Modeling the effects of marketing expenditures from
pharmaceuticals on brand sales: the role of marketing
dynamics within the pharmaceutical market.
by
Erik Plat
MODELING THE EFFECTS OF MARKETING
EXPENDITURES FROM PHARMACEUTICALS ON
BRAND SALES
The role of marketing dynamics within the
pharmaceutical market
Author: Erik Plat
Faculty of Economics and Business
Msc Marketing
Completion date: June 23
d
, 2014
Address: Korreweg 129a
9714 AG GRONINGEN
Phone: +31613216392
Mail: e.plat91@gmail.com
student number: 1874470
Abstract
Table of content
Preface……….………..p.4 1 Introduction………..……….p.5 2 Literature review and hypotheses………..………..p.7
2.1 Own marketing effects……….………p.7
2.1.1 Own price 2.1.2 Own detailing
2.1.3 Own journal advertising
2.1.4 Own physician meeting expenditures 2.1.5 Own direct-to-consumer advertising
2.2 Competitive marketing effects………..p.10
2.2.1 Competitive price 2.2.2 Competitive detailing
2.2.3 Competitive journal advertising
2.2.4 Competitive physician meeting expenditures. 2.2.5 Competitive direct-to-consumer advertising
2.3 Lagged effects……….………p.13
2.3.1 Lagged price 2.3.2 Lagged detailing
2.3.3 Lagged direct-to-consumer advertising
2.3.4 Lagged journal advertising and lagged physician meeting expenditures
2.4 Interaction effects………..p.14
3 Data and methodology………..………p.16
3.1 Conceptual model………..p.16 3.2 Functional form………p.17 3.3 Descriptive statistics……….………p.18 4 Model specification……….………….………p.24 4.1 Model comparison………..……….………..p.24 4.2 Data preparation………..p.25
4.2.1 Correlation between explanatory variables 4.2.2 Unit root test
4.2.3 Autocorrelation 4.2.4 Heteroscedasticity 4.2.5 Nonnormality 5 Estimation………..p.29 6 Validation………p.31 6.1 Face validity……….p.31 6.2 Predictive validity………..……….p31 6.3 Multicollinearity………..……….p32
7 Conclusions and limitations………..……….p.33
7.1 Limitations………..……….p.33 7.2 Managerial recommendations………..………p.33 7.3 Research recommendations……….………p.34
Preface
This Master Thesis is the final part of my study Msc Marketing Intelligence at the
Rijksuniversiteit Groningen. The useful skills and knowledge that I got from the marketing courses are now used by me in writing my thesis. I would like to thank dr. Maarten Gijsenberg for giving me input for choosing this thesis topic and for supervising my work while I was working on my thesis. I also want to thank prof. dr. Jaap Wieringa for providing me the data about the pharmaceutical industry in the United States, where I based my research on.
Erik Plat
1. Introduction
The pharmaceutical market in the United States is one of the largest and most competitive markets in the country (Leeflang & Wieringa, 2013). While the pharmaceutical markets in most European countries are still much regulated by their government, pharmaceutical companies in the United States are given much freedom in their marketing efforts for prescription drugs (Fischer & Albers, 2010). Since the early 1990s, direct-to-consumer advertising expenditures in the United States exploded from $ 985 million in 1996 to $ 4.2 billion in 2005 (Bradford et al, 2010). Detailing expenditures, which are visits from pharmaceutical sales representatives to physicians, are even higher with an amount of $ 5.8 billion in 2004 (Mizik & Jacobson, 2004). Together with the enormous investments in development of new drugs, the pharmaceutical industry accounted for 17% of the GDP in the United States by 2011 (Leeflang & Wieringa, 2013).
The magnitude of the expenditures in the pharmaceutical industry has led to some public controversy since the last decades (Kremer et al, 2008). Opponents disapprove the marketing expenditures of pharmaceutical companies by saying they are way too excessive, especially because precise effects of these marketing efforts are still debated among marketing researchers (Kremer et al, 2008). Furthermore, critics say that the marketing efforts encourage physicians to prescribe expensive drugs over cheaper generic drug, which comes at the cost of patient welfare (Mizik & Jacobson, 2004). Supporters of pharmaceutical marketing however state that marketing actions provide an opportunity to recover the enormous R&D costs that come with the development of new products. Besides this, pharmaceutical marketing is a good way to communicate with physicians about drug information and to improve knowledge and health outcomes for customers when they use or are interested to use the company’s products (Mizik & Jacobson, 2004).
From a marketing point of view, the pharmaceutical industry differs substantially from other industries. First, sales of prescription drugs to customers can only be made via prescriptions of physicians. Because of this, both physicians and consumers are targeted by pharmaceutical companies (Kremer et al, 2008). This is especially the case for the pharmaceutical market in the United States, where one part of the marketing actions are particularly directed at phycisians, like advertising via medical journals, arranging meetings with sales representatives and organizing conventions for physicians (Narayanan et al, 2005). On the other hand, pharmaceutical companies also target customers by advertising their products via TV, billboards or internet (Fischer & Albers, 2010). Directly advertising to customers is however still uncommon in European countries and other countries outside of the United States (Stremersch & Lemmens, 2008). The unique
provider-customer structure between physicians and patients is likely to cause different market dynamics than in other industries (like fast-moving consumer goods or electronics), and provides challenges to pharmaceutical companies in how to target both of the groups.
Because of the developments and controversy about the marketing expenditures of pharmaceutical companies, managers desire to get better insights in the effectiveness of promotional instruments in the pharmaceutical industry (Gonul et al, 2001; Narayanan et al, 2004; Leeflang & Wieringa, 2013). Moreover, the nature of the pharmaceutical industry gained interest from researchers in the marketing field. Besides that pharmaceutical promotions are directed to both physicians and consumers, pharmaceutical companies spend a large percentage of their revenues on marketing (Kremer et al, 2008). Lastly, the pharmaceutical industry differs from other industries by a much larger number of product introductions (Kremer et al, 2008). These unique features of the pharmaceutical industry makes it interesting to specifically investigate the effectiveness of
Until now, some different types of studies within the marketing research field have developed in the investigation of the effectiveness of pharmaceutical marketing efforts on sales (Kremer et al, 2008). One stream of literature focuses on the physician prescriptions, unit sales or category sales as dependent variables to assess the effectiveness of pharmaceutical marketing efforts. The other types of studies about pharmaceutical marketing specifically focus on the role of price, new product introductions or the information and persuasion function of pharmaceutical advertising (for a complete overview: see Kremer et al, 2008). This thesis primarily focuses on the first aspect of pharmaceutical marketing research. The last three types are beyond the scope of this research and are only mentioned on a basic level further in this research.
Most of the articles that investigate the influence from pharmaceutical marketing expenditures on sales mainly look at the own marketing efforts of pharmaceutical companies (Fischer & Albers, 2010). Concerning literature that uses physician prescriptions as unit of analysis, they specifically focus on marketing instruments directed to physicians. These marketing instruments are detailing, sampling, medical journal advertising and physician meeting expenditures (Kremer et al, 2008; Narayanan et al, 2005). Literature that focuses on drug category sales or brand sales also investigates the effect of own direct-to-consumer advertising, together with the aforementioned variables.
