Strategic Advertising: Evidence in the Pharmaceutical

Industry∗

YUTIAN CHEN† WEI TAN‡

May 24, 2011

Abstract

We investigate the change in advertising patterns of competing pharmaceutical firms in the short

period after a new entrant, Crestor, is petitioned to be removed from the Statin drug market. Us-

ing physician-level data, we find that Crestor’s main competitor, Lipitor, increases its advertising

intensity the most among physicians with whom Crestor already has a medium share of total pre-

scriptions. Moreover, such change in advertising patterns does not occur with Crestor’s other

smaller opponents. These findings support the view that Lipitor strategically advertises in order

to maximize its long run profit, while raise doubt on the hypothesis that Lipitor’s advertising is a

normal competitive response to the bad news.

Keywords: Strategic advertising, Predation, Exit

JEL Classification: D43, L11, L13

∗We would like to thank Steven Berry, Hugo Benitez-Silva, Sandro Brusco, Judy Chavalier, Iain Cockburn, Pradeep Dubey,

Sara Ellison, Avi Goldfarb, Ginger Jin, Fiona Morton, Marc Rysman, Matthew Shum, Konstantinos Serfes, Yair Tauman for helpful

discussions and comments. We also appreciate the suggestions from seminar participants at Yale University, Queens College, Drexel

University, 2007 International Industrial Organization Society meetings, 2007 Econometric Society Summer meetings and 2008 AEA

meetings.†Department of Economics, California State University, Long Beach, CA 90840, USA. Email: ychen7@csulb.edu‡Department of Economics, SUNY-Stony Brook, NY 11794, USA. Email: wtan@notes.cc.sunysb.edu

1

1 Introduction

The strategic usage of advertising has been thoroughly studied in theoretical works, where sev-

eral rationales have been identified for strategic advertising to affect firms’ entry/exit decisions.

However, the empirical study on strategic advertising is just getting underway (Bagwell (2007)).

In this work, we analyze the change in advertising patterns of incumbent firms after a new entrant

was petitioned to be removed from the market. We find evidence supportive to the argument that

the dominant incumbent firm strategically advertised after this bad news to the new entrant, and

its intent is to induce the new entrant’s “withdrawal” thereby maximizing its long run profit.

Our investigation is based on a high-profile case in the Statin drug market. Crestor, a brand that

belongs to a therapeutic class of Statin drug, was approved by FDA in August 2003 and entered the

Statin drug market in late 2003. Shortly after its entry, several cases of severe side effects occurred

from the use of this brand. Alarmed by the potential risk, on March 4, 2004, Public Citizen, a very

influential consumer advocacy group, submitted a petition to the Food and Drug Administration

(FDA) to formally remove the brand from the market. This bad news was immediately picked

up by major news networks such as CBS. Interestingly, a different Statin brand, Baycol, was also

petitioned by Public Citizen in 2001. As a consequence of a similar bad news, Baycol withdrew

from the market.

In the period immediately following the bad news of Crestor, all Statin drugs increased their

physician detailing level.1 Among Crestor’s opponents, Lipitor is the dominant brand accounting

for more than 40 percent of the total sale of Statin drug, and is Crestor’s major rival. After the bad

news on Crestor, Lipitor increased its detailing visits by 12 percent.

The increase in the detailing visits by Crestor’s competitors gives rise to a number of research

questions. Since the bad news negatively affects the market demand of Crestor, is the post-news

detailing of Crestor’s competitors a normal competitive response to the news event, namely, a1Detailing is the major form of advertising in the pharmaceutical industry, referring to the visit of pharmaceutical

sales representatives at physicians’ offices. It is commonly viewed as a means of relationship building with a particular

physician. Throughout the work, the advertising of any Statin drug refers to the detailing visits to physicians’ offices

by the Statin drug. For more details on detailing, see Section 3.1.

2

short run profit maximizing behavior holding the market structure unchanged? Or, is there any

strategic consideration underlying their detailing visits such that they are exploiting the bad news

by maximizing their long run profits? In particular, since the bad news does not generate the

pressure for Crestor to fully pull out from the Statin drug market, Crestor’s competitors may

strategically advertise (through detailing) to induce Crestor to cease relationship building with

certain group of physicians, so that they can benefit in the long run when Crestor takes a less

aggressive stance in their competition. Such a strategic intent in essence is “predatory” if each

physician is treated as a “physician market” and Crestor’s ceasing of relationship building with a

physician is treated as the “exit” of Crestor from that physician market.2

If the above “predatory” incentive exists in the post-news advertising of Crestor’s competitors,

they may practice predation only with certain group of physicians. The intuition is articulated

as below. While the bad news negatively affects physicians’ willingness to prescribe Crestor, the

profitability for its competitors to predate varies across physicians. For physicians whose demand

of Crestor is high prior to the bad news, the post-news demand is likely to be large enough for

Crestor to stay in profit through relationship building, making predation impossible. On the other

hand, for physicians whose demand of Crestor is low prior to the bad news, the post-news demand

is likely to be so low that Crestor will naturally give up expensive relationship building, making

costly predation unnecessary. Therefore, the strongest incentive for Crestor’s competitors to prey

shall arise among physicians whose prior demand of Crestor lies in the medium range.

Since Lipitor is the dominant brand in the Statin market and also Crestor’s major rival, we

focus on Lipitor’s advertising behavior to detect strategic considerations in the post-news adver-

tising. The data set we use is a physician-level panel data set covering the periods preceding and

following the bad news. For each physician, we observe the prescription level and the number of

sales representatives’ visits of each Statin brand. A novel feature of our work is that we treat each2It is expensive for a pharmaceutical firm to send out sales representatives to visit and thus build up relationship

with physicians (see Section 3.1). Once Crestor “exits” by giving up relationship-building with certain physicians, it

is reasonable to expect that Crestor faces high re-entry costs, especially in the period when there are skeptics among

physicians on its safety due to the bad news.

3

physician as a market and Crestor’s ceasing of relationship building (i.e., detailing) with a physi-

cian as its withdrawal from the physician market. The relatively large number of physician markets

allows us to investigate the advertising pattern of Lipitor among different physician markets.3

The method we use in the baseline empirical work includes a Difference in Difference study

and a parametric regression model. Both methods give us consistent findings. After the bad news,

Lipitor demonstrates a significantly larger increase in its advertising among physicians with whom

Crestor has a medium share of prescriptions, relative to the increase in its advertising among

physicians where Crestor’s share is either low or high. These results are robust under various

model specifications. Interestingly, we also find that such a pattern in Lipitor’s advertising is more

prominent among physicians who have relatively small number of total Statin drug prescriptions.

One possible explanation to the above empirical findings is based on a strategic view of Lipi-

tor’s advertising. Namely, Lipitor’s detailing is “predatory” in the sense that it is aimed at induc-

ing Crestor to give up relationship building with physicians where Crestor has a medium share

of prescriptions, so that Lipitor can maximize its long run profit. Chen and Tan (2011) offer a

theoretical justification to the “predatory” response of Lipitor to the bad news.4 They consider

each physician as a physician market and that Crestor exits from a physician market if it gives up

relationship building with the physician. By formalizing the post-news competition between Lip-

itor and Crestor into a signaling game, they show that the above empirical findings are consistent

with the theoretical prediction on Lipitor’s predation. Namely, among physicians where Crestor

takes a medium share of prescriptions prior to the bad news, Lipitor advertises with the intent of

inducing Crestor to exit those physician markets so that it can maximize its long run profit.

There can be other explanations to the above empirical results. One may suggest that the

change in Lipitor’s advertising patterns is simply a normal competitive response to the bad news.

That is, holding the structure of the Statin market as fixed, Lipitor increases its advertising to

maximize its post-news short run profit without taking into consideration how Crestor will respond

to its advertising. There exists major difference between the “strategic” view and the “normal3From now on, we use “market” or “physician market” to refer to the individual physician market and “Statin

market” to refer to the aggregate market of Statin brands.4Chen and Tan (2011) is available at http://www.csulb.edu/ ychen7/.

4

competitive” view. In the strategic view, Lipitor is sacrificing part of its short run profit in order

to maximize its long run profit when Crestor is induced into a softer competitor in the future;

whereas such a strategic consideration is missing in the normal competitive view.

To investigate which view is better supported by empirical evidences, we carry out more em-

pirical tests. The first test examines Lipitor’s advertising in physician markets where other Statin

brands take medium market shares. Since the news affects only Crestor but not other Statin

brands, Lipitor’s advertising among physicians where other Statin brands take medium shares

should mainly capture a normal competitive behavior. We find that, Lipitor does not exhibit a sim-

ilar upward distortion in its advertising with physicians where other Statin brands have a medium

share in prescriptions. While this finding to some extent favors the “strategic” view, one caveat is

that the absence of such an upward distortion in Lipitor’s advertising could be purely driven by the

fact that these brands are not affected by the bad news, rather than by a lack of predatory intention.

Therefore, we carry out a second test. We explore the change in advertising patterns of the second

and the third seller of Statin brands, Zocor and Pravachol, in physician markets where Crestor

take different market shares. We note that the market share of these two brands is much smaller

than that of Lipitor; besides, their patents are expiring soon at the time of the bad news. Zocor

and Pravachol’s patents expire in 2006, while the patent of Lipitor expires in 2010 (see Table 1).

