Pharmacie sans ordonnance livraison rapide 24h: acheter viagra en ligne en France.
Microsoft word - price dispersion jpe.doc
PRESCRIPTION DRUG PRICE DISPERSION
IN HETEROGENEOUS MARKETS
School of Economic Sciences, Washington State University
Vincent H. Smith
Montana State University
September 6, 2008
Prescription drugs are homogeneous commodities that typically can be purchased from several different sources. However, prices for those drugs vary substantially among pharmacies within any given market. This study identifies three major causes of price dispersion: differing amounts of search effort, differing degrees of competition in each drug market, and different production costs. Extending previous research, this study shows that differences in price dispersion for the same commodity in different markets are inversely related to variables that reflect differences in the cost of consumer search effort in each market.
Prescription drugs are almost ideal examples of homogeneous commodities. A
patient can fill her prescription at any pharmacy she chooses and expect to obtain exactly
the same medicine. A priori
, therefore, it would seem that if any commodity obeyed the
law of one price then a prescription drug would be that commodity. That is not the case.
Using data from two townships in up-state New York, Sorenson (2001) provided
compelling evidence that the prices of individual drugs varied substantially among
pharmacies within each of the two communities. Sorenson reported price ranges for
some drugs where the highest reported price was more than 50 percent greater than the
lowest reported price. Further, within the two communities, the extent of price dispersion
differed among individual drugs in ways that were consistent with the predictions of
models that explain price dispersion on the basis of costly consumer search.
These findings are not unique. Other studies have examined the determinants of
price dispersion in the markets for gasoline (Adams 1997), water (Yoskowitz 2002),
automobiles (Dahlby and West 1986; and Goldberg and Verboven 2001), and grocery
products (Aalto-Setala 2003). They have also generally concluded that prices in those
markets vary in ways that cannot be accounted for solely by heterogeneity in product
attributes with respect to physical characteristics, space, time, or, in cross country studies,
government regulation. In addition, apparent violations of the law of one price have been
observed in commodity markets for agricultural products and raw materials that are
relatively homogeneous with respect to their physical characteristics (see, for example,
Goodwin et al). However, Sorenson’s study of pharmaceutical drugs is of particular
interest because he examines prices in two clearly defined markets for commodities that
Nevertheless, as Sorenson observes, his empirical analysis of the effects of search
costs is incomplete in at least one important respect. Sorenson’s data are for only two
markets that are also geographically adjacent (in fact, less than 30 miles apart). Thus, he
could not investigate the effects of differences between the two markets with respect to
the characteristics of the communities being served by the pharmacies in those markets.
Yet the costs of search for important groups of consumers within different markets may
be very different. For example, the proportions of both the elderly and the poor in a
population served by a given market may affect the amount of search in that market
because of differences in opportunity costs of time and expected benefits from search.
This paper re-examines price dispersion for pharmaceutical drugs in five
geographically isolated markets in Montana using data obtained from cross section
survey by the authors on the pricing of 75 different drugs by individual pharmacies. The
five markets are a minimum of eighty miles apart from one another and have distinctly
different demographic and other socio-economic characteristics. The new data set
permits a more extensive evaluation of the effects of search and search costs on price
Perhaps the most astonishing finding of this study, as Sorenson also reported, is
simply the degree to which the prices of individual drugs continue to vary among
pharmacies within a given market. Proportionally large variations occur for relatively
low priced drugs: for example, the difference in the highest and lowest prices for
alprazolam (0.5 mg) for a prescription of 20 tablets in the Kallispell market was $16.28,
about 135% of the average price of $12.21. Large absolute variations occur for relatively
expensive drugs: for example, the difference between the highest and lowest prices for
augmentin 875 (in prescriptions of 20 tablets) in the Billings market was $43.88 and the
average price was $121.13. A priori
, these differences seem unlikely to be driven only
by variations in pharmacy costs. Models of price dispersion based on costly search for
information by consumers, however, seem to provide plausible alternative explanations,
especially as many prescription drugs are not widely used and their prices are not
The major contribution of this paper is to show that differences in community
characteristics which serve as indicators of costs of search and search effort have
significant effects on the degree of price dispersion. As several models of price
dispersion predict, this study provides clear evidence that in markets where, for
consumers, costs of search are lower and expected benefits are higher, price dispersion
for pharmaceutical drugs is lower. In addition, the empirical findings provide additional
support for Sorenson’s conclusion that, within a given market, price dispersion is not
explained fully by differences in costs of production associated with differences in
pharmacy services, costs of supplying drugs, or location. The results also confirm that
individual drug characteristics that affect the returns from search also influence price
dispersion. In addition, variations among pharmacies with respect to quality of service
and costs of providing drugs do affect price dispersion in predictable ways; that is, the
more variation or heterogeneity there is among suppliers, the more variation there is
This study also addresses a related but different issue. Three of the cities included
in the analysis are relatively close to Canadian cities (within about a two hour drive).
Consumers in those communities could therefore have possibly acquired prescription
drugs directly from Canadian pharmacies where prices for many of those drugs were
apparently much lower and also available to U.S. citizens at the time of the survey (the
summer of 2004).1 Thus, we also examine whether proximity to Canada affects the level
and dispersion of drug prices within the five markets for which data are available.