Besides the research on direct effects of own marketing efforts on sales of pharmaceuticals, some researchers attempt to include marketing dynamics in their models. For example, Narayanan et al (2004) and Kremer et al (2008) investigate the presence of interaction effects between different types of marketing expenditures. Amaldoss and He (2009) analyze effects of competitive marketing actions on own sales and Mizik & Jacobson (2004), Kolsarici & Vakratsas (2010) and Sethuraman et al (2011) analyze possible lagged effects of the different marketing efforts of pharmaceutical
companies on brand sales.
Despite the growing amount of literature on the effects of pharmaceutical marketing efforts, some important managerial and research insights are still missing in existing literature. First, the retail prices of the drugs in the United States are fairly high. This has led to a relatively high entry of generic drugs, which are cheaper alternatives and can be a competitive threat for branded drugs in the market (Fischer & Albers, 2010; Leeflang & Wieringa, 2013). Although the increased competitive threat from generic drugs is acknowledged by some researchers, specific effects of marketing efforts from generic drugs versus those from branded drugs are only sparsely investigated in the
pharmaceutical market (Fischer & Albers, 2010; Kolsarisi & Vakratsas, 2010).
Lastly, despite that there is some attention for lagged effects of pharmaceutical marketing variables, the marketing literature does not focus on lagged effects of all of these marketing instruments. More specifically, only the lagged effects of price, detailing and direct-to-consumer advertising are
investigated. To the best of my knowledge, lagged effects of other marketing instruments like physician meetings and journal advertising are not investigated. This is remarkable because in general, all marketing mix variables can have a lagged effect on brand sales (Doyle & Saunders, 1985). In other words, it is not unthinkable that journal advertising and physician meetings have a significant lagged effect on brand sales. Therefore, this research wants to focus on lagged effects of all marketing variables.
The goal of this research is to contribute to the existing research about the effectiveness of pharmaceutical marketing expenditures on brand sales via physicians. Unlike most of the current literature, this research wants to place most emphasis on marketing dynamics within the
effects between marketing efforts of generic drugs and branded drugs. By including both own marketing effects and dynamic marketing effects, this research hopes to give a better view and more complete view about the marketing dynamics within the pharmaceutical market.
The outline of this thesis is formed by first doing a literature review of the current literature about pharmaceutical promotions in the next chapter. After the literature is reviewed and hypotheses are formed, the collected data is described and illustrated with graphs and tables to get a first view about the marketing expenditures from the pharmaceutical companies. In chapter four, a final regression model is specified and the parameters of the variables are estimated. In the fifth chapter, the results will be interpreted and validated. In the final part, conclusions of the results are formed and academic and managerial implications will be formed. In addition, a discussion and limitations from the research will be covered in the last part of the thesis.
2. Literature review and hypotheses
2.1 Own marketing efforts
In the pharmaceutical industry, the marketing efforts of companies are somewhat different than the traditional four Ps known in marketing (Fischer & Albers, 2010). The most important factors that influence drug sales are detailing (e.g. visits from pharmaceutical sales representatives to physicians), journal advertising and direct-to-consumer advertising (Fischer & Albers, 2010; Leeflang & Wieringa, 2013). Besides these three factors, pharmaceutical companies also organize physician meetings and events as an instrument to promote their products to physicians (Narayanan et al, 2005).
2.1.1 Own price
In general, one of the marketing mix variables that is most often investigated in marketing literature is the price variable. The influence of price on sales is also investigated in the pharmaceutical industry (e.g. Fischer et al, 2011; Narayanan et al, 2004; Kremer et al, 2008). Compared with other industries, price plays a special role in the pharmaceutical market. The first reason for this is that sales of prescription drugs are made via physicians, who decide which drug is prescribed to their patients (Kremer et al, 2008). Therefore, the price sensitivity of physicians is more important than the price sensitivity of customers, who usually do not compare prices about different drugs that can be prescribed for their treatment.
The physician-patient structure encouraged researchers to particularly look at the price sensitivity of physicians (Kremer et al 2008; Gonul et al, 2001). Intuitively, one would say that physicians want to prescribe the best quality drugs to their patients and are therefore are not much effected by the price of the drug. Existing research finds that just like with end customers, physicians also are price sensitive (Gonul et al, 2001). In accordance with consumers in general, price sensitivity differs across physicians (Montoya et al, 2009) and the price sensitivity of physicians can be lowered by detailing and other marketing efforts (Kremer et al, 2008). The difference in price sensitivity of physicians is on one hand explained by the drug category. In other words, price sensitivity for cancer treatment drugs is far lower than for drugs that lower cholesterol levels (Amaldoss & He, 2009). On the other hand, personal characteristics of physicians are also causing differences in their price sensitivity (Montoya et al, 2009). This means that some physicians naturally react more negatively to price increases than other physicians.
pharmaceuticals (Kremer et al, 2008). Most articles however find that price has a significant and negative impact on sales of pharmaceutical companies (Fischer et al, 2011; Narayanan et al, 2004; Sethuraman et al, 2011). Besides that most articles agree that the influence of drug prices in generally negative for drug sales, marketing scholars also acknowledge that the price elasticity from physicians can be lowered by increasing marketing expenditures (Kremer et al, 2008). In addition, when the health insurance of patients does not cover the costs of using the medicine, direct-to-consumer advertising can decrease price sensitivity of patients (Leeflang & Wieringa, 2013). In conclusion, price has a somewhat less straightforward role on sales than in other industries. Nevertheless, most articles find that physicians in general respond negatively to price increases. This means that like in other industries, price also has a negative effect on sales of pharmaceutical companies. Therefore, hypothesis 1a is stated as follows:
H1a: Own price has a negative impact on brand sales.
2.1.2 Own detailing
Detailing is one of the most important marketing instruments for pharmaceutical companies (Gonul et al, 2001; Bala et al, 2013). Pharmaceutical companies have the opportunity arrange meetings between salespersons and physicians in order to persuade physicians to make use of their products (Gonul et al, 2001). This marketing instrument is also one of the most persuasive instruments that pharmaceutical companies can use, since the meetings between salespersons and physicians are on a direct one-to-one basis (Bala et al, 2013). Therefore, many pharmaceutical companies make use of detailing and spend very high budget on this type of marketing instrument with the goal to increase drug sales (Gonul et al, 2001; Kremer et al, 2008).
Findings about the effects of detailing are a bit mixed in the current literature (Kremer et al, 2008). For example, Kolsarici & Vakratsas (2010) find a strong positive relationship between detailing and drug sales. On the contrary, Fischer & Albers (2010) do not find a significant influence of detailing on brand sales. An remarkable note with these contrary findings is that at the physician level, e.g. the number of prescriptions from physicians, detailing has the strongest and most positive effect relative to all other pharmaceutical marketing instruments (Kolsarici & Vakratsas, 2010). When the focus of the analysis is on unit sales however, findings of significant detailing effects are more mixed (Kremer et al, 2008).
