected by the new requirements, or if certain subgroups are more or less affected. To assess this, in Table 7, I report results from regressions that include interaction terms between the new requirement and group membership, focusing on groups for which there are plausible reasons to imagine that there could be differences in the settlement negotiation process. I first examine whether the impacts of reporting vary according to whether the claimed injury was a sprain or strain; because such injuries are harder to objectively verify, they are often considered a potential marker for claim buildup or fraud (Derrig et al. 1994), and are therefore settled differently than other claims (Crocker and Tennyson 2002; Loughran 2005). For both strain and nonstrain injuries we see evidence of a delay effect of similar magnitude. Table 7. Estimated Effects by Subgroup . DD . |$P$|-value H |$_{0}$|⁠: . . estimate . Groups equal . 1. Type of injury  Sprain/strain ( |$N=$| 6,156) 0.169*** 0.943 (0.060)  Other ( |$N =$| 3,199) 0.174** (0.069) 2. Claimed loss amount  Below median ( |$N =$| 4,209) 0.200*** 0.290 (0.067)  Above median ( |$N=$| 4,258) 0.132** (0.052) 3. Legal regime  No-fault ( |$N =$| 1,607) 0.164* 0.870 (0.087)  Tort ( |$N =$| 7,812) 0.178*** (0.057) 4. Claimant shared fault for crash  Yes ( |$N =$| 729) 0.316*** 0.165 (0.111)  No ( |$N =$| 8,689) 0.164*** (0.054) 5. Reported economic losses other than medical  Yes ( |$N =$| 2,006) 0.285*** 0.087 (0.085)  No ( |$N =$| 6,294) 0.143*** (0.051) . DD . |$P$|-value H |$_{0}$|⁠: . . estimate . Groups equal . 1. Type of injury  Sprain/strain ( |$N=$| 6,156) 0.169*** 0.943 (0.060)  Other ( |$N =$| 3,199) 0.174** (0.069) 2. Claimed loss amount  Below median ( |$N =$| 4,209) 0.200*** 0.290 (0.067)  Above median ( |$N=$| 4,258) 0.132** (0.052) 3. Legal regime  No-fault ( |$N =$| 1,607) 0.164* 0.870 (0.087)  Tort ( |$N =$| 7,812) 0.178*** (0.057) 4. Claimant shared fault for crash  Yes ( |$N =$| 729) 0.316*** 0.165 (0.111)  No ( |$N =$| 8,689) 0.164*** (0.054) 5. Reported economic losses other than medical  Yes ( |$N =$| 2,006) 0.285*** 0.087 (0.085)  No ( |$N =$| 6,294) 0.143*** (0.051) Notes: This table reports coefficient estimates from alternatives to the baseline specification that include interactions between required MSP reporting and membership in the listed group. Each numbered specification presents results from a unique regression. The specification is as in column V of Table 2; see notes for Table 2. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Table 7. Estimated Effects by Subgroup . DD . |$P$|-value H |$_{0}$|⁠: . . estimate . Groups equal . 1. Type of injury  Sprain/strain ( |$N=$| 6,156) 0.169*** 0.943 (0.060)  Other ( |$N =$| 3,199) 0.174** (0.069) 2. Claimed loss amount  Below median ( |$N =$| 4,209) 0.200*** 0.290 (0.067)  Above median ( |$N=$| 4,258) 0.132** (0.052) 3. Legal regime  No-fault ( |$N =$| 1,607) 0.164* 0.870 (0.087)  Tort ( |$N =$| 7,812) 0.178*** (0.057) 4. Claimant shared fault for crash  Yes ( |$N =$| 729) 0.316*** 0.165 (0.111)  No ( |$N =$| 8,689) 0.164*** (0.054) 5. Reported economic losses other than medical  Yes ( |$N =$| 2,006) 0.285*** 0.087 (0.085)  No ( |$N =$| 6,294) 0.143*** (0.051) . DD . |$P$|-value H |$_{0}$|⁠: . . estimate . Groups equal . 1. Type of injury  Sprain/strain ( |$N=$| 6,156) 0.169*** 0.943 (0.060)  Other ( |$N =$| 3,199) 0.174** (0.069) 2. Claimed loss amount  Below median ( |$N =$| 4,209) 0.200*** 0.290 (0.067)  Above median ( |$N=$| 4,258) 0.132** (0.052) 3. Legal regime  No-fault ( |$N =$| 1,607) 0.164* 0.870 (0.087)  Tort ( |$N =$| 7,812) 0.178*** (0.057) 4. Claimant shared fault for crash  Yes ( |$N =$| 729) 0.316*** 0.165 (0.111)  No ( |$N =$| 8,689) 0.164*** (0.054) 5. Reported economic losses other than medical  Yes ( |$N =$| 2,006) 0.285*** 0.087 (0.085)  No ( |$N =$| 6,294) 0.143*** (0.051) Notes: This table reports coefficient estimates from alternatives to the baseline specification that include interactions between required MSP reporting and membership in the listed group. Each numbered specification presents results from a unique regression. The specification is as in column V of Table 2; see notes for Table 2. Robust standard errors are reported in parentheses. * indicates an estimate that is statistically significant at the 10% level, ** at the 5% level, and *** at the 1% level. Open in new tab Specification 2 examines whether delay varies for smaller