I

LIBRARY

I Michigan State

Unlverslty

 

 

PLACE IN RETURN Box
to remove this checkout from your record.
TO AVOID FINES return on or before date due.

 

MTE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1/98 WWW“

THE ROLE AND CONTRIBUTION OF FEDERALLY FUNDED
BASIC RESEARCH
IN PHARMACEUTICAL INNOVATION
By

Andrew A. Toole

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Economics

1 998

ABSTRACT
THE ROLE AND CONTRIBUTION OF FEDERALLY FUNDED
BASIC RESEARCH
IN PHARMACEUTICAL INNOVATION
By

Andrew A. Toole

Research and development investment is an important factor in technical
progress and productivity growth. Recent theoretical research has suggested
that the stock Of knowledge created through cumulative research and
development feeds new innovation and economic growth. When knowledge
creation is funded by public institutions, the research results are both non-rival
and non-excludable. Non-rivalry means that the use of public research by one
agent does not preclude its use by another while non-excludability guarantees
that the results from public research are accessible to all agents. Because of
these Characteristics, the public stock of knowledge can feed innovation across
different sectors and industries in the economy.

This dissertation analyzes the role and contribution of publicly funded
basic research in pharmaceutical innovation. Drawing on conversations with
industry scientists, it is argued that public funding of biomedical research
facilitates advances in public scientific understanding and thereby creates new
avenues to therapeutic outcomes and research opportunities for new drug
discovery. As new opportunities emerge, private firms use this information in the

pharmaceutical innovative process to develop new drug concepts and define

therapeutic outcomes. It is in this manner that public basic research contributes
to the discovery of new therapeutic compounds.

The discovery of new compounds is modeled using the economic
production framework. In this framework, basic research can make both a direct
contribution to the discovery of new therapeutic compounds and an indirect
contribution by stimulating private research and development (R&D) investment.
The sum of these individual effects leads to an estimate of the total impact of
public basic research on pharmaceutical innovation.

The quantitative analysis explores the timing, magnitude, and significance
Of the impact that public basic research has on pharmaceutical innovation and
investment. A new panel data set is constructed using medical therapeutic
Classes over time. Detailed data on Public Health Service awards between 1955
and 1985 are matched by therapeutic Class and year with measures of
pharmaceutical innovation, industry R&D investment, Food and Drug
Administration regulatory stringency, and pharmaceutical demand.

The analysis finds strong evidence for an economically and statistically
significant impact of public basic research on pharmaceutical innovation and
investment. The total marginal impact indicates that a $1 million investment in
the stock of public basic research produces 0.07 new chemical entities in each
therapeutic class after an average of seventeen years. This would yield an
average discounted cash flow of $14.2 million in each therapeutic Class at the

time of introduction.

Copyright by
Andrew A. Toole
1 998

To Karen, Brittni, Kevin and my parents.

ACKNOWLEDGMENTS

First and foremost I would like to thank Karen for her incredible patience
and understanding during the completion of this research. Brittni and Kevin
were also supportive. Everyone helped in one way or another to bring this
project to a successful close and I am very thankful and blessed to have them in
my life.

I would also like to thank my parents. Their endless support and positive
attitude allowed me to overcome some of the darkest moments. My father
listened to countless hours of “visionary economics” and helped me clarify my
thoughts on many points.

I am also in debt to John Goddeeris, Jeff Wooldridge, and Wally Mullin for
making this research stronger and more rigorous. I thank each of you for
encouraging careful work, providing timely feedback and supporting me over the

several years of commitment that doctoral work requires.

vi

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................ ix
LIST OF FIGURES ......................................................................................... xi
CHAPTER 1
Introduction ...................................................................................................... 1
1.1 Overview .................................................................................................... 1
1.2 Plan of the Dissertation ............................................................................. 5
1.3 A Primer on Public Basic Research ........................................................... 6
1.4 Literature Review. Multi-Industry Studies ............................................... 10
1.5 Literature Review. Studies of the Pharmaceutical Industry .................... 13
CHAPTER 2
Basic Research and the Process of Pharmaceutical Innovation ................... 18
2.1 Overview .................................................................................................. 18
2.2 The Pharmaceutical Innovative Process: Standard View ....................... 18
2.3 Basic Research and FDA Regulation: Industry Interviews ..................... 23
2.4 Development of the Hypotheses .............................................................. 27
2.5 The Captopril Example ............................................................................ 32
CHAPTER 3
The Analytical Framework, Data, and Estimation Method ............................. 35
3.1 The Analytical Framework ....................................................................... 35
3.11 Issues Regarding Possible Endogeneity .................................... 39
3.2 Data Sources and Construction ............................................................... 42
3.21 Productivity Measure .................................................................. 43
3.22 Industry R&D Measure ................................................................ 44
3.23 Measure of Government Funded Basic Research ...................... 51
3.24 Measure of FDA Regulatory Stringency ...................................... 56
3.25 Measures of Pharmaceutical Demand and Medical Need .......... 58
3.26 Time Effects and Therapeutic Class Effects ............................... 62
3.3 The Estimation Technique ....................................................................... 66
3.31 The Direct Effect ......................................................................... 66
3.32 The Indirect Effect ....................................................................... 70

vii

CHAPTER 4
The Direct Impact Of Public Basic Research on

Pharmaceutical Innovation ............................................................................ 72
4.1 Overview .................................................................................................. 72
4.2 Model Specification ................................................................................. 75
4.3 The Timing of the Relationship ................................................................ 78
4.4 The Magnitude of the Public Research Impact ........................................ 85
4.5 The Statistical Significance of Public Research ...................................... 86
4.6 Alternative Depreciation Rates for Public Research ................................ 89
4.7 The Industry Elasticity and Net Present Value of Public Research ......... 90
4.8 Substitution Possibilities, Returns to Scale, and FDA Regulation ........... 94
4.9 Conclusion ............................................................................................... 97
CHAPTER 5
The Indirect Impact of Public Basic Research on Pharmaceutical Innovation:
Investment in Response to Research Opportunities ................................... 100
5.1 Overview ................................................................................................ 100
5.2 Model Specification ............................................................................... 104
5.3 Functional Form ..................................................................................... 110
5.4 The Regression Results and Discussion ............................................... 112
5.41 Timing of the Relationship ........................................................ 112
5.42 The Magnitude and Significance of PHS Basic Research ........ 114
5.43 Public Regulation and Industry Demand ................................... 116
5 5 Wiggins (1983) Revisited ...................................................................... 118
5. 6 Endogeneity Test for Scientific Opportunity .......................................... 121
5. 7 Conclusion ............................................................................................. 125
CHAPTER 6
The Total Impact of Public Basic Research on Pharmaceutical
Innovation .................................................................................................... 128
6.1 Overview ................................................................................................ 128
6.2 The Total Impacts .................................................................................. 129
6.3 Re-Cap Dissertation Findings ................................................................ 132
6.4 Future Research .................................................................................... 136
APPENDIX A
Keywords Used to Construct PHS Basic Research Variables ..................... 138
APPENDIX 8
Regression Tables ....................................................................................... 144
LIST OF REFERENCES .............................................................................. 182

viii

LIST OF TABLES

Table 3.1 - Tabulation of Counted NCES ...................................................... 45
Table 3.2 — Descriptive Statistics on Counted NCEs by Therapeutic Class..46
Table 3.3 — Pharmaceutical Industry R&D (millions of 1986 $) ..................... 48
Table 3.4 - PHS Biomedical Basic Research (millions of 1986 S) ................ 53
Table 3.5 - Statistics on FDA Regulatory Delay Times (months) .................. 59
Table 3.6 — Pharmaceutical Industry Sales (millions of 1986 S) .................... 63
Table 3.7 - Incidence of Disease (hospital discharges in thousands) .......... 64
Table 3.8 — “Severity" of Disease (hospital days in thousands) .................... 65
Table 4.1 - PHS Basic Research (15% Dep.) Non-Nested Regression

Results with 12 Year Industry Lag .............................................. 80
Table 4.2 - PHS Basic Research (15% Dep.) Non-Nested Regression

Results with 14 Year Industry Lag .............................................. 81
Table 4.3 — PHS Basic Research (15% Dep.) Nested Regression

Results with 12 Year Industry Lag .............................................. 83
Table 4.4 — PHS Basic Research (15% Dep.) Nested Regression

Results with 14 Year Industry Lag .............................................. 84
Table 4.6 - 10% Depreciation - PHS Basic Research Non-Nested

Regression Results for 12 Year Industry Lag ............................. 91
Table 4.5 - 20% Depreciation — PHS Basic Research Non-Nested

Regression Results for 12 Year Industry Lag ............................. 92
Table 4.7 - CES Production Function Estimation .......................................... 95
Table 5.1 - Regression Results for the Indirect Effect of

PHS Basic Research (15% Dep.) ............................................. 113
Table 5.2 - Re-Examination of the Wiggins Model ..................................... 119
Table 5.3 — PHS Basic Research ................................................................ 124
Table A1 - Listing of Class Filter Keywords or Character Strings .............. 139
Table A2 - Listing of Exclusion Filter Keywords or Character Strings ....... 140
Table A3 - Activity Code Breakdown ......................................................... 141
Table 81 — Industry Lag 12, PHS Lag 15 ................................................... 144
Table 8.2 — Industry Lag 12, PHS Lag 16 ................................................... 145
Table 8.3 - Industry Lag 12, PHS Lag 17 ................................................... 146

Table 8.4 - Industry Lag 12, PHS Lag 18 ................................................... 147

Table 8.5 — Industry Lag 12, PHS Lag 19 ................................................... 148
Table 8.6 - Industry Lag 12, PHS Lag 20 ................................................... 149
Table 8.7 — Industry Lag 14, PHS Lag 15 ................................................... 150
Table 8.8 - Industry Lag 14, PHS Lag 16 ................................................... 151
Table 8.9 — Industry Lag 14, PHS Lag 17 ................................................... 152
Table 8.10 - Industry Lag 14, PHS Lag 18 ................................................. 153
Table 8.11 - Industry Lag 14, PHS Lag 19 ................................................. 154
Table 8.12 - Industry Lag 14, PHS Lag 20 ................................................. 155
Table 8.13 - Industry Lag 12, PHS Nested 12,17,22 .................................. 156
Table 8.14 — Industry Lag 12, PHS Nested 9,17,25 .................................... 157
Table 8.15 — Industry Lag 14, PHS Nested 12, 17, 22 ................................ 158
Table 8.16 — Industry Lag 14, PHS Nested 19, 17, 25 ................................ 159
Table 8.17 — Industry Lag 12, PHS Lag 15 (10% Depreciation) ................. 160
Table 8.18 - Industry Lag 12, PHS Lag 16 (10% Depreciation) ................. 161
Table 8.19 — Industry Lag 12, PHS Lag 17 (10% Depreciation) ................. 162
Table 8.20 — Industry Lag 12, PHS Lag 18 (10% Depreciation) ................. 163
Table 8.21 - Industry Lag 12, PHS Lag 19 (10% Depreciation) ................. 164
Table 8.22 - Industry Lag 12, PHS Lag 20 (10% Depreciation) ................. 165
Table 8.23 — Industry Lag 12, PHS Lag 15 (20% Depreciation) ................. 166
Table 8.24 - Industry Lag 12, PHS Lag 16 (20% Depreciation) ................. 167
Table 8.25 - Industry Lag 12, PHS Lag 17 (20% Depreciation) ................. 168
Table 8.26 — Industry Lag 12, PHS Lag 18 (20% Depreciation) ................. 169
Table 8.27 - Industry Lag 12, PHS Lag 19 (20% Depreciation) ................. 170
Table 8.28 - Industry Lag 12, PHS Lag 20 (20% Depreciation) ................. 171
Table 8.29 — CES Estimation, Industry Stock, PHS Lag 17 ........................ 172
Table 8.30 - Indirect Effect, PHS Concurrent Lag ...................................... 173
Table 8.31 - Indirect Effect, PHS Lag 1 ...................................................... 174
Table 8.32 - Indirect Effect, PHS Lag 2 ...................................................... 175
Table 8.33 - Indirect Effect, PHS Lag 3 ...................................................... 176
Table 8.34 - Indirect Effect, PHS Lag 4 ...................................................... 177
Table 8.35 - Indirect Effect, PHS Lag 5 ...................................................... 178
Table 8.36 - Indirect Effect, PHS Lag 6 ...................................................... 179
Table 8.37 - Indirect Effect, PHS Lag 7 ...................................................... 180
Table 8.38 - Indirect Effect, PHS Lag 8 ...................................................... 181

LIST OF FIGURES

Figure 1.1 — Basic Research Funding (real 1987 S) ........................................ 8
Figure 1.2 - Public Health Service Study Sections, Activity Codes and

Institutes ...................................................................................... 9
Figure 2.1 — The Pharmaceutical Innovative Process

(Source: Bloom (1976)) ............................................................ 19
Figure 2.2 - Where Basic Research Fits into the Pharmaceutical

Innovative process ..................................................................... 28
Figure 2.3 - Captopril Example ..................................................................... 33
Figure 3.1 - Cardiovascular, Anti-infective and GastrolGenito-urinary

Counted NCEs ........................................................................... 47
Figure 3.2 - Cardiovascular, Anti-infective and GastrolGenito-urinary

Industry R&D (millions of 1986 S) .............................................. 49
Figure 3.3 - Cardiovascular, Anti-infective and GastrolGenito-urinary

PHS Basic Research (millions of 1986 S) .................................. 54
Figure 3.4 - Cardiovascular, Anti-infective and GastrolGenito-urinary

FDA Review Times (months) ..................................................... 60
Figure 3.5 — Cardiovascular, Anti-infective and GastrolGenito-urinary

Industry Sales (millions of 1986 S) ............................................ 63
Figure 3.6 — Cardiovascular, Anti-infective and GastrolGenito-urinary

Hospital Discharges ................................................................... 64
Figure 3.7 - Cardiovascular, Anti-infective and GastrolGenito-urinary

Disease Inpatient Days .............................................................. 65
Figure 4.1 — Pharmaceutical Innovation Levels with Slower Real

Basic Research Funding Growth (0.06% slower) ...................... 87

xi

Chapter 1

Introduction

...the most important aspect of research management is the ability
to access the scientific information emanating from academic and
industrial laboratories throughout the world, to discern its
importance, and to integrate it rapidly into our research.

Edward M. Scolnick, MD.

President of Human Health
at Merck & Co., Inc.

1 .1 Overview

Following World War II, the United States Government established itself
as the leading financial supporter of basic research. Fueled by the belief that a
strong basic research foundation would lead to productivity increases and
greater national welfare, federal funding of basic research grew at a real annual
rate of 11.8% through the 1960s. However, beginning with the economic
challenges of the 1970s and the concomitant pressures on the Federal budget,
policy debates resulted in the reduction in the growth rate of federal funding for
basic research. For the 1970-1989 period, the real annual growth rate dropped
to 2.8% (NSF (1992)).

It is evident from current and past debates that competing demands for
federal funds have caused policy makers to re-evaluate national research policy.
The traditional belief in basic research as a means to increase national and

industrial production has been questioned. At least in part, this policy

debate is perpetuated by a shortage of economic analyses describing and
quantifying the impact of federally funded basic research. Since basic research
is that research which builds fundamental knowledge within a scientific
discipline, its connection and contribution to industrial innovation is less obvious
than applied research and development. Nevertheless, the impact of basic
research can be examined using the standard production relation described in
the economics literature.

The production framework identifies two related channels through which
federally funded basic research can influence industry innovation. First, basic
research can contribute directly to private industry's product innovation. In this
case, basic knowledge is used as a direct input to create the new product
introduced by private industry. For example, pharmaceutical firms use the basic
research results on biological and chemical processes to conceive therapeutic
outcomes and synthesize new therapeutic compounds. Second, federally
funded basic research can contribute indirectly to private industry's product
innovation. This indirect contribution recognizes the role that basic knowledge
plays in stimulating additional private R&D investment. Often times additional
private R&D is needed to strengthen and extend the foundation of knowledge
created by public financing. Accounting for both the direct and indirect effects
leads to an estimate of the overall impact of federally funded basic research on
industrial product innovation.

In general, US. support for basic research has gone to universities and

colleges to fund academic research. A recent survey by Mansfield (1995)

reports that federal funds constitute two-thirds of the total research dollars used
by academic researchers. Outside of the agricultural sector, econometric
analyses of basic research have focused primarily on the link between academic
research and manufacturing innovation and productivity. In this vein, several
probing multi-industry studies have been done in recent years. Jaffe (1989)
uses the production framework to estimate the elasticity of corporate patents
with respect to the stock of academic research spending. Grouping industries
and academic departments into "technological areas," Jaffe finds that academic
research makes a significant contribution to commercial innovation. Using
Jaffe's model and a measure of new business innovation, Acs et al. (1991) also
find that academic research contributes to industry innovation. Adams (1990),
using knowledge stocks constructed with weighted counts of journal articles,
finds that fundamental knowledge either increases or decreases manufacturing
productivity depending on the lag specified.

Although the generality of the multi-industry approach is one of its
strengths, it is also one of its weaknesses. It is difficult to infer from these
studies how basic research has impacted an individual industry. The best we
can do is acknowledge the general importance of this research to the various
technological areas studied. As will be seen below, our focus on an individual
industry motivates the mapping of research between sectors and allows us to
control for industry specific characteristics. Moreover, previous studies do not
distinguish between federal and non-federal funding sources. In order to gauge

the impact of federally supported basic research on industrial innovation and

gather information for policy evaluation and formulation, we must explicitly
analyze the impact of this research.

This dissertation provides an estimate of the “total impact” of US.
Government funded basic research on pharmaceutical innovation by combining
separate estimates of the direct and indirect effects. Using insights gained from
interviews with industry scientists, the analysis begins by describing the role of
basic biomedical research in the pharmaceutical innovative process. This
qualitative discussion provides the basis of the four primary hypotheses tested in
the empirical analysis: first, public basic research has a positive and significant
direct impact on pharmaceutical innovation; second, public basic research has
its greatest impact in the formulation of "drug concepts" which come in the
earliest stage of drug discovery; third, the distinctive contributions of public basic
research and private industry R&D to the pharmaceutical innovative process
suggest limited substitution possibilities between these two types of research;
and fourth, public basic research indirectly impacts pharmaceutical product
innovation through its affect on industry R&D investment. Furthermore, the
potential for increasing returns to scale in pharmaceutical innovation and the
effects of product quality regulation by Food and Drug Administration (FDA) are
also explored.

This study introduces basic research by therapeutic class into the
economic model of pharmaceutical innovation. The data used to measure basic
biomedical research consists of all extramural grant and contract awards given

by the Public Health Service (PHS) between 1955 and 1985. The PHS funds the

largest portion of all basic biomedical research, whether one considers public,
private, or both sectors. In 1985, the PHS financed 80% of all US. Government
obligations for national health R&D, covering 66% of all non-industry funding
(NIH data book (1989)). Further, this study incorporates two key characteristics
Of basic research. First, basic research is allowed to be cumulative over time.
This captures the important quality of Ieaming that builds on previous research
knowledge. Second, basic research is lagged so that its hypothesized timing
can be tested and explored statistically.

The federal awards data are combined with standard measures of
pharmaceutical innovation, industry R&D expenditure, and regulatory stringency.
Approved new chemical entities (NCEs), a measure of important patents, are
used to represent pharmaceutical innovative success. Industry R&D
expenditure figures were obtained from the Pharmaceutical Research and
Manufacturers Association (PhRMA). The FDA supplied the necessary data to

calculate measures of regulatory stringency.

1.2 Plan of the Dissertation

Chapter two describes the role of public basic research in the
pharmaceutical innovative process and outlines the main hypotheses to be
tested. Chapter three presents the general modeling framework, describes the
data and discusses the estimation techniques. Chapter four describes the

empirical specification and results of the direct impact of public basic research

on pharmaceutical innovation. Chapter five presents the empirical specification
and results of the indirect impact of public research on pharmaceutical
innovation. Chapter six concludes the analysis and calculates the total impact of

publicly funded basic research.

1.3 A Primer on Public Basic Research

An important part of understanding the relationship between federally
funded basic research and industry innovation is understanding the institutional
base from which this research is created. In general, the stock of public basic
research is the output of the US. national research enterprise. The US.
national research enterprise is the set of government, industrial and academic
institutions that conduct research in the United States. With respect to basic
research, the US. national research enterprise essentially refers to the
aggregate of US. academic institutions. Under federal sponsorship, the
academic research enterprise has grown into a tremendously diversified
institutional base supplying the largest portion of the stock of public basic
research.

The National Science Foundation defines basic research as that which is
directed toward “a fuller knowledge or understanding of the subject under study,
rather than a practical application thereof” (NSF, 1985, P. 221). It is the
research which builds fundamental knowledge within a scientific discipline. The

definition of basic research used in this study encompasses the NSF definition

but also adds that the research must be non-clinical and relevant to the
pharmaceutical industry. Figure 1.1 plots overall federal basic research
expenditures and the PHS basic research awards (across all classes) used in
this study for the years 1960 - 1985. For comparative purposes, both series are
presented in real 1987 dollars using the implicit GDP deflator. It is worth noting
that the PHS awards trend is not characterized by the rapid increases and
decreases that are evident in the trend of overall federal basic research funding.
The Public Health Service is the key governmental body in charge of the
allocation of federal research money for biomedical and pubic health projects.
This is accomplished through the use of “study sections” or peer review groups
which recommend approval of grant applications. Each recommended grant will
have an activity code which describes the nature of the proposal in broad terms
such as research, fellowship, training, etc. The set of recommended
applications are sent on to get final committee approval from the particular
institute that will fund the study. Figure 1.2 plots the total number of PHS activity
codes, the number of institutes and the number of study sections from 1955 to
1985. There are two interesting points with regard to this plot. First, the rapid
expansion of PHS funding through the 19603 is evident in the growth of the
number of study sections, activity codes and institutes during this period.
Likewise, the steady-state which characterized the 19708 is also evident.
Finally, it is important to remember the public goods aspect of federally

financed basic research. Projects funded with federal monies are part of the

cocaomem
23m wIa l+l

583mm
23m .825“. =< IAYI

 

 

 

 

3 “we .3: 059.5... cocmowcm 06mm - : 9:9“.

 

ozonQezozezQQQQQQ
zoe@%e(o(a(e(o(%oo%eooo
T, r TLTT- TLITT TL.-- .T-+l+ +LI TITLni ITTTITTI o
\+ l+tl+ I+.€.....+I+I+I+I+I+ I

+l+l+\+l+l+r Tilt/f

ooow

so?c

oooo

- .. ooom

- oooow

 

. ooom F

 

8.2.3:. .25 880 5.2.94 @8595 Sam 8.28 5.8... 2.8.. - we 9:9“.

     

Av! Av! 01 Au! 1. I. I.
0v. 0. I? as coco «90 coco

 

 

8552..— u a
mmuoo Ed.“ I
mm». I

00..

om;

 

 

.. DON

 

0mm

 

 

10

public domain. As such, the results of this research add to the public stock of
knowledge and possess two key characteristics: non-rivalry and non-
excludability. Non-rivalry means that the use of public research by one agent
does not preclude its use by another while non-excludability guarantees that the
results from public research are accessible to all agents. Because of these
characteristics, the public stock of knowledge can feed innovation across

different sectors and industries in the economy.

1.4 Literature Review: Multi-Industry Studies

The focus of this dissertation draws on two lines of research in the
economics literature. The first is the literature regarding the measurement of
spillovers from government and academic research to industry research and
development. In this vein, the central focus has been on studying multi-industry
flows.

Mansfield (1991) uses survey data from 76 firms in seven manufacturing
industries, one of which is the ethical drug industry, to look at the extent to which
technological innovations have been based on recent academic research; the
time lags between academic research discoveries and commercial utilization;
and the social rate of return from academic research. Mansfield finds that the
drug industry has the highest percentage of new products and processes that
could not have been developed (without substantial delay) in the absence of

recent academic research. However, his study was not designed to produce an

11

estimate of the extent of these spillovers. A second study directed by Mansfield
appears in his book, Research and Innovation in the Modern Cormration,
published in the early 19708. In the chapter on the ethical pharmaceutical
industry, Jerome Schnee identifies the discoverer of sixty-eight drug innovations
spanning two separate time periods, 1935-1949 and 1950-1962, and presents
some descriptive statistics. One of the major conclusions of this exercise is that
academe, as a source of drug innovations, has accounted for fewer drug
discoveries relative to industry sources over time. He defines discovery as the
first identification of a drug's biological activity. That is, he considers the first
identification of the therapeutic action of the drug.

Link (1981) appears to be the first study to analyze the impact of
government financed basic research on industry productivity. The analysis
separates both company financed and government financed research into
applied and basic portions. Using data from fifty-four manufacturing firms, Link
finds that government financed basic research has a positive and significant
effect on total factor productivity. His elasticity estimate for government financed
basic research is 1.17. However, it is important to note that Link measures
direct government payments to the firms themselves and not the stock of public
research available outside of the firms.

Jaffe (1989), in an important study relating academic research to industrial
research, looks at the spatial relationship between academic and industrial
research and development. It belongs to a class of R&D spillover models that

use some measure of "technological distance" separating the origin and

12

destination (the receiving industry) of the innovation. He estimates a three
dimensional simultaneous equation panel data model in which observations are
indexed by state, "technological area," and time. The innovative output measure
is corporate patent counts and the innovative inputs are measures of industry
and academic R&D expenditure aggregated to the level of the technological
area.

Importantly, Jaffe distinguishes between basic and applied R&D in his
analysis. He finds significant spillovers from university R&D to industrial R&D.
While the multi-industry Character of his study is both a strength and weakness,
his work serves as an important step in understanding the relationship between
external knowledge and commercial innovation.

Adams (1990) is a multi-industry study of the relationship between
manufacturing total factor productivity and academic research. There are three
unique elements to his analysis: first, he uses a labor weighted count of journal
articles to measure the stock of “fundamental” knowledge; second, he analyzes
the lag between the research and its impact on productivity; and third, he
includes knowledge stock measures of within industry knowledge and external
knowledge. Although he finds that fundamental knowledge either increases of
decreases manufacturing productivity depending on the lag specified, the data

reveal that the “science only” spillover lag is about thirty years.

1.5 Literature Review: Studies of the Pharmaceutical Industry

The second line of research consists of those studies which focus
exclusively on the pharmaceutical industry. This research falls mainly into two
camps: studies of FDA regulatory stringency and studies of research investment
and productivity. With regard to the studies of FDA regulation, Martin Baily
(1972) and Grabowski et al. (1978) (GVT) estimate the direct effect of regulation
using [new chemical entities/R&D] as their measure of pharmaceutical
productivity. Having a continuous dependent variable, they use OLS to estimate
their models. The only difference in their approaches is that GVT use the United
Kingdom as a control for non-regulatory factors.

There are three important non-regulatory factors identified: the
thalidomide disaster; advances in pharrnacologic science; and the depletion of
research opportunities. The idea of the depletion of research opportunities is of
direct interest. Baily attempts to capture this using a moving average of past
NCE introductions; however, GVT show that this measure is insignificant when a
larger sample period is used. GVT prefer to use the UK. as a control for the
relationship between basic biomedical research and pharmaceutical research by
assuming that advances in basic research affect both the US. and UK
pharmaceutical industries the same. The final important point regarding the
GVT analysis is their method of measuring the regulatory stringency. GVT use
the mean time from submission of the new drug application to approval in each

year of their sample. Their sample covered 1954-1974.

13

14

Wiggins (1979) expands on this work in two ways: (1) he estimates the
relationship for each therapeutic class;1 and (2) he estimates the total effect not
just the direct effect. Wiggins notes the importance of “scientific knowledge” in
the pharmaceutical firms’ decision process. In fact, he states, “In conclusion, it
must be reemphasized that the most important consideration [in choosing a
research project] is still the scientific one.” (Wiggins, 1979, p. 69) This being
said, Wiggins has no way to account for changing scientific knowledge nor can
he estimate its relationship to pharmaceutical drug innovation. Similar to GVT,
Wiggins uses the mean time from NDA submission to approval for each class for
each year. Finally, because Wiggins uses NCES as his dependent variable, he
uses a Tobit estimation technique in an attempt to account for the truncation at
zero.

Using firm level data, Jenson (1987) improves on the estimation
technique for the model by using a Poisson specification along the lines of
Hausman, Hall, and Griliches (1984). She uses the same measure of regulatory
stringency as GVT and includes a time trend and other interacted variables in an
attempt to account for scale economies in pharmaceutical R&D.

Econometrically, Jenson maintains the nominal variance assumption of the
Poisson model (that is the mean = variance property of the Poisson distribution).
Jenson does inspect the off diagonal elements of the covariance matrix formed

using the standardized residuals for evidence of serial correlation and concludes

 

1 A therapeutic class is a grouping of compounds based on their treatment
indication.

15

that it “does not appear to be a serious problem.” Her results indicate that
regulatory stringency decreases the expected number of new chemical entities
while firm size has no significant affect on the marginal productivity of research
expenditure.

There are two existing studies that focus on pharmaceutical research effort
and productivity: Ward and Dranove (1995) and Henderson and Cockbum
(1996). Ward and Dranove measure R&D spillovers from government—funded
basic research to pharmaceutical applied research. The authors are interested
in the magnitude and lag structure characterizing R&D spillovers between basic
research and applied research. They construct a panel data set of therapeutic
classes covering the period 1966-1988.

They measure spillovers in two ways. First, by regressing industry R&D
expenditure on a count of journal articles and other variables, they find that a 1%
increase in journal articles results in a 0.22%—0.36% increase in industry R&D
expenditure. Second, the authors regress industry R&D expenditure on the
National Institutes of Health (NIH) obligations broken-down by institute and
aggregated, as closely as possible, into therapeutic classes. Within the same
therapeutic class, a 1% increase in NIH obligations leads to an increase in
industry R&D expenditure by 0.57%-0.76% (cumulative over all lags). Further,
an "indirect" spillover is associated with NIH obligations in other therapeutic
classes. This effect has industry R&D expenditure increasing 1.26%-1.71% in
response to a 1% increase in NIH obligations.

Two things are important to note about the Ward and Dranove analysis.

16

First, their study explores the determinants of R&D inputs not outputs.
Consequently, while this study is interesting and informative, the role of R&D
spillovers in the successful deveIOpment of usable output remains an open
question. Presumably, the spillovers of inputs contribute to individual firm
success in drug development by either reducing cost or increasing the likelihood
of success by providing better leads in drug discovery. Second, the NIH funding
data that they use consists of the total obligations of the NIH for a particular
institute in a given year. Research obligations correspond to present and future
funding commitments while awards measure the actual financial outlays in a
given year. Although there is not much difference between the two, using
obligations might affect the timing of the relationship since obligations typically
lead awards by one year.

Henderson and Cockbum (1996) look at firm specific economies of scope
and scale as well as inter-firm spillovers in the drug discovery stage of
pharmaceutical R&D.2 They use disaggregated proprietary firm data at the level
of the research program. It is organized as a panel data set which contains
detailed R&D input (investment) data from ten pharmaceutical firms spanning a
period of twenty years. Innovative output is measured by a count of "important"
Composition of Matter patents. This is defined as a patent granted after the

discovery stage of research in two of three major markets.

 

2 In the Henderson and Cockbum analysis, economies of scope exist within the
firm if physical assets or personnel can be used in more than one application at
no extra cost. Economies of scale exist if fixed costs can be distributed over a

17

Their analysis incorporates both direct and indirect inter-firm spillovers.
Direct spillovers occur between firms that are involved in research in the same
therapeutic class. For instance, this is the case when two firms both conduct
R&D on drugs for the cardiovascular system. Indirect spillovers between firms
are those which occur across related therapeutic classes. In this case, research
on blood and blood forming organs may uncover a result useful in
cardiovascular drug R&D. Both measures were found to have positive and

significant coefficients indicating the presence of inter-firm spillovers.

 

larger research effort or if the firm can hire specialized researchers as their total
research effort grows.

Chapter 2

Basic Research and The Process of Pharmaceutical Innovation

2.1 Overview

This chapter develops the qualitative basis for the hypotheses that are
tested in the empirical section of the dissertation. The chapter begins with a
description of the standard schematic of the pharmaceutical innovative process.
As will become clear, this schematic does not explicitly describe the role and
contribution of public basic research in the discovery and development of new
therapeutic compounds. To gain an understanding of where basic research fits
in the pharmaceutical innovative process, previous research is supplemented by
interviews with industry scientists and administrative personnel. Following a
description of the interview responses, a new schematic showing the role of
basic research is presented and the four main hypotheses regarding public basic
research are described. The chapter ends with a description of the discovery of
captopril, an example of how public basic research can contribute to

pharmaceutical innovation.

2.2 The Pharmaceutical Innovative Process: Standard View

Figure 2.1 gives a detailed breakdown of the pharmaceutical innovative

process. This process is divided into two broad stages: drug discovery and

18

A85: E85 .8509 $805 o>:m>o:c_ _mo_SoomE.m.._n_ of. .. ...N 959”.

19

 

OZ—
m3o_>om ‘II

(on.

 

 

 

 

 

beam
.3
5.529"...

.952

 

 

 

 

38%.
.8
:o_fi:_m>m
.mE_c<

 

 

 

 

£8556
$289.0

   

MOE—nuan— fihmmmmm

 

 

 

008.305.

20

<D z‘l

 

 

5 Sea

 

 

 

 

.. $2,.

 

 

 

 

. $2.“.

$.83 - rm 2:9“.

 

 

 

 

ooEEEoo
Bosom
58.2.2.

 

 

21

drug development. Discovery is represented by the first page of Figure 2.1 while
development is shown on the following page of Figure 2.1. The diagram
indicates that pharmaceutical firms need a set of inputs - knowledge, skilled
manpower, research facilities and funding - in order to begin the innovative
process. Having these resources, firms begin drug discovery by defining a
therapeutic outcome and synthesizing potential compounds that may lead to the
desired outcome. After testing or “screening” different chemical combinations, a
candidate compound, or group of compounds, is identified. Following Chemical
analysis to establish certain stability and reproduction properties necessary for
mass production, the firm begins animal tests for efficacy and safety.

The two indicators of success in the drug discovery stage are patents and
investigational new drug applications (lNDs). Patent applications are filed
following proof of efficacy in animals. It is commonplace for firms to patent large
groups of related compounds, referred to as a “patent estate.” This is done both
as a strategic move to block out potential competitors and because of the
uncertainty regarding which of the compounds will be best suited for the
development stage of research. As soon as safety requirements are satisfied,
the firm must submit an IND to obtain FDA permission for human testing.

The three pre-marketing clinical development phases appear in the
continued portion of Figure 2.1. The initial phase (phase I) provides the first
information in human subjects on the tolerance, absorption, and elimination of
the compound. Phase II are the first investigations into the drug’s dose-

response relationship to efficacy in human patients. The final pre-marketing

22

phase, phase III, uses large-scale clinical trials to establish efficacy and safety in
the target patient population. If all phases are successful, the firm will file a new
drug application (NDA) while continuing long-term clinical trials during the FDA
approval period. Once the firm receives the approved NDA, market sales are
legal (DiMasi et al. (1991), OTA (1993)).

It is fundamental to recognize that the pharmaceutical innovative process is
highly knowledge intensive. The standard schematic shows the progression
from chemical synthesis to NDA submission. Along this progression, knowledge
describing the characteristics of both the candidate compound and its close
chemical relatives is continuously accumulated. This additional knowledge is
used to revise expectations concerning a project’s scientific feasibility and
ultimate market importance (see Wiggins (1981) for a detailed discussion of the
decision-making process).

Given the highly knowledge-intensive nature of the innovative process, it is
hardly surprising that pharmaceutical innovation requires long periods of time.

In a comprehensive analysis on the cost of new drug development, DiMasi et al.
(1991) find that the average time from synthesis to FDA approval is nearly
twelve years for the 1979-1989 period. The Pharmaceutical Research and
Manufacturers Association also report a twelve year lag between synthesis and
approval. Further, in a study looking at the relationship between drug
importance and time to market, Dranove and Meltzer (1994) suggest the period
from patent application to FDA approval is twelve to fourteen years.

It is further evident that the FDA, a component of the Public Health Service,

23

has imposed several regulatory requirements at various stages of the process.
The two most obvious requirements are the investigational new drug application
(IND) and the new drug application (NDA). Although broad guidelines exist,
these applications require specific types of data and procedures depending on
the compound involved. Moreover, within the application, each claim must be
supported by extensive documentation. There has been a fair amount of
research investigating the impact of FDA product quality regulation on
pharmaceutical innovation (Baily (1972), Peltzman (1973), Grabowski et al.
(1978), Wiggins (1981 , 1983), Jensen (1987), Thomas (1990)). Although
opinions on timing and magnitude differ among the studies, all conclude that

regulation has lowered pharmaceutical innovation.

2.3 Basic Research and FDA Regulation: Industry Interviews

Within the standard schematic of the pharmaceutical innovative process
(Figure 2.1), the role of basic research is not explicit. Interviews with industry
scientists revealed that the primary role and contribution of basic research
comes in the drug discovery stage of industry research. In fact, they identify
basic research as feeding an independent step in the discovery stage called the
“rock turning” or “drug concept” period. This is the very first point in the
pharmaceutical innovative process and necessarily precedes chemical
synthesis. By providing greater understanding of biological and chemical

processes and structures, basic research creates a foundation of knowledge

24

which Opens up new avenues to therapeutic outcomes.

In the typical case, the basic research that leads to the discovery of new
therapeutic compounds is a combination of publicly available biomedical
knowledge and a firm’s own basic research. In a recent paper, Henderson and
Cockbum (1997) point out that drug discovery is characterized by a high degree
of public and private interaction in research. While this is undoubtedly true, our
analysis suggests that the “net floW’ of ideas is from publicly funded basic
research to pharmaceutical industry research. This not the “simple waterfall
model” of innovation. There is a high degree of complexity and creativity in the
process of drug discovery. Nevertheless, there is a progression in research and
learning. To the extent that firms monitor and use the advance of public medical
understanding in their research, they can begin the process of drug innovation
with something other than a “blank chalkboard.”

Industry scientists point out that the continually expanding stock of public
basic research knowledge creates both new opportunities for therapeutic
outcomes and new approaches to chemical screening. The new opportunities
stem mainly from advances in our understanding of metabolic processes in
normal and disease states while, in the chemical screening step, more clearly
defined therapeutic outwmes are combined with structural design methods that
utilize computer and electronic equipment. By monitoring the advances in public
basic science, the pharmaceutical industry absorbs and extends this “core
knowledge” with an eye toward the ultimate commercial products that may be

produced.

25

~ Since the structure and objectives of the pharmaceutical innovative
process are determined in part by the regulatory requirements of the Food and
Drug Administration, industry regulatory personnel were asked to comment on
the impact of this regulation. In particular, given that NDA review times are the
traditional measure of FDA regulation used in the economics literature, they
were asked to comment on the strengths and weaknesses of the review time
measure.

Interviews revealed that there are some real limitations with the average
review time measure of FDA regulatory stringency. First, pharmaceutical firms
must keep FDA requirements in mind beginning with the animal tests early in the
drug discovery stage of the innovative process. The results and procedures
used in animal efficacy and safety tests are reviewed by the FDA before clinical
(human) tests can begin. Moreover, the clinical testing protocols used in the
three development phases are closely reviewed. This supports the standard
argument that FDA regulation has, over time, contributed to longer delays in the
overall pharmaceutical innovative process. Since the early 19603, synthesis to
approval times have increased from about 3 to 14 years (DiMasi et al. (1991 )).

Second, and perhaps more importantly, industry regulatory personnel
point out that, since at least the late 19703, the industry has followed a broad
regulatory strategy to minimize expected review times. This strategy is designed
to achieve shorter NDA review times by establishing a relationship with the FDA
very early in the innovative process and consulting with the FDA about a

particular compound. The strategy is intended to lower overall costs by

26

eliminating some uncertainty regarding the FDA approval criteria as well as to
increase revenues by realizing sales more quickly for these completed
therapeutic compounds. Overall, this seems to indicate that a large proportion
of regulatory cost is borne by pharmaceutical firms before the NDA review
period.

Third, the regulatory impact is not only occurring earlier but it appears
that pharmaceutical firms have internalized much of the cost of FDA regulation.
The creation and growth of the “government affairs” departments, which are
intended to specialize in regulatory affairs, illustrate the increasing internal
resources devoted to regulation. Although these resource costs could be
measured with detailed proprietary firm data, public data include some portion of
the regulatory resource costs as part of the overall research and development
expenditure figures. This highlights the problems with the composite public
numbers used in the subsequent empirical analysis and should be kept in mind
when interpreting the magnitudes of both industry R&D and FDA regulation
coefficients.

All of these limitations notwithstanding, average NDA review times do
provide some measure of FDA regulatory stringency. This period still represents
the largest regulatory hurdle confronting pharmaceutical firms trying to bring new
compounds to the market. Moreover, to the extent that firms are either
unsuccessful with their minimizing strategies or do not anticipate changes in
regulatory requirements, NDA delay times capture part of the burden of

regulation.