In summary, regarding the articles that do significant results of detailing, they all agree that the effect of detailing on brand sales is positive. In line with these results, this research also expects that own detailing will have a positive effect on brand sales. Since this type of marketing variable is one of the most persuasive marketing instruments for pharmaceutical companies, it is likely that detailing will positively influence physicians in their choice to prescribe the promoted drug. Therefore, hypothesis 1b is stated as:
H1b: Detailing has a positive impact on brand sales.
2.1.3 Own journal advertising
physicians information about the products of the company (Montoya et al, 2009). Second, medical journals can also have an persuasive function to physicians by promoting drugs to physicians with strong arguments about the product quality, benefits etcetera (Kremer et al, 2008; Montoya et al, 2009). One advantage from journal advertising over other marketing instruments is that it reaches a large audience of physicians (Kremer et al, 2008). Because of this, pharmaceutical companies are encouraged to use medical journals as a medium to advertise their products to physicians.
Like other industries however, advertising through online channels in the pharmaceutical industry is also increasing at the cost of traditional media (Sethuraman et al, 2011). Nevertheless, marketing literature in general still finds evidence that advertising through traditional media can positively influence brand sales (Sethuraman et al, 2011). Regarding the pharmaceutical industry, existing literature about the effects of medical journal advertising on brand sales provide some mixed results. For example, Narayanan et al (2004) do not find a significant influence of medical journal advertising on brand sales. In contrast, Kolsarici & Vakratsas (2010) do find a significant positive influence of journal advertising on brand sales. However, they find that the effect of journal advertsing is only low. Others, like Fischer & Albers (2010) and Kremer et al (2008) find a stronger influence of own journal advertising on brand sales. In other words, researchers do not fully agree with each other about the exact strength of medical journal advertising as a marketing instrument.
In conclusion, this research expects a positive relationship between own journal advertising and brand sales, in line with the findings Kolsarici & Vakratsas (2010), Fischer & Albers (2010) and Kremer et al (2008). This research expects that journal advertising can have a persuasive effect on physicians and increase the chance that physicians will prescribe the company’s brand. Hence, hypothesis 1c is formed as:
H1c: Own journal advertising has a positive impact on brand sales.
2.1.4 Own physician meeting expenditures
Physician meeting expenditures (PME) is the fourth promotional instrument where this research pays attention to. PME are promotional activities from pharmaceutical companies directed at physicians during organized meetings, conventions or symposia (Narayanan et al, 2005; Manchanda et al, 2005). During these events, pharmaceutical companies send experts to give information and promote their products to physicians. Unlike detailing, this type of promotion is not used on an one-to-one basis, but is used for groups of physicians (Narayanan et al, 2005). Therefore, the nature of physician meeting expenditures as a marketing instrument is less persuasive than detailing.
Regarding the literature about pharmaceutical marketing, one can see that research about physician meeting expenditures is substantially lower than with detailing and journal advertising. This is remarkable, because pharmaceutical companies do spend budget on this marketing instrument on a regular (monthly) basis (Narayanan et al, 2005; Manchanda et al, 2005). Nevertheless, Narayanan et al (2005) and Manchanda et al (2005) do not explicitly focus on the influence of physician meeting expenditures on brand sales. Kremer et al (2008) classify the expenditures on this marketing instrument as ‘other marketing expenditures’ where it is combined with other expenses. In their results, they find a small positive impact of ‘other marketing expenditures’ on sales of
H1d: Own physician meeting expenditures have a positive impact on brand sales.
2.1.5 Own direct-to-consumer advertising (DTCA)
Direct-to-consumer advertising is the last promotional instrument that pharmaceutical companies can make use of. This type of advertising is targeted directly towards patients, with advertisements of drugs via TV, Internet, radio, newspapers, etcetera. This means that patients can be informed and persuaded about the company’s drugs without a physician who normally stands between the pharmaceutical company and the patient. Because of this, DTCA has become increasingly important and has become the second most important marketing expenditure for pharmaceutical companies, after detailing (Kremer et al, 2008). As a result, studying the effects of DTCA has also become a popular topic in pharmaceutical marketing literature (Kremer et al, 2008).
Almost all literature about direct-to-consumer adverting has found that this marketing instrument positively influences sales of the drug category (Kremer et al, 2008). In other words, the overall demand for a drug increases with an increase in DTCA expenditures from pharmaceutical companies in that particular drug category. Research findings about specific effects of direct-to-consumer advertising on brand sales are however a bit more inconclusive. For example, Fischer & Albers (2010) do not find a significant influence from DTCA on brand sales. Moreover, Kremer et al (2008) find that for some drug categories, effects of DTCA are even negative. This is because the Food and Drug Administration in the United States forces pharmaceutical brands to mention the negative side effects of their products in their advertising. According to Kremer et al (2008), this may be be caused by negative reactions from customers when they get to know these side effects and discourages them from using the drug. In contrast to the previous finding, Narayanan et al (2004) and Kolsarici & Vakratsas (2010) find that DTCA does have a significant positive effect on brand sales.
In summary, there are some mixed results about the effects of direct-to-consumer advertising on brand sales. This research expects that DTCA has a positive influence on brand sales, in line with Narayanan et al (2004) and Kolsarici & Vakratsas (2010). I expect that the positive informational and persuasive aspects of direct-to-consumer advertising will outweigh the negative aspect of having to display side effects. Hence, hypothesis 1e is stated as:
H1e: Own DTCA has a positive impact on brand sales.
2.2 Competitive effects
The literature about competitive marketing efforts in the pharmaceutical industry mainly focuses on effects on competitive price (Fischer et al, 2011) and competitive direct-to-consumer advertising (Amaldos & He, 2009; Stremersch et al, 2013). In addition, Dong et al (2009) and Fischer & Albers (2010) focus on competitive detailing and competitive journal advertising, respectively. As far as I know, no one has yet investigated effects of competitive physician meeting expenditures.
Literature about competitive marketing efforts says competitive marketing efforts either can have a substitution effect on the one hand and can increase overall demand for the product category on the other hand (Fischer & Albers, 2010; Osinga et al, 2010; Anderson & Simester, 2013). Depending on the persuasiveness of the marketing efforts, either the substitution effect or the category expansion effect is stronger, which determines that the impact of this marketing effort is negative of positive for own brand sales, respectively (Fischer & Albers, 2010).
When determining the role of competitive price on sales, two aspects have to be taken into account (Fischer & Albers, 2010). The first aspect is that competitive prices (or own price) can have a
substitution effect (Fischer & Albers, 2010). This means that when competitive prices increase, your own sales will increase as customers will substitute the competitive product for your product. The other aspect is that price levels can generate a market-expanding or market-declining effect (Fischer & Albers, 2010). This means when price levels increase, overall demand in the market will decline. Sethuraman et al (2011) found that in general, increasing competitive prices generate stronger substitution effects than market-declining effects. This means that normally, increasing competitive prices will have a positive influence on a company’s own brand sales.