Price dispersion may be observed in a market because of heterogeneity among
sellers with respect to production costs (Reinganaum, 1979) or product attributes
(Besancourt and Vranceau, 2004). In addition, however, as Stigler (1961) showed, price
dispersion may occur in a market because some or all consumers lack information about
product prices and must incur search costs in order to obtain that information. Several
subsequent models have extended Stigler’s insights. Salop and Stiglitz (1977), for
example, show that a two-price equilibrium can occur where increasing returns to scale
exist at the firm level and consumers have identical search costs. Burdett and Judge
(1983) present a model in which price dispersion occurs because consumers are
heterogeneous and therefore collect different amounts of price information. MacMillan
and Morgan (1988), in a dynamic context, find that that price dispersion can arise when
search costs differ among consumers and, in consequence, some consumers search more
1 Several new articles from 2004 reported that imports of Canadian drugs were a major concern for pharmacy manufacturers, consumers, and policy makers (New York Time, 2004). In 2005, the Canadian health minister, Uijal Dosanjh, avowed that Canada intended to ban the bulk export of prescription drugs (Associated Press, 2005).
intensively and become better informed than others. In addition, Wilder and Schwarz
(1979) obtain price dispersion from a model in which consumers differ with respect to
tastes and propensities for search. MacMinn (1980) and Carlson and MacAffee (1983)
also show that price dispersion equilibriums exist in models in which search costs differ
among consumers and production costs differ among firms.
The theoretical literature indicates that price dispersion in a market will be
influenced by several factors. These factors include sources of heterogeneity among
sellers with respect costs and the services they offer consumers. Price dispersion is also
likely to be lower for commodities when the expected benefits to consumers from search
are larger. Further, variables that affect the extent to which heterogeneous consumers in
the market place have incentives to search for information because of differences in
search costs are also important; that is, price dispersion is likely to be lower in markets
where more consumers have lower search costs. Thus, price dispersion is likely to be
associated with (a) seller heterogeneity, (b) product characteristics that affect the benefits
from search, and (c) consumer characteristics that affect the cost of search by consumers.
Measuring price dispersion requires price data from different sellers of the
commodities of interest in each market. In contrast to pharmacies in New York markets,
pharmacies in Montana are not required to post prices for drugs in a public location
within each place of business. Thus data on pharmacy prices for 75 different drugs were
obtained through a survey instrument.2 The survey instrument was administered to 58
2 Sorenson examined price dispersion in two towns in up-state New York where, at the time of his study, each pharmacy was required to post prices publicly in the store for the top 150 drugs in terms of state-wide
pharmacies in five geographically distinct markets over the period July 1, 2004 to August
15, 2004. A total of 33 completed surveys were completed and returned, yielding a
response rate of 57 percent. The five cities - Billings, Bozeman/Belgrade, Great Falls,
Kalispell, and Missoula - are a minimum of eighty miles apart from each other and no
other town with a pharmacy is closer than 20 miles away from these markets.3 In each of
these markets, price data were obtained from a minimum of five pharmacies for each of
The degree of price dispersion for each drug is likely to be affected by the
characteristics of that drug that affect consumer benefits from search, as well as the
availability of substitutes. These characteristics include a drug’s frequency of use. For
each drug, therefore, the typical dosage and duration of therapy were collected from
Mosby’s 2004 Drug Guide. In addition, Thompson Healthcare’s 2003 Drug Topics
Redbook was used to collect information on each drug’s average wholesale price and the
number of manufacturers and re-packagers of the drug.5 The United States National
prescriptions. Thus he was able to obtain data on prices for 150 separate pharmaceutical products from each outlet. In this study, data had to be obtained through a survey instrument that was filled out by each participating outlet. Seven pharmacies located in markets not included in the study were asked to pretest a preliminary survey instrument which included the top 150 most frequently prescribed drugs in 2003. These pharmacies expressed concerns about response burden. Thus, to lower the response burden for each pharmacist (which was still substantial) and increase response rates, the survey asked pharmacists to provide data on prices for 75 of those drugs.
3 Bozeman and Belgrade are ten miles apart and linked by an interstate highway (I 90). Average travel time from the center of Bozeman to Belgrade is less than 15 minutes. Further, Belgrade serves as a dormitory community for Bozeman and all major “big box” stores such as Walmart and supermarkets such as Safeway are located in Bozeman. In addition, all major health care facilities for the two communities and over 90 percent of physicians are located in Bozeman. Thus it is reasonable to view Bozeman and Belgrade as a single market for prescription drugs.
4 Six of 11 pharmacies surveyed responded in Billings, five of 11 responded in Bozeman/Belgrade, five of 11 responded in Great Falls, eight of 11 responded in Kalispell, and nine of 13 responded in Missoula. Additional price data were obtained from another eleven pharmacies in Montana located in other towns (Livingston, Butte, and a series of small towns along the most northerly major highway) but too few responses were received from these locations to construct measures of prices dispersion for those markets.
5 Thompson Healthcare publishes its estimate of the average wholesale price (AWP) for a drug based on values reported by the manufacturers, re-packagers, and private labelers. Thompson calculates the AWP using a markup specified by the manufacturer for wholesalers when such a markup is provided. If the
Library of Medicine and the National Institute of Health served as sources for
information on brand name drugs that have generic substitutes, the number of those
substitutes, brand name drugs that do not have substitutes, and the class of treatment for
Differences among pharmacies with respect to services may also affect price
dispersion. Respondents to the survey were asked to provide information on their
operations, including whether they offered a delivery service and whether it was free,
location with respect to hospitals, physicians, medical centers, and retirement
communities, availability of discounts for senior citizens, whether the pharmacy was
located in a grocery store or a department store, whether or not it was a member of a
chain or United Buying Group,7 and the hours the pharmacy was open for business. In
addition, the authors visited each pharmacy included in the study and obtained data on
the proportion of thirty eight commonly marketed non-prescription items sold by each
Price dispersion is also likely to be affected by specific market characteristics that
serve as indicators of the amount of consumer search in each market. Demographic
variables on the population proportions of the elderly (aged 65 and over) and the poor
(individuals below the federal poverty line) were collected for each of the five markets
from the 2001 Census (United States Census Bureau). Both of these groups have
relatively low opportunity costs of time and/or relatively high expected benefits from
manufacturer does not provide a recommended mark up, then the AWP is set equal to the wholesale acquisition cost estimated from the values reported by manufacturers, repackagers, and private labelers.