27

2.4 Development of the Hypotheses

Figure 2.2 presents the basic research augmented schematic of the
pharmaceutical innovative process. The essential change to the standard
schematic is the addition of a drug concept stage or step which precedes
Chemical synthesis in drug discovery. Although this step is implicit in the
standard view, making this step explicit is important for two reasons: first, it is
absolutely fundamental in that it defines the therapeutic outcomes which
determine the goals of the entire innovative process; and second, it moves away
from the danger of seeing drug discovery as the mechanistic application of
physical and financial capital and toward the view that human capital or
knowledge capital drives the innovative process. Within the drug concept step,
public basic research is combined with proprietary basic research and creative
thinking to develop a set of potentially valuable therapeutic outcomes. Along
with the underlying biomedical and chemical basic research, these drug
concepts are fed into the chemical synthesis step where computer design,
Chemical engineering and screening technologies are applied.

Figure 2.2 shows that the stock of public basic research, which is produced
mainly by academic researchers, is made possible through the financing from
four primary sources: Public Health Service (the major component of the

Department of Health and Human Services), other federal agencies, state and

28

$305 o>um>o=£ Rosana—ESE DE SE SE £0383. Diem 22;? - Nd EsmE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

¢
565.com
Egan
Eoiéo Z
BBZE
moi—620m . EoEEoEO
“3:309:00 . 33:35 . Good 28 88m
IV 388m
maroofiwcm 5.833— 850
3282—0 . Diem BBZE
98 235m
wE—oooz 5253 838m
DDSQEoU . 58885 . 5:3:
0:95
mica—Him 3252.0 E850 wen

 

 

 

:oSomcm 06.3
0:95 .8 03on

 

29

locate government, and private nonprofit institutions.‘ This analysis focuses on
PHS funded extramural basic research. The PHS accounted for 80% of US.
Government funds for national health R&D in 1985, which is 66% of all non-
industry funding (NIH data book (1989)). Of the remaining 34% of non-industry
health R&D, other federal institutions account for 17%, state and local
government account for 11% and private nonprofit institutions make up 6%. In
terms of basic research, the NIH (a component of the PHS) accounts for 39% of
all federal basic research obligations with 67% going to colleges and
universities.

Economic analysts have just recently begun to investigate the relationship
between advances in medical science and the pharmaceutical industry
(Henderson(1994), Grabowski and Vernon (1994), Henderson and Cockbum
(1996), and Gambardella (1992, 1995)). Using a detailed example drawn from
cardiovascular drug discovery, Henderson (1994) argues that those
pharmaceutical firms which possess a greater ability to integrate research
information have a competitive advantage in the industry. Grabowski and
Vernon (1994) note its importance in the rise of biotechnology firms. Henderson
and Cockbum (1996) and Gambardella (1992) analyze research spillovers
between firms and describe how public medical research has improved the

screening process used in drug discovery.

 

‘ To a much lesser degree, publicly available basic research results also come
from firms within the industry and foreign sources. lntra-industry sources are
small and subject to long lags due to the high degree of secrecy regarding
research opportunities while little is known about foreign sources.

30

Unlike previous analyses, this study focuses on the empirical relationship
between the stock of publicly funded biomedical research and drug discovery.
Since this is the dominant funding source creating the research that feeds drug
concept development, public basic research serves as a core driver of industry
innovation. Although specifying the mechanisms through which public research
is monitored, integrated and utilized by pharmaceutical firms is a high research
priority, the first step toward understanding begins with identifying, measuring,
and testing the broader empirical importance of this relationship. Our objective
is to test the timing, magnitude and significance of the impact that public
research has on pharmaceutical innovation and industry R&D investment.
When combined with the qualitative evidence, this objective leads to the
following testable hypotheses concerning core knowledge created by publicly

funded basic research:

Hyppthesis 1: Core knowledge produced by public basic research funding
has an economically and statistically significant direct effect on pharmaceutical
innovafion.

Hypothesis 2: Core knowledge produced by public basic research funding
feeds the drug concept stage of industry research and, therefore, has its
greatest impact in the earliest stages of drug discovery.

Hypothesis 3: The distinctive contributions of public basic research and
private industry R&D to the pharmaceutical innovative process suggest limited
substitution possibilities between these two types of research.

Hyppthesis 4: Core knowledge produced by public basic research funding
has an economically and statistically significant indirect effect on pharmaceutical
innovation by inducing industry R&D investment.

31

Like public basic research, FDA regulation can have a direct effect on
pharmaceutical innovation and an indirect effect on innovation through its impact
on industry R&D expenditure. In studying each of these effects much of the
previous economic research has used average NDA review times as a proxy to
measure the impact of regulatory stringency on pharmaceutical innovation
(Grabowski and Thomas (1978), Wiggins (1981 ,1983) and Jensen (1987)).
Using lags of review times to account for changing expectations concerning FDA
review criteria, these studies find that regulation has a negative and significant
impact on pharmaceutical innovation up to five years prior to NDA submission.
However, these studies also focused on FDA review data covering the 19603
and 19703. In light of the interview results that claim the industry is adjusting to
FDA regulation earlier in the innovative process, one would expect that the
direct impact of FDA review times on innovation would show up in lags greater
than five. To account for this, the cumulative direct impact of FDA regulatory
stringency is allow to extent nine years prior to an approved NDA.

The indirect effect of regulation is more problematic. Some studies, Wiggins
(1981,1983) and Jensen (1987), find NDA review times reduce industry R&D
investment while a more recent study, Ward and Dranove (1995), finds that NDA
review times increase industry R&D investment. In the former studies, longer
FDA review is interpreted as reducing the number of candidate compounds
entering the innovative process while, in the later study, longer FDA review
times are interpreted as leading to increased expenditure on those projects

already in process. Aside from covering different sample periods, there is one

32

notable methodological difference: Instead of using observed average review
times, Ward and Dranove (1995) use the predicted review times from a first
stage regression. By reducing the random variation in the review times
measure, it seems the Ward and Dranove approach should lead to a more
precise estimate rather than a change in the sign of the estimate. Nevertheless,

each of these findings is theoretically possible.

2.5 The Captopril Example

There is a growing case study literature describing the successful interaction
between public biomedical research and private industry R&D (see Maxwell et
al. (1990), Gambardella (1992, 1995), Henderson (1994), and Henderson and
Cockbum (1996, 1997). Although each story is unique in its details, there are
two major themes that emerge. First, public research knowledge develops or
matures into a body of knowledge that can be extended and utilized by the
pharmaceutical industry. Second, private industry is in a position to transform
public basic research into new compounds. The story of captopril illustrates
these themes.

Captopril is an important drug for regulating blood pressure. Figure 2.3
illustrates that captopril inhibits the conversion of angiotensin I to angiotensin II.
It is referred to as an ACE inhibitor because it blocks the conversion to
angiotensin II and thereby prevents high blood pressure. The scientists at

Squibb synthesized captopril in the early 19703. The patent was granted in

33

cafiaxm 23390 - mN ocswi

MOEMEIZH m0<

 

 

 

238$

coo—m All All 2555 $53250 Al
.3:

= _
52.3635. :_m:86_w=<

 

 

 

 

 

 

34

1977, which is the same date as the publication of their research underlying its
discovery. In 1981, the FDA approved captopril for market sales.

Notwithstanding the creative work of the Squibb scientists, their discovery is
built on two lines of public research. The first line involved the identification and
description of the renin-angiotensin system. While this public research dates
back to at least 1934, it was the late 19503 when the key scientific papers which
identified angiotensin I and angiotensin II were published. The second line of
important public research originated in Brazil. Research into the cause of death
from snake venom identified a natural substance which acts on its victim by
fatally lowering blood pressure. In 1965, it was shown that this natural
substance blocks the conversion of angiotensin I to angiotensin II. Thus, armed
with this public knowledge, the scientists as Squibb were able to discover and
develop a drug for human consumption that Iovvers blood pressure.

The discovery of captopril illustrates several points. First, the advance of
public basic research knowledge can open up new avenues to therapeutic
outcomes. This knowledge can be used directly in the discovery of new
compounds. Second, basic research is cumulative and reaches some maturity
point as which private firms can usefully apply their skills. There is a lag
between this “critical point” and the ultimate FDA approval of a new compound.
If one were to describe the study published in 1965 as the critical point, then the
lag between the key public research discoveries and FDA approval of captopril
would be sixteen years. Third, it was the combination of public basic research

and private industry R&D that created this new therapeutic compound.

Chapter 3

The Analytical Framework, Data, and Estimation Method

3.1 The Analytical Framework

This analysis uses the production framework to model pharmaceutical
technology and new drug innovation. This framework has been used to study
the impact of research and development on innovation by Pakes and Griliches
( 1984). Their “knowledge production function” (KPF) model has also been used
to study knowledge spillovers from academic research to corporate patents
(Jaffe (1989)) and manufacturing innovation (Acs et al.(1991)). While these
models take patent counts as their metric of innovation, this analysis uses a
measure of important patents as its metric of innovation. Important patents in
the pharmaceutical industry are defined as the number of approved new
Chemical entities. Recent research by Henderson and Cockbum (1996) use the
KPF to analyze a measure of important pharmaceutical patents.

The KPF is particularly well suited for an investigation of pharmaceutical
innovation due to the industry’s research intensive Character. It focuses on the
relationship between research expenditures and innovation. One limitation of
this framework, however, is its reduced-form approach. Many of the complex
interactions and details of the pharmaceutical innovative process are not
explicitly modeled. The lack of sufficiently detailed public data prevents the

specification of a more sophisticated structural model. Nevertheless, the KPF

35

36

provides a useful framework to extend our understanding of pharmaceutical
innovation and to provide evidence on the impact of government funded
research.

The KPF is used to estimate the direction and magnitude of
knowledge spillovers or transfer from public medical research to pharmaceutical
innovative output. This type of productivity relationship is not embodied in a
purchased product but instead in the free communication of useful research

results. The model utilizes the “therapeutic class” as its map between federally

funded research and industry research and output.1 The therapeutic class is a
grouping of compounds based on their treatment indication and has been used
by the Pharmaceutical Research and Manufacturers Association to group
industry R&D since the early 19603. For example, cancer basic research is
grouped and related to industry cancer drug research and not to industry
cardiovascular drug research. To capture the differences across therapeutic

Classes, an individual intercept (indexed by the subscript i) is specified for each

, 2
of seven therapeutrc classes.

 

1 Although federally funded basic research can be usefully grouped into
therapeutic classes, these classes are not always mutually independent.
Occasionally, basic research from one class will feed the discovery of
compounds in another class.

2 The classes considered in this study are: endocrine/neoplasm, central
nervous system, cardiovascular, anti-infective, gastrointestinallgenitourinary,
dermatologic, and respiratory. These classes cover over 80% of the industry
R&D investment.

37

The general form of the direct relationship determining pharmaceutical

innovation is represented as:

(30 Yn=qmbfime

where i represents the medical therapeutic class, t represents time, Yit is a
count of pharmaceutical innovative output, 'it is industry R&D, Bit is the stock of
Public Health Service funded basic research, Rit is FDA product quality
regulation. The empirical specification of the functional form, q, will be
discussed in chapter four.

Based on the qualitative discussion in chapter two, the expected signs of
the partial direct effects (marginal productivities) are:

<11 >0.Q2>0.<13<0

The first relation, q1 > 0, is the expectation that industry R&D increases
pharmaceutical innovation. The second relation, q2 > 0, is the statement of
hypothesis one discussed in chapter two. That is, public basic research
contributes to pharmaceutical innovation by providing new avenues to
therapeutic outcomes and facilitating the chemical screening step of drug
discovery. The third relation, q3 < 0, is consistent with previous research on
FDA regulation that shows increased regulatory stringency reduces

pharmaceutical innovation.

38

In addition to the direct production relationship, basic research and
regulation have an indirect effect acting through their influence .on industry R&D
investment. The pharmaceutical investment decision is modeled using a net
present value (NPV) framework. While the details of this model are discussed in

Chapter five, the structural investment relation can be summarized as follows:

(3.2) 'it = g (Bib Rit: Expected Demand“, Cost of Capitalt)

where subscripts i and t represent therapeutic class and time, respectively; Bit
and Rit are defined as in equation (3.1); Expected Demandit are other
explanatory variables affecting the rate of return to pharmaceutical investment
by influencing the expected revenue stream, and the cost of capital variable(s)
account for the influence of the opportunity cost of capital. The NPV model and
each of the explanatory variables will be discussed fully in chapter five.

Here, the expected partial effects are:

91 >0.92<0.93>0.94<>0

The first relation, 91 > 0, corresponds to hypothesis four of chapter two. This is
the inducement effect of public basic research on pharmaceutical R&D. The
second relation, 92 < 0, follows previous research on FDA regulation that shows
increased regulatory stringency reduces pharmaceutical investment. The third
relation, 93 >0, states that increased expected demand for pharmaceutical
compounds in a therapeutic class will lead to greater industry R&D investment.

The final relation, 94 <> 0, indicates that the cost of capital can either increase

39

or decrease investment. Generally speaking, a fall in the cost of capital implies
greater investment since more projects become profitable.

The estimate of the total effect of basic research is calculated by
estimating the direct effect (chapter four) and the indirect effect (chapter five)
separately and combining the results to calculate the total effect (chapter six).
The overall or total effects of government funded basic research and FDA

regulation on pharmaceutical innovation are given by:

(int/dBrt)=q2+91*91

( int / dRit ) = C13 + (:2 *92

where the subscripts represent partial derivatives of the indicated functions.
Combining the expected signs of the partial effects discussed above implies the
following signs on the overall impact of public basic research and FDA review

times:

(int/dBit)>0

(dYIt/dRit)<0

3.11 Issues Regarding Possible Endogeneity

The theoretical production function model embodied in equation (3.1)

defines the direction of causation as running from the inputs to outputs. This

40

means that industry R&D, government funded basic research and FDA
regulation all combine to cause the flow of pharmaceutical innovative output, Yit-
Yet it is reasonable to believe that innovative output may spur or cause changes
in any three of the defined inputs. For example, a newly approved drug by one
firm may cause other firms to undertake related research and, hence, stimulate
industry R&D. For basic research, a new FDA approved drug may spur greater
academic interest in exploring some biological process and, consequently,
cause some academic researchers to write up and submit grant proposals for
federal support. For regulation, a previously approved drug may begin to show
negative long-term side-effects that cause the FDA to review and possibly
change its approval criteria. In all of these cases there is feedback from the
pharmaceutical innovative output to future values of the input variables.
Econometrically, we are saying that the input variables are not strictly
exogenous. Failure of strict exogeneity can be the cause of inconsistent
estimators (see Wooldridge( 1994) and Blundell et al. (1995)) . However, this
analysis does not impose strict exogeneity. The pooled estimation technique
used to calculate the direct effect allows for feedback of an arbitrary nature from
current realizations of the dependent variable to future values of the explanatory
variables.

The direction of causation postulated between industry R&D investment
and government funded basic research in equation (3.2) has been subject to
some debate among researchers. The current research literature on spillovers

points out the possibility that a bias may arise due to Changes in "scientific

41

opportunity." That is, researchers have noted that a promising development in a
particular scientific discipline may lead to correlation between industry and
public R&D funding that is not correctly characterized as spillovers. It arises
when research funding agents respond to the same “scientific break-through”
information when making their funding decisions. Further, if scientific
opportunity is a problem , then estimates of the relationship between public and
private funding of research are biased by the endogenous response of funding
agents. One way to conceptualize the scientific opportunity problem is as an
omitted variable bias. In this scenario, equation (3.2) has an omitted “scientific
opportunity” variable and, consequently, the basic research variable is
endogenous. This would lead to bias and inconsistent estimates of the
parameters.

As a practical matter, any bias due to scientific opportunity will always be
a possibility in models of research spillovers. In our context, the issue rests on
three basic questions. First, how fine a line can be drawn between basic
research and applied research? The National Science Foundation definition
provides a broad guide to follow when delineating between the two; however, it
does not allow one to construct mutually exclusive sets of research. To the
extent that public and industry research overlap, the possibility of a scientific
opportunity bias remains. As with the present study, researchers must use great
care when breaking research projects into separate categories.

Second, at what stage of research “maturity” is the promising

development in the scientific discipline? A fundamental assumption in models of

42

R&D capital is that R&D is effective over many periods and has an evolutionary
Character; Ieaming and research are cumulative processes that are best
measured as stocks. One idea underlying the model of equation (3.2) is that
there exists some level of research maturity at which point basic biomedical
research becomes useful to pharmaceutical drug discovery. A scientific
opportunity bias is less likely when the academic researcher and the industry
researcher focus on different areas of research.

Third, what proportion of pharmaceutical research effort is devoted solely
to the exploration of basic science? Although an exact figure is not publicly
available, pharmaceutical firms do conduct basic research. By all accounts it is
a small but growing element of the industry’s research strategy. Again, to the
extent that research overlaps, a scientific opportunity bias is possible. Although
it is expected that any bias related to scientific opportunity would be minimal in
this analysis, chapter five explores the endogeneity issue between industry R&D

and public basic research.

3.2 Data Sources and Construction

The well known difficulties of measuring the spillover or transfer of
knowledge across sectors has made the data collection and preparation a very
important aspect of this project. The three primary sources of data, the Public
Health Service, Food and Drug Administration, and the Pharmaceutical

Research and Manufacturers Association, are matched by year and therapeutic

43

Class.3 Together these sources represent the most recent and comprehensive

public data available to address the hypotheses in this paper.

3.21 Productivity Measure

The measure of innovative output, Yits is a count of new chemical entities
(NCES) approved by the FDA. These are defined as new molecular compounds
that have never before been tested or used in humans. The FDA supplied the
data on approved NCES for the years 1960-1994. As used here, the counts of
NCES exclude diagnostic and certain biological agents, new dosage
formulations, surgical and other materials such as contact lens and devices.

To construct the counts of NCEs, the FDA data were grouped by year of
approval and therapeutic class. Although the year of approval was supplied by
the FDA, the compounds had to be assigned to the various therapeutic classes.
This was accomplished by using the clinical pharmacology and treatment
indication descriptions from the Physician’s Desk Reference, Merck Index, and
Martindale The Extra Pharmacopoeia. These sources were also used to

eliminate any compounds not fitting the definition of NCES used in this project.

 

3 Each measure used in this analysis has been grouped according to the
therapeutic classification scheme used by the US. Department of Commerce
Current Industrial Reports, Pharmaceutical Preparations except Biologicals.

44

Both the tabulation of counted NCE and some descriptive statistics by
therapeutic class are presented in Table 3.1 and Table 3.2, respectively. Figure
3.1 graphs the number of counted NCES against time for the cardiovascular,
anti-infective and gastrolgenito-urinary classes.

Although NCES are only one possible productivity measure of
pharmaceutical innovation, DiMasi et al. identify NCES as the “most
therapeutically and economically significant” (DiMasi et al. (1991), p. 108). Of
course, patents are the traditional indicator of economically productive
knowledge in KPF types of studies. It should be kept in mind that even though
simple patent counts are an alternative, approved NCEs are a measure of
economically important patents because they have proven therapeutic value.
“Important” patent measures address one of the key criticisms of patent
measures as an indicator of innovative output—the wide variation in the
economic value of individual patents (Griliches (1984)). Trajtenberg (1990)

provides an informative discussion on this issue.

3.22 Industry R&D Measure

The public data on pharmaceutical industry R&D expenditure, 'it. come
from the Pharmaceutical Research and Manufacturers Association’s Annual
Survey Report. Domestic US. company-financed R&D expenditures for human-
use (dosage form) ethical drugs were obtained for the period 1963-1994. Table

3.3 presents summary statistics for industry R&D for 1963-1994 while Figure 3.2

45

Table 3.1 - Tabulation of Counted NCEs

 

 

 

 

 

 

 

 

 

 

Year Total Number of Number of Number of Other Number of]
Approved Material and Diagnositc Repeat Excluded Counted
NCES Device NCES NCEs DosagLe NCES NCES NCES
1964 24 1 0 0 1 22
1965 12 0 0 0 0 12
1966 18 0 0 4 1 13
1967 22 0 0 3 0 19
1968 9 0 1 0 0 8
1969 17 0 2 1 0 14
1970 20 0 0 0 0 20
1971 16 2 1 0 0 13
1972 11 0 2 1 0 8
1973 28 1 11 2 0 14
1974 36 3 3 0 0 30
1975 20 0 1 5 0 14
1976 26 1 9 0 0 16
1977 21 3 1 0 0 17
1978 20 2 2 0 0 16
1979 14 1 0 0 0 13
1980 12 0 0 0 0 12
1981 25 0 3 0 0 22
1982 28 0 5 0 0 23
1983 14 0 1 0 0 13
1984 20 0 1 0 0 19
1985 28 0 3 0 0 25
1986 19 0 2 0 0 17
1987 20 0 3 0 0 17
1988 19 0 3 0 0 16
1989 22 0 2 0 0 20
1990 22 0 4 0 0 18
1991 30 0 0 0 0 30
1992 32 0 3 5 0 24
1993 24 0 3 0 0 21
1994 22 0 4 0 0 18
—Totals 651 14 70 21 2 544

 

 

46

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

m9: dddm ddd mod? on 099. Ed? on d 2mm; ..(Bommflo =<
Ed cod ddd Ed ddN ood Ed dod dod Eodgaom
and cod ddd dud 2: Odd omd cod dod 03229500
Ed... cod cod 9N Odd cod Eé 8.? Odd Bactsogcow>«52.252590
and cod cod and cod cod ddN ooE ddd $2685-22
Nod ooE ddd mod 09E cod v ...F 006 ood 53035.0th
56 cod do... and cod dd... 3.0 ood ddN Sufism m30202 .9200
EEd cod ddé d w d CON do._. end cod do... Emu—moozEctooucm

2<w_2 x32 2.5. Z<m=2 x32 2:2 Z<w_2 x32 2:)—
wad 739 :3 TdEdp End ..-vodw mam—o canon—Sock

 

 

 

 

820 0.5822: .3 802 8:58 8 8.8.88 2.5.880 - mm «.8»

 

47

mmoz umano baccaégcogozmmw can wEomE.-_E< ..m_:omm>o_2mo - fin 2:9“.

vamp Nmm— camp name wmmw vwmw Name mum? ohm? ticsow N59. Ono?

 » a

 

 

 

 

 

 

48

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

$00 N 00.0nvm $N— .0 0v.000v $v0 .0 000000 $ 5.0 00 .00: 00306 =<
$00.0 0 .00 _. $00.0 NO0N0 $v0.0 00. Fm? $0NNr N000 0.900300”.
$00.0 00.05 $000? 00;: $000 N700 $00.0? 0N; 20.299500
$3 .0 0v.00F $00 .N 00.00N $00K 00 .0: $00 .0 N550 b0c_59_c00\_0:6000505000
$00.0 00 dov $NN .0 50 NNO $00.0 V0 00¢ $~.0 N 00.0NN 0020000: 12.2
$00 .0 r0. :0 $004. 0.1.: F $0 F Ne 0v.00v $vnd Nth .m_:omm>o_EmQ
$ 3.0 00.000 $00.0w 00.0v0 $00N 0v.000 $NNO 00.00N E00050 mu30202 _0=c00
$00K 0NON0 $ ..NNP 00 .0 For $NON 309‘ $00 .v 00. SN Emm_0002\0ctoo.ucm
000m 539.0 00m 0002 0.0m 5390 00m c005. 0001 530.0 0.0m c005. 000m 526.0 0.01 c005.
v00 7009 .50 7000? 000 Tmmmw 0N0 7000—. 0006 033000.02»

 

A0 000? 00 205E: 0.0m b00305 503308.5ch - 0.0 030...

 

49

 

A0 002 00 mco==EV 0.0m b00305 b0c_5.o:c0m¥ozmmmu 0cm 028005-006. 0030092200 - N0 050E

.ll.l .l «l .I: sll.v.!l l Illl l .. ll|.ll ll. l..l

 

.50.. 000? 000' 50— 000— 000? 000—. soar @00— 000. v8? 000' N02 30— 0009 0kg 0kg Rae 0&2 chap fla—

T!.|1l . .lTIIllTI- -.JT + l. 0 f .r llll+ll f . . T [r

 

 

 

i 4 l 8.0
I I I I I I I
I I I I I I I I I . 8.8“
I I I . i . \
l‘

 

Dex-0+ \ll . .7 880
.51! + . . .
, 0.2.0ch \ .. 88..

 

 

 

 

50

graphs industry R&D against time for the cardiovascular, anti-infective and
gastrolgenito-urinary therapeutic classes.

Perhaps the biggest problem with the PhRMA data is its level of
aggregation. Firms are aggregated into industry figures to protect the strategic
position of individual firms. Moreover, the R&D investment totals are a
composite measure of industry resources used in drug discovery and
development (the cost of manufacturing appears to be in these figures as well).
The PhRMA definition of R&D investment is as follows:

The total cost incurred for all pharmaceutical research and development

activity, including cost of salaries, other direct costs, service, routine

supplies, and supporting costs, plus a fair share of overhead

(administration, depreciation, space charges, etc.). Cost of drugs or

medical research and development conducted on grant or contract for

other companies are excluded. Conversely, total outlays for all research
and development work contracted to others (manufacturers, independent

research laboratories, academic institutions, etc.) are included.

(PhRMA, Annual Survey Report)

Consequently, due to the nature of the industry data, nothing can be said about
the relative productivity of individual components of industry R&D expenditure.
Therapeutic class totals were computable for most years in the data.
Although yearly totals were always available, some therapeutic class totals had
to be imputed (this amounted to five individual years of data). All figures include
expenditure on research failures as well as expenditures in both discovery and
development. Further, these figures were adjusted for inflation using the
Biomedical Research and Development Price Index (BRDPI) supplied by the

National institutes of Health.

51

3.23 Measure of Government Funded Basic Research

The annual federally funded basic research flows were constructed from a
comprehensive data set which includes all extramural grant and contract awards
by the US. Public Health Service since l955. For each grant and contract
award, the data includes: the title, the identification number (activity code,
institute code, and grant or contract number), the fiscal year of award, the award
amount, and the scientific review group that recommended its approval. The
individual grant level data allow the construction of annual flow and stock
measures of PHS research awards by therapeutic class. The final set of
research projects, called “core” research, consists of all those scientific studies
that represent basic research relevant to the pharmaceutical industry. Of the
PHS research grants, basic research excludes non-medical, instrumentation and
clinical (human) studies.4

Beginning with the total set of grants and awards for an individual year,
the annual basic research flow series are constructed using several data filters.
Four main filter levels are defined and used. The first eliminates awards based
on activity code. Generally speaking, this filter purged activities such as
training, education, construction, demonstration, and institutional block grants

from the data. The second filter eliminates awards based on institute code.

 

4 Industry scientists helped with the data construction process due to the
complexity of medical scientific terminology.

52

There are several institutes or divisions that contribute nothing or only a
negligible amount to core research. Examples of eliminated institutes include
the National Institute on Dental Research, the National Institute on
Environmental Health Sciences, the National Library of Medicine, the Food and
Drug Administration, and several others. The third filter eliminates scientific
review groups that consider research outside the definition of core basic
research. These groups are reviewed in every year and the filter is modified to
be year specific. (This was necessary due to the splitting, adding, and
discontinuance of scientific review groups over time.) The fourth filter level,
which is really a group of filters, analyze awards based on keywords contained
in the title of the grant or contract. This group consisted of seven keyword
“inclusion” filters that help identify awards belonging to specific therapeutic
classes as well as an “exclusion” filter designed to further sift out inappropriate
awards missed in the earlier levels. The primary sources of information in the
construction of the class filters are the Department of Commerce Current
Industrial Reports and medicine’s scientific vocabulary. Table 3.4 shows the
mean expenditure and growth rate by therapeutic class over the sample period
(all figures are in millions of 1986 dollars). Figure 3.3 plots PHS basic research
flows against time for the cardiovascular, anti-infective and gastrolgenito-urinary
classes. Appendix A contains an explanation and copies of worksheets and filter
programs used in the data construction effort.

The filter levels separate the total universe of PHS grants into seven

therapeutic classes of core research and a group of all other research. For a

53

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

$30 ~38 $3... 5.5: $m§ 8.8m $6.8 8.8.. 3330 __<
$8.8 $2 $8... $8 $0: 2.8 $5.8 8... 32938”.
$3.0 8.0 $3... E.» $Bm- go $2.3m one 2.0.23.8
$23 8.3 $93 8...: $2.0 Rem? $2.8 8.8 52.500828..mscafimo
$2.: 8.2: $30 «$8 $05 $0.3 $8.8 No.2 32.8.5.2...
$80 2.5. $30. 30.0% $2 .N 8.0? $8- 8.3 5336.28
$3... 3%. $30 «$8 $8.0 8am: $m§~ 3.5 505.6 «.8202 .9050
$98 5 .mmm $m . .m 58m $03 2 .mmm $2.. E 8.9: 58.082.05.820
Sum 5390 0.8m 322 2mm 5390 28m :85. 2.2. £2.90 0.3m :85. 23. 5590 05mm :82
$2.82 32.32 mg 0-82 82-82 $8.0 c.5399:

 

A0 000? 00 2.05:5 20.00030 2000 60.00890 01““. - v0 0300.

 

 

.0 000. mco_.__E. 20.0000”. 0.000 0.5 00556....0990000 0cm 0>_60.c_-_.c< 003803.200 - 0.0 050.“.

000w 000' ..00. 0N0? Row 0N0? 0&0? 2.0. 000? N00? 000.. 000.. «00? 000? N00. 000..

 

 

.l-|T|:T| .l |+ll.+l. ..l-i+l ll-Tli .l-T- -+- 1+ ll ll+ -..-f - 0 lfl ..T: ITllTlllii » .« + l T _ 080
II
I\.
U .
. L-qu.00
. +\ .
39.0.6: . . . . . . \ .088.
+ I I
.5 . . . . . 8‘
2050+ . . . . . . . .
. l‘lr . $83.
I .

, 95.80

 

.9680

 

55

class total, the awards are summed across all grants. This is a simple method
that does not attempt to apply different weights to individual research projects
based on “importance” or scientific discipline.

Having completed the construction of the annual basic research series
the next step is to construct the cumulative stock measure, Bit. used in the
analysis. The government funded basic research stock is created using the
standard perpetual inventory model described by Hall et al. (1988). For each

therapeutic class, the stock of basic research is given by:

(3.3) Bit = (Annual Flow)“ + ( 1-8 )*( Bit-1 )

where Bit is the stock of PHS funded basic research in class i and year t,
(Annual F low)" is the annual flow of PHS funded basic research in class i and
year t, and 8 is the depreciation rate of “knowledge capital”. With 8 constant
over time we are assuming a geometric form of depreciation. In this case, a
given year’s flow of basic research losses its “productive capacity" at a constant
percentage rate each year. Using a 15% rate of depreciation we see a 48%
decline in the productivity of basic research over four years. In the literature,
Hall et al. (1988) assume a 15% depreciation rate of R&D capital and
Henderson and Cockbum (1996) use a 20% rate. The sensitivity of the
estimated coefficient to changes in the rate of depreciation is explored in chapter
four. The geometric form has the additional property that assets never “retire.”

Unlike the case of physical capital, this makes sense when applied to research

56

capital (it avoids the issue of having to estimate the useful life of research
results—which we expect are long anyway).

To implement the perpetual inventory model the stock measure must begin
at a benchmark. The benchmark for the Bit variable is the year 1944. This is
the year Congress passed the Public Health Service Act authorizing the surgeon
general to award research grants. Using the 1955-1985 sample growth rate by
therapeutic class, the annual flow series were projected back to 1944. These
nominal flows were also adjusted for inflation using BRDPl. Finally, equation

(3.3) is used to construct the stock series.

3.24 Measure of FDA Regulatory Stringency

Since the passage of the 1962 Kefauver-Harris amendments to the
Federal Food, Drug and Cosmetic Act of 1934 economists have been interested
in the relationship between regulation and pharmaceutical innovation. The two
most popular approaches are: first, to use a proxy variable for FDA regulation
(Ward and Dranove (1995) and Wiggins (1979, 1981)), and second, to use the
international residual from comparing the US. experience with that of the United
Kingdom. The studies in the second group assume that US. firms and UK.
firms differ only in their regulatory environment (Grabowski et al. (1978)) or that
they differ in their regulatory environment after controlling for firm size (Thomas
(1990)). They ignore the possibility of that public knowledge can affect

pharmaceutical innovation. While the technical determinants of pharmaceutical

57

innovation are just beginning to be understood, it is clear that the public stock of
basic biomedical research is one such factor. The contention that UK.
pharmaceutical firms have the same access as US. firms to US. Government
funded biomedical research is unsubstantiated. In fact, there is some interesting
research on the role of geography in mediating knowledge spillovers (see Jaffe
(1989) and Trajtenberg et al. (1993)). Moreover, Gambardella (1992) presents
some initial evidence that public research results may be utilized with different
degrees of effectiveness depending on firm human capital.

This paper follows the first approach by specifying a proxy variable for
FDA regulation. In particular, it uses the same method of measuring regulatory
stringency as Wiggins (1981). In his method, firms are assumed to form
expectations of regulatory stringency based on current and past observed
delays between the submission of a new drug application and its final FDA
approval. These average review times are calculated by year and therapeutic
class. The therapeutic class distinction is appropriate since it corresponds fairly
well with the review structure of the FDA.

Using the new chemical entity data supplied by the FDA the review times
are calculated for each approved NDA in months. The appropriate therapeutic
class for each compound was determined in the process of constructing the NCE
productivity measure. For those years in which no approved NDA is observed in
a particular therapeutic class it is assumed that firms adjusted their expectations
by increasing or decreasing the last observation in that class by the change in

the overall average review time across all classes. The resulting regulatory

58

measures capture the cost associated with NDA review segment of FDA
regulation. Table 3.5 shows the minimum, maximum and mean delay times by
therapeutic class over the period 1973-1994. Figure 3.4 plots average FDA
delay times by year for the cardiovascular, anti-infective and gastrolgenito-

urinary classes.

3.25 Measures of Pharmaceutical Demand and Medical Need

The expected demand for a therapeutic compound in an important
component of the investment decision for pharmaceutical firms. It is the firm’s
forecast of actual sales over the life cycle of the therapeutic compound. Informal
interviews reveal that many firms purchase expensive market information that is
specific to individual market segments within therapeutic classes. Purchased
market data is combined with firm specific information on product characteristics
and sales infrastructure to estimate the revenue stream.

The main variable used in this analysis to proxy for expected demand is
actual industry sales by therapeutic class and year. Wiggins (1983) and
Henderson and Cockbum (1996) also use industry sales to proxy for demand.
Sales by therapeutic class convey information on current market size, including
demand from chronic medical conditions, as well as the distribution of industry
sales resources (human and physical). This makes actual sales an important
variable for conveying information about the market. Human-use, dosage-fonn

ethical pharmaceutical industry sales figures broken down by therapeutic class

59

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0500 {.09 00.00 00.2. 00.00 00.00. 00.00 (.00.. R00 00.0..ng
0.10 00.00 050 00.vm 00. 3 05.0 0 F NV 00.00 0....0F 0.0220650
50.00 0509 00.0 00.00 00.9. 00.0 00. ’0 0500— 00.0? 00c..aogcoo\.0c.mc2c.ofi0o
9.00 00.03 00.0 00.8 00.9 8.0 00.00 00.03 00.0. wo>.60.c.-.E<
5.0V 00.0NF 00.0 0 00.00 0N.00 00. E 00.00 00.00? 00.0 F 030090.200
00.0w 0». F I. 00.0 000v 004K 00.0 00.00 05 P : 00.0? 50.000 030202 .0950
00.00 00.00 00.0 V 05.00 00.00 00. : N000 00. P0 00.0.. Efiaoozxmcgoucw
:02). 838.585. 53.55.). C005. 535.58.). EJEE...‘ c022 E:E.x0.2 5255.2

«$1.8. :8 EB. km 2.8. 23.0 0.5892h

 

 

 

 

3585. 85: 58 $0.38”. <9. 8 8.0.50 - mm 033

 

. _. 33?

60

 

v00.

.259... 8E.» 3.5. <9. races-ogcuoasso .25 9.6%....2 53892.23 - ...m 2%:

 

 

 

 

82 «8. 52 82 82 82 $2 82 82 3.: 82 4.2: .2: 82 22 as. E... 2.: as. :2 2.:
+ 0 . T . 4. l4 0 0 LI 0' . .r . Till+l.. JT-- Fl . T o
, . l ..8
5" I. . . t A.
. . 4‘ . 1..
. ..8
agcll
.51.. +8
3900+
-8.
.8.

 

 

 

 

 

 

 

61

are available from PhRMA.5 Table 3.6 shows the mean and growth rate of
industry sales from 1971 to 1985. Figure 3.5 graphs the series for the
cardiovascular, anti-infective and gastrolgenito—urinary classes.

In addition to industry sales, disease incidence and “severity” measures
are developed. Data on the incidence and “severity” of disease are supplied by
the National Discharge Survey. This survey covers non-Federal short-stay
hospitals and is collected from a national sample of hospital records of
discharged inpatients. Although the survey dates back to 1965, a fairly
consistent time series is only available since 1971. It should be noted, however,
that a new classification scheme was introduced in 1979. The lntemational
Classification of Diseases, 9th Revision, Clinical Modification (lCD-9-CM)
replaced the earlier edition. This change caused a small shift in some of the
therapeutic class totals.

A simple count of first-listed diagnoses, which are equivalent to the
number of discharges, are grouped by therapeutic class and year. These counts
represent the incidence of disease requiring short-stay hospital visits for the
civilian non-institutionalized US. population. “Severity” is measured by a count
of the number of hospital inpatient days. The basic notion is that more severe
illness requires longer hospital stays. Because the hospital sample is inflated to

estimates for the civilian non-institutionalized US. population, each of these

 

5 Industry sales to Federal hospitals were purged from the annual sales totals
prior to computing therapeutic class totals. As an empirical matter, this
correction had no effect on the results discussed in chapter five.

62

counts cover most of the US. population. The measures are less accurate for
the older segment of the population since many senior citizens move to nursing
homes. Tables 3.7 and 3.8 present the mean and growth rate of disease
incidence and severity by class for 1971-1985, respectively. Figures 3.6 and 3.7
plot disease incidence and severity against time for the cardiovascular, anti-
infective and genito-urinary therapeutic classes.

Before turning to the estimation methodology for the analysis, it is
important to note a few limitations with these measures. First, disease incidence
is measured not disease prevalence. Since incidence measures the number of
new cases over a period of time, it is not equivalent to prevalence, which
measures the number of disease cases at a point in time. Although these
concepts differ, incidence may be a good substitute for disease prevalence.
Second, measuring severity by the length of hospital stays is fairly crude. In
fact, the length of stay may be more related to the recovery period of illness or

surgery.

3.26 Time Effects and Therapeutic Class Effects

Time effects have been treated inconsistently in the current research on
pharmaceutical innovation. Some researchers, such as Ward and Dranove
(1995), do not include a time trend in their model while other researchers, such
as Henderson and Cockbum (1996) and Thomas (1990), do not specify a full set

of year dummies in their analyses. Dating back to Bailey’s analysis, all

63

a 82 a 20.55. 8.8 $89.. 225-2...8358 2.... 3.835%... 5.28:2an - on 2:9“.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

08' v8. 08’ map .80 08* 050. 2.0— K? 0B0. 050. 3.0. 9.00 «mg {.0—
. + . +111.-.- -- 1T1 .....-T.. ..-l .1.--- .1.- -.-.+ 1..-1.-+...1.---. .11 1+ I- +111 8.0
104.1019. 8.8m
=c<|+lu
0.28141-
\.\ \
+\.\
33... 00.8%! 8820 __<
$80 8.43.. 028.800
:3... 408... 0323.8
#80 2.3.... aoctaogcao=mc_mco.c_ofia0
$3.... 8.080 $832.35.
$20 no.3.“ 5383.28
#8.» a .08; 522.0 3952 .258
$3... 2.83 5308235685
30¢ 539.0 00.0% 0002
new 7:2 80.0 25322:

 

3 82 .o 25%... 8.8 .382. 8.5835,... - a.» 0.8»

 

$90235 .933... Emacs-2.50.2300 2.0 03.83:.-.05 003038.200 - 0.0 2:9“.

l l ll lll-l lll ll l l lll llll ll Ill 111 .l. ll - :-ll .

 

 

 

 

 

1(- l -l111|. l

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mom. 4mm. mom. mom. Pom. one. mum. aka. 55m. one. mumr 40m. mum. «km. rum.
Ti 4.- T- L. .-l-lT -1111T1l- T- 1-- + 1-1-1+ l 11.1-1-1. ---1. 1+1- .1..lT-lll+ ill... 0
.. 88
.1 p p p T p r
+1 l . . . L.lll|+/..lll.T T . . . +l\\-. 0000
20:01?

