The Paradox of Inventors’ Job Security

1. Introduction

Technological innovation is a prototypical human capital–intensive activity that is primarily conducted within firms today. Research suggests that inventors’ human capital plays a central role in explaining variation in innovation productivity across organizations (Bhaskarabhatla et al. 2021). This places inventors and their behavior at the center of the innovation process in research and development (R&D)–intensive firms. Hence, understanding the motivations shaping R&D workers’ actions is crucial for innovative organizations. Because innovation often involves experimentation, designing incentives that effectively foster innovation is particularly challenging. Given the uncertain and long-term nature of innovation, job security is widely regarded as a key factor influencing individuals’ incentives to innovate.

Several streams of research examine the relationship between job security and innovation. They mainly focus on the context of the broader workforce—workers, managers, or entrepreneurs—who may incorporate innovation into their routine activities or reallocate effort from those activities toward innovative pursuits. Drawing on psychological research, Amabile and Conti (1999) argue that termination threats alter the work environment in ways that inhibit creativity, a core input to innovation. In the economics tradition, Manso (2011) suggests that insufficient job security undermines the exploratory behavior required for innovation by eliminating tolerance for early failure. Contracts that mandate termination following poor short-term performance reduce incentives to take risks and test unfamiliar approaches. Supporting this view, experimental evidence involving managers (Ederer and Manso 2013) and observational data on academic scientists (Azoulay et al. 2011) show that tolerance for failure promotes exploratory innovation. Similarly, Acharya et al. (2013, 2014) argue that job security encourages workers to innovate rather than to focus on routine tasks by protecting them from being held up by employers after they exert inventive effort. Empirical evidence from Acharya et al. (2014) and Ekinci and Wehrheim (2024) aligns with this view, showing that stronger employment protection is associated with higher patenting activity.

By focusing on the specific case of corporate R&D employees, this paper introduces an alternative mechanism through which job security influences innovative behavior. Drawing on human capital theory, we posit that increased job security incentivizes employee inventors to allocate more effort toward acquiring firm-specific human capital, which implies mastering technological knowledge whose economic returns are primarily realized within their current employer. This leads to a fundamental change in the nature of the innovations they generate: as inventors acquire more firm-specific human capital, their projects tend to draw more heavily on internal knowledge and exhibit lower novelty. Contrasting this mechanism with prior research, we identify a job security paradox: whereas employment protection provides the necessary safety to fail that facilitates exploration (Manso 2011), it simultaneously strengthens incentives for firm-specific human capital investment in ways that tilt inventive activity toward exploitation rather than novelty.

Whereas the safety-to-fail mechanism operates broadly in settings in which workers choose among actions with different risk–return profiles, the human capital channel emphasized in our study is particularly relevant for R&D professionals employed by innovative firms. First, for these individuals, innovation is not an occasional activity but a core mandate. Their productivity depends critically on their accumulated technological knowledge, which subsequently shapes their future career trajectory. As a result, career concerns influence the direction of inventors’ human capital investments. Second, some of the technological knowledge available to these R&D employees is relatively firm-specific because of the existence of complementary assets within the organization. Because the returns to mastering this firm-specific knowledge are largely tied to continued employment at the focal firm, an increase in inventors’ prospects of staying with the company in the future increases their incentives to invest in it. Consequently, higher job security may shift corporate inventors’ focus toward firm-specific knowledge, leading to more localized search approaches that limit technological novelty and reduce the likelihood of breakthrough innovations.

To test our arguments, we exploit the staggered adoption of employee-friendly labor laws across U.S. states between the 1970s and the 1990s. Known as wrongful discharge laws (WDLs), these statutes introduce exceptions to the employment-at-will doctrine and restrict employers’ ability to terminate employees (Autor et al. 2006). This institutional setting allows us to examine the effects of plausibly exogenous increases in job security on the nature of inventive activity. Using patent data, we identify inventors employed by firms, their locations, and the technological characteristics of their R&D projects. Our sample tracks more than 82,000 unique inventors who are listed as inventors on granted U.S. patents, are employed by publicly traded firms, and work at U.S.-based facilities. We implement a staggered difference-in-differences (DiD) design using a stacked regression approach (Cengiz et al. 2019) and control for employer–employee match fixed effects, allowing us to isolate how changes in job security affect the behavior of a given inventor within a given firm. Our results show that increased job security shifts the nature of inventive activity: inventors rely more heavily on their firms’ prior technological knowledge, produce less novel innovations on average, and generate inventions with higher private economic value for their employers.

