Printer Friendly

Insurance choice and the demand for prescription drugs.

I. Introduction

Over the past decade or so, expenditure on outpatient prescription drugs has become one of the fastest-growing components of health care expenditure in the U.S. [11]. Much of the debate over the Medicare Catastrophic Coverage Act of 1988 and the current debate over health care reform is concerned with the rising cost and utilization of prescription drugs particularly among the elderly. There is of course a general debate over the extent to which the problems of adverse selection and/or moral hazard arising from universal coverage will render futile the cost estimates for the sundry forms of health insurance to be covered under any reform. The case of prescriptions is particularly interesting, as witness the pressure placed on pharmaceutical companies by administration officials in 1993.

This paper analyzes the influence that health insurance has on elderly individuals' decisions to use prescription drugs. We create a data base from a survey of health insurance and medicine use in the Commonwealth of Pennsylvania conducted during the summer of 1990 by researchers affiliated with the Medicine, Health, and Aging Project at Penn State [23]. Pennsylvania is particularly interesting in this regard because of the generous provisions of the PACE (Pharmaceutical Assistance Contract for the Elderly) program, which pays for all outpatient prescriptions for low income elderly, less a $4.00 copayment per 30 days dosage.

Our analytic framework is dictated by three considerations. First, we wish to produce quantitative estimates of insurance effects over a wide range of prescription benefit provisions. Most prior research on prescription drug demand has focused on the effects of relatively minor alterations in insurance benefits such as the addition of a $.50 or $1.00 copayment. Findings from these studies are not generalizable to situations in which individuals gain (or lose) prescription coverage or face other major changes in prescription benefits.

Second, we wish to address the question of whether proscription demand is driven more by the own-price effects of drug coverage or by the cross-price effects of Medicare supplementation for ambulatory physician visits. Outpatient physician visits are covered under Medicare Part B, but are subject to deductible and coinsurance provisions which may reduce their use. Since physicians control access to prescription medicine, it follows that changes in the price of physician care may affect drug demand as well.

The third consideration is methodological rather than substantive. Elderly who have prescription coverage obtain it from three basic sources: employer-sponsored plans, individual Medigap policies, or public programs including Medicaid and pharmaceutical assistance plans like PACE. Except for employer-sponsored plans, the elderly must seek coverage on their own. It is reasonable to suspect that those who do are more likely to need prescription drugs than those who do not seek coverage. Unless controlled for, self selection can lead to serious bias in models designed to measure the moral hazard engendered by insurance.

Most of the empirical research on the demand for prescription drugs involve studies of differences in utilization following small changes in insurance coverage. Several studies conducted in the 1960s and early 1970s concluded that Medicaid copayments reduce drug utilization rates [7; 2; 21], but methodological shortcomings make the results difficult to generalize. Later studies using interrupted time-series designs have produced similar findings. A South Carolina study observed a significant decline in drug use following imposition of a $.50 drug copayment in that state's Medicaid program in 1977 [12; 17]. A study by Soumerai et al. [22] found significant reductions in prescription fill rates among New Hampshire Medicaid recipients following imposition of a three-per-month [R.sub.x] limit and a $1.00 copayment.

Similar before-and-after studies (some with controls, some without) have been conducted in Great Britain following changes in National Health Service patient cost-sharing provisions for prescription drugs that occurred at various times from 1969 through 1986 [16; 20; 1; 8; 14; 19]. Estimated [R.sub.x] price elasticities from these studies range from a low of -.06 [19] to a high of -.64 [14], with a preponderance of estimates in the -.10 to -.20 range.

Changes in cost-sharing for prescription medicine among privately insured groups in the United States have also been shown to have significant effects on drug utilization in the expected direction. An early study by Weeks [25] found that the introduction of prepaid drug benefits in an employee health plan increased the average number of prescriptions filled. A recent study by Harris, Stergachis and Ried [6] discovered that progressively higher copayment levels led to proportionate reductions in drug use among non-aged HMO enrollees.