El _I 11 DOOM
LE . . . . . . . . . . q . ._ ..ss
. . . . . + 88

. . -- 88

-- 82
$80 02.8 830.0 .2

$20 83 322.8%

$3...- omw gagsaémo

$00.0- 000.0 02.52.50:053252.000
$00.0 50. F m0>..om.c.-=c<

08; 08;. 0.333.200

350.0 V006 805.6 26202 .2230

$20 mam...” Emmaoozwctooucm

20m 5320 :022
000.32.? 30.0 0.52.0.2.»

 

 

 

 

$2532.. :. 00909.86 .5335 303.0 .0 8:00.05 - 50 033

65

 

90o E00000. 00000.0 300.560.503.030 0:0 002000.505 003305.200 - N0 050.”.

wear 3a.. 08.. Nmmw 5a.. 89 mm? 059 KB. mum? mumr 3.2 009. N59. Km?
1 -1. 11 13.! .11-T-1- .511. 0

TI TI- TTI. -1- T11 501.-

T+111 - T- .1 +1..

1 1| |11114 i

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

- 0080
$00.0 5 F00:— 00000.0 ..<
$00. _. 3.0.0— 520.300”.
$3.0. F0~.~ oaoT§0E00o
$3M. 000.~N 30000005002055.0500
$550 30.: 0020000505..
$0 to 000.00 050003.200
$00.. 03.00 E00000 000202 .00000
$00.0- RN00 800.802.05.805

200 5320 :85.
000 7 :0 .. 000.0 0.50qu

 

.0283... a. 0.2. 3085 00000.0 .o .2550. - 0.0 030»

 

66

indications are that time has an important and somewhat unexplainable impact
on pharmaceutical innovation. In light of this fact, the current analysis specifies
individual year dummy variables. Allowing each year to have its own intercept is
the most general way to account for variation over time due to inflation, industry
shocks, and other unobserved changes.

As discussed above, the therapeutic classes provide the mapping of
research knowledge used in this paper. Each of these classes encompasses the
research from a cross-section of medical disciplines. Depending on the
distribution of industry research investment as well as scientific and managerial
skill, one class may feed into drug discovery more easily than another. For this
reason, it is important to include some control for therapeutic class effects. This

has been done using fixed-effect dummy variables for each class.

3.3 The Estimation Technique

3.31 The Direct Effect

Since approved NCEs only take on non-negative integer values, a discrete
dependent variable model is appropriate. In this analysis the conditional
expectation of approved NCEs, E[NCEit| Xit], is assumed to be generated by a
Poisson process for each therapeutic class, i, and each year, t. The Poisson
distribution has been the workhorse of the empirical literature studying

innovation counts. Due to advances in the econometrics literature on count

67

variable estimation, the actual conditional distribution of approved NCEs is not
restricted to the Poisson form. In particular, the model can be estimated without
imposing any restriction on the conditional variance of approved NCEs. The
robustness properties of the estimation method are discussed below.

The Poisson parameter, A“, is represented as a exponential function of all

explanatory variables, X, and the parameters, b, as follows:

(34) EINCEitI Xit] = 711t= exp(Xitb)

where i indexes therapeutic class, t indexes time, an X are the explanatory
variables.

There are two important time series characteristics that must be considered
when estimating models for panel data. The first is the strict exogeneity
assumption discussed in section 3.11. Once again, the estimation technique
used in this analysis does not impose this assumption. Second, dynamic models
that do not include lagged dependent variables are not necessarily “dynamically
complete.” The dynamic completeness assumption says that the model

specifies the lag structure correctly so that further lags add no new information:

(3.5) E(NCEitI xitvl’it-1s"it-1v--) = E(NCEit| xitvxit-1:---:Xit-k)

where Yit—1 are lagged dependent variables, xit are the explanatory variables

and, possibly, lags of other variables. Since there is no guarantee that dynamic

68

completeness holds, serial correlation is not ruled out. A Lagrange multiplier
test is used to identify serial correlation in the model that may arise due to
neglected dynamics in the conditional mean. Based on the results of this test,
the heteroscedasticity and serial correlation robust asymptotic standard errors
are calculated following Wooldridge (1991).

A pooled quasi-maximum likelihood estimation technique (QMLE) is used to
estimate equation (3.1). The technique is quasi-likelihood because the specified
likelihood function, a Poisson likelihood function in this case, is not required to
correspond to the actual distribution of the explained variable (NCEs) conditional
on the explanatory variables, X. This leads to certain robustness properties.
For instance, the actual distribution need not have the exact properties of the
Poisson distribution, such as the mean equals the variance. In fact, Gourieroux,
Monfort, Trognon (1984) show that an entire class of quasi-likelihoods in the
linear exponential family (LEF) produce consistent estimates for the parameters
of a correctly specified conditional mean. It is important to note that the Poisson
QMLE produces consistent estimates under any variance assumption and
arbitrary serial correlation in the observations.

Under reasonable assumptions, the Poisson QMLE is relatively efficient.
The asymptotic variance was estimated under three alternative assumptions
about the conditional variance of E[NCEit| xid- The first method is “fully robust”
in that it is valid under any conditional variance (or distributional) assumption
(Wooldridge (1996)). The second method, the Generalized Linear Model (GLM)

assumption, requires that the conditional mean and variance be proportional.

69

This requirement is represented as follows:

(3.6) Var[NCEit| X11] = (:2 E[NCEit| X11] Where 02 > o

The assumption implies that the ratio Var[NCEit| xit] l E[NCEit| xit] is
constant but allows them to be different from each other by some positive
number. The case where 02 > 1 is referred to as overdispersion whereas the
case when 02 < 1 is called underdispersion. If there is overdispersion in the
model, then the GLM standard error will be larger than the MLE and the resulting
test statistic will show lower significance levels.

The third method is the most restrictive because it requires the Poisson

nominal variance assumption to hold.

(37) VarINCEitl Xitl = EINCEitl Xit]

This assumption imposes a standard property of the Poisson distribution, namely
that the first and second moments of the distribution are equal. In most cases,
Poisson models that do not relax the nominal variance assumption find
spuriously high levels of significance.

In the standard cross-sectional problem, calculating the standard errors
under these three alternative conditional variance assumptions would be all that
is needed for correct inference. However, models that incorporate a time

dimension need to account for possible failure of the dynamic completeness

70

assumption. Under the GLM variance assumption, a Lagrange Multiplier statistic
was computed to test for serial correlation. For equation (3.1), the test statistic
is computed by estimating the model under the null hypothesis of dynamic
completeness; calculating weighted residuals and weighted explanatory
variables; running an auxiliary regression of these weighted residuals on the
weighted explanatory variables and lagged weighted residuals. If k lags of
residuals are included in the auxiliary regression, then this is equivalent to
testing for kth order serial correlation. The sample size multiplied by the
uncentered r-squared from this regression is distributed chi-squared with k
degrees of freedom. Further, the significance of the individual coefficients on
the lagged residuals can be tested using the standard t-statistic (for details on
this test see Wooldridge (1996)). If the null hypothesis of dynamic
completeness can not be accepted, the heteroscedasticity/seriaI-correlation

(H/SC) robust standard errors should be used for inference (Wooldridge (1991)).

3.32 The Indirect Effect

The estimation of equation (3.2) is carried out using pooled ordinary least
squares (OLS). As with the pooled Poisson QMLE, the asymptotic properties of
this estimator do not impose the strict exogeneity assumption. This allows
current industry R&D expenditures to influence future values of public basic

research, FDA regulation and pharmaceutical demand.

71

Robust inference requires that the standard errors be adjusted to account
for the possible failure of the dynamic completeness assumption. Once again, a
Lagrange multiplier statistic is used to test for serial correlation. It is the same
procedure as used for the direct effect except the residuals and explanatory
variables are not weighted. If serial correlation is detected, then the
heteroscedasticity/serial correlation (HISC) robust standard errors will be

computed following Wooldridge (1989) and used for inference.

Chapter 4

The Direct Impact of Public Basic Research on Pharmaceutical Innovation

4.1 Overview

This chapter explores the direct determinants of pharmaceutical
innovation. It is postulated that pharmaceutical product innovation is a function
of publicly funded basic research, industry R&D and FDA regulatory stringency.
Previous empirical research has failed to link public basic research with
pharmaceutical innovation. This link, however, is quite important. The extent to
which economic agents can exploit productive external information has
fundamental implications for theories of productivity and growth. Consider, for
instance, that knowledge extemalities are one key to unbounded growth in
endogenous growth theory. This chapter will describe how knowledge
extemalities from publicly funded biomedical research impact pharmaceutical
innovafion.

Strong evidence is found for an economically and statistically significant
impact of public research on pharmaceutical innovation (hypothesis one). Public
research “capital” is intended to mimic the creation of scientific knowledge by
using a cumulative stock form of measurement. The elasticity of the number of
new chemical entities with respect to the stock of public basic research is found

to lie in the range of 2.2 to 2.5. These point estimates are large and imply

72

73

increasing returns to scale in the pharmaceutical innovative process due to
publicly funded basic research.

If public basic research is used to formulate new drug concepts, then the
impact of this research should come early in the drug discovery stage of the
pharmaceutical innovative process (hypothesis two). In fact, the data show that
public research has its most significant impact on pharmaceutical innovation
seventeen years before an approved NCE. This finding implies that an average
of seventeen years elapse between federal award and new product approval.
Because basic research is cumulative and measured as a stock variable, this
result is perhaps best interpreted as a seventeen year lag between the time
basic research reaches a “maturity” point or “critical mass” and the approval of a
new chemical entity.

When one looks at the nature of the research that is supported by the
federal government and private industry, it is clear that these two types of
research are quite distinct. While pharmaceutical firms invest in basic research,
by far the largest portion of their R&D investment is “applied.” The bulk of this
money is spent on testing specific drugs and documenting their findings.
Publicly supported basic research, on the other hand, is more diverse and
general. The analysis finds that both these types of research are important in
the pharmaceutical innovative process. Their distinctive contributions suggest
limited substitution possibilities in the production technology (hypothesis three).
Using a constant elasticity of substitution production function, the elasticity of

substitution (E.) between industry R&D and public basic research is found to be

74

0.29. Since the substitution possibilities between two inputs approach the fixed
proportions Leontief technology as values of E. approach zero, it can be inferred
that public basic and private industry research do not easily substitute for each
other in the pharmaceutical innovative process.

Using the results from Grabowski and Vernon (GV) (1996) regarding the
average discounted stream of revenues for a NCE, the marginal contribution of
the public research stock can be given an explicit present value. The estimated
marginal productivity of federally funded basic research on the number of
approved new chemical entities is 0.07 per $1 million in each therapeutic class.
This marginal product has a present discounted value of $13 million in real 1986
dollars at the time of introduction. In fact, the marginal contribution of public
research is greater that the marginal contribution of private R&D, which has a
marginal product of 0.05 per $1 million. Using the GV data and the seventeen
year lag between research award and product approval, the net present value of
publicly funded basic research at the time of introduction is $5582.368 million in
real 1986 dollars.

Further, the results hold up to robust statistical inference. Using the
heteroscedasticity and serial correlation robust standard errors, the public
research stock variable becomes more statistically significant. In addition to
calculating the robust standard errors, the analysis allows for alternative lags
describing the pharmaceutical innovative process as well as for lagged effects

from FDA regulatory stringency. In both of these cases, using a finite distributed

75

lag is a robust specification since it does not restrict the coefficient values over

time.

4.2 Model Specification

Two empirical specifications of the KPF described by equation (3.1) are
implemented. The first is the Cobb-Douglas. This formulation has been used in
most studies of R&D and productivity. Besides assuming that inputs are strictly
separable, the major limitation of the Cobb-Douglas functional form is that the
elasticity of substitution between inputs is restricted to unity. This restriction
implies that a percentage increase in the relative price of private R&D will induce
the same percentage decrease in the industry’s relative quantity of private R&D
input without any loss of productive output. To the extent that a dollar of public
basic research buys something different than a dollar of private R&D investment,
the Cobb-Douglas elasticity of substitution is probably too restrictive.

To order to explore the potential substitutability between private and
public research, a constant elasticity of substitution (CES) production
formulation is used. As its name implies, the CES requires the elasticity of
substitution between inputs to be constant; however, it is not restricted to unity.
In terms of the estimation, the essential difference is that an additional term,
called the “substitution term,” is added as an explanatory variable. It is perhaps

important to point out that the CES formulation requires a stock measure of

 

76

private R&D in order to avoid problems of collinearity between multiple

substitution terms. The industry level Cobb-Douglas production function is:

(4.1) Yit = e‘“""°(1,t)“‘ , . .. :('it-k)ak(Bit-m)m(Rit)711---»(Rit-h)7hew

where i represents the medical therapeutic class, t represents time, Yit is a

count of pharmaceutical innovative output, 71. and n capture the exogenous
effects of time and therapeutic class on innovation, a1 - ak are the output

elasticities of the industry R&D distributed lag, 'it Jit-ki B1 is the output
elasticity for the stock of Public Health Service funded basic research BM“; and

71 - yh are the output elasticities for the predetermined distributed lag of FDA

regulation (where h= 9), Ritw vRit-h The term, 1.1, is a random error that is

assumed to have a zero mean and constant variance. The distributed lag for the
pharmaceutical innovative process is specified to be length R (where k = 12 and
k = 14 alternatively). The appropriate lag for PHS basic research (m) will be
determined empirically.

The production function is assumed to hold across therapeutic class and
time. The coefficients estimates are long-run industry elasticities for each of the
variables. For instance, the coefficient, B1, tells us the percentage change in the
number of approved NCEs in each therapeutic class for an exogenous change in
the stock of government funded basic research. The industry level results
demonstrate the general importance of PHS funded research in pharmaceutical

innovation. Other industry level estimates concerning the effectiveness of

77

industry R&D dollars and the industry average FDA regulatory delay will also be
of interest. Since the research parameters (01,13) are not restricted to the
constant returns to scale case, these estimates will shed some light on possible
industry-wide increasing returns to scale stemming from the contribution of
public research.

The dynamics of the R&D and regulatory relationships in pharmaceutical
innovation are not known with certainty and have been changing over time. In a
comprehensive analysis on the cost of new drug development, DiMasi et al.
(1991) find that the average time from synthesis to FDA approval is nearly
twelve years for the 1979—1989 period. Moreover, in a study looking at the
relationship between drug importance and time to market, Dranove and Meltzer
(1994) suggest the period from patent application to FDA approval is twelve to
fourteen years. Based on these studies, the analysis allows for both a twelve
year and fourteen year industry lag specification.

Following Grabowski et al. (1978), Wiggins (1981) and Jensen (1987),
regulatory stringency is measured by the average delay between submission of
a new drug application and its approval by the FDA. Both Wiggins (1981) and
Jensen (1987) included their annual stringency measures lagged up to five
periods prior to pharmaceutical innovation. Since it is common for
pharmaceutical firms to anticipate regulatory requirements several years before

submitting new drug applications, the specification of equation (4.1) allows

78

regulatory stringency to have its impact as far back as nine years prior to FDA
approval.1

Chapter two suggests that PHS basic research feeds the creative
conception of new therapeutic compounds with its greatest impact coming in the
early stages of drug discovery. Combined with the existing studies on industry
lag lengths, this implies that firms draw on the pool of available public
knowledge at least thirteen years prior to the drug's approval by the FDA.
Further, because research awards are used to measure the public stock of
knowledge it is necessary to allow for the lag between research award and
research output as well as for the lag between research completion and its
utilization by industry. In the case that two years are enough time for these lags
to work themselves out, the impact of basic research should occur at least fifteen
years prior to an approved NCE. Using this as a starting point, the timing of the

effect is explored in the next section.

4.3 The Timing of the Relationship

The average timing of the impact of PHS basic research is established

using both non-nested and nested regression criteria. The non-nested criteria

 

1 No structure is imposed on the distributed lag of FDA regulatory stringency
variables. As in the case of industry R&D, this is intended to provide a better
econometric control by not restricting the annual coefficients.

 

 

79

are appropriate for evaluating those regressions that include a single lag of PHS
basic research. One advantage of this approach is that it avoids problems of
collinearity between alternative stock variables. The nested regression criteria
are applied to an additional set of regressions that include other lagged stocks
as control variables. Taken together, these approaches determine the empirical
timing of the impact of PHS basic research. Tables 4.1 and 4.2 present the
single PHS lag regression results assuming a twelve and fourteen year industry
R&D distributed lag, respectively. Each basic research stock is assumed to
have a 15% depreciation rate (alternative depreciation rates are explored later in
the chapter). Columns (1) through (6) show the results for the fifteenth through
the twentieth lags of PHS basic research. The rows of Tables 4.1 and 4.2 show
the values of the coefficient estimates and the relevant statistics for each
regression. Because of space considerations not all of the regression
coefficients are show in the tables. The full regression output can be seen in
Appendix B.

In each table, all three non-nested regression criteria , the sum of
squared residuals (SSR), the coefficient of determination, and the value of the
log likelihood function, indicate that PHS basic research feeds pharmaceutical
innovation seventeen years prior to NCE approval. Also note that the values of

these criteria are not much different for the eighteenth lag of PHS basic

80

.96. .0... o... .0 8:80:90:

.25. $0 .5 s 8:85:90.
«02-20. “28>
... €385 c. 2.. «80030.. 3806.2
.8. c. an 830...: __< ”202

 

 

 

3008 800.6 28.6 0000.6 32.6 98.8 82.30184
0000.0 030.0 «mt-0 ~83 Rt... 8&0 3300-0
89.00. 030.5 008.5 83.00. 80.09 .0800. 000
...—800.0. ....830. ...00000. ...00000. 2.002.... .0980. 80.32 .380 00.:
.83.. F. 02000. 2.5030. ..5000. .0880. ...3000. 88%.. 8:30 0...“.
.803. .58... .520. ...6000. 0205. .39.: 0.533 5:0
:80... ...-.35. .500... .550. :80... .800... 80.53 88.8
303 00000 803-. 00.00 nth... 800.. 239080 03.05.30
8 a... 2 0.... 2 05 t 0.: 2 0.... 2 a... .680 02380 280 0:0
5000. 2.00.0. 3000. 080.0. 9.8.0. 2.5.0. 38.90.80 020.390 .0 .50
.8: k8; 9.00.0 68.0 85.0 2000 05.9080 000 0.2.2. .o :50
.88.... 855-. 58..-. .800... :80.-. .85.-. 80.0%-. 88.8
800.2- 0. 6.:- ~000.2- 23.:- 00§07 008.0. .5380
0 0 v 0 N 0 80.50 s is;
88300

 

0002 a .500 “300...; 2880100

00-. 53.0.... 80> N. 5.; 8.38m gaom 03002-002 6000 $03 30.00001 0.000 01a - ...v 0.000.

81

.96. 0... 9.. .0 8:80.090.-

..96. 0.0 o... .0 8080.00.0-
000..0.0. .28..
... 0.9.8... c. 0.0 80000.9. 00205.2
.00. c. 2.. 8.0009 ..< 0.02

 

 

 

000000 0000. .0 0.00. .0 0000. .0 80.. .0 0.0000 080.2... 03
0020 01-00 02-00 000.0 0.000 00000 02300-0
000.0. 000000. 000000. 0. .000. 0000.00. 0000.00. 000
.0800. 2.80.0.0. ...0000. 2.000... 2.0.00.0. ...-08.... 0000....-. .880 00...
.00.... .. .00000. ...0000. 2.0000... ... .020. ..00000. 000050.. .380 ......
.0000... .000... .3000. .0080. .00.0... .0000... 000080.. ...-.0
..000... .0000... .0000... .0000... .0000... .000... 000080.. 88.00
300.. 0.00.0 8000 00000 0.00.. .00... 50.9080 00.05000
00 0.... 0. 0.3 0. 00.. 0. 0.3 0. 0.... 0. 0.... .8... 02280 0.20 0:0
.800- 00000. 00000- 0000? 0.000. 0000.? 0.8.0.080 00.0.0000 .0 .50
.00... 0.000 00000 E00 00000 .0000 0.8.0.080 000 0.80... .0 200
.0000. .-. .0000.-. 5.0..-. .0000.-. .0000.-. .200.-. 0.2.0.0.. 080.00
0000.0.- k. ..0.. 0000....- 800.0.. 00000.- 0.....0- .0880
0 0 0 0 0 . 00.030 a 0.08.3
gnaw

 

0002 .0 .500 0.000; 0.008.000

00-. .5800. .8> 0. 5.3 2.080 8.08.000 0808-82 .000 $0.. 09880 0.80 0:0 - 0 0 0.0...

82

research. Further, the timing of the basic research contribution is consistent
across both industry lag specifications, either the twelve year or fourteen year.
Tables 4.3 and 4.4 present the alternative PHS stock regression results.
Each table shows two alternative regressions in which the seventeenth lag of
PHS basic research is set against alternative lags. The first regression includes
the twelfth, seventeenth, and twenty-second lags. The second regression
expands the range to ninth, seventeenth, and twenty-fifth lags. In each of these
specifications only the seventeenth lag of PHS basic research falls within the
range hypothesized in chapter two. Although the collinearity of these stock
variables lowers the precision of the estimates, the seventeenth lag stands out
as the relevant timing. Both the generalized linear model (GLM) and fully robust
standard errors show statistical significance at conventional levels.
Consequently, both regression approaches support the timing hypothesis
outlined in chapter two. Moreover, the results are consistent with previous
studies on the lag in pharmaceutical innovation (see DiMasi et al. (1991) and
Dranove and Meltzer(1994)). Since an F-test found that the thirteenth and
fourteenth lags of industry R&D do not add explanatory value to the equaiton,
the remainder of the discussion will focus on the twelve year industry lag results

shown in Table 4.1.

83

._o>o_ x; 05 «a 8:85:99.

..26. {on 2: E 8:35:99.
33.03? H83>

.a 29.35 5 2a 8:383 0:05.52
duo. a. 2a 85%? .< ”202

 

 

 

 

82.8 m r38 3239.: 8.4
3:6 326 uoaaamé
83.3 88¢? mwm
5.8. 2 L883 883. .982 Luzon 8803 0363.. $33. as...
53.2 ..an 5. .582 .882 L38. 2 85.3 80.53.. 2.6
3R0... .52.. a fixed. .38.... .88. z .833 03.22 88.8
8.3 .53 «~30 8:8 v83 «83. 220580 8555
an 2.. t a... o 8.. a a... t 9.. «P 8.. goo.» =288¢ 23m 9:
80¢. 82¢- 38.0580 bassoon .0 25
3.3.6 53 25.0580 Sm £32. a .56
NE... E 394.. Z 083%.“ 88.8
58.2- 85.2- 2588
N F 88% a 03%.;
Boson—ow

 

502 .o ano 65%; E3530

8.. 5.8... 30> m. 5.; 238m 538.com 8.82 28 gm: 558cm comm wxa - 3 22¢

.26. ...; as a 858.com:

.65. 3m 9.. «a 8:85:99
39.39 65;
... 89.85 5 2a 83.83 0.63.5.3
.80. c. 05 3.3.3.. .4. ”202

 

 

 

 

8m :8 2.: .3 82.23.34
ES 3:... 853m-..
338. 5.58. «mm
.832 2.532 .832 89.2 ...882 882..-. 88%.. .253. 2...“.
5.2.2 .85.. z .882 .852 ...:fi. 2 .88..-. 03%.. 2.6
.88.... .88. a .332 .832 :3... 2 HBO... 8%.». 838
8.3 88a 88... 8:3 Show 2%.... 220580 Begum
on 3.. t a... a on. a 2.. t a... 3 a... .8.» 5288: 2.8 mza
33.... :5... 25.0580 €5.33. .0 55
~33 .38... 95.0580 9E £32. a saw
.88.... .82..-. 8%.. Samoa
88.2- 32.2. 232.8
m . 888m 3 22...;
2.333

 

£02 .0 ano 6.3:; Eoucaoo

9.. ESE. as. 3 5.; 258m 5.829.. 2.82 .30 ....m 5 5.88m ovum wIa - 3 2%»

4.4 The Magnitude of the Public Research Impact

Given the fact that PHS funded research is an intermediate input into an
innovative process that is subject to long lags, it is not surprising that its
economic impact has been hotly debated in policy circles. Looking at the
estimated coefficients in columns (3) and (4) of Table 4.1, the economic
significance of PHS basic research as a determinant of pharmaceutical
innovation becomes clear. An estimated constant elasticity ranging between 2.2
and 2.5 implies a 10% increase in the stock of PHS basic research will, on
average, lead to a 22% to 25% increase in the number of approved NCEs in
each of the seven therapeutic classes.

The following two examples provide a sense of the important impact of
PHS basic research on pharmaceutical innovation. First, consider the impact of
a 10% increase in the basic research stock in 1976 on the number of approved
NCEs in 1993. In 1976 the PHS granted $705 million in awards for basic
research across all classes. This is approximately 26% of all PHS extramural
awards given in that year. A 10% increase in the stock of basic research
available in 1976 would have required an additional $216 million—an 8%
reallocation of the total value of PHS awards to basic research. The effect, on
average, would have been five additional therapeutic compounds in each class
in 1993. It should be noted that this impact will continue to feed pharmaceutical

innovation beyond 1993.

85

86

Unlike overall federal basic research, PHS funding of basic research
relevant to the pharmaceutical industry has not experienced a dramatic decline
in real annual growth. This contrast in funding trends allows us to ask what
pharmaceutical innovation would have been had PHS basic research followed
the overall trend. The counter factual experiment illustrates both the importance
of federally funded basic research in pharmaceutical innovation as well as the
potential impact of changes in the real growth rate of basic research funding.

Using the estimated model, Figure 4.1 shows the impact of imposing a
slower growth rate in PHS basic research funding on pharmaceutical innovation.
The solid line represents the observed (and predicted) number of approved
NCEs in the sample period 1978-1994. The dotted line represents the predicted
number of NCEs with an imposed 0.6% decline in the real growth rate of PHS
basic research. Holding all else constant, this decline leads to a drop in the
average number of approved NCEs from 19 to 11. Clearly, evidence from the
pharmaceutical industry suggests that a decline in the growth rate of federal

basic research funds will lead to a significant drop in measured innovation.

4.5 The Statistical Significance of Public Research

The tests for statistical significance involve alternative assumptions on

the conditional variance of NCEs as well as accounting for potential serial

correlation in the implied disturbances. Until recently, the most common test for

87

.5266 $8.8 £380 @595“. 2058mm 23m Ema 532w 5.2, 293 8.66:... _8_.:mom§mcn. - _..v 8:9...

 

93>

QQQQQQQQQQQQQQQQQ
aces/020002..@ovcesooaeezeooo(o(

 

 

11il ill... Tl+ .. .1..i!-.|--rll+il+llT|.iiTl..-ilT ._ . c

nmoz
.5803. 38.3.... I l l

mwoz case}.
octomnOIIl

genome”.

28m 38.8... c.

:32... 833 ...?
£30.. congoccT . .l . .

 

SBON penciddv JO tunog

 

.302 3353‘.
«.33 5230:...

V1. --

 

 

 

 

 

 

88

statistical significance in count data models assumed the Poisson nominal
variance property in which the mean and variance are restricted to equality.
Under this assumption, which is shown in column (3) of Table 4.1, the
seventeenth lag of PHS basic research has a positive coefficient with a t-statistic
equal to 2.02 implying a p-value < 0.025 (one-sided alternative). The weakness
of this assumption is that it rarely holds in empirical applications and,
consequently, can lead to spuriously high levels of significance.

The next two assumptions generalize the Poisson property. The GLM
assumption allows a proportional relationship between the mean and the
variance. Since the conditional distribution of NCEs is characterized by
underdispersion, estimated c2 < 1, the variance is less than the mean.
Accounting for the underdispersion in the conditional distribution of NCEs leads
to a smaller standard error. So, PHS basic research has a t-statistic equal to
2.25 implying a p-value < 0.025. The underdispersion finding is consistent with
previous research on pharmaceutical innovation. In particular, Thomas (1990)
also found underdispersion in his analysis of NCEs and pharmaceutical
regulation. Finally, the fully robust standard error does not impose any
restriction on the conditional variance of NCEs. The fully robust t-statistic is
equal to 3.56 implying a very small p-value, less than 0.0001. Consequently,
regardless of which assumption on the conditional variance of NCEs one feels is

most appropriate, PHS basic research is strongly statistically significant.

89

To be completely robust, it is necessary to account for the time dimension
of the data.2 A Lagrange Multiplier test is used to detect serial correlation in the
implied disturbances. Both the magnitude and significance of the serial
correlation indicator variables show that negative serial correlation is present in
the model. The LM statistic for the regression is (T-4) * (uncentered R-Squared)
= 35.19. Using the chi-squared distribution with four degrees of freedom, the
null hypothesis of no serial correlation is rejected at a 5% level. Based on this
finding, the heteroskedasticity-serial correlation (HISC) robust standard error is
calculated accounting for up to fourth order serial correlation. The adjustment
for serial correlation reduces the standard error to 0.2508. As seen in column
(3) of Table 4.1, the resulting t-statistic on PHS basic research is 8.95 showing a

high statistical significance.

4.6 Alternative Depreciation Rates for Public Research

In general, the impact of PHS basic research on pharmaceutical
innovation is robust to the assumed depreciation rate of knowledge capital. Up
to this point the empirical analysis has assumed a 15% depreciation rate of PHS
basic research. Although this is the depreciation rate used Hall et al. (1988), if

the effect of federally funded basic research is important, we would expect a

 

2 Adjusting the standard errors for serial correlation ex post has the advantage of
not imposing auxiliary assumptions on the model. Using a FGLS AR(1) or
higher order model would have introduced common factor restrictions into the
statistical model. See Wooldridge (1991) for further discussion.

90

change in the depreciation rate to only affect the magnitude of the effect and not
its timing or statistical significance. The magnitude of the impact will depend on
the size of the effective “knowledge stock.” To evaluate the sensitivity of the
timing, magnitude, and significance of the PHS basic research results to the
assumed depreciation rate, the non-nested regression analysis is repeated
using both a 10% and a 20% depreciation rate for the basic research stock. .The
results of the sensitivity analysis can be seen in Tables 4.5 and 4.6. As before,
the non-nested criteria point to the seventeenth and eighteenth lags as the
relevant timing of the effect. The pattern of statistical significance remains
unchanged. The magnitude of the effect, however, does change slightly when
alternative depreciation rates are used. Under the 10% rate, the estimated
impact is greatest with a coefficient of 2.6 as opposed to a coefficient of 2.0
under a 20% depreciation rate. The pattern implies a slightly smaller impact

when the effective stock of basic research “knowledge capital” decays rapidly.

4.7 The Industry Elasticity and Net Present Value of Public Research

The estimated elasticity of industry R&D can be seen in row 3 of Tables
4.1 and 4.2. Focusing on seventeenth lag of basic research in column (3), it
can be seen that the magnitude of the cumulative effect of pharmaceutical R&D
does not change dramatically between the two lag specifications. Under the

twelve year specification, industry R&D has an elasticity estimate of 0.69 while,

.26. .... o... .- 83.5w:

.26. ...... o... .a 8:82:98.
.838. ”88>
... 8.8... c. 2.. 8.85.... 8885.2
68. ... a. 83...? ..< 5.02

 

91

 

88.8 .818 88. ... 88.8 «new. a 88.8 82.9... .5
82.... «at... 8k... «83 8t... 2t... 8.38-..
88.8. 9.5.8. $8.8. 88.8. 28.5 88.8. «8.
L882 ...582 ...882 ...882 :88... ...882 8859. .88.. cm...
:8... s ..832 3.882 Lot-n2 ...882 .2822 88.8-. .88”. ......
.83... ...-NE... .582 .882 ...83. Sam... 88...... ...-.0
.38... 28:... ...-88... ...-.382 .88. z .883. 88s.... 88......
88.. 89... 82..“ 83a 58.. 88.. Egg-:80 8.23
8 a... 2 a... 2 ...: t a... . 2 a... 2 a... ...-.8» ‘22.... 2.8 2:
89.... 88.... 8:8. 88.... 88.9 88...- 839880 82.38.. .o 58
«SN. .8... 88... $2... 8:... «8:... £58.80 2:. .382. .o 58
.85..-. .82..-. .88..-. .83..-. .98..-. .85.-. 883... 88.9".
92.9- 898.- «837 88.:- §..«.. 88....- .588
o n v n u . 888 a 28...;
£883

 

 

omoz .0 .58 5.8.; E8880

...... .326... as. m. a. 3.3.... 5.828.. 8.82.82 8.88m 0.8m 9.... .- 83888 .8. - m... 038

.26. .... 6:. ... 8:8...85:

.26. .8 o... .a 8:8...:...w.
v8.8... .28..

... 26.8... :. e... 8.6.5.... 883.2
.68. :. 2.. 8.8:? ..< 682

 

92

 

 

88.8 8%.... .8... ... 88.... .895 8...... 82.3%
v.8... 8.... .8... 8.... ...... 8.... 8.26m-..
8... .n. 88.8. ..88. 29.8. .898. 88.8. «mm
2.8.88. ...888. ..........m. ...88... 2.8:... 2.88... 88...... .263. om...
.88. .. ..888. 2.68.8. :28... 2.8.88. ...83... 88.6.. .88.. 2...“.
.... .... .. .....8. .. ...... . .8. .. 8.8.8. .88. .. .88. .. 9.8.6.. 2.0
.88... .68.... .88... 8.8... .88.... .88... 88....-. 88.8
88.. 8.... {-8.8 88.8 8.0. 1..... 26.8.80 3.2.3
8 ...... a. a... .. ...... .. 2.. o. a... a. a... .8.» 2288.. 28m ......
8.8. 88.9 8.3- 68.? 8...? 8...... 8:38.80 3853...... .o :5».-
.8... ..8... 8m... 88... 88... 68... 3:28.80 Sm E52. .0 saw
.88..-. ..8...-. .38..-. .88..-. .88. .-. .82..-. 8886-. 58.6..
88.... .83.- «83.. 8887 88...- B8..- .588
o n v a a . 0.6.6.... a 6.8:;
68.33

 

802 .258 ”2......28288

on. .532. 56> a. a. 8.3:. 8.68.8... 38.82 5.88.. 0.8m mxa .- 886280 .88 - o... 6......

93

under the fourteen year lag, the elasticity is 0.72. These estimates are similar to '
the findings of Thomas (1990) whose elasticity estimate is 0.66.

The results indicate that the marginal contribution of PHS funded basic
research (MP = 0.067 per $1 million) is larger than the marginal contribution of
industry R&D (MP = 0.046 per $1 million) leading to approved NCEs. This
makes sense given the different nature of public basic research and industry
R&D. Whereas public research expands fundamental knowledge about disease
processes, the bulk of industry R&D effort is devoted to animal and human
testing of specific compounds for safety and efficacy. Basic research has
broader applicability and may contribute to the discovery and development of
many therapeutic compounds.

An alternative method for evaluating the relative merit of our national
investment in basic biomedical research is to calculate the net present value
(NPV) of this investment. A recent study by Grabowski and Vernon (1996) uses
a sample of new chemical entities and a product life cycle model to estimate the
present value of cash flows for an average NCE. Their present value stream of
cash flows for an average NCE is $193.1896 million in real 1986 dollars. If we
take an initial 10% increase in the stock of PHS basic research, say in 1976,
then our investment in basic research will produce an average of five approved
NCEs in each therapeutic class by 1993. Across all classes, this gives a total
NPV of cash flows of $6761.636 million [(5 NCEs) * (7 classes) *(193.1896 cash
flow per NCE)] at the time of drug introduction. To calculate the NPV, we

capitalize our initial investment of $216 million at 105% (this is the GV (1996)

94

cost of capital) and subtract this from the present value of cash flows. Doing this
calculation indicates that our public investment in basic research has a positive

NPV of $5582.368 million in real 1986 dollars at the time of introduction.

4.8 Substitution Possibilities, Returns to Scale, and FDA Regulation

Given that the marginal productivity of public research is greater than that
of private R&D, one might suggest that public funding should be substituted for
private funding in order to increase the number of new therapeutic compounds
available to fight disease. This suggestion is overly simplistic. A closer look at
the nature of the research funded with federal monies and the research funded
with private industry monies reveals that these separate funding agents are
“buying” different types of research. Most obviously, public basic research does
not include the uncertain and expensive animal and human testing for safety and
efficacy that must be completed to obtain FDA approval. In order to investigate
the potential substitutability between publicly funded basic research and private
industry R&D, a substitution parameter is introduced into the model based on the
CES production function described by Kmenta (1967).

In this extension of the model, industry R&D is specified as a stock
variable using the methodology established by Thomas (1990). The results of
this regression appear in Table 4.7. The estimated elasticity of substitution (E5)

of industry R&D for public basic research is 0.29. Since this value is near zero,

95

Table 4.7 - CES Production Function Estimation

Dependent Variable: Count of NCEs

 

 

Ecmation
Variable or Statistic 1
Constant -12.5714
Poisson t-statistic {-1.7121]
Industry R&D Stock Coefficient 0.3773
Poisson t-statistic [0.625]
GLM t-statistic [0.7238]
Fully Robust t-statistic [42677]“
"Substitution" Term: (ln l/B)"2 -0.3996
Poisson t-statistic {-1.0344]
GLM t-statistic [-1 . 1 979]
Fully Robust t-statistic [-2.1903]‘
Sum of Regulatory Coefficients -0.4941
PHS Basic Research Stock (15% dep.) Lag 17
Estimated Coefficient 2.1769
Poisson t-statistic [20306]"
GLM t—statistic [21852]"
Fully Robust t-statistic [42099]“
SSR 157.3997
R-Squared 0.7209
Log Likelihood 85.6068
Elasticity of Substitution 0.2869
Scale Parameter 2.5542

 

Note: All variables are in logs.
Asymptotic t-statistics are in brackets [1
Years: 1978-1994

'Significance at the 5% level.
“Significance at the 1% level.

96

public basic and private industry R&D are substitutable, however, they are poor
substitutes. In the CES model, the elasticity of substitution can vary between 0
and so. When Es —> 0, the isoquants become L shaped indicating a fixed
proportions Leontief technology. The result implies that pharmaceutical
technology is characterized by limited substitution possibilities between private
R&D and public basic research. A 1% increase in the relative price of private
R&D investment, holding output constant, will only allow a 0.29% decease in the
relative use of private R&D (its factor share). And, since the relationship is
symmetric, a 1% increase in the relative price of public basic research, holding
output constant, will only allow a 0.29% decrease in the industry’s relative use
of this research. This finding is consistent with the development of our national
research enterprise following World War II. Our national commitment to
financing an expanding institutional foundation producing basic research, largely
through universities, has created a profitable and productive opportunity for the
pharmaceutical industry to specialize in more applied research. One may argue
that what has evolved is more or less an institutional division of labor.

Both the Cobb-Douglas and CES estimation results indicate strong
increasing returns to scale in the discovery and development of approved new
chemical entities. Under the twelve year industry distributed lag, the sum of the
research coefficients (a, B) is 2.89, while the scale parameter in the CES
regression is 2.55. Without the contribution of PHS basic research, the
estimates indicate that pharmaceutical innovation is characterized by decreasing

returns to scale. Most previous studies, including Henderson and Cockbum

97

(1996), do not find evidence of increasing returns to scale in pharmaceutical
innovation in the post-1978 sample period. However, it is important to keep in
mind that previous studies have not included a measure of basic biomedical
research. If fact, it is the contribution of this research that leads to industry
increasing returns to scale.

The cumulative effect of FDA regulatory stringency can be seen in row 4
of Tables 4.1 and 4.2. Regardless of the industry lag specified, the cumulative
elasticity is -0.59. In the CES estimation, the elasticity has a value of -0.49.
Although there are no other estimates of FDA regulation for the sample period
used in this study, the estimate is slightly lower than the estimate reported by
Grabowski el al. (1978) in their study of the regulatory impact of the Kefauver-
Harris Amendments of 1962. FDA regulation appears to have a fairly strong

negative impact on pharmaceutical innovation.

4.9 Conclusion

Strong evidence is found for an economically and statistically significant
impact of public research on pharmaceutical innovation (hypothesis one). The
elasticity of the number of new chemical entities with respect to the stock of
public basic research “capital” is found to lie in the range of 2.2 to 2.5. These
point estimates are large and imply increasing returns to scale in the

pharmaceutical innovative process due to publicly funded basic research.

98

The chapter finds that public research has its most significant impact on
pharmaceutical innovation seventeen years before an approved NCE. This is
consistent with the idea that pharmaceutical scientists monitor public research
and use public research results in drug discovery. It is the drug concept stage of
the pharmaceutical innovative process in which publicly funded basic research
has its impact (hypothesis two). This result is perhaps best interpreted as a
seventeen year lag between the time basic research reaches a “maturity" point
or “critical mass” and the approval of a new chemical entity.

