Developing Countries and Scientific Collaboration in Pharmaceuticals
Richa Srivastava[1], Department of Humanities and Social Sciences, Indian Institute of Technology, Kanpur
Abstract
This paper analyses the nature and causes of international scientific collaboration between developed and developing countries in pharmaceuticals. The effect of proximity factors such as contiguity, common language, colonial links and distance, and of institutional factors such as ease of clinical research and the presence of private organisations on collaboration is estimated. Poisson count data model is used for the analysis, in which evidence indicating the impact of the presence of private institution, common language and colonial links on collaborations is found. However, the effect of geographical distance and ease of clinical research on pharmaceutical collaborations is not very clear.
Keywords: Scientific collaboration, pharmaceuticals, developing countries, clinical research, private firms, Poisson model
Introduction
Scientific collaboration is frequently used to solve complex scientific problems and promote various political, economic and social agendas. It involves the sharing of research data, equipment sharing, joint experimentation, building of databases and conferences. With scientists from different countries working towards common goals, international scientific collaboration has grown in recent years: research teams are not only bigger in size but are also more diversified in terms of nationalities. The increasing trend towards team formation has been attributed to increasing scale, complexity, costs of big scientific projects and the fact that teams produce more influential (highly cited) work than individual authors (Wutchy et al., 2007 :1036-37).
International collaboration has also emerged as a preferred method for building scientific capacity in developing countries (Wagner et al., 2001: 9). Innovation has always been an important source of economic growth and increased welfare. Advanced countries spend huge amounts to fund merit-based collaborations in order to exploit the strong link between knowledge generation, productivity enhancement and economic growth. Developing countries, on the other hand, strive to build up their scientific capacities and employ the knowledge generated abroad to move up the development ladder. However, with the emergence of new international trade, investment and intellectual property rules, developing countries have faced difficulties in using the technology developed abroad. This has served as a trigger for innovation (knowledge generation) and international collaboration (knowledge flow) (Mytelka, 2006: 415-16).
From the findings of literature on scientific collaboration, it can be observed that collaboration mainly takes place between countries which share similar research profiles. Two developed countries can, therefore, collaborate in all major scientific fields. Developing countries, however, tend to focus on specialising in a few specific areas of science which are related to their national need - disease control, for example. North-south collaboration in pharmaceuticals is, therefore, of interest.
Developing countries are not only the largest producers of generic drugs but also hold large markets. Clinical trials of pharmaceutical and device companies are increasingly being conducted in developing countries. An important factor affecting this trend is the increasing bureaucratic and expensive regulatory environment in advanced countries. The large number of potential participants and lower cost of research have attracted the pharmaceutical companies towards developing countries such as India and China, where the population size alone offers the promise of an expanding market (Glickman et al., 2009: 818).
Moreover, in the pharmaceutical industry, the knowledge produced by the public sector does not spill over free of cost to the downstream researcher. Thus, the private sector not only needs to invest in basic research but also to collaborate with the public sector to gain commercial advantage. The pharmaceutical companies produce a large number of publications which form a rich subset of collaboration data (Cockburn and Henderson, 1998:158). However, the effect of private sector research and clinical research on international collaboration remains largely unknown. The paper aims to address these issues.
In this paper, the publication data from Web of Science Databases (WoS) for the period 1974-2008 is used to estimate the effect of various institutional and proximity factors on collaboration between developed and developing countries. The analysis focuses on observing country-specific factors like institutional interaction, nature of scientific research and therapeutic areas of interest, and on policies which play an important role in shaping the way these factors are perceived.
The rest of this paper consists of 4 sections. Section 2 reviews literature and builds conceptual hypotheses for the analysis. In section 3, the descriptive and comparative analysis to test the proposed hypotheses is performed. Section 4 lists and explains the main findings. Section 5 contains some concluding remarks.
Literature Review
Collaboration for building capacity in developing countries
Wagner et al. (2001:1-9) examined the role of merit-based collaboration between developed and developing countries in building scientific capacity in the latter. They identified the need for expertise and presence of particular research equipment, databases and laboratories as factors influencing collaboration.
