Recalls and Innovation: Own and Competitor Firm Response George P. Ball Jeffrey T. Macher Ariel D. Stern
Working Paper 19-028
Working Paper 19-028
Copyright © 2018, 2019, 2020 by George P. Ball, Jeffrey T. Macher, and Ariel D. Stern.
Working papers are in draft form. This working paper is distributed for purposes of comment and discussion only. It may not be reproduced without permission of the copyright holder. Copies of working papers are available from the author.
Funding for this research was provided in part by Harvard Business School. ADS gratefully acknowledges support from the Kauffman Junior Faculty Fellowship. Any errors or omissions are entirely our own. Author order is alphabetical; all authors contributed equally
Recalls and Innovation: Own and Competitor Firm Response
George P. Ball Indiana University
Jeffrey T. Macher Georgetown University
Ariel D. Stern Harvard Business School
RECALLS AND INNOVATION:
OWN AND COMPETITOR FIRM RESPONSES*
MARCH 2020
George P. Ball Assistant Professor Kelley School of Business Indiana University [email protected] 812-856-0625
Jeffrey T. Macher Professor McDonough School of Business Georgetown University [email protected] 202-687-4793
Ariel D. Stern Associate Professor Harvard Business School Harvard University [email protected] 617-495-2332
Abstract: Innovation is the lifeblood of firms that operate in R&D-intensive industries. While strong in-centives to innovate in these industries exist, product recalls create challenges for impacted firms and op-portunities for their competitors. Using the U.S. medical device industry as the empirical setting, we de-velop predictions and provide evidence that own firm recalls slow innovation activities, while competitor firm recalls accelerate innovation activities. We find that these relationships are most pronounced when recalls and innovation directly overlap in product area. We further unpack the mechanisms that drive these dynamics. Own firm recalls slow innovation activities for all firm types, but competitor firm recalls are only associated with accelerated innovation for broad product scope firms and public firms—indicating that organizational resources and incentives are key determinants of whether firms can effectively capitalize on market opportunities. In post-hoc analyses, we find that longer post-recall submission times do not predict subsequent quality—suggesting observed delays are driven more by distractions than by learning. Firms thus react strategically and rationally to competitor firm failures by speeding innovations to market.
Keywords: Innovation, Recalls, Product Failures, Medical Devices, FDA, Health Care
* The authors thank participants at the NBER Summer Institute, the Industry Studies Association Conference, The Wharton Empirical Operations
Management Conference, the Society of Institutional and Organizational Economics Conference, INFORMS Healthcare, Boston University (Questrom) and the University of Utah (Eccles), as well as Tim Bresnahan, Joshua Krieger, and Michael Toffel for helpful comments and suggestions. Melissa Ouellet and Lila Kelso provided excellent research assistance. ADS gratefully acknowledges support from the Kauffman Junior Faculty Fellowship. Any errors or omissions are entirely our own. Author order is alphabetical; all authors contributed equally.
2
1. INTRODUCTION
Innovation and new product development are the lifeblood of firms in a wide range of research
and development (R&D)-intensive industries, including software, microprocessors, automobiles,
and pharmaceuticals. Firm incentives to innovate in these industries are substantial, given the ben-
efits of innovation and new product development to profitability and survival. Nevertheless, the
impact of faulty or dangerous commercialized products can be ruinous: software bugs can com-
promise sensitive customer data; automotive component failures can result in passenger injury and
death. Such “first-order” effects are salient—harming customers and negatively affecting firm per-
formance (Wowak et al. 2015). But other “second-order” effects also present strategic challenges.
For instance, product failures are heavily publicized and scrutinized events (Jarrell & Peltzman
1985), and may influence subsequent innovation activities by firms and their competitors.
A well-documented effect of product recalls is revenue losses (Thirumalai & Sinha 2011).
When products are found unsafe, sales and distribution are reduced or halted completely
(Krumholz et al. 2007). Recalls can also be costly to manage with external stakeholders, as nega-
tive publicity can amplify sales downturns and generate shareholder losses (Jarrell & Peltzman
1985; Rhee & Haunschild 2006). Operations are also affected as internal resources must be redi-
rected to correct recall-related problems (Ball et al. 2018). We propose that recalls also influence
a previously unexplored but important firm activity: innovation. We examine whether own firm
recalls create internal disruptions that slow subsequent innovation and whether competitor firm
recalls create market opportunities that accelerate subsequent innovation (KC et al. 2013; Krieger
2017). We further consider whether these effects are more pronounced when product recalls and
innovation activities occur in the same product area.
We explore these phenomena by developing predictions and providing empirical evidence of
how innovation activity changes in response to product recalls in the U.S. medical device (“med-
tech”) industry. By leveraging comprehensive device submission and recall data from the Food
and Drug Administration (FDA), we examine the following research questions: first, does the
source of a recall (i.e., own vs. competitor firm) influence subsequent innovation? And second,
does the proximity of a recall (i.e., same vs. different product area) influence subsequent innova-
tion? While recall source and proximity might impact future innovation activities, the effects are
unlikely to be homogenous across the population of firms. We thus consider two firm-level mod-
erating factors: product scope and ownership. Broad product portfolio firms and public firms may
3
possess resource and organizational advantages or have different incentives to innovate in response
to own and competitor recalls, in comparison to narrow product portfolio firms and private firms.
An additional feature of our empirical setting is the ability to examine distinct innovation
types. The FDA categorizes medical device innovation using the concept of product similarity.
When a new device is similar to an approved and marketed product by the same or another firm,
it receives a 510(k) submission-type designation. When a new device is dissimilar to an approved
and marketed product or is of extremely high risk, it receives a Pre-Market Approval (PMA) des-
ignation. We examine our hypothesized relationships using 510(k) submissions for three reasons.1
First, the vast majority of FDA medical device submissions are 510(k) products. Second, 510(k)
innovation projects are more “nimble” than PMA innovation projects, suggesting that they are
more likely to respectively slow down or speed up following own and competitor firm recalls.
Third, 510(k) data are well-suited to examining this phenomenon without confounding explana-
tions: well-established product areas create settings where there is limited uncertainty around tech-
nical viability but clear certainty around firm mistakes. Competitor firm failures in 510(k) product
areas thus constitute information about potential market opportunities; unlike certain competitor
failures in other industries that signal insurmountable risk and thereby lead to innovation retreat
(Krieger 2017). By focusing on 510(k) submissions, we are thus able to explore the boundary
conditions and mechanisms that explain these relationships. In particular, we ground our theory in
organizational resource allocation and incentives more so than learning from failures because we
examine innovation in product areas already proven safe. This approach allows us to generalize to
other recall-intensive industries where market opportunities are driven more by resources and in-
centives than by product area viability – for example, automobiles and consumer products. While
510(k) devices somewhat narrows the scope of the empirical setting, it broadens the relevance and
applicability of our findings to other industries.
We collect data on all med-tech submissions and recalls over 2003-2015. Using matching soft-
ware and novel algorithms, we assign these data to a set of standardized firm names and FDA-
designated product areas and then construct detailed innovation and recall histories that provide
precise definitions of the relevant set of firms and competitors in each product area and across
time. We incorporate these detailed histories into recurrent-event accelerated failure time (AFT)
1 510(k) submissions represent roughly 99 percent of all med-tech devices. PMA submissions are examined in robustness analyses.
4
models to determine how recall source and proximity—as well as product scope and ownership—
affect the timing of firms’ subsequent innovation activities.
Our empirical findings are informative and largely in-line with our hypotheses: (1) own firm
recalls slow subsequent innovation – an effect explained primarily by product area overlap between
recalls and innovation activity; (2) competitor firm recalls quicken subsequent innovation – an
effect that is also explained by product area overlap; and (3) product scope and ownership amplify
the respective innovation responses to recalls.
In post-hoc analysis, we explore two juxtaposing mechanisms that help explain this recall-
innovation relationship. One the one hand, slowing innovation after own firm recalls may be ben-
eficial if it enhances product quality learning. On the other hand, innovation delays after own firm
recalls may have no association with subsequent product quality risk reductions and more likely
represent recall-related distractions. As a corollary, accelerating innovation after competitor firm
recalls may mean rushing products to market to the detriment of quality, or may represent an at-
tempt to capture market share with no tangible quality downside. We find no association between
post-recall innovation timing and subsequent product quality, suggesting: (i) innovation delay
from own firm recalls is due to distractions and not quality learning; and (ii) innovation accelera-
tion from competitor firm recalls is rational and strategic with no discernable quality downside.
We contribute to several research streams in strategy and innovation, as well as research in
product recalls. First, our theoretical lens adds to new product development research (Brown &
Eisenhardt 1995) by examining a largely overlooked but important determinant of innovative ac-
tivity: product failures—in particular, product recalls. Second, we add to the resource-based view
(Barney 1991; Wernerfelt 1984) by finding organizational resources and incentives effectively
moderate the recall and innovation relationship. Third, the empirical approach contributes to prod-
uct recall research by establishing novel ramifications that predict future innovation activity. Our
results suggest that there are additional externalities that are unlikely to be fully captured in the
extant literature related to estimating product recall costs. While this research has identified several
recall effects, such as firm learning (Haunschild & Rhee 2004), market share losses (Jarrell &
Peltzman 1985), and consumer confidence reductions (Rhee & Haunschild 2006), no study of
which we are aware has associated product recalls with subsequent innovation. Fourth, the empir-
5
ical methodology builds upon research that explores innovation and competition at a highly gran-
ular level of analysis. Our comprehensive data and algorithms allow for the dynamic identification
of relevant competitors that vary across firms, across product areas, and over time.
Our theoretical predictions and empirical results have implications to industry and policy. For
managers, we offer evidence that own firm recalls crowd out innovation without providing product
quality learning benefits. Moreover, firms experiencing recalls face a double penalty: internal
challenges related to root cause identification and innovation delay and external challenges related
to competitor firm incentives and innovation acceleration. Recall prevention and correction are
thus likely to be more important than extant research suggests. For regulators, our results demon-
strate that – within med-tech – prior recalls and subsequent innovative activity are closely related.
Greater coordination and information exchange between product approval activities and quality
surveillance activities likely provide benefits that are not fully internalized by the FDA.
2. LITERATURE REVIEW AND HYPOTHESES
We first hypothesize that recall source—own or competitor firm—influences firm innovation ac-
tivity but in opposite directions. We then incorporate recall proximity—same or different product
type—into the recall source hypotheses, arguing that these respective relationships strengthen with
greater product area overlap between recalls and innovation. We then consider product scope and
ownership as moderating factors that further shape these hypothesized relationships.
2.1 RECALL SOURCE
Empirical product recall research is largely divided into two categories: effects and causes. Most
of the research to-date resides in the former category and predominately examines stock market,
market share, and customer loyalty effects. For example, Jarrell and Peltzman (1985) provide the
first major empirical study: using a nine-year panel of automotive and pharmaceutical industry
recalls, the authors determine that the costs incurred by shareholders following recalls exceed the
costs incurred by firms to rework or replace defective products. Similar findings related to recall
costs are documented by Davidson and Worrell (1992) in the automotive industry; by Cheah et al.
(2007) in the pharmaceutical industry; and by Chen et al. (2009) in the consumer products industry.