6 For example, some drugs in the sample served as beta blockers, some as anti-depressants, and others as anti-viral drugs. In this study, each of the 75 drugs is allocated to one of sixteen classes of treatment use.
7 Independent pharmacies may purchase drugs from wholesalers through a United Buying Group with other independent pharmacies to increase their bargaining power. Alternatively, they may simply purchase their drugs directly from wholesalers.
search. Data on the distance of each market from the closest competing market in Canada
Empirical Measures of Price Dispersion
Price dispersion for a commodity within a given market can be measured in
several different ways. Two obvious direct measures are the range and standard
deviation of the price of a drug obtained using the actual price data from the samples.
These measures can then be regressed on a comprehensive set of explanatory variables
that account for heterogeneity among sellers, drug characteristics that affect incentives
for search, and market characteristics that affect incentives for search.
The price dispersion measures obtained directly from the price data indicate that,
within individual markets, prices for pharmaceutical drugs vary substantially. Table 1
presents illustrative values for these two measures of price dispersion and, in addition, the
coefficient of variation in each of the five markets for six of the 75 drugs included in the
study. Among these six drugs, on a per prescription basis, four were relatively
inexpensive, with average prices in the five markets ranging from $7.91 to $13.58, and
two were relatively expensive, with average prices in the five markets ranging from
Within each market, prices for some of these drugs exhibited considerable
variation. For example, among the low priced drugs in table 1, for alprazolam (0.5MG),
the average price in the Billings market was $12.80 but the difference between the
highest and lowest price was $21.34 and the coefficient of variation was 64%. In
contrast, other drugs with similar average prices in the same market may exhibit much
less price variation. For example, in the Billings market, the average price of acetamine
codeine #3 was $9.80, the difference between the highest and lowest price was only
$1.74, and the coefficient of variation was 6%. Further, the same drug may exhibit much
more price variability in one market than in another. For example, the coefficient of
variation for the price of acetemine codeine ranged from 5% in the Billings market to
Finally, while the data on coefficients of variation in Table 1 suggest that high
price drugs exhibit less variation than low priced drugs,8 the price range for a high price
drug can be substantial, even when the coefficient of variation is relatively small. For
example, the coefficient of variation for augmentin (873 mg) ranged from 5% in the
Bozeman market to 15% in the Great Falls market. However, the price range for that
drug was $43.88 in the Billings market and exceeded $16 in all five markets.
Measures of price dispersion obtained directly from the sample data reflect the
effects of all sources of price variation among pharmacies. In addition to factors that
affect incentives for search, these include pharmacy specific effects that arise from
differences in costs associated with heterogeneous services, variations in the wholesale
costs incurred by different pharmacies, and effects arising from variations in competition
in the market for specific drugs. To account for differences among suppliers with respect
to services, econometric models can be estimated to predict the level of the price of each
drug in each pharmacy in terms of heterogeneous pharmacy characteristics, wholesale
costs of each drug, and other variables unrelated to search. The residuals for the price
8 The coefficients of variation for the two high priced drugs range from 5% for Allegra-D in Billings to 15% for Augmentin (875 mg) in Great Falls. Coefficients of variations for the four lower priced drugs range from 6% for acetemine codeine #3 in Billings to 64% for alprazolam, also in the Billings market, and in all but 3 of 16 cases exceed 20% and in all but one case exceed 15%.
level model can therefore be regarded as purged of pharmacy and other sources of
heterogeneity effects and used to construct range and standard deviation measures of
price variability for each drug that represent the variation that cannot be explained by
pharmacy heterogeneity. These measures can then be regressed on drug characteristics
and market characteristics that affect incentives for search to examine the effects of
search on price dispersion. Here, following Sorenson, we examine the behavior both of
measures of price dispersion obtained directly from the price data and of measures of
price dispersion obtained from the residuals of price determination models.9
To account for the effects of drug specific, pharmacy and other potential sources
of heterogeneity, price levels for individual drugs are regressed on a set of drug
characteristic, pharmacy characteristic, and other variables that affect price levels.
Parameter estimates for two sets of price level models are reported in Tables 2(a) and
2(b). The two sets of models differ only in that the second set of models includes sixteen
drug treatment class variables that identify the types of ailments for which each drug is
used. In both Tables 2(a) and 2(b), model 1 includes only drug characteristics, model 2
includes drug and pharmacy characteristics, model 3 includes drug and pharmacy
variables and two interaction variables between drug and pharmacy characteristics, and
model 4 includes three additional market characteristic variables as well as all of the
variables incorporated in model 3. Model parameters were estimated using the OLS
procedures in Stata. Cook and Weisberg (1983) tests indicated the presence of
heteroskedesticity in the error terms. Robust standard errors were therefore estimated
9 Sorenson also examines price dispersion measures obtained directly from the data and price dispersion measures obtained from the residuals of empirical models that explain price levels in terms of variables that account for pharmacy heterogeneity.