They recognised the role of information and communication technology in boosting international collaboration but mentioned that it cannot alone motivate or enable collaboration. They suggested that presence of a baseline level of scientific capacity (which is different for different fields) is necessary for any sort of collaboration. Agrawal and Goldfarb (2005:1-3), on the other hand, found an 85% increase in the likelihood of collaboration after the adoption of Bitnet (an earlier form of the internet) by universities. Similar evidence was found by Adam et al. (2005: 275-77) who reported rapid growth in university-level and university-firm collaboration. They attributed this growth to the decline in collaboration costs due to the deployment of the National Science Foundation's program to promote advanced networking in the US (NFSNET) and its connection to networks in Europe and Japan. This evidence lowers the significance of geographical distance as a determinant of international collaboration.
Wagner and Leydesdorff (2005: 186-88) mapped the networks created by international co-authorships for the years 1990 and 2000. They analysed the observed linkages at the global and regional level and witnessed a pronounced expansion of the global network along with emergence of regional hubs. Using factor analysis, they found that large countries compete with each other for developing partners in the global network.
Maina-Ahlberg et al. (1997: 1229-30) found that most of the collaborative projects were initiated from the North and that some disagreements concerning remuneration/compensation existed, due to different policies and remuneration rates set by institutions. In addition, loopholes in financial management, legal and regulatory obstacles, the absence of a common language and spillover effects of international diplomacy were the main factors which hindered international cooperation.
Policies for International Collaboration
Mytelka (2006: 420-22) examined the experiences of India, Cuba, Iran, Taiwan, Egypt and Nigeria to study the country-specific drivers and triggers of pharmaceutical innovation processes. He suggested that policies play an important role in shaping the way these triggers are perceived and how they drive the innovation process, arguing that simply increasing the 'supply of researchers' does not ensure a process of innovation. Rather, complementary policies which provide incentives to these researchers to acquire knowledge and focus on domestic problems are needed for the development of an innovation dynamic.
Cockburn and Henderson (1996: 159-63) found evidence of excessive co-authoring between pharmaceutical researchers belonging to the public and private sectors. They suggested that collaboration between public and private sectors is an essential determinant of the productivity of the latter.
Based on the literature review the following four hypotheses are proposed:
- Hypothesis 1: Developing countries collaborate more with developed countries than with other developing countries.
- Hypothesis 2
- A: Common Language increases collaboration between countries.
- B: Contiguous countries collaborate more.
- C: Colonial links between countries increase collaboration.
- D: The greater the distance between two countries, the less they collaborate.
- Hypothesis 3: Developed countries with a larger share of clinical research publications prefer to collaborate with developing countries.
- Hypothesis 4: Developed countries with more publications under private research organisations collaborate with developing countries.
Data
The data for this study has been taken from various sources. BioPharmInsight was used to draw up a list of 215 medical indications which were then assigned to one of the 12 therapeutic areas. These therapeutic areas were defined according to a system of an organism or a general disease group. A list of the 12 therapeutic areas can be found in the appendix (A.1). The list of medical indications was used to search for corresponding scientific pharmaceutical publications in the Web of Science databases (WoS). It consists of 7 databases, the most important one being the Science Citation Index Expanded. It covers the scientific fields of biochemistry, medicine and pharmacology and indexes more than 6500 scientific journals. Information concerning the scientific publications themselves, like the title, the year of publication, the journal, author's affiliation (including the country of respective organisation), cited references, categorisation of research fields that a publication can be assigned to, and further bibliographic information can be obtained using the WoS database. Information concerning the authors' affiliations is matched with WHO Regions and World Bank income groups in order to include the geographical region a country is located in and the wealth level of the countries in our sample.
In this study, all publications included in categories related to pharmaceutical research are considered. Articles from the subcategories "Biochemistry & Molecular Biology", "Biotechnology and Applied Microbiology", "Chemistry, Applied", "Chemistry, Medicinal", "Medicine, Research and Experimental", "Pharmacology and Pharmacy" and "Toxicology" are included in our dataset. The sample is restricted to journal articles and excludes publications that are labelled as meeting abstracts, editorials or reviews as well as other non-journal publications and conference proceedings.
In order to determine whether researchers affiliated to private companies publish, authors' affiliations were searched for the occurrence of companies' legal forms and the names of big pharmaceutical companies. Also, articles originating in universities and public research institutions were classified on the authors' affiliations.