Empirical research also finds that past recalls influence future recalls (Thirumalai & Sinha 2011),
especially when recalls are voluntarily-initiated by firms (Haunschild & Rhee 2004). A small but
6
growing research stream examines recall predictors in various industry settings: higher R&D in-
tensity (Thirumalai & Sinha 2011), more product and plant variety (Shah et al. 2016; Ball et al.
2018) and adverse inspection outcomes (Ball et al. 2017) are all found predictive of future recalls.
The extant literature is largely silent, however, on whether a relationship between product re-
calls and innovation exists. Some research suggests that firms learn from their own recalls and
make quality improvements, which can accelerate or decelerate subsequent innovation
(Haunschild & Rhee 2004). Other research suggests that firms observe and learn from the pre-
market product development failures of their competitors, which may also influence subsequent
innovation efforts (Krieger 2017). Our empirical setting differs from these studies in that we ex-
amine the impact of post-market product recalls from both own and competitor firm perspectives.
Our approach is thus similar to research that examines the determinants of firm performance once
innovations are already commercialized (Haunschild & Sullivan 2002; Baum & Dahlin 2007; Kim
& Miner 2007), but is distinct in that it considers own and competitor firm failures as predictors—
rather than consequences—of innovation.
Operations research examines both the sources—e.g., supply-chain problems (Demirel et al.
2017; Yang et al. 2009), natural disasters (Kim et al. 2010)—and the solutions—e.g., insurance
(Serpa & Krishnan 2016), buffer inventory (Dong & Tomlin 2012)—to operational disruptions.
These studies in aggregate unsurprisingly find that disruptions are harmful to firm performance. A
narrower research stream examines disruption effects on new product development: Sterman et al.
(1997) finds that when firms focus on quality improvement initiatives, product development speed
suffers; Benner and Tushman (2002) reach similar conclusions. Quality disruptions are thus likely
to negatively affect innovation, as the resources used in activities related to new product develop-
ment and regulatory submission are simultaneously tasked with addressing and correcting prob-
lems across other functional areas—e.g., operations, legal, managerial, public relations, etc.
Product recalls are significant operational disruptions that likely reorient firm attention (Ocasio
1997) and generate unanticipated capacity constraints (Levinthal and Wu, 2010). Beyond manag-
ing customers, firms must identify the root causes of recalls and correct these problems. Technical
and operational resources are redirected and repurposed to implementing the requisite product or
process changes. As one med-tech industry executive we interviewed explained, “recalls are a
shock to the system. Everyone tries to avoid them. But when they happen, everyone works together
7
to recover as quickly as possible. Recall is the preeminent ‘four-letter word’ in the med-tech in-
dustry.” We therefore expect resources and attention to divert away from innovation activities
following own firm recalls. This diversion should increase the time to subsequent innovation ac-
tivity, ceteris paribus.2 We examine the following hypothesis:
H1A: Own firm recalls delay the time to innovation submission, ceteris paribus.
Med-tech firms operate in well-defined product areas, with limited information asymmetries
within and across these areas. Firms are well informed with respect to the product status—both
successes and failures—of competitors (Porter and Heppelmann 2014; Wu 2013; Thirumalai and
Sinha 2011). We suggest that this information awareness influences subsequent firm activities: in
particular, how firms respond in innovation activities to competitor firm recalls.
This argument has strong analogs to pharmaceuticals—a similarly R&D-intensive and FDA-
regulated health care product setting. Pharmaceutical innovation studies have shown product mar-
ket demand shocks that increase profitability generally lead to more product market innovation.
Examples include exogenous patient population increases (Acemoglu and Linn 2004; Dubois et
al. 2015), regulatory rule changes (Finkelstein 2004), and reimbursement modifications (Blume-
Kohut and Sood 2013). With competitor firm recalls, similar demand shocks are experienced by
firms as defective products are removed from the market for some period of time. These shocks
are particularly salient in the med-tech industry, given the extremely high profit margins and large
revenue opportunities available that tend to overwhelm any potential quality concerns of rushing
new product to market.3
Corollaries to this argument are also found in related research. For instance, the R&D races
literature suggests robust product market competition reduces innovation (Dasgupta and Stiglitz
1980). Reduced product market competition via competitor firm failures should have the opposite
effect by accelerating innovation submissions. We therefore suspect that med-tech firms accelerate
innovation activity when competitors experience product recalls, given the large margins and po-
tential market share available. We examine the following hypothesis:
2 While some research indicates firms learn from their own failures (Rerup 2009; Madsen & Desai 2010), these studies do not explore how product
failures affect innovation efforts. We contend that if learning does occur following own firm recalls, is it unlikely to manifest in faster innovation as recall recovery efforts are likely to redirect time, resources and attention away from ongoing innovation activities.
3 Med-tech is documented as one of the highest margin industries, with gross profit margins of 80-95 percent and net profit margins of 20-30 percent on average. See https://www.forbes.com/sites/liyanchen/2015/09/23/the-most-profitable-industries-in-2015/#1c3bf8216b73 and https://www.mddionline.com/three-medical-device-manufacturers-highest-profit-margins.
8
H1B: Competitor firm recalls accelerate the time to innovation submission, ce-teris paribus.
2.2 RECALL PROXIMITY
It is unlikely that own and competitor firm recalls influence innovation equivalently, given product
area differences. We propose that recalls influence innovation efforts differently depending upon
the degree of overlap: in particular, whether recalls occur in the same product area, in a related
product area, or in an unrelated product area to current innovation activity. A clear reason to expect
such a relationship is that med-tech firms – similar to many other product manufacturers – are
typically organized into separate strategic business units (SBUs) comprised of related product ar-
eas (e.g., cardiovascular devices in one SBU; orthopedic devices in another SBU), given technol-
ogy, product and process, and resource overlap.
This organizational approach suggests that when an own firm recall occurs in a particular prod-
uct area, expertise to assist in recall resolution is drawn from the same product area—leveraging a
common set of resources with a high degree of product familiarity. These “direct” resources are
arguably most effective in finding and implementing corrective solutions—given focused experi-
ence (Holmqvist 2004). For example, cardiac stent R&D engineers can be used to assist in the
recovery efforts of cardiac stent recalls, as they are most attuned to the technical details of the
particular product quality problems. Other resources within the firm might also provide recall-
related benefits if they possess similar – but not necessarily directly related – expertise. These
“proximate” resources are also likely effective in facilitating corrective solutions due to similar
product experience (Argote et al. 1990), adaptation experience (Eggers 2012), or knowledge spill-
overs (Henderson and Cockburn 1994). For example, cardiac stent R&D engineers can assist in
cardiac defibrillator recalls, given their recent and related experience. Other firm-level resources,
however, are unlikely to provide much recall-related benefits. These “distant” resources are simply
ineffective to the recall situation and tasks at hand, given unrelated knowledge (Haas and Hansen
2004). An example is the low likelihood that cardiac stent R&D engineers could effectively aid
with hip implant product recalls.
In summary, we suggest that the more proximate own firm recalls are to a particular product
area, the more influence they have on slowing subsequent innovation activities as resources are
redirected and repurposed. We further suggest that the innovation response to competitor firm
9
recalls is similar in product area proximity. In particular, the more proximate competitor firm re-
calls are to current innovation activity, the more influence they have on innovation timing; the less
proximate to innovation activity, the less influence on innovation timing. Direct and proximate
resources within the firm are again redirected and repurposed—but with the intention of acceler-
ating innovation into those product markets where new opportunities are present.
Research related to this proximity proposition is found in other industries. Kim and Miner
(2007) suggest that banks learn vicariously from failures, but the impact depends on geographic
and industry origin conditions: local experience provides survival-enhancing value in comparison
to non-local experience. Aranda et al. (2017) find EU travel companies decrease performance tar-
get weighting and shift learning from comparable regional branches to the focal branch as it ma-
tures but depending upon the successes and failures. Kalnins and Mayer (2004) find pizza restau-
rants enjoy survival benefits from the local experience of franchisees. The net effect of operational
disruptions on innovation is thus likely greatest from proximity: in our setting, in those product
areas where recalls and innovation activity overlap.
We partially substantiate these claims via discussion with a med-tech industry executive, who
described the cardiac device division of her firm as organized into discrete teams with each team
focused on a specific product type: e.g., pacemakers, defibrillators, stents, catheters, etc. She de-
scribed a situation in which a pacemaker recall diverted R&D engineers from new product devel-
opment for several months because manufacturing engineers needed assistance in identifying the
root cause of the product failure. She confirmed that this organizational practice was typical in the
industry, suggesting that resolving own firm product failures and responding to competitor firm
product failures most directly impact those business units and resources that are closely-related.
H2A: Product similarity drives the relationship between own firm recalls and in-novation submission delay, ceteris paribus.
H2B: Product similarity drives the relationship between competitor firm recalls and innovation submission acceleration, ceteris paribus.
2.3 RECALL-INNOVATION MECHANISMS
Recall source and proximity are unlikely to have the same effects on innovation across the popu-
lation of firms, given heterogeneity across firm types. We examine two firm-level moderating
factors that potentially shape how innovation activities are affected: product scope and ownership.
10
Scope economies exist when performance benefits are realized when multiple activities are
conducted jointly in comparison to when these activities are conducted separately (Panzar and
Willig, 1981). The standard analysis of production suggests scope economies result when activities
share inputs with no to limited additional cost. Henderson and Cockburn (1994) identify internal
spillovers of knowledge as another source of returns that results from greater product scope: in
particular, knowledge developed and accumulated in one R&D activity can be transferred to other
activities at little cost but with significant performance benefits.
Scope economies from knowledge spillovers are found in a variety of R&D-intensive indus-
tries, including pharmaceutical drug development (Macher and Boerner 2006; Arora et al. 2009),
beverage production (Brahm et al. 2017), microcomputer software development (Cottrell and
Nault 2004), and semiconductor process development (Macher, 2006), among others. Knowledge
generated in one technology area is not only informative, but also beneficial to the development
of other technology areas. For instance, Nerkar and Roberts (2004) find firms with more diverse
market experience produce higher quality new products; Desai (2015) suggests hospitals more
effectively learn from failures with broad origin distributions. Such actions may require marshal-
ling a large volume of product development resources, however, which subsequently create capac-
ity constraints and limit overall effectiveness (Levinthal and Wu 2010). As Eggers (2012) notes,
broad (vs. focused) and deep (vs. shallow) prior experience shapes the temporal dimensions (e.g.,
prior or concurrent) and speed of subsequent new product development. Timing-related challenges
and capacity constraints can thus affect overall innovation performance.
We suggest that knowledge spillovers resulting from broad product scope should similarly
affect the pattern and pace of firm innovation responses to own firm and competitor firm recalls.
In general, firms with broad product portfolios can more readily shift and reallocate resources—
especially among similar and related product areas—in comparison to firms with narrow product
portfolios. Firms with broad product portfolios, however, might face more requirements than their
more focused brethren with respect to own firm recalls. In particular, recalls in one product area
might have spillover implications for other similar and related product areas. If problem contain-
ment is a concern across product areas, then broad firms may instead slow innovation submissions
more to ensure that any recall issue in a particular product area is not endemic across other related
product areas.
