Statement of
Peter R. Orszag
Director

Performance Budgeting:
Applications to Health Insurance
Programs and Tax Policy

before the
Committee on the Budget
U.S. House of Representatives

September 20, 2007

Chairman Spratt, Ranking Member Ryan, and Members of the Committee, thank you for inviting me to testify today on performance budgeting. It is difficult to argue with the proposition that a program’s budget should be linked to its effectiveness in achieving its objectives. But the concept of performance-based budgeting has been applied mostly to discretionary spending programs, which represent less than 40 percent of federal spending. That effort has taken many forms over the years, with the most recent version being the Program Assessment Rating Tool (PART).

Rather than assessing those past and ongoing efforts, my testimony extends the concept of performance budgeting to two other parts of the federal budget that have not been examined as deeply using that perspective: health insurance programs (and the health care costs that they cover) and tax expenditures. In both cases, the amount of federal funds at stake is substantial, and important questions exist about the cost-effectiveness of those expenditures.

• Health Care Costs. Rising health care costs represent the central fiscal challenge facing the country. Currently, net federal spending on Medicare and Medicaid totals about $560 billion; it accounts for about 20 percent of federal outlays and about 5 percent of gross domestic product (GDP). If spending per enrollee continued growing at the same average rate as it has grown over the past 40 years, federal costs for those two programs would amount to about 20 percent of GDP by 2050 (see Figure 1). Yet very little analysis is undertaken of whether that spending is generating corresponding gains in the health of enrollees, which presumably is the ultimate objective of those programs. To the extent that those programs are paying for services that make little or no contribution to health, opportunities exist to improve their efficiency.
  
  
• Tax Expenditures. Tax expenditures comprise various exceptions to the general tax rules—such as the income tax deduction that is allowed for interest paid on home mortgages—and total several hundred billion dollars per year. Those exceptions are often designed to encourage the tax-favored activities (such as home ownership). Tax expenditures may appear to reflect a reduced impact of federal policies on taxpayers, but they are effectively equivalent to collecting taxes at ordinary rates on the full potential tax base and then subsidizing the preferred behavior through outlays. Because selective tax reductions are like expenditures for specific economic activities, they can and should be evaluated in the same way as spending programs are. Applying performance budgeting to tax expenditures would therefore involve assessing whether they were achieving their purposes and whether they were doing so in a cost-effective manner.

Figure 1.

Source: Congressional Budget Office.

Note: The health cost growth differential refers to the number of percentage points by which the growth of annual health care spending per beneficiary is assumed to exceed the growth of nominal gross domestic product per capita, after an adjustment for the growth and aging of the Medicare and Medicaid populations.

In examining those two components of federal spending from a performancebudgeting perspective, four themes emerge:

• A substantial share of Medicare spending—and of health spending for the country as a whole—does not seem to be improving the health of enrollees. Average costs in Medicare vary significantly from one region of the country to another, but the additional spending and use of services that occur in the higher-cost regions do not appear to yield improvements in life expectancy or other available measures of health. Some researchers have concluded that about 30 percent of the money spent on health care services is not contributing to better health— an estimate that could probably be applied to the Medicaid program and to the whole health sector. Capturing those potential savings, however, is a substantial challenge.
  
  
• Many of the special preferences in the tax code do not appear to be costeffective in achieving their objectives. In some cases, tax expenditures appear to be in conflict with the purposes of other tax expenditures or explicit spending programs. In many more cases, preferences could be redesigned to reduce their costs substantially with minimal or no impact on the degree to which they were achieving their objectives.
  
  
• Both federal health insurance programs and tax expenditures could be subject to performance budgeting. The executive branch and the Congress generally engage in various forms of effectiveness reviews through their annual budgeting processes, but performance budgeting seeks to make that consideration more systematic and more consistent among programs. Assessing the effectiveness of any federal program in a more precise and systematic manner, however, is a complex undertaking, both because of disagreements about the goals of such programs and difficulties in measuring their impact accurately.
  