The CHI classification (Hamilton, 2003) was used to distinguish between "clinical observation", "clinical mix", "clinical investigation", and "basic biomedical research" journals. CEPII (Centre d'Etudes Prospectives et d'Information Internationales) was used to obtain data on the distance measures and proximity measures, like presence of a common language, colonial link and common border etc. (Mayer and Zignago, 2006).
The dataset was restricted to journal publications in which authors from at least one country assigned to the low-income or the lower-middle income group (according to the World Bank classification) were involved. For the period from 1974 to 2008, 13,126 publications were obtained. Each publication was assigned to the respective countries mentioned in the authors' affiliations. Since co-publications represent undirected links, each pair of countries was included only once in the analysis.
Data description and summary statistics
The patterns of collaboration between developed and developing countries can be significantly different from those between two developed countries. The difference may arise depending on the type of research (clinical/ basic) and the institutions involved (firm/universities). Our main focus here is to observe such differences and report them.
Dummy variable 'developing_2' is used to differentiate the developed-developing country pair from a developing-developing country pair. Another source of difference is the therapeutic area. Table 1 presents the main variables used in the descriptive and regression analysis.
| Variables | Description and Source |
|---|---|
| Collaborations | Total number of collaborations between countries in a pair in a particular therapeutic area. |
| distance | Distance between two countries based on the largest (population wise) cities of those countries |
| shareclinPub_developing / shareclinPub_developed | Share of total number of publications of the respective country that are published in journals assigned to CHI category "clinical research", "clinical mix" or "clinical observations". |
| sharefirmPub_developing / sharefirmPub_developed | Share of publications of the respective country in the respective therapeutic area that can be assigned to firms |
| Developing _2 | A dummy variable which takes value 1 if both the collaborating countries are developing |
| common_lang_official | A dummy variable which indicates whether two countries share a common official language |
| colony | A dummy variable which indicates whether two countries ever had a colonial link |
| contiguity | A dummy which indicates whether two countries are neighbours |
| Period_dummy | A dummy which indicates whether the collaboration takes place in period 1(1998-2002) or period 2(2002-2008) |
Table 1: Variable Description and Source
The publication data is divided into two periods (Period 1: 1998-2002 and Period 2: 2002-2008). Table 2 presents the pooled summary statistics for the variables.
| Variables | Mean | S.D. | Max | Min | Skewness | N |
|---|---|---|---|---|---|---|
| collaborations | 1.309 | 30.167 | 3021 | 0 | 71.447 | 17362 |
| Distance | 7173.47 | 4200.062 | 19772.34 | 56.99 | 0.621 | 17362 |
| sharefirmPub_developed | .0199 | .0686 | 1 | 0 | 7.952 | 17362 |
| sharefirmPub_developing | .0111 | .0866 | 1 | 0 | 10.680 | 17362 |
| shareclinPub_developed | .4898 | .3059 | 1 | 0 | -.2685 | 17362 |
| shareclinPub_developing | .6085 | .3307 | 1 | 0 | -.4749 | 17362 |
Table 2: Summary Statistics
The dataset comprises 121 countries forming 6486 pairs over the 2 periods. Each country pair in the dataset necessarily includes at least one developing country. This characteristic of the data makes it highly skewed. The collaboration variable, for example, is highly skewed towards zero and even more so when both the collaborating countries are developing.
The collaboration data contains a large number of zeros which leaves the classical tests and regression techniques inappropriate. Taking a closer look at the data (with collaborations > 50 for developed-developing pairs), the major collaborators among the developing countries are found to be China, India, Kenya, Nigeria and Thailand (see Figure 1).
Figure 1 also shows that "infectious disease" is one of the most researched therapeutic areas amongst the developing countries; however, the maximum numbers of collaborations take place in "cancer". Cancer research is one of the major concerns of the developed world and such observations in data may point towards the growing potential of China as a cancer research centre.
Figure 1: Plot showing the major developing collaborators and the concerned Therapeutic Areas
With such a skewed distribution, the collaborations data fails to follow the normality assumption for the t-test (mean comparison test). Therefore for the comparative statistical analysis here, a two-sample median test (Wilcoxon-Mann-Whitney test) is used.
