1 See Sherry A. Glied and Dahlia K. Remler, "What Every Public Finance Economist Needs to Know about Health Economics: Recent Advances and Unresolved Questions," National Tax Journal, Vol. 55, 2002, pages 771-88.

2 See "National Health Expenditures Tables," Centers for Medicare and Medicaid Services, Table 9, http://www.cms.hhs.gov/statistics/nhe/historical

3 U.S. Department of Health and Human Services, Report to the President: Prescription Drug Coverage, Spending, Utilization, and Prices, April 2000.

4 From The Effects of H.R. 2473 on Prescription Drug Costs of Medicare Beneficiaries by the Office of the Assistant Secretary for Planning and Evaluation, Department of Health and Human Services, June 24, 2003; http://aspe.hhs.gov/search/health/HR2473.htm. The prescription drug provisions of the bill passed by the House, H.R. 1, are similar to those in H.R. 2473.

5 The impact of increasing patient prescription drug cost sharing on drug expenditures, and also the quantity side of the demand, have received much attention in literature. See for example, A, Leibowitz, Willard G. Manning Jr. and Joseph P. Newhouse, "The Demand For Prescription Drugs as a Function of Cost-Sharing," Rand Note, N-2278-HHS, October 1985; R.E. Johnson et al., "The Impact of Increasing Patient Prescription Drug Cost Sharing on Therapeutic Classes of Drugs Received and on the Health Status of Elderly HMO Members," Health Services Research, Vol. 32, No. 1, pages 103-22, April 1997; A. Street, A. Jones and A. Furuta, "Cost-sharing and Pharmaceutical Utilization and Expenditure in Russia," Journal of Health Economics, Vol. 18, No. 4, pages 459-72, 1999; Teresa B. Gibson et al., "Cost-Sharing for Prescription Drugs," Journal of the American Medical Association, Vol. 285, pages 2328-29, 2001; and Jessie X Fan, Deanna L. Sharpe and Goog-Soog Hong, "Health Care and Prescription Drug Spending by Seniors," Monthly Labor Review, Vol. 126, No. 3, pages 16-26, March 2003. The issues under consideration include: adverse selection in the health insurance market; differences in utilization and spending between patients with and without drug coverage across different income levels, ages, health status, and other categories; the influences of drug coverage on the type of medications used, etc. In contrast, less attention has been paid to the impact of drug insurance and cost sharing on the market price variation faced by patients. It is our hope that this short note will provide some preliminary evidence on this issue.

6 To further distinguish between the two competing theories about the effects of third-party payment and out-of-pocket spending in the market for prescription drugs, we estimated several alternative regressions that are presented in Appendix B. Note: As the share they pay rises, patients tend to continue searching up to the point at which the cost of searching equals the price reduction they expect. But since the patient's search costs are not reflected in the price he or she pays for the drug, these costs are largely hidden and are not included in this analysis. Similarly, third-party payers also are expected to search for and negotiate low prices for their claimants, and those costs also not known.

7 The initial set of sample restrictions requires valid values for pill strength in milligrams and number of tablets. Figures I and II are based on the 39 drugs for which there are at least 1,000 prescriptions after these restrictions. A second restriction is that both the total cost of the prescription and the amount paid out of pocket are not imputed.

8 Since Medicaid total payments are imputed, they are not included in the analysis. See Appendix A for additional information and for the definition of unit price, which is implemented as the price per milligram.

9 Table I provides additional details behind Figures I and II.

10 The number of prescriptions (observations) per drug ranged from 28 to 2,059. The average unit cost of the heavily prescribed drugs was more than double that of the other drugs, and patients pay a larger share of the cost.

11 An alternative specification of out-of-pocket share is also used in Appendix B. That specification divides out-of-pocket share into six categories, ranging from prescriptions paid entirely by the patient to those paid entirely by a third party. The alternative specification yields similar qualitative results.

12 Insurers also have the incentive to lower spending of their beneficiaries. The negative relationship between out of pocket spending and pill size in six cases may be consistent with insurers encouraging pill splitting. We also tested for the relationship between out of pocket spending and quantity buying. Higher out of pocket spending was negatively and significantly related to the purchase of 30 or more tablets for 34 of the 39 drugs considered. This indicates that bulk buying is not used as a means to lower unit costs among those with higher out-of-pocket spending.

13 Alan T. Sorensen, "Equilibrium Price Dispersion in Retail Markets for Prescription Drugs," Journal of Political Economy, Vol. 108, No. 4, 2000, pages 833-50.

14 See, for example, Devon M. Herrick, "Shopping for Drugs," Policy Report No. 262, June 2003, National Center for Policy Analysis. Available at http://www.ncpa.org/pub/st/st262/.