For the pharmaceutical industry however, Fischer & Albers (2010) find that competitive prices have a significant negative impact on own brand sales. This finding is again confirmed by a later article of Fischer et al (2011). These both findings indicate that the market-declining effect of price in the pharmaceutical market is stronger than the substitution effect of price, which causes a negative influence on competitive price on own brand sales (Fischer & Albers, 2010). This may seem a bit counterintuitive, because substitution effects are stronger than market-declining effects in many other markets (Sethuraman et al, 2011). Fischer & Albers (2010) motivate their findings however that the price sensitivity of customers in the pharmaceutical market are often inelastic. This makes it less likely that sales significantly increase with increasing competitive prices (Fischer & Albers, 2010). In line with the findings of Fischer & Albers (2010) and Fischer et al (2011), this research also expects a negative relationship between competitive price and brand sales. Hence, hypothesis 2a is stated as follows:
H2a: competitive price has a negative impact on sales.
2.2.2 Competitive detailing
Competitive detailing is the third competitive element that this research investigates. Looking only at the persuasiveness of detailing, it is known that this marketing instrument is one of the most
persuasive instruments for pharmaceutical companies (Gonul et al, 2001). This makes it is likely that the substitution effect of this marketing instrument will be stronger than its market expansion effect (Anderson & Simester, 2013). Moreover, the main goal of detailing is not to increase overall demand, but only to persuade physicians to prescribe the company’s drugs (Kremer et al, 2008). This
intuitively means that a possible market expansion effect will be really weak, if not absent. Hence, when competitive detailing has a significant impact, it probably has a (negative) substitution effect on own brand sales.
In the current literature, Gonul et al (2001) and Dong et al (2009) have done research on the effect of competitive detailing on sales of pharmaceutical companies. Both Gonul et al (2001) and Dong et al (2009) analyze the effect of this competitive marketing variable on the prescription behaviour of physicians. Both studies agree that competitive detailing has a strong substitution effect on the frequency of drug prescriptions. In other words, increases in competitive detailing have a negative influence on the number of own drug prescriptions.
With the results of these two studies, it is indeed confirmed that competitive detailing has a
significant negative effect on brand sales. When a company thus wants to attract patients from their competitors, it is beneficial to increase the amount of detailing. In line with the results of Gonul et al (2001) and Dong et al (2009), this research expects a negative relationship between competitive detailing and brand sales. Therefore, hypothesis 2b is stated as follows:
2.2.3 Competitive journal advertising
Unlike detailing, the medical journal advertising marketing instrument is not always used in a persuasive manner. Hence, it may be interesting to know if competitive journal advertising has a negative effect on own brand sales or that competitive journal advertising has a positive effect on own brand sales by increasing overall demand for the drug category. Fischer & Albers (2010) and Kolsarici & Vakratsas (2010) made an attempt to answer this question by analyzing the effect of competitive journal advertising on brand sales. Kolsarici & Vakratsas (2010) did not find any
significant effects of competitive journal advertising on brand sales in their study. However, Fischer & Albers (2010) found that competitive journal advertising has a negative effect on brand sales. Their result implies that medical journal advertising is often used in a persuasive manner. Fischer & Albers (2010) find that this is especially the case for competitors who enter the market in a later stage, in an attempt to encourage physicians to switch to their brand.
Concluding, while Fischer & Albers (2010) find a significant negative impact from competitive journal advertising on brand sales, there is no overwhelming evidence that confirms these findings, simply because so few articles have investigated the relationship between competitive journal advertising and pharmaceutical sales. Nevertheless, this research expects that competitive journal advertising will have a significant negative impact on brand sales, in line with the results of Fischer & Albers (2010). Therefore, hypothesis 2c is stated as:
H2c: competitive journal advertising has a negative impact on brand sales.
2.2.4 Competitive physician meeting expenditures
Competitive physician meeting expenditures is the fourth competitive variable that is included in this research. Remarkably, no articles have yet investigated the impact of competitive physician meeting expenditures on own brand sales. This makes physician meeting expenditures the only marketing instrument where researchers do not pay attention to its competitive effect. Therefore, this
hypothesis is formed on the basis of the generalizations that are found for the previous competitive variables. This research expects that physician meetings are not as persuasive as detailing, because physician meetings and events are not on a one-to-one basis like detailing. Moreover, the nature of physician meetings and events are more for informing physicians about product benefits than pharmaceutical companies want to persuade physicians to that they should prescribe the company’s drug (Narayanan et al, 2005). Because of this, this research expects that physicians meeting
expenditures are more likely to increase the overall category demand than that they substract sales from competitive brands. Therefore, hypothesis 2d is stated as follows:
H2d: Competitive physician meeting expenditures have a positive impact on brand sales.
2.2.5 Competitive direct-to-consumer advertising
With the large increase in overall direct-to-consumer advertising expenditures, researchers are not only interested in its direct effects, but also in the competitive effects of this marketing instrument. Amaldoss & He (2009) and Stremersch et al (2013) both investigated if competitive
direct-to-consumer advertising had a positive or negative effect. The findings of Amaldoss & He (2009) showed that an increase in competitive DTCA caused an increase in own brand sales. This finding thus proved that DTCA expenditures caused an increase in overall demand for the drug category. Hence,
direct-to-consumer advertising is indeed less persuasive than for example detailing, so it will not be very likely that customers are attracted from competitors with this marketing instrument (Anderson &
Simester, 2013).
In conclusion, there is consensus in the literature about the positive influence from competitive direct-to-consumer advertising on brand sales. Therefore, hypothesis 2e will be formed as: H2e: competitive DTCA has a positive impact on brand sales.
2.3 Lagged effects.
Besides that own marketing efforts of pharmaceutical companies have a direct effect on brand sales, they also can have a lagged effect (Mizik & Jacobson, 2004; Kolsarici & Vakratsas, 2010). In this section, sparse literature about lagged effects of the different marketing instruments is reviewed.
2.3.1 Lagged price
First, there will be taken a look at lagged price. In general, marketing literature acknowledges that price does have a lagged effect, but that there is some discussion in the magnitude of the lagged effect of price (Sethuraman et al, 2011). For example, Pauwels et al (2002) find that the lagged effect of price is less strong than the direct effect of price. One reason of this is that in other markets (e.g. consumer packaged goods) can have post promotional dips, which causes lower responsiveness to price changes after a period of price promotions (Pauwels et al, 2002). However, price promotions normally do not exist in the pharmaceutical market (Dekimpe & Hanssens, 1999). Therefore, it is not likely that the lagged effect of price will be substantially lower than the direct effect of price. Instead, there often is price competition between cheap generic drugs and branded drugs. Dekimpe & Hanssens (1999) find a significant negative lagged effect of price, which is even more negative than the direct price effect for brand sales. So, hypothesis 3a is stated as follows:
H3a: Lagged price has a negative impact on brand sales.
2.3.2 Lagged detailing
Mizik & Jacobson (2004) investigate lagged effects of detailing on new prescriptions by physicians for three different drug brands. The researchers investigate lagged effects of detailing in six direct lag variables, in which each lag term represents one month (Mizik & Jacobson, 2004). In their results they find that lagged detailing and lagged sampling effects are positive for all three brands and increase for new drugs over time and deteriorate over time for older drugs (Mizik & Jacobson, 2004). Moreover, they prove that current detailing does not only positively affect brand sales with a lag of one month, but also in more lags (Mizik & Jacobson, 2004). In line with these results, hypothesis 3b is stated as follows:
H3b: Lagged detailing has a positive impact on brand sales.