Publicly funded basic research and the bulk of industry investment are
very different in their objectives and results. Public basic research is typically
more diverse and general while private industry R&D tends to be directed at
testing specific drugs and documenting their findings. Each type of research is
important in the pharmaceutical innovative process. Their distinctive
contributions imply limited substitution possibilities in the production technology.
The estimated elasticity of substitution (E) of 0.29 supports this hypothesis.

it is found that the marginal productivity of federally funded basic research
is 0.07 per $1 million in each therapeutic class. A $7 million investment in public
basic research produces, after a lag of seventeen years, a present discounted
value of $91 million in real 1986 dollars at the time of drug introduction. Further,
publicly funded basic research has a large net present value. The NPV equals
$5582.368 in real 1986 dollars at the time of introduction.

The results hold up to robust statistical inference. As assumptions

regarding the conditional variance of NCEs are relaxed, the estimated impact

99

becomes more significant. The heteroscedasticity and serial correlation robust
asymptotic t—statistic is 8.96. Further, the impact is robust to alternative industry
lag specifications. A twelve year innovative lag is preferred by the data. Finally,
the PHS impact is robust to variation in FDA regulatory stringency. This effect

was allowed to reach as far back as nine years prior to FDA approval.

CHAPTER 5
The Indirect Impact of Public Basic Research on Pharmaceutical

Innovation: Investment in Response to Research Opportunities

5.1 Overview

This chapter explores the indirect impact of public basic research and
FDA regulatory stringency on pharmaceutical innovation. Each of these factors
indirectly impacts the number approved new chemical entities by influencing the
level of industry R&D investment. Hypothesis four, developed in chapter two,
postulates that public funding of basic biomedical research facilitates advances
in public scientific understanding and thereby creates research opportunities for
new drug discovery. As new avenues to therapeutic outcomes emerge, private
industry attempts to exploit these opportunities by investing in research and
development. The induced private R&D investment makes some contribution to
the successful introduction of new therapeutic compounds. This is the indirect
effect of public basic research on pharmaceutical innovation.

Similar to public funding of basic research, FDA regulatory stringency
indirectly impacts pharmaceutical innovation by influencing the level of industry
R&D investment. In this case, however, one would expect that increases in
regulatory stringency reduce industry research and development investment.

Empirical support for this hypothesis was provided by Wiggins (1979, 1983).

100

101

Because of the parallel approach taken here, it is expected that increases in
regulatory delay lower industry R&D investment. Both analyses use the same
methodology to construct the regulatory stringency proxy variable and use the
same source for industry research and development data. This similarity
provides an opportunity to compare results and test the robustness of Wiggins’
analysis. Direct comparisons with Wiggins (1983) appears in section 5.5 of the
chapter.

The central objective of this chapter is to empirically describe the timing,
magnitude, and significance of the indirect impact of public research on
pharmaceutical innovation. Of the previous research, no study has linked
publicly funded research with pharmaceutical innovation. Ward and Dranove
(1995) explore the inducement effect of NIH obligations on pharmaceutical
investment but stop short of linking this with innovation. Henderson and
Cockbum (1996) and Grabowski and Vernon (1981) have used “research
program” and firm level data to look at the determinants of pharmaceutical R&D
investment without including any measures of public research. As regards FDA
regulation, Wiggins (1979) was the first to use therapeutic class distinctions to
show that FDA regulatory stringency depressed pharmaceutical investment and
innovation. Ward and Dranove (1995), on the other hand, use a somewhat
different approach and find that “expected” approval delays increase
pharmaceutical R&D investment.

This chapter advances our understanding of the determinants of

pharmaceutical industry investment and the indirect contribution of public

102

biomedical knowledge to pharmaceutical innovation. Whereas previous
analyses have treated public research investment in a broad and somewhat
simplistic fashion, this analysis is based on detailed public research awards data
that have been carefully classified into meaningful therapeutic categories for
each year between 1955 and 1985. Additionally, the contribution of biomedical
knowledge is measured as a capitalized stock. This is intended to mimic the
cumulative Ieaming process that characterizes the advancement of scientific
research. Both of these aspects improve the empirical results by reducing the
measurement error and matching limitations found in previous research.

The chapter finds that the elasticity of industry R&D investment with
respect to the stock of publicly funded basic research lies in the range of 0.42 to
0.46. Although publicly funded research begins to stimulate industry investment
as early as three years following award, the data reveal that its effect is
statistically dominant in the seventh year following research award. Using the
estimated marginal effect of industry R&D on approved NCEs from chapter four,
a $1 million increase in the PHS research stock produces a marginal physical
product of 0.01 approved NCEs in each of the seven therapeutic classes. This
amounts to a total increase in pharmaceutical innovation of 0.07 additional
approved NCEs (across all classes). If this increase in innovative output earns
the average discounted revenue, then a discounted present value of $10.3
million in additional revenue will be produced at the time of introduction.

As regards FDA regulatory stringency, the point estimates are statistically

insignificant and small in magnitude. These elasticities, one for the current year

103

through the fourth lag of regulatory stringency, were not found to be statistically
different from zero either individually or jointly. Taken literally, this implies that a
marginal increase in FDA approval delay time does not affect the level of
industry R&D. Although this may be true, collinearity among the included
stringency variables is one cause of the imprecise estimates. If we put any faith
in the estimated coefficients, then the cumulative impact of FDA regulatory
stringency has an overall elasticity value of -0.1023.

The chapter contains several other ancillary results. Perhaps the most
interesting of these pertains to industry sales. With an estimated elasticity of
0.24, a 1% increase in pharmaceutical sales in 1985 would have increased
industry investment by $7.1 million in that year. Using the estimated impact of
industry investment on the production of NCEs found in chapter four, this
increase produces 0.08 new therapeutic compounds across all classes after
twelve years. Again, if these marginal NCEs each earn the average discounted
revenue, then this increase in sales will bring fourth nearly $14.8 million in new
revenues at the time of introduction.

It is also found that industry R&D investment increases with disease
incidence and falls with disease “severity.” A 1% increase in disease incidence
in the forty-five to sixty-four year old age group leads to an 0.4% increase in
industry investment. The estimated effect of disease “severity,” which is
measured as the number of hospital days, is negative and significant. Thus,

longer hospital stays are associated with lower pharmaceutical investment.

104

The last section of the chapter presents some exploratory analysis
regarding the potential simultaneity bias between publicly funded research and
private industry R&D investment. As discussed in chapter three, this is the
“scientific opportunity” problem. It arises when research funding agents respond
to the same “scientific break-through“ information when making their funding
decisions and thereby induce a correlation that is not a spillover of knowledge.
Using instruments for public research implied by the Ward and Dranove (1995)
analysis, no evidence is found that supports an endogeneity bias due to
scientific opportunity. The intended instruments, however, are poorly correlated
with public funding of basic research. This makes the endogeneity test
uninformative as to the presence of a scientific opportunity problem.
Nevertheless, scientific opportunity is unlikely to affect lagged public funding.
The analysis finds that the relevant public and private funding decisions are
separated by a seven year period. In order for scientific opportunity to be
causing a simultaneity bias, the relevant scientific opportunities would need to

remain constant over the seven year period. This seems unlikely.

5.2 Model Specification

Based on previous research as well as interviews with industry marketing
personnel, this section follows the net present value (NPV) methodology for
modeling pharmaceutical R&D investment decisions. This method is consistent

with the interview responses given to Wiggins (1979) and used in his analysis.

105

It is also the approach taken by Grabowski and Vernon (1981) in their firm level
analysis of the determinants of pharmaceutical R&D investment.

The basic NPV relation is shown in equation (5.1). The calculated NPV is
the difference between a project’s discounted present value stream of expected
revenue and its the discounted present value stream of expected costs. If the
NPV is positive, then it is profitable to invest in the R&D project. If the NPV is

less than or equal to zero, then it is not profitable to invest.

(5.1) NPV = PV[expected revenues] - PV[expected costs]

To make projects comparable, the discounted present value of cash flows
for revenues and costs are calculated using a “discount factor" that adjusts for
the time value of money and for risk. Alternatively, the internal rate of return for
a proposed R&D project can be calculated. This return is the value of the
discount rate that equates the present value of expected revenues and expected
costs. Comparing the calculated internal rates of return to the opportunity cost
of capital for any project is one way of making an R&D investment decision.1 If
the firm calculates the internal rate of return for all projects and ranks them in
descending order, then they can fund projects down to the point where the return
on the marginal project just equals the cost of capital. Notice that this assumes

pharmaceutical firms are not liquidity constrained.

 

1 The opportunity cost of capital is the expected rate of return to investors
offered by all other investments of similar risk.

106

It is useful to image a downward sloping “marginal return to investment”
schedule plotted with the rate of return on the vertical axis and the level of
investment on the horizontal axis. Given the distribution of returns, the
downward slope of the curve reflects that R&D investment increases as
additional R&D projects are funded. Along this curve, all variables influencing
the rate of return on new chemical entities are held constant. Because the
opportunity cost of capital is assumed to be exogenous, it is represented by a
horizontal line emanating from the vertical axis and extending across all levels of
investment. The equilibrium level of investment is determined by the
intersection of the “marginal return” schedule and the “opportunity cost of
capital” schedule.

The empirical analysis will be done on an industry level using the
observed level of investment in each medical therapeutic class. It is postulated
that public basic research, FDA regulatory stringency, and demand factors
influence the rate of return and shift the marginal return schedule for each
therapeutic class. Macro changes that shift the opportunity cost of capital
schedule are modeled using yearly time dummy variables. Allowing each year to
have its own intercept is the most general way to account for variation over time
due to inflation, industry shocks and other unobserved “macro” changes.
Therapeutic class dummies are included to account for the fixed differences in
technological opportunity across medical classes. Depending on the distribution

of research investment and scientific skill, one class may feed into drug

107

discovery more easily than another. The structural investment equation is as

follows:

(5.2) In = (Bitm)"‘(Rit)",... ,(Rit_h)’"(Salesit)°‘(lncit)°2(Sevit)°3e‘”""°e°°

where i represents the medical therapeutic class, t represents time. 'it is the

level of industry research and development investment; [31 is the long-run
investment elasticity with respect to the lagged stock of PHS funded basic

biomedical research, Bit-m ; y1-yh are the investment elasticities with respect to
the distributed lag of regulatory stringency, Rit - RM, (length h = 4); 01-03 are

the elasticities of investment with respect to industry sales, U.S. disease
incidence and severity, respectively. The M (one for each year) are the semi-
elasticities capturing changes in the opportunity cost of capital while the ni
correspond to each therapeutic class. The term, It. is a random disturbance with
mean zero and constant variance. The lag, m, characterizing the stock of PHS
basic research will be determined empirically.

It is postulated that increases in the stock of public basic research

knowledge, BM”, lead to increases in the level of pharmaceutical investment.

In the present framework, this effect works by increasing the rate of return for all
projects in a given therapeutic class. Although research opportunities may affect
the rate of return through alternative mechanisms, it is likely that the effect of

public research is some combination of lower costs and a higher the probability

108

of success. Either by lowering expected costs or increasing the expected payoff
to private R&D, public research increases the rate of return. This implies [31 > 0.
It was noted above that increases in FDA regulatory stringency are

expected to lower pharmaceutical R&D investment, *1" < 0. FDA regulation can

lower the rate of return through two mechanisms. First, changes in regulatory
stringency may increase the risk of FDA approval in a therapeutic class. In our
framework, an increase in the risk of approval implies a lower probability of
approval. With a lower probability of approval, fewer R&D projects will be
profitable because the expected revenue stream will be lower. If fewer projects
are pursued, then investment will be lower. The second mechanism works
through the costs of approval and holds the risk of approval constant. In this
case, an increase in the expected stream of costs associated with getting an
FDA approved drug lowers the rate of return to all projects. Assuming the
opportunity cost of capital remains constant, fewer projects will be pursued and
R&D investment will be lower.

In a multi-period setting, FDA product quality regulation may either
increase or decrease pharmaceutical research investment. To the extent that
increases in regulatory stringency require more testing and documentation, the

cost of those compounds already in the pipeline will increase, yj > 0. However, if

firms are able to adjust their portfolios of research investment, then increased
regulatory stringency would reduce the number of profitable candidate

compounds. Fewer profitable compounds, in turn, lead to reduced spending as

109

these potential new therapies are not pursued, yj < 0. It is this later effect that

Wiggins (1983) found to be dominant between 1971 and 1976. However, Ward
and Dranove (1995) find the former effect to be dominant.

Industry sales and measures of US. disease incidence and severity are
included in the structural equation to represent the influence of expected
demand on the pharmaceutical investment decision. Increases in industry sales
are expected to increase the level of pharmaceutical investment, 01 > 0. This
effect can work through two alternative channels. First, sales represent an
important source of funds for financing R&D projects. To the extent that
pharmaceutical firms are liquidity constrained, greater sales allow firms to fund
additional projects and increase their level of R&D investment. It should be
noted, however, that the rate of return framework used here assumes that firms
are not liquidity constrained. The second channel works through the market
information provided by drug sales. Here, the level of sales acts as an indicator
of market size and the potential for future sales. To the extent that observed
sales in a therapeutic class affect the forecasted revenue stream for projects in
that therapeutic class, sales will influence the level of R&D investment. It is
expected that greater sales signals increasing market demand. This, in turn,
shifts the marginal return schedule up and stimulates investment.

This analysis includes two additional measures of expected demand,
disease incidence and “severity” for the civilian non-institutionalized US.
population. Based on the first listed diagnosis on the hospital discharge sheet,

the total number of discharges are summed across disease category into

110

therapeutic classes by year and age group. Disease “severity,” which is
measured as the total number of nights in the hospital, is constructed using the
same methodology. For each variable, four age groupings are used: 0-14
years, 15-44 years, 45-64 years, and 65 and older. It is expected that increases
in disease incidence and severity lead to greater demand for therapeutic
compounds. As demand increases, the expected revenue streams are
increased and the return to all projects in that therapeutic class shift up. Holding

all else constant, industry R&D investment will increase, 0j > 0.

5.3 Functional Form

Previous studies of pharmaceutical investment and regulation have used
either the linear model (Wiggins (1979, 1983), Grabowski and Vernon (1981),
Jensen (1987), Henderson and Cockburn ( 1996)) or the log-log functional form
(Ward and Dranove (1995)). In all cases, the authors did not claim that their
choice of functional form was based on a statistical test. In this analysis,
however, two alternative econometric tests were used to determine the
statistically preferred functional form. The first, developed by Davidson and
MacKinnon (1981) (DM), is referred to as the P5 test. The second test was
developed by Wooldridge (1991) and is robust to heteroscedasticity and serial
correlation.

The results of all tests indicate that the log-log specification is preferred.

Two DM tests were performed. The first test takes the linear model as the null

111

hypothesis and involves evaluating the statistical significance of a nested
indicator variable (the indicator is calculated as the difference between predicted
values of the alternative models). With the linear model as the null, the indicator
variable has a asymptotic t-statistic of 6.0879 (p-value < 0.0005). Thus, the null
linear hypothesis is rejected in favor of the alternative log-log hypothesis. In the
second DM test, the log-log model served as the null hypothesis. In this case,
the indicator was statistically insignificant with a t-statistic of -1.3127.
Consequently, the log-log null hypothesis is not rejected.

The limitation of the DM test is that it assumes a correctly specified
conditional variance. The test proposed by Wooldridge (1991), however, is
robust to conditional variance mis-specification, including both
heteroscedasticity and serial correlation.2 He proposes a Lagrange Multiplier
test which produces a chi-squared statistic under the null hypothesis. Taking
the log-log model as the null hypothesis, the value of the test statistic is 0071.
This is less than the chi-squared critical value at a 1% significance level and one
degree of freedom. Thus, the robust test fails the reject the null log-log

hypothesis.

 

2 To be completely precise, the correspondence between the linear and log-
linear models requires the additional assumption that E(l|X) is proportional to
exp(E[log(l)|x]). This, however, is not restrictive and is usually assumed
implicitly.

5.4 The Regression Results and Discussion

Table 5.1 presents the regression results for equation (5.2) using
alternative lags of the PHS basic research stock. The leftmost column lists the
variables and the relevant statistics while columns (1) through (9) of the table
show the results for the alternative lags of public research. Notice that the
heteroscedasticity (Hetero) and the heteroscedasticity/serial correlation robust
(HISC) standard errors have been calculated for the PHS basic research
variable following Wooldridge (1989). The HISC standard errors were
calculated accounting for up to 1st order serial correlation. It should be noted
that this is not the same as using a feasible generalized least-squares (FGLS)
procedure with an AR(1) model for the errors. Unlike F GLS, the procedure used
here does not impose any common factor restrictions on the model (see

Wooldridge (1991) for further details).

5.41 Timing of the Relationship

In order to describe the magnitude and significance of the indirect impact
of public research, the timing of its relationship with pharmaceutical investment
must be considered. This is important because there is a time lag between the
financial award to researchers and when this research becomes available to the

broader research community. The empirical identification of the appropriate lag

112

113

.92 .9 2.. ... 8:85:29:
.26. :8 o... s 8:85:29.
.26. $9 2.. ... 8:82:99

89.4an n800>

... 82:2: :. a: 8.2.8-. 2.0192

.30. c. 8: 332.9. __< ”:82

 

 

8. 8. 8. 8. 8. 8. 8. 8. 8. 82583983232
83.: 93.: 93.: 83.: 83.: 83.: 93.: 93.: 83.: 8.3.9.3:
98.: 98.: 98.: 98.: 88.: 88.: 83.: 83.: 83.: 8.2.8-:
8:... 89.. 89... 88.. 82.... 89.. 89.. 93.. 93.. 2::

9.93.9... .8238. 3.9-:
99.: 5.88: 28: 9.:
Co .. 2.2.8: 8.35.20 >882:
:8 .oz .2: .oz .2: 82 :8 .oz .2: .02 .:.m :52... .:.m 8:8 .2: «so: :8 2:0: 8.6.56 2::
.:.m 2:8 .2: 2:8 :8 2:8 :8 2:8 .2: 2:8 .2: 2:8 .:.m .82 .:.m .8... .2: .8... 8.558 320
.989. .989. .839. .839. .989. .939. .839. .899. .889. 2.2.8..
:9:9 85.9 859 88.9 88.9 8.:9 35.9 95.9 85.9 8.3.3:... v :3
.89.... .89.... .839. .999. .889. ...:89. .989. .899. .859. 2.2.8..
889 889 989 8.:9 859 9:9 839 ..89 98.9 8.33:3: : :3
.389. .889. .699. .999. .599. .88.:. .9 .:.9. .939. 8.9. 2.2.8..
95.9 86.9 9 .:.9 98.9 58.9 58.: 88.9 389 889 5.238: N :3
.889. .389. :89. .989. .899. .389. .88 9. .989. .889. 2.2.29.
«8:9 889 989 :39 85.9 8.:9 98.9 859 38.9 8.8.8:: . :3
.839. .899. .859. .889. :89. .889. .9 89. .839. .889. 2.2.89.
839 93.9 98.9 93.9 839 88.9 88.9 33.9 989 83:8: .556
2.88.9. ....889. ....881 588.1 ....891 ....88...-. ....8:...1 ....881 ....881 2.2.8-.
8 .:.9 98.9 989 :89 33.9 8 6. .. 33. .. ..9o. .- 88. .. .31... 8... .828.
...98. .. ..39. .. ...:9. .. ..89. .. ..33. .. ...9:..~. ...... .m.~. ....93.~. ....m3..:. 2.2.8..
3.3.: .83 88.: 89.: 89.: 9 .m: 88.: 93.: 3 9.: 999 8:. 8:88.
..9 .o. .. ...33. .. ... ...8. .. ...83. .. ..88. .. ..88. .. .99... .88. .. .89... 2.2.8..
.93 89.: 99.: 98.: 39.: 83.: 89.: 89.: 89.: 8...: 2.26:.
...88. .. ...9.~.~. ....8..~. ...38. .. ..88. .. ...:9. .. .5... .. 999:. 393. 2.2.8.. .38: 09:
...88.~. ...88.~. ...89a. ...98a. ...89. .. ..88. .. .89. .. .88.:. 58:. 2.2.8-. .88: 23:1
....88.:. 2.9.0:. ....«8:.:. ....83.~. ...:k...~. ...3. ..N. .38. .. .99... .88.:. 2.2.8.. 80
89.: 89.: 89.: 38.: .80: 99.: 9.9.: 89.: 3.9.: 8.83 8.22.8
: :3 9 :3 o :3 u :3 v :3 n :1. u :3 . :3 .556 28.: 8.88: 28: 9.:
....88.~. ....motd. ....:~8.~. ....98.:. 3.39.... ....9..~.:. ....98.:. ....93.n. ....8.m.:. 2.2.8..
83.: :89: 38.: 88.: 98:: .8... 8.3 99.9 «8:9 .588
: o 9 m m .. n N . 2.2.2: .0 28:;
9.2.9....m

 

 

82.23:. on: :88. :3 28:9 28:88

.80 8.. 6.88: 28: 9.: .0 88m .828. 2.. a. 2.8:: 8.828: . ..m 28»

114

is determined by comparing the sum of squared residuals and adjusted
coefficient of determination across alternative specifications.

Looking at the non-nested regression criteria, the seventh lag of the stock
of PHS funded basic research has the smallest sum of squared residuals and
the largest adjusted r—squared. While this is the preferred timing and will be
used in all subsequent discussion, it should be noted that the values of the non-
nested regression criteria are not much different for the sixth and eighth lags of

PHS basic research.

5.42 The Magnitude and Significance of PHS Basic Research

Table 5.1 reveals that the stock of basic research begins to influence
pharmaceutical investment as early as three years following research award.
The constant elasticity estimate is smallest and insignificant in the concurrent
year while, with each passing year, the magnitude and significance of the effect
increases until the seventh lag. Given that research awards are used to
measure the PHS public knowledge stock, it is sensible that its immediate impact
would be small relative to longer lags. Longer lags allow for the process of
doing scientific research and the process of communicating results.

Referring to column (8) of Table 5.1, it is clear that the coefficient on the
PHS funded stock of basic research is statistically significant. Without
correcting for heteroscedasticity or serial correlation in the model, publicly

funded research has a t-statistic of 3.32 (p-value < 0.001 ). Accounting for

115

heteroscedasticity reduces the t-statistic to 2.29 (p-value < 0.05) while the HISC
robust t-statistic drops further to 2.14 (p—value < 0.05).

The magnitude of the impact indicates that industry R&D investment
responds strongly to the stock of PHS basic research. A 1% increase in this
stock will result in a 0.46% increase in industry R&D after seven years. At the
sample mean, the marginal effect on pharmaceutical investment is about $166
thousand per million of public basic research. Alternatively, a 1% increase in the
stock of basic research in 1978 ($70.6 million) would have induced $13.7 million
in pharmaceutical investment by 1985.

Using this estimate with the marginal impact of industry R&D on approved
NCEs calculated in chapter four, a $1 million increase in the PHS research stock
produces a marginal physical product of 0.01 approved NCEs in each of the
seven therapeutic classes. This amounts to a total increase in pharmaceutical
innovation of 0.07 additional approved NCEs. Using the present discounted
stream of revenue earned by the average NCE (estimated by Grabowski and
Vernon (1996)), the additional 0.07 approved NCEs will produce a discounted
present value of $10.3 million in additional revenue at the time of introduciton.

The results suggest that industry firms are investing in response to
research opportunities created through federal extramural funding of research.
The emergence of our national research enterprise following World War II has
created productive and profitable research opportunities for industry. Public
basic research, which is conducted largely in academic institutions, creates a

foundation of public knowledge that contributes to industry efforts to conceive

116

new therapeutic outcomes. This is not to say that the interaction between
publicly funded research and industry research is unidirectional. Case study
evidence has been used to shed light on the complexity of the relationship
(Henderson and Cockbum (1997)); however, the purpose here is to generalize
beyond a handful of specific case analyses and test whether there exists a
broader relationship. Moreover, until now, we have not had any broader

evidence on the timing, magnitude and significance of this relationship.

5.43 Public Regulation and Industry Demand

Table 5.1 shows that the estimates of the impact of FDA review time are
not statistically significance either individually of jointly. This finding differs from
the findings of Wiggins as well as with some of the findings of Ward and
Dranove. In his linear model, Wiggins found that each of the second through the
fifth lags had a significantly negative impact on industry R&D (a comparison with
Wiggins’ results appear in section 5.5). His interpretation describes regulation
as reducing the number of new compounds entering the industry’s innovative
process. Ward and Dranove (1995), using the log of expected approval times
calculated from a first stage regression, find that review times are either
insignificant or have a positive and significant effect on industry R&D
investment. Their preferred interpretation characterizes their estimate as a
measure of the cost of FDA compliance plus the increase in rents associated

with intellectual property protection. The interesting contrast between the

117

findings here and those of Ward and Dranove is that their regulatory measure
became positive and significant in their regression that included the distributed
lag of NIH obligations. In Table 5.1, the signs of all regulatory variables are
negative. Although no statistical significance is found, this may suggest that
Wiggins’ interpretation is most relevant here. It is also the case, however, that
excessive noise and collinearity in the average review time measures are
leading to imprecise estimates.

These results provide an interesting contrast to Wiggins (1983) results
obtained using the period of 1971-1976. In his regressions the second through
the fifth lags of regulatory stringency were significant. This implies that there
has been some type of change in the behavior of pharmaceutical firms toward
FDA regulation. It may be that interacting with the FDA early in the discovery
stage and investing in “governmental affairs” departments have transformed
FDA regulation into a fixed cost rather than a variable cost of innovation.
Further investigation into the impact of FDA regulation seems warranted.

Column (8) of Table 5.1 shows the elasticity of industry investment with
respect to industry sales. With a point estimate of 0.24, a 1% increase in sales
leads to a 0.24% increase in industry R&D investment. In 1985, for example, a
1% increase in sales implies a $7.1 million increase in industry R&D investment
in that same year. Moreover, the sales variable is statistically significant with a
t-statistic of 1.98 (p-value < 0.05).

The regression results in Table 5.1 include US. disease incidence for the

45-64 age group and US. disease severity for the 15-44 age group. The other

118

age groupings were insignificant and were dropped from the equation. The
elasticity estimates indicate that a 1% increase in disease incidence results in a
0.40% increase in pharmaceutical investment. For disease severity, on the
other hand, a 1% increase in the number of hospital days results in a 0.90%
decrease in pharmaceutical investment. The impact of disease incidence on
investment is in the expected direction. The impact of severity, however, was
not expected to lower pharmaceutical investment. It is unclear why longer
hospital stays would lower investment in drug research. Yet, the effect is

strongly significant was a t-statistic of -3.99 (p-value < 0.0001 ).

5.5 Wiggins (1983) Revisited

Because of the similarity between this analysis and Wiggins (1983), it is
interesting to re-examine the original Wiggins (1983) results it light of the new
data available. In Table 5.2 his original model is estimated using the current
data. For these regressions, the model is linear in the levels, all variables are in
nominal dollars and the sample period is 1971-1976. Wiggins (1983) results
appear in the right most column while the regressions with the current data are
in columns (2) through (7). Although there may be slight differences in the
measurement of PhRMA sales and R&D, the primary difference here is the
introduction of a public basic research measure that was unavailable to Wiggins.
Also, these regressions include an additional therapeutic class in the cross-

sectional dimension, namely the gastrointestinal/genito-urinary class.

Table 5.2 - Re-Enmindion of the Wiggins Model

119

Dependen Variable = Level of lndustg R&D lnvedment

 

 

 

Equations
Wiggins 2 3 4 5 6 7
Intercen nla 84.417 81.948 83.712 22.764 25.853 26.69
[5757]“ [5563]“ [5714]“ [0.931] [0.993] [1.200]
PHS Basic Research 0.234 0.217 0.216
[2867]“ [2572]“ [3.170 "
Sales 0.047 0.080“ 0.0645 0.0727 0.046 0.052 0.043
[2.47] [4861]“ [3019]“ [3219]“ [2.3141' [2.409}‘ [2.347]'
Reguldion (curred) -0.067 -0.049 0056 -0.064 -0.068 -0.058
[-1 .041] [-O.744} [-0.840] {-1.084} {-1 .148} {-1 .028]
Reguldion (leg 1) -0.070 -0.048 -0.052 -0.057 -0.058 -0.05
[-1.310] [0837] [-0.914] [-1.122] [-1.142] {-1 .059]
Reguldion (1392) -0.486 -0.131 -0.102 -0.103 -0.114 -0.111 -0.105
{-2 .38] [2589]“ [-1 .817}' {-1 .807}' [-2.269]' {-2.170]‘ [-2.341]'
Reguldion (leg 3) -0.773 -0.131 -0.109 -0.141 -0.129 -0.146 -0.121
[~3.65] [-2.422]“ [-1 .9241' [-2.161]‘ {-2.529]“ [2498]" {-2.580]“
Reguldion (leg 4) -0.885 -0.056 -0.05 -0.066 -0.076 -0.075 -0.072
[-3.44} {-0.938} {0825] {-0.927] {-1.395} [-1.170] {-1.364]
Reguldion (leg 5) -0.481 0.008 0.007 -0.52 -0.005 -0.038 -0.004
{-1.80] [0.111] [0.101] {0585} [-0. .79] [-0.474} {-0.068]
CNS dummy n/a -29.179 -20.333 -23.341 32.798 26.95 31.071
[-1.650] {-1.062] [-1.211] [1.303] [1.033] [1.271]
Cardio dummy nla 15.769 10.022 16.428 48.484 48.802 43.93
[1.531] [0.881] [1.229] [2886]“ [2808]“ [3482]“
Anti-inf dummy n/a 15.934 17.853 18.052 61.6111 58.266 58.733
{1.905]‘ [2.106}' [2.125]‘ [3620]" [3.351)“ [3830]“
Gl/GU dummy -54.094 -57.85 -56.112 -8.841 -11.74 -13.672
[6325]“ [6358]“ [5960]“ [-0.467} {-0.611] {-0.926}
Derm dummy nla -53.172 -61.01 -56.14 15.251 12.246 7.191
[5975]" {-5.477}" [4645]“ [0.537] {0.427} [0.350]
Resp dummy n/a -26.431 -37.951 -26.077 44.383 44.237 34.849
{-1.708]‘ {-2.071]' [-1.166] [1.344] [1.305] [1.479]
Time 2.426 -0.921
[1.158] [-0.419]
YR 1972 5.946 2.163
[0.892] [0.351]
YR 1973 3.485 -4.182
[0.462] [0566]
YR 1974 12.457 0.92
[1.518] [0.107]
YR 1975 14.891 1.932
[1.596] [0.198]
YR 1976 5.423 -8.077
[0.482] l-0-7101
R-Squared 0.69 0.964 0.965 0.971 0.974 0.977 0.973
Rber-Squered nla 0.96 0.947 0.948 0.958 0.958 0.96
Num Obs 42 42 42 42 42 42 42

 

Note: All variables are in logs.
Asymptotic t-stdistics are in brackets {1.

Years: 1971-1976

“Significance I the 5% level.
“Significance a the 1% level.

120

Columns (2) to (4) explore the relationship without incorporating research
opportunities from publicly funded basic research. The estimates in column (2)
show that concurrent sales and FDA regulation are significant and have the
signs found by Wiggins. The impact of sales is twice that found by Wiggins and
more statistically significant. Also, only the second and third lags of regulation
are significant whereas the second through the fifth lags were significant in
Wiggins’ regression. When a linear time trend and year dummies are
introduced, Columns (3) and (4), the coefficients generally show lower
significance and are slightly smaller. Although these differences are probably
due to differences in the data sets, it is broadly confirmed that average review
times did lower industry R&D expenditure in this period.

Columns (5) through (7) introduce publicly supplied research
opportunities into the relationship using the seventh lag of PHS funded research.
The change in the estimates is quite interesting. First, the intercept and the
therapeutic class dummies change dramatically. This is to be expected since
these variables were postulated to capture the effects of research opportunities
by Wiggins. Second, the estimate of industry sales falls and is now in line with
the his original estimate. Third, the second and third lags of regulation are now
more significant than previously although they are quite a bit smaller in
magnitude than Wiggins’ original estimates.

The most preferred specification appears in column (7). Here, the

insignificant time variables are excluded. Public basic research has a strong

121

and large impact on the level of industry R&D investment. With an estimate of
0.22, each additional dollar of public basic research leads to a 22 cent increase
in industry R&D investment seven years hence. This contrasts with the 4 cent
increase resulting from an addition dollar of concurrent sales. Average review
times have a significant impact in this sub-period although the fourth and fifth
lags do not show any significance using the current data.

Overall, Wiggins’ results are broadly confirmed. His exclusion of public
research does not appear to have biased his sales and regulatory estimates.
Revisiting his model has highlighted the importance of incorporating a measure
of public research as well as shown that the effects of FDA review times on

industry expenditure have changed over time.

5.6 Endogeneity Test for Scientific Opportunity

It has been suggested that estimates of the relationship between public
and private funding of research are biased by the endogenous response of
funding agents to the same set of scientific opportunities. To explore this issue,
a reduced form equation for the stock of PHS funded basic research is specified
using disease incidence and severity instruments for public research. The

reduced form equation is:

(53) Bit = h('"Cit. SeVit)

122

where i represents the medical therapeutic class, t represents time. Bit is the
stock of PHS funded basic research; lncit and Sevit are the measures of disease

incidence and severity not already included in the estimation of equation (5.2).

Ward and Dranove (1995) maintain that government funding is a function
of disease prevalence and severity.3 While this is the basis for choosing the
instruments for equation (5.3), there are two complications with Ward and
Dranove’s view. The first involves their measures of publicly funded research
while the second involves causation. Apart from timing issues related to their
use of financial obligations versus awards, Ward and Dranove use a very broad
measure of public health research that includes many special and large
programs designed to address immediate health problems. One would
reasonably expect that the funding of these programs has been driven more by
measured prevalence and severity, however, one would also expect that these
programs have been somewhat successful in lowering targeted disease
prevalence and severity. Clearly, the causation problem comes from
categorizing prevalence and severity as determining funding while they are also
an outcome of this funding.

Public research measured in this analysis is quite different from Ward and

Dranove’s measure. Here, publicly funded basic research is analyzed. While it

 

3 Ward and Dranove (1995) measure disease incidence by the number of
physicians in a particular specialty. While this measure is problematic for the
obvious reason that physician specialties have only an indirect correspondence
to disease prevalence. Here, disease incidence is used because systematic
prevalence measures are not available.

123

is still conceivable that basic research funding responds to perceived incidence
and severity, it is difficult to know exactly how scientific review groups choose
projects to fund. Presumably, these decisions are based on some notion of
scientific merit and not on short-term goals to lowering disease incidence and
severity. In the long-run, funded basic research may well lead to lower
incidence and severity indirectly through therapeutic compounds or other
embodiments of this knowledge. Thus, the issues of how public projects are
picked for funding and the direction of causation cloud the relations specified in
equation (5.3). The most simplistic a priori expectations imply that funding of
basic research responds positively to increases in incidence and severity, h1 > 0
and h; > 0.

Using the heteroscedasticity robust version of the Hausman specification
test, column (1) of Table 5.1 shows the calculated test statistic has a value of
2.351. Comparing this with the chi-squared (one degree of freedom) critical
value reveals that the null hypothesis of exogeneity cannot be rejected at a 10%
level or higher. Consequently, no evidence is found for endogeneity of publicly
funded research and industry research.

Table 5.3 shows the regression results of equation (5.3). As with industry
sales, the regression includes the omitted measures of disease incidence and
severity along with the other variables from equation (5.2). It is clear from this
table that all the incidence and severity variables are insignificant. This
contrasts with the maintained view of Ward and Dranove (1995). Their position

holds that NIH funding of research responds positively to these variables in a

Table 5.3 - PHS Basic Research

Dependent Variable: PHS Basic Research (15% Dep.)

 

 

 

Equation
Variable or Statistic 1
Constant 1.8992
t-statistic [0.8674]
Industry Sales -0.0670
t-statistic [-0.7004}
Incidence (age 0-14yrs) 0.5866
t-statistic [0.3221]
Incidence (age 15-44yrs) 0.1280
t-statistic [0.56447]
Incidence (age 45-64yrs) 0.3543
t-statistic {1 .0703]
Incidence (age 65+ yrs) 0.1202
t-statistic [0.3876]
Severity (age 0-14yrs) 0.0047
t-statistic [0.0340]
Severity (age 15-44yrs) 0.1081
t-statistic [0.5421]
Severity (age 45-64yrs) —0.0124
t-statistic [-0.0425]
Severity (age 65+ yrs) 0.1450
t-statistic [0.5025]
Current Regulation -0.0413
t-statistic [-2. 1 778]"
Lag 1 Regulation -0.0288
t-statistic {-1 6333]"
Lag 2 Regulation -0.0255
t-statistic [-1 .3660]
Lag 3 Regulation -0.0221
t-statistic [-1 .1 935]
Lag 4 Regulation -0.0223
t-statistic [-1.1628]
Lag 5 Regulation -0.0118
t-statistic [-0.6127}
Class Dummies Sig.
Time Dummies Sig.
SSR 0.5370
R-Squared 0.9970
Number of Observations 105

 

Note: All variables are in logs.
Asymptotic t-statistiqs are in brackets [1.
Years: 1971-1985

'Significance at the 10% level.
“Significance at the 5% level.

125

given year. The notion that funding responds to incidence and severity makes
sense if one can properly account for the problems of timing and causation,
which have not been adequately addressed in this specification. The timing
problem comes from relating incidence and severity in a given year to funding in
a given year. In our context that implies researchers are receiving awards for
grant proposals prepared at roughly the same time as incidence and severity are
being observed. Clearly, researchers must prepare proposals in advance and
this relationship should not be expected to hold. The problem of causation was
mentioned earlier. Consequently, the simple a priori expectations that h1>0 and
h2>0 are not realized.

Since the chosen instruments for public basic research are weak, this
violates one of the properties that instruments must have in order to test for
endogeneity. Consequently, the econometric test is uninformative.
Nevertheless, scientific opportunity is unlikely to affect lagged public funding.
The analysis finds that the relevant public and private funding decisions are
separated by a seven year period. In order for scientific opportunity to be
causing a simultaneity bias, the relevant scientific opportunities would need to

remain constant over the seven year period. This seems unlikely.

5.7 Conclusion

This chapter has explored the indirect impact of publicly funded research

on pharmaceutical productivity. This effect works through industry R&D

126

investment to stimulate new product innovation. The analysis finds statistical
support for an inducement effect of public basic research on industry R&D
investment. Federally funded research begins to impact pharmaceutical
investment as early as three years following financial award with its statistically
strongest impact coming seven years after financial award. The effect is
statistically significant with a heteroscedasticity/serial correlation robust t-
statistic of 2.23 (p-value < 0.05). The magnitude of the impact is in the range of
0.42 to 0.46.

The results imply a 10% increase in the stock of public basic research
leads to a 4.2% to 4.6 % increase in industry R&D investment after seven years.
When combined the marginal impact of pharmaceutical investment on approved
NCEs, a marginal ($1 million) increase in the PHS research stock produces a
marginal physical product of 0.01 approved NCEs in each of the seven
therapeutic classes. This amounts to a total increase in pharmaceutical
productivity of 0.07 additional approved NCEs. If this increase in innovative
output earns the average discounted revenue, then a discounted present value
of $10.3 million in additional revenue will be produced.

The empirical results indicate that FDA review delays no longer affect
industry R&D investment. This is not to say that FDA regulation imposes no
costs on the industry. Review delays still have a direct impact on innovation
(see chapter four) but do not seem to be increasing R&D expenditure for those
compounds in the pipeline. The bulk of compliance costs appear to be borne

early in the innovative process and are probably best measured using

127

proprietary data. Overall, Wiggins’ idea that FDA regulation keeps potential
compounds from ever entering the innovative pipeline seems most accurate

The elasticity of pharmaceutical R&D investment with respect to industry
sales is 0.24. A increase of 1% in sales leads to a 0.24% increase in R&D
expenditure. This would have increased industry investment by $7.1 million in
1985 and produced nearly $14.8 million in new revenues.

It is also found that industry sales increase with disease incidence and fall
with disease “severity.” A 1% increased incidence in the forty-five to sixty-four
year old age group leads to an 0.40% increase in investment. The finding that
disease “severity” in the fifteen to forty-four year old age group lowers industry
investment is in the opposite direction of the expected effect.

Finally, there was no evidence found that supports an endogeneity bias
due to scientific opportunity between public basic research and private industry
R&D investment. Although the econometric test is not informative, the fact that
public research is lagged seven years makes a simultaneity bias due to scientific

opportunity unlikely in our context.