  
• Even with a systematic and well-designed assessment of performance, determining what specific policy steps to take as a result may be difficult. Unlike discretionary appropriations, the costs of federal health programs and tax expenditures are not set directly by policymakers but, instead, depend greatly on the actions of individuals and firms. The effects of any reforms aimed at improving performance thus depend on the responses to the policy changes. Despite those caveats, applying the concepts behind performance budgeting to health insurance programs and tax expenditures provides a useful framework for assessing those two major components of the federal budget.

Federal Health Insurance Programs

TheMedicare and Medicaid programs currently provide health insurance coverage to nearly 100 million people.⁽¹⁾ Both of those programs are relatively effective in achieving the goal of providing insurance to their eligible populations: elderly people, disabled individuals, and poor mothers and their children. Prior to the advent of Medicare, most elderly Americans lacked meaningful health insurance, and before Medicaid came into existence, low-income, disabled, and institutionalized people generally faced significant difficulty finding affordable health insurance on the private market.⁽²⁾ (Those programs also have the effect of transferring resources to enrollees who would otherwise have purchased their own insurance or paid for their own care.)

Whether the services that are financed by Medicare and Medicaid are always achieving the more fundamental goal of improving the health of enrollees is less clear, however. Indeed, concerns about the degree to which health care spending produces better health outcomes are not limited to public health care programs; they arise more generally in the health sector. Although many treatments improve enrollees’ health, and the benefits from health spending probably exceed the costs, on average, evidence suggests that much spending is not cost-effective—and in many cases does not even improve health.⁽³⁾

One reason is that relatively little rigorous evidence is available about which treatments work best for which patients or whether the benefits of more-expensive therapies warrant their additional costs. As a result, decisions about what treatments to use often depend on the experience and judgment of the individual physicians involved, as well as on anecdotal evidence and conjecture. In many cases, that basis may be sufficient, but in other cases, that approach is not adequate to select the most effective choice of treatment. Although estimates vary, some experts believe that less than half of all medical care is based on or supported by firm evidence about its effectiveness.

Geographic Variation in Health Care Spending

Partly as a result of limited evidence about their absolute or comparative effectiveness, the use of certain treatments and the types of care provided can vary widely from one area of the country to another. For example, even after adjusting for differences in age, sex, and race, researchers at Dartmouth College found about a fourfold variation in the share of Medicare beneficiaries receiving a coronary artery bypass graft; those differences were not correlated with rates of heart attacks in each region.⁽⁴⁾ Geographic differences in the types of care provided can remain substantial even among patients who turn out to be in their last six months of life. (That analytic approach has been used in an effort to control for differences in the prevalence and severity of illness, on the grounds that large groups of patients who are nearing death are likely to have comparable health problems regardless of where they live.) For example, patients in the final six months of their life spend nearly 20 days in the hospital, on average, in the highest-use areas of the country, compared with an average of about six hospital days in the lowest-use areas; the average number of visits to physicians in that period also varies by a factor of three.⁽⁵⁾

Those variations in the use of care correspond to substantial differences in Medicare spending per enrollee in different parts of the country, even after adjustments are made to account for differences in the age, sex, and race of enrollees in different areas (see Figure 2). In 2003, average costs ranged from about $4,500 per enrollee in the areas with the lowest spending to nearly $12,000 in areas with the highest spending. Some of those differences in spending reflect varying rates of illness as well as differences in the prices that Medicare pays for the same service, which are adjusted on the basis of local costs for labor and equipment in the health sector. But according to the Dartmouth researchers, differences in illness rates account for less than 30 percent of the variation in spending, and price differences can explain perhaps another 10 percent—meaning that more than 60 percent of the variation results from other factors.⁽⁶⁾ Other studies have found that a larger share of the variation in spending (as much as 70 percent) can be accounted for by differences in health status and demographic factors, but even so, the remaining differences are large in dollar terms.⁽⁷⁾

Figure 2.