11
At the same time, firms with broad product portfolios can more readily shift resources between
(especially related) product areas in attempts to accelerate innovation submissions to capitalize on
potential market opportunities from competitor firm recalls, in comparison to firms with narrow
product portfolios. The ability to respond rapidly to competitor firm failures is thus arguably great-
est among broad product scope firms, given their resource and organizational advantages. We
therefore examine the following set of hypotheses related to product scope:
H3A: Own firm recalls delay the time to innovation submission more for broad product scope firms, ceteris paribus.
H3B: Competitor firm recalls accelerate the time to innovation submission more for broad product scope firms, ceteris paribus.
Firm incentives to innovate are myriad and include internal pay schemes (Yanadori and
Cui, 2013), complementary assets (Wu et al. 2014), demand conditions (Fabrizio and Thomas
2011), and competitive heterogeneity (Leiblein and Madsen 2009; Boudreau et al. 2011),
among others. We focus here on a single and readily observable factor that potentially affects
innovation incentives within organizations: public versus private ownership.
Management research has long observed that public and private organizations differ in
strategies and management processes (Trostel and Nichols 1982). More recent research con-
siders the effects of going public on financial outcomes—such as stock price and operating
performance (Mikkelson et al. 1997)—or on organizational outcomes—such as survival and
growth (Fischer and Pollack, 2004). Other research considers innovation-related differences
between public and private firms, with mixed findings. Some studies suggest that financial
markets (Benner, 2010) and institutional investors (Kochhar and Parthiban, 1996) condition
public firm responses to innovation, while other studies argue that external market pressures
do not necessarily discourage R&D investment (Hall and Lerner 2010). Related research con-
siders whether external market pressures influence the type and direction of innovation, but
again with mixed results: for instance, innovation quality—as measured by patent counts and
forward patent citations—is found to be higher for private versus public biotechnology firms
(Aggarwal and Hsu 2014); but no evidence of creativity differences is found between public
and private advertising agencies (Von Nordenflycht 2007).
Any innovation effect difference from product recalls likely depends upon underlying fi-
nancial and organizational mechanisms that differ between public and private firms (Wu 2012).
12
First, public firms generally have greater access to financial resources, in comparison to private
firms. Such resources can more effectively fund R&D efforts, increase financial slack (Greve
2007), and enhance visibility (Pollock and Gulati 2007). Second, public firms arguably have
more formal policies, procedures and structures in place, in comparison to private firms. Se-
curities and Exchange Commission (SEC) regulatory requirements, investor relations outreach
programs, and corporate governance protocols suggest more formalized systems are in place
that govern innovative activities. Third, public firms face greater external market pressures to
manage earnings expectations, in comparison to private firms.
Public ownership may therefore amplify the effects of own and competitor recalls: on the
one hand, greater financial resources are realized; on the other hand, more rigid organizational
structure and formalization and external market pressures are present. First consider own firm
recalls: public firms arguably face greater public trust concerns, in comparison to private firms.
If reputation is paramount, then public firms have larger incentives to slow subsequent inno-
vation activities and “get it right” by correcting any and all recall-related problems prior to
undertaking new innovation activities. The greater external market discipline and more rigid
formalization that public firms face suggests innovation following own firm product failures
is slowed, in comparison to private firms. Next consider competitor firm recalls: public firms
have superior abilities and larger incentives to accelerate innovation submissions to capitalize
on the market opportunities present – in comparison to private firms – for two reasons: first,
financial resources availability that can fund such R&D initiatives; and second, external market
pressures that reward efforts to capitalize on potentially profitable market opportunities. We
therefore examine the following set of hypotheses related to ownership:
H4A: Own firm recalls delay the time to innovation submission more for public firms, ceteris paribus.
H4B: Competitor firm recalls accelerate the time to innovation submission more for pubic firms, ceteris paribus.
3. EMPIRICS
3.1 SETTING
Innovative activity and product recalls in the med-tech industry have increased in recent years:
over 2003-2015, the number of FDA device submissions increased by 11 percent while the number
13
of device recalls increased by nearly 50 percent. Understanding how product failures potentially
impact future innovation efforts in this setting is not only critical for managers and firms, but also
important for investors, regulators and policy makers, and health care providers and patients.
Medical devices are regulated by the Center for Devices and Radiological Health (CDRH)
within the FDA via two primary approaches: pre-market gatekeeper and post-market regulator. In
its role as pre-market gatekeeper, CDRH reviews new product submissions to determine whether
these devices are safe and effective for use in and by patients. Federal statutes make it illegal to
market and sell a medical device in the U.S. without regulatory approval. CDRH assigns medical
devices to product areas based on intended use and to regulatory approval pathways based primar-
ily on approved product similarity. Devices are categorized by regulatory medical specialty areas
(product classes) and product codes (products types). Same product class devices are related by
their area of application (e.g., cardiology; orthopedics) and largely correspond to med-tech firm
business units.4 Same product type devices (e.g., coronary stents) are effective substitutes as they
serve the same function, are used in identical ways, and are reviewed by the same regulators. Fig-
ure 1 depicts how FDA devices delineate into mutually exclusive product classes and product types
(denoted by three letter codes).
In its role as post-market regulator, CDRH ensures that approved and marketed devices per-
form in a safe and effective manner and present no unnecessary patient risk. CDRH performs on-
going surveillance of approved products for continued safety and effectiveness. When medical
devices are defective or pose health risks—due to design, manufacturing or technological factors—
med-tech firms and healthcare facilities (e.g., hospitals, physician offices) are required to report
this information to CDRH. When a systematic pattern of product defects or safety issues arises,
firms must initiate voluntary recalls that are overseen by the FDA. In cases where product safety
concerns emerge, federal statutes mandate devices that “present a risk of injury, gross deception,
or are otherwise defective” be recalled and removed from the market by the infringing firm.5 Re-
call classifications range from Class I (most severe) to Class II (moderately severe) and Class III
(least severe). Class I recalls are for what FDA terms “violative” medical device failures that have
a reasonable probability of serious adverse health consequences or death. An example would be a
faulty implantable heart valve. Class II recalls occur when the use of a device may cause medically
4 Our review of the top-ten U.S. medical device firms by revenue indicate each is organized by product class.5 While the recalls in our data are all voluntarily-initiated, FDA maintains the legal authority to mandate recalls but seldom does. Both market
corrections and removals are considered as recalls by FDA because they entail modifications to approved and marketed products.
14
reversible adverse health consequences, such as a malfunctioning hearing aid. Class III recalls
occur when a device is unlikely to cause adverse health consequences, but should nevertheless be
corrected, such as a product labeling error. We examine Class I and II recalls in the main empirical
analysis as they constitute the most significant patient health risks and firm disruptions, but incor-
porate Class III recalls in robustness analysis.
3.2 DATA
We download FDA recall data over 2003-2015 and FDA device submission data over 2003-2016:
the respective windows for which these data are available. We assign each recall and submission
to a standardized firm name based on the information included. We clean firm names in each
database and then match between databases.6
Innovation Submissions – The 510(k) Clearance data provide detailed information, including
unique identification numbers, submission and approval dates, applicant firm identities, and device
class and product details.7
Recalls – The recall data include unique recall event numbers, severity classifications, dates,
and applicant firms. We use a digital text-scraping program to identify FDA submission numbers
in recall reports that are not included in the downloaded data, which allows recalls to link directly
to product classes and product types, to firms, and over time.
3.3 VARIABLES
Our empirical analysis considers how product recalls affect innovation submission times. Model
estimation is from the own firm perspective. As in the peer effects literature (Sacerdote, 2001;
2014), data elements are reflexive: that is, when Firm B is a competitor of Firm A, Firm A is a
competitor of Firm B. Our empirical setting differs, however, in that we consider dynamic com-
petitor definitions. Our coding algorithms require that firms must have a submission or a recall in
at least one overlapping product area within a five-year window to be considered competitors.8
Dependent Variables – Healthcare innovation research considers myriad determinants, includ-
ing how potential market size positively predicts innovation in pharmaceutical markets (Acemoglu
6 Firm names are cleaned and matched using matchIT, a software package for “fuzzy matching” of text strings. matchIT creates match keys to
search for duplicates and grades matching records. 7 We also download the Pre-Market Approval (PMA) database from the FDA website. The PMA data provide similarly detailed information as
the 510(k) Clearance data. We examine these data in robustness analysis. 8 We utilize five-year windows because the average medtech device life cycle is roughly three years and the average product development cycle
is roughly two years (Wizemann 2010; Nazarian 2009). Our empirical results are robust, however, to other time windows.
15
and Linn 2004; Dubois et. al. 2015) and how expected time-to-market shapes R&D activities and
new drug commercialization (Budish et. al. 2015). In the specific context of FDA regulatory ap-
proval processes, Carpenter et al. (2010) examine FDA review times for new pharmaceutical prod-
ucts and Stern (2017) examines these dynamics in the context of new high-risk medical devices.
In the med-tech setting, management scholars have studied other innovation determinants and firm
performance (Chatterji 2009, Chatterji and Fabrizio 2016, Wu 2013). We are not aware of any
empirical studies, however, that use product recalls to predict subsequent innovation. In the tradi-
tion of using “shocks” to examine healthcare innovation effects (Blume-Kohout and Sood 2013;
Krieger et al. 2018; and Krieger 2017), we consider own firm recalls as negative shocks and com-
petitor firm recalls as positive shocks. The dependent variable measures the time since last new
product submission in elapsed calendar days. Given its construction, the data include established
incumbents over de novo entrants.
Independent Variables – The main independent variables are recall counts defined by source
and proximity. For example, REC: Own tallies all own firm recalls and REC: Competitor does the
same for competitor firm recalls. Recalls are disaggregated into mutually exclusive categories us-
ing FDA-defined product classes and types: (1) Direct – same product class and product type; (2)
Proximate – same product class but different product type; and (3) Distant – different product
classes and types. For example: REC: Competitor-Direct is a count of competitor firm recalls in
the same product class and type; REC: Own-Proximate is a count of own firm recalls in the same
product class but different product type; and REC: Competitor-Distant is a count of competitor
firm recalls in different product classes and types. Our primary analysis considers recall counts
over 24-month windows prior to the submission event—reflecting the typical med-tech product
development timeline (Nazarian 2009). We examine 36-month estimation models in robustness
tests. Because the distributions are left-skewed, recall counts are log transformed. Data are avail-
able in 2003, but recall time windows determine the relevant years and data used in empirical
analyses. The main analysis thus narrows the sample to 2005–2016, inclusive.
Moderating variables are firm-level measures of product scope and ownership. Product Scope
is a count of the number of active product areas for a firm over a five-year window and is log
transformed. Public (Private) is an indicator of whether the firm is publicly-held (privately-held).
Table 1 describes the independent and moderator variables in more detail.
16
Other Variables – Research suggests that innovation and recall propensities can be partially
explained by products, by firms, and over time (Thirumalai and Sinha 2011; Wowak et al. 2015;
Shah et al. 2016; Ball et al. 2017). We therefore include product, firm and year fixed effects in all
estimations. We also include controls for own and competitor firm submission counts at the prod-
uct class-level over five-year windows. These submission controls are of the form SUB: Firm-
Class, where SUB indicates submissions; Firm designates own or competitor; and Class designates
same or different product class with respect to current innovation activity.