versus larger claims, where the size of claim is measured based upon the reported loss in dollars. Again, we observe similar delay effects for both categories of claims, and fail to reject the null of no group differences. Similarly, allowing the effects of reporting to vary across no-fault states—which see different types of bodily injury claims due to statutory restrictions on the types of claims that can be brought within the tort system (Anderson et al. 2010)—and tort states yields little evidence of heterogeneous effects. Specification 4 separately considers claims in which the claimant was estimated to have some level of fault for the accident from claims where the fault lay solely with the defendant. These former claims may be more difficult to resolve due to disagreements over culpability. However, for both types of claims there is evidence of a delay effect of mandatory Medicare reporting. Finally, in Specification 5, I consider whether reporting differentially affects cases involving claimed economic losses other than medical care (e.g. lost wages) versus purely medical losses. When medical care is the sole source of losses and therefore highly influences the amount of general damages available to the plaintiff, one might expect plaintiffs and their attorneys to be particularly protective against Medicare encroaching on the settlement. Although the point estimates suggest, contrarily, that there may be somewhat higher delays for combined medical/nonmedical claims, the data do not clearly reject equal impacts across the two categories of claims. Overall, the patterns in Table 6 indicate that the unintended delay effects of reporting are experienced across a wide range of different types of claimants and across claims of varying complexity, with little concrete evidence of significant heterogeneity. 5. Conclusions The analysis indicates that the introduction of a new reporting requirement for Medicare slowed the resolution of auto injury claims by 19%, or about 2 months. Importantly, this is the average effect across all individuals subject to the new reporting requirement, including claimants for whom there ultimately was no recovery for Medicare. While reporting and subrogation may benefit Medicare fiscally by allowing it to recoup otherwise unreimbursed outlays for medical care that by law are assignable to other parties, it comes at the cost of delaying the resolution of injury settlements for Medicare beneficiaries. It is instructive to compare the estimates here to those of Helland and Klick (2018). Helland and Klick (2018) employ a similar DD research design, but apply it to a somewhat idiosyncratic sample of claims from a single large insurer. Their dataset includes accidents that occurred at varying points in time, some prior the reform and some after, but all of which triggered queries by the insurer to Medicare to determine if the claimant was a Medicare beneficiary. The average claim duration for claims involving preperiod accidents in the Helland and Klick (2018) sample is over 800 days, nearly double that of my sample, and nearly half of their claims had not been resolved as of 2014. This likely reflects a different composition of their sample—during the pre-reform period, when reporting was not yet required, it seems plausible to imagine that the insurer would have queried Medicare mostly in cases where claim adjusters anticipated the possibility that resolving the claim might take sufficiently long that the new reporting requirement would come into effect, leaving comparatively complex claims in their sampling frame. Although coming to similar qualitative conclusions, the new results here thus generalize the finding of Helland and Klick (2018) to a broader and arguably more representative set of claims. In documenting a similar delay effect in data covering closed medical malpractice claims, I also demonstrate that the impacts of Medicare reporting appear to extend to a broader set of personal injury cases. While on the surface a 19% increase in case resolution time may seem relatively benign, simple calculations suggest that such delay could be costly. One simple way to assess the magnitude of the time costs of delay is to monetize the delays using traditional time value of money calculations. While to my knowledge there is no data source that measures the value of auto claims payments directed specifically to Medicare beneficiaries, applying