Twenty three characteristic variables were included in models 2-4 in tables 2(a)
and 2(b). They include the total number of hours a pharmacy is opened, two dummy
variables indicating whether the pharmacy is open on Saturdays and Sundays, seven
dummy variables to account for the use of eight different primary wholesalers, five
dummy variables to account for the use of six different secondary wholesalers, two
dummy variables to account for the availability of any delivery services and free delivery,
and five variables to account for other differences among pharmacies in their business
operations (whether or not the pharmacy is a member of a chain or buying group, located
in a grocery or big box department store, near a hospital,10 or as a stand alone operation,
and provides a wide or limited range of non-pharmaceutical products).
Individual parameter estimates are not reported for these 23 pharmacy
characteristics variables in Tables 2(a) and 2(b). However, the general pattern of results
is as follows. Drug price levels appear to be lower in pharmacies located within a mile of
a hospital, higher at pharmacies that open on Saturdays, and lower at pharmacies that
open on Sundays (the coefficients for these variables are consistently statistically
significant at the ten percent level). The pharmacy’s choice of primary and secondary
wholesaler may also affect price levels.11 Other pharmacy characteristics such as
delivery services and the scope of non-pharmacy products offered affect drug price levels
do not appear to have any systematic effects on pharmaceutical drug pricing decisions.
10 This variable has a value of 1 if the pharmacy is located within a mile of a hospital and zero otherwise. Pharmacies located in hospitals were not included in the survey because many are owned and operated by the hospital and pricing structures in those operations may be linked to pricing structures for the other services that hospitals provide, in part because of the captive nature of many hospital patients.
11Coefficients for the dummy variables for three primary wholesale sources and two secondary wholesale sources are statistically significant from zero in many of the price level models.
Each empirical price level model includes the following drug characteristics
variables. Indicator (zero-one) variables account for whether the drug is sold under a
brand name (Brand
) and whether the drug is a brand name that faces competition from
generic drugs (Brand Substitutes
). The variable Substitutes
identifies the number of drugs
that provide alternative therapies for condition treated by the drug of interest,
measures the number of manufacturers producing the drug, and the
variable Average Cost
measures the drug’s estimated average wholesale cost to
The degree of competition in the market for a specific drug is likely to inversely
affect the price at which the drug is sold. Thus, if the drug is a brand name drug (Brand
equals one), then its price is expected to be higher. However, if that branded drug faces
competition from generic substitutes (Brand Substitutes
equals one), the price is expected
to be lower. The results presented in Tables 2(a) and 2(b) support this hypothesis. In all
eight models, the coefficient for the variable Brand
is positive while the coefficient for
the variable Brand Substitutes
is negative and both are statistically significant at the one
Similarly, as competition increases from other drugs that are substitutes in
treatment, the price of a drug is expected to be lower. The parameter estimates for the
reported in Table 2, which are negative and significant at the one
percent level, are also consistent with this hypothesis. As the number of manufacturers
providing the same drug increases and competition among them increases, the price of
that drug might be expected to fall. However, an increase in the number of manufacturers
may result in differences in production costs. These cost differences among
manufacturers my result in differences in wholesale prices that are passed through to the
retail level. The parameter estimates for the variable Number of Manufacturers
in Table 2 are positive and significant at the one percent level and, therefore, suggest that
latter influence may be more important than competition effects.
The retail price of a drug is expected to increase as its average cost to a pharmacy
increases. As expected, the parameter estimates reported in Table 2 for the variable
are also positive and statistically significant at the one percent level.12
Market characteristics are also included in price level model 4 in Tables 2 (a) and
2 (b). One measure of competition among sellers within a market is the number of sellers
per thousand people in that market. Parameter estimates for the variable Pharmacies per
are negative but not statistically significant.
Markets in Canada represent another potential source competition for the
pharmacies included in this study. The variable Distance to Canada
, included in the
regressions with both linear and quadratic terms,13 measures the distance from each of the
five markets in Montana to the closest Canadian market. The variable is hypothesized to
have a net positive effect on the price level for each drug; that is, drug prices may be
lower in markets closer to Canadian than in markets more distant from Canada because of
potential competition from Canadian pharmacies. While the parameters for both the
12The Average Cost
variable was contructed as follows. In 2004, Montana’s Meducaid reimbursements to pharmacies were the equivalent of 85 percent of a drug’s average wholesale price (AWP), considered to be the cost of obtaining the drug, plus a fixed dispensing fee of $4.70 per prescription. Often contracts between insurance companies and pharmacies use formulas similar to those established by Medicaid. Thus the variable Average Cost
, which serves as a proxy for the average cost to a pharmacy of acquiring a drug, is measured as the sum of the drug’s AWP and the Medicaid prescription fee of $4.70. The information on links between drug cost reimbursements for pharmacies in Montana under the Medicaid and insurance contracts was provided to the authors through personal communications from two practicing pharmacists who requested that they remain anonymous.
13 A quadratic term for distance from Canada is included to allow for the possibility that the marginal effect of distance diminishes as total distance from Canada increases.
linear and squared terms for this variable are positive, they are not statistically significant
and p-values are extremely small. Thus there is no evidence that proximity to Canada
had any effect on the level of drug prices in Montana markets in the summer of 2004.
One possible explanation for this result is that transportation costs (including travel time)
are generally quite large relative to expected benefits from traveling to Canada to obtain
drugs, even for expensive drugs. For example, the shortest distance between a Montana
and Canadian market was over seventy miles and would involve at least a two and half
hour round trip journey. In addition, in the summer of 2004, some Canadian companies
were marketing drugs to U.S. customers via the internet and postage costs were relatively
Finally, interaction effects between the pharmacy characteristic of belonging to a
chain of stores and two drug characteristics are considered. The variable Chain_Brand
equals one if the pharmacy is a member of a chain and the drug is a patented brand.