Chapter 6

The Total Impact of Public Basic Research on Pharmaceutical Innovation

6.1 Overview

This chapter calculates the total impact of publicly funded basic research
and FDA regulatory stringency on pharmaceutical innovation. Recall that the
production framework identifies two related channels through which federally
funded basic research can influence industry output. First, basic research can
contribute directly to private industry’s product innovation. In this case, basic
knowledge is used as a direct input to create the new product introduced by
private industry. Second, this research can contribute indirectly to private
industry’s product innovation. This indirect contribution recognizes the role that
basic knowledge plays in stimulating additional private R&D investment. The
channels of influence for FDA regulatory stringency are analogous to those of
public research, however, having a negative impact instead of a positive impact.
The total impacts are the sum of the direct effects (chapter four) and the indirect

effects (chapter five).

128

6.2 The Total Impacts

Recall from chapter three that the total impact is calculated as the sum of
the direct and indirect marginal impacts. This relationship for PHS basic

research is given by the following relation:

(6.1) (int / dBit ) = ( int / dBit ) + ( int / dllt ) * ( dlit I dBit)

where i is medical therapeutic class and t represents time; Yit is a count of
approved new chemical entities; Bit is the stock of PHS funded basic research;
'it is the annual flow of industry R&D investment. The symbol, d, stands for the

mathematical derivative.
Estimates of the marginal impacts are obtained from the elasticity
estimates. For example, the direct marginal impact of public research is

calculated as follows:

(6.2) (MP)B = (Elasticity)y3 * [ (Sample Average)yl (Sample Average)B ]
(MP)B = (2.2) * [ (19) l (633.9879) ]

(MP)B = 0.0659

Using the other estimated elasticities and sample period averages, the total

impact of public basic research is:

129

130

(6.3) (Total MP)B = (0.0659) + (0.0458) " (0.1659)

(Total MP)B = 0.0735

Further, using Grabowski and Vernon (1996), if the total marginal impact earns
an average return, then the value of the total marginal product in each

therapeutic class is:

(6.4) (Value Total MP)B = (Total MP)B * (Avg. Discounted Value (1986$))
(Value Total MP)B = (0.0735) * ($193.1896)

(Value Total MP)B = $14.20 Million

80, across all seven therapeutic classes:

(6.5) (Value Total MP)a" masses = (7) * ($14.20) = $99.4 million

It is evident that the total marginal impact of PHS basic research is
determined mainly by the direct effect. The direct marginal impact (roughly 0.07)
is seven times greater than the indirect marginal impact (roughly 0.01) on the
number of approved NCEs. The indirect impact of PHS basic research must

work through the marginal effect of private R&D before having an effect on the

131

number of approved new chemical entities. An increase in the marginal impact
of private R&D will also increase the indirect effect of PHS basic research.
Analogous to publicly funded basic research, the total marginal impact of

FDA regulatory stringency is given by:

(6.6) ( int I dRit ) = ( int I dRit ) + ( int I (“it ) * ( dlit I dRit )

where all terms are defined as before and Rit represents the annual regulatory

delay (measured in months).

Because none of the indirect elasticities were found to be statistically
significant, the second term on the right-hand side of equation (6.6) is zero. This
implies that the total marginal impact of FDA regulatory stringency is equal to the
direct marginal impact. Because FDA regulatory stringency enters the empirical
specification in chapter four as a nine period distributed lag, the total marginal
impact of FDA regulatory stringency is calculated as the sum of the marginal
impact of each Iag. Using estimated and sample values, this total marginal

impact is:

(6.6) (Total MP)R = Z (Elasticity)YR * [ (Sample Avg.)y I (Sample Avg.)R ]

(Total MP)R = -0.284

132

Wiggins (1983) found the total impact of regulation came primarily from its
direct effect. His estimate of the total effect is -0.156, which is quite a bit smaller
than the direct effect found here. There are two reasons for the different
findings. First, whereas Wiggins eliminated insignificant lags of stringency in his
formulation, all lags were included in this analysis regardless of their statistical
significance. There are no good a priori reasons to claim only certain lags
matter. For instance, how sensible is it to say that lags two and five matter but
that lags three and four are irrelevant? Second, a longer distributed lag was
included in this analysis based on the interviews with industry regulatory

personnel. Overall, the estimate of regulatory stringency is probably too large.

6.4 Recap of Dissertation Findings

This dissertation has described the role of publicly funded basic research
in the discovery and development of new therapeutic compounds. It is found
that public basic research is a very important factor in pharmaceutical innovation
and that knowledge extemalities from this research lead to industry-vvide
increasing returns to scale. It was noted in the introduction that federal funding
of basic scientific research is one of the traditional priorities of our national
research policy. The objective has been to build a productive foundation of
knowledge that will stimulate national and industry innovation and growth. This
dissertation has provided some evidence that, at least in part, this objective has

been achieved.

133

Overall, this analysis has found evidence that the basic research
component of our national research policy has been successful in stimulating
industrial innovation. The analysis suggests that the decline in the growth rate
of federal basic research spending will have a negative impact on
pharmaceutical innovation. Although this result is found within the context of
biomedical research and the pharmaceutical industry, it could be that federal
R&D is quite important to industrial innovation and economic growth.

The main results from the estimation of the direct impact of PHS basic
research and FDA regulatory stringency on the production of new chemical

entities in the pharmaceutical industry are:

1. Publicly funded basic research has an economically and statistically
significant direct effect on pharmaceutical innovation. The elasticity of the
number of new chemical entities with respect to the stock of public basic

research is found to lie in the range of 2.2 to 2.5.

2. Public basic research has a distinct role in the pharmaceutical innovative
process. PHS basic research contributes to the creative conception of new
avenues to therapeutic outcomes and helps guide the parameters used in
chemical screening. Because public basic research affects the drug concept
period of pharmaceutical innovation, its impact comes early in the

pharmaceutical innovative process. The data indicate that an average of

134

seventeen years elapse between the stock of federal awards and the

introduction of new therapeutic compounds.

3. While still substitutes, public basic research and private industry R&D have
limited substitution possibilities in the pharmaceutical innovative process. The
estimated elasticity of substitution of 0.29 indicates that the pharmaceutical
innovative process is closer to a fixed proportions technology rather than a more
flexible substitution technology like the Cobb-Douglas. This reflects the different
nature of the public basic research and private R&D and lends support to the
hypothesis that a division of labor has emerged in the biomedical research

SGCtOf.

4. Using the Grabowski and Vernon (1996) estimate of the average return to an
approved NCE, the public basic research has a net present value of $5582 in

real 1986 dollars at the time of introduction.

5. Pharmaceutical innovation is characterized by increasing returns to scale at
the industry level. This result is a direct consequence of the availability of

publicly funded basic biomedical research.

6. Product quality regulation by the Food and Drug Administration continues to

have a negative direct effect on new chemical entity innovation. This is a pure

135

production effect, however, and does not include any of the possible benefits

stemming from increased safety and efficacy.

The main results from the estimation of the indirect impact of public basic

research and FDA regulatory stringency are:

1. Public basic research has an economically and statistically signification
indirect effect on pharmaceutical innovation by acting to induce private industry
R&D investment. The magnitude of this impact is in the range of 0.42 to 0.46.
This implies a 1% increase in the stock of public basic research leads to a .43%

to .46% increase in industry R&D investment after seven years.

2. The empirical results find no significant impact of regulatory delay times (in
months) on pharmaceutical investment. This result contrasts with the findings of
Wiggins (1983). It may be that, in the sample period, compliance costs are more

accurately thought of as fixed costs rather than variable costs of innovation.

3. The elasticity of pharmaceutical R&D with respect to industry sales is
estimated to be 0.24. A 1% increase in industry sales in 1985 ($202 million)
would have increased industry investment by $7.1 million in that year. Using the
estimated impact of industry R&D on the number of approved NCEs, this
increase produces 0.01 additional therapeutic compounds in each therapeutic

class after twelve years.

136

4. It is also found that industry investment increases with disease incidence and
fall with disease “severity.” A 1% increased incidence in the forty-five to sixty-
four year old age group leads to an 0.40% increase in investment. With regard
to disease “severity,” the analysis indicates that pharmaceutical investment falls
as the number of inpatient days increase. This underlying reason for this is

unclear.

5. There was no evidence found that supports an endogeneity bias due to
scientific opportunity between public basic research and private industry R&D.
Unfortunately, because of the weakness of the instruments, the econometric test
is uninformative. Nevertheless, any bias due to scientific opportunity is unlikely

since the funding decisions are separated by a seven year period.

6.5 Future Research

As one might image, there are numerous unanswered questions that
deserve research. Here are a few that come to mind:
1. What characteristics help or hinder a firm’s ability to exploit the pool of public
research?
2. How do these factors affect their competitive position in the industry and their

productivity?

137

3.. How do NCEs affect health output measures and, in turn, affect GDP and
national economic growth? Further, how can these insights improve our
estimate of the social return to public basic research?

4. How does federally funded basic research, or any other type of federal

sponsorship, affect other industries in our economy?

APPENDICES

APPENDIX A

APPENDIX A

Keywords Used to Construct PHS Basic Research Variables

This appendix lists the keywords and character stings used in the data
filters. Copies of worksheets were not able to be formatted for this appendix.

138

139

Table A.1 - Listing of Class Filter Keywords or Character Strings

Note: After running each of these filters, the results must be inspected.
Character strings typically capture more than is desired.

Anti-lnfective Class:

IAMEBICIDEI, ITRICHOMONICIDEI, [ANTHELMINTIC], /ANT|B|OTICI,
[CYCLINES[, ICEPHAU, ICILLINI, ['THROMYCINI, /STREPTO[, IISONIAZIDI,
[MALARIA], NIRAU, NIRUSI, IFUNGI, [BACTERIA], [PARASIT/

Endocrine/Neoplasm Class:

[HORMONE], [CORTIC], [ANDRO/, [ESTRO], [PROGESTO], IDIABETI,
ITHYROIDI, [INSULIN], IANABOLICI, IENDOCRINI, INEOPLASMI, [CANCER],
[TUMOR], [CARCINI, [CHEMOI

Cardiovascular Class:

[HEART], ICARDIOI, ICOAGUI, DIGITALISI, [HEMOSTAT], [HYPOTENSIVE],
NASO], IHEPARINI, IARRYTHMIC], [CALCIUM]

Central Nervoufs Svstem Clas_s:

IPARASYMPATHI, [MUSCLE], INARCOTICI, IOPII, [SALICY[, [ASPIRIN],
]ACETAMIN[, ICONVUU, [DEPRESS], [BRAIN], [NERV[, [TRANQUIL],
IAMPHETAI, [ANOREX], [SEDAT], [ANESTHE[, [ARTHR|], [HYPNO], ] EYE ],
[OPHTHA], [MYDRIA], [MIOTIC], ] EAR I, IAURII, [RHEUMA], [NEURO/

Gastro-intestinallGenito-Urinam Class:

] RENAU, [URINE], [KIDNEY], [NEPH/, [UREML lBlLE/, [GASTR], [INTESTINE[,
[COLIN], [CHOLERETIC], [CHOLESTEROU, [EMETIC], lHISTAMlNE/,
[CHOLINERGICL [LIPID], [GENITO], [CHOLAGOGUE], ['I'HIAZIDEI, [DIURETIC]

Denhatologic Class:
ISKINI, IDERM/, IQUTANI, ISEBACEI, IDANDRUFFI, ISEBOR/

Respiratom Class:
ILUNGI, [PULMONARY], lBRONCHl, [RESPIR/

140

Table A2 - Listing of Exclusion Filter Keywords or Character Strings

Note: After running each of this filter, the results must be inspected. Character
strings typically capture more than is desired.

[OGRAPHI, [OGRAM[, [SURGICAU, [DETECTION], [INTOXICATION[, [ BCGI,
[POPULATION], [EXERCISE], ['l'OXICOLOGYl, [PATIENT], [RHEUMAT|C[,
[CONTRAST], [CLINICAL], [DIAGNOS[, [HUMAN], [INVESTIGAT[, [ DRUG],
[RADI], [ MAN ], [INDICATOR], [ AGE [, [DEATH], [TEACH], [AGEING],
[FITNESS], [THERAPY], [PEOPLE], [PERFORMANCE], [ADOLESCENT],
[NATIONAU, [SUPPORT], [REHAB], [ENVIRON], [DENTAL], [EMOTION],
[CENTER], [LITERATURE], [MORTALITY], [CONSUM[, [FOLLOW-UP],
[PREGNAN/, [INFORMATION], [PROCEDURE], [COMMUNITY], ] AGENT],
[TECHNIQUE], [ DIET], [SCREEN], [X-RAY], [COOPERAT[, [PREVAU,
[LEARN], [URBAN], [AREA], [NURS[, [LIFE], [CHILD], [ EEG [, [ EKG ], [VCG l,
[FETAL], [INFANT], [DATA], [OMETER[, [PANEU, NACCINEI, [NEONATAL],
[RESEARCH], [INSTRUMENT], [EQUIPMENT], [ULTRASOUND], [PUBLIC],
[FELLOWSHIP], [PROGRAM], [DIRECTORY], [TREATMENT], [NUTRITION],
[THERAPEUTIC], [EVALUATION], [MEASUREMENT], / YEAR],
[COLLABORAT[, [ TRAIN], [PREVENT], [INDUSTR], [COMMITTEE],
[INSTITUTE], [CONGRESS]

141
Table A.3 - Activity Code Breakdown

1ST Round Elimination: Activity Code Breakdown
Year:

 

Activity Code Number of Grants Total Value of
Grants

 

Total Value - All Grants

 

A?? - Applied Training

 

0?? - Construction Grants

 

D?? - Environmental Demo

 

E?? - Gen Support Education

 

F04 - Nursing Fellowship
F09 - Scientific Evaluation
F10 - Fellowship Traineeship
F11 - Direct Traineeship

F13 - Hlth Science Scholars
F15- Fogarty Scholarships
F17 - Clinical Fellowship
F35 - Visiting Scientist Fello

 

G?? - National Lib of Med

 

H?? - Staffing Grants

 

J?? - Joint Facilities Grants

 

K08 - Clinical Investigator
K09 - Scientific Evaluation
K10 - Special Proj. (NLM)

 

M01 - Gen. Clinical Res. Ctr

 

N43 - Sm. Bus. Innovation

 

 

 

 

U?? - Cooperative Agreemnts

 

 

1 42
Table A3 (con’t)

 

P02 - Categ. Clinical Res Ctr
P06 - Animal Resources

P07 - Biotech. Resources
P09 - Scientific Evaluation
P10 - Envir. Hlth Centers
P11 - Pharma - Toxic Centers
P13 - Dental Res Inst. Prog
P15 - Outpatient Clin. Res.
P16 - Hlth Services Res. Ctr.
P17 - Spec. Centers of Res.
P18 - Sickle Cell Centers
P20 - Exploratory Grants
P30 - Center Core Grants
P40 - Animal Resources

P41 - Biotech. Resource
P50 - Specialized Center
P60 - Comprehensive Ctr

 

R02 - Nursing

R04 - Anthropology

R06 - Translation

R07 - Int’l Ctr for Res. & Tr.
R09 - Scientific Evaluation
R10 - Coop. Clinical Res.
R11 - MH Project Grants
R12 - MH Special Grants
R13 - Conferences

R14 - Psycopharm Conferen
R15 - MH Project Conferen
R16 - MH Special Conferen
R18 - Hosp. & Med. Facility
R20 - MH Hosp. lmprovemt
R21 - Comm. Hlth Explore
R24 - Biotech. Resources
R25 - Drug Abuse Education
R26 - Oregan Project NCI
R27 - Computer Technology
R43 - Sm Bus. Innovation
R44 - Sm Bus. Innovation

 

 

 

 

 

143
Table A3 (con’t)

 

S?? - Gen. Research Support

 

T?? - Training Grants

W?? - Foreign Currency Prog

Activity Code - Rnd 1 Total

 

 

 

 

 

 

 

After Eliminating inappropriate Institutes

 

File Name # of Grants Value of Grants

 

 

 

 

 

After Direct Edit of Some Study Sections

 

File Name # of Grants Value of Grants

 

 

 

 

 

 

After Eliminating inappropriate and Individual review Study Sections

 

File Name # of Grants Value of Grants

 

 

 

 

 

 

 

After Exclusion Grants are eliminated

 

File Name # of Grants Value of Grants

 

 

 

 

 

 

After Eliminating Class Filtered Grants

 

File Name # of Grants Value of Grants

 

 

 

 

 

 

APPENDIX B

Table 3.1 -Industry Lag 12. ms Log 15

Num of Obs 119
Log Likelihood @8616
Sum of Sqr Resid 12920239
R-Squared 0.77136
Sigma Squared 0.83227
Parameter

one 5.98m
curird 1.15545
lg1ird -2.31443
I92ird 1.66644
Igaird omeee
lg4ird 1.10461
lgSird -1 792$
lgeird 0.27066
lngrd 1 £331 6
I98ird -1.(9250
ngird 0.682(9
IgIOird 07666)
lg11ird -1.14263
Igl 2ird 1.77680
curreg -0.17164
lagreg1 -0.®19
Iagreg2 015877
lagrega 0.15317
lagreg4 0.20658
lagregs 00.1
lagregG 013434
lagreg7 0.16862
IagregB 031037
IagregQ 0.122%
PHS15 1.2656
y79 0&1 7
y80 068464
y81 0.625%
y82 -0.72731
y83 -1.44171
y84 -1.19476
yes 0.6175
y86 -1.72244
y87 -1 .65000
we -1 .44204
y89 -1 58:5
ya) -1.751$
y91 -1 .42518
y92 -1 56855
m -1.81371
y94 -2.24712
cns 1&82
cer 1.2288)
ant 1.181(1)
glu 1.-15
der 3.61557
res 3.13959

APPENDIX B

Estimate Poisson Std Err

7%
0%
1.32721
1.29748
1.2%16
1.26134
1.29478
1.3131)
1.28813
1.2%
1.1381
1.1164)
1.11Cm
1.12%
0.18201
0.18278
0.17m2
0.191%

144

Regresion Tables

1 .3452
0.74371
-1 .19070
-1 .‘76
-1 .1 1-
-1 some
-1 4%
0.76431
-2.027CB
-1 .8337
-1 64%
-1 cm
-1 .644!)
-1 .41 159
-1 .47322
-1 .621m
-1 .0“

Robust Std Err

3.25441
0.581 71
1 .1 2%4
0.74268
0.56412
0.45215
0.7700
1 .2253

Tabb 8.2 - Industry Lag 12. PHS Lag 16

Num of Obs 1 19
Log Likelihood 91.17339
Sum of Sqr Resid 128.15357
R-Squared 0.77272
Sigma Squared 0.82233
Parameter

one -10.74545
curird 1.10170
Ig1ird -2.34392
Ig2ird 1.77701
Ig3ird 0.08301
lg4ird 1.07848
IgSird -1.79714
Ig6ird 0.29062
lg7ird 1.09895
IgBird -1.15806
Ig9ird 0.69966
lg1 Oird -0.77551
Ig11ird -1.1553O
lg12ird 1.77984
curreg -0. 16304
lagreg1 -0.08536
lagreg2 0.06286
lagregB 0.16140
lagreg4 -0.19819
IagregS -0.06001
Iagreg6 013222
Iagreg7 0.16182
IagregS —0.31305
lagregQ -0. 12147
PHS16 1.57734
y79 -0.44796
y80 -O.88680
y81 -0.86420
y82 -1.00715
y83 -1.79105
y84 -1.621 13
y85 -1.09807
y86 -2.20950
y87 -2. 12652
y88 -1 .94485
y89 -1 .98685
y90 -2.25026
y91 -1.87157
y92 -2.10300
y93 -2.34943
y94 -2.77261
cns 1.67014
car 1 .39096
ant 1.37417
giu 1.89649
der 4.65524

res 3.95849

145

Estimate Poisson Std Err

7.35637
0.96847
1.32962
1.30730
1.29939
1.26202
1.29521
1.31893
1.28128
1.25546
1.12952
1.11262
1.10405
1.12311
0.18212
0.18284
0.17980
0.19239
0.18609
0.17476
0.15075
0.14378
0.13943
0.13694
1.00720
0.49728
0.62393
0.65756
0.73431
0.87077
0.95747
0.98151
1.02092
1.03901
1.04986
1.06272
1.11290
1.14285
1.23565
1.28616
1.36080
0.65256
0.73990
0.72820
1.08711
3.58166
2.56357

Poisson t-stat

-1.46070
1.13756
-1.76285
1.35930
0.06389
0.85456
-1.38752
0.22035
0.85769
-0.92242
0.61943
-0.69701
-1.04642
1.58475
-0.89521
-0.46687
0.34963
0.83890
-1.06504
-0.34339
-0.87706
1.12546
-2.24516
-0.88702
1.56606
-0.90082
-1.42131
-1 .31425
-1.37156
-2.05686
-1.69314
-1.1 1876
-2. 16423
-2.04669
-1.85248
-1.86960
-2.02198
-1.63763
-1.70194
-1.82670
-2.03748
2.55935
1.87993
1.88707
1.74453
1.29974
1.54413

Robust Std Err

3.84133
0.59789
1.10986
0.75980
0.56013
0.43524
0.75457
1.21700
1 .07382
0.40658
0.22364
0.52962
0.93413
0.66734
0.12105
0.17461
0.04080
0.09765
0.15463
0.13123
0.1 1838
0.08381
0.04196
0.03960
0.43897
0.23666
0.49789
0.43536
0.39571
0.52323
0.41416
0.65359
0.55356
0.56337
0.65170
0.63232
0.51058
0.55630
0.69441
0.82054
0.76994
0.25996
0.31543
0.22826
0.52699
1 .80658
1.33418

Robust t-stat

-2.79733
1.84265
-2.1 1 190
2.33878
0.14820
2.47791
-2.38166
0.23880
1.02340
-2.84827
3.12847
-1.46428
-1.23677
2.66707
-1.34687
-0.48888
1.54073
1.65282
-1.28172
-0.45730
-1.1 1686
1.93064
-7.46028
-3.06723
3.59331
-1.89283
-1.781 12
-1.98501
-2.54514
-3.42307
-3.91424
-1.68007
-3.99141
-3.77467
-2.98427
-3.14216
-4.40728
-3.36435
~3.02848
-2.86326
-3.60106
6.42456
4.40971
6.02016
3.59869
2.57683
2.96697

Table 3.3 - Industry Lag 12, PHS Lag 17

146

Num of Obs 1 19

Log Likelihood 91.95852

Sum of Sqr Resid 123.9466

R-Squared 0.78018

Sigma Squared 0.80209

Parameter Estimate Poisson Std Err
one -14.94784 8.03942
curird 0.99723 0.97669
lg1ird -2.31566 1.33139
ngird 1.88129 1.31563
Ig3ird 0.09163 1.30071
Ig4ird 1.08264 1.26561
lgSird -1.83958 1.29377
lg6ird 0.32968 1 .32360
lg7ird 1 .05712 1.27798
lg8ird -1 . 18441 1.24679
ngird 0.72595 1.1 1794
Ig10ird -0.75210 1.10622
lg11ird -1.16089 1.09329
lgIZird 1.78054 1.11184
curreg -0. 13099 0.18373
lagreg1 -0.06742 0.18312
IagregZ 0.06974 0.17928
lagregB 0.16879 0.19272
lagreg4 -0.17713 0.18749
IagregS -0.04882 0.17568
Iagreg6 -0.11712 0.15160
lagreg7 0.15523 0.14378
IagregB -0.32239 0.14006
lagregQ -0.12015 0.13674
PHS17 2.24577 1.11445
y79 -0.66396 0.52246
y80 -1.29293 0.69995
y81 -1.42364 0.78793
y82 -1 .62679 0.87784
y83 -2.52620 1 .04281
y84 -2.50901 1.18256
y85 -2.08686 1 .24720
y86 -3.27177 1.31121
y87 -3.19974 1.32920
y88 -3.02900 1 .34432
y89 -3.11266 1.37219
y90 -3. 35031 1.40223
y91 -3.00123 1.43927
y92 -3.19322 1.50167
y93 -3.57331 1 .60249
y94 -3.99364 1.66661
cns 2.03827 0.70079
car 1.74313 0.76437
ant 1.80280 0.78715
giu 2.48982 1.18288
der 7.00553 3.96835
res 5.76062 2.89935

Poisson t-stat Robust Std Err

-1.85932
1.02103
-1.73927
1.42995
0.07045
0.85543
-1.42188
0.24907
0.82718
-0.94996
0.64936
-0.67988
-1.06183
1.60143
-0.71295
-0.36814
0.38900
0.87584
-0.94476
-0.27790
-0.77253
1.07967
-2.30181
—0.87868
2.01514
-1 .27084
-1.84716
-1.80682
-1.85317
-2.42249
-2.12168
-1.67324
-2.49523
-2.40726
-2.25319
-2.26839
-2.38927
-2.08524
-2.12644
-2.22985
-2.39627
2.90854
2.28047
2.29030
2.10489
1.76535
1.98687

5.05233
0.62046
1 .05362
0.78339
0.52276
0.42929
0.76288
1.19204
1.04266
0.40098
0.21318
0.53674
0.91383
0.67709
0.1 1670
0.17089
0.03666
0.09681
0.15500
0.13300
0.1 1778
0.08135
0.04350
0.03799
0.63143
0.26043
0.55439
0.55694
0.54240
0.65850
0.64261
0.89240
0.84657
0.81316
0.89355
0.89592
0.78376
0.851 12
0.97763
1.13946
1.11520
0.35530
0.40091
0.35561
0.67470
2.45558
1.82553

Robust t-stat

-2.95860
1 .60724
-2.19782
2.40147
0.17529
2.52194
-2.41 137
0.27656
1 .01 387
-2.95376
3.40534
-1 .40124
-1 .27036
2.62970
-1 . 12246
-0.39451
1 .90229
1 .74363
-1 . 14275
-0.36707
-0.99437
1 .90833
-7.41 1 88
-3.16276
3.55667
-2.54946
-2.33217
-2.55617
-2.99926
-3.83631
-3.90442
-2.33848
-3.86474
-3.93495
-3.38985
-3.47424
4.27469
-3.52620
-3.26627
-3.13597
-3.5811 1
5.73674
4.34795
5.06957
3.69024
2.85290
3.15558

Table 3.4 - Industry Log 12, PHS Lag 18

Num of Obs 119
Log Likelihood 91.86173
Sum of Sqr Resid 124.23301
R-Squared 0.77967
Sigma Squared 0.7988
Parameter

one -16.85523
curird 1 .18978
Ig1ird -2.53168
lg2ird 1 .89409
IgSird 0.20049
lg4ird 1 .09517
lgSird -1 .76252
IgBird 0.13493
lg7ird 1 .19615
IgBird -1 .17591
ngird 0.77789
Ig10ird -0.74531
Ig1 1ird -1 .09928
lg12ird 1.65105
curreg -0.1 1481
lagreg1 -0.03943
lagregz 0.091 08
Iagreg3 0.16957
lagreg4 -0. 16014
Iagreg5 -0. 02758
lagre96 -0. 10975
lagreg7 0.16056
IagregB -0. 31 564
lagregQ -0.12226
PHS18 2.45921
y79 -0.75420
y80 -1 .49645
y81 -1 .74101
y82 -2.09261
y83 -3.00767
y84 -3.04816
y85 -2.76017
y86 -4.02915
y87 4.04022
y88 -3.88971
y89 -3.98873
y90 -4.27490
y91 -3.87720
y92 4.10684
y93 -4.40007
y94 4.92882
cns 2.1261 1
car 1 .77451
ant 1 .95389
giu 2.82042
der 7.94473
res 6.53707

147

Estimate Poisson Std Err

9.04179
0.96202
1.33779
1.31407
1.30197
1.26242
1.28799
1.31816
1.28012
1.24536
1.11559
1.09532
1.08902
1.09733
0.18465
0.18529
0.18032
0.19256
0.18793
0.17646
0.15243
0.14422
0.14015
0.13703
1.25845
0.55051
0.78531
0.94415
1.10748
1.28055
1.45812
1.59449
1.69614
1.75212
1.77806
1.81351
1.85980
1.87243
1.94921
2.00878
2.11542
0.75704
0.79651
0.88193
1.32733
4.51613
3.35521

-1.86415
1.23676
-1 .89244
1.44139
0.15399
0.86752
-1.36842
0.10236
0.93440
-0.94423
0.69729
-0.68045
-1.00942
1.50461
-0.62179
-0.21280
0.50510
0.88060
—0.8521 0
-0.15631
-0.72004
1.1 1329
-2.25222
-0.89218
1.95415
-1 .36998
-1.90555
-1.84400
-1.88953
-2.34873
-2.09047
-1.73107
-2.37548
-2.30591
-2.18761
-2.19945

-2.29858,

-2.07068
-2.10693
-2.19042
-2.32994
2.80846
2.22784
2.21547
2.12489
1 .75919
1 .94834

Poisson t-stat Robust Std Err

7.00221
0.54366
1.0031 1
0.80259
0.47088
0.44884
0.74645
1.13471
1.00713
0.39642
0.25455
0.55102
0.92808
0.64404
0.11402
0.17627
0.03927
0.09498
0.15268
0.13543
0.11962
0.08696
0.04473
0.03085
0.86351
0.31133
0.66652
0.68365
0.81186
0.88875
0.99182
1.25378
1.24355
1.25579
1.31417
1.37869
1.28076
1.33789
1.50169
1.59649
1.60156
0.42078
0.47012
0.53657
0.93097
3.34622
2.44434

Robust t-stat

-2.40713
2.18848
-2.52383
2.35996
0.42578
2.43999
-2.361 19
0.1 1891
1 . 18768
-2.96633
3.05592
-1 .35260
-1 . 18447
2.56358
-1 .00699
-0.22369
2.31938
1 .78523
-1 .04883
-0.20365
-0.91751
1 .84633
-7. 05667
-3.96297
2.84790
-2.42251
-2.24516
-2.54665
-2. 57754
-3.38416
-3.07330
-2.20148
-3.24004
—3.21727
-2.95982
-2.89314
-3.33778
-2.89799
-2.73481
-2.75609
-3.07751
5.05284
3.77457
3.64143
3.02956
2. 37424
2.67437

Tabb 8.5 - Industry Lag 12, PHS Lag 19

Num of Obs 119
Log Likelihood 91.53659
Sum of Sqr Resd 127.28431
R-Squared 0.77426
Sigma Squared 0.80302
Parameter

one -17.01 104
curird 1 .36312
|g1ird -2.50595
Ig2ird 1 .63343
lgBird 0.33780
lg4ird 1.13510
IgSird -1.64920
Ig6ird -0. 01 038
Ig7ird 1 .21742
Ig8ird -1.06137
IgQird 0.74666
Ig10ird -0.67758
lg11ird -1.08803
Ig12ird 1 .59672
cu rreg -0. 12167
lagreg1 -0.03537
IagregZ 0.1 1999
lagre93 0.17161
lagreg4 -0. 1 5797
lagregS -0.01 744
IagregG -0.09120
lagreg7 0.16485
lagregB -0.29141
IagregQ -0. 10866
PHS19 2.32217
y79 -0.74175
y80 -1 .48300
y81 -1 .78965
y82 -2.22354
y83 -3.22140
y84 -3.20792
y85 -2.97106
y86 -4.32853
y87 4.42292
y88 434573
y89 -4.46267
y90 -4.75709
y91 439654
y92 -4.57769
y93 -4.90143
y94 -5.32041
cns 2.00642
car 1 .59971
ant 1 .88666
giu 2.93351
der 7.79649

res 6.46526

148

Estimate Poisson Std Err

9.90278
0.95082
1.33868
1.29347
1.30286
1.26394
1.28275
1.32103
1.28293
1.24554
1.11634
1.09615
1.08610
1.09397
0.18412
0.18626
0.18394
0.19331
0.18742
0.17675
0.15527
0.14465
0.14063
0.13684
1.32597
0.56714
0.83386
1.05446
1.28440
1.51646
1.68699
1.87307
2.03202
2.13794
2.20778
2.25517
2.30701
2.33759
2.39065
2.46881
2.51723
0.76816
0.78214
0.93698
1.47742
4.91207
3.68214

-1.71780
1.43363
-1.87196
1.26283
0.25928
0.89807
-1.28568
-0.00786
0.94894
-0.85214
0.66885
-0.61815
-1.00178
1.45957
-0.66080
-0.18990
0.65235
0.88777
-0.84289
-0.09869
-0.58734
1.13961
-2.07224
-0.79406
1.75130
-1.30788
-1.77847
-1.69721
-1.73119
-2. 12429
-1.90157
-1.58620
-2.13017
-2.06878
-1.96837
-1.97886
-2.06202
-1.88080
-1.91483
-1.98534
-2.11359
2.61197
2.04530
2.01356
1.98556
1.58721
1.75584

Poisson t-stat Robust Std Err

8.99025
0.46346
0.99654
0.74303
0.41995
0.42791
0.67260
1.10127
0.99495
0.37775
0.24815
0.49535
0.88948
0.60602
0.11728
0.18127
0.05348
0.09255
0.15646
0.13891
0.12564
0.09248
0.05139
0.03413
1.09466
0.37673
0.79073
0.85757
1.10249
1.27914
1.39818
1.70891
1.71372
1.84680
1.94086
2.02662
1.95646
2.00283
2.15861
2.26342
2.21860
0.51980
0.55163
0.71113
1.26145
4.34977
3.18631

Robust t-stat

-1.89217
2.94120
-2.51466
2.19634
0.80438
2.65268
-2.45199
-0.00943
1.22360
-2.80974
3.00884
-1.36788
-1.22322
2.63476
-1.03741
-0.19512
2.24376
1.85428
-1.00966
-0.12557
-0.72584
1.78265
-5.67071
-3.18384
2.12137
-1.96891
-1.87547
-2.08687
-2.01684
-2.51841
-2.29436
-1.73857
-2.52581
-2.39491
-2.23908
-2.20203
-2.43147
-2.19516
-2.12066
-2.16550
-2.39810
3.85997
2.89997
2.65304
2.32549
1.79239
2.02907

Table B.6 - Industry Lag 12, PHS Lag 20

Num of Obs 1 19
Log Likelihood 90.83429
Sum of Sqr Resid 131.57881
R-Squared 0.76664
Sigma Squared 0.81429
Parameter

one -1 3.25281
cu rird 1 .53099
I91 ird -2.52755
ngird 1.54559
lgBird 0.29911
lg4ird 1 .19621
lg5ird -1.59851
lg6ird -0.07934
lg7ird 1 .27504
lgBird -0. 99498
ngird 0.75558
Ig10ird -0.71985
Ig11ird -1 .04268
lg12ird 1 .55646
curreg -0. 16055
lagreg1 -0.06735
lagregZ 0.10799
lagreg3 0.16713
lagreg4 -0.17005
IagregS -0. 02772
lagreg6 -0.11083
lagreg7 0.17732
lagregB -0.28141
IagregQ -0.09461
PHS 20 1.63038
y79 -0.53695
y80 -1.1 1344
y81 -1.31782
y82 -1.71974
y83 -2.65581
y84 -2.59026
y85 . -2.26236
y86 -3.58942
y87 -3.69891
y88 -3.63858
y89 -3.79473
y90 4.09076
y91 -3.72420
y92 -3.92292
y93 4.18404
y94 4.62526
cns 1.62135
car 1.17360
ant 1.441 13
giu 2.49809
der 5.64062

res 4.86635

149

Estimate Poisson Std Err

10.03849
0.94410
1.33895
1.28736
1.30127
1.26402
1.28582
1.32678
1.28927
1.24845
1.12858
1.10261
1.09804
1.09478
0.18127
0.18440
0.18604
0.19368
0.18666
0.17585
0.15455
0.14534
0.14138
0.13770
1.26380
0.54990
0.80466
1.04192
1.32695
1.58786
1.78136
1.96933
2.15845
2.31025
2.41646
2.50667
2.57070
2.60451
2.67893
2.72711
2.79686
0.72937
0.71539
0.91119
1.55845
4.95643
3.72179

Poisson t-stat

-1.32020
1.62163
-1.88770
1.20059
0.22986
0.94635
-1.24318
-0.05980
0.98896
-0.79698
0.66950
-0.65286
-0.94958
1.42171
-0.88570
-0.36526
0.58046
0.86293
-0.91099
-0.15762
-0.71711
1.22002
-1.99044
-0.68708
1.29006
-0.97644
-1.38375
-1.26480
-1.29601
-1.67257
-1.45409
-1.14879
-1.66296
-1.60109
-1.50575
-1.51385
-1.59130
-1.42991
-1.46436
-1.53424
-1.65373
2.22294
1.64050
1.58159
1.60293
1.13804
1.30753

Robust Std Err

9.60317
0.42105
1 .03896
0.74496
0.43503
0.46578
0.66673
1 .12551
0.96059
0.37470
0.26688
0.48519
0.94429
0.59145
0.12646
0.18161
0.05746
0.08989
0.15254
0.13796
0.12405
0.09526
0.05250
0.04260
1.15563
0.38395
0.82843
0.90749
1 .27345
1 .48028
1.65556
1.95984
1 .99851
2.21879
2.32786
2.48762
2.42938
2.45295
2.61353
2.65734
2.64491
0.56739
0.56131
0.76637
1.45539
4.78334
3.53034

Robust t-stat

-1.38004
3.63610
-2.43277
2.07473
0.68756
2.56819
-2.39753
-0.07049
1.32735
-2.65540
2.83117
-1.48366
-1.10420
2.63162
-1.26953
-0.37086
1.87931
1.85918
-1.11477
-0.20092
-0.89345
1.86150
-5.36050
-2.22075
1.41081
-1.39848
-1 .34403
-1.45216
-1.35045
-1.79413
-1.56458
-1.15436
-1.79605
-1.66709
-1.56306
-1.52545
-1.68387
-1.51825
-1.50100
-1.57452
-1.74874
2.85759
2.09083
1.88047
1.71644
1.17922
1.37843

Table 8.7 40110867ng 14, PHS Lag 15

Num of Obs

Log Likelihood
Sum of Sqr Resid
R-Squared

Sigma Squa'ed

Parameter

one
curird
Ig1 ird
IgZird
lg3ird
Ig4ird
IgSird
IgSird
lg7ird
l98ird
IgQird
lg1 Oird
lg1 1 ird
lg1 2ird
lg1 3ird
lg1 4ird
CUM
harem
|a9r092
lagreg3
lagreg4
lagr095
lagre96
lagreg7
IagregS
lagreg9
PHS 15
y79
y80
y81

giu

res

119

90.90157
1 29.37992

0.77054
0.85197

Estimate

0.14760
1.18817
-2.28047
1 .63929
0.02083
1.14889
-1 .78536
0.27812
1 .04768
-1.03038
0.74237
0.76909
-1 .17704
1 .92368
0.23783
0.04975
0.17698
0.09382
0.05876
0.15790
0.20435
0.05617
0.13363
0.16615
0.32172
0.12534
1.17209
0.34948
0.6563)
0.5852
0.68141
-1 .39859
-1 .13686
0.53664
-1 .66302
-1 .57337
-1 .34334
-1 .43101
-1 .68380
-1 .35843
-1 .49529
-1 .73125
-2.17736
1 .45506
1.13054
1.13643
1 .54721
3.20605
2.82824

150

PoissonStdErr

7.74675
0.97789
1 .35333
1.31338
1.32602
1.29618
1.32314
1.31809
1 .28783
1 .27786
1.17137
1.11853
1.12055
1.27742
1.17846
0.92015
0.18561
0.18410
0.17955
0.19264
0.18646
0.17598
0.15245
0.14398
0.14489
0.13968
1 .01325
0.48777
0.58310
0.61270
0.67683
0.78062
0.83772
0.86089
0.88566
0.91755
0.94408
0.96065
0.98424
1 .03888
1.09698
1.15693
1 .23115
0.64CBS
0.82768
0.71928
1.1209
3.72663
2.65649

Poisson t-stat

-1.05174
1.21504
-1 .68508
1.24814
0.01571
0.88637
-1.34934
0.211%
0.81352
08(333
0.63377
0.68759
-1 .05042
1.50591
0.20182
0.05407
0.95346
0.50965
0.32727
0.81%3
-1.09599
0.31917
0.87660
1.15394
-2.2053
0.89735
1.15677
0.71648
-1.12657
0.95514
-1 11575
-1.79164
-1.35709
0.62335
-1.87771
-1 .71475
-1.4290
-1 .48962
-1.71076
-1.30759
-1.36309
-1 .49643
-1 .76855
2.2729
1.36590
1.57997
1.37887
0.86031
1 .W65

RobustStd Err

2.87593
0.70003
1 .29656
0.75374
0.63257
0.53177
0.55585
1 .18991
1 .1052
0.40464
0.33471
0.49002
0.97076
0.69324
1 .2204
1 .02329
0.08888
0.15855
0.04236
013801
0.14782
0.13546
0.143%
0.07684
003142
0.05331
0.29357
0.24348
0.44814
0.35249
0.27533
0.45676
0.27926
0.503%
0.33010
0.47532
0.56413
0.46764
0.39292
0.41547
0.52023
0.65664
0.56398
0.19888
0.2794
0.1626
0.42390
1 .26424
0.91642