The Rand Health Insurance Experiment (HIE) produced two papers on the relationship between drug utilization and insurance coverage [10; 9]. Unfortunately, the health insurance packages offered to HIE participants did not vary according to prescription drug benefits, so that it proved impossible to estimate directly the price effect of [R.sub.x] coverage on utilization levels. Moreover, the sample frame excluded the elderly. Despite these shortcomings, the HIE provides important clues to the possible impact of insurance on medicine use. In their 1985 paper, Leibowitz, Manning, and Newhouse [10] examined the relationship between [R.sub.x] utilization and patient cost-sharing (for all medical services, not just drugs). They found that HIE enrollees with generous insurance filled significantly more prescriptions than did those with less generous coverage; in fact, the degree of price responsiveness did not differ much between drugs and other medical services. Their conclusion that "drugs, like medical care expenditures in general, respond to cost-sharing faced by consumers" [10, 1068] has been widely debated on grounds that the HIE results cannot distinguish between the own-price effect of insurance on the covered service in question (prescription drugs) and the cross-price effect of coverage for services that complement drug therapy (physician visits). A second Leibowitz paper [9] reported no significant relationship between insurance plan generosity and utilization rates for over-the-counter medicine (which may potentially substitute for [R.sub.x] products).

Finally, we take note of a study by Cameron et al. [3] which uses Australian data to estimate a series of health services demand equations including equations for prescription drug use. This study is relevant to our purposes mainly because it demonstrates the difficulty of estimating insurance effects in the presence of self-selected coverage (variation in own-price for prescribed medicine in Australia arises from private insurance supplements of the national insurance plan). The authors first estimate a model with endogenous insurance variables and find that more generous insurance coverage has a positive and significant effect on prescription utilization. They then reestimate the models using instrumental variables for the insurance options and find implausibly large insurance coefficients and a sign reversal (persons subject to a $2.00 prescription copayment are predicted to use nearly double the number of prescribed medicines compared to those for whom drugs are free).

II. Empirical Model

The estimation of insurance effects on the demand for health care in choice-based samples is difficult because the determinants of insurance coverage are strongly correlated with the determinants of use. Persons with poor health have higher utilization rates, but are also likely to purchase larger amounts of insurance. For the analyst this creates a modeling problem since the insurance variables will capture both the demand-inducing effect of better coverage and residual effects associated with insurance selection. To make matters worse, the standard econometric remedies for endogeneity - instrumental variables and Heckman selection-control models - perform poorly in this context. Experience similar to Cameron's has been reported by Newhouse and Phelps [13] and others. A recent Monte Carlo experiment reported by Hartman [6] shows that two-stage estimators are inherently unstable. The problem appears to arise from the fact that the instrument created in the first-stage stage equation is highly collinear with the exogenous variables in the second stage. The coefficient estimates produced in the second stage are thus extremely sensitive to minor changes in the specification of either equation.

Our approach to modeling the demand for prescription drugs controls for self-selectivity in another way. It is a variation of a longitudinal model developed by Wolfe and Goddeeris [26] in which past utilization is used to proxy for the unobserved determinants of insurance choice. We test for the endogeneity of insurance choice using Hausman tests and find that insurance choice is correlated with the error term when our proxy ms excluded and uncorrelated when it is included. Thus the Hausman test is able to discriminate between the two models, and justifies, to that extent, the use of the designated proxies. On that basis we can measure the moral hazard associated with insurance coverage directly from the insurance coefficients.

Formally, we propose a model of prescription demand of the following sort,

[] = [[Beta].sub.1][] + [[Beta].sub.2][] + [] (1)

where M is the number of prescription drugs reported purchased by the ith individual within a certain time period (t), X is a vector of personal characteristics, and I is a vector of insurance dummies. Although [[Beta].sub.2] captures the moral hazard effect, OLS estimates of equation (1) are biased because the error term is of the form:

[] = [] - [] (2)

where v is random noise and h is health status over the period in which the insurance is in force. The value of [] is not directly observable until after the insurance decision has been made but is presumed by the individual (and the modeler) to be related to []. The better the health the fewer the M, but also the less likely is insurance. Hence I is correlated with the error term.