Similarly, the variable Chain_Brand_Substitutes
equals one if the pharmacy is a chain
and the drug is a brand that faces competition from generic substitutes. These variables
are included to account for the possibility that chains may have different mark-up rates
for patented brands, brands with generic competitors, and generic drugs. However,
parameter estimates for these interaction terms are not statistically significant.
The price level models explain some of the observed within-sample differences in
prices. Unadjusted coefficients of determination (R-squared) are about 0.80 for the
models presented in Table 2(a) that exclude the sixteen treatment class variables and
about 0.91 for the models presented in Table 2(b) that include them. However, a
considerable amount of variation remains to be explained. Among the 75 drugs, the
average estimated standard deviation of the price of a drug within a given market
obtained directly from the price data obtained in the survey is $5.85 and the average
range or difference between the highest and lowest price drug is $15.84. The average
estimated standard deviation for all 75 drugs based on the residuals obtained from the
pharmacy-specific forecasts of the most comprehensive price level model (model 4 in
Table 2(b)) is $5.27 and the average range is $14.15. The price level model which
accounts for pharmacy, manufacturer, and other sources of heterogeneity therefore
reduces both the average standard deviation and range estimates of price dispersion
within the five markets by about ten percent.
Models of Price Dispersion
Price dispersion within a market for a specific drug is likely to be a function of
physical drug characteristics that affect incentives for search, characteristics of the supply
side of the market that affect competition, and characteristics of consumers in the market
that affect incentives for search. Each of the four measures of price dispersion is
assumed potentially to be a function of drug characteristics, supply side characteristics,
The duration of a drug’s therapy affects the benefits consumers expect to obtain
from search for low prices. Returns to search will generally be greater for drugs whose
prescriptions have to be filled several times or even many times to complete a therapy
than for drugs whose prescriptions have to be filled only once or twice. Thus price
dispersion is assumed to depend on annual frequency of use as measured by the variable
.14 Similarly, drugs that on average are more expensive are assumed
to offer greater returns from consumer search efforts. However, given that the measures
of price dispersion used in this study are ranges and standard deviations, including an
indicator of average cost in the price dispersion model may serve to normalize for
differences in price levels. The estimated average cost to a pharmacy of obtaining a drug,
, is used as a proxy for the average price of the drug to a consumer and price
dispersion may either be positively or negatively associated with this variable.
The amount of consumer search is a function of both expected search benefits and
expected search costs. The opportunity costs of search time are likely to be lower for the
elderly, defined as individuals over 65, who are likely to be retired and members of low
income households. Thus, in communities with larger proportions of the elderly and
households in poverty, search effort is likely to be greater in the market for
pharmaceutical drugs. In the case of the elderly, an additional factor may be important.
On average, individuals over 65 take more medications than other consumers (HHS
Weekly Report). Thus the benefits from any give level of search are likely to be greater
for that group because search is being carried out over multiple commodities and
economies of scope are likely to be achieved. To account for these potential effects, the
two variables Percent 65
and Percent Poverty
are included in the price dispersion
14 Purchase frequency is measured as follows. Each drug’s estimated typical dosage per day is multiplied by the typical length of treatment (where both are reported by Moseby), which is then divided by the typical quantity provided in a prescription. This approach provides an estimate the number of times in a year the typical patient has to purchase the drug. For example, Atenolol is a beta-blocker that is typically taken once a day for the rest of a person’s life. The prescription in this survey is for a quantity of 30 pills. Thus the estimated annual frequency of purchase for Atenolol is 12.17 (= 1* 365/ 30). Drugs with an estimated annual purchase frequency of less than one are allocated a purchase frequency of one, because an individual must fill a prescription for a drug at least once in order to use it.
models. Price dispersion within each market is expected to be inversely related to both of
Characteristics associated with competition from other drugs are also likely to
affect price dispersion. The number of manufacturers of a single drug may increase price
dispersion because of differences in production costs among the competing firms that
results in different wholesale prices. Given that pharmacies tend to purchase their drugs
from one wholesaler and that wholesalers often obtain a given drug from a single
manufacturer, these cost differences are on potential source of price dispersion. By the
same token, a patented branded drug produced by a single manufacturer is likely to
exhibit less price dispersion among pharmacies because there is only one manufacturer
from which wholesalers can purchase the drug and, absent price discrimination among
wholesalers by the manufacturer, only one wholesale price. Thus price dispersion is
likely to be directly related to the variable Manufacturer
and inversely related to be
A branded drug that faces competition from generic substitutes may exhibit less
price dispersion than one that does not, because competition from generics reduces the
incidence of high prices by increasing the probability that a low price is observed for any
given level of consumer search. An increase in the number of generic substitutes is likely
to further increase the probability that a given amount of search will result in the
observation of a low price. Thus the frequency of high prices is likely to diminish and
price dispersion to decrease. Thus price dispersion is likely to be inversely related to the
variables Brand Substitutes
Finally, potential competition from other markets in which prices for all drugs are
known to be lower may affect price dispersion in a given market by lowering prices at the
high end of the distribution. To examine the potential for this effect, the variable
Distance to Canada
is included in the empirical models. This hypothesis implies that
markets closer to Canada should experience less price dispersion and therefore that price
dispersion should be inversely related to this variable.
Price Dispersion Estimation Models and Results
The effects of drug characteristics and market characteristics on price dispersion
are examined using four different measures of price dispersion with each market for each
drug. These four measures are (a) the range and (b) the standard deviation for each of the
75 drugs computed directly using the prices obtained from the pharmacy survey, and (c)
the range and (d) the standard deviation for each drug estimated using the residuals of the
forecasts of prices obtained from the price level models presented in Table 2(a) and 2(b).