Certain robustness estimations include additional product-level measures: Product Age is the
age of the product area—defined as the time in days between the establishment of a product type
and the submission date in question; Product Competition is the number of active competitors in
the product area over a five-year window. Other robustness estimations include additional
measures only available for public firms: Employees is total number of employees; Revenues is
total revenues; R&D is total R&D spending; and R&D Intensity is R&D divided by Revenue.
3.4 DESCRIPTIVE STATISTICS
Table 2 provides descriptive statistics for the variables used in estimation. The average elapsed
time between submissions is roughly 186 days. Firms experience 11.42 recalls on average, while
their competitors experience 262.38 recalls on average. Most own and competitor firm recalls oc-
cur in different product classes and types than current innovative activity. The average med-tech
firm has roughly 22 unique products, but substantial heterogeneity is present. About half of the
observations are from public firms. Table 3 provides pairwise correlations for the dependent, in-
dependent, and moderator variables. Multicollinearity is not a concern, as recall measures with
high correlations are not used in the same estimation. Appendix Tables A-1 and A-2 respectively
provide similar descriptive and correlation statistics for PMA data used in robustness analysis.
3.5 EMPIRICAL METHODOLOGY
Our empirical methodology accounts for the unique characteristics of the industry setting and our
research questions. The data consist of all med-tech firms that are active in innovation submissions
and experience recalls within at least one product type. Our empirical objective is to examine how
recalls—by source and proximity—and firm-level moderating factors—in particular, product
scope and ownership—impact innovation as measured by the elapsed calendar time between FDA
product submissions. We therefore implement survival analysis using Accelerated Failure Time
17
(AFT) models due to their suitability to our setting and enhanced interpretability of the resulting
estimates. Other prevalent survival models, such as the Cox Proportional Hazard, estimate the
instantaneous hazard rate of an event occurrence at any point in time. An advantage of AFT models
over Cox models is that estimates can be used to examine how independent variable changes in-
fluence time to an event. AFT models have also been used by other innovation research that models
time-to-event data (Harhoff and Wagner 2009). In our empirical setting, the AFT model estimates
the time to innovation submission for firms based on observable factors that change over time.
Because firms in our data experience multiple submissions and recalls, we employ recurrent-
event AFT models with an exponential distribution and clustered standard errors at the firm level
(Harhoff and Wagner 2009; Box-Steffensmeier and Jones 2004).9 The AFT estimation model fol-
lows the following generalized equation:
𝐿𝑜𝑔(𝑡&) = 𝛽* + 𝛽𝑋& +𝑢&
where 𝑡& is the elapsed time between submissions for firm i, 𝛽* is an intercept term, 𝛽 is a vector
of regression coefficients, 𝑋& is a vector of covariates, and 𝑢& is an error term with an exponential
distribution. The number of observations is the sum of recalls and submissions observed.
4. EMPIRICAL RESULTS
4.1 INTERPRETING COEFFICIENTS
The interpretation of estimated coefficients in AFT models is as follows: a one percent change in
a logged recall count is associated with a (0.01×(expβ-1) multiplicative effect on the time to the
event of interest (Harhoff and Wagner, 2009; Stock and Watson 2012; Wooldridge 2010). A pos-
itive (negative) β coefficient signifies a longer (shorter) time to the event, which in our empirical
setting translates to a slower (faster) time to submission. As recall counts are highly varied and
dependent upon category and context, we consider two meaningful time to submission bench-
marks: (i) a one standard deviation recall increase; and (ii) a single recall increase. These bench-
marks not only show how reasonable variation in recalls influence innovation activity, but also
demonstrate how (in some cases) a single recall can have meaningful impact.
9 We use STREG with the dist(exp) time option in STATA. Results are robust to the other available distribution choices, including Weibull and
Lognormal.
18
4.2 MAIN RESULTS
Table 4 provides the main AFT model results that test the hypotheses. All models include product,
firm, and year fixed effects and submission controls: Model (1) includes own and competitor firm
recalls; Model (2) disaggregates these recalls by product class and type; Model (3) adds Product
Scope to Model (2); Models (4) and (5) respectively examine focused and broad product scope
firms; and Models (6) and (7) respectively examine private and public firms. Model (1) indicates
that: (i) REC: Own does not significantly impact innovation submission time (β = 0.037; p =
0.340), failing to support Hypothesis H1A; and (ii) REC: Competitor has a negative and significant
effect on innovation submission time (β = -0.084; p = 0.005), lending support to Hypothesis H1B.
Model (2) indicates that: (i) REC: Own-Direct is positive and significant (b = 0.240; p = 0.000);
(ii) REC: Own-Proximate is positive and moderately significant (b = 0.070; p = 0.088); and (iii)
REC: Own-Distant is not significant (b = -0.050; p = 0.128)—indicating that own firm recalls
delay innovation more when they directly overlap in product area and providing strong support for
Hypothesis H2A. Model (2) also indicates that: (i) REC: Competitor-Direct is negative and sig-
nificant (b = -0.050; p = 0.002); and (ii) REC: Competitor-Proximate is not significant (b = 0.021;
p = 0.132). These results suggest that competitor firm recalls accelerate innovation submission
times more when they directly overlap in product area, supporting Hypothesis H2B. We also find
that REC: Competitor-Distant is negative and significant (β = -0.071; p = 0.005): a result that
suggests more competitor firm recalls in unrelated product areas accelerate innovation activity
within a product area. We consider this result in more detail via Models (3) – (5).
Model (3) adds a logged measure of firm product scope. Several findings are noteworthy. First,
the main independent variables are largely similar in terms of magnitude and statistical signifi-
cance. Second, Product Scope is negative and significant (β = -0.365; p = 0.000): firms with
broader product portfolios appear better able to accelerate innovation submissions on average—
indicating that product scope impacts the speed of innovation and suggesting further exploration
of is warranted. Third, REC: Competitor-Direct is negative and significant (β = -0.047; p = 0.004),
while REC: Competitor-Proximate is positive and significant (β = 0.035; p = 0.014). These results
indicate how firms respond to competitor firm recalls: accelerating innovation activity in product
areas where competitors falter by actively engaging direct resources (i.e., same class and same
type) and by repurposing proximate resources (i.e., same class but different type).
19
To more directly examine Hypotheses 3A/B, we disaggregate the sample by product scope
terciles: Model (4) considers the bottom tercile (“focused” firms; ≤ 9 products); Model (5) consid-
ers the top tercile (“broad” firms; ≥ 38 products). The results suggest the following: First, REC:
Own-Direct maintains sign and significance across terciles, but magnitudes markedly increase for
broad firms. There is a more dramatic slowing in innovation submission time for own firm recalls
by broad firms—in support of Hypothesis H3A. Second, broad firms accelerate innovation sub-
missions with direct competitor firm recalls (REC: Competitor-Direct); slow innovation submis-
sions with proximate competitor firm recalls (REC: Competitor-Proximate); and accelerate inno-
vation submissions with distant competitor firm recalls (REC: Competitor-Distant). In comparison
to narrow firms, the results suggest broad firms respond to competitors’ failures by engaging direct
resources, repurposing proximate resources, and endowing distant resources—perhaps to focus
new product development in areas that are perceived to be of lower risk in the wake of these recalls.
These results suggest that product scope moderates the effects of competitor firm recalls on inno-
vative submission times, supporting Hypothesis H3B.
Models (6) and (7) examine the moderating effect of firm ownership by respectively consider-
ing private and public firms. The results indicate privately-held firms are negatively and signifi-
cantly impacted by REC: Own-Direct (β = 0.191; p = 0.000)—innovation submissions times are
longer for private firms when recalls and innovation directly overlap. All other recall counts do
not reach statistical significance, however, suggesting that private firms face resource constraints,
organizational challenges, and/or lack strong incentives to respond to recalls in innovation activi-
ties. The public firm results differ in important ways: REC: Own-Direct is positive and significant
(β= 0.269; p = 0.000), but REC: Own-Proximate is positive and moderately significant (β= 0.075;
p = 0.069). Innovation submissions times thus lengthen for public firms when recalls directly and
proximately overlap with current innovation activity. The difference in coefficient magnitudes for
REC: Own-Direct between private and public firms, moreover, lends support Hypothesis H4A.
Further, competitor firm recalls have large effects on public firms: (i) REC: Competitor-Direct is
negative and significant (β = -0.067; p = 0.000); (ii) REC: Competitor-Proximate is positive and
significant (β = 0.047; p = 0.006); and (iii) REC: Competitor-Distant is not significant (β = -0.010;
p = 0.684). In comparison to private firms, public firms appear to accelerate innovation responses
to competitor firm recalls by engaging direct resources and repurposing proximate resources to
better capitalize on market opportunities. These results provide strong support to Hypothesis H4B.
20
Table 5 shows the effects of a one standard deviation increase and a unit increase in recalls on
innovation submission times using the Model (1) and (2) results in Table 4. Innovation submission
times are largest for disaggregated recalls. We discuss these economic effects in detail below.
4.3 ROBUSTNESS RESULTS
We present several robustness tests to our main results. We first demonstrate that the timeframe
for past recall counts does not affect the main empirical results much. Appendix Table A-3 repli-
cates the Table 4 results using 36-month windows in place of 24-month windows. The results are
largely consistent, but some coefficient estimates see minor statistical significance degradation
with longer time windows. The relationship between past recalls and future innovation submission
thus appears to weaken temporally—a sensible and intuitive result.
We next demonstrate robustness around key model estimation assumptions and variable inclu-
sions via Appendix Table A-4. Model (1) replicates the Model (2) results of Table 4 for comparison
purposes. Model (2) adds product class X year fixed effects as additional controls. The results are
robust, with no loss in sign or significance to the main variables of interest. Model (3) instead adds
logged Product Age as a control measure. The main variables of interest are robust to its inclusion.
Product Age is positive and significant (β = 0.027; p = 0.024), indicating older product areas have
longer submission delays—likely a feature of well-established product areas, as fewer updates are
introduced later in their existence. Model (4) instead includes logged Product Competition (i.e.,
the number of firms in a product type) as a control measure. The main variables of interest are
robust to its inclusion. Product Competition is negative and significant (β = -0.121; p = 0.011),
suggesting more competitive product areas are associated with accelerated innovation submission
times—a sensible result given the setting. Finally, Model (5) replaces Class 1-2 recall counts with
all class (i.e., 1-3) recall counts. The findings are robust to these variable changes.
We next demonstrate that the inclusion of additional public firm controls do not change the
results much via Appendix Table A-5. Models (1) and (2) respectively replicate Models (6) and
(7) in Table 4 for comparison purposes. Models (3) – (6) add respective controls to Model (2):
Revenue, Employees, R&D, and R&D Intensity. The signs and significance levels of the main var-
iables remain, although none of the added controls are statistically significant.