Two sample Wilcoxon rank-sum Test
The Wilcoxon-Mann-Whitney (or rank sum) test is the non-parametric alternative of a t-test. It does not assume normality. In this approach, the two samples to be tested are combined and ranked. The sum of ranks from each sample then acts as the test statistic, thereby avoiding assumption of a specific distribution for the original data. In this study, this test is used to observe the differences in collaborations occurring due to the difference in the income group of partnering countries, their official language, common coloniser and contiguity.
Comparison based on Income group
| Developing_2 | Obs. | Rank sum | Expected |
|---|---|---|---|
| 0 | 14606 | 1.263e+08 | 1.268e+08 |
| 1 | 2756 | 24401762 | 23926214 |
| Combined | 17362 | 1.507e+08 | 1.507e+08 |
Table 3a: Rank-sum test for developed and developing country collaboration
Ho: collaborations (Developing_2==0) = collaborations (Developing_2==1)
z = -4.004
Prob > |z| = 0.0001
P{collaborations(Developing_2==0) > collaborations(Developing_2==1)} = 0.488
The results suggest that there is a statistically significant difference between the underlying distribution of collaborations between two developing countries and those between a developed and developing country (p= 0.0001). The probability-order suggests the chances of collaborations between a developed-developing country pair as being 48.8 % more as compared to a developing-developing country pair.
This can be explained by the need for the developing countries to collaborate with the developed ones in order to build capacity (Wagner et al., 2001). Factors such as cost sharing, presence of better research equipment, databases and laboratories as well as a need for expertise has led to such international collaborations.
Also, many developed countries which previously donated resources to 'research-for-aid' are not willing to continue without reciprocity or some clear benefit. This has led to a collaboration system which is attractive to both the partnering countries.
Comparison based on Official language
| Common_lang_official | Obs. | Rank sum | Expected |
|---|---|---|---|
| 0 | 15269 | 1.313e+08 | 1.326e+08 |
| 1 | 2093 | 19434058 | 18170380 |
| Combined | 17362 | 1.507e+08 | 1.507e+08 |
Table 3b: Rank-sum test for Common Official Language
Ho: collaborations (common_lang_official==0) = collaborations (common_lang_official ==1)
z = -11.942
Prob > |z| = 0.0000
P {collaborations (common_lang_official==0) = collaborations (common_lang_official ==1)} = 0.460
There is a statistically significant difference between the underlying distribution of collaborations between countries which share a common official language and those which do not. The probability indicates that the chances of collaborations between countries not sharing an official language are 46% more than between countries which do have a common official language.
As seen from the previous result, the collaborations between a developed-developing country pair are more as compared to a developing-developing country pair. Secondly, the probability of a developing country like India or one of the African nations, not sharing an official language with their developed partner like U.S. is quite high. Hence, it can be said that international collaborations take place irrespective of a common official language.
Since our sample consists of only those country pairs which contain at least one developing country, the above results seem to hold.
Comparison based on contiguity
| Contiguity | Obs. | Rank sum | Expected |
|---|---|---|---|
| 0 | 16986 | 1.471e+08 | 1.475e+08 |
| 1 | 376 | 19434058 | 3264244 |
| Combined | 17362 | 1.507e+08 | 1.507e+08 |
Table 3c: Rank-sum test for contiguity
Ho: collaborations (contiguity==0) = collaborations (contiguity ==1)
z = -6.985
Prob > |z| = 0.0000
P {collaborations (contiguity==0) = collaborations (contiguity ==1)} = 0.448
The probability here indicates that the chances of collaborations between countries sharing a common border are 44.8% less than between countries which are not contiguous.
This result can again be tracked back to the higher number of collaborations between developed-developing countries pairs. Since it is unlikely that a developed and developing country share a common border, it can be said that factors influencing collaborations between two countries are: research facilities and expertise, rather than contiguity.
The Wilcoxon rank-sum test thus helps to gain an intuition about the data and the difference in patterns of collaboration between various country pairs.
Methodology
Poisson Regression for Count Data Model
The Poisson regression is used to model count data where the error structure does not follow the normal distribution. It assumes that the response variable Y follows the Poisson distributions, i.e.:
for µ>0. The mean and variance of this distribution can be shown to be
Since the mean is equal to variance, the usual assumptions of homoscedasticity do not hold for the Poisson data.