Table 3 presents parameter estimates for models in which the 16 different treatment class
dummy variables are omitted where the measures of price dispersion based on residuals
from price level models are obtained from price level model 4 in Table 2(a), from which
the sixteen different treatment class dummy variables were also omitted. Table 4
presents results that include residual measures of price dispersion obtained from price
level model 4 in Table 2(b) that included dummy variables for the sixteen different
classes of treatment. The price dispersion models for which results are presented in
Table 4 also include those sixteen treatment class variables.15
15 A total of 375 observations (75 drugs multiplied by five markets) are used to estimate the price dispersion models for which results are presented in tables 3 and 4.
In Tables 3 and 4, two regression models are estimated to explain variations in
each of the four price dispersion measures. The first model (model 1) includes only drug
characteristic variables: Purchase Frequency, Brand, Brand Substitutes, Substitutes,
and Average Cost
. The second model (model 2) includes the three
market characteristic variables: Percent Poverty
, and Distance to Canada
The central concern of this study is the role of search costs in determining price
dispersion. In all eight of the models presented in Table 3, regardless of which price
dispersion measure is used, the parameter estimate for the variable Purchase Frequency
is negative and statistically significant at the 1 percent confidence level. This result is
entirely consistent with the hypothesis that higher expected benefits from search increase
search effort and reduce price dispersion. In Table 4, in which the estimated models
include indicator variables for sixteen treatment classes, the parameter estimates for
are positive but not statistically significant different from zero.
These results indicate that frequency of purchase is highly correlated with class of
treatment, a conclusion that accords with medical common sense. Drugs needed to treat
similar illnesses (for example, beta blockers or pain killers) are likely to be prescribed in
similar doses and with similar treatment protocols.
In all sixteen models presented in Tables 3 and 4, price dispersion is an increasing
function of Average Cost
. Parameter estimates are consistently positive and statistically
significant at the 1 percent level. The dependent variables in these models involve
absolute measures of price dispersion (ranges and standard deviations). These results
indicate that, ceteris paribus, as the average price of a drug increases, the absolute amount
of price dispersion increases. However, in each of these models, the ratio of the measure
of price dispersion divided by average cost declines. Thus, to the extent that average cost
is a proxy for average price, the results indicate that relative measures of price dispersion
such as the coefficient of variation decreases as average prices increase.16 This finding is
also consistent with the hypothesis that as benefits from search increase relative price
dispersion decreases, even though absolute price dispersion increases with average cost.
Other drug characteristics may also be important. In all sixteen models presented
in Tables 3 and 4, the estimated coefficients for the variable Brand
negative and in 12 of those models, are statistically significant at the ten percent level or
better. These results are consistent with the hypothesis that branded drugs produced by
only one manufacturer exhibit less price dispersion because production costs do not vary
as they could if more than one manufacturer made a drug. The parameter estimates for
the variable Manufacturers
are positive in all 16 models presented in Tables 3 and 4 and
statistically significant at the ten percent level or better in six of the models presented in
Table 4. These results provide some additional support for the hypotheses that variations
in costs among manufacturers are a source of price dispersion among retailers.
In the models presented in Table 3, which exclude the 16treatment class dummy
variables, parameter estimates for the variable Brand Substitutes
are not statistically
significant from zero. In the models presented in Table 4, which include the treatment
class variables, the parameter estimates for this variable are positive and statistically
significant. These latter findings are not consistent with the hypothesis that where
branded drugs face competition from generic substitutes, less price dispersion is
16 Assuming values for all other explanatory variables are fixed, the measure of price dispersion, y, is simply a linear function of the proxy for average price, x; that is, y = α + β x. Hence y/x = (α/x) + β. Given that α > 0, as x increases α/x decreases and thus y/x decreases. In all of the models presented in Table 3, α is strictly positive when variables are set at their mean levels or, in the case of dummy variables, equal to unity.
observed. They suggest that the advent of generic substitutes simply increases price
variability, perhaps because the markets for the branded drug and the generics becomes
segmented between consumers who are concerned about the quality of generics and those
who are not. In contrast, parameter estimates for the variable Substitutes
negative and, in twelve of the 16 models presented in Tables 3 and 4, statistically
significant. This result is consistent with the hypothesis that when a drug faces increased
competition from other drugs that provide competing therapies then prices dispersion
Search effort is related both to the benefits consumers expect to receive and the
search costs they incur. Two market specific indicators of search costs, Percent Poverty
, are included in model 2 but not in model 1. The elderly and the poor are
assumed to have relatively low opportunity costs of time. In addition, many of the
elderly are likely to obtain larger benefits from search because they have to obtain
multiple prescriptions on a regular basis. Further, low income households are less likely
to be insured and therefore more likely to pay the full price of a prescription drug. Thus,
they also may obtain relatively large benefits from search. The parameter estimates for
both these variable presented in Tables 3 and 4 are uniformly negative and in four of the
eight models for which results are reported (the models that explain the standard
deviation measures of price dispersion) are statistically significant at either the ten
percent or five percent confidence level. Thus, the results are consistent with the
hypothesis that price dispersion is lower in markets where costs of search are low and
expected benefits relatively substantial for larger proportions of the population.