the patterns from the closed claim data to data on aggregate auto liability payments from the Insurance Information Institute (2017) suggests that there were roughly |${\$}$|6 billion in auto liability payments in 2014 made to claimants ages 65 and older. While different stakeholders may have different views regarding the appropriate rate of time preference, at a lower bound we might apply a risk-free market interest rate of 3% per year to this sum, in which case the delays documented here would account for roughly |${\$}$|30 million in aggregate losses due to delay. If one wished to adopt a more plaintiff-centric view, an alternative logical time preference benchmark would be to apply rates derived from real market transactions in the structured settlement market, a market in which individuals with claims on future payment streams (including tort awards) can transform these into lump sum payments. Hindert and Ulman (2005) document an average implied annual interest rate in structured settlement transactions of 20%, which would equate to |${\$}$|200 million in annual delay costs. To put these amounts in perspective, in its CMS Statistics publication, CMS (2009, 2015) reports that it obtained a total of |${\$}$|465 million in secondary payer recoveries from auto and liability insurers in FY2007, and this amount had risen to |${\$}$|770 million by FY2015. Assuming generously that this entire increase was attributable to the new reporting provisions, Medicare gained about |${\$}$|300 million per year in additional revenues from reporting. The most conservative estimate above of |${\$}$|30 million puts the waiting costs at roughly 10% of Medicare’s total recovery. Although many stakeholders might view a 10% as an acceptable cost to enable Medicare to collect funds to which it is legally entitled, it is important to note that these costs are borne by injured parties, insurers, and their attorneys rather than Medicare. Moreover, under the upper bound estimates discussed above, waiting costs represent a much larger share of Medicare’s total recovery. Additionally, these cost estimates are conservative for several reasons. First, the 19% estimate used above is itself likely an understatement of the true impacts of reporting for the reasons discussed in Section II. Second, delays from reporting apply not only to those aged 65 and over, but all Medicare recipients. Individuals under age 65 constitute about 16% of the overall Medicare population, and there are logical reasons to imagine that reconciliation and negotiations during the subrogation process for this group might even be more difficult than for those aged 65 and older, since these beneficiaries by definition have chronic health conditions that may be difficult to disentangle from the conditions brought about by the tort in question. Third, the above calculations apply only to private passenger autorelated injuries, but reporting requirements generally apply to the universe of bodily injury claims within the tort system—for example, commercial auto payments or bodily injury payments under homeowners or commercial liability policies. Because the ICD coding systems commonly used for capturing medical diagnoses for billing purposes include codes that make it comparatively easy to identify care associated with an auto injury than would be possible for these other types of claims, if anything it seems likely that effects of new reporting requirements would be more pronounced for nonauto tort claims. Finally, the calculations above assume that the only costs of delay are the time costs of money, when in actuality there are administrative costs associated with querying the system, reconciling bills, and continuing to monitor open claims. Thus, while delay costs of |${\$}$|30– |${\$}$|200 million are not insubstantial, these numbers may be appreciably below the true systemwide cost of the new reporting and subrogation requirements. While understanding delay and its costs are important, delay is but one among a complex set of policy considerations relevant to decisions regarding the most appropriate rules for subrogation. In addition to improving Medicare’s fiscal health, the reporting rule also promoted greater parity between the government and private health insurers such as Medigap plans that may already pursue subrogation. Changing subrogation practices also can have tax implications, as the tax treatment of benefits payments