Health Savings Accounts and Tax Subsidies
How Effective Can They Be?
About This Event

Will Health Savings Accounts (HSAs), made possible by the Medicare Modernization Act of 2003, bring about a revolution in the health insurance market? Will they lower the cost of health insurance and help to reduce the number of people without coverage?

While little research has been done until now about the probable effects of HSAs, a new study, “Health Savings Accounts: Early Estimates of National Takeup,” by University of Minnesota scholars Stephen Parente, Roger Feldman, Jean Abraham, Jon Christianson, and Ruth Taylor offers us a first look at the effects of health savings accounts and tax subsidies on health coverage and costs. Two of the authors will present their findings at this AEI event and two experts with extensive knowledge of the health insurance market will analyze the methodology and policy implications of the research.

This discussion is sponsored jointly by Health Affairs and the American Enterprise Institute. The study was recently published in the November/December 2005 issue of Health Affairs.


1:30 p.m.




Par Atwal, Health Affairs


Stephen T. Parente, University of Minnesota

Roger Feldman, University of Minnesota


Melinda Beeuwkes Buntin, The RAND Corporation

Jack Rodgers, PricewaterhouseCoopers


Robert B. Helms, AEI


Wine and cheese reception

Event Summary

November 2005

Health Savings Accounts and Tax Subsidies: How Effective Can They Be?

Will health savings accounts (HSAs), made possible by the Medicare Modernization Act of 2003, bring about a revolution in the health insurance market? Will they lower the cost of health insurance and help to reduce the number of people without coverage? While little research has been done until now about the probable effects of HSAs, a new study, “Health Savings Accounts: Early Estimates of National Takeup,” by University of Minnesota scholars Stephen Parente, Roger Feldman, Jean Abraham, Jon Christianson, and Ruth Taylor offers a first look at the effects of health savings accounts and tax subsidies on health coverage and costs. At a November 18 panel discussion, two of the authors presented their findings, and two experts with extensive knowledge of the health insurance market analyzed the methodology and policy implications of the research.

Stephen T. Parente
Roger Feldman
University of Minnesota

In 2003, Congress passed the Medicare Prescription Drug, Improvement, and Modernization Act (MMA), and incited a surge of interest in consumer-directed health plans (CDHPs). But will high-deductible insurance plans, coupled with tax-advantaged health savings accounts (HSAs), have enough national appeal to significantly increase the number of insured? Would a tax credit, as proposed by the Bush administration, or other possible subsidies encourage additional HSA takeup? Furthermore, will these proposals be efficient in terms of reducing both health costs and the number of the uninsured?

This paper addressed these questions by analyzing data from the Medical Expenditure Panel Survey, individual HSA data from, and health plan choice data from three large employers that offer consumer-directed plans. These data were used to predict national HSA takeup under four simulated policy proposals, all of which were compared to the 2005 baseline situation under MMA alone. The first simulation incorporated the administration’s proposal for a refundable tax credit of up to 90 percent of the insurance premium. Other subsidies for part or all of the premium, which could more aggressively tackle the uninsured problem, were also simulated. As a follow on to this proposal, the full price of the high-deductible insurance with an HSA was set to zero for the prototype HSA plan and for a more generous plan containing a smaller gap in coverage between the HSA amount and the deductible (doughnut hole). The final simulation provided only the unemployed with a full subsidy regardless of income.

Under the right incentives, widespread adoption of HSA plans is possible. MMA alone, without tax-credit inducements or additional changes in policy, led to a takeup of 9 percent in the individual market. At an annual cost of $8.1 billion, the administration’s tax-credit would double HSA takeup and consequently reduce the number of uninsured by 2.9 million. Offering a low-income subsidy would require $12.2 billion in annual outlays and would result in a 16.5 percent reduction in the uninsured.