Therefore, predatory incentive with these two brands, if there is any, should be much weaker than

that of Lipitor. On the other hand, these two brands still possess the competitive incentive to ex-

ploit the bad news and maximize the post-news short run profits. Therefore, advertising patterns of

Zocor and Pravachol mainly capture a normal competitive response to the bad news in the lack of

a predatory intention. We find that, their advertising patterns change from before to after the news

in a significantly different way compared to Lipitor. In particular, neither brand increases its ad-

vertising the most among physicians where Crestor has a medium share. To address the issue that

these differences may result from the fact that Zocor and Pravachol take a much smaller market

share relative to Lipitor, we also examine the advertising behavior of Zocor and Pravachol when

we only look at physician markets where Zocor/Pravachol has a relatively large share. Our results

indicate that similar findings persist with the reduced sample. Namely, neither of them increases

5

advertising the most among physicians where Crestor has a medium share.

The rest of the paper is organized as follows: Section 2 reviews the related literature. Section

3 introduces the background regarding the Statin drug market and the bad news event. Section 4

describes the data. After that, Section 5 provides the empirical strategy together with a Difference-

in-Difference study. Section 6 gives the empirical model and results. Section 7 compares two

possible explanations to the empirical results: the strategic view and the normal competitive view.

More empirical tests are given in Section 8 to see which view is better supported. Then Section 9

concludes.

2 Literature Review

Our work is closely related to empirical studies on predatory practice. To our knowledge, only

a few empirical evidences of predation exist in contrast to the extensive theoretical studies on

predation.5 Among the empirical works, Morton (1997) examines how the British shipping cartel

responds to the entry of new shipping lines. She finds that incumbent cartels are more likely to

start a price war after the entry of “weaker” firms, which supports the “long purse” theory of

predation. Genesove and Mullin (2006) study the price war following the entry into the American

sugar refining industry. By first comparing the price to the marginal cost and then comparing the

predicted competitive price-cost margin to the observed margins, they suggest that the price wars

following the entry are predatory. Complementing the existing empirical studies on predation, our

work detects predatory behavior by examining the link between predatory incentives and market

properties. With a relatively large sample size, we are able to test whether Lipitor’s advertising5Beginning in the 1950s, “Chicago School” economic analysis, assuming complete information, rightfully pointed

out the irrationality of monopolizing a market through predation (McGee (1958, 1980)). On the other hand, modern

strategic analysis have resurrected the logic of predatory practice. The key assumption is the absence of perfect infor-

mation; under such conditions, predation becomes a rational strategy even for a predator without efficiency or financial

advantages over its prey. Many intrinsic incentives have been identified in favor of predation. To cite several of these

works, see Milgrom and Roberts (1982), Kreps and Wilson (1982), Fudenberg and Tirole (1986), Roberts (1986),

Saloner (1987), etc..

6

pattern across different markets is consistent with the prediction of predation.

Another line of related literature to our work is the empirical study of entry deterrence with

pricing/advertising strategy. Ellison and Ellison (2007) analyze the behavior of pharmaceutical

incumbents in the period prior to the expiration of their patents. They identify evidences of entry

deterrence by investigating whether the incumbent’s behavior is non-monotonic in the size of the

market. While large markets are so attractive to the entrant that deterring entry is never optimal,

small markets inherently defy entry thereby making entry-deterring investment unnecessary. Con-

sequently, the strongest incentive of entry deterrence shall arise within markets of medium size.

In a manner closely related to Ellison and Ellison (2007), our work suggests that the strongest

incentive for Lipitor to predate shall arise within physician markets where the demand of Crestor

is in the medium range. More related works include Morton (2000), where she tests whether in-

cumbents facing generic entry increase advertising in order to deter entry and finds no evidence

supportive of brand advertising as constituting an entry barrier; Goolsbee and Syverson (2008)

study how incumbent carriers of major airlines respond to the threat of the entry of Southwest

Airlines. They find evidence consistent with theories claiming that incumbents try to construct

longer-term loyalty before entry occurs.

Our work is also related to the large empirical literature on market entry and exit. The major-

ity of empirical studies on entry and exit belong to the class of “equilibrium entry/exit models”

pioneered by the influential studies of Bresnahan and Reiss (1988) and Berry (1992). A common

feature of those studies is their assumption that the observed market structure is at equilibrium.

Thus even though the studies try to understand entry and exit, no actual entry or exit is observed

in the data. In addition, equilibrium entry/exit models often ignore how firm strategies – such as

price and advertising – affect entry and exit. Instead, our work focuses on the possibility of the

strategic usage of advertising to induce exit.

There is literature empirically testing strategic advertising to signal product quality. Using

U.S. automobile industry data, Thomas et al. (1998) find an upward distortion in advertising with

products whose prices are also upward distorted. This finding is consistent with signaling game

theory, which says that firms with higher quality would use advertising together with price to

7

signal the quality of unobserved attributes. Horstmann and MacDonald (2003) test if advertising

is a signal of product quality in the compact disc player market. They find that the advertising and

pricing patterns can be explained by signaling game involving persistent consumer uncertainty

and learning.

The theoretical literature offers several justifications for strategic advertising. One argument is

based on the “goodwill” effect of advertising. Schmalensee (1983), Fudenberg and Tirole (1984),

Doraszelski and Markovich (2007) show that when advertising generates “captive” consumers,

incumbents may either under-invest or over-invest in advertising to deter entry. A second rationale

is based on the “noise” effect of advertising. Hilke and Nelson (1984) argue that an incumbent

may overinvest in advertising to jam consumer’s message space, therefore inducing rival’s exit.

The third line of explanation is based on the “signaling” effect of advertising. Bagwell and Ramey

(1988, 1990), Linnermer (1998) show that an incumbent may distort its advertising to signal its

private information regarding cost/demand/quality, aimed at deterring entry.

3 Background

3.1 The Statin Drug Market

Statins are drugs that help lower cholesterol levels. With at least 12 million Americans taking

cholesterol-lowering drugs — most of them Statins — and experts recommending that another

23 million take them, Statin drugs are certainly very important to the pharmaceutical industry. In

many cases, Statin drugs are the number one seller of their company. With sales for 2004 reaching

US $10.8 billion and 5.2 billion respectively, the top two brands in the market, Lipitor and Zocor,

are not only the largest seller for their maker, Pfizer and Merck & Co, but also the largest two

selling drugs in the world.

Table 1 tabulates for the major Statin brands including chemical content, production method,

generic availability, makers, market share in 2004 and patent expiration date. Based on chemi-

cal content, there are six competing products in the market:6 Crestor (rosuvastatin), Lescol (flu-6If two drugs have the same chemical content, they should be considered the same product from a chemical point

8

vastatin), Lipitor (atorvastatin), Pravachol (pravastatin), Zocor (simvastatin) and Mevacor (lovas-

tatin). Only lovastatin has a number of generic products. During our study period, although

generic brands are available, the market is dominated by the branded drug. The four main brands

in the market are Lipitor, Zocor, Pravachol and Crestor. Their makers are Pfizer, Merck, Bristol-

Myers Squibb and Astra Zeneca respectively. Among them, Lipitor and Crestor are considered

closer substitutes as they use the same production method (Synthetic). Lipitor is the dominant

brand counting for 43 percent Statin market share. The second and third largest brand, Zocor and

Pravachol, count for 23 percent and 11 percent of the total sale, respectively. Among the major

brands, Zocor and Pravachol’s patents expire in 2006, while Lipitor’s patent expires in 2010 and

Crestor’s in 2012.

[Insert Table 1 Here]

Pharmaceutical companies mainly compete through advertising and marketing promotions.

Excepting direct to consumer advertising, the most commonly used marketing methods are phar-

maceutical sales representatives’ visits to physicians’ offices, referred to as “detailing”. Detailing

is widely viewed as a means of building a relationship with a particular physician. During the of-

fice visit, a sales representative will show the physician various findings that support the use of the

relevant drug. The sales representative will also leave free samples for the physician to distribute

to patients. According to industry convention, each sales representative is only allowed to visit

the same physician once a month. However, if a firm wants to have more than one detailing per

month at a given physician’s office, it can employ several sales representatives, having each visit

the physician at a different time during the month.

3.2 News Event

The drug that suffered the bad news is Crestor. It first obtained FDA approval on August 12th, 2003

and entered the Statin market shortly afterward. After entering the Statin market, several severe

of view. Of course consumers may have different preference towards drugs of the same content. This difference may

be caused by various reasons such as brand image effect, advertising and different ways of packaging the product, etc.

9

cases of side effects occurred. Alarmed by the findings, a very influential consumer advocacy

group known as Public Citizen submitted on March 4th, 2004 a formal petition to FDA to have the

drug removed from the Statin market. According to its petition, “three patients in the United States

who were taking approved doses of rosuvastatin developed kidney failure or muscle damage. One

of those patients, a 39-year-old woman, died of kidney failure and rhabdomyolysis, or muscle

damage. Data obtained from the United States, the United Kingdom and Canada show seven cases

of rhabdomyolysis and nine case of kidney damage or failure occurred after FDA approval.” Major

news networks such as CBS immediately broadcasted the news about the petition from Public

Citizen. Crestor’s maker Astra Zeneca wouldn’t comment on the deaths or any other serious side

effects, except to say that “the safety profile is totally comparable” to what pre-marketing studies

had predicted.

Interestingly, this was not the first time that Public Citizen issued a petition for the removal

of a Statin drug. Baycol, another Statin drug, withdrew from the Statin market in the fall of 2001

following a petition by Public Citizen to FDA. The side effects of Baycol were very similar to

those of Crestor in that they were both likely to cause kidney damage. Not surprisingly, Public

Citizen emphasized the similarity of the two drugs and noted the withdrawal of Baycol following

their petition to FDA.