Wolfe and Goddeeris deal with this difficulty by assuming that h is first-order autoregressive. Performing the Cochrane-Orcutt transformation yields an equation in which lags of all the variables are included on the right hand side and the error term consists only of [] and the residual from the AR(1) model of [] and hence is uncorrelated with the regressors.

This model is a specific example in the class of models where health status is forecasted linearly from some information set Z:

E([][where][]) = [[Beta].sub.3][] (3)

where Z is the set of information available to the ith individual at time t - 1. In the above scenario, Z consists of all lagged variables, but because of the autocorrelation, [[Beta].sub.3] is constrained. Our data set, described in the next section, is limited in that we do not have access to all the variables at time t - 1 (specifically, we do not have longitudinal data on prescription drug use). We do, however, have data on a number of health indicators which we believe will encompass the information available at time t - 1. Because we eschew the parameterization suggested by the Cochrane-Orcutt transformation, the estimates of [[Beta].sub.3] are unconstrained.

The regression model is simply:

[] = [[Beta].sub.1][] + [[Beta].sub.2][] + [[Beta].sub.3][] + [] + [] (4)

where u is the prediction error on h. Under the current error structure, the estimate of [[Beta].sub.2] is consistent.

We can test the appropriateness of equation (4) via a Hausman test. We run this regression without [Z.sub.t-1] and use Hausman procedures to test for endogeneity of the insurance choice variables. It will turn out (as described below) that exogeneity can be rejected, and from this we infer that the difficulties discussed above indeed exist. We then add [Z.sub.t-1] to the list of regressors and reperform the test, which in this case cannot reject exogeneity. Hence the Hausman test is able to discriminate between the two specifications and leads to our belief that the expanded model provides consistent estimates.

We estimate three versions of equation (4). First is an OLS model with observations on prescription drug use for all individuals (including those with no reported use). Second is a Probit probability-of-use model with prescription utilization coded as a binary variable (1 = reports use of prescription drugs; 0 = reports no drug use). Third is an OLS level-of-use equation estimated over the subset of individuals reporting one or more prescription during the study period.

III. Data

Our data set is a panel constructed from survey responses and Medicare claims records for 4066 elderly Pennsylvanians who were asked about their demographic characteristics, current health status, insurance coverage and medicine use in a mail questionnaire administered in the late summer and early fall of 1990. The survey sample was randomly selected from Medicare enrollment files for Pennsylvania. Disabled beneficiaries under age 65 were excluded from the sample frame as were elderly HMO enrollees.(1) The survey questionnaire was sent to 6,502 persons. After reminders and repeat mailings 4,508 responded (69.3 percent). Further detail on the survey instrument and its implementation is contained in Stuart et al. [23].

We selected calendar year 1988 as the t - 1 period for the Z variables. Thus when constructing the analytical file for the study we eliminated 442 individuals who were not covered for Medicare Parts A and B continuously from January, 1988 through the survey date. The characteristics of the reduced sample are shown in Table I. Respondents reported an average of 1.23 prescriptions or prescription refills during a two-week recall period. Approximately half (54 percent) filled at least one prescription and among that group use averaged two-and-a-half [R.sub.x] scripts. These utilization rates are considerably above those reported in the most current National Medical Expenditure Survey [11], but then the percentage of elderly Pennsylvanians with prescription coverage is also significantly above the national mean [23].

The X vector is represented by various demographic characteristics including age, sex, race, education, annual income, and marital status. The income breakdown includes two categories used to identify individuals putatively eligible for public programs, Medicaid (income [less than] $6,000) and PACE (income [less than] $12,000 if single, [less than] 15,000 if married). The next income bracket ($12,000 to $15,000) is limited to single persons given the PACE restriction just noted. The X vector also contains measures of self-reported health status contemporaneous to the survey date. These included general questions about physical and mental health, questions about specific health conditions (here summarized into two measures), and queries regarding smoking and drinking behavior.

