NBER Working Paper 22802
Intertemporal Substitution in Health Care Demand: Evidence from the RAND Health Insurance Experiment
By Haizhen Lin and Daniel W. Sacks
National Bureau of Economic Research, November 2016
Nonlinear cost-sharing in health insurance encourages intertemporal substitution because patients can reduce their out-of-pocket costs by concentrating spending in years when they hit the deductible. We test for such intertemporal substitution using data from the RAND Health Insurance Experiment, where people were randomly assigned either to a free care plan or to a cost-sharing plan which had coinsurance up to a maximum dollar expenditure (MDE). Hitting the MDE — leading to an effective price of zero — has a bigger effect on monthly health care spending and utilization than does being in free care, because people who hit the MDE face high future and past prices. As a result, we estimate that sensitivity to short-lasting price changes is about twice as large as sensitivity to long-lasting changes. These findings help reconcile conflicting estimates of the price elasticity of demand for health care, and suggest that high deductible health plans may be less effective than hoped in controlling health care spending.
From the Introduction
Studying intertemporal substitution is important for at least two reasons. First, with the important exceptions of Einav et al. (2015) and Cabral (2016), most of the literature on health care demand and the effect of insurance on spending has neglected intertemporal substitution, and estimated in a wide range of price elasticities, from as small as -0.2 to as large as -1.5 (for example, Manning et al. (1987); Eichner (1998); Zweifel and Manning (2000); Cardon and Hendel (2001); Bajari et al. (2014); Dalton (2014) and Kowalski (2015, 2016))). Most papers assume that health care decisions are made statically on an annual basis, meaning that there is no scope for future prices to affect current demand. Estimated. However, if there is intertemporal substitution, then patients may respond very differently to a temporary price change than to a long-lasting one. Therefore, depending on the sources of price variation used for identification, one might draw dramatically different conclusions regarding price sensitivities. Accounting for intertemporal substitution makes it possible to separate elasticities with respect to permanent or temporary price changes, and may help reconcile the disparate elasticity estimates in the literature.
Second, intertemporal substitution affects the response to high deductible health plans, which are now common in the American health insurance landscape. Regulators, insurers, and policymakers tolerate the weak risk protection of high deductible health plans out of the hope that they will reduce health care spending. This view implicitly assumes that care foregone in one year because of the high deductible represents a permanent reduction in health care spending. But if patients are deferring needed care, then their spending may be higher in future years, either because deferrable problems become so severe that they must be addressed, regardless of the cost, or because once patients finally do hit the deductible, they stock up on care, retiming deferrable procedures to a year when their price is low. Thus a key question for the effectiveness of high deductible health plans is whether patients intertemporally substitute in their demand for health care. Evidence of intertemporal substitution would suggest that high-deductible health plans may not be as effective as hoped in controlling overall health care spending.
Reconciliation with original HIE findings
We have argued that intertemporal substitution is an important part of how patients respond to nonlinear cost-sharing, causing them to stock up on health care when it goes on “sale,” with especially large anticipatory responses. Neglecting these dynamics can lead to biased estimates of the long-run effect of cost-sharing on utilization. The original RAND investigators, however, argued that intertemporal substitution was not an important part of the response to cost-sharing, and found little evidence for anticipatory effects. Here we reconcile these different conclusions.
The key difference between our analysis and the original investigators is in the timing of when we look for intertemporal substitution and anticipatory responses. They focus on the period around hitting the MDE, before and after. We focus on the end of the coverage year. As Keeler and Rolph acknowledge, it is likely difficult to detect anticipatory effects or pent up demand by focusing on fine timing around hitting the MDE. Households may not know exactly when they hit the MDE (as has been pointed out by the original RAND investigators), and may not appreciate the link between their current and future spending (Einav et al., 2015; Dalton et al., 2015; Abaluck et al., 2015). On the other hand, by the end of the coverage year, most families who hit the MDE will have seen a bill which makes clear their financial position, and it is not hard to understand that in the future, prices will be higher. Indeed, providers may help make this clear. Thus the myopia or limited understanding of the insurance contracts may have made it difficult for the original investigators to identify intertemporal substitution; by looking at the end of the year, we avoid this difficulty.
From the Conclusion
Studying data from the RAND Health Insurance Experiment, we found striking patterns of health care spending over the insurance coverage year. In most months, spending is lower in cost-sharing plans than in the free care plan. But in the last 1-3 months of the coverage year, spending rises quickly in cost-sharing relative to free care, and by the end of the year spending in the two plans is roughly equal. On the other hand, spending in free care is roughly flat over the coverage year, but it is particularly high early in the first coverage year, and it spikes dramatically at the end of the experiment.
These patterns are inconsistent with the standard model of demand for health care, which assumes away intertemporal substitution. Instead they suggest that patients can retime their care, especially for medically deferrable procedures and dental care, to reduce their out-of-pocket expenses in the face of nonlinear cost-sharing rules. To quantify the importance of intertemporal substitution, we estimate how health care spending responds to lag and lead prices as well as current prices. The estimates suggest that short-run moral hazard — the response to a one time, unanticipated price change — is substantially larger than the long-run response.
These results have important implications for health care spending and insurance design. First, they help reconcile some of the disparate estimates of the price elasticity of health care demand, since they imply that health care spending is more responsive to temporary price changes — for example, hitting the deductible — than to permanent price changes, for example, from insurance plan changes. Second, they suggest that high deductible health plans may not be as effective as hoped in controlling health care spending. These plans can reduce health care spending as long as patients do not hit the deductible. But in years when patients do hit it — as they eventually will — the large short-run response means that spending will make up for lost time, as patients stock up on care.
By Don McCanne, M.D.
At the risk of over-simplification, we can interpret this complex study to show us that previous conclusions from the RAND Health Insurance Experiment (RAND HIE) ignored the responsiveness of year-end health care spending once hitting the deductible, and that particular increase in spending suggests that “high deductible health plans may not be as effective as hoped in controlling health care spending.”
The primary purpose of high deductibles in insurance plans is to reduce health care spending by insurers, employers and government plans. The tradeoff is that they create financial hardships and impair health care access for patients. But isn’t health care reform supposed to be about the patients?
How much are deductibles really saving? For the 20 percent of people using 80 percent of health care, the savings are negligible since the deductible is met early on for those individuals and thus has no further impact. For the four-fifths of us who use only one-fifth of the nation’s health care, many of us have deductible coverage through Medicaid, Medigap or MA plans, or health savings accounts, and thus are not affected.
So who is negatively impacted by high deductibles? It is individuals who do not qualify for Medicaid nor for ACA subsidies and who often cannot fund a health savings account. This is our workforce and their families. As a percent of our national health expenditures, the amount saved is very small, and may be even smaller than believed based on this NBER study. This seems to be too heavy of a price to pay for such a relatively small savings.
So high deductibles don’t work very well and they harm patients. In contrast, single payer tools do control spending while helping patients. Do we really have to continue contemplating making the change that would finally fix our system? Why not action now?