The news breakout is clearly not anticipated by a regular physician or a patient. Although it

is possible that before the news event, physicians may have heard about the incidence of severe

side effects from other sources such as sales representative visits, medical journals and profes-

sional conferences, etc., the formal petition of Public Citizen to FDA and subsequent mass media

coverage still brings additional information to physicians. Moreover, it is very likely that this is

the first time that a patient will receive any news about Crestor’s side effect. As many studies

have argued that patient also plays an important role in medical treatment decision, patient’s re-

action may influence physicians’ prescription decisions. Therefore, the news event provides us a

unique opportunity, i.e., a “natural experiment”, to study the changes in advertising strategies of

pharmaceutical firms after the news event.

Although the news event imposes negative impact on the demand of Crestor, it is not severe

10

enough to force Crestor to fully pull out from the Statin market. While Crestor remains active

in detailing after the news, its competitors, especially Lipitor, may employ predatory strategy to

convince Crestor that continuing to detail certain physicians is no longer profitable. If this is the

case, Lipitor’s incentive to prey shall vary in physicians’ prior demand of Crestor. In the following

part, we first describe our data; then we investigate the change in Lipitor’s advertising before and

after the news among groups of physicians where prior demand of Crestor differs.

4 Data

The data we use for the empirical analysis is from a category of prescription drugs used to treat

high levels of blood cholesterol; such drug is commonly referred to as Statin drug. The data are at

the level of physician and are based on a panel of representative physicians from the United States.

These data are obtained from a pharmaceutical consulting firm, which also provided a similar

dataset to a marketing study by Narayanan and Manchanda (2005). For each physician, we observe

a sample of prescriptions between January 1st, 2004 and December 31st, 2004. In addition, we

also have a record of all the detailing visits made by pharmaceutical sales representatives during

the same period. The dataset is unique in that we observe both the prescription level and the

detailing frequencies of all Statin brands at the physician level.7

Since each sales representative can visit a physician’s office only once per month according

to the industry convention, it is likely that pharmaceutical firms make their advertising decision

on a monthly basis. Therefore, we create a monthly physician panel data. For each physician,

we calculate the total number of sales representatives’ visits and total prescriptions for all Statin

brands in each month. Since the news event happens on March 4th, 2004, which is the 9th week

in our dataset, we decide to exclude week 9 from our study to give firms enough time to adjust to

the news event. Our study focuses on changes in detailing patterns right after the news, so we use

a short time window and exclude observations after week 13, which is five weeks after the news7A limitation of this data set is that we do not observe the minutes of each detailing visit. Therefore, we can not

tell whether actual physician contact occurs or not during a detailing visit. However, we think that number of detailing

visits still serve as a good proxy for detailing efforts.

11

event. We make some adjustment in our definition of a month. We divide the time before and after

the news into three periods of equal length: week 1-4 (January), week 5-8 (February) and week

10-13 (March). An advantage of constructing the time period this way is that we make sure our

comparisons are based on the same length of periods.

An important part of our study requires us to create measures of monthly market share. During

our sample period, Crestor’s share of total prescription is around 10 percent. However, some

physicians in our data set have a very small number of prescriptions during each month. We are

concerned that including those physicians will make our measure of market share very noisy and

increase the risk of misclassifying a physician’s market share level. For example, if we observe

only one prescription of Statin drug by a physician in a given month and that prescription happens

to be Crestor, we will infer that Crestor has a 100 percent market share in that physician market.

Clearly, if we classify this physician as having high Crestor market share, we are not as confident

as doing so for another physician who prescribes Crestor 10 times out of 10 prescriptions. In order

to have a reasonably large prescription levels to calculate market share, we restrict our sample to

physicians with 10 or more prescriptions in both week 1-4 (January) and week 5-8 (February).

This restricts our dataset to 308 physicians for whom we observe monthly detailing as well as

prescription levels in all three months.

Table 2 summarizes monthly prescriptions and detailing intensities for our sample physicians.

The statistics of prescriptions and market shares are based on the data in January and February;

and the statistics of number of detailing visits are based on the data in February and March. That

means for each physician, Table 2 summarizes two monthly prescriptions (January and February)

and two number of detailing visits (February and March) of each Statin brand. Therefore, the

sample size is 616. The average number of monthly total prescriptions is 20.54 with a maximum

of 75. Crestor’s average physician market share is 9 percent. Lipitor has the largest physician

market share at 43 percent, followed by Zocor at 24 percent and Pravachol at 13 percent. Table

2 also shows the distribution of market shares for each major Statin brand. Among physicians

with at least one prescription of Crestor, the 25th quantile, median and 75th quantile of Crestor’s

market share are 0.07, 0.11 and 0.20 respectively. It is worth mentioning that the distribution of

12

Pravachol’s market share is the closest to that of Crestor.

We do not observe physicians’ personal characteristics such as their education or type of prac-

tice. Nevertheless, we report in Table 2 measures of patients’ mix at physicians’ offices. These

measures include the (physician market) share of new patients, the share of older patients (age

65+), the share of Asian patients, the share of female patients and the share of severe-symptom

patients. On average, new patients, patients aged 65 or above, Asian patients, female patients and

patients with severe symptoms count for 10 percent, 47 percent , 2 percent , 47 percent and 6

percent, respectively.

Table 2 also summarizes the monthly number of detailing visits by each major Statin brand.

In February and March, the average number of detailing visits by Lipitor, Zocor, Pravachol and

Crestor are 0.83, 0.89, 0.58 and 1.29 respectively. Crestor has the highest level of detailing inten-

sity in part due to that it just entered the Statin market about four months ago. In addition, because

of the release of the bad news, Crestor may increase its advertising level to counteract the negative

effect of the bad news. It is clear that all brands exhibit considerable variations in detailing inten-

sity across physicians. For example, the maximum number of Lipitor’s monthly detailing visits to

a physician is 6, while to some physicians there is no detailing visit by Lipitor at all.

[Insert Table 2 Here]

5 Empirical Strategy

The objective of our work is to examine how Crestor’s competitors change their advertising across

different physician markets after the release of the bad news to Crestor. In particular, we investi-

gate whether the change in their advertising is non-monotonic in the prior demand of Crestor, and

whether they increase advertising the most among markets where the prior demand of Crestor is

in the medium range. We focus on Lipitor’s response to the bad news as Lipitor is the dominant

brand in the Statin drug market and also the closest substitute for Crestor. Moreover, patents of

Zocor and Pravachol are expiring soon at the time of the bad news. Therefore, the incentive to

13

practice predatory advertising should be the strongest with Lipitor.8

As the bad news do not generate the pressure for Crestor to fully pull out from the Statin

market, the exiting decision of Crestor is more relevant at the physician level. Therefore, we

treat each physician as a market and define Crestor as having exited from a physician market if it

foregoes relationship building with that physician by ceasing to send sales representatives.9 We

use Crestor’s share in a physician’s total prescriptions before the bad news to capture the prior

demand of Crestor in that physician market.

We want to emphasize that we are examining the change in Lipitor’s advertising strategy

before and after the news event, rather than the actual outcome of such a strategy. The main

reason is that even if Lipitor’s post-news advertising is indeed predatory, it is very difficult to

know how long it takes to have any effect. In addition, the information flow is constant in the

Statin drug market. After the bad news event, there might be other additional information released

to the public. For example, we observe that FDA changed the warning on the label of Crestor three

months after Public Citizen’s petition. These additional information certainly will have effects on

Lipitor’s strategy, which we are not able to control for in our model. As such, we focus on the

short time window around the news event, so that the results are mostly driven by the bad news

rather than additional information released afterwards.

We focus on the changes in detailing strategy rather than price and other competition practice

such as direct to consumer advertising. The reason is that we look at the immediate response of

firms after the news event,10 whereas firms’ change in price and direct to consumer advertising

is likely to take longer time to occur. In addition, we explore how the changes in firms’ strate-

gies differ across different types of physicians. In our data, only detailing activity varies across

physicians, while price and direct to consumer advertising are constant.8We will also examine the changes in Zocor and Pravachol’s advertising patterns in section 8 when we discuss

alternative explanations to our empirical results.9As long as Crestors competitors prefer its exiting (stopping detailing a physician), the intrinsic incentive of preda-

tion still holds under our definition of market exit.10We expect that the effect of the bad news is the strongest in the period immediately following the news event,

together with the incentives for Crestor’s competitors to use predatory strategies.

14

5.1 A Difference in Difference Study

The main idea of a Dif-in-Dif (Difference in Difference) Study is to compare the changes in the

outcome variable of the comparison group with that of the control group. In our baseline Dif-

in-Dif estimation, the outcome variable is number of Lipitor detailing. Ideally, we would like to

look at the change in Lipitor’s number of detailing during a short time window around the date

of the news event (March 4th, 2004). However, we also need a large data set to provide a test

with appropriate power. As the news happens on March 4th (week 9), we compare the number of

detailing in February (week 5-8) with that in March (week 10-13). As described in Section 5, by

defining the time period this way, we guarantee that our comparisons are done over the exact same

length of periods.

In our baseline Dif-in-Dif estimation, the comparison group includes physicians whose pre-

scription level of Crestor before the news falls in the medium range. For the control group, we use

two different specifications: physicians with low prescription level of Crestor and physicians with

high prescription level of Crestor. In this way, we control for the expected differences in detailing

due to other factors that are not associated with the news event such as the seasonal effect. It is

worth mentioning that our test objectives are different from the standard Dif-in-Dif estimation.