can vary depending on whether these costs are borne by Medicare, injured parties, or private liability insurers. Because the analysis here considers only one effect of more aggressive subrogation, it does not definitely demonstrate whether the move to mandatory reporting was beneficial or harmful overall. It does, however, supply an important ingredient required to assess the utility of this policy change. To what extent are the findings of this paper likely to translate to Medicaid reporting? At a conceptual level, a reporting process for Medicaid beneficiaries is likely to contain similar elements to the MSP reporting process studied here—in particular, an initial query process to determine whether a claimant is a Medicaid beneficiary followed by, for those who are, a demand notice in which Medicaid indicates treatment it has reimbursed that it believes to be subject to subrogation, followed by a reconciliation process with the liability insurer and claimant. Under both systems claimants, liability insurers, and the government will face similar negotiation and information-sharing incentives. Thus, the forces that produced delay in the Medicare context seem likely to exist in any states that elect to develop such reporting requirements for Medicaid. However, there are also some important differences between Medicaid and Medicare that could escalate costs in the Medicaid context. First, because Medicaid is operated at the state level, each individual state will have to develop a reporting system for its own claimants; states will likely differ in the resources and expertise available in developing such systems, and it seems plausible to think that some states may develop systems inferior to the current federal system. Multistate liability insurers would also potentially need to develop technology to interact with multiple different reporting systems in different states, which could prove costly. A second reason why delay cost might be higher in the Medicaid context relates to the nature of the Medicaid program itself. Medicaid is targeted towards low-income individuals and families, and Medicaid recipients have fewer assets than the typical claimant. The costs of delayed settlements may be particularly acute for low-income individuals with few resources; indeed, empirical evidence suggests that poor individuals have substantially higher rates of time preference (Lawrence 1991; Paserman 2008). Moreover, policies which delay payments to poor claimants who have recently suffered adverse health shocks may be viewed as particularly problematic by policymakers and the public. It remains to be seen whether states will follow the federal lead and more aggressively pursue reporting and subrogation for their Medicaid programs. This article identifies a drawback of such subrogation efforts—a delay in settling cases induced by the additional reporting and reconciliation requirements. The fact that such delays occur does not mean that increased subrogation is unwarranted—the fiscal pressures facing Medicaid may very well offer sufficient justification to policymakers and the public for new efforts to appropriately pursue reporting and subrogation. However, such options should be pursued only with a clear understanding that reporting carries with it an unintended consequence of delay. For those jurisdictions that do more aggressively pursue recovery from personal injury awards, authorities should seek to develop reporting systems that minimize the adverse effects of delay, by, for example, applying a safe harbor for low-value claims as currently exists in the federal system. More broadly, given the clear costs associated with reporting identified in this article, researchers would do well to develop additional empirical evidence on the potential recovery for Medicaid through additional subrogation, so as to better inform policymakers regarding how the fiscal benefits of additional recovery efforts might compare to the costs. Appendix: Analysis of Choice of Age Bandwidth The baseline specification limits the analysis to those ages 50 and over. In this Appendix, I explore the sensitivity of the results to using different samples by age. In determining the sample, we face a tradeoff—as we narrow the age window, the youngest and oldest individuals grow closer in age, so it becomes more plausible to think that the claim/injury experience of the youngest claimants provides a reasonable