We next consider reverse causality via Appendix Table A-6: in particular, whether the results
are biased by a potential association between past submissions and future recalls. If observed re-
calls are driven by past innovation efforts then the results are potentially confounded (Ingram and
21
Baum 1997). We implement propensity score matching (PSM) to examine this possibility. PSM
models use all independent and control variables to predict the propensity of receiving a certain
treatment, and then match observations according to equivalent propensities. Once matched, the
model examines outcomes of receiving a treatment compared to not receiving a treatment. In our
setting, we estimate how a submission in the prior 24 months (i.e., the treatment) compared to no
submission in the prior 24 months (i.e., no treatment) for two comparable observations—which
are expected to otherwise be equivalent—influences recall likelihoods. The results suggest that
reverse causality does not drive the results, as the treatment effect is not statistically significant.
Prior submissions do not appear to predict the probability that firms experience recalls, supporting
the interpretation that recalls drive subsequent submission behavior and not vice versa.
We finally consider how innovation submission activity is affected by recalls under a different
submission pathway. Appendix Table A-7 replicates the Table 4 results, replacing 510(k) data with
PMA data. The number of observations is reduced, which subsequently limits the controls to year
and product type fixed effects and submission counts (i.e., no firm fixed effects). The results are
similar in direction and magnitude to those in the main estimation, but the number of observations
suggests caution in interpreting these results in significant detail.
4.4 POST-HOC RESULTS
We implement post-hoc analysis to examine whether post-recall submission times are predictive
of subsequent product quality—that is, do longer times between recalls and subsequent innovation
submissions result in product quality learning and, hence, safer products? To conduct this analysis,
we make the dependent variable (510(k) Time) an independent variable and create a new dependent
variable that measures the time from submission to first recall for that specific product. We then
implement the same AFT model estimation approach: positive (negative) β coefficients signify
longer (shorter) times to the observed event of interest, which translates to longer (shorter) times
to subsequent recall. A positive and significant coefficient indicates longer submission times lead
to longer times to recall and is suggestive of firm learning and quality improvement, whereas a
negative and significant coefficient indicates longer submission times lead to shorter times to recall
and is suggestive of firm mistakes and rushing ineffective products to market.
Table 6 provides the estimation results. Model (1) considers only those innovation submissions
that experience subsequent recalls; Model (2) considers all innovation submissions regardless of
whether they experience subsequent recalls. Coefficient estimates are near zero and insignificant
22
in each model, suggesting two managerial implications. First, observed submission delays follow-
ing own firm recalls do not lead to substantive learning or quality improvement, and are instead
driven by recall-related distractions. Second, observed submission accelerations following com-
petitor firm recalls do not reduce product quality and, as such, are rational and strategic attempts
to gain market share when competitors stumble.
5. CONCLUSION
5.1 DISCUSSION
This study examines how product recalls influence subsequent innovation activity. It first exam-
ines how recall source (own vs. competitor firms) and recall proximity (same vs. different product
area) to current innovative activity influence this relationship. It then unpacks the mechanisms and
boundary conditions that shape these relationships. The theoretical arguments and empirical find-
ings contribute to several literatures and have practical implications to both firms and regulators.
First, we demonstrate that recalls affect firm innovation—in particular, when they are more
proximate. Own firm recalls that directly overlap with current innovation activity slow subsequent
innovation submissions, while competitor firm recalls that directly overlap accelerate subsequent
innovation submissions. These findings enhance the body of literature that examines the conse-
quences of recalls (Haunschild and Rhee 2004; Thirumalai and Sinha 2011; Jarrell and Peltzman
1985) by uncovering a highly relevant but largely understudied ramification: recalls by both own
and competitor firms impact future innovation, but in opposite directions. Our findings also expand
upon previous studies that explore factors that influence firm innovation incentives in health care
product markets (Acemoglu and Linn 2004; Dubois et. al. 2015; Budish et. al. 2015; Carpenter et
al. 2010; Stern 2017). We find that in med-tech, in particular, the often protracted “shocks” in-
duced by product recalls drive meaningful responses by competitor firms.
Second, the estimates imply that recalls have meaningful economic effects on med-tech firms.
Table 5 estimates the delay or acceleration effects (in days) of a standard deviation increase and a
unit increase in recalls by source and by proximity. For own firm recalls: a one standard deviation
increase in REC: Own-Direct delays subsequent innovation by nearly five months (169.2 days)
and a single recall in this category slows innovation by over a month (38.3 days). Both measures
imply non-trivial impacts on future innovation and revenue. It is likely that innovation efforts are
more sensitive to direct recalls because product development is often highly specialized. While
23
such an organizational approach offers significant benefits—including greater goal alignment and
focus—there may be a potential downside: when product failures occur, their resolution can tax
functional expertise, which in turn slows subsequent innovation activities. This finding is both
informative and logical when considering the empirical context. For competitor firm recalls: a
standard deviation increase in REC: Competitor is associated with a roughly 15-day acceleration
in innovation submission times and a standard deviation increase in REC: Competitor-Direct ac-
celerates subsequent innovation by 12 days. Firms appear to accelerate innovation submission ac-
tivities to take advantage of market opportunities—not only by actively engaging resources in
directly-related product areas, but also by repurposing resources in proximately-related product
areas. These results are further bolstered by the fact that the largest responses to competitor failures
are observed among broad firms—i.e., those arguably best-positioned—and by public firms—i.e.,
those arguably with the strongest incentives.
Third, we consider two mechanisms that drive these results. The first is product scope. Our
results suggest that own firm recalls slow innovation submission activity for all firms regardless
of product portfolio size. Responses to competitor firm recalls, however, are unique to firms with
broad product scope. In particular, these firms more actively engage direct resources and repurpose
proximate resources to accelerate innovation submissions because they are better positioned to
take such strategic approaches. The second is firm ownership. Our results again suggest that pri-
vate and public firms experience slowed innovation following own firm recalls. Responses to com-
petitor firm recalls, however, are unique to public firms: similar to broad product scope firms (with
which there is significant overlap), these firms more actively engage direct resources and repur-
pose proximate resources to accelerate innovation submissions. Given their greater resources and
external market incentives to pursue these opportunities, public firms respond accordingly.
Fourth, we demonstrate robustness of our main empirical results to a number of different spec-
ifications, control variables, submission data, and explanations. We examine alternative time win-
dows, alternative estimation approaches, additional product area control measures, and additional
public firm controls. In each robustness test, the key findings remain. We further consider alterna-
tive explanations and show: (i) there is no association between past submissions and future recalls,
reducing concerns around reverse causality; and (ii) post-recall submission times are not predictive
of subsequent product quality, suggesting distractions slow innovation responses for own firm
recalls and market opportunities accelerate innovation responses for competitor firm recalls.
24
Our results also have important implications for firms and regulators. For firms, this study
suggests a double penalty associated with product failures: recalls not only slow down own firm
innovation activity, but also accelerate competitor firm innovation activity. These results highlight
an additional reason why firms should seek to avoid product failures in the first place. The temp-
tation to divert resources from innovation activities to help resolve product quality problems is no
doubt strong, but doing so may simply fix a present problem at the cost of future innovation and
subsequent revenue and profits. More concerning perhaps is the fact that product quality issues
represent opportunities for competitor firms. A medical device industry executive suggested two
actions med-tech firms might take in response to our findings. First, creating dedicated product
recall recovery teams that retain significant and broad product area expertise, helping to insulate
new product development from product recall fire-fighting efforts. Second, establishing competi-
tor firm recall surveillance tools, which could integrate market knowledge and take advantage of
such opportunities as they emerge. We note that some med-tech firms are already pursuing such
strategic responses to recalls—whether structured or otherwise.
Regulators – such as the FDA – can also draw insight from this study. Our findings highlight
the link between innovation submissions and recalls. It may benefit regulators to establish formal-
ized coordination and information exchange mechanisms between pre-market product submission
and approval activities and post-market surveillance and compliance activities. In our discussions
with senior FDA personnel as a part of this study, we learned that such coordination does not
currently exist. Among other opportunities, implementing organizational changes to facilitate pre-
and post-market information exchange may help regulators better predict the timing and nature of
future regulatory submissions in those products with potential quality concerns.
5.2 LIMITATIONS
Certain limitations and caveats related to our empirical setting, variables, and econometric analysis
are worth noting. First, we examine a single industry and its innovation- and recall-related activi-
ties. While such a focus potentially limits the generalizability of our findings and implications, it
simultaneously offers greater precision in our measures and estimation. Additionally, many R&D-
intensive industries are subject to product failures and recalls, which suggests that our findings
likely have broad applicability. Second, our primary predictor is product recalls, but other negative
shocks exist within the med-tech industry. These include non-recall-related malfunctions and man-
ufacturing compliance issues. Third, our recall measures are based on source and proximity and
25
potentially do not capture other relevant features that are unavailable in our data, such as size (i.e.,
number of products), media coverage or financial costs. We nevertheless find that the recall char-
acteristics that we do observe are of substantial importance in predicting the forward-looking in-
novation activities of med-tech firms. Fourth, we only examine innovation submissions for those
devices that receive regulatory approval. While selection is potentially a concern, actual rejection
rates for these submissions are minimal.
5.3 CONTRIBUTIONS
Product failures such as recalls present major and often public challenges for firms. Empirical
research has examined both the external market effects and the internal causes or leading indicators
of recalls. Despite these contributions, a dearth of research explicitly examines the relationship
between product failures and innovation. Using detailed firm-level FDA data, we address this gap
by examining the effects of product recalls on subsequent innovation.
We provide novel evidence that competitor firm recalls accelerate subsequent innovation sub-
missions, shedding new light on firms’ strategic responses to their rivals’ product failures. Further,
we demonstrate that more proximate recalls to innovation activity appear to lead to more dramatic
effects in several contexts, indicating the nature of product failures and their relationship to current
R&D efforts is crucial for understanding how and when recalls impact innovation. Moreover, we
document how firm-level factors—such as product scope and ownership—shape organizational
strategies and incentives to respond to recalls. The fact that delayed product submissions following
recalls do not lead to higher quality products also suggests that there is limited evidence of firm
learning in this setting and, as a corollary, responding to competitor firm recalls by accelerating
innovation submissions is both rational and strategic.
Our findings make several contributions to strategy and innovation research. Arguably most
important, we examine product recalls as a largely overlooked but important determinant of inno-
vative activity by R&D-focused firms. No studies of which we are aware have considered the
impact of post-market product failures on subsequent innovation activity by firms and/or their
competitors. Our results suggest that there are additional externalities associated with product re-
calls that are unlikely to be fully captured in the existing literature related to estimating the costs
of product failures. Firms experiencing product recalls therefore face a host of challenges in the
26
form of internal disruptions and opportunistic responses by their closest and most nimble compet-
itors. Product failure prevention and remediation activities are thus likely to be more valuable for
managers than previously thought.
References Acemoglu, D. and Linn, J. 2004. Market size in innovation: theory and evidence from the pharmaceutical
industry. The Quarterly Journal of Economics. 119(3): 1049-1090.
Aggarwal, V.A. and Hsu, D.H. 2014. Entrepreneurial exits and innovation. Management Science. 60(4): 867-887.
Aranda, C., Arellano, J. and Davila, A. 2017. Organizational learning in target setting. Academy of Man-agement Journal. 60(3): 1189-1211.