15 Our results also provide an indirect support for Sorensen's 2000 findings (Alan T. Sorensen, "Equilibrium Price Dispersion in Retail Markets for Prescription Drugs").

16 For a detailed description of this data file, see the Center of Medicare and Medicaid Services' Codebook for 1998 Cost and Use, accessible at http:/cms.hhs.gov/mcbs/codebkCNU/CNU1998/cnu98PME.asp. The procedure used to impute missing total costs does not provide regional variation. Imputed total costs begin with national average wholesale prices based on drug, strength and number of tablets. The wholesale prices are then discounted and assigned fixed dispensing fees based the presence of a third-party payer. Each type of third-party payer, for example, Medicaid, employer-based insurance or HMO, has different discounts and dispensing fees. Because the discounts and dispensing fees are identical within a third-party payer group and are applied to the national average wholesale price, prices for identical prescriptions do not vary. Given that the focus of the present study is the relationship between out-of-pocket spending and unit price, imputed values based on these procedures are not appropriate. For these reasons prescriptions for which total cost or out of pocket spending is imputed have been omitted from our analysis. It should also be noted that there has been some concern that household survey responses may underreport both drug utilization and cost. See John A. Poisal, Lauren A. Murray, George S. Chulis and Barbara S. Cooper, "Prescription Drug Coverage and Spending for Medicare Beneficiaries," Health Care Financing Review, Vol. 20, No. 3, pages 15-27, Spring 1999, and John A. Poisal, "Misreporting of Drug Expenditures in the Medicare Current Beneficiary Survey," Working Paper, Office of Research, Development, and Information, Centers for Medicare and Medicaid Services, 2003, for a discussion of misreporting of drug spending in the Medicare Current Beneficiary Survey.

17 This data truncation also eliminates observations with zero out-of-pocket payments, which is appropriate when considering only the effect of proportional cost-sharing arrangements. Fixed-dollar denominated copayments result in patients having no incentive to search because the marginal benefits of searching are zero. We return to this issue later in the paper.

18 Since the unit costs of different drugs are in different orders of magnitude, the pooled regression of all 247 drugs is likely to result in a heteroskedasticity problem. Therefore, all reported t-statistics are Huber-White consistent estimates. See H. White, "A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity," Econometrica, Vol. 48, pages 817-38, 1980. Given that the natural log of the unit price serves as the dependent variable, the coefficients reported in the table reflect the percentage change in price resulting from a change in the share paid out of pocket. To determine the change in unit price resulting from a change in the out-of-pocket share, the coefficient must be multiplied by the unit price, as discussed earlier. The values reported in the last columns of Tables II and III are the marginal effects evaluated at the means. An alternative specification of the dependent variable based on average daily dose results in the same parameter estimates in the pooled regression, given the controls for each drug. The within-drug regressions will not be affected by our use of cost per milligram as the dependent variable rather than average daily dose.

19 As stated earlier, we exclude drugs with fewer than 100 observations in Sample II. However, upon further elimination of observations involving fixed-dollar copayments, or imputed costs, the final number of observations used in regressions is smaller for each drug.

20 An alternative specification would divide the prescriptions based on the costs of an average daily dose.

21 Recall from Table III that the coefficient of the pooled regression of these 39 drugs (with a total of 31,086 observations) is negative and significant.

22 The most heavily prescribed drug, Furosemide, has a positive slope estimate, but it is small in magnitude (.068) and statistically insignificant (t-value is 1.203).

23 In all regressions, R2, the measure of the ability of out-of-pocket share to explain the variation in the unit cost in model (2), is low. This is due in part to the fact that the process by which prescription drug prices are determined is very complicated, especially when third parties are involved. When we include in the regression two types of regional dummies, metropolitan and census regions, and demographical characters, age, age squared, income, and a control for whether the patient has difficulty in daily activities, the R2 in each drug regression increases significantly. However, the basic results reported in the text remain largely the same. The coefficient estimates are negative for 30 drugs, of which 29 are significant. They are positive in the remaining nine drugs, of which five are significant. The pooled regression also has a significant negative estimate of the variable of out-of-pocket share (although conceivably smaller in magnitude). Furthermore, there are 3,691 observations with missing values of at least one of the above dummy and demographic variables (a reduction of 11.9 percent from 31,086). To preserve degrees of freedom, we chose to concentrate on the simple specification.

24 An alternative set of regressions was estimated in which the fixed-dollar copayments are identified as any integer numbers. With this specification of the explanatory variables, 29 of the 39 regressions produced a negative coefficient (of which 24 are significant), while only six were positive and significant.