As noted, all respondents are Medicare beneficiaries. They supplement Medicare in several ways including Medicaid, PACE, and private insurance plans. We characterized the latter according to source of coverage (employer-sponsored and individual Medigap plans), type of benefit (ambulatory physician visits and prescription drugs), and single or multiple plans. Individuals may be included in more than one category. In fact most are.

The survey did not elicit information on the scope of drug benefits provided in the private plans. Medicaid represents the most comprehensive drug coverage on the list. During the study period elderly Medicaid recipients were charged a $.50 copayment for certain classes of drugs, but most prescription medicine was free. PACE imposed a $4.00 copayment per prescription or prescription refill (both limited to dosages of 30 days). The modal employer-sponsored plan included prescription benefits of some type. Few individual Medigap policies provide this coverage. Most private plans that supplement Medicare Part B coverage of physician visits pay the Medicare 20 percent coinsurance; few cover the Part B deductible or excess charges above Medicare limits.

The [Z.sub.t-1] vector captures prior health status and is proxied by 1988 Medicare utilization. We had a wide choice of Medicare variables for this purpose and selected two, days hospitalized in 1988 and total Medicare Part A payments for that year.

The final set of model variables represent health care resources available to survey respondents. These include measures of physician, hospital, nursing home, and pharmacy supply relative to county population levels. Data for these variables were obtained from the Area Resource File and the Pennsylvania Department of Health and were linked to the person-level analytical file via respondent addresses and the county FIPS code.


IV. Results

We are interested in the effects that the six types of insurance have on the use of pharmaceuticals. Enrollment in employer-sponsored plans is considered exogenous in all models, leaving us with five potential endogenous insurance dummies. Our first task is to use the Hausman test to deliver models free of endogeneity problems. The Hausman test consists of two steps (see e.g., Godfrey [4]). First, five probit regressions are run with the insurance dummies as dependent variables. The residuals from these regressions are collected and added as covariates to the utilization equation (with the insurance choice dummies also included). If the coefficients of the residuals are significantly different from zero, exogeneity is rejected.

Following this procedure, the probits were estimated, using as right-hand side variables all of the variables discussed above. Also included, to aid in identification, were county-specific measures of rurality.(2) The residuals from these models were then inserted into the prescription utilization equations containing the prior utilization ([Z.sub.t-1]) measures and those without them. We rejected exogeneity in every model that did not contain the [Z.sub.t-1] measures. For example, in the version of the first model (use estimated over the entire sample) that contained no [Z.sub.t-1] variables, the F-test for the coefficients on the residuals was 9.307 (p [less than] .0001). We conclude that exogeneity is rejected at any conventional level of significance. When this same model is estimated with the [Z.sub.t-1] variables included, the F-statistic was 0.2015 with a p-value of 0.96. Exogeneity in this case is substantively accepted. The Hausman test has discriminated between the two versions of the utilization model and the [Z.sub.t-1] measures are shown to be jointly adequate for the representation of unobserved determinants of insurance choice.(3)

To then estimate the selection-purged models, we run the regressions with the auxiliary regressors (the probit residuals) removed. Our findings are presented in Table II.

Overall, the models perform quite well, explaining between 15 and 20 percent of the variance in prescription use among respondents. Many of the individual variables, however, have very low significance. These include ethnicity, educational attainment, income,(4) marital status, mental health rating, and health care resources in the county. By contrast, all of the indicators of current physical health are significant in both the full-sample utilization equation (column 1) and the probability of use model (column 2). Moreover, the magnitudes of the health status indicators are in accord with expectations: people in better (self-perceived) health use fewer medications. The prior utilization [Z.sub.t-1] measures also have the expected signs and are statistically significant in two of the three models.

Among the background variables are three idiosyncratic coefficients: alcohol users, smokers, and age all have negative coefficients when positive might have been expected. In the case of age, the result is not entirely unexpected given the evidence of morbidity compression found in our month-to-death models described in Stuart et al. [23]. If very old survivors are healthier than their younger counterparts, then on average older people within any given cross-section will use fewer prescription drugs (even when the time trend for any individual is positive). The signs on the other two variables are unexpected. The highly significant negative effects of drinking and [TABULAR DATA FOR TABLE II OMITTED] smoking on the probability of filling any prescription is particularly problematic. It seems unlikely that these behaviors would have protective effects on health (although there is some evidence that light drinking reduces risk of heart attacks). Perhaps instead these coefficients reflect a general recklessness towards health.