Consumers are remarkably sensitive to the size of the doughnut hole. The most significant increase in takeup, and likewise cost, occurred when a fully subsidized HSA had a smaller doughnut hole. This policy in turn eroded the employer coverage market, with 5.7 and 31.6 million employees opting for standard and more generous HSA plans, respectively. A full subsidy for the unemployed population was less cost-effective than the administration’s proposal and the low-income premium subsidies. This study provides a way to model the effects of various HSA designs and may facilitate the creation of a plan which cost effectively expands insurance coverage. 

Jack Rodgers

This paper has several interesting features. First of all, it simulates the impact of a range of policy choices. This range is important especially when it comes to the employer-offered population, which frequently both shifts among plans and enters or exits the insurance market. Moreover, this model is one of the first to incorporate recent information on actual choices from the developing CDHP market.

The individual market is a very small part of the broader insurance system. It is frequently misunderstood and poorly estimated in many studies. The market does not work very well because there are different offerings and different rules across the country. 

This model avoids many of the problems by using the small firm market as a proxy for the individual market. Although it offers a cleaner scenario, the model may not be realistic. Notable are the differences between the people in small firms and those that are unemployed. These assumptions and differences can lead to a high amount of imprecision.

There are also many problems associated with simulation models, especially in terms of the data not included in the main data set. This study is rich in information from a number of databases and surveys; but much of it is brought in as averages by age group, geographic area, and type of health insurance plan. That averaging results in a loss of much variation in the imputed variables. It is difficult to assess the impact of the averages in the model, but it is clear that it adds another level of imprecision.

It would be interesting to have seen the authors delve into broader issues raised by the implementation of HSAs. For example, do HSAs offer an easier method to provide a modest government subsidy than in a standard low-deductible policy? An interesting model would compare take-up rates when people are offered a choice between a low-deductible, high cost-sharing plan and a high-deductible plus HSA plan. Furthermore, are HSAs appropriate for low-income populations? Typical health insurance programs subsidized for low-income people eliminate the cost-sharing feature; however, one obviously cannot eliminate cost-sharing when dealing with an HSA. Lastly, are HSAs inherently more cost-effective in terms of public subsidies than other types of health plans? To answer these questions, it is necessary to have simulations of low-deductible policies for the sake of comparison.

There are three potential ways to improve the data reporting in the paper. One is to include the motivation behind people’s shifting between plans--in other words, why some types of plans lost more people to HSAs than others. Secondly, reporting a change per unit or change per percentage in the regression tables would have served a much better purpose than merely providing the coefficients or directions of change. Finally, a breakdown of the results into income or demographic characteristics would have further facilitated the analysis of the data.

Melinda Beeuwkes Buntin
RAND Corporation

There are several reasons why HSA takeup is so important. First, there is much hope right now that high-deductible plans plus HSAs will be an efficient form of insurance, offering incentives for consumers to use care wisely. More importantly are the new information tools that will allow consumers to purchase health care efficiently. Secondly, there are 46 million uninsured Americans. Even if altruistic motives are unimportant, the cost of insuring these people is partly borne by taxpayers. Lastly, HSAs and tax subsidies for the individual insurance market are two of the very few options for insurance expansion possible in the current political environment.

The study offers valuable insight into subsidizing individual insurance plans. The authors found that although people are most sensitive to premium level, they are also very sensitive to the deductible level, especially those people who are sick. Another interesting point is that subsidies aimed at the individual market can affect the group market to a surprising extent. Not only will people leave the group insurance market in light of the subsidy (commonly termed crowd-out), but also people who would have purchased individual insurance without the subsidy will now be subsidized. Both of these unintended effects raise the cost per newly insured patient. Nevertheless, the study demonstrates that subsidizing low-income people provides the lowest cost per newly insured regardless of the fact that there are diminishing returns to these subsidies.