Our empirical analysis also provides complementary evidence that supports our proposed mechanism and is difficult to reconcile with alternative explanations. Most importantly, we examine whether the observed changes in innovation reflect inventors’ own decisions rather than shifts in employers’ innovation policies (Griffith and Macartney 2014, Keum 2020). To do so, we exploit differences in responses between inventors working in labs located in their employer’s headquarters (HQ) state and those based in other states. Because inventors outside HQ are typically further removed from centralized corporate decision making, they are likely to have greater autonomy in their innovation activities (Glaeser et al. 2023) and, thus, to respond more strongly to changes in job security. Consistent with this view, the effects are concentrated among inventors working outside corporate HQ. We also test whether innovation responses are driven by changes in employment protection in the inventor’s state of employment or in the firm’s HQ state when the two differ (Bena et al. 2022). Only changes in employment protection in the inventor’s state of employment affect innovation outcomes, suggesting that our results do not reflect broad firm-level policy adjustments. Moreover, we find evidence that the documented change in invention outcomes following increased job security varies according to inventor characteristics such as age or inventive experience. Taken together, these results indicate that our main findings primarily reflect inventors’ own decisions rather than their employers’ decisions.

This study makes several contributions to the literature. First, it provides a complementary perspective on the relationship between job security and innovation incentives by focusing on the career concerns of R&D professionals and the types of innovations they generate. Our results indicate that greater job security encourages inventors to invest in mastering firm-specific technological knowledge, leading to deeper exploitation of the employer’s existing knowledge base and, consequently, to less novel innovations. This finding refines the common view that job security generally favors exploratory innovation (Amabile and Conti 1999, Azoulay et al. 2011, Manso 2011, Ederer and Manso 2013) and documents the existence of an interesting paradox concerning the consequences of inventors’ job security. In settings in which human capital investments are important and can be firm-specific, job security may foster exploitative innovation even as it establishes the safety conditions typically thought to promote exploration. Whereas the safety-to-fail effect is likely to dominate in contexts with limited opportunities for firm-specific investment (e.g., Azoulay et al. 2010, Ederer and Manso 2013), the human capital mechanism we highlight is particularly salient for the innovation activities of R&D employees in the private sector. In these settings, job security increases alignment with internal firm resources but simultaneously discourages the pursuit of exploratory, path-breaking innovation.

Second, we contribute to the strategic human capital literature by showing how individual employee decisions can contribute to shaping firm-level innovation trajectories. Whereas prior research on employment protection primarily focuses on aggregate firm-level patent outcomes (Acharya et al. 2014, Keum 2020, Bena et al. 2022), we examine the microfoundations of this relationship at the inventor level. The study most closely related to ours is Ekinci and Wehrheim (2024), which also analyzes individual inventor outcomes but treats the transferability of inventors’ human capital across employers as exogenously determined. In contrast, we allow for the possibility that inventors actively choose the direction of their human capital investments. This allows us to shed light on how increased job security affects both the degree of firm specificity and the novelty of the innovations that inventors produce. In addition, we make a methodological contribution by evaluating the relationship between wrongful discharge laws and innovation using a stacked DiD design that is robust to staggered policy adoption, addressing concerns about two-way fixed effects estimators in this setting (Cengiz et al. 2019, Baker et al. 2022, Roth et al. 2023).