counterfactual for the claim/injury experience of the older claimants absent a reporting and subrogation requirement. However, as we narrow the age window, we also reduce the sample size—which may decrease the precision of the estimates—and potentially increase the influence of measurement error on our estimates. The measurement error problem arises because we do not actually observe the Medicare status of any particular claimant, we only observe their age, which we use as a proxy for Medicare status. This data limitation introduces several forms of measurement error. First, as discussed in Section 3 of the article, although the majority of individuals who gain Medicare do so upon turning 65, certain classes of individuals, such as disabled, low-income individuals, can be eligible for Medicare at earlier ages. Card et al. (2009) suggest that roughly 10% of adults ages 55–64 are Medicare eligible; this means that there are some younger individuals in the sample labeled as not being subject to reporting and subrogation who in fact are. Second, rather than knowing a claimant’s exact age, we only know their year of birth. This will lead us to label some individuals as treated in 2012 who actually would not have been subject to Medicare reporting and subrogation. As an example, consider a claimant from 2012 born in 1947 who therefore was 65 in 2012 when their claim was closed. For the purposes of the analysis, the individual will be coded as subject to the Medicare reporting and subrogation requirement. However, if their birthdate was actually in December 1947, their claim was closed in February 2012, and they follow the usual pattern of gaining Medicare eligibility at the time they turn 65, their claim would in actuality have been resolved when they were still age 64, and therefore not subject to Medicare reporting or subrogation. If birthdays and claim resolution dates are uniformly distributed across the calendar, we had expect 50% of the individuals who turned 65 in 2012 to in fact not have been subject to the reporting and subrogation requirement. This form of measurement error only affects individuals who turn 65 in the year their claim is closed. Finally, there are lags between the time that treatment is received and the time that claims are settled that may introduce measurement error. Consider, for example, an individual who is involved in a crash in early 2009 when they are 63 years old who receives medical care for their injuries in 2009 and 2010, but negotiations over the settlement amount, lost wages, etc. delay the resolution of the claim until 2012. This person will be 66 in 2012, and thus treated as subject to Medicare reporting and subrogation for the purposes of this analysis. However, because their medical treatment all occurred when they were 63 and 64—before they became eligible for Medicare—they in fact may not be subject to meaningful Medicare subrogation in actual practice. This form of measurement error seems likely to be most prevalent for individuals close to age 65 in 2012—because most claims are resolved within a few years of the initial accident, as individuals get older, it seems less plausible to imagine that their period of medical care for the auto injury would be completely nonoverlapping with the period during which they were Medicare beneficiaries. To summarize, there are reasons to expect that there may be measurement error with regard to actual reporting and subrogation status at and just above age 65, but as the age window expands, the fraction of the treated sample that was not in fact treated diminishes. Appendix Figure A.1 plots coefficients from a series of DD regressions that systematically vary the age bandwidth around 65. These estimates are analogous to those presented in column V of Table 2, but considering samples. Each point on the chart depicts a separate estimate, and the dotted lines depict 95% confidence intervals for each estimate based on clustered standard errors. The point estimate depicted where the horizontal axis equals 5, for example, corresponds to the DD estimate obtained when the age window includes 5 years before normal Medicare eligibility and 5 year after, or ages 60–69. Figure A.1 Open in new tabDownload slide DD Estimates Varying the Age Bandwidth. Figure A.1 Open in new tabDownload slide DD Estimates Varying the Age Bandwidth. Several patterns are apparent from the figure. First, the point estimates are