Argote, L., Beckman, S.L. and Epple, D. 1990. The persistence and transfer of learning in industrial set-tings. Management Science. 36(2): 140-154.
Arora, A., Gambardella, A., Magazzini, L. and Pammolli, F. 2009. A breath of fresh air? Firm type, scale, scope, and selection effects in drug development. Management Science: 55(10): 1638-1653.
Ball, G., Shah, R. and Wowak, K. D. 2018. Product competition, managerial discretion, and manufacturing recalls in the US pharmaceutical industry. Journal of Operations Management. 58-59: 59-72.
Ball, G., Siemsen, E. and Shah, R. 2017. Do plant inspections predict future quality? The role of investigator experience. Manufacturing & Service Operations Management. 19(4): 534-550.
Barney, J.B. 1991. Firm resources and sustainable competitive advantage. Journal of Management 17: 99-120.
Baum, J.A. and Dahlin, K.B. 2007. Aspiration performance and railroads’ patterns of learning from train wrecks and crashes. Organization Science. 18(3): 368-385.
Benner, M.J. 2010. Securities analysts and incumbent response to radical technological change: Evidence from digital photography and Internet telephony. Organization Science. 21(1): 42–62.
Benner, M. J. and Tushman, M. 2002. Process management and technological innovation: A longitudinal study of the photography and paint industries. Administrative Science Quarterly. 47(4): 676-707.
Blume-Kohout, M. E. and Sood, N. 2013. Market size and innovation: Effects of Medicare Part D on phar-maceutical research and development. Journal of Public Economics. 97: 327-336.
Boudreau, K.J., Lacetera, N. and Lakhani, K.R. 2011. Incentives and problem uncertainty in innovation contests: An empirical analysis. Management Science. 57(5): 843-863.
Box-Steffensmeier, J.M. and Jones, B.S. 2004. Event history modeling: A guide for social scientists. Cam-bridge University Press.
Brahm, F., Tarzijan, J. & Singer, M. 2017. The impact of frictions in routine execution of economies of scope. Strategic Management Journal. 38(10): 2121-2142.
Brown, S.L. and K.M. Eisenhardt. 1995. “Product Development: Past Research, Present Findings, and Fu-ture Directions.” Academy of Management Review. 20(2): 343-378.
Budish, E., Roin, B.N. and Williams, H. 2015. Do firms underinvest in long-term research? Evidence from cancer clinical trials. American Economic Review. 105(7): 2044-2085.
27
Carpenter, D. P., Moffitt, S.I., Moore, C.D., Rynbrandt, R.T., Ting, M.M., Yohai, I. and Zucker, E.J. 2010. Early entrant protection in approval regulation: theory and evidence from FDA drug review. Journal of Law, Economics and Organization. 26(3): 515–545.
Chatterji, A.K., 2009. Spawned with a silver spoon? Entrepreneurial performance and innovation in the medical device industry. Strategic Management Journal. 30(2): 185-206.
Chatterji, A.K. and Fabrizio, K.R. 2016. Does the market for ideas influence the rate and direction of inno-vative activity? Evidence from the medical device industry. Strategic Management Journal. 37(3): 447-465.
Cheah, E.T., Chan, W. L. and Chieng, C. L. L. 2007. The corporate social responsibility of pharmaceutical product recalls: An empirical examination of US and UK markets. Journal of Business Ethics. 76(4): 427-449.
Chen, Y., Ganesan, S. and Liu, Y. 2009. Does a firm's product-recall strategy affect its financial value? An examination of strategic alternatives during product-harm crises. Journal of Marketing. 73(6): 214-226.
Cottrell, T. and Nault, B.R. 2004. Product variety and firm survival in the microcomputer software industry. Strategic Management Journal. 23(10): 1005-1025.
Dasgupta, P. and Stiglitz, J. 1980. Uncertainty, industrial structure, and the speed of R&D. Bell Journal of Economics. 11(1): 1–28.
Davidson, W.N. and Worrell, D. L. 1992. Research notes and communications: The effect of product recall announcements on shareholder wealth. Strategic Management Journal. 13(6): 467-473.
Demirel, S., Kapuscinski, R. and Yu, M. 2017. Strategic behavior of suppliers in the face of production disruptions. Management Science. 64(2): 533-551.
Desai, V. 2015. Learning through the distribution of failures within an organization: Evidence from heart bypass surgery performance. Academy of Management Journal. 58(4): 1032-1050.
Dong, L. & Tomlin, B. 2012. Managing disruption risk: The interplay between operations and insurance. Management Science. 58(10): 1898-1915.
Dubois, P., De Mouzon, O., Scott-Morton, F. and Seabright, P. 2015. Market size and pharmaceutical in-novation. The RAND Journal of Economics. 46(4): 844-871.
Eggers, J.P. 2012. All experience is not created equal: Learning, adapting, and focusing in product portfolio management. Strategic Management Journal. 33: 315-325.
Fabrizio, K.R. and Thomas, L.G. The impact of local demand on innovation in a global industry. Strategic Management Journal. 33(1): 42-64.
Finkelstein, A. 2004. Static and dynamic effects of health policy: Evidence from the vaccine industry. The Quarterly Journal of Economics. 119(2): 527-564.
Fischer, H.M. and Pollock, T.G. 2004. Effects of social capital and power on surviving transformational change: The case of initial public offerings. Academy of Management Journal. 47(4): 463–481.
Greve, H. R. 2007. Exploration and exploitation in product innovation. Industry and Corporate Change. 16(5): 945–975.
Haas, M.R. and Hansen, M.T. 2004. When using knowledge can hurt performance: the value of organiza-tional capabilities in a management consulting company. Strategic Management Journal. 26(1): 1-24.
Hall, B.H. and Lerner, J. 2010. The financing of R&D and innovation. In Hall, B.H & Rosenberg, N. eds. Handbook of the Economics of Innovation, Vol. 1. North-Holland, Amsterdam, 609–640.
28
Harhoff, D. and Wagner, S. 2009. The duration of patent examination at the European Patent Office. Man-agement Science. 55(12): 1969-1984.
Haunschild, P.R. and Rhee, M. 2004. The role of volition in organizational learning: The case of automotive product recalls. Management Science. 50(11): 1545-1560.
Haunschild, P.R. and Sullivan, B.N. 2002. Learning from complexity: Effects of prior accidents and inci-dents on airlines' learning. Administrative Science Quarterly. 47(4): 609-643.
Holmqvist, M. 2004. Experiential learning processes of exploitation and exploration within and between organizations: an empirical study of product development. Organization Science. 15(1): 70-81.
Henderson, R. and Cockburn, I. 1994. Measuring competence? Exploring firm effects in pharmaceutical research. Strategic Management Journal 15: 63-84.
Ingram, P. and Baum, J. A. 1997. Opportunity and constraint: Organization’s learning from the operating and competitive experience of industries. Strategic Management Journal. 18(S1): 75-98.
Jarrell, G. and Peltzman, S. 1985. The impact of product recalls on the wealth of sellers. The Journal of Political Economy. 93(3): 512-536.
Kalnins, A. and Mayer, K. J. 2004. Franchising, ownership, and experience: A study of pizza restaurant survival. Management Science. 50(12): 1716-1728.
KC, D., Staats, B.R. and Gino, F. 2013. Learning from my success and from others' failure: Evidence from minimally invasive cardiac surgery. Management Science. 59(11): 2435-2449.
Kim, S.H., Cohen, M.A., Netessine, S. and Veeraraghavan, S. (2010). Contracting for infrequent restoration and recovery of mission-critical systems. Management Science. 56(9): 1551-1567.
Kim, J.Y. and Miner, A.S. 2007. Vicarious learning from the failures and near-failures of others: Evidence from the US commercial banking industry. Academy of Management Journal. 50(3): 687-714.
Kochhar, R. and Parthiban, D. 1996. Institutional investors and firm innovation: A test of competing hy-potheses. Strategic Management Journal. 17(1): 73-84
Krieger, J. 2017. Trials and Terminations: Learning from Competitors’ R&D Failures. MIT Sloan School of Management Working Paper.
Krieger, J.L., Li, D. and Papanikolaou, D. 2018. Developing Novel Drugs (No. w24595). National Bureau of Economic Research.
Krumholz, H.M., Ross, J.S., Presler, A.H. and Egilman, D. S. 2007. What have we learnt from Vioxx? Brit-ish Medical Journal. 334(7585):120-123.
Leiblein, M.J. and Madsen, T.L. 2009. Unbundling competitive heterogeneity: incentive structures and ca-pability influences on technological innovation. Academy of Management Journal. 30(7): 711-735.
Levinthal, D.A. and Wu, B. 2010. Opportunity costs and non-scale free capabilities: profit maximization, corporate scope, and profit margins. Strategic Management Journal. 31(7): 780-801.
Macher, J.T. 2006. Technological development and the boundaries of the firm: A knowledge-based exam-ination in semiconductor manufacturing. Management Science 52(6): 826-843.
Macher, J.T. and C.S. Boerner 2006. Experience and scale and scope economies: Trade-offs and perfor-mance in development. Strategic Management Journal. 27(9): 845-865.
Madsen, P.M. and Desai, V. 2010. Failing to learn? The effects of failure and success on organizational learning in the global orbital launch vehicle industry. Academy of Management Journal 53(3): 451–76.
29
Mikkelson, W.H., Partch, M.M. and Shah, K. 1997. Ownership and operating performance of companies that go public. Journal of Financial Economics. 44(3): 281–307.
Nazarian, R. 2009. Developing Medical Devices in a Consumer-Driven Market. Medical Device and Diag-nostic Industry (MDDI) Newsletter. https://www.mddionline.com/developing-medical-devices-con-sumer-driven-market.
Nerkar, A. and Roberts, P. W. 2004. Technological and product-market experience and the success of new product introductions in the pharmaceutical industry. Strategic Management Journal. 25(8-9): 779-799.
Ocasio, W. 1997. Towards an attention-based view of the firm. Strategic Management Journal. 18: 187-206.
Panzar, J.C. and Willig, R.D. 1981. Economies of scope. The American Economic Review 71(2): 268–72.
Pollock, T. G. and Gulati, R. 2007. Standing out from the crowd: The visibility-enhancing effects of IPO-related signals on alliance formation by entrepreneurial firms. Strategic Organization. 5(4): 339–372.
Porter, M. E. and Heppelmann, J. E. 2014. How smart, connected products are transforming competition. Harvard Business Review 92(11): 64-88.
Rerup, C. 2009. Attentional triangulation: Learning from unexpected rare crises. Organization Science. 20(5): 876-893.
Rhee, M. and Haunschild, P. R. 2006. The liability of good reputation: A study of product recalls in the US automobile industry. Organization Science. 17(1): 101-117.
Sacerdote, B. 2001. Peer effects with random assignment: Results for Dartmouth roommates. The Quarterly Journal of Economics. 116(2): 681-704.
Sacerdote, B. 2014. Experimental and quasi-experimental analysis of peer effects: two steps forward? An-nual Review of Economics. 6(1): 253-272.
Serpa, J.C. and Krishnan, H. 2016. The strategic role of business insurance. Management Science. 63(2): 384-404.
Shah, R., Ball, G. and Netessine, S. 2016, Plant Operations and Product Recalls in the Automotive Industry: An Empirical Investigation. Management Science. 63(8): 2439-2459.