We are not trying to estimate the causal effect of news event on detailing level. Instead, we are

interested in testing if the detailing visits of Lipitor increases more in the comparison group than

in BOTH control groups.

We use the market share of Crestor in week 1-8 to categorize a physician. Unfortunately, there

is no guideline for us to determine the range of market share where predatory intent exists. We

decide to use the quantile of Crestor’s market share distribution to determine the ranges. Based

on week 1-8 prescription data, the 25th percentile of Crestor’s market share is at 0.05, and the

75th percentile is at 0.18. We choose to categorize a physician as low, medium, and high type if

the market share of Crestor is between [0-0.05), [0.05-0.18], and (0.18, 1] respectively. We try to

categorize more physicians in the medium type and avoid the potential problems of misclassifying

a medium type physician into either low or high type. The downside of having a large size of

medium type is that we may include low or high type physicians as medium type, which in turn

15

may weaken our test results. Clearly, the misclassification of market share ranges is a potential

problem. To check the robustness of our results, we also use alternative but similar definitions of

the range of Crestor market to categorize physicians. The results show that our qualitative results

continue to hold under alternative categorizations.11

The baseline Dif-in-Dif results are reported in Table 3. During the four weeks before the

news event (week 5-8, or February), the average number of Lipitor detailing among medium type

physicians is 0.76, while the average number of Lipitor detailing among physicians of low and

high types are 0.73 and 0.82 respectively. After the news event (week 10-13, or March), the

average number of Lipitor detailing increases to 0.92 among medium type physicians. Among the

low and high type physicians, the average number of Lipitor detailing increase to 0.81 and 0.89,

respectively. Based on the first difference results, the increase in number of Lipitor detailing for

the medium type physicians is 0.16, while the increase for low and high types physicians are 0.09

and 0.07 respectively. The Dif-in-Dif estimates (0.07 and 0.09) suggest that Lipitor’s detailing

increases the most among medium type physicians. However, due to the large standard error (0.18

and 0.24), we cannot conclude that the difference is statistically significant.

[Insert Table 3 Here]

A potential concern of the Dif-in-Dif study is the misclassification of Crestor’s medium market

share. We classify a physician to be the medium type if Crestor’s market share is between the 25th

and 75th quantile. In order to check whether the definition of the medium market share affects our

findings, we non-parametrically estimate the detailing level of Lipitor during the four weeks prior

to and after the news event as a function of the quantile of Crestor market share in week 1-8.12

11For example, the qualitative part of the Dif-in-Dif results reported in Table 3 is unaltered if we use [0.04, 0.2] or

[0.06, 0.15] as the range of Crestor market share to categorize a physician as of the medium type.12The Nadaraya Watson kernel regression estimator is:

m(x) =

n∑i=1

wni(x)yi

wni(x) = K(xi − x

h)/(

n∑i=1

K(xi − x

h))

where yi is the observed number of detailings, xi is the observed quantile of market share, and x is the value of quantile

16

Figure 1 plots the nonparametric estimates of Lipitor detailing and the quantile of Crestor

market share. The result shows that the change in Lipitor advertising is non-monotonic in Crestor’s

market share. Moreover, when Crestor’s market share is between the 40th and 70th quantile,

Lipitor increases the detailing level the most after the news event. For physicians where Crestor’s

market share is below 40th quantile or above 70th quantile, Lipitor’s advertising level increases

by a much smaller scale.

[Insert Figure 1 Here]

6 The Empirical Model and Results

Our baseline econometric model uses a regression based model to measure the impact of the bad

news on the detailing pattern of Lipitor. An advantage of a regression based model is that we can

control for additional market characteristics. In addition, as we have a panel data of physicians,

regression based model allows us to control for physician specific effect through random effect

model and fixed effect model.

We use the market share of Crestor in the previous month to categorize a physician. This

means that types of physicians observed in February (week 5-8) are determined according to the

market share of Crestor in January (week 1-4); types of physicians observed in March (week 10-

13) are determined according to the market share of Crestor in February (week 5-8). Similar as in

the Dif-in-Dif study, the quantile of Crestor’s market share is used to determine the high, medium

and low ranges of Crestor’s market share. Based on the prescription data in week 1-4 (January)

and week 5-8 (February), the 25th percentile of Crestor’s market share is at 0.07, the median is at

0.11, and the 75th percentile is at 0.2. We thus categorize a physician as low, medium and high

type if the market share of Crestor is between [0-0.07), [0.07-0.2], and (0.2, 1] respectively.13

of market share where we calculate the expected number of detailings.13The qualitative results of Table 4 is unaltered if we use alternative but similar definitions for physicians of low,

medium and high type. For example, we may categorize a physician as of the medium type if the market share of

Crestor is between [0.08, 0.18] or [0.06, 0.22].

17

To investigate how Lipitor changes its detailing pattern from before to after the bad news, we

run Poisson regression to estimate the basic specification:

yit = exp(β0 + β1CM + β2CH +

(γ0 + γ1CM + γ2CH) ∗ Postevent+ δXit + αi + ϵit).

Here yit is the number of Lipitor detailing to physician i before (t = 1) and after (t = 2) the

news event. It is a non-negative integer. Similar to the above Dif-in-Dif estimation, we use four

weeks of window before and after the event. Thus t = 1 if the week is between the 5th week and

the 8th week, and t = 2 if the week is between the 10th week and the 13th week. CM and CH are

dummy variables indicative of Crestor’s previous market share as in the medium range and high

range respectively (recollect that we use week 1-4 market share to create previous market share for

t = 1 and week 5-8 market share to create previous market share for t = 2). Postevent is a dummy

variable equal to 1 for t = 2. Xit are market characteristics that may affect the detailing decisions.

These market characteristics include the market size (total number of prescriptions), the share of

new patients, the share of older patients (age 65+), the share of Asian patients, the share of female

patients, and the share of severe-symptom patients. These additional variables, all are based on the

data in the previous month (four weeks), help to control for other factors that might influence the

demand for Crestor. And αi is the physician specific effect in the random effect model and fixed

effect model. In all the regressions, we weight observations using the number of prescriptions of

the physician in week 1-8 in order that larger markets should have a greater impact on Lipitor’s

detailing response than smaller ones. Table 4 gives the results from the baseline specification

using random effect regression and fixed effect regression. The estimates we report include γ1,

which is the coefficient of CM ∗ Postevent; and also γ0 and γ2, the coefficients of Postevent

and CH ∗ Postevent respectively.

In Table 4, we first report the regression results in specification 1 and specification 2, where we

do not include Xit (market characteristics) in the regression. The estimated values of γ0, γ1, and

γ2 from the random effect models are 0.06, 0.07 and -0.13 respectively. All coefficient estimates

are significant. The estimates from the fixed effect model are 0.05, 0.05 and -0.11 respectively

18

and only the estimates of γ0 and γ2 are significant. First, the estimates from both specifications

are reasonably similar. They suggest that γ0 and γ1 are positive and γ2 is negative. Namely, after

the news event, Lipitor’s advertising level increases for physicians with low and medium Crestor

market share; while it decreases for physicians with high Crestor market share. In addition, the

magnitude of increase in Lipitor’s detailing is the highest with those physicians where Crestor’s

market share falls in the medium range.

Second, we report the regression results in specification 3 and specification 4, where we in-

clude Xit in the regression. The estimate of γ0 is 0.05 in random effect model and 0.07 in fixed

effect model. Estimates of γ1 are 0.08 and 0.07 respectively for random effect and fixed effect

model; while the estimates of γ2 are -0.05 and -0.12 respectively. The estimates from both specifi-

cations are similar. They show that Lipitor’s advertising level increases the most within physicians

with medium Crestor market share. This suggests that our findings are robust to additional control

variables.

[Insert Table 4 Here]

Here an interesting question arises. If Lipitor’s post-news advertising has predatory intent, will

such intent be more pronounced in markets relatively large (i.e., physicians with a large number

of Statin prescriptions), or in markets relatively small? On the one hand, Lipitor may benefit more

when Crestor exits a large market. Correspondingly, Lipitor may possess a stronger incentive

to prey in markets of relatively large size. On the other hand, advertising is more intensified

among physicians who prescribe a large number of Statins. As the marginal effect of advertising

diminishes, the predation of Lipitor becomes more costly. In addition, advertising may also be

less informative in large markets since those physicians tend to be more knowledgeable about

drug efficacy from their prescription experiences. Therefore, the incentive for Lipitor to prey can

be weaker in large markets as advertising becomes less effective.

To check either a larger market size strengthens or weakens the predatory intention, we sepa-

rate the physician markets into two categories: large and small. The average number of monthly

Statin drug prescriptions in our sample is around 20. We thus define a physician market to be

19

“small” if its monthly total prescriptions both before and after the news event are less than 20.

Otherwise, we categorize it to be “large”. This categorization leaves us with 152 small markets

and 156 large markets.14 We then estimate our baseline model on the two samples separately and

the results are reported in Table 5.

[Insert Table 5 Here]

The top panel of Table 5 presents the baseline results for small physician markets. In speci-

fication 1 and specification 2 where we do not control for additional variables in the regression,

the estimated values of γ0, γ1 and γ2 from the random effect models are 0.14, 0.15 and -0.28

respectively. All coefficient estimates are significant. The estimates from the fixed effect model

are 0.04, 0.47 and -0.15 respectively and only the estimates of γ1 and γ2 are significant. Secondly,

we report the regression results in specification 3 and specification 4, where we include additional

control variables in the regression. The estimate of γ0 is 0.13 in random effect model and 0.08

in fixed effect model. Estimates of γ1 are 0.18 and 0.5 respectively; and the estimates of γ2 are

-0.19 and 0.03 respectively. Both of the estimates of γ1 are significant. The estimates from all

four specifications are reasonably similar. They suggest that estimates of γ0 and γ1 are positive

and significant while estimates of γ2 are either negative or insignificant.