2.3.3 Lagged direct-to-consumer advertising
effect. More specifically, Osinga et al (2010) found significant lagged effects for five lag terms of direct-to-consumer advertising. In conclusion, the majority of the articles has found insignificant results for lagged DTCA. Nevertheless, the research of Sethuraman et al (2011) and Osinga et al (2010) did find a significant positive relationship between lagged direct-to-consumer advertising and brand sales. Therefore, this research still expects that a lagged effect of direct-to-consumer
advertising exists and that it will have a positive impact on brand sales. Hence, hypothesis 3c will be stated as:
H3c: Lagged DTCA has a positive impact on brand sales.
2.3.4 Lagged journal advertising and lagged physician meeting expenditures
Finally, lagged effects of journal advertising and physician meeting expenditures are also included in this research. To the best of my knowledge, no articles have yet investigated a possible lagged effect for each of these two types of expenditures. For journal advertising, it may be the case because journal advertising only has a small impact on sales, so a possible lagged effect is probably even lower, according to Fischer & Albers (2010). Regarding physician meeting expenditures, it was already unpopular among researchers to study the direct effect of the marketing instrument. Nevertheless, both lagged journal advertising and lagged physician meeting expenditures will be included in this research. The first reason for including these variables is because it is not likely that effects of marketing efforts are limited to one period (Mizik & Jacobson, 2004). Also, looking at the effect advertising on sales in general, its lagged effects can be even higher than its direct effect (Doyle & Saunders, 1985). In line with a positive direct effects of journal advertising and physician meeting expenditures, this research expects that lagged journal advertising and lagged physician meeting expenditures also will have a positive impact on sales. Hence, hypothesis 3d and 3e are stated as:
H3d: Lagged journal advertising has a positive impact on brand sales.
H3e: Lagged physician meetings expenditures have a positive impact on brand sales.
2.4 Interaction effects between own and generic marketing expenditures
Generic drugs have the same ingredients as their branded competitors, but they do not carry a brand name (Wieringa et al, 2014). In addition, generic drugs are also substantially cheaper than their branded equivalents. In this way, generic drugs have a bit in common with private label brands in other markets, since private label brands do not carry a brand name and also are typically cheaper than national brands (Lamey et al, 2007).
expenditures from branded drugs, they find no significant results. This research wants to take a better look on interaction effects between generic and branded marketing expenditures by
specifically investigating possible interaction effects per individual marketing instrument of branded and generic drugs.
First, this section looks at the influence of price from generic competitors on own marketing effectiveness. Fischer & Albers (2010) find that price competition from generic drugs makes physicians overall more sensitive to price changes. In other words, generic price competition in the market can cause physicians to react more negatively to own price increases. In line with this result, this research expects that that generic price and own price will have negative interaction with each other. Therefore, hypothesis 4a is stated as follows:
H4a: Generic price has a negative moderating effect between price and brand sales.
Until now, a specific moderating effect from generic detailing on own detailing and brand sales is not investigated. However, Vakratsas & Kolsarici (2008) have investigated a moderating effect of the amount of market saturation between own detailing and the diffusion rate of drugs. They found that when the market becomes more saturated, the effectiveness of detailing becomes insignificant (Vakratsas & Kolsarici, 2008). According to Vakratsas & Kolsarici (2008), the ineffectiveness of detailing is explained due to saturation effects of the detailing efforts when overall detailing expenditures in the market increase.
Furthermore, Shankar et al (1998) found evidence that an increase in the marketing expenditures from later entrants in the market (e.g. generic competitors) can cause the marketing efforts of the earlier entrants to be less effective (Wieringa et al 2014). This can suggest that own detailing effectiveness is negatively moderated by an increase in detailing from (generic) competition. In conclusion, this research expects that when generic competition increases its detailing expenditures, the effect of branded detailing will saturate and therefore will be less effective. Hence, hypothesis 4b is stated as follows:
H4b: Generic detailing has a negative moderating effect between own detailing and brand sales. Generic journal advertising is the last interaction effect that is included in this research. As far as I know, there is no literature available about interaction effects between own journal advertising and generic journal advertising. When marketing literature about other industries is investigated, there is evidence that negative interaction effects between competitive advertising instruments exist (Naik et al, 2005). In line with this finding, I also expect that journal advertising from generic drugs will have a negative moderating effect between own journal advertising and brand sales. Hence, hypothesis 4c is stated as:
H4c: Generic journal advertising has a negative moderating effect between own journal advertising and brand sales.
3. Data and methodology
The data that is used are time series data that contain 96 months of sales from branded and generic drugs to physicians in the antidepressant category of the pharmaceutical market in the United States. The observations cover the period between January 1993 and January 2001. For every month, data is available of the revenues, drug price, detailing expenditures, journal advertising expenditures, physician meeting expenditures and DTCA expenditures. This research will use sales data from one focal brand, data from two competing brands and data from five generic brands in the hypertension drug category. Competitive price is the average price of brands Effexor and Prozac. The remaining competitive variables are the total expenditures of Effexor and Prozac on detailing, journal
advertising, physician meetings and direct-to-consumer advertising. Generic variables in the model are defined in the same way as the competitive variables, with Amitriptyline, Doxepin, Imipramine, Nortripyline and Trazodone as generic competitors. The data will be analyzed with a multivariate regression analysis by ln-transforming the initial equation (see equation 1 on the next page).
3.1 Conceptual model
Own marketing efforts
Price (-) Detailing (+)
Physician meeting exp. (+) DTCA (+)
Journal advertising (+)
Competitive marketing efforts Comp. Price (-)
Comp. Detailing (-)
Comp. Physician meeting exp. (+) Comp. DTCA (+)
Comp. Journal advertising (-)
Lagged effects
Lagged price (-) Lagged detailing (+)
Lagged physician meeting exp. (+) Lagged DTCA (+)
Lagged Journal advertising (+)
Brand sales
Moderating variables
Generic price x own price (-) Generic detailing x own detailing (-)
Generic journal advertising x own journal adv. (-)
Where
i = brand specific term
t (1,..,96) = time period (in months) β0= brand specific intercept Sit = Sales of brand i at time t OPit= Own price
DTLit = own detailing JAD= own journal advertising
PME = own physician meeting expenditures DTCA = own direct-to-consumer advertising CPR= competitive price
CDTL = competitive detailing CJAD = competitive journal advertising
CPME = competitive physician meeting expenditures CDTCA = competitive direct-to-consumer advertising OPit-1 = lagged own price
DTLit-1= lagged detailing
JADit-1 = lagged journal advertising
PMEit-1 = lagged physician meeting expenditures DTCAit-1= lagged direct-to-consumer advertising GPR = generic price
GDTL =generic detailing
GJAD = generic journal advertising
OPGPR = interaction of own price and generic price DTLGDTL = interaction of own detailing and generic detailing
JADGJAD = interaction of own journal advertising and generic journal advertising
eit = error term
Equation 1. Initial sales equation for the focal brand in the hypertension drug category
3.2 Functional form
3.3 Descriptive statistics
In this section, descriptive statistics from the antidepressant drug category are covered. The statistics can be seen in the graphs below.