In this study, there are four types of Poisson regression models. Model 1 gives the result of regression of collaborations on shareclinPub_developed, shareclinPub_developing and distance while controlling for the proximity dummies and the therapeutic dummies. Model 2 gives results of regression of collaborations on shareclinPub_actor, shareclinPub_partner for two developing countries while controlling for other variables. Model 3 regresses collaborations on sharefirmPub_developed, sharefirmPub_developing while Model 4 does the same for two developing countries while regressing on shareclinPub_actor, shareclinPub_partner. Regression results of Model 2 and 4 can be found in the appendix.
Test for Over-dispersion
There is a test of the null hypothesis of equal-dispersion, , against the alternative of over-dispersion which can be given by equation
the variance function being the one of a negative binomial model.
Hence, the hypothesis H0: α = 0 against H1: α > 0 is tested. The results (Appendix A.4) indicate the presence of significant over-dispersion in Model (1) and (3). To model this feature of the data, robust estimate of VCE (Cameron and Trivedi 2009) is used.
Regression Results
The results are obtained by applying the Poisson model with robust VCE estimates (application of OLS yields similar results). Table 4 gives the marginal effects after Poisson regression of collaborations on the share of publications attributed to firms of the partnering developed and developing countries (Model 3). A developed country with higher firm-level collaboration has 93% more chance of collaborating with a developing country. Similarly, a developing country with low firm-level publications has 30.6% higher chance of collaborating with its developed counterpart. These positive and negative coefficients of sharefirmPub_developed and sharefirmPub_developing respectively imply that the pharmaceutical firms in developed countries prefer to collaborate with their counterparts who are less developed in pharmaceutical research. A general upward trend has been observed in firm-level international collaborations, however these results point towards the usage of economic and bureaucratic disparities by the developed countries. While the firms from developing countries collaborate to absorb capacity, their developed counterparts may actually gain by exploiting the wage and regulation disparities between the two nations. Local people who consider themselves to be poorly paid see these externally funded research projects as an added benefit (Maina-Ahlberg et al.,1998).
| Independent Variables (Dependent Variable : Collaborations) |
Model (3) dy/dx (S.E.) |
|---|---|
| sharefirmPub_developed | 0.93*** (0.17 ) |
| sharefirmPub_developing | -0.306** (0 .15 ) |
| distance | -1.40e-06 (0.00002) |
| Common_lang_official | 0.89** (0 .35) |
| Colony | 0.54* (0 .29) |
| contiguity | 0.07 (0.19) |
| Period control | Yes |
| Therapeutic Area control | Yes |
| N | 14606 |
| Log pseudo-likelihood | -39367.039 |
Table 4: Marginal effects after Poisson Regression on Model 3
(*p<0.05, **p<0.01, ***p<0.001)
In Table 4, distance represents the geographic distance between two countries. An insignificant coefficient for distance only supports the fact that developed and developing countries are seldom closely situated. In the presence of new communication technologies, geographical distance does not matter much for collaboration and is found to be insignificant in individual periods.
It is found that a common official language facilities the collaboration between two countries belonging to different income groups. The number of collaborated research articles between partners sharing a common official language were 0.89 times more than those with different common languages.
The variable colony shows a positive association with the number of collaborated research publications while the effect of contiguity is ambiguous.
Table 5 gives the marginal effects after Poisson regression of collaborations on the share of publications assigned to the CHI category of "clinical research", "clinical observation" or "clinical mix", of the developed and developing countries.
The coefficients of shareclinPub_developing and shareclinPub_developed are negative and significant. This can be interpreted as the mismatch of human needs and priorities of different countries. The developing countries with high share of clinical research publication may not prefer to collaborate with their developing counterparts as they easily obtain the research participants and lower cost of research in their home country.
