Snapshots: Illustrating the Potential Impacts of Adverse Selection on Health Insurance Costs in Consumer Choice Models

A current strategy for addressing the cost of health insurance involves consumer-directed health plans (CDHPs). These plans generally are less expensive than more traditional health plan designs, but it is not clear whether the lower costs derive entirely from the new benefit structure or whether some of the savings result because these new arrangements attract a disproportionate share of enrollees in good health who have relatively low health spending. Public discussions of enrollment bias tend to be over-broad, with critics suggesting that CDHPs primarily attract the young and healthy while supporters suggest that CDHP enrollees look like other people with insurance. We illustrate that even a small shift in enrollment among enrollees who end up with high health expenses can have a meaningful impact on plan costs. Additional research will be needed to determine whether there is a health difference between people who enroll in new consumer directed health plans and those who choose more comprehensive plans and, if so, how that difference affects plan costs.

The introduction of consumer-directed health plans (CDHPs) has generated considerable debate about their potential impact on the cost and availability of health insurance. These arrangements include health plans with high deductibles that require significant up-front out-of-pocket expenses that consumers must pay for, either directly or through a tax-preferred savings device such as a health savings account (HSA) or health reimbursement arrangement (HRA).1

Whenever consumers have a choice among different insurance arrangements, there is a tendency for people to choose a level of insurance based on their expected need for what is being covered (in this case health care). At any given price, people with a relatively high perceived need for the covered item are more likely to want coverage (and given coverage, more likely to want generous coverage) than people with a lower perceived need for the covered item. This is often referred to as “adverse selection.” Since high deductible plans generally are less generous than typical insurance policies –- particularly for people with employer-sponsored coverage –- some have raised concerns that CDHPs could disrupt the pooling of health risk in insurance markets.

CDHPs offer consumers a health plan with a lower premium but higher up-front cost sharing as compared with more traditional insurance. This tradeoff is likely be viewed differently by people with different perceived health care needs –- people with lower perceived needs may be more likely to focus on the certain premium savings relative to the potential higher out-of-pocket expenses, while people with higher perceived needs may be more likely to focus on the potential higher out-of-pocket costs relative to the premium savings. As a result, CDHPs could end up with a disproportionate number of people in relatively good health, while more comprehensive benefit plans end up with a disproportionate number of people in relatively poor health. This difference in health status among enrollees will in turn affect the claims costs for CDHPs as compared with other products, and also affect premiums unless they are adjusted to reflect the risk differences between enrollees in the different products.

Public discussions of adverse selection sometimes over-generalize the issue. For example, critics of CDHPs may suggest that they are cheaper because their enrollees are primarily young and healthy, while supporters may maintain that there is no selection because CDHP enrollees have the same characteristics as the population overall. In fact, as discussed below, relatively small differences in the enrollment of higher cost people in different plan options can have a meaningful impact on plan costs. The reason for this is that health spending is highly skewed, with a small share of people in any year accounting for a very large share of overall spending. Figure 1 shows the distribution of health spending across the U.S. population. As the figure shows, the one percent of the population with the highest health spending account for almost 24% of all expenditures; the 5% of the population with the highest health spending account for almost one-half of health expenditures; in contrast, the 50% of the population with the lowest spending account for less than 4% of total expenditures.

If even a relatively small portion of these higher spenders is able to anticipate that they are at higher risk and choose more comprehensive plans instead of less comprehensive plan options, the average claims costs of comprehensive plans would increase and the average claims costs of less comprehensive plans would decrease due to selection. The next section illustrates the potential magnitude of this effect.

Figure 1: Concentration of Health Expenditures, 2003

Concentration of Health Expenditures, 2003

Notes: Health spending is defined as total payments, or the sum of spending by all payer sources.
Source: Kaiser Family Foundation calculations using data from U.S. Department of Health and Human Services, Agency for Healthcare Research and Quality, Medical Expenditure Panel Survey (MEPS), 2003.