Graph 1. Prices of branded drugs in the antidepressant market
From graph 1 can be seen that most brands increased their prices over the years. The price levels of the brands lie fairly close to each other. Two outliers are the prices of Prozac (around $ 115) and of Luvox (around $ 150). Another remarkable point shown by graph 1 is that Lucox and Effexor started highly priced when they entered the market and made a sudden drop in their prices in one of the following periods.
Graph 2. Detailing expenditures of branded drugs per month.
30,000 50,000 70,000 90,000 110,000 130,000 150,000 1 12 23 34 45 56 67 78 89
Pr
ice
in
$
Month
Prices of branded drugs in the
antidepressant market
PAXIL (focal brand) CELEXA EFFEXOR LUVOX PROZAC REMERON SERZONE WELLBUTRIN ZOLOFT 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 1 7 1319253137434955616773798591 Exp e n d itu re (x $ 1000) Month
Detailing expenditures per period
Graph 3. Cumulative detailing expenditures of branded drugs in the antidepressant market.
Graph 2 and 3 show the detailing expenditures form the branded drugs in the antidepressant market. The detailing expenditures of the brands are the largest compared to all their other marketing efforts. Especially Paxil, Prozac and Effexor spend much budget on detailing. Graph 2 shows that on average, brands spend around $ 2 to $ 6 million dollars per month on detailing. During some months, a brand spends even more than $ 10 million dollars. In total, the brands have individually spent more than 500 million dollars on detailing, seen from graph 3. Remarkably, Effexor has spent relatively little budget on detailing, while they are active for a longer period of time than Luvox and Remeron, who entered the market later in time.
,000 100000,000 200000,000 300000,000 400000,000 500000,000 600000,000 1 12 23 34 45 56 67 78 89
De
ti
al
in
g
exp
.e
n
d
itu
re
s
(x10
00
$)
Month
Cumulative detailing expenditures
Graph 4. Cumulative Journal advertising expenditures.
Graph 5. Journal advertising expenditures of brands per month
Graph 4 shows that Zoloft and Effexor spend the most on journal advertising relative to the other brands. Also the high increase from Celexa in their journal advertising is remarkable in graph 4. The journal advertising expenditures from Paxil are also fairly high. Looking at the journal advertising expenditures per period in graph 5, one can see that the expenditures have some peaks every now and then. Especially Celexa has one high peak, around $ 2 million around month 70.
,000 5000,000 10000,000 15000,000 20000,000 25000,000 30000,000 35000,000 40000,000 1 12 23 34 45 56 67 78 89 Exp e n d itu re s (x1000$ )
Month
Cumulative Journal advertising expenditures
PAXIL (focal brand) CELEXA EFFEXOR LUVOX PROZAC REMERON SERZONE WELLBUTRIN ZOLOFT 0 0 500 1000 1500 2000 2500 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 Exp e n d itu re (x $ 1000) Month
Journal advertising expenditures per period
Graph 6. Cumulative physician meeting expenditures from brands in the antidepressant market.
Looking at physician meeting expenditures in graph 6, Zoloft spends the most on physician meetings, followed by Prozac. Paxil has comparable PME spendings with Celexa, Serzone and Effexor. Graph 6 further shows that Luvox and Remeron spend the least on physician meetings.
Graph 7. Cumulative DTCA expenditures from brands in the antidepressant market.
,000 20000,000 40000,000 60000,000 80000,000 100000,000 120000,000 140000,000 1 12 23 34 45 56 67 78 89
Exp
e
n
d
itu
re
s
(x10
00
$)
Month
Cumulative physician meeting
expenditures
PAXIL (focal brand) CELEXA EFFEXOR LUVOX PROZAC REMERON SERZONE WELLBUTRIN ZOLOFT ,000 20000,000 40000,000 60000,000 80000,000 100000,000 120000,000 1 12 23 34 45 56 67 78 89
Exp
e
n
d
itu
re
s
(x10
00
$)
Month
Cumulative DTCA expenditures
Looking at graph 7, no clear pattern can be distinguished between the DTCA expenditures from the brands. The cumulative expenditures are widely separated from each other. Furthermore, the brands do not spend DTCA budget on a regular basis. For example, Effexor and Serzone only spend on direct-to-consumer advertising between periods 34 and 60, and then stopped spending budget on this marketing instrument. Brands like Luvox, Wellbutrin, Zoloft and Remeron do not even spend anything at all on direct to consumer advertising. Paxil, the focal brand, spends most on DTCA, followed by Prozac and Effexor.
Graph 8. Prices of generics in the antidepressant market
The prices of generic drugs are substantially lower than their branded equivalents, as seen from graph 8. Most generics charge around $ 10 till $ 16 for their drugs, except for Nortriptyline. This is a bargain compared to the branded drugs, with prices ranging from $ 70 to $ 150. The most
remarkable point shown by graph 8 is that Nortriptyline started with a high price level, but strongly decreased their price over the years.
Total expenditures (x1000 $) per brand Detailing Journal advertising Physician meetings DTCA Amitriptyline 35 0 0 0 Doxepin 8 0 0 0 Imipramine 101 107 0 0 Nortripyline 0 345 0 0 Trazodone 109 0 0 0
Table 1. Marketing expenditures of generic drugs.
Looking at table 1, it becomes clear that the marketing expenditures of generic drugs are far lower than the marketing expenditures from the branded drugs. Remarkably, no budget from the generic drugs is spent on direct to consumer advertising and physician meetings.
,000 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000 45,000 1 12 23 34 45 56 67 78 89 Pr ic e ( $) Month
3.4 Choice of lag term
Before the model is estimated, the most appropriate lag term is chosen for the model. According to Leeflang et al (1992), lag terms between one and three periods are most suitable for a regression. In order to gain statistical confidence of which lag term is best, six models with different lag terms will be estimated. The results can be seen in table 2 below.
Lag term F-value R-square Log-likelihood CAIC BIC
1 month ,166 ,009 -1152,435 2338,193 2332,193 2 months ,330 ,018 -1138,894 2316,591 2309,591 3 months ,187 ,011 -1126,259 2267,835 2284,245 4 months ,214 ,012 -1113,216 2265,084 2258,084 10 months ,123 ,008 -1036,177 2110,535 2103,535 25 months ,074 ,006 -845,595 1728,028 1721,028
Table 2. Model fit for lag terms
4 Model Specification
4.1 Model comparisonIn order to specify the model with the highest model fit, different models are compared to each other to see which variables have the best explanatory power (Blattberg et al, 2008). Table 4 below shows which variables are included in each model.