The change in Liptior’s advertising patterns in small markets shows a similar pattern compared

to the results using the full sample. Namely, after the news event, Lipitor’s advertising increases

the most among physicians where Crestor takes a medium market share. Moreover, such a non-

monotonic change in Lipitor’s advertising is more prominent when we only look at those small

markets, as the estimates of γ1 in the top panel of Table 5 are highly significant and larger than the

estimates in Table 4.

However, we could not find a similar pattern in the change of Lipitor’s advertising when we

only look at large markets, as shown by the bottom panel of Table 5. The estimates of γ1 in the two

random effect models (specification 1 and 3) are negative and insignificant at -0.01 and -0.02. In

the two fixed effect models (specification 2 and 4), the estimates of γ1 are negative and significant14Our empirical results are robust under alternative but similar definitions of large and small markets.

20

at -0.2 and -0.22 respectively. In large markets, Lipitor does not increase its advertising the most

among physicians where Crestor takes a medium market share.

The results in Table 5 show that the incentive for Lipitor to target markets where Crestor has

a medium market share is even stronger among smaller markets. While other interpretations can

exist, this finding is supportive to the argument that when a market becomes larger and is more

frequently detailed before the news, additional detailing after the news is less effective.

7 Strategic Response vs. Normal Competitive Response

The above empirical tests look at the changes in advertising patterns of Crestor’s major competitor,

Lipitor, from before to after the news event. The results suggest that Lipitor changes its advertising

in a non-monotonic way in the prior market share of Crestor. In addition, Lipitor increases its

advertising intensity the most among physicians with medium market share of Crestor.

There can exist different interpretations to the above empirical results. One possible explana-

tion is based on a strategic view. According to this view, Lipitor’s post news advertising is not

aimed at maximizing its post-news short run profit. Instead, Lipitor advertises with a predatory

intent to induce Crestor to leave certain physician markets so that Lipitor can benefit in the long

run. The basic idea of predation is that a firm may sacrifice part of the profits that could be earned

under competitive circumstances in order to induce the exit of an opponent, therefore gaining ad-

ditional profits in the future. Literature has provided various rationales for predatory behavior

based on game theoretical analysis. A common feature of these theoretical models is the existence

of asymmetric information.

Chen and Tan (2011) offer a justification to Lipitor’s predation based on an asymmetric in-

formation assumption between Lipitor and Crestor. They assume that when the bad news is sud-

denly released to the public, Lipitor possesses superior information regarding how severely the

bad news is going to impact physicians’ prescription behavior vis-a-vis Crestor. Such an assump-

tion is based on the following reasons: first, Crestor is a new entrant at the time of the bad news

and it is Crestor’s first time experiencing such bad news. Therefore, there is likely uncertainty

21

in Crestor’s perception regarding how badly the news will affect each physician’s prescription of

Crestor. Second, Crestor’s competitors witness a similar bad news event on Baycol three years

ago.15 Moreover, they have interacted with physicians for a much longer period compared to

Crestor. Therefore, these competitors are likely to possess superior information regarding how

physicians will respond to the bad news of Crestor.

Chen and Tan (2011) constitute a two-period game to model the post-news advertising compe-

tition between Lipitor and Crestor among independent physician markets. Before the game starts,

the bad news of Crestor occurs. Nature chooses the effect of the bad news as either severe or mild,

and reveals the true effect only to Lipitor but not to Crestor. In period one, Lipitor and Crestor

choose their advertising intensities simultaneously, and each has its demand/profit realized. In

period two, Crestor makes its withdrawal decision based on its period-one demand. If Crestor

chooses to continue to operate, in stage three it infers the real effect of the bad news and compete

with Lipitor under symmetric information.

Chen and Tan (2011) show that in period one when Crestor remains active to infer information

about its future profitability of continuing to detail a physician, Lipitor may distort its advertising

to signal Crestor its private information regarding the effect of the bad news. The model in Chen

and Tan (2011) admits two types of equilibrium: separating equilibrium and pooling equilibrium.

In both of them, Lipitor upward distorts its advertising compared to its normal competitive adver-

tising. By doing so, Lipitor further drives down the period-one demand of Crestor in order to either

signal Crestor an intrinsically unprofitable market situation (separating equilibrium) or to interfere

with Crestor’s reference procedure of the true effect of the bad news (pooling equilibrium).

Moreover, Chen and Tan (2011) show that such a predatory intent of Lipitor exists only in

physician markets where demand of Crestor prior to the news is neither too low nor too high. The

intuition is that while predation is unnecessary in physician markets with low demand of Crestor15Although Crestor’s producer Astra Zenecca must have observed the bad news event on Baycol, it is not operating at

the Statin market at the point of that bad news hence can only observe the event as an outsider. Therefore, its knowledge

regarding how physicians change their prescription behavior as a response to the bad news should be limited relative to

Crestor’s competitors.

22

prior to the news event, it is not possible in physician markets with high prior demand of Crestor.

Thus only in physician markets with medium prior demand of Crestor, Lipitor has the incentive to

employ predatory strategy.16

An alternative interpretation is offered by a normal competitive view to Lipitor’s non-monotonic

change in advertising from before to after the news. According to this view, Lipitor takes a normal

competitive response to the bad news by maximizing its post-news short run profit, and it targets

physicians with medium market share of Crestor while taking the current Statin market structure

as fixed. Recollect that as a profit maximizing firm, Lipitor will choose its advertising level in

each physician market such that the marginal benefit of advertising equals the marginal cost of

advertising. Given that the marginal cost of advertising is constant among physicians, Lipitor will

advertise more towards physicians where the marginal benefit of advertising is larger. Therefore,

if it is the case that the bad news enhances the marginal benefit of advertising more among physi-

cians with medium market share of Crestor, the empirical findings can arise as a result of Lipitor’s

normal competitive response to the bad news. In the following section, we conduct more empirical

tests to investigate either the strategic view or the normal competitive view is better supported by

the empirical results.

8 Testing the Normal Competitive View

As pointed out above, in certain scenario Lipitor’s normal competitive response to the bad news

may lead to a non-monotonic change in its advertising to Crestor’s prior market share. In this

section, we use several additional tests to explore the possibility of this alternative explanation.

First, since the news only affects Crestor but not other Statin brands, Lipitor’s detailing distri-

bution among physicians where other brands take different market shares should capture a normal

competitive behavior. If the non-monotonic change in Lipitor’s detailing is driven by Lipitor’s

predatory intent towards Crestor after the bad news, a similar pattern will not be found with Lipi-

tor’s detailing towards physicians where other brands have medium market share.16The intuition of the results is very similar to that of Ellison and Ellison (2007), where the strategic use of advertising

to deter entry occurs only in markets with intermediate probability of entry.

23

Second, we expect that the two smaller brands, Zocor and Pravachol, with patents expiring

soon, have much weaker incentives (if existing at all) to engage in predatory practice compared

to Lipitor. On the other hand, they have the competitive incentive to maximize their short-run

profits in the post-news period. Therefore, their post-news advertising mainly captures the normal

competitive response to the bad news in the absence of a predatory intent. If the change in their

detailing patterns is significantly different from the change in Lipitor’s detailing patterns from

before to after the news, the evidence then favors the strategic view that a predatory intent underlies

Lipitor’s post-news advertising.

We follow the same approach as in Section 5 and 6 to do the additional tests. We begin with a

Difference-in-Difference study where we compare the difference in detailing intensity between the

comparison group and the control groups before and after the bad news. We then use a regression

based estimator which allows us to control for additional variables.

8.1 Change in Lipitor’s Advertising and Pravachol’s Market Share

As argued above, the predatory strategy of Lipitor can only be aimed at driving Crestor but not

other Statin brands to leave certain physician markets. Since Pravachol’s market share is the

closest to that of Crestor, we examine whether Lipitor increases its advertising intensity the most

among physicians where Pravachol has a medium market share.

We report in Table 6 the results of a Dif-in-Dif study when we use Pravachol’s market share

to construct the comparison group and the control groups, while keeping the number of Liptior

detailing the outcome variable. Namely, the comparison group includes physicians whose pre-

scription level of Pravachol falls in the medium range. The control groups include physicians with

low prescription level of Pravachol and physicians with high prescription level of Pravachol.

After the news event, average detailing number of Lipitor to physicians with medium market

share of Pravachol increases from 0.82 to 0.90; while its detailing increases from 0.75 to 0.93 for

physicians with low market share of Pravachol and from 0.64 to 0.69 for physicians with high

market share of Pravachol. Thus Lipitor’s detailing intensity shows a much larger increase within

physicians where Pravachol has low market share, instead of within physicians where Pravachol

24

has medium market share. The Dif-in-Dif estimates are -0.10 and 0.03, with the first value nega-

tive.

[Insert Table 6 Here]

Figure 2 plots the nonparametric estimates of the number of Lipitor detailing and the quantile

of Pravachol market share. As the figure shows, Lipitor’s advertising increases the most after the

bad news among physicians with low Pravachol market share.