We are, of course, primarily interested in the coefficients of the insurance dummies. The most striking result is for PACE enrollees. According to the full-sample estimate (column 1) PACE beneficiaries fill 0.29 more prescriptions or refills per two-week period than do elderly who are not covered by an employer-sponsored plan and have neither prescription coverage nor Medicare supplementation for ambulatory physician visits (the excluded category for the insurance vector). Since the model has been putatively purged of adverse selection bias, we conclude that this is the result of the price subsidy that PACE beneficiaries enjoy.

The results for other insurance and subsidy programs are less substantive and less precise. In the full-sample estimates, private coverage for ambulatory physician care and prescription drugs also increase pharmaceutical use by about one-tenth of a prescription in a two-week period, although the test statistics indicate only borderline significance. Employer-sponsored insurance appears to have a negative effect on prescription use which is most notable in the probability-of-use equation (column 2). This anomaly is more apparent than real since prescription and physician coverage effects are independently assessed. The negative coefficients on [R.sub.x] use among those with employer-based coverage may reflect some residual health status effects associated with previous employment. It is unlikely that these findings are due to usage restrictions or drug review imposed by employers for the simple reason that such utilization control programs were not common during the study period. The failure to find significant effects associated with multiple Medigap coverage should be interpreted in the same light as the employment insurance coefficients. However, we are surprised by the consistently negative (albeit insignificant) signs on the Medicaid coverage variable. Medicaid offers the most generous package of prescription benefits available to elderly Pennsylvanians and it would be perverse in the extreme for recipients to reduce their use of prescribed medicines because of the benefit. In all likelihood, these coefficients are an artifact of some other characteristic of Medicaid enrollment not captured in the model.

Further insight into the effect of insurance on drug utilization can be gained by comparing the probability-of-use and level-of-use-by-users results in columns 2 and 3 of Table II. Here it becomes evident that insurance does not simply shift the demand curve outward. Rather the demand-inducing effect appears to be limited to those who would not use any prescription medicine in the absence of insurance. For example, at mean values for the background variables, enrollment in PACE raises the probability of prescription use by 11 percent while ambulatory physician coverage raises it by 5 percent. In both cases the effects are significant at p [less than] .001. By contrast, neither form of coverage has a statistically significant effect on the number of medicines used by persons filling one or more prescriptions. Indeed, none of the insurance dummies was significant in the use-by-users equation.

These findings shed new light on the demand for prescription drugs by the elderly. It appears clear that drug use is price sensitive both to direct subsidies and to coverage of complementary physician services. The actual degree of price sensitivity is impossible to compute for most of the insurance variables because of heterogeneous benefits within the coverage groups (even Medicaid recipients face different out-of-pocket prices depending on the drugs prescribed). PACE is an exception. During the study period PACE imposed a flat $4.00 copay per prescription or refill regardless of the actual charge. Between July and December 1990 (when the survey was fielded), the average usual and customary charge per PACE prescription was $24.98 [15]. Most pharmacies offer elderly patrons a standard 10 percent "senior citizen discount" which would reduce the average noninsured prescription charge to $22.48. The average subsidy is thus $18.48 or 82.2 percent. Based on our full-sample regression results, PACE beneficiaries responded to this price reduction with a 27.6 percent increase in the quantity of prescriptions purchased,(5) thus yielding an own-price elasticity of -.34.

V. Concluding Remarks

Whether this elasticity estimate can be generalized beyond the PACE program is an open question. We are confident that the PACE coefficients truly reflect price sensitivity and not adverse selection, but they reflect it within the context of a particular program with unique features. PACE differs from most private Medicare supplements, whether employer-sponsored or individually-purchased, in that the program reimburses pharmacies directly. Some authors have conjectured that "card programs" (like PACE) induce greater demand than indemnity plans with similar benefit coverage because there is no need to pay up front and beneficiaries are protected from the risk of lost or misplaced receipts [18]. PACE also differs from other public-sector pharmaceutical assistance plans and state Medicaid programs in that it places few administrative restrictions on drug benefits, with one notable exception. There is no restrictive formulary. Both branded and generic products are covered. There are no limits on the number of prescriptions that can be filled at one time nor are there limits on refills. However, coverage is limited to one-month's supply at a time; each additional month's supply is subject to another copay.