Variables included Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 1.Own price 2.Own detailing 3.Own journal advertising 4.Own physician meeting expenditures 5. Own DTCA expenditures √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 6. competitive price 7. competitive detailing 8. competitive journal advertising 9. Competitive physician meeting expenditures 10. competitive DTCA - - - - - √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ - - - - - √ √ √ √ √ 11. Lagged own price
12. Lagged own detailing
13. Lagged own journal advertising
14. Lagged own physician meeting expenditures 15. Lagged own DTCA expenditures - - - - - - - - - - √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ - - - - - 16. Generic price * own
price 17. Generic detailing * own detailing 18. Generic journal advertising * own journal advertising - - - - - - - - - √ √ √ √ √ √ √ √ √
NB: Interaction effects of generic DTCA and generic physician meeting expenditures are excluded because all values from generic brands on these variables are zero.
Model fit statistics 1 2 3 4 5 6 R² ,357 ,737 ,799 ,835 ,675 ,850 Adjusted R² ,321 ,705 ,760 ,795 ,622 ,820 F-value 9,868 23,483 20,643 35,178 12,796 27,690 Log-likelihood -81,717 -39,372 -36,854 -27,505 -59,531 -24,420 AIC 177,435 102,744 107,708 95,011 148,702 93,960 BIC 178,737 133,264 150,944 145,876 186,851 130,619 CAIC 202,238 145,264 165,944 165,876 201,851 148,619
Table 5. Model fit statistics
Model 1 includes only own marketing variables; model 2 includes own marketing variables and competitive marketing variables; model 3 includes own marketing variables, competitive marketing variables and lagged variables and model 4 includes own marketing variables, competitive marketing variables, lagged variables and interaction effects. In order to choose the best model, the model fit statistics of these six different models are compared to each other. A first look is taken at the F-values of the models. The F-value is significant for all models. This means that all models at least contain one predictor variable that is statistically different from zero. On the basis of the adjusted R², model 6 has the highest model fit. This model explains 82,0% of the unexplained variance from the variables.
Second, the AIC, BIC and CAIC are used to assess the model fit. These measures are based on the log-likelihood and show which model gives the best approximation to reality. For models with few variables, the AIC is most often used to analyze model fit. The BIC and CAIC are however more preferred for models with many predictor variables, because these statistics give greater penalty to adding variables with poor predictive power. According to the AIC, model 6 has the highest model fit of all models, in line with the adjusted R². When there is looked at the BIC and the CAIC however, there are some conflicting results. The BIC statistic says that model 6 has the highest model fit, since model 6 has the lowest BIC value. The CAIC however says that model 2 has the best model fit.
Therefore, a decision has to be made for choosing model 2 or model 6. This research chooses to work with model 6, because besides the BIC, the adjusted R², AIC and F-value indicate that this is the best model.
The choice of model six however means that the hypotheses about the influence of lagged variables cannot be tested. Since we are still interested in possible lagged effects of the own marketing variables, the initial model is estimated to double check if the lagged variables indeed have poor explanatory power. The results can be seen in Appendix A on page 39. The estimation results of the initial model in Appendix A shows that the lagged variables indeed all have an insignificant influence on sales. Hence, exclusion of the lagged variables in the final regression model will not cause biased parameter estimates.
4.2 Data preparation
between variables is likely to cause multicollinearity (Malhotra, 2010). Also, since some variables seem to show a trend over time, all variables are tested for having a unit root. In the final part of this section, the regression model is checked for issues with autocorrelation, heteroscedasticity and nonnormality.
4.2.1 Correlation between explanatory variables
When running a correlation matrix between all variables, it becomes clear that variable of generic price and the interaction variable of own price and generic price have an extremely high and significant correlation coefficient (at the level of p<0,01) of -0,904. When running the regression of model 5, these variables turn out to be the only variables with severe multicollinearity, with VIF values that are even above 1000 (for details, see appendix A). This indicates extremely high
multicollinearity between these two variables. Therefore, one of these variables has to be excluded from the estimation. Prior research has shown that omitting price variables (e.g own price variable or competitive price variables) can lead to omitted variable bias (e.g. Kremer et al, 2008; Sethuraman et al, 2011). Therefore, the interaction effect between own price and generic price will be excluded from estimation.
4.2.2 Unit root test
To check if the variables are stationary, the test of Phillips and Perron (1988) will be performed. The results of the Phillips-Perron test are displayed in the table below.
Variable Phillips-Perron statistic significance
LnRev -17,501 0,000 Lnprice -2,837 0,1879 lnDTL -11,218 0,000 lnJAD -9,744 0,000 lnMTG -5,802 0,000 lnDTC -2,280 0,438 Lncompprice -2,690 0,243 Lncompdtl -5,588 0,000 Lncompjad -4,060 0,010 Lncompmtg -5,283 0,000 Lncompdtc -3,918 0,015 Lngenprice -0,7349 0,967 LngenDTL -11,265 0,000 lngenJAD -4,132 0,008
InteractionDTL* N/A N/A
InteractionJAD* N/A N/A
*these variables are created after unit-root correction
Table 6. Phillips-Perron test results for all variables
4.2.3 Autocorrelation
Graph 9. Display of residual values per period (t)
Graph 9 represents the residual values over time. The pattern of the residuals seem to imply that there is some positive autocorrelation within the residuals, which means that when a residual at time t is positive or negative, residual t+1 also tends to be respectively positive or negative. The
autocorrelation between the variables is statistically tested with the Durbin-Watson statistic. This statistic has a lower limit and a upper limit, where values below the lower limit give evidence for positive autocorrelation and values above the upper limit give evidence for negative autocorrelation. For a regression model with 15 predictor variables and 95 observations, the lower limit is 1,312 and the upper limit is 2,040. The Durbin-Watson statistic for the residuals of this model is 1,279. Hence, there is statistical evidence that this data faces positive correlation. When a correlation is performed between the residuals and the lagged residuals, a positive correlation coefficient of 0,343 is found. To correct the variables for autocorrelation, they are all corrected with the correlation coefficient between the residuals. In the new situation, the Durbin Watson statistic increased to 1,536. This value is closer to the ideal Durbin Watson value of 2 when there is no autocorrelation. Furthermore, the new value is higher than the lower limit of 1,312, which means the data does not have significant autocorrelation anymore.
4.2.4 Heteroscedasticity
Third, the data is checked for satisfying the homoscedasticity assumption. If the homoscedasticity assumption is violated, the variances of the betas are wrongly estimated. In order to detect heteroscedasticity, the residuals of large and small values of the journal advertising variable are compared to each other with Levene’s test for homogeneity of variances. The choice for the journal advertising variable is made because this variable has a relatively high number of outliers compared to the other explanatory variables. Looking at graph 10, one can see that the residuals are clustered fairly close to each other. In order to test equal variances in the data, the residual values of the two groups are compared to each other. The first group consists of 26 cases with low values of journal advertising (lower than 4) and the second group consists of 70 cases with high values of journal
-1 -0,8 -0,6 -0,4 -0,2 0 0,2 0,4 0,6 0,8 1 1 13 25 37 49 61 73 85 Va lu e o f re sid u al Period (t)
Residual values over time
advertising (equal to and higher than 4). The Levene’s statistic shows a p-value of 0,910, which means that the variances are equal between the two groups and that there are no problems with heteroscedasticity in the data.