[Insert Figure 2 Here]

We conclude that the change in Lipitor’s detailing intensity shows a different pattern in Prava-

chol’s market share compared to its pattern in Crestor’s market share. This implies that a normal

competitive advertising of Lipitor may exhibit a different pattern compared to the pattern we ob-

serve among different Crestor market share. While this finding favors the strategic view to some

extent, there is one caveat. One may argue that such a difference is driven by the fact that the

bad news has no impact on Pravachol, rather than by Liptior’s predatory intent towards Crestor.

Therefore, we carry out a second test in the following part to further explore the possibility of

Lipitor’s normal competitive response to the bad news.

8.2 Change in Zocor/Pravachol’s Advertising and Crestor’s Market Share

The second and the third best sellers of Statin drug are Zocor and Pravachol. They take a much

smaller market share than Lipitor and their patents are expiring soon at the point of the bad news.

Thus their predatory intent towards Crestor, if exists at all, should be much weaker compared to

Lipitor. On the other hand, they have the competitive incentive to maximize their short-run profits

in the period following the bad news.

We run two additional Dif-in-Dif estimations by replacing the outcome variable in the base-

line model (Table 3) with the number of Zocor detailing and the number of Pravachol detailing,

respectively. The comparison group and the control groups are constructed in the same way by

using Crestor’s market share. The results are included in Table 7. The mean value of the number

25

of Zocor detailing within physicians with medium, low and high market share of Crestor is 0.81,

0.72, and 0.79 before the news event; and it increases to 0.94, 0.98 and 1.07 after the news event.

The increases in number of Zocor detailing are higher among physicians where Crestor has low

and high market share (0.26 and 0.28). The Dif-in-Dif results are both negative (-0.13 and -0.15)

for Zocor. Moreover, changes in the number of Pravachol detailing differ from that of Lipitor as

well. The mean value of the number of Pravachol detailing within physicians with medium, low

and high market share of Crestor is 0.60, 0.49, and 0.43 before the news event; and it changes to

0.77, 0.47 and 0.66 after the news event. The Dif-in-Dif estimates are 0.2 and -0.06 respectively.

Changes in Pravachol’s detailing patterns is monotonic in Crestor market share.

[Insert Table 7 Here]

The above results show that Zocor and Pravachol respond to the bad news in a different way

compared to Lipitor. Nevertheless, one may argue that the uniqueness in the change of Liptior’s

advertising is driven by its unique properties (for example, due to its large market share) and its

response is still normal competitive. To investigate this point, we check Zocor and Pravachol’s

response to the bad news using a reduced sample, which includes only physician markets where

Zocor/Pravachol has relatively large market share. In the reduced sample for Zocor, we use physi-

cians where Zocor’s market share in week 1-8 is above its median level (24 percent); and the

reduced sample for Pravachol includes only physicians where Pravachol’s market share in week

1-8 is above its median level (11 percent).

Table 8 presents the Dif-in-Dif results using the reduced samples. Compared to Table 7,

the results with the reduced samples are qualitatively the same. The Dif-in-Dif results are both

negative (-0.13 and -0.23) for Zocor, while for Pravachol, the Dif-in-Dif estimates are 0.21 and

-0.13 respectively. These results show that the changes in Zocor and Pravachol’s detailing patterns

differ from that of Lipitor even among markets where Zocor and Pravachol take a relatively large

market share.

[Insert Table 8 Here]

26

Figure 3 and Figure 4 present the results of nonparametric estimates of Zocor and Pravachol’s

detailing level. Compared to Lipitor, Zocor and Pravachol clearly have different changes in de-

tailing patterns after the news event. The most significant increase in Zocor’s detailing happens to

physicians with high and low Crestor market share, while the most significant increase in Prava-

chol’s detailing happens to physicians with high Crestor market share.

[Insert Figure 3 and Figure 4 Here]

We then use a regression model to look at the changes in Zocor and Pravachol’s detailing. We

replace the dependent variable in the baseline model with the number of Zocor detailing and the

number of Pravachol detailing, respectively. Table 9 gives the results of the random effect model

and the fixed effect model. The top panel reports the results using the number of Zocor detailing

as the outcome variable. In specification 1 and 2, we do not include additional control variables.

According to the random effect model (specification 1), the estimated values of γ0, γ1and γ2

are 0.23, -0.14 and 0.05 respectively. The estimates of γ0 and γ1 are significant. According to the

fixed effect model (specification 2), the estimated values of γ0, γ1, and γ2 are 0.26, -0.14 and -0.09

respectively. All estimates are significant. In specification 3 and 4, we include additional control

variables. According to the random effect model (specification 3), the estimated values of γ0, γ1

and γ2 are all significant at 0.21, -0.14 and 0.17 respectively. According to the fixed effect model

(specification 4), the estimated values of γ0, γ1 and γ2 are 0.26, -0.01 and 0.08 respectively. Only

the estimate of γ0 is significant. Since the estimates of γ1 are either negative or insignificant across

all specifications, it is clear that Zocor does not increase its detailing the most among physicians

where Crestor has medium market share.

The bottom panel of Table 9 reports the results using the number of Pravachol detailing as

the outcome variable. In specification 1 and 2, we do not include additional control variables.

According to the random effect model (specification 1), the estimated values of γ0, γ1and γ2 are

0.07, 0.02 and 0.36 respectively. The estimates of γ0 and γ2 are significant. According to the

fixed effect model (specification 2), the estimated values of γ0, γ1 and γ2 are 0.09, -0.09 and 0.41

respectively. All estimates are significant. In specification 3 and 4, we include additional control

27

variables. According to the random effect model (specification 3), the estimates of γ0, γ1 and

γ2 are 0.05, -0.02 and 0.43 respectively. The estimates of γ0 and γ2 are significant. According

to the fixed effect model (specification 4), the estimates of γ0, γ1and γ2 are 0.1, -0.09 and 0.47

respectively. The estimates of γ0 and γ2 are significant. In all four specifications, the estimates

of γ1 are either negative or insignificant, while the estimates of γ2 are positive and significant.

This suggests that Pravachol increases advertising intensity more to physicians with high Crestor

market share.

[Insert Table 9 Here]

To summarize, both the Dif-in-Dif study and the regression results show that the changes in

the advertising patterns of the two smaller competitors of Crestor, Zocor and Pravachol, differ in

a significant way from that of Lipitor. We do not find a significant increase in Zocor/Pravachol’s

advertising intensity among physicians where Crestor takes a medium market share.

The additional results from the above empirical tests suggest that the upward-distortion in

Lipitor’s advertising among physicians with medium Crestor market share is unique. Lipitor does

not show a similar advertising pattern among physician markets where Pravachol takes a medium

market share; nor do Zocor and Pravachol exhibit similar patterns in their advertising among

physicians where Crestor takes a medium share. These results are consistent with the strategic

view of Lipitor’s response to the bad news described in Section 7, and can hardly be justified by

the hypothesis that Lipitor is in normal competitive advertising after the bad news.

9 Conclusion

We study the advertising behavior of Crestor’s competitors following a petition of withdrawal of

Crestor. Our empirical work shows that the advertising of Crestor’s major competitor, Lipitor,

exhibits certain pattern which differs in a significant way from Crestor’s other competitors. To be

explicit, we find that Lipitor increases its advertising the most among physicians where Crestor

has taken a medium share in total prescriptions. Moreover, such changes in advertising patterns

28

are not found with the second and third seller, Zocor and Pravachol. Neither does Lipitor shows a

similar change in its advertising with physicians where other Statin drugs take a medium market

share. These results thus favor the strategic view of Lipitor’s post-news advertising and argue

against the normal competitive view.

Our empirical results are consistent with the predictions in Chen and Tan (2011), where they

argue that a predatory intention can underlie Lipitor’s post-news advertising. They show that under

asymmetric information between Lipitor and Crestor, in equilibrium Lipitor upward distorts its

advertising in order to induce Crestor to deem staying with certain physicians unprofitable. When

Crestor gradually ceases relationship building with those physicians, Lipitor can benefit the most

from the bad news by maximizing its profit in the long run. Moreover, the upward distortion

in Lipitor’s advertising occurs only among physicians where Crestor has taken a medium market

share, since for the rest of the physicians it is either unnecessary or impossible for Lipitor to induce

the “exit” of Crestor.

In contrast to the extensive theoretical studies on the use of predatory strategies to induce

a rival to exit, empirical study on predation is still rare. One reason for the lack of empirical

work relies with the difficulty in differentiating a predatory behavior from a normal competitive

behavior. Our work is an attempt to fill this blank by empirically testing predatory behavior.

Instead of directly comparing marginal benefit and marginal cost of advertising, our work detects

predation by linking firms’ strategic incentive to market properties. Our approach is based on the

“natural experiment” method. Since the news event is an exogenous shock to the Statin market,

we can detect predatory intent by comparing the changes in advertising from before to after the

news across different physician markets where the incentive to prey varies. While our work offers

a potential avenue of empirically testing predatory behavior, future work is certainly warranted to

further differentiate a predatory behavior from a normal competitive behavior.

29

References

Bagwell, K, 2007. “The Economic Analysis of Advertising.” Handbook of Industrial Organiza-

tion, Volume 3.

Bagwell, K., Ramey, G., 1988. “Advertising and Limit Pricing.” The Rand Journal of Economics,

19, 59-71.

Bagwell, K., Ramey, G., 1990. “Advertising and Pricing to Deter or Accommodate Entry When

Demand is Unknown.” International Journal of Industrial Organization, 8, 93-113.