Source: Dartmouth Atlas Project, The Dartmouth Atlas of Health Care.

Note: Numbers in parentheses refer to the number of hospital referral regions with per capita spending in each range.

Some evidence suggests that the degree of geographic variation in treatment patterns is greater when less of a consensus exists within the medical community about the best treatment to use—further buttressing concerns that services are being paid for that may contribute little to health. For patients who have fractured their hip, for example, the need for hospitalization is plain, and there is relatively little variation in admission rates for Medicare beneficiaries with that diagnosis (see Figure 3). For hip replacements and for knee replacements, more discretion is involved, and the surgery rates vary more widely. There appears to be even more variation in the rates of back surgery—a treatment whose benefits have been the subject of substantial questions. (Determining what share of any geographic variation in the use of procedures is attributable to differences in the treatments that doctors recommend and what share is attributable to differences in the prevalence of the underlying illness is challenging, however, so the comparison of procedures may be sensitive to the manner in which differences in illness rates are estimated.)

Figure 3.

Source: Dartmouth Atlas Project, The Dartmouth Atlas of Health Care.

Notes: In the figure, each point represents a hospital referral region; the country was divided into about 300 such regions on the basis of where Medicare enrollees typically receive their hospital care.

The points indicate how the rate at which the procedure is performed (per 1,000 Medicare enrollees) in each referral region compares with the national average rate (which has been normalized to 1.0). Differences in procedure rates were adjusted to account for differences among regions in the age, sex, and race of enrollees and for measures of illness rates.

The Relationship Between Spending and Health Outcomes

The implications of the observed variations in treatments and spending depend primarily on their relationship to health outcomes. If life expectancy and other measures were better in the areas with higher spending, that outcome might suggest that increased spending in the low-cost areas would yield health benefits. However, a study using Medicare data showed that the higher-spending regions did not have lower mortality rates than the lower-spending regions, even with adjustments to control for different illness rates among patients and regions.⁽⁸⁾ That study also found that higher spending did not slow the rate at which the elderly developed functional limitations. Furthermore, differences in spending are not correlated with simple measures of the quality of care that enrollees receive (see Figure 4). Although more research is needed about the impact of spending differences on morbidity and the quality of life (perhaps using more-extensive measures of health outcomes), the available evidence suggests that spending in the high-cost areas could be reduced without substantial adverse effects on the overall health of residents in those areas.

Figure 4.

Source: Congressional Budget Office based on data from Department of Health and Human Services, Agency for Healthcare Research and Quality, National Healthcare Quality Report, 2005 (December 2005), Data Tables Appendix, available at www.ahrq.gov/qual/nhqr05/index.html, and data from the Centers for Medicare and Medicaid Services’ Continuous Medicare History Sample.

Notes: The composite measure of the quality of care, based on Medicare beneficiaries in the fee-for-service program who were hospitalized in 2004, conveys the percentage who received recommended care for myocardial infarction, heart failure, or pneumonia.

Spending figures convey average amounts by state.

Much of the research that has been done about Medicare spending and its impact on health has focused on the traditional fee-for-service portion of the program, which serves the vast majority of its enrollees. But the concepts behind performance budgeting could also be applied to the Medicare Advantage program, under which enrollees can receive their benefits through a private health insurance plan. Currently, those plans are paid more than the costs for enrollees in the fee-forservice sector—about 12 percent more, on average—and most of the difference is returned to enrollees in the form of lower premiums, lower cost-sharing requirements, or additional benefits. Part of the rationale for the Medicare Advantage program is that enrollees will receive better care because it will be coordinated more effectively; by contrast, the fee-for-service program has no explicit mechanisms to coordinate care. Therefore, estimating the benefits of that coordination and comparing those benefits to their costs could be useful. Unfortunately, however, the data collected on Medicare Advantage plans are not sufficient to do so.⁽⁹⁾