Robust t-stat

-2.83303
1 .67814
-1 .75885
2.17487
0.03294
2.16050
0.21197
0.23373
0.94794
-2.54641
2.21799
-1 .56951
-1 .21249
2.77493
0.19462
0.04862
-1 .99128
0.591 77
1 .38728
1 .61097
-1 .38249
0.41463
0.92866
2.16241

-1 0.2402

-2.35105

3.99259
-1 .43538
-1 .46583
-1 .66025
-2.47483
0.06201
4.07097
-1 .06497
5.03790
0.31010
-2.38124
0.06005
4.28531
0.26961
-2.87429
-2.63655
0.86073

7.31635

4.95987

7.00379

3.64993

2.53596

3.08620

Table an -Industry Lag 14,9113 Lag 16

151

Num of Obs 119

Log Likelihood 91.19025

Sum of Sqr Resid 128.85904

R-Squared 0.77147

Sigma Squared 0.8436

Parameter Estimate Poisson Std Err
one 40.43257 8.23952
curird 1.13276 0.98286
|g1ird 235071 1 .36CB2
1921rd 1.74460 1.32201
lg3ird 0.06326 1.33129
Ig4ird 1.13036 1.29805
lgSird -1 .82372 1 .32520
IgBird 0.29703 1 .31986
lg7ird 1 .09435 1 .28297
Igaird -1.11911 1.27902
ngird 0.70831 1.16940
ig10ird 0.78001 1.11597
1911ird -1.17824 1.11249
1912ird 1.88801 1.27307
lg13ird 0.21392 1.17451
Ig14ird 0.07237 0.93561
curreg 0.16918 0.18576
lagreg1 0.08309 0.18446
IagregZ 0.06281 0.17966
lagreg3 0.16432 0.19328
lagreg4 0.19821 0.18661
lagregS 0.05570 0.17640
lagreg6 0.13423 0.15229
lagreg7 0.15991 0.14408
lagrege 0.31800 0.14482
lagregQ 0.12003 0.13949
PHS 16 1.54099 1.10687
y79 0.44100 0.50046
y80 0.86963 0.64103
y81 0.84877 0.69863
y82 0.99019 0.78320
y83 -1.77162 0.91164
y84 -1 .58874 1.01015
y85 -1 .05943 1.07603
y86 -2.18728 1.09712
y87 -2.08350 1.11443
y88 -1.89390 1.16128
y89 -1.95991 1.16398
y90 -2.21981 1.18642
y91 -1.84057 1.20461
y92 0.08559 1 .30579
y93 -2.30827 1.36432
y94 -2.74053 1 .42389
cns 1.65465 0.68434
oar 1.36014 0.86449
ant 1.35713 0.76233
giu 1.85727 1.18900
der 4.50596 4.05068
res 3.84418 2.94711

Poisson t-etat

-1.2%16
1.15252
-1.72806
1.319%
0.04752
0.87081
-1.37618
0.2504
0.85298
0.87497
0.60570
0.6%95
-1 .05910
1.48304
0.18214
0.07735
0.91076
0.45044
0.34960
0.85019
-1.06215
0.31580
0.88144
1.109%
0.19593
0.86055
1.3920
0.88120
-1.35%3
-1.21492
-1 .26429
—1 .94332
-1 .57 278
0.98457
-1.993%
-1.86956
-1.63086
-1 .68380
-1.87101
-1 .52794
-1 .58187
-1.69189
-1 .92467
2.41788
1.57335
1.78024
1.56205
1.11240
1.30439

Robust Std Err

3.15896
0.71092
1 .28099
0.76883
0.62928
0.52466
0.56067
1 .17478
1 .08230
0.38493
0.36%8
0.49686
0.954%
0.69697
1 .19129
1 .(XJ941
0.08815
0.15754
0.04109
0.1%42
0.14997
0.13935
0.14091
0.07623
0.03318
0.05319
0.34473
0.24%1
0.47987
0.40708
0.33372
0.4%98
0.37210
0.59812
0.44791
0.53625
0.62%4
0.56827
0.44091
0.48%7
0.61979
0.76181
0.65833
0.21945
0.23651
0.1%95
0.44070
1 .44%9
1 .04398

Robust t-etat

0.30254
1 .59337
-1 .83508
2.2691 8
0.10053
2.15448
0.25276
0.25284
1 .01 1 14
-2.90729
1 .96707
-1 56%8
-1 .23419
2.70888
0.17957
0.07169
-1 .91935
0.52741
1 .52851
1 .63639
-1 .32165
0.39975
0.95262
2.09780
0.58485
0.25680
4.47013
-1 .78826
-1 .8124
0.08503
0.96709
0.79374
4.26970
-1 .77127
4.88325
0.88529
0.01078
0.44890
0.03467
0.82914
0.33273
0.03000
4.16288
7.54013
5.75087
7.10721
4.21436
3.12721
3.6824

152

Table ao-IMustryL-g 14,9113 Lag 17

Num of Obs

Log Likelihood
Sum of Sqr Resid
R-Squared

Sigma Squared

Parameter

one
curird
191 ird
192ird
lg3ird
Ig4ird
lgSird
igSird
lgTird
ig8ird
ig9ird
lg1 Oird
191 1 ird
lg1 2ird
lg1 3ird
lg1 4ird
curreg
lagreg1
lagre92
lagreg3
harem
IagregS

119
91.9928

125.01129

0.77829
0.82484

Estimate

-15.63597

1.02072
0.38800
1.85418
0.15004
1.14100
-1.91591
0.33602
1 .07129
-1 .18778
0%209
0.76293
-1 .16380
1.82159
0.16762
0.24676
0.13642
0.06019
0.06991
0.16989
0.17908
0.04777
0.1243
0.15334
0.31928
0.11332
2.33693
0.67546
-1.32054
-1 .48316
-1 .69586
0.594%
0.57953
0.18529
0.38240
0.28448
0.12928
0.23862
0.45537
0.09786
0.28999
0.67990
4.10218
2.08198
1.82986
1.85%1
2.58407
7.37585
6.04764

PoissonStdErr

9.04459
0.99206
1.36630
1.32980
1.33608
1.33233
1.32488
1.32624
1.28178
1.27232
1.16124
1.11092
1.09937
1.26272
1.16713
0.94359
0.18732
0.18492
0.17930
0.19407
0.18797
0.17688
0.15288
0.14420
0.14478
0.13910
1.23232
0.52829
0.72645
0.85184
0.95240
1.1%30
1.26585
1.38281
1.43332
1.44720
1.49931
1.52346
1.52174
1.54484
1.61144
1.72931
1.77749
0.74291
0.9022
0.82971
1.30831
4.51505
3.34453

Poisson t-stat

-1 .72877
1 .02888
-1 .74779
1 .39433
0.1 1230
0.87612
-1 .44610
0.25337
0.83578
0.93355
0.57016
0.68676
-1 .05861
1 .44259
0.14362
0.26152
0.72826
0.32552
0.38992
0.87541
0.95273

, 0.27005

0.80080

1 .06342
0.20532
0.81488

1 .89637
-1 .27858
-1 .81779
-1 .741 13
-1 .78061
0.33901
0.03779
-1 .58032
0.35984
0.26954
0.08714
0.12583
0.270%
00530
0.04165
0.12796
0.30785

2.80249

2.02817

2.2969

1 .97512

1 .63362

1 8%2

Robust Std Err

3.72652
0.71%2
1 .23067
0.79024
0.60590
0.51480
0.58545
1 .14063
1 .04135
0.38%6
0.37129
0.48527
0.91760
0.69658
1 .14625
0.98671
0.08475
0.15262
0.03729
0.10490
0.15250
0.14179
0.14052
0.07095
0.03589
0.05061
0.46741
0.25539
0.51613
0.50219
0.42588
0.52502
0.52899
0.75508
0.66963
0.67713
0.75394
0.71737
0.58778
0.6%82
0.80177
0.97402
0.9026
0.29442
0.27772
0.26815
0.49973
1 .79%0
1 .3206

Robust t-stat

4.19586
1 .42156
-1 .94041
2.34634
0.24763
2.21638
0.27255
0.29459
1 .02875
0.04827
1 .78319
-1 .57219
-1 .26832
2.61503
0.14623
0.25009
-1 .60959
0.39441
1 .87478
1 .61945
-1 .17431
0.33688
0.87128
2.16121
0.89640
0.23908
4.99976
0.64486
0.55854
0.95339
0.98203
4.94205
4.87635
0.89409
0051 13
4.85062
4.15058
4.51455
0.87868
4.64574
4.10338
0.77807
4.54656
7.07137
6.58888
6.89923
5.17098
4.10544
4.57442

153

Table e10 - industry Lag 14,9113 Leg 13

Num of Obs

Log Likelihood
Sum of Sqr Resid
R-Squared
Sigma Squared

Parameter

one
curird
lg1 ird
ngird
lg3ird
lg4ird
lg5ird
lg6ird
lgTIrd
lg8ird
ngird
lg1 Oird
Ig1 1ird
lg1 2ird
lg1 3ird
lg1 4ird
curreg
lagreg1
|a9re92

harem
lagr995
Iagre96
lagreg7

PHS 18

119

91 .%163
125.45525

0.7775
0.81965

Estimate

-17.03649

1.23589
0.58444
1.844%
0.21838
1.179%
-1 .83132
0.14208
1.19965
-1.14207
0.75421
0.76149
-1 .12113
1 .77773
0.311370
0.21574
0.12369
0.03246
0%186
0.17391
0.16133
0.0269
0.11442
0.15753
0.31823
0.11683
2.49%3
0.75%4
-1 .50139
-1 .76582
0.12567
0.03923
0.0%24
0.79147
4.08416
4.%116
0.91344
4.04926
4.32187
0.91777
4.14306
4.43628
4.97599
2.14339
1.%3%
1.97437
2.84743
8%059
6.62497

PoissonStdErr

9.93585
0.97579
1.38251
1.32953
1.34025
1.29877
1.31492
1.32113
1.28491
1.26731
1.15299
1.10151
1.%716
1.25417
1.15088
0.93619
0.18813
0.18753
0.1%26
0.194%
0.18844
0.17739
0.15338
0.14459
0.14501
0.13927
1.35528
0.5562
0.81117
1.%459
1.18402
1.34932
1.53911
1.725%
1.82153
1.87347
1.93250
1.96856
1.98943
1%
2.%821
2.13771
2.23525
0.79042
0.91018
0.91817
1.44502
4.98715
3.7428

Poisson t-stat

-1 .71465
1 .2%55
-1 88339
1 .38740
0.16294
0%840
-1 .39273
0.10754
0.93365
0.901 18
0.65414
0.%132
-1 .02185
1 .41746
0.26389
0.23045
0.65748
0.17310
0.5%62
0.%599
0.85612
0.127%
0.74598
1 .08949
0.19463
0.83888
1 .83728
-1 .3%32
-1 .85088
-1 .75775
-1 .79530
0.25241
-1 .99351
-1 .61824
0.24216
0.16772
0.02507
0.05%6
0.17241
-1 .97188
0.%321
0.07524
0.2615
2.71170
1 .98103
2.15%3
1 .97052
1.61627
1 .77030

RobustStd Err

5.87205
0.65%7
1.20752
0.81178
0.5552
0.51%5
0.56520
1.0%44
1.%4%
0.37404
0.38442
0.49393
0.98403
0.69550
1.07347
1.%572
0.08484
0.15655
0.%9%
0.10161
0.15113
0.14231
0.13875
0.07737
0.0402
0.04850
0.684%
0.29%2
0.%707
0.5%35
0%085
0.68794
0.85362
1.04401
1.01510
1.04381
1.05373
1.16527
1.01971
1.%3%
1.27388
1.34%2
1.3023
0.313%
0.31836
0.44945
0.75507
2.%321
1.86931

Robust t-stat

0.%128
1.88119
0.14028
2.2728
0.39332
2.29688
0.24014
0.13030
1.1%92
0.05334
1.96195
-1 .54171
-1.2%31
2.55%4
0.28292
0.21451
-1 .45799
0.20736
2.31505
1.71151
-1 .06749
0.15942
0.82462
2.03618
-7.91183
0.40%6
3.63567
0.55%1
0.47315
0.10146
0.1210
4.417%
0.59440
0.67379
4.02339
0.89071
0.713%
0.47495
4.23835
0.58344
0.25233
0.29%5
0.82113
6.83052
5.%363
4.39283
3.77106
3.02%4
3.54408

Table 3.11 - Industry Leg 14,9113 Lag 19

154

Num of Obs 119

Log Likelihood 91 .62458

Sum of Sqr Resid 127.25382

R-Squared 0.77431

Sigma Squared 0.82386

Parameter Estimate Poisson Std Err
one 46.11767 10.34160
curird 1 .39385 0.96560
lg1ird 0.46500 1 .37263
ngird 1.59014 1.31422
Ig3ird 0.24547 1 .34035
lg4ird 1 .21546 1 .29829
IgSird 4 .67098 1 .30557
lgGird 0.01540 1.32229
Ig7ird 1 .18550 1 .28314
198ird 0.98330 1 .25869
ngird 0.82051 1.15129
lg10ird 0.69222 1 .10140
lg11ird 4.14519 1.09726
lg12ird 1 .84273 1 .25399
lg13ird 0.39390 1 .14535
lg14ird 0.00249 0.90628
curreg 0.13162 0.18750
lagreg1 0.03491 0.18789
lagregZ 0.11930 0.18375
lagreg3 0.18001 0.19463
lagreg4 0.15725 0.18816
lagreg5 0.00979 0.17789
lagregG 0.09220 0.15624
lagreg7 0.16114 0.14486
lagrega 0.30668 0.14609
lagregQ 0.1 1058 0.13948
PHS 19 2.24172 1.35796
y79 0.73020 0.56909
y80 -1 .44586 0.84286
y81 -1 .73393 1 .07755
y82 0.15597 1 .31522
y83 0.15413 1.54379
y84 0.11338 1.71676
y85 0.83712 1 .92782
y86 4.22023 2.08425
y87 4.27725 2.18816
y88 4.16941 2.27918
y89 4.32351 2.32740
y90 4.62446 2.36300
y91 4.26988 2.38664
y92 4.44010 2.44092
y93 4.75124 2.52443
y94 -5. 19256 2.56489
cns 1 .98243 0.77503
car 1 .51830 0.83617
ant 1.86018 0.94627
giu 2.80734 1 .53762
der 7.3861 1 5.1 1447
res 6.13349 3.85676

Poisson t-stat

4.55853
1.44350
4.79582
1.2%95
0.18314
0.93620
4.27988
0.01165
0.92391
0.78121
0.712%
0.62850
4.043%
1.46950
0.34392
0%275
0.70196
0.185%
0.6492
0.92490
0.83576
0.05502
0.59014
1.11236
0.%932
0.79281
1.65080
4.283(B
4.71542
4 .6%14
4.63924
0.04310
4.81353
-1 .47168
0.02482
4.95472
4.82935
4.85765
4.95703
4 .7%07
4.81%2
4.88211
0.02448
2.557%
1.81579
1.965%
1.82577
1.44416
1.59032

Robust Std Err

8.3182
0.61732
1.21165
0.762%
0.50706
0.50%8
0.48501
1.068%
1 .00276
0.35987
0.36401
0.46983
0.%%4
0.65992
1 .04598
1.03815
0.088%
0.16265
0.05520
0%367
0.15399
0.14517
0.14413
0.08323
0.04542
0.05181
1.%417
0.37513
0.75204
0.7%67
1.02733
1.17594
1.32590
1.58149
1 .57326
1.73074
1.77853
1.88462
1.81421
1 .86341
2.02749
2.11519
2.%729
0.47756
0.46311
0.672%
1.15895
3.95%0
2.85928

Robust t-stat

4 .93764
2.25791
0.03441
2.%655
0.48410
2.38%2
0.44528
0.01441
1 .1823
0.73240
2.2541 1
4 .47337
4 .2631 1
2.79238
0.37659
0.00240
4 .4824
0.21463
2.16134
1 .921 %
4 .021 19
0.06743
0.63971
1 .93610
0.7524
0.13455
2.23241
4 .94650
4 .9258
0.21%
0.%861
0.6822
0.34812
4 .79396
0.68247
0.47134
0.34431
0.294%
0.54901
0.29144
0.1%95
0.24625
0.52399
4.151 16
3.27846
2.76428
2.4230
1 .86570
2.14512

Tabb 8.12 - Industry Lag 14, PHS Lag 20

Num of Obs 119
Log Likelihood 90.83429
Sum of Sqr Resid 131 .57881
R-Squared 0.76664
Sigma Squared 0.81429
Parameter

one 43.25281
curird 1 .53099
lg1ird 0.52755
19de 1 .54559
lg3ird 0.2991 1
lg4ird 1 .19621
lgSird -1 .59851
lg6ird 0.07934
lgTird 1 .27504
Igeird 0.99498
ngird 0.75558
lg10ird 0.71985
lg1 1ird -1 .04268
lg12ird 1 .55646
lg13ird 0.16055
Ig14ird 0.06735
curreg 0.10799
lagreg1 0.16713
lagreg2 0.17005
lagregS 0.02772
lagreg4 0.1 1083
lagregS 0.17732
lagregG 0.28141
lagreg7 0.09461
lagregB 1 .63038
lagregQ 0.53695
PHS 20 -1 .1 1344
y79 4 .31782
y80 -1 .71974
y81 0.65581
y82 0.59026
y83 0.26236
y84 0.58942
y85 0.69891
y86 0.63858
y87 0.79473
y88 4.09076
y89 0.72420
y90 0.92292
y91 4.18404
y92 4.62526
y93 1 .62135
y94 1 .17360
cns 1 .441 13
car 2.498%
ant 5.64062
giu 4.86635
der 7.38611
res 6.13349

155

Estimate Poisson Std Err

10.03849
0.94410
1.33%5
1.28736
1.30127
1.26402
1.28582
1.32678
1.2%27
1.24845
1.12858
1.10261
1.%%4
1.%478
0.18127
0.18440
0.18%4
0.193%
0.186%
0.17585
0.15455
0.14534
0.14138
0.13770
1.263%
0.54990
0.%4%
1.04192
1.32695
1.58786
1.78136
1.96933
2.15845
2.31025
2.41646
2.5%67
2.57070
2.%451
2.67%3
2.72711
2.7%86
0.72937
0.71539
0.91119
1.55845
4.95643
3.72179
5.11447
3.85676

Poisson t-stat

4.32020
1.62163
4.88770
1.21159
0.2986
0.94635
4.24318
0.059%
0.98%6
0.79698
0.%950
0.65286
0.94958
1 .42171
0.88570
0.36526
0.5%46
0.86293
0.91%9
0.15762
0.71711
1.2%2
4.99044
0.68708
1.29%6
0.97644
4.38375
4.26480
4 .29%1
4.67257
4.454%
4.14879
4 .%296
4 .%1%
4.50575
4.51385
4.59130
4.42991
4.46436
4.53424
4.65373
2.2294
1.84050
1.58159
1.%293
1.13%4
1.30753
1.44416
1.59032

Robust Std Err

9.6%17
0.42105
1 .03%6
0.74496
0.43503
0.48578
0.6%73
1 .1 2551
0.96059
0.37470
0.2%88
0.48519
0.94429
0.59145
0.12646
0.18161
0.057 46
0.08989
0. 1 5254
0.13796
0.12405
0.%526
0.05250
0.042%
1 .15563
0.38395
0.82843
0.90749
1 .27345
1 .4%28
1 .65556
1 .95984
1 9%51
2.21879
2. 32786
2.48762
2.42938
2.45295
2.61353
2.65734
2.64491
0.56739
0.561 31
0.7%37
1 .45539
4.78334
3.53034
3.95%0
2.85928

Robust t-stat

4 .38%4
3.63610
0.43277
2.07473
0.68756
2.56819
0.39753
0.07049
1.32735
0.65540
2.83117
4.483%
4.10420
2.63162
4.26953
0.37086
1.87931
1.85918
4.11477
0.20%2
0.%345
1.86150
-5.3%50
0.2075
1.41081
4.39848
4.34403
4.45216
4.35045
4.79413
4.56458
4.15436
4 .79%5
4 .%7%
4.56306
4.52545
4.68387
4.51825
4.50100
4.57452
4.74874
2.85759
2.%083
1.8%47
1.71644
1.1792
1.37843
1.86570
2.14512

156

Table 8.13 - Mushy L89 12, PHS Nested 12,17,22

Num of Obs 119
Log Likelihood 92.15151
Sum of Sqr Resid 124.23083
R-Squared 0.77967
Sigma Squared 0.81582
Parameter Estimate
one 4 6.57264
curird 1 .13937
lg1ird 0.40982
ngird 1 .85343
lgSird 0.22204
lg4‘rd 1 .1 1641
lg5ird 4 .77030
lgGird 0.21469
lg7ird 1 .13635
lg8ird 4 .16502
ngird 0.70150
lg10ird 0.72865
lg1 1 ird 4 .16079
Ig12ird 1 .76550
curreg 0.12725
lagreg1 0.06401
lagregZ 0.08524
lagreg3 0.17551
lagreg4 0.15880
13ng 0.03691
lagregG 0.10397
lagreg7 0.16160
lagreg8 0.30104
lagregQ 0.10120
PHS 12 0.46237
PHS 17 2.40%5
PHS 22 0.44051
y79 0.74499
y80 4 .46200
y81 4 .71579
y82 0.07712
y83 0.10084
y84 0.18227
y85 0.881 16
y86 4.15232
y87 4.14191
y88 4.00751
y89 4.14820
y90 4.43738
y91 4.09781
y92 4.31263
y93 4.71883
y94 0.15364
cns 2.06859
car 1 .69323
ant 1 .90766
giu 2.83321
der 7.77298
res 6.51756

PoissonStdErr

9.50335
1.00743
1.34115
1.32996
1.3249
1.27144
12%
1.34506
1.285%
1.24347
1.11534
1.10284
1.%275
1.10705
0.183%
0.18304
0.181%
0.19375
0.1%91
0.17765
0.15356
0.14445
0.14515
0.14%9
1.31245
1.48615
0.81856
0.53886
0.750%
0.92%7
1.14588
1.404%
1.61743
1.%370
1.95261
2.04427
2.11345
2.21075
2.3%93
2.34751
2.42145
2.51435
2.57437
0.77300
0.%830
0.92714
1.43718
4.68430
3.43154

Poisson t—sut

4 .74387
1 .13%7
4 .79682
1 393%
0.16790
0.878%
4 .36513
0.15961
0.88432
0.93691
0.62896
0.6%70
4 .0626
1 .59477
0.%230
0.34973
0.46941
0.905%
0.83617
0.20774
0.677%
1 .1 1873
0.074%
0.7237
0.3529
1 .62053
0.53815
4 .38253
4 .94762
4 .86302
4 .81268
0.20750
4 .96749
4 .59736
0.12655
0.0261 1
4 .%619
4 .87%8
4 92%4
4 .745%
4 .78101
4 .87676
0.%190
2.67%5
1 .95”
2.05758
1 .97136
1 .65937
1 .%931

RobustStd Err

7.02428
0.57614
1 .%937
0.77%2
0.43465
0.43303
0.71762
1.1%
0.98346
0.37%8
0.2512
0.52%7
0.%825
0.647%
0.116%
0.1682
0.04404
0.%853
0.153%
0.14424
0.124%
0.%011
0.05981
0.05390
0.677%
0.928(5
0.47659
0.316%
0.%041
0.74425
0.%591
1 .14328
1 32%
1 .61648
1 .65633
1.72116
1 .77050
1.91374
1 .8%30
1 .9%44
2.07542
2.191%
2.17501
0.38270
0.38627
0.49541
1 .02323
3.27776
2.52827

Robust t-stat

0.35934
1 977%
0.38744
2.3823
0.51%6
2.57812
0.4%91
0.18495
1 .15546
0.%529
3. 1 1612
4 .39%3
4 .2928
2.72860
4 .%858
0.38053
1 .93536
1 .78120
4 .%1%
0.25587
0.83424
2.01706
0.03323
4 .87753
0.%295
2.%%7
0.92429
0.35%8
0.1 1757
0.30540
0.1%%
0.7123
0.40700
4 782%
0.50694
0.4%47
0.2%49
0.16759
0.41254
0.14720
0.07795
0.1 5371
0.36947
5.40530
4.3%57
3.85%4
2.7%%
2.37143
2. 57788

 

157

Tabb 8.14 - Industry Lag 12, PHS Maud 9.17.25

Num of Obs 119
Log Likeiihood 92.12054
Sum of Sqr Resid 125.45063
R-Squared 0.77751
Sigma Squared 0.81991
Parameter Estimate
one 4 6.02741
curird 1 .07396
lg1ird 0.36723
ngird 1 .80779
lg3ird 0.14869
ig4ird 1 .1 1822
I95ird 4 .78944
igGird 0.30525
Ig7ird 1 .0732
l98ird 4 .18800
IgQird 0.7128
lg10ird 0.74095
Ig1 1ird 4 .16099
lg12ird 1 .75856
curreg 0.13137
lagreg1 0.06604
lagregZ 0.08279
lagregS 0.17849
lagreg4 0.16472
lagregS 0.04074
lagregG 0.1 1 156
lagreg7 0.16105
lagregB 0.30506
lagregQ 0.10979
PHS 9 0.07222
PHS 17 2.07199
PHS 25 0.21326
y79 0.63680
y80 4 .24188
y81 4 .37886
y82 4 .63409
y83 0.56302
y84 0.57929
y85 0.21574
y86 0.45283
y87 0.44877
y88 0.32858
y89 0.44767
y90 0.72268
y91 0.39616
y92 0.62138
y93 4.01339
y94 4.42852
cns 2.0851 1
car 1 .76651
ant 1 .94580
giu 2.67696
der 7.51364

6.15821

Poisson Std Err

10.17516

0.%299
1.33659
1.32527
1.30439
1.27%6
1.30336
1.32584
1.2%19
1.24768
1.11994
1.103%
1.%057
1.11336
0.18405
0.18293
0.181%
0.19373
0.18%4
0.17%9
0.15327
0.14377
0.14319
0.13796
1.33287
1.19470
0.40442
0.52583
0.7%51
0.%271
0.90537
1.07664
1.2854
1.3%18
1.405%
1.45725
1.49764
1.550%
1.61065
1.66339
1.75142
1.85350
1.91478
0.95%9
1.%858
1.10296
1.41714
5.23746
3.48463

Poisson t-stat

4 .57515
1 .%154
4 .771%
1 .36410
0.11399
0.8%44
4 .37295
0.23023
0.83833
0.95217
0.636%
0.6712
4 .%457
1 .57951
0.71378
0.361%
0.45727
0.92136
0.87177
0.23134
0.72783
1 .12019
0.13053
0.79581
0.05418
1 .73432
0.52732
4 .21 1%
4 .75%2
4 .71775
4 .%4%
0.38058
0.%948
4 .69246
0.45748
0.36663
0.2255
0.2417
0.31 129
0.04171
0.%769
0.16530
0.31281
2.19279
1 .6277
1 .76416
1 .88%9
1 .43459
1 .76725

RobustStd Err

4.69136
0.632%
1 048%
0.82163
0.4%28
0.44925
0.77816
1 .18%7
1 .(XBOB
0.39466
0.%%1
0.50838
0.8%%
0.707%
0.121%
0.16744
0.05255
0.10%5
0.155%
0.14767
0.12597

. 0.07599

0.05521
0.04581
0.66274
0.73637
0.19431
0.26652
0.5%28
0.59267
0.59969
0.69953
0.70532
0.93731
0.90468
0.8852
0.92965
0.97414
0.84%1
0.918%
1 .%593
1.18670
1.14341
0.45870
0.49478
0.39674
0.62679
2.43%2
1 .71976

Robust t—stat

0.41637
1 .%717
0.258%
2.2%26
0.303%
2.4%08
0.29957
0.25848
1 .06674
0.01020
3.%673
4 .45746
4 3%32
2.487%
4 .07%0
0.39440
1 .57 544
1 .66%9
4 .%271
0.27586
0.88559
2.1 1942
0.52586
0.39688
0.1%97
2.81378
1 .%749
0.3%26
0.10746
0.32652
0.72488
0.66394
0.65689
0.36393
0.81662
0.%593
0.5%47
0.53919
4.38523
0.69712
0.39739
0.38198
0.87309
4.54570
3.57029
4.90450
4.27%0
3.0%73
3.5%85

158

Tabb 8.15 - industry Lao 14, PHS W 12,17,22

Num of Obs 119
Log Likelihood 92.37649
Sum of Sqr Resid 124.95949
R-Squared 0.77838
Sigma Squared 0.82629
Parameter Estimate
one 4 7.95168
curird 1 .15163
lg1ird 0.47335
ngrd 1 .85827
Ig3ird 0.26457
Ig4ird 1 .19932
lg5ird 4 .86090
Igeird 0.24814
Ig7ird 1 .19762
lgaird 4 .13653
ngird 0.65821
lg10ird 0.74107
lg1 1ird 4 .17209
lg12ird 1 .76555
Ig13ird 0.24360
lg14ird 0.25846
curreg 0.1 1868
lagreg1 0.04694
lagregZ 0.08339
lagreg3 0.17266
lagreg4 0.14569
lagreg5 0.02464
lagregG 0.1 1566
lagreg7 0.14626
lagreg8 0.30533
lagregQ 0.09225
PHS 14 4.28591
PHS 17 3.48370
HPS 22 0.39480
y79 0.85907
y80 4 .65125
y81 4 .96712
y82 0.35278
y83 0.42796
y84 0.61752
y85 0.39579
y86 4.70549
y87 4.66845
y88 4.56161
y89 4.73950
y90 4.9425
y91 4.61647
y92 4.83989
y93 0.28134
y94 -5.70191
cns 2.17936
car 1 .79238
ant 2.01445
giu 3.05981
der 8.57090
res 7.20723

PoissonStdErr

9.74869
1.02342
1.37212
1.34629
1.34586
1.30794
1.33045
1.35371
1.292%
1.26950
1.15877
1.10587
1 .%368
1.26378
1.16387
0.97233
0.18823
0.18468
0.18152
0.19566
0.1925
0.1%16
0.15592
0.14618
0.14828
0.14178
1.75187
2.30255
0.84338
0.56681
0.81391
1113911
1.21746
1.477%
1.74462
1.96870
2.11167
2.19467
2.28291
2.37393
2.4%63
2.46242
2.538%
2.64513
2.68776
0.754%
0.%314
0.%4%
1.47717
4.81120
3.65840

Poisson t-stat

4 .84145
1 .12527
4 .%258
1 .3%29
0.19658
0.91%5
4 .39870
0.183%
0.92688
0.%526
0.56802
0.67012
4 .07169
1 .39704
0.20%0
0.26582
0.63051
0.25417
0.45938
0.88243
0.75779
0.13675
0.741%
1 .%053
0.05914
0.65%7
0.73402
1 .51297
0.46812
4 .51562
0.02879
4 .94937
4 .93253
0.320%
0.07353
4 724%
0.2832
0.12717
4 .9%15
4 .9%47
0.05104
4 .87476
4 .9%30
4 .99663
0.12144
2.88137
211383
2.25256
2.07141
1 .78145
1 .97%5

RobustStd Err

6.92910
0.642%
1.1%44
0.78644
0.48%5
0.48857
0.53133
1.12336
0.95179
0.36677
0.39691
0.48125
0.89938
0.64959
1.07948
1.02633
0.08641
0.15535
0.%965
0.10142
0.15294
0.15%4
0.14181
0.0%59
0.04513
0.06236
0.97279
1.15875
0.53669
0.35647
0.73282
0.75448
1.%266
1.15%3
1.44641
1.7(1155
1.76856
1.87512
1.90585
2.%294
1.97356
2.%445
2.27391
2.3682
2.335%
0.3%16
0.25323
0.53307
1.%591
3.20187
2.42966

7.07795
3.77%6
2.95373
2.67684
2.96636

 

159

Table 8.16 - Nutty Log 14, PHS Nested 9.17.25

Num of Obs 119
Log Likelihood 92.13606
Sum of Sqr Resid 1 26.2564
R-Squared 0.77608
Sigma Squared 0.84255
Parameter Estimate
one 4 5.98923
curird 1 .10184
191 ird 0.39060
ig2ird 1 .77544
lg3ird 0.14984
lg4ird 1 .16995
lgSird 4 .82862
lgSird 0.3121 1
lg7ird 1 .071 56
lgBird 4 .16074
ngird 0.7031 5
191 Oird 0.74939
lg11ird 4.17814
lg12ird 1 .84772
191 3ird 0. 1 9959
Ig14ird 0.1 2063
curreg 0.13730
lagreg1 0.06244
lagreg2 0.08284
lagregS 0.18100
lagreg4 0.16564
lagreg5 0.03763
lagreg6 0.1 1471
lagreg7 0.1 5899
lagre98 0.30772
IagregQ 0.10689
PHS 9 0.07550
PHS 17 2.07086
PHS 25 0.21061
y79 0.63479
y80 4 .2361 1
y81 4 .38103
y82 4 .63709
y83 0.56339
y84 0.56919
y85 0. 20694
y86 0.45776
y87 0.43%5
y88 0.3%29
y89 0.45147
y90 0.71842
y91 0.38864
y92 0.60981
y93 0.99999
y94 4.42214
cns 2.08676
car 1 .77052
ant 1.94693
giu 2.67246
der 7.50740
res 6.1 5053

Poisson Std Err

10.63423

1.%852
1.36%1
1.34458
1.33688
1.30381
1.33742
1.32736
1.28%7
1.27539
1.1%21
1.1%12
1.%929
1.26438
1.17183
0.9729
0.18747
0.18472
0.181%
0.19529
0.1%%
0.17773
0.15475
0.14420
0.146%
0.13982
1.351%
1.39086
0.41851
0.53%?
0.75403
0.%3%
0.99790
1.155%
1.32073
1.44124
1.51079
1.54492
1.%567
1.64371
1.67290
1.71274
1.8%54
1.91470
1.96581
0.96310
1.15833
1.10794
1.47513
5.47754
3.73420

Poisson t-stat

4 .5%56
1.%253
4.74750
1.32044
0.112%
0.%733
4.36727
0.23514
0.83515
0.91011
0.%139
0.67627
4.07173
1.46136
0.17%3
0.12406
0.73240
0.33%1
0.45748
0.92684
0.87325
0.21172
0.74126
1.10257
0.10634
0.76446
0.05586
1.48%1
0.50323
4.18218
4.63934
4.54598
4.64054
0.21921
4.94528
4.53127
0.28871
0.2%0
0.06101
0.%981
0.2274
4.97849
0.%484
0.%910
0.24952
2.16671
1.52851
1.75726
1.81167
1.37058
1.647%

Robust Std Err

4.11581
0.72101
1.23494
0.83412
0.58%3
0.53218
0.%7%
1.13546
1.01361
0.40307
0.3%55
0.45704
0.%193
0.%172
1.173%
1.01527
0.09035
0.15%0
0.05455
0.11279
0.15113
0.15772
0.14884
0.%435
0.044%
0.05496
0.59764
0.54243
0.2338
0.26211
0.54637
0.53749
0.46779
0.55846
0.56841
0.78282
0.71854
0.75850
0.79617
0.79648
0.67439
0.77353
0.92927
1.05239
0.98106
0.46304
0.52882
0.3%61
0.54867
2.27796
1.5262

Robust t-stat

0.88483
1.52819
4.93581
2.12852
0.25820
2.19839
0.0%58
0.27488
1.05716
0.87978
1.90272
4.63965
4.320%
2.67119
0.17011
0.11881
4.51972
0.41542
1.51851
1.%473
4 .%%0
0.23858
0.77065
2.47064
0.99341
4.94468
0.12634
3.81772
0.942%
0.42185
0.26241
0.5%41
0.49963
4.59%8
4.51993
0.81920
4.8123
4.52334
4.15651
4.33339
-5.51375
4.3%74
0.88459
0.8%87
4.50749
4.50663
3.348%
4.%716
4.87083
3.29567
4.03944

160

Table 8.17 - Industry Lag 12, PHS Lag 15 (10% Depreciation)

Num of Obs 119
Log Likelihood 90.95091
Sum of Sqr Resid 128.9766
R-Squared 0.77126
Sigma Squared 0.83023
Parameter Estimate
one 4 0.91476
curird 1 .15814
Ig1ird 0.32845
ngird 1 .67713
IgBird 0.09544
lg4ird 1 .10505
lgSird 4.78012
lgGird 0.26553
lg7ird 1 .06561
lgBird 4.10441
ngird 0.68592
Ig10ird 0.76637
Ig11ird 4.14414
ig12ird 1.78099
cu rreg 0.16976
lagreg1 0.09002
lagreg2 0.06449
lagreg3 0.15772
lagreg4 0.20236
lagregS 0.06025
lagreg6 0.13094
lagreg7 0.16903
IagregB 0.30930
iagregQ 0.12142
PHS 1 5 1 .50200
y79 0.40782
y80 0.77840
y81 0.76025
y82 0.90072
y83 4 .65155
y84 4 .44064
y85 0.89868
y86 0.03542
y87 4 .98985
y88 4 .80748
y89 4 .89399
y90 0.15773
y91 4 .84184
y92 0.00666
y93 0.26822
y94 0.71248
cns 1 .6238
car 1 .34781
ant 1 .33773
giu 1 .86643
der 4.44264
res 3.79072

Poisson Std Err

7.94731
0.96259
1.32797
1.29882
1.29564
1.26067
1.29367
1.31745
1.28613
1.25807
1.13208
1.11497
1.10908
1.12820
0.18204
0.18279
0.18006
0.19226
0.18605
0.17424
0.15097
0.14360
0.13941
0.13700
1.06714
0.49536
0.60383
0.64508
0.72684
0.85416
0.92523
0.94193
0.99763
1.03770
1.04904
1.08616
1.13803
1.20387
1.26566
1.32396
1.41052
0.68044
0.77935
0.77427
1.12456
3.78859
2.69852

4.37339
1.20314
4.75339
1.29127
0.07367
0.87655
4.37603
0.20155
0.82854
0.87786
0.60589
0.68734
4.03162
1.57861
0.93251
0.49246
0.35815
0.82035
4 .08766
0.34580
0.86733
1.17709
0.21859
0.88626
1.40751
0.82328
4.28912
4.17854
4.23922
4.93355
4.55706
0.95408
0.04027
4.91756
4.72298
4.74375
4 .89603
4.52994
4.58547
4.71321
4.92303
2.38429
1.72940
1.72773
1.65970
1.17263
1.40474

Poisson t-stat Robust Std Err

3.66732
0.57500
1 . 12025
0.74274
0.55878
0.45044
0.76228
1 .21405
1 .08098
0.42079
0.21813
0.50874
0.93813
0.66474
0.12195
0.17520
0.04317
0.09996
0.15326
0.13038
0.12291
0.08149
0.04212
0.03838
0.39970
0.23495
0.47007
0.40540
0.35122
0.50959
0.33684
0.59504
0.48478
0.52424
0.61960
0.58077
0.48888
0.53230
0.64062
0.78134
0.74126
0.24671
0.30372
0.20742
0.50547
1 .66626
1 .23895

Robust t-stet

0.97622
2.01415
0.07851
2.25803
0.17081
2.45324
0.33526
0.21872
0.98578
0.62462
3.14454
4 .50640
4 .21960
2.67924
4 .39201
0.51378
1 .49388
1 .57790
4 .32032
0.46212
4 .06528
2.07416
~7.34401
0.16357
3.75780
4 .73579
4 .65593
4 .87533
0.56451
0.24093
4.27688
-1 .51028
4.19864
0.79570
0.91717
0.261 14
4.41361
0.46014
0.13236
0.90298
0.65927
6.57602
4.43768
6.44932
3.69250
2.66624
3.05963

161

Table 8.18 - Industry Lag 12, PHS Lag 16 (10% Depreciation)

Num of Obs 119
Log Likelihood 91 .27322
Sum of Sqr Redd 127.83786
R-Squared 0.77328
Sigma Squared 0.82032
Parameter Estimate
one 4 3.1 3581
curird 1 .1 1907
Ig1ird 0.36109
ngird 1 .77761
lg3ird 0.09465
lg4ird 1 .08091
lgSird 4 .7771 1
lgGird 0.27264
lg7ird 1.11118
lgBird 4.16437
IgQird 0.70596
lg10ird 0.77219
lg11ird 4.15649
Ig12ird 1.78001
curreg 0.16109
lagreg1 0.08236
lagreg2 0.06975
IagregB 0.16652
lagreg4 0.19178
lagregS 0.05527
lagre96 0.12773
lagreg7 0.16285
lagreg8 0.31000
lagregQ 0.1 1905
PHS 16 1.86008
y79 0. 50887
y80 4 .00585
y81 4 .03783
y82 4 .23492
y83 0.06714
y84 4 .94308
y85 4 .46641
y86 0.62191
y87 0.57925
y88 0.43209
y89 0.50734
y90 0.79801
y91 0.44181
y92 0.69251
y93 0.96370
y94 0.40410
cns 1 .821 90
car 1 .52922
ant 1 .56725
giu 2.17316
der 5.68148
res 4.77353