This last feature both lowers the value of the PACE subsidy for maintenance drugs and makes it appear as if program beneficiaries are refilling more prescriptions than are individuals whose prescriptions provide two or three month's supply. Because our survey did not distinguish between initial prescriptions and refills, we cannot rule out the possibility that some of the higher demand attributed to the PACE price subsidy is really an artifact of this coverage restriction. To the extent this bias is present in our results it should be most evident in the use-by-user equation, since there are the people most likely to be refilling maintenance prescriptions. We simply note that the PACE coefficient in this equation is small and insignificant whereas the PACE coefficient in the probability-of-any-prescription-fill equation is large and highly significant.

Caveats aside, the results of this analysis strongly suggest that if prescription drug coverage for the elderly were improved - say through an expansion of Medicare benefits - that the demand for drugs would increase. But by how much? Surely by less than a doubling of drug expenditures on behalf of the previously uninsured, as recently projected by Waldo [24]. Nor is the impact likely to be as large as the 3.4% increase per 10% decline in out-of-pocket expense predicted by our own PACE coefficients (for reasons described above). The importance of our findings lies less in the point estimates than in the ability of the estimates to show that higher utilization rates are due to the insurance coverage itself and are not just an artifact of adverse selection.

1. Medicare HMO enrollees were excluded because HCFA does not maintain complete utilization and payment records for them.

2. The rurality measure is the 10-level Human Resource Profile County Adjacency Code classification scheme compiled by the Health Resources Administration and available on the Area Resource File.

3. A precisely analogous pair of results is obtained using the linear probability model in the first stage.

4. The coefficients on the categorical income variables capture the effect of income net of insurance coverage. As expected, the price effects of insurance - particularly the comprehensive coverage afforded by Medicaid and PACE - are stronger than income effects on prescription demand.

5. The percentage quantity increase is obtained by dividing the PACE coefficient (0.29) by the intercept (1.05).


1. Birch, S., "Relationship Between Increasing Prescription Charges and Consumption in Groups Not Exempt From Charges." Journal of the Royal College of General Practitioners, January 1986, 154-56.

2. Brian, E. and S. Gibbens, "California's Medical Copayment Experiment." Medical Care, December 1974 (supplement), 1-56.

3. Cameron, A. Colin, P. Trivedi, F. Milne, and J. Piggott, "A Microeconomic Model of the Demand for Medical Care and Health Insurance in Australia." Review of Economic Studies, January 1988, 85-106.

4. Godfrey, L. G. Misspecification Tests in Econometrics. Cambridge: Cambridge University Press, 1988.

5. Harris, B., A. Stergachis, and L. Ried, "The Effect of Drug Co-payments on Utilization and Cost of Pharmaceuticals in Health Maintenance Organizations." Medical Care, October 1990, 907-17.

6. Hartman, Richard, "A Monte Carlo Analysis of Alternative Estimators in Models Involving Selectivity." Journal of Business and Economic Statistics, January 1991, 41-49.

7. Knapp, D., "Paying for Outpatient Prescription Drugs and Related Services in Third-party Programs." Medical Care Review, August 1971, 826-59.

8. Lavers, R., "Prescription Charges, the Demand for Prescriptions and Morbidity." Applied Economics August 1989, 1043-52.

9. Leibowitz, Arleen, "Substitution Between Prescribed and Over-the-counter Medications." Medical Care, January 1989, 85-94.

10. -----, W. Manning, and J. Newhouse, "The Demand for Prescription Drugs as a Function of Cost-sharing." Social Science and Medicine, 1985, 1063-69.