An area that has received less attention is the variation in Medicaid spending and what that variation implies about the effectiveness of the services that enrollees receive. Medicaid spending at the state level varies substantially, even among enrollees who have qualified for the program for the same reason. For example, among elderly enrollees, spending averaged about $11,500 in 2004 but ranged from about $5,000 in the lowest-cost state to about $23,000 in the highest-cost state; for disabled enrollees, average costs varied to a similar degree.⁽¹⁰⁾ Those variations, it should be noted, reflect not just differences in prices or the underlying sickness of enrollees but also differences in eligibility rules and in the scope of benefits provided by states. Although it is reasonable to assume that Medicaid spending is subject to the same forces that cause spending variations in the health care sector as a whole, more research on the sources of variation in the program’s costs and its impact on enrollees’ health is warranted.

Addressing the Performance of Medicare and Medicaid

To a certain extent, both Medicare and Medicaid are already subject to evaluations of their performance, but the question is whether or how they might benefit from a more systematic review. Research about geographic differences in spending and other assessments of the programs’ contributions to health have been undertaken without formal requirements, and the executive branch and the Congress are aware of those analyses as they carry out their regulatory and legislative responsibilities. In addition, both Medicare and Medicaid have been assessed by the Administration’s PART initiative.⁽¹¹⁾ But those examinations have not been as rigorous as the level and growth of spending on the programs would seem to warrant. The PART reviews, for example, focus much of their attention on how efficiently the administrative functions of each program are carried out, and both reviews state that the question of tying budget requests to performance goals—which is usually a standard component of such assessments—is not applicable because Medicare and Medicaid are mandatory programs and their "budgetary resources are not driven by performance goals."

One reason that tying spending in those programs to performance is difficult is that policymakers generally have only indirect control over how much is spent. Those programs entitle enrollees to receive coverage for specified medical services, but decisions about which services are medically necessary are largely left to doctors and their patients. Thus, although Medicare and Medicaid officials generally establish the prices that the programs will pay for each service, they have little control over the quantities that are used. Efforts to improve the programs’ performance would therefore have to change the behavior of doctors, other providers, and patients—probably using some combination of information about what services were most effective or most cost-effective and incentives for doctors and patients to use those services.⁽¹²⁾ An additional challenge within Medicaid is that decisions about what services to cover and how much to pay for them are made primarily by state program administrators, subject to limits and guidelines set in federal legislation and regulations.

Those challenges notwithstanding, some efforts have been made to encourage more effective use of resources in the federal health insurance programs—but those efforts to move from a fee-for-service system toward a "fee-for-value" system are at a very early stage. For example, Medicare recently initiated demonstration programs to test "pay-for-performance" approaches for hospitals and large groups of physicians. Initial assessments indicated that the additional payments to hospitals participating in the demonstration were not effective in bringing about improvements in the care they provided (perhaps because the payments were too small), while the demonstration program involving physicians’ group practices may not have been designed in a way to effectively identify its effects. This year, Medicare will also make modest bonuses available to doctors who elect to report information on certain types of care that they provide (although the payments will not depend on their actual performance). Many states recognize the growing fiscal burden posed by Medicaid costs, and several of them have expressed interest in research on treatments’ effectiveness. For example, more than a dozen state Medicaid programs are involved in a project (affiliated with the Oregon Health and Sciences University) assessing evidence about the relative safety and effectiveness of competing drugs in the same therapeutic class.

Federal Tax Expenditures

A second area of the federal budget that is often not examined from a performance-budgeting perspective is tax expenditures. The total size of tax expenditures is large, but because of the interactions among them, it is analytically problematic to sum estimates of the individual expenditures—each of which assumes that the tax code otherwise is unchanged. Nonetheless, some idea of their aggregate importance can be seen from estimates of the largest individual items. The largest single tax expenditure is for contributions to and earnings within employer-sponsored retirement plans, which are exempt from income taxation; according to the Joint Committee on Taxation’s estimate, that exemption cost $110 billion in 2007 (see Table 1). The next largest tax expenditure, costing $100 billion, is for employer-provided health insurance—an amount that reflects only the impact on income tax receipts. The costs of providing that insurance are also exempt from payroll taxes, generating an additional tax expenditure estimated to exceed $60 billion.⁽¹³⁾ Reduced tax rates on capital gains and dividends diminished tax receipts by $95 billion, and the deduction for mortgage interest lowered federal revenues by $76 billion.⁽¹⁴⁾

Table 1.