Finally, parameter estimates for the variable Canada
are negative and statistically
significant at the ten or five percent level in three of the eight models in which it is
included. Distance from Canada is expected to be directly related to price dispersion as
markets closer to Canada may have faced competition from Canadian markets in which
pharmaceutical drugs are generally less expensive. Thus this variable’s estimated
parameters are inconsistent with this hypothesis. However, on balance the evidence
suggests that distance from Canada is not a determinant of price dispersion for
The law of one price implies that commodities with the same physical attributes
should be offered at the same price by different sellers in the same market under the same
conditions of sale. This is often not true even when, as is the case with pharmaceutical
drugs, the commodities offered for sale in a given market have identical physical
characteristics. This paper has reexamined the law of one price and price dispersion for
pharmaceutical drugs in five geographically isolated markets more than eighty miles
apart from one another in Montana, the fourth largest of the lower 48 states, in which
there are less than 1 million people. Price data for 75 drugs were obtained from a “point
in time” survey of 33 pharmacies in the five communities and four measures of price
dispersion for each drug in each market were constructed. Two of the four measures
consisted of the range and standard deviation for each drug in each market estimated
directly using the price data obtained from the survey. The other two measures consisted
of the range and standard deviation for each drug estimated using the residuals from
models of price determination that accounted for differences in pharmacy services.
Prices for the same drug varied substantially among pharmacies in each of the
five markets in ways that could not be explained simply by differences in pharmacy
services. However, price dispersion is lower for a drug when the expected benefits from
search are greater. Price dispersion is lower for drugs that are purchased more frequently
and there is some evidence that relative price dispersion (as measured by the ratio of the
range or standard deviation to the average cost of a drug) is lower for drugs with higher
prices. These results confirm the findings reported by Sorenson in his study of
pharmaceutical drug prices in two communities in up-state New York.
In addition, in the five Montana markets examined in this study, price dispersion
is lower in markets where the proportions of the population that elderly and in poverty
and that are elderly are relatively large. The elderly and the poor both have relatively low
opportunity costs of time, relatively large expected benefits from search, and, therefore,
relatively high levels of search effort. This is a new finding that provides additional
empirical support for the hypothesis that price dispersion is affected by consumer search
effort, which in turn is influenced by economic incentives.
This study also provides evidence that some price dispersion results from
heterogeneity among pharmacies, and the extent of competition among manufacturers
and from substitutes in the market for a prescription drug. These results are generally
consistent with the findings of previous studies.
Finally, this study provides little or no evidence that access to pharmaceuticals
from Canadian markets had any measurable effects on either prescription price levels or
price dispersion in 2004. This is worth noting in light of the extensive media attention
given to the issue of the availability of much lower priced drugs in Canada during 2004
and early 2005. One reason for the finding may have been that during that time many
U.S. consumers could obtain prescription drugs from Canadian outlets through the
internet and, therefore, geographic proximity to Canadian markets was unimportant.
Another may be that for most consumers in the markets examined in this study, travel
costs exceeded any potential savings from buying prescription drugs in Canadian
markets, even though such gains appeared to be quite substantial for some drugs.
Aalto-Setala, Ville, “Explaining Price Dispersion for Homogeneous Grocery Products.”
Journal of Agricultural and Food Industrial Organization. 1 (2003).
Adams, A. Frank, III, “Search Costs and Price Dispersion in a Localized, Homogeneous
Products Market: Some Empirical Evidence.” Review of Industrial Organization. 12 (1997): 801-808.
Besancenot, Damien, and Radu Vranceanu. “Quality and Price Dispersion in an
Equilibrium Search Model.” Journal of Economics and Business. 56.2 (Mar.-Apr. 2004): 99-116.
Burdett, Kenneth and Kenneth Judd, “Equilibrium Price Dispersion.” Econometrica. 51
Carlson, John A., and R. Preston McAfee, “Discrete Equilibrium Price Dispersion.” The
Journal of Political Economy. 91 (June 1983): 480-493.
Cook, R. D. and S. Weisberg. “Diagnostics for heteroscedasticity in regression.”
Dahlby, Bev and Douglas West, “Price Dispersion in an Automobile Insurance Market.”
The Journal of Political Economy. 94 (Apr. 1986): 418-438.
Goldberg, Pinelopi and Frank Verboven, “The Evolution of Price Dispersion in the
European Car Market.” The Review of Economic Studies. 68 (Oct. 2001): 811-848.
Goodwin, Barry K., Thomas J. Grennes, and Michael K. Wohlgenant. "Testing the Law
of One Price When Trade Takes Time." Journal of International Money and Finance, 9(1990):21-40.
HHS Weekly Report, “Science in the News: Almost Half of Americans Use at Least One
Prescription Drug Annual Report on Nation’s Health Shows.” Dec. 5, 2004. http://www.hhs.gov/news/newsletter/weekly/archive/5dec04.html (accessed Jan. 2005).
MacMinn, Richard D., “Search and Market Equilibrium.” The Journal of Political
McMillan, John and Peter Morgan, “Price Dispersion, Price Flexibility, and Repeated
Purchasing.” The Canadian Journal of Economics. 21 (Nov. 1988): 883-902.
Reinganum, Jennifer, “A Simple Model of Equilibrium Price Dispersion.” The Journal
of Political Economy. 87 (Aug. 1979): 851-858.
Salop, Steven and George Stiglitz, “Bargains and Ripoffs: A Model of Monopolistically
Competitive Rice Dispersion.” The Review of Economic Studies. 44 (Oct. 1977): 493-510.
Sorensen, Alan, “Equilibrium Price Dispersion in Retail Markets for Prescription Drugs.”
The Journal of Political Economy. 108 (Aug. 2000): 833-850.
Stigler, George, “The Economics of Information.” The Journal of Political Economy.
United States Census Bureau, “American FactFinder.” Available at
http://factfinder.census.gov/ (accessed August 2004).
United States National Library of Medicine and the National Institute of Health,
“Medline Plus.” Available at http://www.nlm.nih.gov/medlineplus/druginformation.html (accessed Oct. 2004).