Examining Medical Claims

To demonstrate that a relatively small share of enrollees can have a meaningful impact on average claims costs, we present an exercise using a database of medical claims complied by the Society of Actuaries.2 The database summarizes paid health insurance claims from seven insurers for 1997, 1998, and 1999. We used the 1999 data, which includes almost 1.6 million claimants representing almost $2.6 billion in paid claims. The average paid claim amount across all categories of spending was $1,633. For the exercise, we used 51 categories of claimants based on their level of paid claims (e.g., spending between $1,000 and $2,000; spending between $50,000 and $55,000) and computed the average number of claimants in each category and the average spending in each category.

As mentioned above, when looking at health care spending patterns, a small percentage of claimants account for a large share of the spending. In the database that we use here, the 11% of claimants with the highest paid claims accounted for 60% of the paid claims.We refer to them as top spenders.

To look at the impact of adverse selection by even a small number of the top spenders, we split the database into two identical pools, each with same number of claimants (795,869) and the same average spending for each category of paid claims. The average paid claim amount in each pool is $1,633 (the average in the database). Splitting the database in half represents what would occur if people distributed themselves randomly into two different insurance arrangements, with no tendency towards adverse selection based on perceived health needs. To then look at the potential impact of selection, we varied the percentage of the top spenders that enroll in each pool, leaving the remaining enrollment in each pool unchanged. We looked at five scenarios: (1) the initial 50%/50% split of high spenders between the pools; (2) a 52%/48% split; (3) a 54%/46% split; (4) a 56%/44% split; and (5) a 60%/40% split. What this means is that, for example, in the 52%/48% scenario, one pool enrolls 52% of the top spenders and the other pool enrolls 48% of the top spenders. The altered percentages were applied equally to each spending category for the top spenders (i.e., the top spenders that move from one pool to the other represent the average cost of all top spenders).4

Smaller Scale Example

To bring this illustration down to a more manageable scale, we will assume an employer covers 5,000 workers and dependents with an expected claims distribution that matches the one that we used for illustration shown in Figure 2. We will further assume the employer has two health plans, one that covers 80% of its workers and dependents (4,000 enrollees) and another that covers the remaining 20% (1,000 enrollees). If the two plans get an even distribution of risk, each would have the same average paid claims cost of $1,633. If, however, one pool ends up with an additional 10 of the top spenders, the average paid claim amount in the smaller plan would decrease by $94 to $1,539, while the average paid claim in the larger plan would increase by just over $20 to $1,656. As a result of shifting just 10 out of 1,000 enrollees from smaller plan to the larger plan, the average cost in the larger plan become 7.6% higher than the average cost in the smaller plan. A shift of just 20 of the top spenders would produce over a 16% difference in the average costs between the two plans.

Figure 2 show the difference in paid claims in the two the pools under each of the scenarios. For example, in the 52%/48% scenario, one pool –- which we will refer to as the adverse selection pool –- attracts an additional 2% of the top spenders from the other pool, which is an additional 3,463 claimants added to the 795,869 existing claimants in the pool. The average paid claims amount is 5% higher in the adverse selection pool than in the pool that gets the lower percentage of top spenders, which we will refer to as the favorable selection pool. Under the 54%/46% scenario, an additional 6,926 claimants join the adverse selection pool, and the difference in the average paid claims amounts in the two pools is 10%. Table 1 shows the results for each of the four scenarios, including the change in the average paid claim amount for the adverse selection pool compared with the average paid claim amount in the 50%/50% baseline scenario.