Poisson Std Err

8.35885
0.96557
1.33056
1.30729
1.29974
1.26254
1.29381
1.31818
1.27986
1.25321
1.12726
1.11060
1.10171
1.12077
0.18212
0.18276
0.18003
0.19281
0.18631
0.17486
0.15100
0.14373
0.13941
0.13683
1.14088
0.50929
0.66071
0.72465
0.82880
0.98588
1.09739
1.14785
1.20717
1.24374
1.27253
1.30038
1.35956
1.39845
1.49846
1.55677
1.63369
0.71588
0.79324
0.81353
1.18360
4.04225
2.93894

Poisson t-stat

4.57149
1.15897
4.77450
1.35977
0.07282
0.85614
4.37355
0.20683
0.86821
0.92912
0.62626
0.69529
4.04973
1.58820
0.88449
0.45063
0.38742
0.86364
4.02935
0.31610
0.84590
1.13307
0.22370
0.87007
1.63038
0.99919
4.52238
4.43217
4.49001
0.09674
4.77063
4.27753
0.17195
0.07378
4.91 123
4.92816
0.05803
4.74608
4.79686
4.90375
0.08369
2.54498
1.92781
1.92647
1.83606
1.40552
1.62423

Robust Std Err

4.49612
0.58642
1.09730
0.75981
0.55053
0.43400
0.74531
1.20209
1.06256
0.39961
0.22533
0.52706
0.92793
0.66663
0.12142
0.17373
0.04049
0.09771
0.15453
0.13286
0.1 1932
0.08212
0.04275
0.03955
0.51878
0.24487
0.52008
0.47048
0.45151
0.57869
0.49576
0.74824
0.66248
0.67519
0.77077
0.76094
0.64931
0.70529
0.85161
0.98609
0.94740
0.29438
0.34663
0.28488
0.59749
2.09785
1.55770

Robust t—stat

0.92159

1.90829
0.15173
2.33956
0.17192
2.49056
0.38439
0.22680
1.04576
0.91377
3.13302
4.46510
4.24632
2.67016
4.32674
0.47405
1.7274
1.70417
4.24105
0.41601
4.07044
1.98313
-7.25115
0.01024
3.58550
0.07817
4.93402
0.20590
0.73508
0.57211
0.91936
4.95981
0.95775
0.82005
0.15541
0.29507
4.30917
0.46215
0.16167
0.00550
0.59310
6.18884
4.41162
5.50135
3.63713
2.70824
3.06447

162

Table 8.19 - Industry Lag 12, PHS Lag 17 (10% Depreciation)

Num of Obs

Log Likelihood
Sum of Sqr Resid
R-Squared
Sigma Squared

Parameter

one
curird
lg1 ird
ngird
Ig3ird
lg4ird
lgSird
lgSird
lg7ird
lgBird
ngird
lg10wd
Ig1 1ird
Ig12wd
curreg
lagreg1
lagreg2
lagregB
lagreg4
IagregS
lagreg6
lagreg7
IagregB
lagregQ
PHS 17
y79
y80
y81
y82
y83
y84
y85
y86
y87
y88
y89
y90
y91
y92
y93
y94
one

car

ant

giu

der

res

119

92.02087
123.94608

0.78018
0.80159

Estimate

47.92620

1.05714
0.34999
1.86026
0.11775
1.08890
4.80206
0.28072
1.08680
4.17647
0.73812
0.74629
4.15571
1.76650
0.13171
0.06521
0.07908
0.17432
0.16816
0.04064
0.11054
0.15749
0.31529
0.11536
2.57085
0.73340
4.43028
4.63346
4.91719
0.88038
0.91798
0.56219
0.80895
0.80041
0.68455
0.81694
4.10515
0.78834
4.01869
4.41857
4.86673
2.20918
1.88761
2.03364
2.84938
8.23793
6.75882

Poisson Std Err

9.19330
0.97031
1.33203
1.31372
1.30104
1.26601
1.29154
1.32239
1.27718
1.24438
1.11571
1.10341
1.09086
1.10893
0.18349
0.18299
0.17954
0.19320
0.18787
0.17582
0.15203
0.14375
0.13997
0.13665
1.25760
0.53764
0.74282
0.86661
0.99205
1.18263
1.34872
1.44527
1.53507
1.57974
1.61929
1.66785
1.71600
1.76573
1.84129
1.94920
2.01969
0.76896
0.82001
0.88307
1.30609
4.48470
3.32466

4.94992
1.08948
4.76422
1.41603
0.09050
0.8601 1
4.39528
0.21228
0.85094
0.94543
0.66157
0.67635
4.05945
1.59297
0.71783
0.35634
0.44047
0.90230
0.89508
0.231 16
0.72710
1.09563
0.25253
0.84418
2.04426
4.36412
4.92547
4.88490
4.93255
0.43557
0.16351
4.77281
0.48129
0.40573
0.27541
0.28854
0.39229
0.14548
0.18254
0.26686
0.40964
2.87295
2.30193
2.30292
2.18161
1.83689
2.03293

Poisson t—stat Robust Std Err

6.22171
0.59540
1.03860
0.78162
0.50765
0.42978
0.75055
1.17084
1.02858
0.39135
0.21789
0.52927
0.90620
0.67135
0.11808
0.17009
0.03730
0.09756
0.15538
0.13520
0.12024
0.08028
0.04389
0.03791
0.76250
0.27794
0.59404
0.61760
0.64471
0.76539
0.79143
1 .05849
1 .03062
1.01468
1.10939
1.13549
1.04374
1.12123
1.26514
1 .43076
1.41573
0.40979
0.45198
0.45491
0.81 1 17
2.94873
2.19678

Robust t-stat

0.88123
1.77552
0.26265
2.38000
0.23194
2.53360
0.40099
0.23976
1 .05660
0.00615
3.38761
4.41004
4 .27533
2.63127
4 .1 1549
0.38336
2.1 1994
1.78673
4 .08227
0.30061
0.91933
1.96172
-7. 18286
0.04272
3.37160
0.63867
0.40771
0.64485
0.97374
0.76330
0.68698
0.42060
0.69580
0.74542
0.32125
0.36148
0.93312
0.37873
0.17648
0.08826
0.43760
5.39106
4.17635
4.47038
3.51266
2.79372
3.07669

163

Table 8.20 -lndus1ry Leg 12, PHS Lag 18 (10% Depreciation)

Num of Obs

Log Likelihood
Sum of Sqr Resid
R-Squared
Sigma Squared

Parameter

one
curird
lg1 ird
lg2ird
Ig3ird
lg4ird
IgSird
Ig6ird
Ig7ird
lgBird
lg9ird
lg1 Oird
Ig11kd
Ig120d
cu rreg
lagreg1
lagreg2
lagregS
lagreg4
IagregS
lagreg6
lagreg7
lagregB
iagregQ
PHS 1 8
y79
y80

y81

y82
y83
y84
y85
y86
y87
y88
y89
y90

y91

y92
y93
y94
cns

car

ant

giu

der

res

119

91.86941
124.87125

0.77854
0.79892

Estimate

4 9.52824

1.27215
0.55308
1.84067
0.23537
1.10546
4.72034
0.07425
1.22581
4.14997
0.78619
0.73152
4.09419
1.63519
0.11973
0.04169
0.09924
0.17302
0.15230
0.01950

.0.10249

0.16390
0.30528
0.11456

2.71591
0.80041
4.59192
4.90134
0.33254
0.31322
0.39997
0.17506
4.50647
4.58421
4.49658
4.65301
4.99556
4.64545
4.91479
0.24417
0.78203

2.25807

1.87502

2.14925

3.15569

8.98716

7.40153

Poisson Std Err

10.26496

0.95653
1.33879
1.30885
1.30215
1.26308
1.28617
1.31914
1.28113
1.24311
1.11444
1.09316
1.08759
1.09549
0.18400
0.18483
0.18083
0.19296
0.18839
0.17675
0.15302
0.14422
0.14016
0.13690
1.39315
0.56416
0.82398
1.01880
1.22240
1.42655
1.62934
1.79958
1.93171
2.01957
2.07627
2.13918
2.21090
2.24524
2.33918
2.41406
2.52297
0.82200
0.84618
0.97812
1.47045
5.03945
3.79242

4 .90242
1.32996
4.90700
1 .40633
0.18075
0.87521
4.33757
0.05629
0.95682
0.92508
0.70546
0.66918
4 .00607
1 .49265
0.65074
0.22559
0.54880
0.89662
0.80841
0.11033
0.66979
1.13652
0.17818
0.83685
1.94948
4.41877
4.93198
4.86624
4.90817
0.32253
0.08671
4.76434
0.33289
0.26989
0.16570
0.17514
0.25951
0.06902
0.10107
0.17234
0.29175
2.74704
2.21588
2.19733
2.14607
1.78336
1.95166

Poisson t-stat Robust Std Err

8.48044
0.50885
0.99253
0.79105
0.45338
0.45065
0.73083
1.11417
0.99074
0.38482
0.25863
0.53225
0.92102
0.63825
0.1 1706
0.17585
0.04288
0.09693
0.15408
0.14016
0.12319
0.08595
0.04652
0.03199
1.03097
0.33469
0.72313
0.76432
0.95173
1.04155
1.19487
1 .47834
1.49619
1.53991
1.61719
1 .71720
1.64028
1.71440
1.89915
1.99825
2.00806
0.49591
0.53816
0.66632
1.10965
3.96852
2.92295

Robust t-stat

0.30274
2.50004
0.57229
2.32688
0.51914
2.45304
0.35397
0.06664
1 .23726
0.98831
3.03987
4 .37440
4 .18802
2.56201
4 .02283
0.2371 1
2.31462
1 .78499
0.98846
0.13913
0.83203
1 .90697
0.56252
0.58124
2.63432
0.39150
0.20142
0.48760
0.45085
0.18103
0.84547
0.14772
0.01 196
0.97694
0.78049
0.70965
0.04555
0.70966
0.58788
0.62438
0.87941
4.55336
3.48415
3.22556
2.84387
2.26461
2.53221

164

Table 8.21 - Industry Lag 12. PHS Lag 19 (10% Depreciation)

Num of Obs 119
Log Likelihood 91 .49206
Sum of Sqr Resid 128.31454
R0quared 0.77243
Sigma Squared 0.80293
Parameter Estimate
one 4 8.48680
curird 1 .45076
lg1ird 0.52730
ngird 1 .58924
lg3ird 0.35322
lg4ird 1 .15253
IgSird 4.61858
lgGird 0.06387
Ig7ird 1 .24704
lg8ird 4 .03344
ngird 0.75256
lg10ird 0.67003
lg11ird 4.07854
Ig12ird 1.58153
curreg 0. 13069
lagreg1 0.04305
lagreg2 0.12032
lagreg3 0. 17069
lagreg4 0.15378
lagreg5 0.01285
lagreg6 0.08942
lagreg7 0.16854
lagregB 0.28138
lagregQ 0.10040
PHS 19 2.40882
y79 0.73779
y80 4 .4821 0
y81 4 .81409
y82 0.29579
y83 0.32898
y84 0.34442
y85 0.14609
y86 4.541 1 5
y87 4.68400
y88 4.65444
y89 4.82310
y90 -5. 17207
y91 4.85263
y92 0.08082
y93 -5.43146
y94 -5.87297
cns 2.05001
car 1 .61527
ant 1.9741 1
giu 3.12039
der 8.24732

res

6.87557

Poisson Std Err

10.84420

0.94811
1.33912
1.29063
1.30255
1.26425
1.28280
1.32366
1.28452
1.24463
1.11714
1.09604
1.08724
1.09360
0.18316
0.18531
0.18411
0.19330
0.18774
0.17691
0.15548
0.14461
0.14101
0.13697
1.40569
0.57028
0.84504
1.08676
1.34864
1.60526
1.79679
2.01010
2.19396
2.32883
2.42715
2.50425
2.58633
2.64042
2.71870
2.81162
2.87167
0.80798
0.80632
1.00255
1.59596
5.27601
3.99614

4.70476
1.53017
4.88728
1.23137
0.27117
0.91163
4.26175
0.04825
0.97082
0.83032
0.67365
0.61132
43.99200
1.44617
-o.71353
0.23230
0.65353
0.88302
0.81910
0.07265
0.57514
1.16547
4.99541
4.73300
1.71363
4.29374
4.75388
4.66927
4.70230
0.07380
4.86133
4.56514
0.06984
-2.01131
4.91766
4.92597
4.99978
4.83782
4.86884
4.93179
2.04514
2.53720
2.00326
1.96910
1.95518
1.56317
1.72055

Poisson t-stat Robust Std Err

9.90491
0.42888
0.99643
0.73368
0.41514
0.43644
0.66675
1.09165
0.97572
0.37187
0.25373
0.47688
0.89161
0.60349
0.12190
0.17974
0.05515
0.09446
0.15628
0.14373
0.12700
0.09142
0.05417
0.03737
1.17855
0.37871
0.81110
0.88043
1.17262
1.34703
1.51402
1.83188
1.86686
2.02614
2.13228
2.26403
2.20720
2.27510
2.45248
2.55165
2.51288
0.55505
0.57592
0.78995
1.37031
4.67835
3.45119

Robust t-stat

4.86643
3.38266
0.53635
2.16613
0.85084
2.64077
0.42759
0.05851
1.27808
0.77901
2.96596
4.40503
4.20966
2.62066
4.07206
0.23951
2.18167
1.80698
0.98401
0.08942
0.70411
1.84369
0.19477
0.68696
2.04388
4.94819
4.82727
0.06044
4.95782
0.47135
0.20897
4.71741
0.43251
0.31179
0.18285
0.13032
0.34327
0.13293
0.07171
0.12861
0.33715
3.69334
2.80468
2.49902
2.27715
1.76287
1.99223

165

Table 8.22 - Industry Log 12, PHS Log 20 (10% Doptochtion)

Num of Obs

Log Likelihood
Sum of Sqr Resid
R-Squared
Sigma Squared

Parameter

one
curird
Ig1 ird
igZird
lgBird
Ig4ird
lgSird
I96ird
Ig7ird
igBird
ngird
lg100d
lg11ud
|g12ud
curreg
lagreg1
lagregZ
lagreg3
lagreg4
lagregS
lagregG
lagreg7
IagregB
lagregQ
PHS 20
y79
y80

y81

y82
y83
y84
y85
y86
y87
y88
y89
y90
y91
y92
y93
y94
cns

car

ant

giu

der

res

119

90.83492
132.15497

0.76562
0.81439

Estimate

43.76404

1.57795
0.53202
1.50987
0.31569
1.19953
4.58277
0.09813
1.28424
0.97389
0.74747
0.71074
4.04324
1.55021
0.16474
0.07352
0.10489
0.16355
0.16750
0.02514
0.1 1095
0.17781
0.27403
0.08939
1 .62427
0.51507
4.071 16
4.27103
4.68294
0.62751
0.56966
0.26287
0.60325
0.73376
0.69773
-3.87933
4.21215
0.87520
4.10268
4.39225
4.84088
1.61674
1.15476
1.46372
2.55263
5.68986
4.95038

Poisson Std Err

10.45397

0.94665
1.33878
1.28563
1.30123
1.26407
1.28677
1.32936
1.29043
1.24789
1.12923
1.10366
1.09982
1.09583
0.18095
0.18351
0.18545
0.19317
0.18694
0.17597
0.15429
0.14520
0.14199
0.13812
1.26758
0.54195
0.78245
1.01509
1.30848
1.57752
1.77853
1.98411
2.18467
2.35278
2.47836
2.58864
2.68026
2.73665
2.83268
2.90156
2.97728
0.73175
0.70749
0.93173
1.60732
5.03860
3.81527

Poisson t-stat

4.31663
1.66688
4.89129
1.17442
0.24261
0.94894
4.23003
0.07382
0.99520
0.78043
0.66193
0.64399
0.94855
1.41464
0.91043
0.40063
0.56559
0.84663
0.89601
0.14285
0.71909
1.22456
4.92993
0.64718
1.28139
0.95041
4.36898
4.25214
4.28618
4.66560
4.44483
4.14050
4.64933
4.58696
4.49201
4.49860
4.57154
4.41604
4.44834
4.51376
4.62594
2.20940
1.63219
1.57098
1.58813
1.12925
1.29752

Robust Std Err

9.59022
0.41637
1.04546
0.73787
0.43556
0.46871
0.66929
1.13043
0.95171
0.37582
0.26630
0.47087
0.93974
0.59516
0.12884
0.17903
0.05555
0.09181
0.15196
0.14106
0.12345
0.09513
0.05462
0.04589
1.10867
0.35705
0.78359
0.83450
1.20372
1.38543
1.58557
1.88067
1.92882
2.15520
2.26595
2.45982
2.40985
2.45417
2.63021
2.67413
2.65328
0.53313
0.52061
0.76209
1.42799
4.61781
3.41802

Robust t-stat

4.43522
3.78975
0.42193
2.04624
0.72478
2.55923
0.36483
0.08681
1.34940
0.59138
2.80684
4.50942
4 .1 1014
2.60470
4.27869
0.41065
1.88805
1.78137
4.10226
0.17820
0.89876
1.86903
0.01660
4.94787
1.46506
4.44258
4.36698
4.52310
4 .3981 1
4.89654
4.62065
4.20323
4 .8681 1
4.73244
4.63187
4.57708
4.74788
4.57903
4.55983
4.64250
4.82449
3.03256
2.21807
1.92068
1.78756
1.23215
1.44832

166

Table 8.23 - Industry Lag 12. PHS Lag 15 (20% Depreciation)

Num of Obs 119
Log Likelihood 90.77965
Sum of Sqr Resid 129.45617
R-Squared 0.77041
Sigma Squared 0.83427
Parameter Estimate
one -7.66510
curird 1 . 16316
Ig1 ird 0.30456
lg2ird 1 .65123
lg3ird 0.08750
Ig4ird 1 .10484
lgSird 4 .79763
lg6ird 0.26959
lg7ird 1 .05945
lgBird 4 .07895
IgQird 0.67974
lg10ird 0.76571
Ig11ird 4.14118
Ig12ird 1 .77062
curreg 0. 17405
lagreg1 0.09670
lagreg2 0.05416
lagreg3 0.14933
lagreg4 0.20952
lagregS 0.06475
lagreg6 0.13691
lagreg7 0.16851
IagregB 0.31134
lagregQ 0. 12370
PHS 1 5 1 . 101 39
y79 0.32533
y80 0.61573
y81 0.52963
y82 0.60626
y83 4 .29640
y84 4 .02413
y85 0.42380
y86 4 .50866
y87 4 .42197
y88 4 . 19998
y89 4 .25721
y90 4.49125
y91 4 . 16282
y92 4 .29402
y93 4 .53096
y94 4 .95970
cns 1 .40570
car 1 . 14077
ant 1 .07145
giu 1 .51 181
der 3.03698

res 2.69447

Poisson Std Err

6.51472
0.96364
1.32653
1.29571
1.29468
1.25946
1.29561
1.31831
1.28988
1.26014
1.13528
1.11751
1.11253
1.13134
0.18194
0.18270
0.17965
0.19160
0.18587
0.17406
0.15066
0.14369
0.13947
0.13722
0.86322
0.48045
0.55579
0.55562
0.6031 1
0.70539
0.74096
0.72283
0.75776
0.78097
0.77411
0.79856
0.84576
0.90244
0.95587
1.00952
1.09580
0.58600
0.69871
0.65101
1.00301
3.12230
2.16229

4.17658
1 .20704
4 .73729
1 .27438
0.06759
0.87724
4 .38748
0.20450
0.82136
0.85622
0.59874
0.68520
4 .02575
1 .56507
0. 95664
0.52929
0.30146
0.77939
4 . 12723
0.37199
0. 90879
1 . 17272
0.23225
0.90142
1 .27591
0.67713
4 . 10785
0.95322
4 .00522
4 .83786
4 .38216
0.58630
4 .99094
4 .82078
4 .55013
4 .57434
4 .76321
4 .28853
4 .35377
4 .51653
4 .78837
2.39879
1 .63268
1 .64583
1 .50727
0.97268
1 .24612

Poisson t-stat Robust Std Err

2.99916
0.58308
1.13612
0.74172
0.56708
0.45417
0.77618
1.23290
1.09344
0.43470
0.21666
0.50602
0.94680
0.66458
0.12176
0.17593
0.04370
0.10010
0.15361
0.12758
0.12244
0.08463
0.04088
0.03853
0.31034
0.22526
0.44269
0.36247
0.29431
0.44626
0.25928
0.4871 1
0.36217
0.40314
0.48879
0.46031
0.36261
0.38904
0.49040
0.60785
0.54417
0.20817
0.27045
0.15308
0.44198
1.33928
0.98101

Robust t-stat

0.55575
1.99485
0.02844
2.22624
0.15431
2.43265
0.31601
0.21866
0.96891
0.48208
3.13739
4.51321
4.20530
2.66428
4.42941
0.54965
1.23940
1.49191
4.36392
0.50754
4 .1 1824
1.991 16
-7.61647
-3.21074
3.54897
4.44425
4.39087
4 .461 15
0.05993
0.90505
0.94993
0.87002
4.16563
0.52725
0.45498
0.73125
-4.1 1256
0.98892
0.63872
0.51866
0.60127
6.75256
4.21799
6.99921
3.42053
2.26763
2.74662

167

Table 8.24 - Industry Lag 12. PHS Lag 18 (20% Depreciation)

Num of Obs
Log Likelihood

Sum of Sqr Resid

R.Squared
Sigma Squared

Parameter

one
curird
lg1kd
ngkd
|gSkd
Ig4nd
lgsnd
Ig6kd
lg7nd
IgBWd
ngHd
lg100d
Ig11kd
ig12fld
curreg
kxfieg1
kmuegZ
Iagnxx3
kxneg4
kxyegS
kxyeQG
kuyeg7
kuyegB
kuyegQ
PHS16
y79
y80
y81
y82
y83
y84
y85
y86
y87
y88
y89
y90
y91
y92
y93
y94
cns
car
ant

gut
der
res

119

91.08372
128.52412

0.77206
0.82435

Esfinune

0.08930
1.09929
0.33204
1.76961
0.07605
1.07752
4.80922
0.29854
1.09344
4.14877
0.69514
0.77660
4.15351
1.77702
0.16578
0.08889
0.05722
0.15692
0.20293
0.06323
0.13553
0.16125
0.31435
0.12270
1.37294
0.40218
0.79744
0.73626
0.84352
4.59451
4.39203
0.83787
4.91980
4.81219
4.61076
4.63367
4.88489
4.49600
4.72055
4.95287
0.36722
1.56137
1.28880
1.23895
1.70916
3.92801
3.39270

Poisson Std Err

6.70342
0.96950
1.32881
1.30657
1.29903
1.26154
1.29631
1.31945
1.28257
1.25705
1.13141
1.11417
1.10605
1.12502
0.18205
0.18283
0.17965
0.19205
0.18596
0.17468
0.15057
0.14381
0.13944
0.13702
0.9121 1
0.48922
0.59821
0.61071
0.66990
0.79324
0.86234
0.86868
0.89641
0.90513
0.90668
0.91303
0.96201
0.98955
1.07996
1.12795
1.20347
0.60955
0.70314
0.67080
1.02849
3.26695
2.31218

4.35592
1.13387
4.75499
1.35440
0.05854
0.85413
4.39567
0.22626
0.85254
0.91386
0.61441
0.69702
4.04292
1.57955
0.91059
0.48619
0.31849
0.81710
4.09130
0.36195
0.90013
1.12126
0.25432
0.89548
1.50524
0.82209
4.33305
4.20558
4.25917
0.01013
4.61426
0.96453
0.14164
0.00213
4.77654
4.78928
4.95932
4.51179
4.59315
4.73135
4.96699
2.56150
1.83292
1.84696
1.66181
1.20235
1.46732

Poisson t-stat Robust Std Err

3.431 12
0.60204
1 .1 1916
0.75876
0.56662
0.43718
0.761 10
1.22760
1.08139
0.41 174
0.22272
0.52927
0.93888
0.66813
0.121 16
0.17521
0.04133
0.09795
0.15479
0.13018
0.1 1790
0.08516
0.04124
0.03979
0.38340
0.23086
0.48240
0.40991
0.35909
0.48509
0.36141
0.58895
0.48009
0.49033
0.57435
0.55359
0.42673
0.46592
0.59990
0.71750
0.65680
0.23603
0.29315
0.19186
0.48443
1 .60876
1 . 18285

Robust t—stat

0.64908
1.82593
0.08375
2.33224
0.13421
2.46468
0.37712
0.24319
1.01 1 14
0.79004
3.12114
4.46731
4.22861
2.65969
4.36823
0.50731
1.38424
1.60200
4.31105
0.48566
4.14953
1.89338
-7.62265
0.08363
3.58095
4 .7421 1
4.65306
4.79615
0.34903
0.28705
0.85170
4.42264
0.99882
0.69585
0.80447
0.95104
4.41703
0.21084
0.86805
0.72177
0.60418
6.61527
4.39645
6.45740
3.52818
2.44165
2.86824

168

Table 8.25 - Industry Lag 12. PHS Lag 17 (20% Depreciation)

Num of Obs 119

Log Likelihood 91.90392

Sum of Sqr Resid 124.13129

R-Squared 0.77985

Sigma Squared 0.80258

Parameter Estimate Poisson Std Err Poisson t-stat Robust Std Err Robust t-stat
one 42.78483 7.24538 4.76455 4.24606 0.01099
curird 0.96013 0.98105 0.97867 0.63322 1.51626
Ig1ird 0.28754 1.33096 4.71871 1.06388 0.15018
lg2ird 1.89096 1.31650 1.43636 0.78242 2.41682
lg3ird 0.07156 1.30038 0.05503 0.53374 0.13407
Ig4ird 1.08089 1.26525 0.85429 0.43029 2.51198
lgSird 4.86766 1.29568 4.44145 0.77130 0.42146
lgGird 0.36269 1 .32467 0.27380 1 .20668 0.30057
lg7ird 1.03579 1 .27897 0.80987 1 .05234 0.98427
IgBird 4.18560 1.24864 0.94951 0.40862 0.90149
ngird 0.71562 1.1 1989 0.63902 0.20951 3.41576
lg10ird 0.75435 1.10850 0.68051 0.53923 4.39893
Ig11ird 4.16381 1.09549 4.06237 0.92000 4.26501
tg12ird 1.79026 1.11445 1.60641 0.68271 2.62228
curreg 0.13176 0.18382 0.71677 0.11615 4.13436
lagreg1 0.07075 0.18315 0.38629 0.17139 0.41279
lagregZ 0.06153 0.17915 0.34346 0.03714 1.65694
lagreg3 0.16330 0.19232 0.84910 0.09702 1.68311
lagreg4 0.18432 0.18726 0.98429 0.15487 4.19011
IagregS 0.05493 0.17558 0.31282 0.13207 0.41588
IagregG 0.12214 0.15128 0.80741 0.11612 4.05182
lagreg7 0.15371 0.14379 1.06903 0.08195 1.87582
lagrege 0.32673 0.14015 0.33134 0.04331 -7.54406
tagregQ 0.12300 0.13680 0.89914 0.03841 0.20276
PHS 17 2.00149 1.00721 1.98716 0.53468 3.74336
y79 0.60876 0.51 138 4.19044 0.24757 0.45897
y80 4.18364 0.66744 4.77341 0.52601 0.25020
y81 4.25991 0.72861 4.72920 0.51196 0.46097
y82 4.40440 0.79347 4.76994 0.47061 0.98421
y83 0.25796 0.94087 0.39986 0.58173 0.88145
y84 0.20115 1.06118 0.07425 0.53809 4.09071
y85 4.73153 1.10290 4.56998 0.77318 0.23947
y86 0.87240 1.14982 0.49814 0.71892 0.99542
y87 0.75623 1.15076 0.39513 0.67530 4.081 48
y88 0.54898 1.15049 0.21555 0.74521 0.42046
y89 0.60238 1.16677 0.23043 0.73585 0.53657
y90 0.80988 1.18906 0. 3631 1 0.60674 4.63108
y91 0.44597 1.22156 0.00234 0.67203 0.63967
y92 0.61800 1.27922 0.04657 0.78871 0.31935
y93 0.99301 1.37836 0.17144 0.94657 0.16195
y94 0.39641 1.44107 0.35686 0.91526 0.71087
cns 1.91179 0.65194 2.93244 0.31669 6.03681
car 1 .63372 0.72370 2.25747 0.36283 4.50270
ant 1.63656 0.71923 2.27544 0.28464 5.74951
giu 2.23407 1.10257 2.02624 0.58123 3.84371
der 6.09558 3.59599 1.69510 2.09562 2.90872
res 5.03791 2.59854 1.93875 1.56035 3.22871

 

169

Table 8.28 - industry Lap 12, PHS Lag 18 (20% Depreciation)

Num of Obs 119

Log Likelihood 91.83871

Sum of Sqr Resid 124.00166

R-Squared 0.78008

Sigma Squared 0.79865

Parameter Estimate Poisson Std Err Poisson t—stat Robust Std Err Robust t-stat
one 44.62530 8.10106 4.80535 5.79942 0.52186
curird 1.13502 0.96638 1.17450 0.56330 2.01493
Ig1ird 0.51553 1.33699 4.88149 1.01323 0.48270
ngird 1.93091 1.31799 1.46505 0.80842 2.38851
Ig3ird 0.17005 1.30169 0.13064 0.48612 0.34982
lg4ird 1.09162 1.26174 0.86517 0.44889 2.43179
lgSird 4.79441 1.28978 4.39125 0.75738 0.36924
IgGird 0.18083 1.31780 0.13722 1.15162 0.15702
lg7ird 1.17339 1.27969 0.91694 1.01940 1.15107
lgBird 4.19120 1.24737 0.95497 0.40668 0.92912
IgQird 0.7681 1 1.1 1705 0.68762 0.25104 3.05965
lg10ird 0.75517 1.09752 0.68806 0.56342 4.34034
lg11ird 4.10178 1.09088 4.00999 0.93444 4.17908
Ig121rd 1 .66074 1.09925 1.51080 0.64976 2.55592
curreg 0.11238 0.18507 0.60725 0.11228 4.00088
lagreg1 0.03954 0.18554 0.21309 0.17641 0.22411
lagreg2 0.08283 0.17992 0.46035 0.03679 2.25119
lagregB 0.16480 0.19210 0.85785 0.09422 1.74910
lagreg4 0.16725 0.18758 0.89163 0.15134 4.10511
lagregS 0.03440 0.17622 0.19520 0.13245 0.25971
lagreg6 0.11651 0.15190 0.76704 0.11651 4.00005
lagreg7 0.15791 0.14421 1.09495 0.08773 1.79994
IagregB 0.32277 0.14021 0.30208 0.04387 -7.35771
lagregQ 0.12757 0.13718 0.92994 0.03076 4.14735
PHS 18 2.22707 1.14431 1.94621 0.71642 3.10862
y79 0.70661 0.53832 4.31261 0.28920 0.44331
y80 4.39918 0.75000 4.86558 0.61730 0.26662
y81 4 .58584 0.87796 4 .80629 0.60937 0.60241
y82 4.87098 1.00911 4.85409 0.69140 0.70606
y83 0.73041 1.15768 0.35853 0.75513 0.61584
y84 0.73209 1.31488 0.07783 0.82002 0.33174
y85 0.39363 1.42476 4.68003 1.06172 0.25449
y86 0.61297 1.50355 0.40295 1.03457 0.49224
y87 0. 57332 1.53642 0.32575 1.02471 0.48716
y88 0.37570 1.54018 0.19176 1.06784 0.16124
y89 0.43330 1.55694 0.20516 1.11111 0.08997
y90 0.68232 1.58845 0.31819 0.99422 0.70374
y91 0.25393 1.58877 0.04808 1.04544 0.11250
y92 0.46216 1.65799 0.08816 1.19605 0.89467
y93 0.73489 1.71048 0.18354 1.28682 0.90242
y94 4.26318 1.81928 0.34333 1.28698 0.31254
cns 2.00817 0.70413 2.85198 0.35502 5.65648
car 1.67831 0.75413 2.22549 0.40711 4.12245
ant 1.78954 0.80604 2.22017 0.42833 4.17791
giu 2.54355 1.22134 2.08258 0.78458 3.24193
der 7.03314 4.09172 1.71887 2.80765 2.50499
res 5.80281 3.00824 1.92897 2.03939 2.84536

 

170

Table 8.27 - Industry Lag 12, PHS Lag 19 (20% Depreciation)

Num of Obs

Log Likelihood
Sum of Sqr Resid
R-Squared
Sigma Squared

Parameter

one
curird
Ig1ird
ngird
IgBird
Ig4ird
igSird
IgGird
lg7ird
igBird
ngird
lg10ird
lg11kd
|g120d
cu rreg
lagreg1
lagreg2
lagreg3
lagreg4
lagregS
lagreg6
lagreg7
IagregB
lagregQ
PHS 19
y79
y80

y81

y82
y83
y84
y85
y86
y87
y88
y89
y90

y91

y92
y93
y94
cns

car

ant

giu

der

res

119

91.54301
126.70392

0.77529
0.8033

Estimate

4 5.32770

1.29249
0.48482
1.66535
0.31886
1.12336
4.67524
0.04258
1.19066
4.08068
0.73515
0.68537
4.09708
1.60645
0.11564
0.03112
0.11702
0.16980
0.16329
0.02262
0.09516
0.16101
0.29975
0.11542
2.17987
0.72509
4.44291
4.71068
0.08911
0.04384
0.99686
0.71884
4.03762
4.08695
0.96782
4.03989
4.28990
0.89840
4.04472
4.35183
4.75623
1.93510
1.55081
1.77606
2.71681
7.17141
5.94521

Poisson Std Err

9.01573
0.95450
1.33821
1.29580
1.30260
1 .26350
1.28327
1.31934
1.28196
1.24669
1.1 1650
1.09722
1.08628
1.09505
0.18490
0.18695
0.18360
0.19305
0.18706
0.17652
0.15478
0.14466
0.14039
0.13683
1.23593
0.56046
0.81283
1.00843
1.20625
1.41373
1.56382
1.72459
1.86124
1.94249
1.98891
2.01229
2.04154
2.05585
2.09231
2.16267
2.20560
0.72543
0.75153
0.87266
1.36791
4.53728
3.37069

4.70011
1.35411
4.85682
1.28519
0.24479
0.88909
4.30544
0.03227
0.92878
0.86684
0.65844
0.62464
4.00995
1.46701
0.62544
0.16644
0.63735
0.87956
0.87297
0.12813
0.61480
1.11301
0.13516
0.84354
1.76376
4.29374
4.77517
4.69638
4.73191
0.15306
4.91638
4.57651
0.16931
0.10397
4.99497
0.00761
0.10130
4.89625
4.93314
0.01225
0.15644
2.66754
2.06355
2.03522
1.98610
1.58055
1.76380

Poisson t-stat Robust Std Err

7.87571
0.49233
1.00124
0.74968
0.43051
0.42246
0.67945
1.11622
1.01 186
0.38435
0.24228
0.51 135
0.88839
0.61107
0.1 1339
0.18220
0.05047
0.09106
0.15595
0.13454
0.12299
0.09375
0.04910
0.03253
0.97196
0.36251
0.75053
0.79864
0.99253
1.161 10
1.23434
1.53096
1.50641
1.61697
1.69710
1.74760
1.66271
1.69430
1.83363
1.94164
1.89042
0.46442
0.50453
0.61548
1.12220
3.88467
2.82734

Robust t-stat

4.94620
2.62526
0.48174
2.22140
0.74068
2.65907
0.46558
0.03815
1.17671
0.81 174
3.03430
4.34031
4.23491
2.62890
4.01989
0.17077
2.31862
1.86466
4.04707
0.1681 1
0.77373
1.71747
-6. 10520
0.54841
2.24276
0.00021
4.92253
0.14198
0.10483
0.62152
0.42791
4.77591
0.68029
0.52754
0.33800
0.31 168
0.58007
0.30089
0.20586
0.24131
0.51596
4.16674
3.07379
2.88564
2.42095
1.84608
2.10276

171

Table 8.28 - Industry Lag 12. PHS Lag 20 (20% Depreciation)

Num of Obs
Log Likelihood

Sum of Sqr Resid

R-Squared
Sigma Squared

Parameter

one
curird
lg1ird
igZird
Ig3ird
lg4ird
IgSird
lgGird
lg7ird
lgBird
ngird
lg100d
lg11Wd
ig12ud
curreg
lagreg1
iagregZ
lagreg3
lagreg4
IagregS
IagregG
lagreg7
lagreg8
iagregQ
PHS 20
y79
y80

y81
y82
y83
y84
y85
y86
y87
y88
y89
y90
y91
y92
y93
y94
cns

car

ant

giu

der

res

119

90.80243
131.17883

0.76735
0.81526

Estimate

12.30732
1.48427
0.51883
1.57639
0.27568
1.19411
4.61488
0.04915
1.26486
4.01220
0.75695
0.73141
4.04548
1.55875
0.15795
0.06368
0.10829
0.16834
0.17418
0.03152
0.11298
0.17580
0.28904
0.10000
1.56730
0.53959
4.11464
4.30447
4.67868
0.59117
0.50752
0.14928
0.45511
0.53853
0.45178
0.58104
0.84200
0.44891
0.62304
0.86059
4.29798
1.59139
1.15699
1.38120
2.37813
5.33482
4.60217

Poisson Std Err

9.48287
0.94267
1.33877
1.28928
1.30078
1.26379
1.28550
1.32402
1.28837
1.24956
1.12868
1.10265
1.09740
1.09468
0.18156
0.18519
0.18641
0.19394
0.18630
0.17564
0.15448
0.14537
0.14089
0.13740
1.23102
0.55346
0.81256
1.04424
1.31254
1.55869
1.74040
1.90730
2.08253
2.21690
2.30403
2.37482
2.41374
2.42800
2.48543
2.51744
2.58530
0.71469
0.71017
0.87898
1.48991
4.77761
3.56235

4.29785
1.57454
4.88145
1.22270
0.21193
0.94486
4.25623
0.03712
0.98175
0.81005
0.67065
0.66332
0.95269
1.42393
0.86996
0.34387
0.58095
0.86799
0.93497
0.17944
0.73132
1.20930
0.05145
0.72780
1.27317
0.97495
4.37176
4.24920
4.27895
4.66240
4.44078
4.12687
4.65909
4.59616
4.49815
4.50792
4.59172
4.42047
4.45771
4.53354
4.66247
2.22668
1.62919
1.57136
1.59615
1.11663
1.29189

Poisson t—stat Robust Std Err

9.21307
0.43518
1.03669
0.75135
0.441 11
0.46437
0.66763
1.12763
0.97071
0.37596
0.26654
0.50087
0.94735
0.59135
0.12406
0.18402
0.05838
0.08820
0.15282
0.13427
0.12381
0.09589
0.05016
0.03994
1.14751
0.39935
0.84802
0.93747
1.28448
1.50247
1.64563
1.94772
1.97013
2.18073
2.28224
2.40847
2.33788
2.34123
2.48611
2.52810
2.52146
0.57342
0.57437
0.74137
1.42265
4.73159
3.48257

Robust t-stat

4.33585
3.41067
0.42968
2.09807
0.62497
2.57146
0.41881
0.04358
1.30303
0.69234
2.83992
4.46027
4.10359
2.63594
4.27317
0.34607
1.85499
1 .90867
4.13978
0.23472
0.91249
1.83340
0.76214
0.50344
1.36583
4 .351 17
4.31441
4.39148
4.30690
4.72461
4 .52374
4.10348
4.75375
4.62264
4.51245
4.48685
4.64337
4.47311
4.45731
4.52707
4.70456
2.77524
2.01436
1.86305
1.67162
1.12749
1.32149

 