Graph 10. Plot of Residuals against journal advertising expenditures 4.2.5 Nonnormality
The last element that has to be satisfied is that the disturbances of the data are normally distributed. These disturbances need to be normally distributed for the standard test statistics and confidence intervals to be reliable. The normality assumption is assessed by generating a Q-Q plot of the residuals and with the Kolmogorov-Smirnov test.
Looking at the Q-Q plot, the residuals look fairly normally distributed, since the plot does not show many outliers in the residual values. The
Kolmogorov-Smirnov test statistic shows an insignificant p-value, which means that there also is statistical evidence that the residuals are normally distributed. In conclusion, the normality assumption is satisfied, which means that estimating a regression model for this type of data is appropriate.
Graph 11. Q-Q plot of residuals Table 7. Kolmogorov-Smirnov test statistic
-1 -0,8 -0,6 -0,4 -0,2 0 0,2 0,4 0,6 0 2 4 6 Va lue o f re sidu al
Journal advertising expenditures
5 Estimation of the final model
Now that the data is tested and corrected for unit roots, heteroscedasticity, autocorrelation and nonnormality, the regression model can be estimated. The regression results are displayed in table 8 below.
Variable Interpretation Hypothesized effect
Parameter estimate
Std. Error p-value
ΔlnPrice Own price - -6,062 1,969 ,003***
lnDTL Own detailing + ,250 ,122 ,049*
lnJAD Own Journal advertising + -,167 ,050 ,001*** lnMTG Own physician meeting expenditures + ,097 ,041 ,020** ΔlnDTC Own DTCA expenditures + n.s. ,017 ,908
ΔlnCompPrice Competitive price - n.s. ,298 ,515
lnCompDTL Competitive detailing - n.s ,167 ,328 lnCompJAD Competitive Journal advertising + n.s ,044 ,394 lnCompMTG Competitive physician meeting expenditures + ,235 ,056 ,000*** lnCompDTC Competitive DTCA expenditures + ,033 ,008 ,000***
ΔlnGenPrice Generic price 4,658 1,695 ,007***
lnGenDTL Generic detailing -,079 ,007 ,001***
lnGenJAD Generic journal advertising n.s. ,022 ,524 InteractionDT L Interaction effect of own detailing and generic detailing - n.s. ,128 ,778 InteractionJA D Interaction effect of own journal advertising and generic journal advertising - -,136 ,058 ,000***
Table 8. Estimation results of final model
Since some variables are the first difference operator of the original variables, one should be
cautious about the interpretation of these variables. Pauwels et al (2007) explain in their article that the first difference operator of a logged variable can be interpreted as the growth rate of the
so that the variable has the same value relative to the previous period (Pauwels et al, 2007). Further, the regular ln-transformed variables can be interpreted as elasticities (Pauwels et al, 2007).
The estimation results show that price has a significant and strong negative effect of -6,062 on brand sales, as expected. The estimation shows that when the growth rate of price increases with 1%, revenues will drop with 6,062%. Also as expected, detailing has a positive influence on brand sales. When detailing expenditures increase with 1%, own brand sales will increase with 0,250%.
Remarkably, journal advertising has a significant negative parameter of -0,167. This means that when journal advertising expenditures increase with 1%, the brand sales will decline with 0,167%. Physician meeting expenditures have a small positive impact on sales of 0,097, as was hypothesized. So, when physician meeting expenditures increase with 1%, it will positively affect sales with a 0,097%
increase, given that all other variables are held constant. Lastly, direct-to-consumer advertising does not have a significant effect on brand sales, despite the fact that this has become a popular
marketing instrument.
Looking at competitive effects, the estimations results show that competitive physician meeting expenditures and competitive DTCA spendings have a significant positive effect on brand sales. Respectively, a 1% increase in competitive PME and competitive DTCA spendings cause a 0,235% and 0,033% increase in own brand sales. This proves that these marketing instruments indeed mainly have a category expansion effect. Remarkably, competitive price, competitive detailing and competitive journal advertising do not have a significant impact on own brand sales. This indicates that there are no serious competitive threats for brands in this category of the pharmaceutical industry. Regarding competitive effects of generics, the direct effect of generic price on brand sales is 4,658. This implies that buyers of generic drugs are highly price-sensitive, which makes sense
considering the low prices of generics. The estimation shows that when increases in generic prices change with 1%, brand sales will go up with 4,658%. Regarding generic detailing, the results show a negative effect of -0,079 on brand sales. This means that increases in detailing from generics can be detrimental for own brand sales, but the negative effect is not very large.
Lastly, table 8 shows that the interaction effect between own detailing and generic detailing is not significant. In contrast, the interaction effect of own journal advertising and generic journal
Graph 12. Interaction effect between own journal advertising and generic journal advertising
6 Validation
6.1 Face validity
The face validity of the model is good. Many of the hypothesized effects were confirmed by the estimation. For example, the effects of price, detailing, physician meeting expenditures, competitive physician meeting expenditures and competitive DTCA show results in line with their hypothesized effects. However, own journal advertising shows a negative effect while a positive effect was expected. Furthermore, three of the five competitive variables do not seem to have a significant influence on brand sales. Nevertheless, the model looks applicable to a real life situation, since it includes significant marketing dynamics as competitive effects and an interaction effect besides the direct effects of the own marketing instruments.
6.2 Predictive validity
In order to test the predictive validity of the model, the data is split up in a estimation sample and validation sample. The estimation sample consists of the first 84 periods (months) of observations. With the estimation sample, predicted values are computed for period 85 until 96. The predicted values are then compared to the actual values of period 85 until 96. With these values, the RASPE is computed to judge the predictive validity of the model.
√∑ ̂
The value of the RASPE statistic should be larger than the standard deviation of the residuals. Since the standard deviation of the residuals is 0,226, the model has sufficient predictive validity. To test the errors in the forecasting ability of the model in absolute terms, the MAPE statistic is calculated. The outcome can be seen below.
Benchmark models with good predictive validity mostly have a MAPE around 10% (Gijsenberg, 2014). The MAPE value for this model is substantially higher than the benchmark models, but an error percentage around 35% is still reasonable according to Gijsenberg (2014). In conclusion, the RASPE and MAPE show that that the predictive validity may not be excellent, but they give evidence that the forecasting ability of the model is sufficiently valid.
6.3 Multicollinearity
Table 9 below displays the VIF values for all the variables in the model. The table shows that none of the variables show moderate (VIF > 4) or severe multicollinearity (VIF > 10). All variables have VIF values that are well below these values. The variable with the highest VIF score is the competitive journal advertising variable with a value of 3,144. In conclusion, the results shown in table 9 clarify that the parameter estimates are not biased by multicollinearity problems (Malhotra, 2010).
Variable VIF values ΔlnPrice 1,563 ΔlnDTL 1,720 lnJAD 1,595 lnMTG 1,488 ΔlnDTC 1,039 ΔlnCompPrice 1,136 ΔlnCompDTL 2,623 ΔlnCompJAD 3,144 ΔlnCompMTG 1,811 ΔlnCompDTC 1,414 ΔlnGenPrice 1,355 ΔlnGenDTL 1,444 lnGenJAD 2,859 InteractionDTL 2,545 InteractionJAD 1,734