fairly stable, similar to the baseline reported in Table 2, and economically meaningful for bandwidths above 2 years. Second, the estimates are attenuated and much less precise when the bandwidth is only 1 or 2 years, consistent with the discussion regarding measurement error above. Overall these results suggest that the paper’s main conclusions are not particularly sensitive to the choice of age bandwidth. Acknowledgement This work was supported by the RAND Institute for Civil Justice (ICJ) core funding program. Special thanks to Eric Helland, Allison Hoffman, and Jon Klick who provided comments on the research and David Corum who assisted with Insurance Research Council (IRC) data access. The editor J.J. Prescott and two anonymous referees provided very helpful comments and suggestions. All content is the responsibility of the author and may not reflect the views of RAND or the ICJ. Footnotes 1. See, for example, American Association for Justice (2011), Jordan (2017), and Medicare Advocacy Recovery Coalition (2017). 2. While it may seem at first glance as though insurers should benefit from holding money longer before transferring it to injured parties, such benefits are illusory in practice. Insurers are legally required to hold risk-free reserves against future claim payments, and are therefore constrained from deploying any capital tied up in the claims process towards more productive investments. A long claims process thus impedes them from maximizing returns on their capital. 3. As a benchmark, in my data the average claim resolution time is 422 days, and there is no censoring. 4. See, for example, § 116 of Public Law 97-248, the Tax Equity and Fiscal Responsibility Act of 1982 (Medicare) and § 1902(a)(25) of Public Law 74–271, The Social Security Act (Medicaid). 5. Such recoveries are often referred to colloquially as “Medicare liens,” although they are not liens in the traditional sense. 6. 42 U.S.C. § 1395y. 7. See, for example, Government Accountability Office (2012) and Kirchoff (2014). 8. For example, in response to empirical evidence demonstrating that there were likely to be some claims for which the costs of reporting exceeded Medicare’s recovery (Helland and Kipperman 2012), Congress and CMS ultimately established a de minimis threshold below which reporting was not required. 9. See Section 53102 of the Bipartisan Budget Act of 2018. 10. https://ri-mais.com/index.html. (accessed November 29, 2020) 11. See http://ncoil.org/wp-content/uploads/2016/04/MedicaidInterceptionModelamended.pdf. (accessed November 29, 2020) 12. In this article’s discussion I refer to claim resolution generically as “settlement,” as the vast majority of claims settle, but the data and my analysis also include claims that go to trial. 13. The tables report robust standard errors. Conceptually it seems most logical to think of each claim as representing a fresh realization of the outcome, particularly given that factors such as the nature of the accident and injury have by far the largest influence on claim complexity. Clustering on age/year—the level of treatment—leads to smaller standard errors, and other forms of clustering (e.g. by state) yield standard errors of similar magnitude to the robust standard errors. 14. This estimate arises directly from estimating equation 1, but one might also want to adjust the estimate to account for the fact that we do not directly observe Medicare eligibility in the data, and age is an imperfect proxy for eligibility. In particular, because roughly 10% of older adults below age 65 are Medicare eligible, and 6% of those over age 65 are not eligible, the actual increase in the fraction of claimants subject to the reconciliation process is below 100%. Adjusting for these patterns slightly increases the point estimate to 23%. 15. Table 1 reveals that many claims take a long time to resolve; if a claim is going to take several years to resolve in any case, an increase in resolution time of a couple of months on the margin may not have a big effect on incentives to file. 16. In particular, I add indicators for the deciles of this variable interacted with log payment amount to allow for a non-constant elasticity of claim time with respect to claim amount. 17. I express appreciation to an anonymous referee for suggesting this analysis. 18. Mullan et al. (1992) provide background on the origin and characteristics of the NPDB. Mello and Studdert (2016) discuss some of the NPDB’s limitations. 