In each of those cases, the reduction in receipts gives the appearance of a reduced impact on the federal budget—indeed, most presentations of the budget omit any mention of tax expenditures. But tax preferences are effectively equivalent to collecting taxes at ordinary rates on the full potential tax base and then subsidizing the preferred behavior through outlays. The reductions are achieved either by excluding some kinds of income from the tax base, allowing deductions for certain activities in arriving at taxable income, applying lower-than-normal tax rates to some kinds of income, or providing credits against tax liability for engaging in certain activities. Because selective tax reductions operate as expenditures for specific economic activities, their effectiveness can and should be evaluated in the same way as is done for spending programs.

Performance Limitations of Current Tax Expenditures

Given the size of many tax expenditures, it is striking that they are subjected to little analysis of their effectiveness in achieving their objectives. Indeed, in many instances, the underlying goal of the tax preference has never been made explicit. Superficially, it might be assumed that part of the purpose of the preference is to provide tax relief. But the burden of taxes could be reduced by having lower tax rates, rather than providing special treatment to certain activities. Tax reductions for specific activities, as opposed to a general reduction in tax rates, also affect the incentives to engage in those activities; therefore, the underlying goal of each preference is presumably to affect outcomes such as home ownership, health insurance coverage, charitable giving, and so on.⁽¹⁵⁾

In many cases, however, the specific outcome that is desired is unclear or appears to conflict with the objectives of other policies. And in instances in which the goal is both apparent and not in conflict with other policy measures, many existing tax preferences seem to fail to achieve that goal or to be relatively inefficient in achieving it. The following examples seek to illustrate those categories of tax expenditures, although the distinctions between them can be blurry:

• Tax Expenditures with Objectives That May Conflict with the Objectives of Other Policies. Various business tax expenditures appear to be in conflict with other spending programs or tax incentives. Some production incentives in the energy sector, for instance, run counter to the intent of other expenditures intended to reduce demand for energy, so the fundamental policy objective for both sets of programs is difficult to discern.
  
  
• Tax Expenditures That May Fail to Achieve Their Objectives. In a number of instances, the policy goal is clear and not in conflict with other policies—but tax incentives do not appear to yield the desired effect. For example, questions exist about whether tax preferences for certain kinds of savings vehicles, such as individual retirement accounts, have significantly boosted private saving or merely subsidized saving that would have occurred anyway.⁽¹⁶⁾ People’s decisions about saving may not be affected much by tax incentives, or ceilings on contributions to such accounts may largely limit the impact of those incentives on total saving. Another example is the deductions for property taxes and mortgage interest. Although those tax expenditures are intended to help lessen the cost of home ownership, evidence suggests that much of their value is reflected in higher prices for homes rather than accruing to home buyers.
  
  
• Tax Expenditures That May Be Inefficient in Achieving Their Objectives. Even in instances in which tax expenditures are likely to have achieved policy goals, many do so at substantially higher costs than may be necessary. For example, exemptions and deductions are commonly employed for purposes that could often be more efficiently achieved with flat-rate (and potentially refundable) credits, as in the case of employer-provided health insurance. In addition, some tax preferences would simply function better if provided in the form of outlays, such as tax credits for low-income housing.