Sterman, J.D., Repenning, N.P. and Kofman, F. 1997. Unanticipated side effects of successful quality pro-grams: Exploring a paradox of organizational improvement. Management Science. 43(4): 503-521.
Stern, A.D. 2017. Innovation under regulatory uncertainty: Evidence from medical technology. Journal of Public Economics. 145: 181-200.
Stock, J.H. and Watson, M.W. 2012. Introduction to econometrics: Global edition. Boston, MA: Pearson Education.
Thirumalai, S. and Sinha, K.K. 2011. Product recalls in the medical device industry: An empirical explora-tion of the sources and financial consequences. Management Science. 57(2): 376-392.
Trostel, A.O. and Nichols, M.L. 1982. Privately-held and publicly-held companies: A comparison of stra-tegic choices and management processes. Academy of Management Journal. 25(1): 47–62.
Von Nordenflycht, A. 2007. Is public ownership bad for professional service firms? Ad agency ownership, performance, and creativity. Academy of Management Journal. 50(2): 429–445.
Wernerfelt, B. 1984. A Resource-Based View of the Firm. Strategic Management Journal. 5(2): 171-180.
30
Wizemann, T. ed. 2010. Public health effectiveness of the FDA 510 (k) clearance process: balancing patient safety and innovation: workshop report. National Academies Press.
Wooldridge, J. M. 2010. Econometric analysis of cross section and panel data. MIT press.
Wowak, A.J., Mannor, M.J. and Wowak, K.D. 2015. Throwing caution to the wind: The effect of CEO stock option pay on the incidence of product safety problems. Strategic Management Journal. 36(7): 1082-1092.
Wu, B. 2013. Opportunity costs, industry dynamics, and corporate diversification: Evidence from the car-diovascular medical device industry, 1976–2004. Strategic Management Journal. 34(11): 1265-1287.
Wu, B., Wan, Z. and Levinthal, D.A. 2014. Complementary assets as pipes and prisms: Innovation incen-tives and trajectory choices. Strategic Management Journal. 35(9): 1257-1278.
Wu, G. 2012. The effect of going public on innovative productivity and exploratory search. Organization Science. 23(4): 928-950.
Yanadori, Y. and Cui, V. 2013. Creating incentives for innovation? The relationship between pay dispersion in R&D groups and firm innovation performance. Strategic Management Journal. 34(12): 1502-1511.
Yang, Z., Aydın, G., Babich, V. and Beil, D. R. 2009. Supply disruptions, asymmetric information, and a backup production option. Management Science. 55(2): 192-209.
31
Figure 1. Product Classes and Product Types
Table 1. Key Independent Variable Definitions
Own Firm Definition
REC: Own Recalls experienced by own firm
REC: Own-Direct ...within same product class and same product type
REC: Own-Proximate ...within same product class and different product type
REC: Own-Distant ...within different product class and different product type
Competitor Firm Definition
REC: Competitor Recalls experienced by competitor firm
REC: Competitor-Direct ...within same product class and same product type
REC: Competitor-Proximate ...within same product class and different product type
REC: Competitor-Distant ...within different product class and different product type
Moderators Definition
Product Scope Number of Unique Product Types of Firm
Public / Private Indicator for Public / Private Firm
Multiple product Classes(Regulatory Medical Specialties)
CardiovascularRadiology Dental …
Multiple product Types(FDA Product Codes)
DSS DXQ DPS OBI …
(Blood Pressure Cuff)
(Electrocardiograph) (Cardiac Pressure Monitoring Catheter)
(Others)
Multiple Firms(Manufacturers)
(Vascular Clip)
Alivecor(3 devices)
…Bionet(5 devices)
Philips (9 devices)
Medtronic(3 devices)
32
Table 2: Descriptive Statistics VARIABLE MEAN ST DEV MIN MAX 510(k) Time 186.32 416.72 1.00 4611.00 REC: Own 11.42 25.75 0.00 165.00 REC: Competitor 262.38 261.98 0.00 1491.00 REC: Own-Direct 1.32 4.42 0.00 54.00 REC: Own-Proximate 5.35 15.64 0.00 128.00 REC: Own-Distant 4.75 13.88 0.00 162.00 REC: Competitor-Direct 65.31 86.68 0.00 533.00 REC: Competitor-Proximate 67.48 87.26 0.00 617.00 REC: Competitor-Distant 129.58 184.62 0.00 1121.00 Product Scope 22.05 23.31 0.00 107.00 Public 0.54 0.50 0.00 1.00 SUB: Own-Same 6.67 17.93 0.00 132.00 SUB: Own-Different 4.75 13.88 0.00 162.00 SUB: Competitor-Same 132.80 134.15 0.00 638.00 SUB: Competitor-Different 129.58 184.62 0.00 1121.00 Product Age 2663.29 1308.05 0.00 5567.00 Product Competition 29.31 38.76 0.00 229.00 Employees (public firms) 88605.05 122319.60 0.00 475000.00 Revenues (public firms) 31100.27 43924.99 -385.97 216626.60 R&D (public firms) 1452.86 2436.32 0.00 13563.90 R&D Intensity (public firms) 0.09 1.24 0.00 91.92 Total 510(k) Clearances 16489
Total 510(k) Recalls 5752
Total Unique Products 1319
Total Unique Firms 940
Table 3: Correlation Statistics
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (1) 510(k) Time 1.00 (2) REC: Own -0.16 1.00 (3) REC: Competitor -0.22 0.53 1.00 (4) REC: Own-Direct -0.07 0.56 0.25 1.00 (5) REC: Own-Proximate -0.12 0.83 0.40 0.42 1.00 (6) REC: Own-Distant -0.13 0.75 0.45 0.25 0.27 1.00 (7) REC: Competitor-Direct 0.00 0.35 0.40 0.30 0.41 0.09 1.00 (8) REC: Competitor-Proximate -0.18 0.46 0.75 0.29 0.44 0.27 0.19 1.00 (9) REC: Competitor-Distant -0.22 0.37 0.88 0.08 0.17 0.47 0.01 0.50 1.00 (10) Product Scope -0.32 0.52 0.64 0.16 0.41 0.46 0.08 0.45 0.67 1.00 (11) Public -0.31 0.33 0.37 0.18 0.26 0.27 0.13 0.27 0.34 0.46 1.00
33
Table 4: Main Results Hypothesis H1 H2 H3 H3 H3 H4 H4 Model (1) (2) (3) (4) (5) (6) (7) Firm Type All All All Focused Broad Private Public REC: Own 0.037 (0.340) REC: Competitor -0.084 (0.005) REC: Own-Direct 0.240 0.240 0.089 0.246 0.191 0.269 (0.000) (0.000) (0.035) (0.000) (0.000) (0.000) REC: Own-Proximate 0.070 0.075 -0.078 0.097 0.077 0.075 (0.088) (0.066) (0.123) (0.143) (0.213) (0.069) REC: Own-Distant -0.050 -0.030 -0.163 0.027 0.027 -0.010 (0.128) (0.348) (0.050) (0.667) (0.630) (0.684) REC: Competitor-Direct -0.050 -0.047 -0.032 -0.063 -0.008 -0.067 (0.002) (0.004) (0.265) (0.040) (0.761) (0.000) REC: Competitor-Proximate 0.021 0.035 -0.003 0.123 -0.005 0.047 (0.132) (0.014) (0.883) (0.002) (0.804) (0.006) REC: Competitor-Distant -0.071 -0.068 -0.051 -0.272 -0.061 0.007 (0.005) (0.008) (0.120) (0.035) (0.090) (0.799) LN(Product Scope) -0.365 (0.000) Constant 6.270 6.617 6.514 24.008 3.650 24.428 6.355 (0.000) (0.000) (0.000) (0.000) (0.031) (0.088) (0.000) Year Fixed Effects X X X X X X X Product Fixed Effects X X X X X X X Firm Fixed Effects X X X X X X X Observations 22241 22241 22241 7971 7641 10315 11926 Notes: The dependent variable is the time to 510(k) submission. All models use 24 months for the analysis window. All models include controls for same product class and different product class submissions by own and competitor firms. Standard errors are clustered by firm. p-values (in parentheses) are presented below the coefficients. Model 3 includes a logged measure of Product Scope. Model 4 examines the first product scope tercile (≤ 9 products); Model 5 examines the third product scope tercile (≥ 38 products); Model 6 examines Private firms; Model 7 examines Public firms. Models 6 and 7 use firm fixed effects for firms with twelve or more observations for conversion purposes.
34
Table 5: Hypotheses Results and Interpretation
Table 6: Post-Hoc Results Model (1) (2) LN(510K Time) 0.001 -0.011 (0.965) (0.591)
Constant 7.849 7.630 (0.000) (0.000)
Year Fixed Effects X X
Firm Fixed Effects X X
Observations 2826 19592
Notes: The dependent variable is the time to a post-submission Class 1–2 recall. Controls for same product class and different product class submissions by own and competitor firms are included. Standard errors are clustered by firm. p-values (in parentheses) are presented below the coefficients. Model 1 examines submissions that experience future recalls. The number of observations (2826) represents how many of these occur. Model 2 examines submissions that do or do not experience future recalls. The number of observations (19592) represents how many of these occur, but excludes pre-submission recalls (i.e., recalls related to submissions that pre-date the sample).
DAYS DAYS
ONE SD REC ONE REC REC: Own 15.8 0.6 REC: Competitor*** -15.0 -0.1 REC: Own-Direct*** 169.2 38.3 REC: Own-Proximate+ 39.9 2.8 REC: Own-Distant -26.6 -2.0 REC: Competitor-Direct** -12.1 -0.1 REC: Competitor-Proximate 5.1 0.1 REC: Competitor-Distant** -18.2 -0.1
+ p < 0.10; * p < 0.05; ** p < 0.01; *** p < 0.001 Notes: We describe the determination of economic effects using REC: Competitor as an example. All interpretations are done in an equivalent manner. The mean of REC: Competitor = 262.38 and the mean 510(k) time is 186.32 days from Table 2. A one percent change in REC: Competitor is 2.6238, which is associated with a (0.01 x (expβ-1) percent change in mean days to submission. The coefficient for REC: Competitor is -0.084: 0.01 x (exp(-0.084)-1) = -0.000805687. We multiply this number by the mean number of days to submission to determine the effect of a one percent change in REC: Com-petitor: 186.32 days X -0.000805687 = -0.150115684 days. The standard deviation of REC: Competitor is 261.98 from Table 2. The effect of a one standard deviation change in REC: Competitor is found by scaling the number of recalls in one standard deviation by the number of recalls in a one percent change and multiplying that by the effect of a one percent change on submission time: 261.98 recalls/ 2.6238 recalls x -0.15 days = -14.99 days. The effect of a single competitor firm recall on submission times is found by dividing one standard deviation in REC: Competitor by the number of those types of recalls in one standard deviation: -14.99 days / 261.98 recalls = -0.06 days per recall.
35
Appendix Table A-1: Descriptive Statistics (PMA Sample)
VARIABLE MEAN ST DEV MIN MAX PMA Time 593.45 816.78 1.00 4542.00 REC: Own 7.09 8.38 0.00 35.00 REC: Competitor 30.19 18.55 0.00 104.00 REC: Own-Direct 0.94 1.88 0.00 11.00 REC: Own-Proximate 1.97 3.22 0.00 17.00 REC: Own-Distant 4.19 6.33 0.00 32.00 REC: Competitor-Direct 8.47 10.18 0.00 47.00 REC: Competitor-Proximate 8.51 11.03 0.00 49.00 REC: Competitor-Distant 13.21 13.68 0.00 63.00 Product Scope 17.19 12.51 0.00 40.00 Public 0.77 0.42 0.00 1.00 SUB: Own-Same 2.91 3.91 0.00 17.00 SUB: Own-Different 4.19 6.33 0.00 32.00 SUB: Competitor-Same 16.98 14.43 0.00 56.00 SUB: Competitor-Different 13.21 13.68 0.00 63.00 Product Age 2531.70 1394.27 0.00 6301.00 Product Competition 2.62 2.13 0.00 10.00 Employees (public firms) 65687.91 70941.61 0.00 475000.00 Revenues (public firms) 23496.04 22621.60 0.00 108827.00 R&D (public firms) 2497.02 2792.44 0.00 9893.88 R&D Intensity (public firms) 0.12 0.34 0.00 7.57 Total PMA Submissions 200 Total PMA Recalls 423 Total Unique Products 134 Total Unique Firms 60
Table A-2: Correlation Statistics (PMA Sample)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (1) PMA Time 1.00 (2) REC: Own -0.40 1.00 (3) REC: Competitor -0.40 0.48 1.00 (4) REC: Own-Direct -0.20 0.43 0.14 1.00 (5) REC: Own-Proximate -0.25 0.59 0.42 0.12 1.00 (6) REC: Own-Distant -0.35 0.90 0.39 0.21 0.24 1.00 (7) REC: Competitor-Direct -0.03 0.04 0.27 0.07 0.22 -0.08 1.00 (8) REC: Competitor-Proximate -0.11 0.25 0.64 -0.02 0.46 0.10 -0.08 1.00 (9) REC: Competitor-Distant -0.43 0.42 0.64 0.15 0.03 0.50 -0.32 0.12 1.00 (10) Product Scope -0.58 0.83 0.68 0.34 0.50 0.75 0.06 0.31 0.62 1.00 (11) Public -0.42 0.39 0.25 0.17 0.21 0.36 -0.08 -0.05 0.44 0.53 1.00
36
Table A-3: Robustness Results (36 Month Windows) Hypothesis H1 H2 H3 H3 H3 H4 H4 Model Firm Type
( 1) All
(2) All
(3) All
(4) Focused
(5) Broad
(6) Private
(7) Public
REC: Own 0.051 (0.217) REC: Competitor -0.065 (0.062) REC: Own-Direct 0.223 0.225 0.170 0.238 0.203 0.251 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) REC: Own-Proximate 0.068 0.071 -0.065 0.116 0.107 0.059 (0.089) (0.077) (0.238) (0.087) (0.051) (0.115) REC: Own-Distant -0.030 -0.013 -0.103 0.030 0.164 -0.006 (0.451) (0.736) (0.191) (0.681) (0.009) (0.835) REC: Competitor-Direct -0.056 -0.054 -0.036 -0.084 0.003 -0.077 (0.002) (0.003) (0.227) (0.010) (0.914) (0.000) REC: Competitor-Proximate 0.027 0.037 -0.006 0.145 0.000 0.048 (0.073) (0.014) (0.731) (0.001) (0.999) (0.008) REC: Competitor-Distant -0.061 -0.059 -0.059 -0.305 -0.065 0.020 (0.036) (0.047) (0.079) (0.062) (0.095) (0.544) LN(Product Scope) -0.344 (0.000) Constant 5.785 6.101 6.075 24.130 3.040 24.831 6.261 (0.000) (0.000) (0.000) (0.045) (0.088) (0.047) (0.000) Year Fixed Effects X X X X X X X Product Fixed Effects X X X X X X X Firm Fixed Effects X X X X X X X Observations 22241 22241 22241 7971 7641 10315 11926 Notes: The dependent variable is the time to 510(k) submission. All models use 36 months for the analysis window. All models include controls for same product class and different product class submissions by own and competitor firms. Standard errors are clustered by firm. p-values (in parentheses) are presented below the coefficients. Model 3 includes a logged measure of Product Scope. Model 4 examines the first product scope tercile (≤ 9 products); Model 5 examines the third product scope tercile (≥ 38 products); Model 6 examines Private firms; Model 7 examines Public firms. Models 6 and 7 use firm fixed effects for firms with twelve or more observations for conversion purposes.
37
Table A-4: Robustness Results (Estimation Assumptions/Placebo Tests) Model (1) (2) (3) (4) (5) Firms All All All All All REC: Own-Direct 0.240 0.251 0.225 0.238 0.239 (0.000) (0.000) (0.000) (0.000) (0.000) REC: Own-Proximate 0.070 0.072 0.069 0.070 0.071 (0.088) (0.054) (0.084) (0.084) (0.064) REC: Own-Distant -0.050 -0.056 -0.029 -0.050 -0.043 (0.127) (0.072) (0.462) (0.127) (0.193) REC: Competitor-Direct -0.050 -0.044 -0.057 -0.030 -0.049 (0.002) (0.006) (0.002) (0.082) (0.002) REC: Competitor-Proximate 0.021 0.024 0.028 0.018 0.023 (0.128) (0.090) (0.059) (0.230) (0.097) REC: Competitor-Distant -0.071 -0.073 -0.063 -0.071 0.026 (0.005) (0.003) (0.031) (0.005) (0.010) LN(Product Age) 0.027 (0.024)
LN(Product Competition) -0.121 (0.011) Constant 4.190 4.174 5.994 6.686 6.628 (0.000) (0.000) (0.000) (0.000) (0.000)
Year Fixed Effects X X X X X Product Fixed Effects X X X X X Firm Fixed Effects X X X X X Class X Year Fixed Effects X
Observations 22241 22241 22241 22241 22241
Notes: The dependent variable is the time to 510(k) submission. All models use 24 months for the analysis window. All models include controls for same product class and different product class submissions by own and competitor firms. Standard errors are clustered by firm. p-values (in parentheses) are presented below the coefficients. Model 1 is the baseline model from Table 4. Model 2 includes product class X year fixed effects. Model 3 includes a logged measure of Product Age. Model 2 includes a logged measure of Product Competition. Model 5 replaces Class 1 – 2 recalls with all Class (i.e., 1 – 3) recalls.
38
Table A-5: Robustness Results (Firm Ownership)
Hypothesis Model Firm Type
H4 (1)
Private
H4 (2)
Public
H4 (3)
Public
H4 (4)
Public
H4 (5)
Public
H4 (6)
Public REC: Own-Direct 0.191 0.269 0.269 0.269 0.269 0.268 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) REC: Own-Proximate 0.077 0.075 0.072 0.073 0.072 0.080 (0.213) (0.069) (0.078) (0.073) (0.081) (0.059) REC: Own-Distant 0.027 -0.010 -0.012 -0.011 -0.011 -0.004 (0.630) (0.684) (0.631) (0.645) (0.649) (0.856) REC: Competitor-Direct -0.008 -0.067 -0.067 -0.066 -0.067 -0.059 (0.761) (0.000) (0.000) (0.000) (0.000) (0.002) REC: Competitor-Proximate -0.005 0.047 0.048 0.048 0.047 0.038 (0.804) (0.006) (0.006) (0.006) (0.006) (0.032) REC: Competitor-Distant -0.061 0.007 0.003 0.006 0.008 0.009
(0.090) (0.799) (0.922) (0.849) (0.781) (0.787) LN(Revenue) -0.027 (0.304) LN(Employees) -0.011 (0.477) LN(R&D Spending) -0.030 (0.249) R&D Intensity -0.001
(0.897) Constant 24.428 6.355 6.475 6.42 6.42 6.414
(0.088) (0.000) (0.000) (0.000) (0.000) (0.000) Year Fixed Effects X X X X X X Product Fixed Effects X X X X X X Firm Fixed Effects X X X X X X
Observations 10315 11926 11916 11926 11926 11599
Notes: The dependent variable is the time to 510(k) submission. All models use 24 months for the analysis window. All models include controls for same product class and different product class submissions by own firms and competitor firms. Standard errors are clustered by firm. p-values (in parentheses) are presented below coefficients. Models 1 and 2 are Models 7 and 8 from Table 4. Model 3 includes a logged measure of Revenue for public firms. Model 4 includes a logged measure of Employees for public firms. Model 5 includes a logged measure of R&D for public firms. Model 6 includes a measure of R&D Intensity for public firms. All models utilize firm fixed effects for those firms with twelve or more obser-vations for conversion purposes.
39
Table A-6: Robustness Results (PSM: Recall Likelihood & Past Submissions) Treatment Control ATE
Main Sample 5374 1289 0.032 (0.412)
Notes: p-value (in parentheses) is presented below the ATE coefficient.
40
Table A-7: Robustness Results (PMA Sample) Hypothesis (H1) (H2) (H3) (H3) (H3) (H3) (H3) Model ( 1) (2) (3) (4) (5) (6) (7) Firm Type All All All Focused Broad Private Public REC: Own 0.120 (0.549) REC: Competitor 0.237 (0.334) REC: Own-Direct 0.939 0.933 0.572 0.629 22.778 0.908 (0.000) (0.000) (0.612) (0.522) (0.000) (0.000) REC: Own-Proximate -0.279 -0.138 0.546 0.319 -2.185 -0.334 (0.079) (0.345) (0.562) (0.755) (0.061) (0.025) REC: Own-Distant 0.179 0.250 1.282 -0.885 17.985 -0.012 (0.291) (0.163) (0.005) (0.000) (0.000) (0.938) REC: Competitor-Direct -0.390 -0.388 0.044 -1.449 -0.666 -0.444 (0.000) (0.000) (0.904) (0.003) (0.385) (0.000) REC: Competitor-Proximate 0.008 0.067 0.032 -0.681 -0.473 0.036 (0.952) (0.629) (0.918) (0.687) (0.544) (0.861) REC: Competitor-Distant 0.121 0.132 1.118 -1.390 0.641 -0.118 (0.584) (0.528) (0.000) (0.266) (0.037) (0.397) LN(Product Scope) -0.548 (0.028) Constant 22.865 23.182 23.291 25.743 13.002 31.806 8.087 (0.000) (0.000) (0.000) (0.000) (0.033) (0.000) (0.000) Year Fixed Effects X X X X X X X Product Fixed Effects X X X X X X X Firm Fixed Effects Observations 623 623 623 217 239 143 480 Notes: The dependent variable is the time to PMA submission. All models use 24 months for the analysis window. All models include controls for same product class and different product class submissions by own and competitor firms. Standard errors are clustered by firm. p-values (in parentheses) are presented below the coefficients. Model 3 includes a logged measure of Product Scope. Model 4 examines the first product scope tercile (≤ 10 products); Model 5 examines the third product scope tercile (≥ 29 products); Model 6 examines Private firms; Model 7 examines Public firms.