| Independent Variables (Dependent Variable : Collaborations) |
Model 1 dy/dx (S.E.) |
|---|---|
| shareclinPub_developed | -0.240** (0.06) |
| shareclinPub_developing | -0.391*** (0.13) |
| distance | 5.81e-06 (0.00002) |
| Common_lang_official | 0.955** (0.348) |
| Colony | 0.533* (0.29) |
| contiguity | 0.113 (0.20) |
| Period control | Yes |
| Therapeutic Area Control | Yes |
| N | 14606 |
| Log pseudo-likelihood | -39203.142 |
Table 5: Marginal effects after Poisson Regression on Model 1
(*p<0.05, **p<0.01, ***p<0.001)
Developed countries, on the other hand, may not find suitable partners who wish to collaborate in clinical research of conditions such as allergic rhinitis or over-reactive bladder, which are of their interest. Also considering the fact that the genetic make-up of the population differs widely across nations, the safety and effectiveness of drugs may not be the same in the trial sample and in the user sample. This can defer collaboration.
The variables distance and contiguity are again insignificant, while common_lang_official and colony show a positive association with the number of collaborated research articles.
Differences in pattern over the 12 Therapeutic areas
Figure 2: Data distribution over the Therapeutic Areas
Figure 2 illustrates the data distribution over the 12 therapeutic areas. As is evident, the share of infectious diseases surpasses all other shares and is more than one-third of the whole sample. This is justifiable because infectious diseases are a global problem and they ought to drive collaboration. However, the more basic therapeutic areas which are of research interest to the developing countries like HIV infection or gastro-intestinal or cardiovascular have a small share in the sample. This may be a reason behind the counter-intuitive results which say that collaborations decrease with higher level of clinical research publications of both the developed and developing countries.
Table 6 gives the regression results for different therapeutic areas. The sign of shareclinPub_developing is positive for the therapeutic areas cardio-vascular and gastrointestinal while it is negative for the areas like cancer and infectious diseases. However, the sign of shareclinPub_developed remains negative whenever it is significant.
| Independent Variables (Dependent Variable : Collaborations) |
TA1 Coef. (S.E.) |
TA2 Coef. (S.E) |
TA7 Coef. (S.E) |
TA13 Coef. (S.E) |
TA17 Coef. (S.E) |
|---|---|---|---|---|---|
| shareclinPub_developed | -2.35* (1.21) |
-0.619 (0.58) |
0.302 (0.574) |
-0.78*** (0.226) |
0.43 (0.29) |
| shareclinPub_developing | -1.13 *** (0.24) |
0.904*** (0.13) |
1.19** (0.361) |
-0.55** (0.20) |
0.34 (0.25) |
| distance | .00004 (0.0001) |
0.00002 (0.00006) |
-0.0001* (0.00007) |
6.55e-06 (0.00003) |
-0.00005 (0.00006) |
| Common_lang_official | 1.53* (0.848) |
1.0* (0.57) |
2.18*** (0.44) |
0.88** (0.29) |
1.32* (0.57) |
| Colony | -1.02 (1.04) |
0.314 (0.54) |
0.234 (0.69) |
1.69*** (0.27) |
0.25 (0.76) |
| contiguity | 0.468 (1.03) |
0.39 (0.76) |
-0.99 (0.74) |
-0.34 (0.45) |
-0.04 (0.53) |
| Period control | Yes | Yes | No | Yes | No |
| N | 1528 | 1689 | 1515 | 5081 | 1107 |
| Log pseudo-likelihood | -12876.627 | -3631.77 | -1434.13 | -13343.9 | -1210.51 |
Table 6: Poisson regression for 5 therapeutic areas
(*p<0.05, **p<0.01, ***p<0.001)
Conclusions
In this study, descriptive statistics and Poisson regression model were used to study the patterns of collaboration between developing countries and their developed counterparts, while focusing on the firm-level publications and clinical-research publications.
The preliminary results suggest that developing countries prefer to collaborate with developed countries and even more so with countries sharing a common official language and colonial links. The Poisson regression results for the firm-level collaborations support the hypothesis that developed countries with higher number of firm-level publications collaborate more. The results for clinical research, however, do not support the hypothesis and are ambiguous (different for the various therapeutic areas). A reason for such results can be the absence of data for collaborations specific to clinical research. Using the Poisson model, an attempt has been made to explain the total collaborations between two countries by the share each has in clinical research. However, the model fails to control for size, as would be the case of a gravity regression model.