Berry, S., 1992. “Estimation of a Model of Entry in the Airline Industry.” Econometrica, 60,

889-917.

Bresnahan, T., Reiss, P., 1988. “Entry and Competition in Concentrated Markets.” Journal of

Political Economy, 99, 977-1009.

Chen, Y., Tan W., 2011. “A Theory on Predatory Advertising After A Demand-reducing Shock.”

working paper.

Cho, I., Kreps, D., 1987. “Signalling Games and Stable Equilibria.” Quarterly Journal of Eco-

nomics, 102, 179-221.

Doraszelski, U., Markovich, S., 2007. “Advertising Dynamics and Competitive Advantage.” The

Rand Journal of Economics, 38, 557-592.

Ellison, G., Ellison, S., 2007. “Strategic Entry Deterrence and the Behavior of Pharmaceutical

Incumbents Prior to Patent Expiration.” working paper.

Fudenberg, D., Tirole, J., 1984. “The Fat-cat Effect, The Puppy-dog Ploy, and The Lean and

Hungry Look.” American Economic Review, Papers and Preceedings, 74, 361-366.

Fudenberg, D., Tirole, J., 1986. “A ‘Signal-Jamming’ Theory of Predation.” The Rand Journal of

Economics, 17, 366-376.

Genesove, D., Mullin, W., 2006. “Predation and Its Rate of Return: The Sugar Industry, 1887-

1914.” The Rand Journal of Economics, 1, 47-69.

Goolsbee, A., Syverson, C., 2008. “How Do Incumbents Respond to the Threat of Entry? Evi-

30

dence from the Major Airlines.” The Quarterly Journal of Economics, 123, 1611-1633.

Hilke, J.C., Nelson, P.B.,1984. “Noisy Advertising and The Predation Tule in Antitrust Analysis.”

American Economic Review, Papers and Preceedings, 34, 367-371.

Horstmann, I., MacDonald, G., 2003. “Is Advertising A Signal of Product Quality? Evidence from

the Compact Disc Player Market, 1983-1992.” International Journal of Industrial Organization,

21, 317-345.

Kreps, D., Wilson, R., 1982. “Reputation and Imperfect Information.” Journal of Economic The-

ory, 27, 253-279.

Linnermer, L., 1998. “Entry Deterrence, Product Quality: Price and Advertising as Signals.”

Journal of Economics and Management Strategy, 7, 615-645.

McGee, J., 1958. “Predatory Price Cutting: The Standard Oil (N.J.) Case.” Journal of Law and

Economics, 1, 137-169.

McGee, J., 1980. “Predatory Pricing Revisited.” Journal of Law and Economics, 23, 443-459.

Milgrom, P., Roberts, J., 1982. “Predation, Reputation, and Entry Deterrence.” Journal of Eco-

nomic Theory, 27, 280-312.

Morton, F., 1997. “Entry and Predation: British Shipping Cartels: 1879-1929.” Journal of Eco-

nomics & Management Strategies, 679-724.

Morton, F., 2000. “Barriers to Entry, Brand Advertising, and Generic Entry in the US Pharmaceu-

tical Industry.” International Journal of Industrial Organization, 18, 1085-1104.

Narayanan, S., Manchanda, P., 2005. “Heterogeneous Learning and the Targeting of Marketing

Communication for New Products.” Working Paper.

Roberts, J., 1986. “A Signaling Model of Predatory Pricing.” Oxford Economic Papers, 38, 75-93.

Saloner, G., 1987. “Predation, Mergers, and Incomplete Information.” The RAND Journal of

Economics, 18, 165-186.

Schmalensee, R., 1983. “Advertising and Entry Deterrence: An Exploratory Model.” Journal of

Political Economy, 91, 636-653.

31

Thomas, L., Shane, S., and Weigelt, K., 1998. “An Empirical Examination of Advertising as A

Signal of Product Quality.” Journal of Economic Behavior & Organization, 37, 415-430.

32

Brand Content Method Type Generic Available Maker Market Share 1

Patent Expiration

Advicor niacin/lovastatin Fermentation Branded No Kos 1.6

Altocor lovastatin Fermentation Branded Yes Andrx 0.79

Crestor rosuvastatin Synthetic Branded No Astra Zeneca 10.89 2012

Lescol/XL fluvastatin Synthetic Branded No Reliant 4.38 2011

Lipitor atorvastatin Synthetic Branded No Pfizer 43.15 2010

Lovastatin lovastatin Fermentation Generic 2.7

Mevacor lovastatin Fermentation Branded Yes Merck 1.12

Pravachol pravastatin Fermentation Branded No Bristol-Myers Squibb 11.44 2006

Zocor simvastatin Fermentation Branded No Merck 23.42 2006

Table 1. Major Brand of Statin Drugs

1 Market Share=number of prescriptions of the brand/(total number of prescriptions in 2004).

Sample Size Mean Sd Min Max 25th quantile median 75th quantile

Total Number of Prescriptions 616 20.54 10.39 10 75

Crestor Market Share 616 0.09 0.13 0 0.94 0.07 0.11 0.20

Lipitor Market Share 616 0.43 0.20 0 1.00 0.29 0.42 0.58

Zocor Market Share 616 0.24 0.16 0 0.91 0.14 0.24 0.36

Pravachol Market Share 616 0.13 0.13 0 0.73 0.08 0.14 0.22

Share of New Patients 616 0.10 0.12 0 1

Share of Older Patients (Age 65+) 616 0.47 0.20 0 1

Share of Asian Patients 616 0.02 0.07 0 1

Share of Female Patients 616 0.47 0.16 0.09 1

Share of Severe-symptom Patients 616 0.06 0.1 0 0.83

Number of Crestor Detailing 616 1.29 1.48 0 11

Number of Lipitor Detailing 616 0.83 1.12 0 6

Number of Zocor Detailing 616 0.89 1.15 0 5

Number of Pravachol Detailing 616 0.58 0.95 0 6

Table 2. Summary Statistics of Monthly Prescription and Detailing Visits

\

1 Brand Market Share=number of prescriptions of the brand/(total number of prescriptions). The quantile market share of a brand is calculated using markets with non-zero prescriptions

of that brand.

2 Number of Brand Detailing is the number of physician office visits by sales representatives of that brand.

[1] [2] [3] [4] [5] [6] [7]= [8]= [9]=

Before2

After3

Before After Before After [2]-[1] [4]-[3] [6]-[5] [7]-[8] [7]-[9]

Mean 0.76 0.92 0.73 0.81 0.82 0.89 0.16 0.09 0.07 0.07 0.09

Sd 0.90 1.22 1.06 1.10 0.97 1.13 1.18 1.04 0.95 0.18 0.24

N 126 126 176 176 61 61

Median Crestor share Low Crestor share High Crestor share

1 The outcome variable is the number of Lipitor detailing . We use Crestor's market share in the first 8 weeks of 2004 to construct the control group and the comparison groups. The

control group includes physicians with medium Crestor market share [0.05-0.18], and the comparison groups include physicians with low Crestor market share [0-0.05) and physicians

with high Crestor market share (0.18-1].

2 Before represents week 5-8 (February), the period before the bad news.

3 After represents week 10-13 (March), the period after the bad news.

Table 3. Difference in Difference Result I, Changes in Lipitor Detailing1

Dif-in-DifFirst Dif

Coef Sd Coef Sd Coef Sd Coef Sd

CM 1

0.12*** ( 0.03 ) 0.31*** ( 0.04 ) 0.09*** ( 0.03 ) 0.29*** ( 0.04 )

CH 0.02 ( 0.04 ) 0.4*** ( 0.05 ) -0.11*** ( 0.04 ) 0.48*** ( 0.06 )

CM*Postevent 0.07** ( 0.03 ) 0.05 ( 0.04 ) 0.08*** ( 0.04 ) 0.07 ( 0.04 )

CH*Postevent -0.13*** ( 0.05 ) -0.11*** ( 0.05 ) -0.05 ( 0.05 ) -0.12*** ( 0.05 )

Postevent 0.06*** ( 0.02 ) 0.05*** ( 0.02 ) 0.05*** ( 0.02 ) 0.07*** ( 0.02 )

Random Effect

Fixed Effect

Additional Control 2

Sample size

***,**, * denotes statistical significance at the 1 percent, 5 percent and 10 percent level respectively. 1

CM represents medium Crestor market share [0.07-0.2]; CH represents high Crestor market share (0.2-1]. Market share is based on

prescriptions in the previous month.

2 Additional control variables include: total number of prescriptions (market size), share of new patients, share of older patients (age

65+), share of Asian patients, share of female patients, share of severe-symptom patients.