11. Moeller, J. and N. Mathiowetz, Prescribed Medicines: A Summary of Use and Expenditures by Medicare Beneficiaries. Washington, D.C.: National Center for Health Services Research and Health Care Technology Assessment, September 1989.

12. Nelson, A., C. Reeder, and W. Dickson, "The Effect of a Medicaid Drug Copayment Program on the Utilization and Costs of Prescription Services." Medical Care, August 1984, 724-36.

13. Newhouse, Joseph, and C. Phelps. "New Estimates of Price and Income Elasticities of Medical Care Services," in The Role of Health Insurance in the Health Services Sector. New York: NBER, pp. 261-313.

14. O'Brian, B., "The Effect of Patient Charges on the Utilisation of Prescription Medicines." Journal of Health Economics, March 1989, 109-32.

15. Pennsylvania Department of Aging. PACE: Pharmaceutical Assistance Contract for the Elderly. Biannual Report to the Pennsylvania General Assembly, July 1990-December 31, 1990. Harrisburg, Penn.: Pennsylvania Department of Aging, 1991.

16. Phelps, Charles and J. Newhouse, "Coinsurance, the Price of Time and the Demand for Medical Services." Review of Economics and Statistics, August 1974, 334-42.

17. Reeder, C. and A. Nelson, "The Differential Impact of Copayment on Drug Use in a Medicaid Population." Inquiry, Winter 1985, 396-403.

18. Rubinstein, E., "Employer Strategies in Managing Prescription Drug Costs." Business and Health. January 1991, 2-6.

19. Ryan, M. "Estimating the Effects of Prescription Charges on the Use of NHS Prescription Drugs in England, 1979-1985." Discussion Paper 3/89, Health Economics Research Unit, University of Aberdeen.

20. Schnell, B., "Pharmaceutical Services Under the British National Health Service." Drugs and Health Care, 1975, 70-71.

21. Smith, M. and D. Garner, "Effects of a Medicaid Program on Prescription Drug Availability and Acquisition." Medical Care, July 1974, 571-81.

22. Soumerai, S., D. Ross-Degan, J. Avorn, et al. "Payment Restrictions for Drugs under Medicaid: Effects on Therapy, Cost, and Equity." New England Journal of Medicine, August 27, 1987, 550-56.

23. Stuart, Bruce, et al. An Analysis of Determinants and Consequences of Prescription Drug Coverage for Pennsylvania Elderly. Final report to the Pew Charitable Trust under grant n. 88-01439-00. Vol. 1. University Park, Penn.: Medicine, Health, and Aging Project, The Pennsylvania State University, 1991.

24. Waldo, Daniel, "Estimating the Cost of a Medicare Outpatient Prescription Drug Benefit." Health Care Financing Review, Spring 1994, 103-12.

25. Weeks, H., "Changes in Prescription Drug Utilization After the Introduction of a Prepaid Drug Insurance Program." Journal of the American Pharmaceutical Association, 1973, 205.

26. Wolfe, John and J. Goddeeris, "Adverse Selection, Moral Hazard, and Wealth Effects in the Medigap Insurance Market." Journal of Health Economics, November 1991, 433-59.
COPYRIGHT 1995 Southern Economic Association
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1995, Gale Group. All rights reserved. Gale Group is a Thomson Corporation Company.

Article Details
Printer friendly Cite/link Email Feedback
Author:Stuart, Bruce C.
Publication:Southern Economic Journal
Date:Apr 1, 1995
Previous Article:Speed of adjustment in commercial real estate markets.
Next Article:Exogeneity and the export-led growth hypothesis: the case of China.

Related Articles
[R.sub.X] for Seniors.
Managed choice: health insurers are responding to consumer demand for more options with tiered pricing for doctors and hospitals patterned after...
The other prescription drug benefit plan. (View on Washington).
Drug plan draws overflow crowd.
Medicare benefit meeting offered.
Special delivery: most studies show mail-order pharmacies save money for insurers and consumers, but retail stores are demanding to be an option,...
Extend the deadline.
Silver lining: 'co-logo contracts' can help health insurers glide though the clouds enveloping Medicare Part D.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters