Make, buy, organize: The interplay between research, external knowledge, and firm structure†

TL;DR: This article explored the interplay between R&D, external knowledge, and organizational structure of a firm's innovation strategy, and found that centralized firms invest more in research, and patent more per research dollar, than decentralized firms.

Abstract: We bridge current streams of innovation research to explore the interplay between R&D, external knowledge, and organizational structure�three elements of a firm's innovation strategy, which we argue should logically be studied together. Using within-firm patent assignment patterns, we develop a novel measure of structure for a large sample of American firms. We find that centralized firms invest more in research, and patent more per R&D dollar, than decentralized firms. Both types access technology via mergers and acquisitions, but their acquisitions differ in terms of frequency, size, and integration. Consistent with our framework, their sources of value creation differ: while centralized firms derive more value from internal R&D, decentralized firms rely more on external knowledge. We discuss how these findings should stimulate more integrative work on theories of innovation. Copyright © 2013 John Wiley & Sons, Ltd.

Summary

1 Introduction

• Using a novel largescale dataset, the authors explore whether rms demonstrate distinct and coherent combinations of R&D organizational structure and knowledge-sourcing strategies, as would be expected given the concatenated predictions of these emerging theories of innovation.
• The authors ndings extend and clarify prior results.
• The implied coherence, however, does not necessarily imply a particular causal structure.

2 Three Pillars of Innovation Strategy

• The authors draw upon important streams in the innovation literature that have explored dyadic relationships between research, external knowledge and structure.
• Firms that invest heavily 3 in basic research have been shown to have more centralized R&D, whereas decentralized R&D, managed by business units and divisions, tends to be more applied and incremental (Argyres and Silverman, 2004).
• Though Argyres and Silverman (2004) hint at this issue by exploring how structure conditions a rm s propensity to "build on.
• Nonetheless, Microsoft has struggled with large acquisitions, such as aQuantive, an online advertising technology rm bought in 2006.
• A rm will perform well (and survive) if organizational structure, internal research, and knowledge acquisition are aligned to support each other if they are coherent (Siggelkow, 2011;.

3 Sample and Data

• The authors match rms to patents by matching assignee names and addresses.
• Ownership data consists of two parts: cross-sectional ownership information from Icarus for 2008, and M&A data from SDC Platinum and Zephyr.
• By matching to SDC, the authors can see that WebTV was purchased in 1997 by Microsoft, then dissolved and absorbed into Microsoft s MSN Networks.

3.1 Patent Assignment as a Proxy for Decentralization

• Whereas a rm s research focus and external knowledge-sourcing activities can be tracked using patents, R&D spending or alliances (Arora, Fosfuri, and Gambardella, 2001; Henderson and Cockburn, 1994; Mowery, Oxley, and Silverman, 1996), the internal organization of R&D is extremely di¢ cult to observe.
• As their new measure does precisely this, it promises to open a new window into the internal organization of rms R&D function.
• As with any proxy measure that lends itself to large-scale empirical analysis, their measure is practical but imperfect, and the authors admit that there are trade-o¤s to consider.
• The authors can see that these patterns are markedly di¤erent.

3.2 Other variable de nitions and measures

• Scienti c publications are a commonly accepted measure of a rm s basic science orientation (Gambardella, 1995; Stern, 2004), and rms such as DuPont, IBM, Merck, and Microsoft, which have traditionally relied upon internally generated innovations, have also tended to produce a great number of scienti c publications.
• The authors use three supplemental measures to probe the robustness of their results to alternate measures, since the literature suggests a variety of empirical proxies.
• Typically, rms that rely upon internal research to fuel growth have higher levels of R&D intensity.
• The authors classify rms according to tertiles of share of patents assigned, and operationalize using categorical variables.
• The authors use the centralized category as their baseline in all regressions.

3.3 Descriptive statistics and evidence of persistence

• Average value of sales in their sample is $3.4 billion, and market value is $5.9 billion (of which $3 billion are in physical assets).
• An analysis of variance (not presented in a table) indicates that between- rm variation accounts for 87%, 81%, and 88% of the total variance for share patents acquired, share patents assigned, and publication intensity, respectively.
• This strongly supports the view that there are reinforcing interactions among the various choices, which may make it di¢ cult or undesirable for rms to abruptly change any of their core strategies.
• This way of assessing persistence obviates the need to control for changes in the environment or changes in rm size or other such variables.
• About 4 percent of all rms move down the distribution, but none drops by more than a single quartile.

3.4.1 Non-parametric analysis

• The authors begin by exploring the relationships between organization, acquisitions, and internal research in Table 2.
• Speci cally, the authors look at how both share patents acquired and publication intensity vary across rms with varying levels of decentralization.
• Centralized rms are smaller than decentralized rms in terms of sales, but they have close to double the number of patents.
• It is important to note that this relationship is also driven by how the rm deals with acquisitions, not simply by how acquisitive the rm is.
• These ndings are important because structural integration is one of the levers that managers use to shape both the nature of research and the structure of the rm (Haspeslagh and Jemison, 1991; Puranam, Singh, and Zollo, 2006).

3.4.2 Parametric analysis

• The authors ndings thus far serve as large-scale validation and extension of earlier studies.
• In unreported speci cations the authors explore the extent to which structure conditions the publications-acquisitions relationship.
• This supports the notion that centralized rms would acquire more nascent external technology to integrate into their existing research (Capron et al., 1998; Karim, 2006), whereas decentralized rms may acquire more developed technology that is closer to being commercialized (e.g., Cisco s "acquire and develop" model).
• The authors also nd a negative relation between publications and share patents acquired (column 9), however, it is not statistically signi cant when controlling for structure.
• Organization structure thus is signi cantly associated with investment in internal research, even after conditioning on external knowledge sourcing.

3.5 Firm Market Value

• The authors have shown that centralization is associated with investment in basic research while decentralization is associated with a strong emphasis on accessing external innovations.
• In fact, this drop is mostly attributed to controlling for sales growth: when excluding sales growth but still controlling for sales, the coe¢ cient estimate on external patents stock is 0.07 and is statistically signi cant at the 1 percent level.
• Among the set of technically diversi ed rms, centralized rms derive considerable value from internal R&D, whereas decentralized rms derive little value from internal R&D (coe¢ cient estimates of 0.16 versus -0.01).
• Conversely, external patents are associated with value in decentralized rms but not in centralized rms (coe¢ cient estimate of 0.10 compared to 0.03).
• 21 Overall, their ndings support the interpretation that rms that rely upon internal research to create value are best served by a centralized organization, in contrast to rms that rely upon external knowledge, especially in large or technically diversi ed rms.

3.6 Robustness

• The authors checked the sensitivity of their results to rm size by excluding very small and very large rms from the sample (lowest and highest sales deciles).
• Though there is a growing literature on the geographical location and management of R&D activities (Leiponen and Helfat, 2011; Singh, 2008), the question of geography is logically distinct from the question of internal organization.
• As Singh (2008) puts it, a rm could have a decentralized formal organization even with relatively small number of R&D locations, while another rm might have a much more centralized organization despite having a much greater number of R&D locations.
• To test whether their set of relationships is driven by the distinction between rms with and without a¢ liates, the authors estimate the main speci cations 22 for a sample that includes only rms with at least one a¢ liate, regardless of whether the a¢ liate patents or not.
• Demonstrating that their results continue to hold also within a sample of rms that have at least one a¢ liate eliminates the concern that the results are driven by comparing rms with and without a¢ liates.

4 Discussion and Conclusion

• External knowledge sourcing, and internal research focus.the authors.
• It is likely that rms that innovate primarily by developing knowledge internally favor investments in more basic, long-term research and do not rely much on incremental research that merely improves existing goods and services.
• Given the role of structure in conditioning the relationship between internal development and external knowledge 24 integration, it is unlikely that innovation strategy can be charted using a simple "make vs. buy" logic, if this does not take into account the complex role played by organizational structure.

Figures

Table 5. Firm Market Value: Diversity and Size

Table 2. Organizational Structure, External Orientation, and Internal Research

Table 1. Summary Statistics for Main Variables

Figure 2: Measures for external orientation vs. internal research across tertiles of decentralization

Figure 1b: An example of a centralized pattern of patent assignment. Top 30 affiliates shown

Table 3. How External Orientation and Internal Research are Related to Organizational Structure

Table 4. Market Value Estimation: Contributions From Internal and External Research Across Varying Organizational Structure

Full text

&61>-:;1<A7.!-66;A4>)61)&61>-:;1<A7.!-66;A4>)61)
$+074):4A75576;$+074):4A75576;
)6)/-5-6<!)8-:; '0):<76)+=4<A#-;-):+0

)3-=A :/)61B-%0-6<-:84)A*-<?--6#-;-):+0@<-:6)4)3-=A :/)61B-%0-6<-:84)A*-<?--6#-;-):+0@<-:6)4
67?4-,/-)6,1:5$<:=+<=:-67?4-,/-)6,1:5$<:=+<=:-
;01;0:7:)
$0):76-4-6B76
=1;#17;
&61>-:;1<A7.!-66;A4>)61)
7447?<01;)6,),,1<176)4?7:3;)<0<<8;:-87;1<7:A=8-66-,=5/5<(8)8-:;
!):<7.<0-=;16-;;,5161;<:)<176)6)/-5-6<)6, 8-:)<176;75576;=;16-;;6)4A<1+;
75576;=;16-;;)6,7:87:)<-755=61+)<176;75576;=;16-;;)?!=*41+#-;876;1*141<A)6,
<01+;75576;)6)/-5-6<6.7:5)<176$A;<-5;75576;)6)/-5-6<$+1-6+-;)6,"=)6<1<)<1>-
-<07,;75576; 8-:)<176;)6,$=884A0)16)6)/-5-6<75576; :/)61B)<176)4-0)>17:)6,
%0-7:A75576;)6,<0-$<:)<-/1+)6)/-5-6<!741+A75576;
#-+755-6,-,1<)<176#-+755-6,-,1<)<176
:7:)-4-6B76$#17;)3-=A :/)61B-%0-6<-:84)A*-<?--6#-;-):+0@<-:6)4
67?4-,/-)6,1:5$<:=+<=:-
$<:)<-/1+)6)/-5-6<7=:6)4


0<<8,@,717:/
;52
%01;8)8-:1;87;<-,)<$+074):4A75576;0<<8;:-87;1<7:A=8-66-,=5/5<(8)8-:;
7:57:-16.7:5)<17684-);-+76<)+<:-87;1<7:A87*7@=8-66-,=

)3-=A :/)61B-%0-6<-:84)A*-<?--6#-;-):+0@<-:6)467?4-,/-)6,)3-=A :/)61B-%0-6<-:84)A*-<?--6#-;-):+0@<-:6)467?4-,/-)6,
1:5$<:=+<=:-1:5$<:=+<=:-
*;<:)+<*;<:)+<
'-*:1,/-+=::-6<;<:-)5;7.1667>)<176:-;-):+0<7-@847:-<0-16<-:84)A*-<?--6#-@<-:6)4
367?4-,/-)6,7:/)61B)<176)4;<:=+<=:-D<0:---4-5-6<;7.)E:5;1667>)<176;<:)<-/A?01+0?-):/=-
;07=4,47/1+)44A*-;<=,1-,<7/-<0-:&;16/?1<016E:58)<-6<);;1/65-6<8)<<-:6;?-,->-478)67>-4
5-);=:-7.;<:=+<=:-.7:)4):/-;)584-7.5-:1+)6E:5;'-E6,<0)<+-6<:)41B-,E:5;16>-;<57:-16
:-;-):+0)6,8)<-6<57:-8-:#,744):<0)6,-+-6<:)41B-,E:5;7<0<A8-;)++-;;<-+06747/A>1)
5-:/-:;)6,)+9=1;1<176;*=<<0-1:)+9=1;1<176;,1..-:16<-:5;7..:-9=-6+A;1B-)6,16<-/:)<176
76;1;<-6<?1<07=:.:)5-?7:3<0-1:;7=:+-;7.>)4=-+:-)<176,1..-:?014-+-6<:)41B-,E:5;,-:1>-57:-
>)4=-.:7516<-:6)4#,-+-6<:)41B-,E:5;:-4A57:-76-@<-:6)4367?4-,/-'-,1;+=;;07?<0-;-
E6,16/;;07=4,;<15=4)<-57:-16<-/:)<1>-?7:376<0-7:1-;7.1667>)<176
-A?7:,;-A?7:,;
,-+-6<:)41B)<1767:/)61B)<176)4;<:=+<=:-5-:/-:;)6,)+9=1;1<176;8)<-6<;#5):3-<>)4=-
1;+18416-;1;+18416-;
=;16-;;,5161;<:)<176)6)/-5-6<)6, 8-:)<176;C=;16-;;6)4A<1+;C=;16-;;)6,7:87:)<-
755=61+)<176;C=;16-;;)?!=*41+#-;876;1*141<A)6,<01+;C)6)/-5-6<6.7:5)<176$A;<-5;C
)6)/-5-6<$+1-6+-;)6,"=)6<1<)<1>--<07,;C 8-:)<176;)6,$=884A0)16)6)/-5-6<C
:/)61B)<176)4-0)>17:)6,%0-7:AC$<:)<-/1+)6)/-5-6<!741+A
%01;27=:6)4):<1+4-1;)>)14)*4-)<$+074):4A75576;0<<8;:-87;1<7:A=8-66-,=5/5<(8)8-:;

Make, Buy, Organize: The Interplay Between Research,
External Knowledge, and Firm Structure

Ashish Arora
y
Sharon Belenzon
z
Luis A. Rios
x
December 11, 2012
Abstract
We bridge current streams of innovation research to explore the interplay between
R&D, external knowledge, an d organizational structure–three elements of a …rm’s inno-
vation strategy which we argue should logically be studied together. Using within-…rm
patent assignment patterns, we develop a novel measure of structure for a large sample of
American …rms. We …nd that centralized …rms invest more in research and patent more
per R&D dollar than decentralized …rms. Both types acc ess technology via mergers and
acquisitions, but their acquisitions di¤er in terms of frequency, size, and integration. Con-
sistent with our framework, their sources of value creation di¤er: while centralized …rms
derive more value from internal R&D, decentralized …rms rely more on external knowl-
edge. We discuss how these …ndings s hould stimulate more integrative work on theories of
innovation.
Keywords: decentralization, organizational structure, mergers and acquisitions, patents,
R&D, market value
JEL Classi…cation: D23 D83 L22
1 Intro duction
How do …rms allocate resources between R&D and external technology in order to maximize
value and drive growth? And in turn, do their ensuing growth trajectories themselves shape
such future resource allocation? Over the past three decades, the …eld of innovation strategy

All authors contributed equally and are listed in alphabetical order.
y
Duke University, Fuqua School of Business, 100 Fuqua Drive, Durham, NC 27708, U.S.A. E-mail:
ashish.arora@duke.edu
z
Duke University, Fuqua School of Business, 100 Fuqua Drive, Durham, NC 27708, U.S.A. E-mail:
sharon.belenzon@duke.edu
x
Corresponding author. Duke University, Fuqua School of Business, 100 Fuqua Drive, Durham, NC 27708,
U.S.A. E-mail: luis.rios@duke.edu
1

has isolated a set of important dyadic relationships in an e¤ort to understand these interrelated
questions. For example, a substantial body of work has advanced our understanding of the
relationship b etween internal R&D and external knowledge, (e.g. Cohen and Levinthal, 1990;
Pisano, 1990; Katz and Allen, 1982). More recently, a small literature on the structure of R&D
has explored how the resource allocation decision is related to the centralization or decentraliza-
tion of R&D (e.g. Argyres and Silverman, 2004; Lerner and Wulf, 2007). Separately, work on
structural integration and resource recon…guration has lo oked at how organizations are shaped
by acquisitions and absorption (e.g., Ahuja and Katila, 2001; Puranam, Singh, and Zollo, 2006;
Karim and Mitchell, 2000).
Surprisingly, there remains little integration of the aforementioned streams. This lack of
synthesis may be due to data constrains, since most work that considers such organizational
dynamics tends to rely on small samples (Karim and Mitchell, 2004, and Cohen and Levinthal,
1990 are notable exceptions). Understandably, researchers must often cede the pursuit of a richer
understanding of strategic interrelationships in exchange for "analytical precision and theoretical
rigor" (Zollo and Singh, 2004). Nonetheless, this lacuna is an important and understudied
limitation, since the very word "organization"–from the Greek organon ("tool, instrument, set
of rules")–denotes a coherent system or unit where interdependent parts work as one. In fact,
a central tenet of organization theory holds that the structures, systems, and processes of a …rm
should be interdependent and must be mutually supportive and coherent (Drazin and Van de
Ven, 1985; Nadler and Tushman, 1997; Rivkin and Siggelkow, 2003; Siggelkow, 2011).
It is the pursuit of this coherence logic that motivates our paper. Using a novel large-
scale dataset, we explore whether …rms demonstrate distinct and coherent combinations of
R&D organizational structure and knowledge-sourcing strategies, as would be expected given
the concatenated predictions of these emerging theories of innovation. We propose that …rms
pursuing a particular approach to innovation (e.g., a strong focus on internal research like
IBM or an "acquire and develop" approach like Cisco) also need a well-matched supporting
organizational structure (e.g., centralized or decentralized). Empirically, we exploit a sample
that includes nearly all patenting public American …rms, and develop a new measure of R&D
organizational structure which uses the ratio of patents assigned to a¢ liates versus corporate
parents as a proxy for the decentralization of R&D. This involves matching 576,052 patents
to 1,014 publicly traded American corporations and their 2,768 a¢ liates. By do cumenting the
types of choices that …rms make, we bridge streams of the literature that have previously studied
dyadic relationships b etween internal and external knowledge sourcing, between organizational
2

structure and innovation, and between acquisitions and structure.
Our …ndings extend and clarify prior results. We …nd evidence that strongly supports the
coherence logic. Knowledge sourcing strategies appear to be systematically related to orga-
nizational structure. Moreover, the market valuation of these knowledge sourcing strategies
strongly correlates with structure. We …nd con…rmatory large-scale evidence that research-
oriented …rms are signi…cantly more centralized than others, consistent with earlier small-sample
…ndings (Hoskisson, et al. 1993; Kay, 1988; Argyres and Silverman, 2004). But we also …nd
that organizational structure seems to strongly condition the relationship between …rms’ re-
search focus and their external knowledge acquisition strategy. Though both centralized and
decentralized …rms acquire external technology, centralized …rms do so less frequently and tend
to make smaller acquisitions. Moreover, they manage acquisitions di¤erently. Acquisitions by
centralized …rms frequently undergo full structural integration (Puranam et al., 2006), whereas
decentralized …rms tend to keep acquisitions as discrete entities.
Importantly, the logic underlying these patterns of choice is re‡ected in measurable di¤er-
ences in the composition of …rms’market value. Whereas centralized …rms draw most of their
intangible value from internal R&D stocks, decentralized …rms derive relatively more value
from externally acquired patents. This …nding is especially strong for large …rms and …rms with
higher technological diversity.
Our results imply that a successful innovation strategy requires careful alignment both be-
tween internal and external knowledge sourcing, and between the internal/external mix of inputs
and organizational structure. The implied coherence, however, does not necessarily imply a par-
ticular causal structure. Establishing causality is important, but given the nature of our data,
it is beyond the scope of our project. Furthermore, it is just as important to develop a fuller
theory of innovation that accounts for the dynamics we highlight in this study. By developing a
new empirical measure and systematically exposing the relationships between internal research,
external knowledge, and organizational structure among a nearly comprehensive set of …rms,
we take an important step towards the development of such theory.
2 Three Pillars of Innovation Strategy
We draw upon important streams in the innovation literature that have explored dyadic relation-
ships between research, external knowledge and structure. The …rst explores how the nature of
research inside a …rm is related to how the activity itself is organized. Firms that invest heavily
3

Frequently asked questions

Q1. What are the future works in "Make, buy, organize: the interplay between research, external knowledge, and firm structure" ?

Future work should further exploit their novel measure of decentralization, as well as time and exogenous variation in order to better explain the systematic patterns the authors have described. For example, given the role of structure in conditioning the relationship between internal development and external knowledge 24 integration, it is unlikely that innovation strategy can be charted using a simple `` make vs. buy '' logic, if this does not take into account the complex role played by organizational structure. 

Q2. What contributions have the authors mentioned in the paper "Make, buy, organize: the interplay between research, external knowledge, and firm structure" ?

The authors bridge current streams of innovation research to explore the interplay between R & D, external knowledge, and organizational structure–three elements of a firm 's innovation strategy which they argue should logically be studied together. The authors discuss how these findings should stimulate more integrative work on theories of innovation. Using within-firm patent assignment patterns, the authors develop a novel measure of structure for a large sample of American firms.