Figure 2: Average Paid Claims in Two Pools Under Four Selection Scenarios

Average Paid Claims in Two Pools

Table 1: Change in Enrollment and Differences in Average Paid
Claims Amounts Initially and Under Four Scenarios






Total Claimants in Adverse Selection Pool






Additional Claimants in Adverse Selection Pool (relative to 50%/50%)






Average Paid Claims Amount In Favorable Section Pool






Average Paid Claims Amount In Adverse Section Pool






Percentage Difference in Average Paid Claims between Adverse and Favorable Selection Pools






Percentage Increase in Average Paid Claims Amount in Adverse Selection Pool and 50%/50% Scenario






What we can see from this exercise is that adverse selection can occur and have a meaningful impact on claims cost even where there are no extreme changes in enrollment. The enrollment shifts portrayed here would be far too small to substantially change the demographic make up of either pool in terms of age, gender, income, or other observable factors –- even the 60%/40% scenario only increases the overall census of the adverse selection pool by 2%. Generalizations about enrollment would not reveal the real differences in underlying claim costs. If bias enrollment patterns such as these persist and insurers base premiums on claims experience, or if insurers can anticipate enrollment bias and base their premiums on expected claims, then these small enrollment differences would translate into fairly significant premium differences.

Separating the impact of selection from the impact of plan design is not a new problem for health policy, nor is it one that is easily solved. For example, the issue of whether enrollees in Medicare managed care plans are healthier than enrollees in Medicare fee-for-service has been a prominent one in health policy since the mid-1980s, with numerous studies considering whether bias selection exists and how plan payments might be adjusted to account for it.Researchers will need to undertake similar studies to identify whether there is a health difference between people who enroll in new consumer directed health plans and those who choose more comprehensive plans and, if so, how that difference affects plan costs. Such research can be time-consuming and expensive, but without it, policy makers will not be able to judge the true effectiveness of new plan designs.


1. For a discussion of how HSAs and HRAs work, see The consumer incentives tend to be stronger in HSAs than in HRAs because HSAs provide consumers with far more discretion over funds that accrue in the savings device than do HRAs.

2. See K.L. Grazier and W. G. Sell, “Group Medical Insurance Claims Database Collection and Analysis,” Society of Actuaries Research Project, September 2004, at We use the claims from the year 1999.

3. The top 11% of claimants had paid claims in excess of $3,000.

4. There are 48 spending categories among the top spenders. For the 52%/48% scenario, we assumed that 52% of the claimants in each category enrolled in one pool and 48% of the claimants enrolled in the other pool, each with a value equal to the average for the category.

5. There is a large volume of literature over a period of years. For example, see: Brown, Randell S, Dolores Gurnick Clement, Jerrod W. Hill, et al. “Do Health Maintenance Organizations Work for Medicare?” Health Care Financing Review 15(1) (1993): 7-23; Lubitz, J., J. Beebe and Gerald Riley. “Improving the Medicare HMO Payment Formula to Deal with Biased Selection.” In Advances in Health Economics and Health Services Research, vol. 6, edited by R. Scheffler and L. Rossiter. Greenwich, Conn: JAI Press, 1985; Physician Payment Review Commission. “Risk Selection and Risk Adjustment in Medicare.” In Annual Report to Congress, ch. 15. Washington, D.C.: Physician Payment Review Commission, 1996; Blumberg, Linda J., and Allison Evans. “Reform of the Medicare AAPCC: Learning from Previous Proposals.” Inquiry 35(1) (1998): 62-77; Newhouse, Joseph P., Melinda Beeuwkes Buntin and John D. Chapman. “Risk Adjustment and Medicare: Taking a Closer Look.” Health Affairs 16(5) (1997): 26-43; Mehrotra, A., Sonya Grier and R. Adams Dudley. “The Relationship Between Health Plan Advertising and Market Incentives: Evidence of Risk Selection Behavior.” Health Affairs 25(3) (2006): 5-24; Medicare Payment Advisory Commission. Medicare Brief: Medicare Advantage Benchmarks and Payments Compared with Average Medicare Fee-for-Service Spending. Washington D.C.: Medicare Payment Advisory Commission, June 2006.


KFF Headquarters: 185 Berry St., Suite 2000, San Francisco, CA 94107 | Phone 650-854-9400
Washington Offices and Barbara Jordan Conference Center: 1330 G Street, NW, Washington, DC 20005 | Phone 202-347-5270 | Email Alerts: | |

The independent source for health policy research, polling, and news, KFF is a nonprofit organization based in San Francisco, California.