Wilde, Louis and Alan Schwartz, “Equilibrium Comparison Shopping.” The Review of
Economic Studies. 46 (July 1979): 543-553.
Yahoo!® Maps. Available at http://maps.yahoo.com/dd (accessed August 2004).
Yoskowitz, David W., “Price Dispersion and Price Discrimination: Empirical Evidence
from a Spot Market for Water.” Review of Industrial Organization. 20 (2002): 283-289.
2003 Drug Topics® Redbook. Thomson PDR. 2003. 2004 Drug Guide. Mosby, Inc. 2004.
Table 1: Price Levels and Measures of Price Dispersion for
Selected Drugs in Five Montana Markets
Allegra –D Alprazolam
Tabs (20 caps)
0.5 Mg (30
Table 2(a). Price Level Models Excluding Class VariablesA
and Drug &
Drug & Pharmacy,
and City EffectsB
A Robust standard errors are presented in parentheses below each parameter estimate. The superscripts ***, **, * indicate that parameter estimates are statistically significant at the 1%, 5%, and 10% confidence levels, respectively. B Models 2, 3 and 4 include a set of 23 pharmacy characteristic variables for which parameter estimates are not reported
Table 2(b). Price Level Models Including Treatment Class VariablesA
and Drug &
Drug & Pharmacy,
and City EffectsB
A Robust standard errors are presented in parentheses below each parameter estimate. The superscripts ***, **, * indicate that parameter estimates are significant at the 1%, 5%, and 10% confidence levels, respectively. The models presented in Table 2(b) also include dummy variables for sixteen different treatment categories of drugs. Parameter estimates for these treatment class variables are not reported.
B Models 2, 3 and 4 include a set of 23 pharmacy characteristic variables for which parameter estimates are also not reported.
Table 3. Price Dispersion Models (excluding treatment class variables).A
Price Dispersion Model (1)
Price Dispersion Model (2)
(0.1130) (0.0400) (0.1074) (0.0385) (0.1129) (0.0398) (0.1068) (0.0380)
(1.3970) (0.4953) (1.3196) (0.4717) (1.3899) (0.4930) (1.3146) (0.4706)
(1.6051) (0.5489) (1.4937) (0.5422) (1.6055) (0.5499) (1.5000) (0.5457)
(0.0150) (0.0055) (.0129) (0.0046) (0.0149) (0.0054) (0.0127) (0.0044)
0.1014*** 0.0379*** 0.0846*** 0.0327*** 0.1014*** 0.0379***
(0.0175) (0.0063) (0.0164) (0.0062) (0.0174) (0.0063) (0.0164) (0.0062)
(0.1820) (0.0646) (0.1738) (0.0603) (0.1831) (0.0637) (0.1713) (0.0583)
13.0994*** 4.7244*** 12.7267*** 4.6549*** 27.9347
(1.7926) (0.6531) (1.5908) (0.5673) (31.2445) (12.0775) (31.4233) (11.9855)
0.2587 0.2738 0.1843 0.2025 0.2659 0.2874
A Robust standard errors are presented in parenthesis. The superscripts ***, **, * indicate that parameters are statistically significant at the 1%, 5%, and 10% confidence levels, respectively.
Table 4. Price Dispersion Models (including treatment class variables).A
Price Dispersion Model (3)
Price Dispersion Model (4)
(0.1368) (0.0484) (0.1369) (0.0448) (0.1360) (0.0479) (0.1358) (0.04404)
-2.1889*** -3.0204* -1.2249** -4.9638*** -2.1889***
(1.5990) (0.5740) (1.6786) (0.6114) (1.5901) (0.5682) (1.6462) (0.5968)
4.5625*** 1.6907*** 5.8269*** 2.3231*** 4.5625*** 1.6907*** 5.8269*** 2.3231***
(1.6789) (0.5638) (1.8232) (0.6456) (1.7092) (0.5662) (1.7834) (0.6240)
(0.0157) (0.0055) (0.0167) (0.0057) (0.0155) (0.0055) (0.0164) (0.0057)
0.1855*** 0.0673*** 0.1596*** 0.0603*** 0.1855*** 0.0673*** 0.1596*** 0.0603***
(0.0220) (0.0077) (0.0232) (0.0081) (0.0221) (0.0076) (0.0224) (0.0078)
0.3046* 0.1071* 0.2192 0.07754 0.3046* 0.1071* 0.2192 0.07754
(0.1744) (0.0621) (0.1885) (0.0654) (0.1750) (0.0608) (0.1857) (0.0632)
(6.2167) (2.2092) (5.0803) (1.4729) (27.0557)
0.4145 0.4302 0.3536 0.3738 0.4218 0.4438 0.3659 0.3923
A Robust standard errors are presented in parenthesis. The superscripts ***, **, * indicate statistical significance at the 1%, 5%, and 10% confidence levels, respectively. These models also include sixteen treatment class dummy variables for which parameter estimates are not reported.
Surgical Biology for the Clinician Biologie chirurgicale pour le clinicien Thromboprophylaxis in surgical patients Martin O’Donnell, MB; Jeffrey I. Weitz, MDVenous thromboembolism is the most common preventable cause of death in surgical patients. Throm-boprophylaxis, using mechanical methods to promote venous outflow from the legs and antithromboticdrugs, provides the most effective m
Pregnancy Testing Schedule & Checklist First and Second Prenatal Visit (6-12 weeks) o History and physical o Pap, gonorrhea and chlamydia screening o Confirmation ultrasound o Review medical and genetic history of both parents. o Prenatal labs: CBC, RPR, HIV, HepB, rubella, ABO Rh and urine culture o Schedule first-trimester screening at 12 weeks (if desired). o Start prenatal vitamins