The limitations of the paper are also important to note. The first limitation is the fact that the paper ignores that health status is an important determinant of HSA takeup. A study by Emmett B. Keeler et al. from 1996 found that healthy individuals and those exceeding their deductibles are more likely to be attracted to account-linked high-deductible plans. Health status will also affect one’s ability to obtain insurance and the premium offered. These factors are important because not only do we care about more people getting insurance, but we also wish to make sure that those are the right people. For example, even if one is offered a subsidized policy, one’s risk-adjusted premium may be too expensive to take it up. The second limitation is that the model is based on behavior at firms that are early adopters of health reimbursement arrangements (HRAs). Because HSAs could be more attractive, the takeup estimates could be biased downwards. However, this select set of firms is already committed to the consumer-directed concept; therefore, these firms probably offered extensive education and information to their employees. This fact would bias the takeup estimates upwards. Lastly, as noted earlier, the workers involved in the model differ not only from the general population but also from the uninsured.

If the study’s results are taken as our best estimate to date, the HSA subsidy plan offers a somewhat moderate cost per newly insured patient of about $2,700. Other studies found somewhat higher costs at $5,495 (Joseph Antos) and $4,956 (David Lewin). Jonathan Gruber’s study on Medicaid expansion found the cost to be around $1,750, although there would be a limited number of providers due to the nature of the plan (SCHIP and Medicaid). Therefore, if the purpose is to increase access to care at the lowest cost, perhaps subsidized HSAs do not accomplish that goal.

HSAs are predicated on the belief in the increased reliance on the individual market. If this reliance fractures the employer market and healthier people are pulled out of that market, it could have adverse effects on the rest of society’s ability to obtain insurance. In other words, there is some value to employer pooling and the choice of plans.

A further consideration of the study is how high-deductible plans and HSAs affect health outcomes. There is limited evidence about the effects of consumer-driven health care on cost, use, quality and appropriateness of care. The results from the RAND Health Insurance Experiment suggest both appropriate and inappropriate care will be reduced when people face higher cost-sharing, even though preventative care is covered 100 percent. Early work by these authors and others finds mixed effects of consumer-driven health care on the use of different types of care. It has been shown that people are pretty savvy about substituting generics for prescription drugs, but also there is evidence of increased inpatient admissions rates for people with serious illnesses.

There is no evidence about the effects of consumer-driven health care on quality in general. There is much excitement that increased consumer engagement will spur providers to improve quality. Nevertheless, there is fear that increased financial exposure could lead to foregone care, fragmentation of care, and adverse selection issues. These fears are especially strong for the sick and poor who have not been enrolled in such plans, so there is little evidence of its effects.

There are some key areas of future research to be addressed before we can conclude that HSAs and high-deductible plans plus HSAs are where we want to invest our money in subsidization. There are few studies that compare a range of consumer-directed designs to traditional HMOs and PPOs in terms of costs, utilization, quality, and appropriateness of care. Measuring costs and utilization initially and over time is important because although we have data on early experience, we have seen that over time, people’s expenses can grow to that of traditional plans and PPOs.

Also, it would be useful to have studies that investigate the effects of consumer-directed designs on the poor and the sick. This task is difficult because most of these plans are offered mainly by employers whose employees are a significantly different group of people than the poor and the sick.

Finally, an invaluable asset would be investigations of alternative consumer-directed designs, including the following: plans in which premiums/cost-sharing/account contributions vary with income; HSAs paired with care management or incentives for primary and secondary prevention; plans that provide comparative information about provider costs and quality to educate consumers; and HRAs versus HSAs, especially the way that consumers behave with different account balances.

AEI research assistants Elizabeth DuPre and Jonathan Stricks prepared this summary.

View complete summary.
Also Visit
AEIdeas Blog The American Magazine

What's new on AEI

Poverty in America—and What to Do About It
image GDP for second quarter: Strong headline, weak innards
image Paul Ryan and the emerging conservative reform agenda in higher education
image Democrats' impeachment fixation
AEI Participants


Robert B.
AEI on Facebook
Don't Miss...