19. Examples of recent research using the NPDB include Studdert et al. (2016), Schaffer et al. (2017), and Gupta et al. (2018). 20. In addition to the measurement error arising from the lack of exact age in the data, there is also measurement error arising from the fact that some unknown group of claimants likely gained Medicare coverage while the claim was pending, both because the claim latency for malpractice claims is long enough that many claimants could age in while the claim was being resolved, and because adults under age 65 can gain Medicare eligibility if they become permanently disabled. 21. In an unreported analysis, I also estimated an event-study version of the DD regressions for the NPDB data, and found no evidence of a meaningful pre-implementation trend, and a pattern suggestive of an increase in case resolution time coinciding with the new reporting requirement. 22. There are several reasons to expect Medicare to have greater involvement in the settlement process for medmal claims than auto claims, even absent a reporting requirement. Because the source of injury in a medmal case is medical care, Medicare may be able to infer the existence of a malpractice claim involving one of its beneficiaries (and therefore seek recovery) based upon information available from its medical records. Moreover, when there will be ongoing medical payments after the claim is settled—a not uncommon occurrence in medmal cases—it has been longstanding practice for Medicare to be involved in negotiations to ensure its interests are satisfied through so-called Medicare Set-Asides. This means that the “dosage” of treatment associated with the new reporting requirement was likely lower for medical malpractice claims than for auto claims, which could also explain the diminished impact. 23. One reason for this is that PIP and Medpay policies generally do not receive the large discounts from medical provider list prices that are available to large health insurers, so providers get higher reimbursements from auto insurers than they would from traditional health insurers. 24. See Powell (2019) for a more detailed discussion of the interpretation of quantile regression coefficients in the presence of covariates. References American Association for Justice. 2011 . Medicare Secondary Payer: How Streamlining a Broken Bureaucracy Will Protect Seniors and Taxpayers . Washington DC . Andrews, Michele . “ Long Waits For Consumers When Medicare Is ‘Secondary Payer’, ” Kaiser Health News , Available at https://khn.org/news/012913-michelle-andrews-medicare-secondary-payer/ (accessed November 29, 2020). Anderson, James M. , Heaton Paul, and Carroll Stephen J.. 2010 . The US Experience with No-Fault Automobile Insurance: A Retrospective. Santa Monica, CA : RAND Corporation . Barnett, Jessica C. , and Vornovitsky Marina S.. 2016 . Health Insurance Coverage in the United States: 2015. US Census Bureau, Current Population Reports , Report P60-257 . Card, David , Dobkin Carlos, and Maestas Nicole. 2009 . “ Does Medicare Save Lives?, ” 124 Quarterly Journal of Economics 597 - 636 . Google Scholar OpenURL Placeholder Text WorldCat Centers for Medicare and Medicaid Services. 2009 . 2009 CMS Statistics . Available at https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/CMS-Statistics-Reference-Booklet/Downloads/CMS_Stats_2009.pdf (accessed November 29, 2020). Centers for Medicare and Medicaid Services. 2015 . 2015 CMS Statistics . Available at https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/CMS-Statistics-Reference-Booklet/Downloads/2015CMSStatistics.pdf (accessed November 29, 2020). Congressional Budget Office. 2016 . The 2016 Long-Term Budget Outlook . Available at https://www.cbo.gov/publication/51580 (accessed November 29, 2020). Crocker, Keith J. , and Tennyson Sharon. 2002 . “ Insurance Fraud and Optimal Claims Settlement Strategies, ” 45 Journal of Law and Economics 469 – 508 . Google Scholar OpenURL Placeholder Text WorldCat Derrig, Richard A. , Weisberg Herbert I., and Chen Xiu. 1994 . “ Behavioral Factors and Lotteries Under No-Fault with a Monetary Threshold: A Study of Massachusetts Automobile Claims, ” 61 Journal of Risk and Insurance 245 - 275 . Google Scholar Crossref Search ADS WorldCat Government Accountability Office. 