Scope of future work lies in improving on the methodology (using gravity regression model) while focusing on a particular therapeutic area. Such a study would provide a better picture of 'why' does a developing nation collaborate with a developed one and 'what' role does it play. The dataset, however, will then have to be more refined, i.e. at the 'Therapeutic area' level.
Acknowledgements
The DAAD WISE program and GSBC - EIC have jointly funded this project. I thank Professor Uwe Cantner, Chair of Microeconomics at the University of Jena for giving me the opportunity to work on this topic. I particularly thank Bastian Rake for sharing the data with me and for helpful comments, expressed interest and concerns. I also sincerely thank Dr Ljubica Nedelkoska for her very helpful comments and suggestions.
List of Figures
Figure 1: Plot showing the major developing collaborators and the concerned Therapeutic Areas
Figure 2: Data distribution over the Therapeutic Areas
List of Tables
Table 1: Variable Description and Source
Table 2: Summary Statistics
Table 3a: Rank-sum test for developed and developing country collaboration
Table 3b: Rank-sum test for Common Official Language
Table 3c: Rank-sum test for contiguity
Table 4: Marginal effects after Poisson Regression on Model 3
Table 5: Marginal effects after Poisson Regression on Model 1
Table 6: Poisson regression for 5 therapeutic areas
Table 7: List of the 12 Therapeutic Areas
Table 8: Cross-correlation between variables
Table 9: Poisson Regression results for Model 2
Table 10: Poisson regression results for Model 4
Table 11: Test for over-dispersion
Appendix
A.1 List of Therapeutic Areas and their IDs
| Therapeutic Area | ID |
|---|---|
| Cancer | 1 |
| Cardiovascular | 2 |
| Central Nervous System | 3 |
| Eye and Ear | 6 |
| Gastrointestinal | 7 |
| Genitourinary | 8 |
| Haematological | 9 |
| HIV Infections | 10 |
| Immune System | 12 |
| Infectious Diseases | 13 |
| Musculoskeletal | 15 |
| Respiratory | 17 |
Table 7: List of the 12 Therapeutic Areas
A.2 Correlations
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | ||
| 1 | Collaborations | 1 | ||||||||
| 2 | shareclinP~g | -0.014 | 1 | |||||||
| 3 | shareclinP~d | -0.018 | 0.054 | 1 | ||||||
| 4 | sharefirmP~d | 0.01 | 0.002 | 0.091 | 1 | |||||
| 5 | sharefirmP~d | -0.003 | 0.02 | -0.02 | -0.004 | 1 | ||||
| 6 | distance | -1 | 0.057 | 0.093 | 0.162 | 0.022 | 1 | |||
| 7 | colony | 0.023 | 0.003 | 0.002 | 0.027 | 0.0002 | -0.062 | 1 | ||
| 8 | contiguity | 0.004 | -0.019 | -0.002 | -0.034 | -0.008 | - .1834 | 0.142 | 1 | |
| 9 | common_lang_official | 0.039 | 0.011 | 0.059 | 0.091 | 0.014 | 0.008 | 0.197 | 0.043 | 1 |
Table 8: Cross-correlation between variables
A.3 Additional Regressions
Model 2
| Independent Variables (Dependent Variable : Collaborations) |
Model 2 dy/dx (S.E.) |
|---|---|
| shareclinPubactor | -0.063 (0.059) |
| shareclinPubpartner | -0.042 (0.034) |
| distancew | -0.0001*** (0.00002) |
| Common_lang_official | -0.158*** (0.041) |
| Comcol | -0.194*** (0.045) |
| contiguity | -0.129*** (0.028) |
| Period control | Yes |
| Therapeutic Area Control | Yes |
| N | 2756 |
| Log pseudo-likelihood | -31801.44 |
Table 9: Poisson Regression results for Model 2
Model 4
| Independent Variables (Dependent Variable : Collaborations) |
Model (4) dy/dx (S.E.) |
|---|---|
| sharefirmPub_actor | -0.568 (0.85 ) |
| sharefirmPub_partner | -0.038 (0 .79 ) |
| distancew | -0.0007*** (0.00004) |
| Common_lang_official | -2.16*** (0.421) |
| comcol | -2.77*** (0 .468) |
| contiguity | -2.40*** (0.558) |
| Period control | Yes |
| Therapeutic Area control | Yes |
| N | 2756 |
| Log pseudo-likelihood | -32178.144 |
Table 10: Poisson regression results for Model 4
A.4 Test for Over-dispersion