172

Table 8.29 - CES Estimation, Industry Stock, PHS Lap 17

Num of Obs 119

Log Likelihood 85.60683

Sum of Sqr Resid 157.39969

R-Squared 0.72085

Sigma Squared 0.86354

Parameter Estimate Poisson Std Err Poisson t-stat Robust Std Err Robust t-stat
one 42.57139 7.34271 4.71209 3.43194 0.66306
pmastk 0.37727 0.60367 0.62495 0.08840 4.26772
curreg 0.00160 0.16680 0.00959 0.11096 0.01441
lagregt 0.07798 0.17226 0.45271 0.19918 0.39151
lagregZ 0.01255 0.17057 0.07356 0.06404 0.19591
IagregS 0.10685 0.17085 0.62539 0.08246 1.29580
lagreg4 0.10738 0.17491 0.61392 0.11282 0.95177
IagregS 0.04804 0.16374 0.29341 0.08669 0.55420
IagregS 0.17629 0.13971 4 .26185 0.08480 0.07888
lagreg7 0.00773 0.13460 0.05746 0.06765 0.11432
lagrega 0.21319 0.13068 4.63131 0.07941 0.68475
IagregQ 0.09284 0.12469 0.74457 0.06190 4.49975
Sub. Term "In IB" 0.39956 0.38627 4.03441 0.18243 0.19025
PHS 17 2.17691 1.07206 2.03059 0.51710 4.20987
y79 0.80972 0.47188 4.71595 0.19330 4.1 8897
y80 4.29032 0.62151 0.07609 0.41322 0.12261
y81 4 .18566 0.72201 4.64216 0.30088 0.94057
y82 4.44470 0.84000 4.71987 0.33353 4.33158
y83 0.23738 0.97755 0.28875 0.44679 -5.00762
y84 0.28497 1.07709 0.12143 0.37905 0.02818
y85 4 .98746 1.14403 4.73724 0.50197 0.95934
y86 0.67459 1.20855 0.21305 0.50347 0.31229
y87 0.75211 1.25028 0.20120 0.49959 0.50873
y88 0.88482 1 .27791 0.25745 0.54848 0.25969
y89 0.83874 1.31414 0.16016 0.63553 4.46674
y90 0.07269 1.36072 0.25814 0.55788 0.50781
y91 0.76602 1.40481 4.96897 0.58017 4.76759
y92 0.01832 1.44518 0.08854 0.67811 4.45111
y93 0.38523 1.54156 0.19597 0.82557 4.10047
y94 0.68344 1.58429 0.32499 0.68446 0.38155
cns 1 .63057 0.68381 2.38452 0.19438 8.38861
car 1.18282 0.71853 1.64617 0.19423 6.08970
ant 1.39464 0.78660 1.77300 0.27105 5.14535
giu 2.29671 1.07898 2.12859 0.49002 4.68699
der 5.94926 3.71685 1 .60062 1.62456 3.66208
res 4.72136 2.65658 1.77724 1.11399 4.23824

‘3-

173

Table 8.30 - Indirect Eflect, PHS Concurrent Lag

Total SS: 62.081
R-squared: 0.979
Residual SS: 1.324
F(29,75): 118.708
Durbin-Watson: 1 .169
Degrees of Freedom: 75
Rbar-squared: 0.97
Std error of Est: 0.133
Probability of F: 0

Dependent Variable: Log of Industry R&D Investment

 

Standard Standardized Cor with
Variable Estimate Error t-value Prob > It) Estimate Dep Var
CONSTANT 7.9002 2.2464 3.5169 0.0010
PHS BASIC Concurrent 0.1744 0.1760 0.9909 0.3250 0.2890 0.8707
LOGSALE 0.1488 0.1251 1.1896 0.2380 0.1256 0.8624
INCTHRE (age 4504) 0.7132 0.2266 3.1475 0.0020 0.9299 0.6558
SEVTWO (age 1544) 4 .0936 0.2361 4.6326 0.0000 4 .3408 0.5758
CURREG 0.0248 0.0292 0.8492 0.3980 0.0201 0.1778
LAGt REG 0.0167 0.0263 0.6351 0.5270 0.0145 0.2097
LAGZREG 0.0031 0.0270 0.1159 0.9080 0.0029 0.2836
LAG3REG 0.0223 0.0274 0.8155 0.4170 0.0205 0.2937
LAG4REG 0.0158 0.0288 0.5483 0.5850 0.0140 0.3334
CNS 0.9411 0.2959 3.1807 0.0020 0.4283 0.2878
CAR 0.3710 0.1902 4 .9501 0.0550 0.1688 0.3282
ANT 0.5022 0.2464 2.0376 0.0450 0.2285 0.3460
GIU 0.0935 0.2181 0.4287 0.6690 0.0426 0.2333
DER 4 .3017 0.6507 0.0005 0.0490 0.5924 0.6446
RES 4 .0855 0.3764 0.8837 0.0050 0.4940 0.3747
YR72 0.0825 0.0757 1.0906 0.2790 0.0268 0.1031
YR73 0.1165 0.0815 1.4292 0.1570 0.0378 0.0815
YR74 0.1942 0.0848 2.2904 0.0250 0.0630 0.0456
YR75 0.1853 0.0901 2.0564 0.0430 0.0601 0.0396
YR76 0.0258 0.0943 0.2732 0.7850 0.0084 0.0588
YR77 0.0847 0.0977 0.8662 0.3890 0.0275 0.0363
YR78 0.0963 0.1010 0.9530 0.3440 0.0312 0.0254
YR79 0.0693 0.1053 0.6587 0.5120 0.0225 0.0110
YR80 0.1540 0.1087 1.4169 0.1610 0.0500 0.0211
YR81 0.1697 0.1113 1.5247 0.1320 0.0551 0.0363
YR82 0.2622 0.1166 2.2484 0.0270 0.0850 0.0772
YR83 0.3291 0.1215 2.7095 0.0080 0.1068 0.1135
YR84 0.2998 0.1345 2.2286 0.0290 0.0973 0.1318
YR85 0.3604 0.1465 2.4610 0.0160 0.1169 0.1550

Table 8.31 - Indirect Effect. PHS Lac 1

Total SS:

R-squared:

Residual SS:
F(29.75):
Durbin-Watson:
Degrees of Freedom:
Rbar-squared:

Std error of Est:
Probability of F :

Dependent Variable: Log of Industry R&D Investment

62.081
0.979
1.317

119.297

1.174
75
0.971
0.133
0

174

 

Standard Standardized Cor with
Variable _ Estimate Error t-value Prob > [t] Estimate Dep Var
CONSTANT 7.7872 2.2426 3.4723 0.0010
PHS BASIC LAG 1 0.1853 0.1595 1.1623 0.2490 0.3070 0.8727
LOGSALE 0.1630 0.1267 1.2864 0.2020 0.1375 0.8624
INCTHRE (age 4504) 0.6746 0.2349 2.8716 0.0050 0.8796 0.6558
SEVTWO (age 1544) 4 .0734 0.2343 4.5809 0.0000 4 .3160 0.5758
CURREG 0.0237 0.0291 0.8158 0.4170 0.0192 0.1778
LAGt REG 0.0155 0.0263 0.5915 0.5560 0.0135 0.2097
LAG2REG 0.0021 0.0270 0.0772 0.9390 0.0019 0.2836
LAG3REG 0.0211 0.0274 0.7695 0.4440 0.0194 0.2937
LAG4REG 0.0143 0.0289 0.4968 0.6210 0.0127 0.3334
CNS 0.9181 0.2832 3.2420 0.0020 0.4178 0.2878
CAR 0.3492 0.1888 4 .8497 0.0680 0.1589 0.3282
ANT 0.4650 0.2497 1.8626 0.0660 0.2116 0.3460
GIU 0.1024 0.1951 0.5249 0.6010 0.0466 0.2333
DER 4 .3085 0.6166 0.1219 0.0370 0.5955 0.6446
RES 4 .0673 0.3562 0.9962 0.0040 0.4857 0.3747
YR72 0.0893 0.0761 1.1729 0.2450 0.0290 0.1031
YR73 0.1218 0.0818 1.4894 0.1410 0.0395 0.0815
YR74 0.2071 0.0860 2.4075 0.0190 0.0672 0.0456
YR75 0.1969 0.0911 2.1612 0.0340 0.0639 0.0396
YR76 0.0478 0.0952 0.5019 0.6170 0.0155 0.0588
YR77 0.0982 0.0968 1.0140 0.3140 0.0319 0.0363
YR78 0.1099 0.0990 1.1109 0.2700 0.0357 0.0254
YR79 0.0853 0.1001 0.8526 0.3970 0.0277 0.0110
YR80 0.1662 0.1032 1.6104 0.1120 0.0539 0.0211
YR81 0.1794 0.1046 1.7155 0.0900 0.0582 0.0363
YR82 0.2693 0.1098 2.4532 0.0160 0.0874 0.0772
YR83 0.3367 0.1131 2.9775 0.0040 0.1092 0.1135
YR84 0.3061 0.1239 2.4709 0.0160 0.0993 0.1318
YR85 0.3617 0.1355 2.6699 0.0090 0.1173 0.1550

Table 8.32 - Indirect Effect. PHS Lag 2

Total SS:

R-squared:

Residual SS:
F(29,75):
Durbin-Watson:
Degrees of Freedom:
Rbar-squared:

Std error of Est:
Probability of F:

Dependent Variable: Log of Industry R&D Investment

62.081
0.979
1.299

121.027

1.181
75
0.971
0.132
0

175

 

Standard Standardized Cor with
Variable Estimate Error t—value Prob > It] Estimate Dep Var
CONSTANT 7.5153 2.2318 3.3674 0.0010
PHS BASIC LAG 2 0.2317 0.1485 1.5604 0.1230 0.3832 0.8742
LOGSALE 0.1862 0.1277 1.4575 0.1490 0.1571 0.8624
INCTHRE (age 4504) 0.6050 0.2409 2.5114 0.0140 0.7889 0.6558
SEVTWO (age 1544) 4 .0487 0.2331 4.4995 0.0000 4 .2857 0.5758
CURREG 0.0232 0.0285 0.8119 0.4190 0.0188 0.1778
LAG1 REG 0.0146 0.0260 0.5639 0.5740 0.0127 0.2097
LAG2REG 0.0003 0.0268 0.0113 0.9910 0.0003 0.2836
LAG3REG 0.0189 0.0273 0.6916 0.4910 0.0173 0.2937
LAG4REG 0.0127 0.0286 0.4431 0.6590 0.0113 0.3334
CNS 0.9074 0.2752 3.2970 0.0010 0.4130 0.2878
CAR 0.2929 0.1890 4 .5500 0.1250 0.1333 0.3282
ANT 0.4133 0.2528 1.6347 0.1060 0.1881 0.3460
GIU 0.0851 0.1796 0.4740 0.6370 0.0387 0.2333
DER 4 .2426 0.5952 0.0878 0.0400 0.5655 0.6446
RES 0.9818 0.3466 0.8326 0.0060 0.4468 0.3747
YR72 0.0942 0.0756 1.2459 0.2170 0.0306 0.1031
YR73 0.1324 0.0819 1.6178 0.1100 0.0430 0.0815
YR74 0.2175 0.0859 2.5310 0.0130 0.0706 0.0456
YR75 0.2151 0.0922 2.3342 0.0220 0.0698 0.0396
YR76 0.0633 0.0957 0.6616 0.5100 0.0205 0.0588
YR77 0.1227 0.0976 1.2562 0.2130 0.0398 0.0363
YR78 0.1209 0.0982 1.2310 0.2220 0.0392 0.0254
YR79 0.0945 0.0982 0.9621 0.3390 0.0306 0.0110
YR80 0.1759 0.1002 1.7564 0.0830 0.0571 0.0211
YR81 0.1830 0.1006 1.8196 0.0730 0.0594 0.0363
YR82 0.2667 0.1057 2.5237 0.0140 0.0865 0.0772
YR83 0.3298 0.1090 3.0263 0.0030 0.1070 0.1135
YR84 0.2962 0.1179 2.5136 0.0140 0.0961 0.1318
YR85 0.3433 0.1287 2.6688 0.0090 0.1114 0.1550

Table 8.33 - Indiect Effect. PHS Lag 3

Total SS: 62.081
R-squared: 0.98
Residual SS: 1.265
F(29,75): 124.306
Durbin-Watson: 1.188
Degrees of Freedom: 75
Rbar—squared: 0.972
Std error of Est: 0.13
Probability of F: 0

176

 

Dependent Variable: Log of Industry R&D Investment

 

Standard Standardized Cor with
Variable Estimate Error t-value Prob > It) Estimate Dep Var
CONSTANT 7.1554 2.2052 3.2448 0.0020
PHS BASIC LAG 3 0.2976 0.1405 2.1187 0.0370 0.4913 0.8762
LOGSALE 0.2168 0.1276 1.6982 0.0940 0.1829 0.8624
INCTHRE (age 45-64) 05118 0.2434 2.1026 0.0390 0.6674 0.6558
SEVTWO (age 1544) 4 .0150 0.2311 4.3920 0.0000 4 .2444 0.5758
CURREG 0.0233 0.0279 0.8338 0.4070 0.0189 0.1778
LAG1 REG 0.0159 0.0254 0.6263 0.5330 0.0138 0.2097
LAG2REG 0.0001 0.0263 0.0055 0.9960 0.0001 0.2836
LAG3REG 0.0162 0.0269 0.6004 0.5500 0.0149 0.2937
LAG4REG 0.0106 0.0282 0.3747 0.7090 0.0094 0.3334
CNS 0.8976 0.2690 3.3368 0.0010 0.4085 0.2878
CAR 0.2133 0.1887 4.1308 0.2620 0.0971 0.3282
ANT 0.3461 0.2539 1.3631 0.1770 0.1575 0.3460
GIU 0.0586 0.1686 0.3474 0.7290 0.0267 0.2333
DER 4 .1405 0.5782 4 .9724 0.0520 0.5190 0.6446
RES 0.8507 0.3418 0.4891 0.0150 0.3872 0.3747
YR72 0.0894 0.0739 1.2097 0.2300 0.0290 0.1031
YR73 0.1283 0.0797 1.6094 0.1120 0.0416 0.0815
YR74 0.2219 0.0842 2.6344 0.0100 0.0720 0.0456
YR75 0.2180 0.0898 2.4266 0.0180 0.0707 0.0396
YR76 0.0751 0.0945 0.7946 0.4290 0.0244 0.0588
YR77 0.1290 0.0962 1.3412 0.1840 0.0418 0.0363
YR78 0.1358 0.0973 1.3949 0.1670 0.0440 0.0254
YR79 0.0912 0.0969 0.9415 0.3490 0.0296 0.0110
YR80 0.1709 0.0988 1.7307 0.0880 0.0554 0.021 1
YR81 0.1786 0.0984 1.8153 0.0730 0.0579 0.0363
YR82 0.2517 0.1039 2.4238 0.0180 0.0817 0.0772
YR83 0.3073 0.1076 2.8560 0.0060 0.0997 0.1135
YR84 0.2650 0.1167 2.2718 0.0260 0.0860 0.1318
YR85 0.3032 0.1269 2.3883 0.0190 0.0984 0.1550

Table 8.34 - Indirect Eflect. PHS Lag 4

Total SS:

R-squared:

Residual SS:
F(29,75):
Durbin-Watson:
Degrees of Freedom:
Rbar-squared:

Std error of Est:
Probability of F:

Dependent Variable: Log of Industry R&D Investment

62.081
0.98
1.24

126.942

1.22
75
0.972
0.129
0

177

 

Standard Standardized Cor with
Variable Estimate Error t-value Prob > [t] Estimate Dep Var
CONSTANT 6.9067 2.1866 3.1587 0.0020
PHS BASIC LAG 4 0.3331 0.1344 2.4779 0.0150 0.5485 0.8783
LOGSALE 0.2327 0.1266 1.8385 0.0700 0.1964 0.8624
INCTHRE (age 4504) 0.4562 0.2427 1.8797 0.0640 0.5949 0.6558
SEVTWO (age 1544) 0.9797 0.2302 4.2561 0.0000 4 .2012 0.5758
CURREG 0.0236 0.0275 0.8552 0.3950 0.0191 0.1778
LAG1 REG 0.0189 0.0250 0.7558 0.4520 0.0164 0.2097
LAG2REG 0. 0031 0.0259 0.1201 0.9050 0.0029 0.2836
LAGBREG 0.0166 0.0265 0.6258 0.5330 0.0152 0.2937
LAG4REG 0.0093 0.0279 0.3325 0.7400 0.0082 0.3334
CNS 0.8799 0.2651 3.3194 0.0010 0.4004 0.2878
CAR 0.1623 0.1881 0.8630 0.3910 0.0739 0.3282
ANT 0.3087 0.2526 1.2220 0.2260 0.1405 0.3460
GIU 0.0556 0.1609 0.3456 0.7310 0.0253 0.2333
DER 4 .0700 0.5700 4 .8772 0.0640 0.4870 0.6446
RES 0.7540 0.3426 0.2007 0.0310 0.3431 0.3747
YR72 0.0799 0.0729 1.0959 0.2770 0.0259 0.1031
YR73 0.1057 0.0782 1.3518 0.1810 0.0343 0.0815
YR74 0.1993 0.0819 2.4338 0.0170 0.0647 0.0456
YR75 0.2030 0.0875 2.3215 0.0230 0.0659 0.0396
YR76 0.0575 0.0918 0.6262 0.5330 0.0187 0.0588
YR77 0.1204 0.0944 1.2745 0.2060 0.0390 0.0363
YR78 0.1204 0.0959 1.2554 0.2130 0.0391 0.0254
YR79 0.0899 0.0959 0.9376 0.3510 0.0292 0.0110
YR80 0.1520 0.0986 1.5406 0.1280 0.0493 0.0211
YR81 0.1601 0.0984 1.6268 0.1080 0.0519 0.0363
YR82 0.2356 0.1035 2.2766 0.0260 0.0764 0.0772
YR83 0.2828 0.1082 2.6129 0.0110 0.0917 0.1135
YR84 0.2342 0.1179 1.9869 0.0510 0.0760 0.1318
YR85 0.2640 0.1291 2.0452 0.0440 0.0857 0.1550

 

Table 3.35 - Indirect Effect. PHS Lac 5

Total SS:

R-squared:

Residual SS:
F(29,75):
Durbin-Watson:
Degrees of Freedom:
Rbar-squared:

Std error of Est:
Probability of F:

Dependent Variable: Log of Industry R&D Investment

62.081
0.981
1.202

130.943

1.231
75
0.973
0.127
0

178

 

Standard Standardized Cor with
Variable Estimate Error t-value Prob > It] Estimate Dep Var
CONSTANT 6.5958 2.1575 3.0572 0.0030
PHS BASIC LAG 5 0.3807 0.1295 2.9405 0.0040 0.6255 0.8805
LOGSALE 0.2442 0.1240 1 .9699 0.0530 0.2061 0.8624
INCTHRE (age 4504) 0.4094 0.2362 1.7335 0.0870 0.5338 0.6558
SEVTWO (age 1544) 0.9471 0.2277 4.1588 0.0000 4.1611 0.5758
CURREG 0.0218 0.0271 0.8036 0.4240 0.0177 0.1778
LAG1 REG 0.0211 0.0246 0.8571 0.3940 0.0183 0.2097
LAGZREG 0.0076 0.0254 0.2975 0.7670 0.0070 0.2836
LAGBREG 0.0199 0.0259 0.7677 0.4450 0.0183 0.2937
LAG4REG 0.0090 0.0274 0.3294 0.7430 0.0080 0.3334
CNS 0.8759 0.2608 3.3587 0.0010 0.3986 0.2878
CAR 0.1062 0.1846 0.5752 0.5670 0.0483 0.3282
ANT 0.2906 0.2469 1.1769 0.2430 0.1322 0.3460
GIU 0.0395 0.1555 0.2537 0.8000 0.0180 0.2333
DER 0.9435 0.5651 4 .6698 0.0990 0.4294 0.6446
RES 0.6174 0.3456 4 .7866 0.0780 0.2810 0.3747
YR72 0.0636 0.0718 0.8855 0.3790 0.0206 0.1031
YR73 0.0746 0.0777 0.9598 0.3400 0.0242 0.0815
YR74 0.1535 0.0813 1.8878 0.0630 0.0498 0.0456
YR75 0.1571 0.0858 1.8305 0.0710 0.0510 0.0396
YR76 0.0206 0.0899 0.2290 0.8190 0.0067 0.0588
YR77 0.0774 0.0927 0.8344 0.4070 0.0251 0.0363
YR78 0.0862 0.0950 0.9075 0.3670 0.0280 0.0254
YR79 0.0484 0.0962 0.5026 0.6170 0.0157 0.0110
YR80 0.1257 0.0985 1.2765 0.2060 0.0408 0.0211
YR81 0.1138 0.1007 1.1303 0.2620 0.0369 0.0363
YR82 0.1894 0.1060 1.7864 0.0780 0.0614 0.0772
YR83 0.2384 0.1105 2.1575 0.0340 0.0773 0.1135
YR84 0.1799 0.1215 1.4812 0.1430 0.0584 0.1318
YR85 0.2000 0.1338 1.4946 0.1390 0.0649 0.1550

Table 8.38 - Indirect Effect. PHS Lag 6

Total SS:

R-squared:

Residual SS:

F (29,75):
Durbin-Watson:
Degrees of Freedom:
Rbar—squared:

Std error of Est:
Probability of F:

Dependent Variable: Log of Industry R&D Investment

62.081
0.981
1.175

134.056

1.238
75
0.974
0.125
0

179

 

Standard Standardized Cor with
Variable Estimate Error t-value Prob > It) Estimate Dep Var
CONSTANT 6.3384 2.1399 2.9620 0.0040
PHS BASIC LAG 6 0.4204 0.1292 3.2552 0.0020 0.6896 0.8825
LOGSALE 0.2419 0.1213 1.9941 0.0500 0.2041 0.8624
INCTHRE (age 4504) 0.3954 0.2291 1.7261 0.0880 0.5156 0.6558
SEVTWO (age 1544) 0.9248 0.2257 4.0969 0.0000 4 .1339 0.5758
CURREG 0.0219 0.0268 0.8158 0.4170 0.0177 0.1778
LAG1 REG 0.0218 0.0244 0.8952 0.3740 0.0189 0.2097
LAGZREG 0.0115 0.0252 0.4561 0.6500 0.0106 0.2836
LAGBREG 0.0248 0.0255 0.9726 0.3340 0.0228 0.2937
LAG4REG 0.0120 0.0270 0.4460 0.6570 0.0107 0.3334
CNS 0.8867 0.2579 3.4377 0.0010 0.4035 0.2878
CAR 0.0677 0.1826 0.3709 0.7120 0.0308 0.3282
ANT 0.3059 0.2407 1.2710 0.2080 0.1392 0.3460
GIU 0.0207 0.1535 0.1349 0.8930 0.0094 0.2333
DER 0.8009 0.5689 4 .4077 0.1630 0.3645 0.6446
RES 0.4799 0.3570 4 .3441 0.1830 0.2184 0.3747-
YR72 0.0548 0.0712 0.7700 0.4440 0.0178 0.1031
YR73 0.0447 0.0783 0.5710 0.5700 0.0145 0.0815
YR74 0.1069 0.0833 1.2835 0.2030 0.0347 0.0456
YR75 0.0937 0.0884 1.0603 0.2920 0.0304 0.0396
YR76 0.0426 0.0916 0.4652 0.6430 0.0138 0.0588
YR77 0.0218 0.0942 0.2318 0.8170 0.0071 0.0363
YR78 0.0236 0.0976 0.2418 0.8100 0.0077 0.0254
YR79 0.0024 0.0988 0.0245 0.9810 0.0008 0.01 10
YR80 0.0674 0.1024 0.6588 0.5120 0.0219 0.0211
YR81 0.0741 0.1032 0.7182 0.4750 0.0240 0.0363
YR82 0.1282 0.1119 1.1454 0.2560 0.0416 0.0772
YR83 0.1781 0.1166 1.5276 0.1310 0.0578 0.1135
YR84 0.1244 0.1268 0.9809 0.3300 0.0403 0.1318
YR85 0.1351 0.1408 0.9591 0.3410 0.0438 0.1550

 

 

 

Table 8.37 - Indirect Effect, PHS Lag 7

Total SS:

R-squared:

Residual SS:
F(29,75):
Durbin-Watson:
Degrees of Freedom:
Rbar-squared:

Std error of Est:
Probability of F:

Dependent Variable: Log of Industry R&D Investment

62.081
0.981
1.169

134.714

1.245
75
0.974
0.125
0

180

 

Standard Standardized Cor with
Variable Estimate Error t-value Prob > It] Estimate Dep Var
CONSTANT 6.0088 2.1641 2.7766 0.0070
PHS BASIC LAG 7 0.4562 0.1375 3.3179 0.0010 0.7478 0.8839
LOGSALE 0.2383 0.1205 1.9767 0.0520 0.2010 0.8624
INCTHRE (age 4504) 0.4001 0.2265 1.7664 0.0810 0.5218 0.6558
SEVTWO (age 1544) 0.9046 0.2263 0.9968 0.0000 4 .1090 0.5758
CURREG 0.0212 0.0267 0.7934 0.4300 0.0172 0.1778
LAGi REG 0.0233 0.0243 0.9604 0.3400 0.0203 0.2097
LAGZREG 0.0133 0.0251 0.5295 0.5980 0.0122 0.2836
LAG3REG 0.0286 0.0254 4 .1238 0.2650 0.0263 0.2937
LAG4REG 0.0159 0.0268 0.5919 0.5560 0.0141 0.3334
CNS 0.8946 0.2575 3.4749 0.0010 0.4071 0.2878
CAR 0.0409 0.1866 0.2191 0.8270 0.0186 0.3282
ANT 0.3403 0.2373 1.4343 0.1560 0.1549 0.3460
GIU 0.0024 0.1552 0.0153 0.9880 0.0011 0.2333
DER 0.6324 0.5908 4 .0705 0.2880 0.2878 0.6446
RES 0.3363 0.3867 0.8698 0.3870 0.1531 0.3747
YR72 0.0483 0.0712 0.6791 0.4990 0.0157 0.1031
YR73 0.0275 0.0795 0.3455 0.7310 0.0089 0.0815
YR74 0.0657 0.0874 0.7520 0.4540 0.0213 0.0456
YR75 0.0360 0.0945 0.3815 0.7040 0.0117 0.0396
YR76 0.1163 0.0990 4 .1748 0.2440 0.0377 0.0588
YR77 0.0527 0.1017 0.5189 0.6050 0.0171 0.0363
YR78 0.0417 0.1045 0.3993 0.6910 0.0135 0.0254
YR79 0.0747 0.1076 0.6945 0.4900 0.0242 0.01 10
YR80 0.0083 0.1099 0.0756 0.9400 0.0027 0.021 1
YR81 0.0077 0.1127 0.0687 0.9450 0.0025 0.0363
YR82 0.0829 0.1188 0.6978 0.4870 0.0269 0.0772
YR83 0.1099 0.1284 0.8559 0.3950 0.0357 0.1135
YR84 0.0583 0.1389 0.4201 0.6760 0.0189 0.1318
YR85 0.0757 0.1525 0.4968 0.6210 0.0246 0.1550

Table 8.38 - Indirect Effect. PHS Lag 8

Total SS:

R-squared:

Residual SS:
F(29,75):
Durbin-Watson:
Degrees of Freedom:
Rbar-squared:

Std error of Est:
Probability of F:

Dependent Variable: Log of Industry R&D Investment

62.081
0.981
1.195

131.757

1.246
75
0.973
0.126
0

181

 

Standard Standardized Cor with
Variable Estimate Error t-value Prob > It) Estimate Dep Var
CONSTANT 5.9795 2.2150 2.6996 0.0090 —
PHS BASIC LAG 8 0.4620 0.1527 3.0259 0.0030 0.7581 0.8850
LOGSALE 0.2191 0.1209 1.8116 0.0740 0.1849 0.8624
INCTHRE (age 4504) 0.4484 0.2262 1.9823 0.0510 0.5846 0.6558
SEVTWO (age 1544) 0.9154 0.2290 0.9982 0.0000 4.1223 0.5758
CURREG 0.0228 0.0270 0.8439 0.4010 0.0185 0.1778
LAG1 REG 0.0234 0.0246 0.9533 0.3430 0.0204 0.2097
LAG2REG 0.0149 0.0255 0.5867 0.5590 0.0137 0.2836
LAG3REG 0.0300 0.0257 -1 .1658 0.2470 0.0276 0.2937
LAG4REG 0.0185 0.0271 0.6846 0.4960 0.0165 0.3334
CNS 0.9186 0.2612 3.5168 0.0010 0.4180 0.2878
CAR 0.0591 0.1932 0.3058 0.7610 0.0269 0.3282
ANT 0.3972 0.2371 1.6755 0.0980 0.1808 0.3460
GIU 0.0099 0.1624 0.0611 0.9510 0.0045 0.2333
DER 0.5486 0.6278 0.8738 0.3850 0.2497 0.6446
RES 0.2741 0.4298 0.6377 0.5260 0.1247 0.3747
YR72 0.0270 0.0732 0.3690 0.7130 0.0088 0.1031
YR73 0.0028 0.0839 0.0328 0.9740 0.0009 0.0815
YR74 0.0320 0.0952 0.3361 0.7380 0.0104 0.0456
YR75 0.0190 0.1071 0.1775 0.8600 0.0062 0.0396
YR76 0.1856 0.1145 4 .6210 0.1090 0.0602 0.0588
YR77 0.1385 0.1205 4 .1499 0.2540 0.0449 0.0363
YR78 0.1272 0.1237 4 .0277 0.3070 0.0413 0.0254
YR79 0.1488 0.1260 4 .1813 0.2410 0.0483 0.0110
YR80 0.0722 0.1301 0.5551 0.5800 0.0234 0.021 1
YR81 0.0569 0.1311 0.4338 0.6660 0.0185 0.0363
YR82 0.0121 0.1398 0.0868 0.9310 0.0039 0.0772
YR83 0.0641 0.1459 0.4393 0.6620 0.0208 0.1135
YR84 0.0073 0.1630 0.0448 0.9640 0.0024 0.1318
YR85 0.0176 0.1774 0.0992 0.9210 0.0057 0.1550

BIBLIOGRAPHY

BIBLIOGRAPHY

Acs, Zoltan J., David B. Audretsch, and Maryann P. Feldman. (1991) “Real
Effect of Academic Research: Comment,” American Economic Review,
March 1991, pp. 363067.

Adams, James D. (1990) “Fundamental Stocks of Knowledge and Productivity
Growth,“ Journal of Political Economy, Vol. 98, No. 4, 1990, pp. 673-702.

Baily, Martin N. (1972) "Research and Development Costs and Returns: The
US. Pharmaceutical Industry," Journal of Political Economy,
January/February 1972, pp. 70-85.

Bloom, Barry M. (1976) “Socially Optimal Results from Drug Research,” in
Impact of Pgblic Policy on Drug Innovation and Pricing, Samuel A.
Mitchell (editor), Washington, DC: The American University, 1976.

Blundell, Richard, Rachel Griffith, and John Van Reenen. (1995) “Dynamic
Count Data Models of Technological Innovation,” The Economic Journal,
Vol. 105, March 1995, pp. 333-344.

Cockbum, Iain and Rebecca Henderson. (1994) “Racing to Invest? Dynamics of
Competition in Ethical Drug Discovery,” Journal of Economics 8
Management Strategy, Vol. 3, No. 3, Fall 1994, pp. 481 -519.

Davidson, R. and J. MacKinnon. (1981) “Several Tests for Model Specification in
the Presence of Alternative Hypotheses,” Econometrica, Vol. 49, 1981,
pp. 781 -793.

DiMasi, Joseph A., Ronald W. Hansen, Henry G. Grabowski, and Louis
Lasagna. "Cost of innovation in the pharmaceutical industry," Journal of
Health Economics, Vol. 10, 1991, pp. 107-142.

Dranove, David, and David Meltzer. (1994) “Do important drugs reach the
market sooner?” Rand Journal of Economics, Vol. 25, No. 3, Autumn,
1994, pp. 402423.

182

183

Gambardella, Alfonzo. (1992) “Competitive advantages from in-house scientific
research: The US pharmaceutical industry in the 1980s,” Research
Policy, Vol. 21, 1992, pp. 391407.

Gambardella, Alfonzo. (1995) Science and Innovation: The US pharmaceutical
industm during the 19803, Cambridge: Cambridge University Press,

1 995.

Grabowski, Henry, and John Vernon. (1981) “The Determinants of Research and
Development Expenditure in the Pharmaceutical Industry,” in Drugs and
Health, Robert B. Helms (editor), Washington DC: The AEI Press, 1981.

Grabowski, Henry, and John Vernon. (1994) “Innovation and Structural Change
in Pharmaceuticals and Biotechnology,” Industrial and Corporate Change,
Vol. 3, No. 2, 1994, pp. 435-449.

Grabowski, Henry, and John Vernon. (1996) “Prospects for returns to
Pharmaceutical R&D under Health Care Reform,” in Competitive
Strategies in the Pharmaceutical Industry, Robert B. Helms (editor),
Washington DC: The AEI Press, 1996.

Grabowski, Henry 6., John M. Vernon, and Lacy Glenn Thomas. (1978)
“Estimating the Effects of Regulation on Innovation: An lntemational
Comparative Analysis of the Pharmaceutical Industry,” Journal of Law and
Economics, Vol. 21, No. 1, April 1978, pp. 133-163.

Griliches, Zvi. (1979) “Issues in Assessing the Contribution of Research and
Development to Productivity Growth,” Bell Journal of Economics, Vol. 10,
Spring 1979, pp. 92416.

Griliches, Zvi. (1984) R&D. Paients and Productivity, (ed.) Chicago: Chicago
University Press, 1984.

Griliches, Zvi. (1986) “Productivity, R&D, and Basic Research at the Firm Level
in the 1970s,” American Economic Review, March 1986, pp. 1141-154.

Griliches, Zvi. (1991) "The Search for R&D Spillovers," NBER Working Paper,
No. 3768, July 1991.

Govemment-University-Industry Research Roundtable. (1989) Science and

Technology in the Academic Enterprise: Status, Trends, and Issues,
Washington DC: National Academy Press, October, 1989.

184

Hall, Bronwyn H., Clint Cummins, Elizabeth S. Laderman, and Joy Mundy.
(1988) “The R&D Master File Documentation,” NBER Technical Working
Paper Series, December 1988.

Hall, Bronwyn H., Zvi Griliches, and Jerry Hausman. (1986) “Patents and R&D:
Is there a Lag?” lntemational Economic Review, Vol. 27, No. 2, June
1986, pp. 265-283.

 

Hausman, Jeny, Bronwyn H. Hall, and Zvi Griliches. (1984) “Econometric
Models for Count Data with an Application to the Patents-R&D
Relationship,” Econometrica, Vol. 52, No. 4, July 1984, pp. 909-938.

Henderson, Rebecca. (1994) “The Evolution of Integrative Capability: Innovation
in Cardiovascular Drug Discovery,” Industrial and Corporate Change, Vol.
3, No. 2, 1994, pp. 607-627.

 

Henderson, Rebecca and Iain Cockbum. (1996) "Scale, Scope and Spillovers:
The Determinants of Research Productivity in the Pharmaceutical
Industry," Rand Journal of Economics, Vol. 27, No. 1, Spring 1996, pp.
3209.

Henderson, Rebecca and Iain Cockbum. (1997) “Public-Private Interaction and
the Productivity of Pharmaceutical Research,” NBER Working Paper
Series, No. 6018, April 1997.

Jaffe, Adam B. (1989) "Real Effects of Academic Research," American Economic
Review, December 1989, pp. 957-970.

Jenson, Elizabeth J. (1987) “Research Expenditures and the Discovery of New
Drugs,” The Journal of Industrial Economics, Vol. 36, No. 1, September
1987, pp. 83-95.

Kmenta, Jan. (1967) “On Estimation of the CES Production Function,”
International Economic Review, Vol. 8, No. 2, June 1967, pp. 180489.

Link, Albert. (1981) “Basic Research and Productivity Increase in Manufacturing:
Additional Evidence,” American Economic Review, Vol. 71, No. 5,
December 1981, pp. 1111-1112.

Mairesse, Jacques and Mohamed Sassenou. (1991) “R&D and Productivity: A
Survey of Econometric Studies at the Firm Level,” NBER Working Paper
Series, Working Paper No. 3666, March 1991.

185

Mansfield, Edwin, John Rapoport, Jerome Schnee, Samuel Wagner, and
Michael Hamburger. (1971) Research and Innovation in the Modern
Comoration, New York: WW. Norton 8 Company Inc., 1971.

Mansfield, Edwin. (1980) “Basic Research and Productivity Increase in
Manufacturing,” American Economic Review, December 1980, pp. 863-
873.

Mansfield, Edwin. (1991) "Academic Research and Industrial Innovation,”
Research Policy, Vol. 20, 1991, pp. 142.

Mansfield, Edwin. ( 1995) "Academic Research Underlying Industrial Innovations:
Sources, Characteristics, and F inancing,” Review of Economics and
Statistics, March 1995, pp. 5505.

Maxwell, Robert A., and Shohreh B. Eckhardt. (1990) Drug Discovegy: A Case
Book and Analysis, Clifton, NJ: Humana Press, 1990.

Gourieroux, C., A. Monfort, and A. Trognon. (1984) “Pseudo Maximum
Likelihood Methods: Applications to Poisson Models,” Econometrica, Vol.
52, 1984, pp. 701-720.

National Institutes of Health. (1989) NIH Data Book 1989, US. Department of
Health and Human Services, 1989.

National Science Foundation. Science and Technology Data Sock,
Washington DC. NSF, 1985.

National Science Foundation. National Patterns of R&D Resources:
1992, by J.E. Jankowski, Jr., NSF 92030 (Washington DC, 1992).

Pakes, Ariel, and Zvi Griliches. (1980) "Patents and R&D at the Firm Level: A
First Look," in R&D, Patents, and Productivity, Zvi Griliches (editor),
Chicago: Chicago University Press, 1984.

Peltzman, Sam. (1973) An Evaluation of Consumer Protection Legislation: The
1962 Drug Amendments,” Journal of Political Economy, Vol. 81, No. 5,
September/October 1 973, pp. 1 0494 091 .

Pharmaceutical Manufacturers Association. "Annual Survey Report," PMA,
Washington DC, various years.

 

186

Romer, Paul M. (1990) "Endogenous Technological Change," Journal of Political
Economy, Vol. 98, 871-8102.

Romer, Paul M. (1994) “The Origins of Endogenous Growth,” Journal of
Economic Perspectives, Vol. 8, No.1, Winter 1994, pp. 1302.

Scolnick, Edward M. (1990) "Basic Research and its Impact on Industrial R&D,"
Research Technology Management, Vol.33, November/December 1990,
pp. 2206.

Scherer, Fredric M., and William S. Comanor. (1996) “Commentary on Part
Three,” in Competitive Strategies in the Pharmaceutical Industgy, Robert
B. Helms (editor), Washington DC: The AEI Press, 1996.

Thomas, Lacy Glen. (1990) “Regulation and firm size: FDA impacts on
innovation,” Rand Journal of Economics, Vol. 24, No. 4, Winter 1990, pp.
497-517.

Trajtenberg, Manuel. (1990) “A Penny for your quotes: patent citations and the
value of innovations,” Rand Journal of Economics, Vol. 21, No. 1, Spring
1990, pp. 172-185.

Trajtenberg, Manuel, Rebecca Henderson, Adam Jaffe. (1992) “Ivory Tower
Versus Corporate Lab: An Empirical Study of Basic Research and
Appropriability,” NBER Working Paper Series, No. 4146, 1992.

US. Department of Commerce, Current Industrial Reports: Pharmaceutical
Preparations except Biologicals. Washington DC: US. Government,
September, 1994.

US. Congress, Office of Technology Assessment, Pharmaceutical R&D: Sosts,
Risks and Rewards, OTA-H022 (Washington, DC: US. Government
Printing Office, February, 1 993).

 

Ward, Michael R., and David Dranove. (1995) “The Vertical Chain of Research
and Development in the Pharmaceutical Industry," Economic Inquiry, Vol.
33, 1995, pp. 70-87.

Wiggins, Steven. (1979) Regulation and Innovation in the Pharmaceutical
Industm, Unpublished MIT Ph.D. Dissertation, 1979.

187

Wiggins, Steven. (1981) “Product Quality Regulation and Drug Introductions,”
Review of Economics and Statistics, Vol. 63, November 1981, pp. 615-
619

Wiggins, Steven. (1983) “The Impact of Regulation on Pharmaceutical Research
Expenditures: A Dynamic Approach,” Economic Inquiry, Vol. 21, January
1983, pp. 115-128.

Wiggins, Steven. (1987) “The Effect of US. Pharmaceutical Regulation on New
Introductions," in Arne Rvngvmposipm on Pharmaceutical Economics,
B. Lindgren (editor), Stockholm, Swedish Institute for Health Economics,
1987.

Wooldridge, Jeffrey M. (1989) “A Computationally Simple Heteroscedasticity and
Serial Correlation Robust Standard Error for the Linear Regression
Model,” Economic Letters, Vol. 31, 1989, pp. 239-243.

Wooldridge, Jeffrey M. (1991) “On the Application of robust, regression-based
diagnostics to models of conditional means and conditional variances,”
Journal of Econometrics, Vol. 47, 1991, pp. 5-46.

Wooldridge, Jeffrey M. (1994) “Multiplicative Panel Data Models Without the
Strict Exogeneity Assumption,” mimeo, Michigan State University,
January 1994.

Wooldridge, Jeffrey M. (1996) "Quasi-Likelihood Methods for Count Data," in
Handbook of Applied Econometrics, M.H. Pesaran and Peter Schmidt
(eds), Volume II, forthcoming.