Yes Yes

408

Yes

Yes

616 616408

Table 4. Baseline Results for Lipitor

Specification 4

Yes

Specification 1 Specification 3Specification 2

Yes

Coef Sd Coef Sd Coef Sd Coef Sd

Panel 1: Small Markets 1

CM 2

0.15*** ( 0.05 ) 0.22*** ( 0.07 ) 0.12*** ( 0.05 ) 0.2*** ( 0.07 )

CH 0.22*** ( 0.05 ) 0.37*** ( 0.09 ) 0.09 ( 0.06 ) 0.47*** ( 0.09 )

CM*Postevent 0.15*** ( 0.06 ) 0.47*** ( 0.07 ) 0.18*** ( 0.06 ) 0.5*** ( 0.08 )

CH*Postevent -0.28*** ( 0.07 ) -0.15*** ( 0.07 ) -0.19*** ( 0.07 ) 0.03 ( 0.08 )

Postevent 0.14*** ( 0.03 ) 0.04 ( 0.04 ) 0.13*** ( 0.04 ) 0.08*** ( 0.04 )

Random Effect

Fixed Effect

Additional Control 3

Sample size

Panel 2: Large Markets

CM 0.15*** ( 0.04 ) 0.49*** ( 0.05 ) 0.12*** ( 0.04 ) 0.44*** ( 0.05 )

CH -0.07 ( 0.05 ) 0.70*** ( 0.08 ) -0.20*** ( 0.05 ) 0.78*** ( 0.08 )

CM*Postevent -0.01 ( 0.04 ) -0.20*** ( 0.05 ) -0.02 ( 0.04 ) -0.22*** ( 0.05 )

CH*Postevent -0.05 ( 0.07 ) -0.14* ( 0.07 ) -0.01 ( 0.07 ) 0.03 ( 0.08 )

Postevent 0.04* ( 0.02 ) 0.06*** ( 0.02 ) -0.03 ( 0.02 ) 0.07*** ( 0.02 )

Random Effect

Fixed Effect

Additional Control

Sample size

Yes

Specification 1 Specification 2 Specification 3 Specification 4

Yes Yes

304 206 304 206

***,**, * denotes statistical significance at the 1 percent, 5 percent and 10 percent level respectively 1

Small markets represent physicians with no more than 20 total prescriptions in both January and February. All other physicians are

categorized as large markets. 2 CM

represents medium Crestor market share [0.07-0.2]; CH represents high Crestor market share (0.2-1]. Market share is based on

prescriptions in the previous month.

3 Additional control variables includes: total number of prescriptions (market size), share of new patients, share of older patients (age

65+), share of Asian patients, share of female patients, share of severe-symptom patients.

Table 5. Baseline Results for Lipitor: Small vs. Large Markets

Yes Yes

312 202 312

Yes

202

Yes

Yes Yes

Yes

Yes Yes

[1] [2] [3] [4] [5] [6] [7]= [8]= [9]=

Before2

After3

Before After Before After [2]-[1] [4]-[3] [6]-[5] [7]-[8] [7]-[9]

Mean 0.82 0.90 0.75 0.93 0.64 0.69 0.08 0.18 0.05 -0.10 0.03

Sd 0.93 1.12 1.08 1.25 0.94 1.03 1.04 1.15 1.03 0.18 0.20

N 157 157 126 126 80 801 The outcome variable is the number of Lipitor detailing . We use Pravachol's market share in the first 8 weeks of 2004 to construct the control goup and the comparison groups. The

control group includes physicians with medium Pravachol market share [0.06-0.20], while the comparison groups include physicians with low Pravachol market share (0-0.06) and

physicians with high Pravachol market share (0.20-1].

2 Before represents week 5-8, the period before the bad news.

3 After represents week 10-13, the period after the bad news.

Median Pravachol share Low Pravachol share High Pravachol share

Table 6. Difference in Difference Result II, Changes in Lipitor Detailing1

Dif-in-DifFirst Dif

[1] [2] [3] [4] [5] [6] [7]= [8]= [9]=

Before2

After3

Before After Before After [2]-[1] [4]-[3] [6]-[5] [7]-[8] [7]-[9]

Panel 1: Outcome variable is number of Zocor detailing

Mean 0.81 0.94 0.72 0.98 0.79 1.07 0.13 0.26 0.28 -0.13 -0.15

sd 1.16 1.24 1.02 1.23 0.95 1.29 1.23 1.10 1.17 0.19 0.26

N 126 126 176 176 61 61

Panel 2: Outcome variable is number of Pravachol detailing

Mean 0.60 0.77 0.49 0.47 0.43 0.66 0.17 -0.03 0.23 0.20 -0.06

sd 1.06 1.20 0.76 0.81 0.69 1.05 1.38 0.76 1.07 0.16 0.23

N 126 126 176 176 61 611 The outcome variables are the number of Zocor detailing and the number of Pravachol detailing , respectively. We use Crestor's market share in the first 8 weeks of 2004 to construct the control

group and the comparison groups. The control group includes physicians with medium Crestor market share [0.05-0.18], and the comparison groups include physicians with low Crestor market

share [0-0.05) and physicians with high Crestor market share (0.18-1].

2 Before represents week 5-8, the period before the bad news.

3 After represents week 10-13, the period after the bad news.

Median Crestor share Low Crestor share High Crestor share

Table 7. Difference in Difference Result III, Changes in Zocor and Pravachol Detailing1

First Dif Dif-in-Dif

[1] [2] [3] [4] [5] [6] [7]= [8]= [9]=

Before2

After3

Before After Before After [2]-[1] [4]-[3] [6]-[5] [7]-[8] [7]-[9]

Panel 1: Outcome variable is the number of Zocor detailings

Mean 0.86 1.05 0.85 1.17 1.00 1.42 0.19 0.32 0.42 -0.13 -0.23

sd 1.19 1.38 1.17 1.37 1.00 1.54 1.41 1.20 1.58 0.30 0.48

N 63 63 93 93 19 19

Panel 2: Outcome variable is the number of Pravachol detailings

Mean 0.73 0.75 0.77 0.58 0.80 0.95 0.02 -0.19 0.15 0.21 -0.13

sd 1.26 1.08 0.90 0.94 0.83 1.36 1.08 0.52 1.57 0.25 0.42

N 63 63 77 77 20 20

Median Crestor share Low Crestor share High Crestor share

1. The sample in Panel 1 includes only markets where Zocor's market share is above its median level. The sample in Panel 2 includes only markets where Pravachol's market share is above its

median level. The outcome variables are the number of Zocor detailing and the number of Pravachol detailing , respectively. We use Crestor's market share in the first 8 weeks of 2004 to construct

the control group and the comparison groups. The control group includes physicians with median Crestor market share [0.05-0.18], and the comparison groups include physicians with low Crestor

market share [0-0.05) and physicians with high Crestor market share (0.18-1].

2 Before represents week 5-8, the period before the bad news.

3 After represents week 10-13, the period after the bad news.

Table 8. Difference in Difference Result IV, Changes in Zocor and Pravachol Detailing (Reduced Sample) 1

First Dif Dif-in-Dif

Coef Sd Coef Sd Coef Sd Coef Sd

Panel 1: Outcome Variable: Number of Zocor Detailing

CM 1

-0.06*** ( 0.03 ) -0.22*** ( 0.04 ) -0.04 ( 0.03 ) -0.32*** ( 0.04 )

CH -0.12*** ( 0.04 ) -0.26*** ( 0.05 ) -0.12*** ( 0.04 ) -0.22*** ( 0.05 )

CM*Postevent -0.14*** ( 0.03 ) -0.14*** ( 0.04 ) -0.14*** ( 0.03 ) -0.01 ( 0.04 )

CH*Postevent 0.05 ( 0.04 ) -0.09** ( 0.04 ) 0.17*** ( 0.04 ) 0.08 ( 0.05 )

Postevent 0.23*** ( 0.02 ) 0.26*** ( 0.02 ) 0.21*** ( 0.02 ) 0.26*** ( 0.02 )

Random Effect

Fixed Effect

Additional Control 2

Sample size

Panel 2: Outcome Variable: Number of Pravachol Detailing

CM 0.31*** ( 0.03 ) 0.37*** ( 0.04 ) 0.4*** ( 0.03 ) 0.4*** ( 0.04 )

CH -0.3*** ( 0.04 ) -0.31*** ( 0.06 ) -0.24*** ( 0.05 ) -0.36*** ( 0.07 )

CM*Postevent 0.02 ( 0.04 ) -0.09*** ( 0.04 ) -0.02 ( 0.04 ) -0.09** ( 0.05 )

CH*Postevent 0.36*** ( 0.05 ) 0.41*** ( 0.06 ) 0.43*** ( 0.06 ) 0.47*** ( 0.06 )

Postevent 0.07*** ( 0.02 ) 0.09*** ( 0.02 ) 0.05*** ( 0.02 ) 0.1*** ( 0.02 )

Random Effect

Fixed Effect

Additional Control

Sample size

Yes

Yes

616

Yes

Specification 4Specification 3

Yes

Yes

Specification 1 Specification 2

Yes

Yes Yes

Table 9. Results for Zocor and Pravachol

***,**, * denotes statistical significance at the 1 percent, 5 percent and 10 percent level respectively.

1. CM represents median Crestor market share [0.07-0.2]; CH represents high Crestor market share (0.2-1]. Market share is based on

prescriptions in the last 4 weeks.

2 Additional control variables includes: total number of prescriptions (market size), share of new patients, share of older patients (age

65+), share of Asian patients, share of female patients, share of severe-symptom patients.

616 400 400

616

Yes Yes

322 322

YesYes

616

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

1.05Figure 1: Nonparametric Estimation of Lipitor Detailing

Quantile of Crestor Market Share

Num

ber o

f Lip

itor D

etai

ling

Before

After

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1Figure 2: Nonparametric estimation of Lipitor detailing

Quantile of Pravachol Market Share

Num

ber o

f Lip

itor D

etai

ling

BeforeAfter

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.7

0.8

0.9

1

1.1

1.2

1.3

1.4Figure 3: Nonparametric Estimation of Zocor Detailing

Quantile of Crestor Market Share

Num

ber o

f Zoc

or D

etai

ling

Before

After

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.4

0.45

0.5

0.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9Figure 4: Nonparametric Estimation of Pravachol Detailing

Quantile of Crestor Market Share

Num

ber o

f Pra

vach

ol D

etai

ling

Before

After