That final category of tax expenditures warrants further comment. As a result of the progressive structure of the individual income tax, preferences in the form of exemptions or deductions increase in value as income rises—because higher taxable income places households into higher marginal tax brackets. An income tax deduction of $1, for example, is worth 35 cents to a taxpayer in the 35 percent marginal tax bracket and 15 cents to a taxpayer in the 15 percent bracket. The consequence is that the subsidy rate provided through the tax code for that type of preference also tends to rise with income. Yet unless taxpayers’ reaction to the preference or the social benefit of that response varies systematically according to taxpayers’ income, providing differential subsidy rates is inefficient.⁽¹⁷⁾

Those efficiency considerations are related to, but distinct from, concerns about the distribution of various tax expenditures. An analysis by the Joint Committee of Taxation shows that the benefits of several common high-value tax expenditures accrue primarily to higher-income taxpayers (see Table 2). Part of the reason is that those taxpayers have greater tax liability in the first place. They also tend to engage in more of the subsidized behavior. The distributional pattern, though, results in part from the progressive nature of the tax code and, therefore, the higher subsidy rates provided through deductions and exclusions to higher-income taxpayers than to lower-income taxpayers (which in turn may influence the pattern of subsidized behavior).

Table 2.

Tax Credits and Economic Efficiency

Empirical evidence demonstrating that high-income households are particularly responsive to tax incentives—or that their responses generate disproportionate social benefits—is lacking in most cases, raising questions about the efficiency of providing differential subsidy rates through the tax code. In the absence of such evidence, the most economically efficient approach to subsidizing an activity through the tax code generally involves flat-rate credits available to all households.

As an example, consider the current exclusion for contributions to 401(k) plans (a vehicle for retirement savings offered through employers). The value of that exclusion rises with taxable income because of the link to marginal tax rates. By contrast, a 15 percent or 25 percent credit for 401(k) contributions would have the same marginal value for all taxpayers. If the credits were refundable, they would provide the same subsidy rate to all tax filers. Unless evidence exists that households in different marginal tax brackets are more likely to increase their saving in response to the tax incentive, or to generate larger benefits to society if they did increase their saving, such a flat tax incentive would generally be the most efficient approach to subsidizing retirement saving.

The exemption of income from state and local bonds provides another example. The exemption delivers different marginal benefits to different taxpayers depending on their income bracket. As a result, the subsidy that the exemption delivers to state and local governments is less than the value of revenues lost to the federal government. Less costly means of delivering the same subsidy could be to change the structure of the current tax expenditure from tax-exempt bonds to a variation on tax-credit bonds (discussed further, below) or to provide direct intergovernmental transfers.⁽¹⁸⁾

Perhaps the most prominent example of a tax expenditure that appears to be inefficient is the exclusion for employer-provided health insurance—which allows that form of compensation to avoid both income and payroll taxes. For a typical worker, those tax preferences amount to a subsidy from the government of more than 30 percent of the costs of health care services that are covered by employersponsored insurance. By reducing the price of that insurance, the tax subsidy encourages workers to purchase coverage through their employer and to secure more-extensive policies, increasing the share of costs that is covered and decreasing the share that is paid out of pocket. In turn, that more-extensive coverage puts upward pressure on total spending on health care. Moreover, the value of the exclusion is larger for households with higher income.

The potential effects of redesigning the current exclusion for employer-sponsored health insurance are illustrated by the President’s recent proposal to replace that exclusion (as well as most current tax deductions for health insurance premiums and out-of-pocket costs) with a new standard deduction of $15,000 for couples who file joint tax returns and $7,500 for single filers. Under that proposal, all taxpayers not enrolled in Medicare who purchased qualifying insurance would be eligible for the deduction, and its value would not vary with the costs of the policy that was purchased. According to the Congressional Budget Office’s (CBO’s) estimates, by 2010 the number of people without health insurance would be about 6.8 million lower under the proposal than under current law. In addition, the fixed deduction would cause people to buy plans with less extensive benefits, on average; as a result, benefits for insured workers would decline by roughly 15 percent in 2010. At the same time, the Joint Committee on Taxation has estimated, the proposal would increase federal revenues by more than $300 billion over 10 years.⁽¹⁹⁾