2012 . Medicare Secondary Payer: Additional Steps are Needed to Improve Program Effectiveness for Non-Group Health Plans . GAO-12-333 , Washington DC . Gupta, Ashwin , Snyder Ashley, Kachalia Allen, Flanders Scott, Saint Sanjay, and Chopra Vineet. 2018 . “ Malpractice Claims Related to Diagnostic Errors in the Hospital, ” 27 BMJ Quality & Safety 53 – 60 . Google Scholar Crossref Search ADS WorldCat Hagy, Tom . 2010 . Waters & Kraus Attorney on Negotiating Medicare Liens. Available at: https://litigationconferences.com/waters-kraus-attorney-on-negotiating-medicare-liens/ (accessed November 29, 2020). Helland, Eric and Kipperman Fred. 2012 . “ Recovery Under the Medicare Secondary Payer Act Impact of Reporting Thresholds, ” RAND Occasional Paper OP-332-ICJ , Santa Monica, CA . Google Scholar OpenURL Placeholder Text WorldCat Helland, Eric and Klick Jonathan. 2018 . “ Medicare Secondary Payer and Settlement Delay, ” 15 Journal of Empirical Legal Studies 356 – 377 . Google Scholar Crossref Search ADS WorldCat Hindert, Daniel W. , and Ulman Craig H.. 2005 . “ Transfers of Structured Settlement Payment Rights: What Judges Should Know About Structured Settlement Protection Acts, ” 44 Judges Journal 19-31. Google Scholar OpenURL Placeholder Text WorldCat Insurance Information Institute. 2017 . “ Incurred Losses For Auto Insurance, 2011-2015, ” Available at: http://www.iii.org/fact-statistic/auto-insurance (accessed April 27, 2017). Jordan, Jennifer . 2017 . “ Medicare Secondary Payer Concerns in 2017: A Perfect Storm Could Be Brewing, ” LexisNexis Legal Newsroom , Available at: https://www.lexisnexis.com/legalnewsroom/workers-compensation/b/recent-cases-news-trends-developments/archive/2017/05/31/medicare-secondary-payer-concerns-in-2017-a-perfect-storm-could-be-brewing.aspx (accessed November 29, 2020). Kirchoff, Suzanne . 2014 . Medicare Secondary Payer: Coordination of Benefits . Congressional Research Service Report #7-5700 , Washington DC . Lawrence, Emily . 1991 . “ Poverty and the Rate of Time Preference: Evidence from Panel Data, ” 99 Journal of Political Economy 54 - 77 . Google Scholar Crossref Search ADS WorldCat Loughran, David . 2005 . “ Deterring Fraud: The Role of General Damage Awards in Automobile Insurance Settlements, ” 72 Journal of Risk and Insurance 551 - 575 . Google Scholar Crossref Search ADS WorldCat Matzus, Jason . 2011 . Testimony on “Protecting Medicare with Improvements to the Medicare Secondary Payer Regime” before the Subcommittee on Oversight and Investigations, Committee on Energy and Commerce United States House of Representatives. Available at https://www.govinfo.gov/content/pkg/CHRG-112hhrg74294/pdf/CHRG-112hhrg74294.pdf (accessed November 29, 2020). Medicare Advocacy Recovery Coalition. 2017 . “ SMART Act Implementation, ” Available at: http://www.marccoalition.com/other-smart-act-implementation.html (accessed November 29, 2020). Mello, Michelle M. , and Studdert David M.. 2016 . “ Building a National Surveillance System for Malpractice Claims, ” 51 Health Services Research 2642 – 2648 . Google Scholar Crossref Search ADS PubMed WorldCat Mullan, Fitzhugh , Politzer R. M., Lewis C. T., Bastacky S., Rodak J. Jr, and Harmon R. G.. 1992 . “ The National Practitioner Data Bank, ” 268 JAMA 73 – 79 . Google Scholar Crossref Search ADS PubMed WorldCat Paserman, M. Daniele . 2008 . “ Job Search and Hyperbolic Discounting: Structural Estimation and Policy Evaluation, ” 118 Economic Journal 1418 – 1452 . Google Scholar Crossref Search ADS WorldCat Powell, David . 2019 . “ Quantile Treatment Effects in the Presence of Covariates, ” Review of Economics and Statistics . 1 – 12 . Available at: https://www.mitpressjournals.org/doi/abs/10.1162/rest_a_00858. Schaffer, Adam C. , Jena Anupam B., Seabury Seth A., Singh Harnam, Chalasani Venkat, and Kachalia Allen. 2017 . “ Rates and Characteristics of Paid Malpractice Claims Among US physicians by Specialty, 1992–2014, ” 177 JAMA Internal Medicine 710 – 718 . Google Scholar Crossref Search ADS PubMed WorldCat Studdert, David M. , Bismark Marie M., Mello Michelle M., Singh Harnam, and Spittal Matthew J.. 2016 . “ Prevalence and Characteristics of Physicians Prone to Malpractice Claims, ” 374 New England Journal of Medicine 354 – 362 . Google Scholar Crossref Search ADS WorldCat © The Author 2021. Published by Oxford University Press on behalf of the American Law and Economics Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Journal

American Law and Economics ReviewOxford University Press

Published: Dec 1, 2020

References