| Model | α |
|---|---|
| Model(3) | 739.7* (345.9) |
| Model(4) | 16.04 (340.6) |
| Model(1) | 553.3** (175.5) |
| Model(2) | 14.5 (261.1) |
Table 11: Test for over-dispersion
Notes
[1] Richa Srivastava is a penultimate year undergraduate pursuing Integrated Masters in Economics at the Indian Institute of Technology Kanpur. Her interests lie in the field of Game Theory, Behavioral Economics and Applied Microeconomics. She has received the Academic Excellence Award for being in the top 5% at IITK across all disciplines. In summer 2011, she was selected as DAAD WISE scholar to pursue undergraduate research at Economics of Innovative Change, Jena, Germany. She plans to pursue management after completing her masters program in 2013
References
Adams, J. D., G. C. Black, J. R. Clemmons and P. E. Stephan (2005), 'Scientific Teams and Institutional Collaborations: Evidence from U.S. Universities', Research Policy, 34 (3), 259-85
Agrawal, K. A. and A. Goldfarb (2008), 'Restructuring research: Communication costs and the democratization of university innovation', American Economic Review, 98 (4), 1578-90
Cameron, A. C. and P. K. Trivedi (1990), 'Regression-based Tests for Overdispersion in the Poisson Model', Journal of Econometrics, 46 (3), 347-64
Cockburn, I. and R. Henderson (1996), 'Public-private Interaction in Pharmaceutical Research', Proceedings of the National Academy of Sciences, 93 (23), 12725-30
Cockburn, I. and R. Henderson (1998), 'Absorptive Capacity, Coauthoring Behavior, and the Organization of Research in Drug Discovery', Journal of Industrial Economics, 46 (2), 157-82
Glickman, S.W., J. G. McHutchison, E. D. Peterson, C. B. Cairns, R. A. Harrington, R. M. Califf and K. A. Schulman (2009), 'Ethical and Scientific Implications of the Globalization', The new England journal of medicine, 360 (8), 816-23
Maina-Ahlberg, B., E. Nordberg and G. Tomson (1998), 'North-South health research collaboration: Challenges in institutional interaction', Social Science Medicine, 44 (8), 1229-38
McKelvey, M., L. Orsenigo and F. Pammolli (2004), 'Pharmaceuticals Analyzed Through the Lens of a Sectoral Innovation System', Sectoral Systems of Innovation: Concepts, Issues and Analyses of Six Major Sectors in Europe, Cambridge, UK: Cambridge University Press, pp. 73-120
Mytelka (2006), 'Pathways and Policies to (Bio) Pharmaceutical Innovation Systems in Developing Countries', Industry and Innovation, 13 (4), 415-35
Plotnikova, T. and B. Rake (2010), 'Collaboration in pharmaceutical research: Exploration of country-level determinants', DIME-DRUID ACADEMY Winter Conference 2011, Denmark
Sonnenwald, D. H. (2007), 'Scientific collaboration', Annual Review of Information Science & Technology, 41 (1), 643-81
Wagner, C. and L. Leydesdorff (2005), 'Mapping the network of global science- comparing international co-authorships from 1990 to 2010', International Journal of Technology and Globalization, 1 (2), 185-208
Wagner, C., I. Brahmakulam, B. Jackson, A. Wong and T. Yoda (2001), Science and Technology Collaboration: Building Capacity in Developing Countries?, Monograph Reports, Santa Monica: RAND Publications
Wuchty, S., B.F. Jones and B. Uzzi (2007), 'The Increasing Dominance of Teams in Production of Knowledge', Science, 316 (5827), 1036-39
To cite this paper please use the following details: Srivastava, R. (2012), 'Developing countries and scientific collaboration in pharmaceuticals', Reinvention: a Journal of Undergraduate Research, Volume 5, Issue 1, http://www.warwick.ac.uk/reinventionjournal/archive/volume5issue1/srivastava Date accessed [insert date]. If you cite this article or use it in any teaching or other related activities please let us know by e-mailing us at Reinventionjournal@warwick.ac.uk.