Compared with people who would be uninsured in 2010 under current law, those gaining insurance coverage under the President’s proposal would have higher income and be in relatively good health, on average. The reason is that the value of the new deduction would be greater at higher marginal tax rates, which are associated with higher income, whereas insurance premiums would be lower for people in better health; thus, the net cost of buying health insurance would be lowest for that group. Even so, the majority of newly insured people would come from lowermiddle- and middle-income households, mostly because the uninsured population as a group has relatively low income. At the same time, the estimate of the net reduction in the number of uninsured individuals masks a number of offsetting shifts in coverage; overall, about 8.3 million people would newly gain coverage, CBO estimates, but about 1.5 million people would lose employment-based coverage and would end up uninsured. Those individuals are somewhat more likely to have lower income, as the tax deduction would be moderately less valuable to lower-income households and thus would offset a smaller share of their policy premiums.

As with other tax expenditures, a key issue in assessing the current tax expenditure for health insurance and the options for replacing or redesigning it is whether the responses of households to subsidies or the benefits of insurance coverage vary in proportion to income. Although households with higher income are more likely to have insurance coverage, it is not clear that those households are more responsive to subsidies than are lower-income households; rather, that pattern may simply reflect the larger subsidies provided to higher-income households. Nor do the social benefits of health insurance coverage appear to increase with income. Because most people who currently lack insurance have relatively low income— below 200 percent of the federal poverty level—some observers have argued that reforms of the current exclusion would reduce the number of uninsured by a larger degree if structured as refundable credits.

Other Options for Restructuring Tax Expenditures

Other tax expenditures have not exploited the possibilities afforded by the use of floors and ceilings to increase their "bang for the buck" and limit the degree to which they subsidize activity that would have occurred anyway. For example, a floor of 1 percent or 2 percent of adjusted gross income for deductible charitable contributions would reduce the incentive to make additional contributions by only a negligible amount while substantially reducing the loss of revenues associated with the tax expenditure. By virtue of the fact that income calculations are fundamental to the tax system, that structure lends itself readily to floors and ceilings— whereas such an approach applied to regular expenditures would significantly complicate administration.

Some tax expenditures are inefficient by virtue of being part of the tax system— and thus might be more efficient if converted to regular outlays. The low-income housing credit and tax-credit bonds may provide examples. Because they are subject to ceilings, they do not administer themselves automatically, and a separate institutional mechanism must be set up to allocate them. Incorporating them into the tax code thus appears to generate little benefit in terms of efficiency or administrative simplicity. Moreover, the apparent purpose of the housing credits is to make housing more affordable for low-income people—but the credits go to developers, who are more likely to have tax liability. Similarly, tax-credit bonds are intended to aid local governments, but the credits go to bondholders because local governments do not pay federal taxes. Because the tax benefits have only an indirect effect on the activity that policymakers are trying to encourage, those credits are probably a more costly way to provide support than simply spending the money as outlays.

To be sure, some expenditures undertaken through the tax system may be more efficient than ones delivered directly. In particular, the tax return can be a relatively efficient mechanism for administering programs whose eligibility depends on income because income must be computed for tax purposes. For example, the earned income tax credit is an income assistance program that could be structured as outlays, but building it into the tax code probably enhances its effectiveness significantly while reducing its administrative cost.⁽²⁰⁾ But as the foregoing examples suggest, many tax expenditures do not appear to have been evaluated sufficiently for cost-effectiveness. Consequently, the considerations of performance budgeting could be usefully applied to assessing them, both in terms of the connection between expenditures and outcomes and in terms of the efficiency of tax expenditures and outlays.

Conclusion

To sum up, the growth of health care spending is the central long-term fiscal challenge facing the nation, and existing tax expenditures entail a substantial reduction in the nation’s tax base. In both cases, federal policy might be improved by applying principles of performance budgeting—gauging whether the benefits derived warrant the resources provided. Typically, performance budgeting has been applied to discretionary programs, where it has had a checkered history at best.⁽²¹⁾ But lessons drawn from those efforts might be applied to a more systematic assessment of mandatory spending and tax expenditures.