Finally, we contribute to the literature investigating how institutional constraints on labor mobility shape innovation outcomes. Prior research mostly examines the consequences of restrictions on workers’ voluntary mobility, such as those imposed by trade secret laws or noncompete agreements (see, e.g., Marx et al. 2009, Conti 2014, Contigiani et al. 2018). Whereas that stream of research parallels our study of employment protection, the two institutional contexts trigger fundamentally different strategic mechanisms. Contractual restrictions on workers’ voluntary mobility, such as noncompete agreements, affect firms’ R&D investment decisions by preventing the leakage of critical knowledge to competitors. Conversely, employment protection laws do not interfere with a firm’s ability to prevent knowledge leakage; instead, they alter inventors’ expectations of capturing future returns on firm-specific human capital. Thus, whereas Conti (2014) finds that the protection against leakage afforded by noncompete agreements encourages firms to pursue riskier projects, our findings suggest that protection against involuntary job loss leads inventors to favor exploitative projects of lower inherent novelty. This difference underscores the divergent implications of voluntary and involuntary mobility restrictions for firms’ innovation trajectories.

2. Conceptual Framework

The development of technological innovations requires scientists and engineers in R&D to deliberately invest in acquiring the knowledge and skills needed to build new ideas. As Bhaskarabhatla et al. (2021) show, inventors’ human capital is a primary driver of innovation. Whereas their initial stock of knowledge is largely shaped by formal education (Toivanen and Väänänen 2016), inventors continue to build human capital throughout their careers. In their daily search for solutions, R&D professionals are continually exposed to new technologies and ideas. The tools they use, the projects in which they engage, and the colleagues with whom they collaborate all play a critical role in shaping the evolution of their human capital (Akcigit et al. 2018, Bhaskarabhatla et al. 2021).

As a nonroutine task, R&D grants scientists and engineers considerable autonomy in how they approach problems (Bailyn 1985). This discretion allows inventors to shape, at least in part, the type of human capital they accumulate.¹ This matters because both the quantity and nature of accumulated human capital influence the market value of their skills (Topel 1990). A useful lens for analyzing these investments is the classic distinction in human capital theory between general skills—valuable across employers—and firm-specific skills—valuable primarily within the current employer (Becker 1962). Research on inventors’ interfirm mobility underscores the relevance of this distinction for R&D workers. For example, Marx et al. (2009) find that inventors specialized in narrow technological fields are disproportionately affected by noncompete clauses, which restrict job mobility and effectively increase the firm specificity of their human capital. Similarly, Palomeras and Melero (2010) show that knowledge characteristics associated with high firm specificity—such as complementarity with coworkers’ knowledge—reduce an inventor’s external market value, thereby limiting mobility.

Prior research also illustrates that inventors adjust their behavior in response to factors that affect their career prospects. In environments in which interfirm mobility is restricted by employer-friendly trade secret laws, Contigiani et al. (2018) find that R&D employees have weaker incentives to signal their quality to the external labor market and, as a result, reduce their quality-weighted patent output. At the same time, they shift toward developing general-purpose technologies that are more valuable in markets less affected by such legal constraints. Similarly, Marx et al. (2015) document that inventors in states where noncompete agreements become enforceable tend to relocate to states without such enforcement, seeking greater career flexibility. In a different context, Tandon and Toh (2022) show that inventors who are less concerned with their career prospects are less likely to align their search strategies with their firm’s technological priorities. Building on these insights, we argue that job security, by shaping inventors’ prospects of staying with their current employer, may affect their inventive behavior.

In particular, by increasing inventors’ expected tenure at their current firm, job security may influence their decisions regarding human capital accumulation. Employment protection does not guarantee continued employment but raises the cost of dismissal, thereby altering the probability and timing of job separation and increasing expected employment duration (Lazear 1990).² Even when baseline job stability is relatively high, such changes in expected tenure can meaningfully affect incentives to invest in firm-specific human capital (MacLeod and Nakavachara 2007). When workers anticipate a longer employment relationship, they are more willing to commit to their employers and to invest in firm-specific skills that would otherwise be suboptimal because of the potential holdup problem associated with early termination (Wang and Barney 2006). A reduced fear of job loss may offset this effect to some extent by diminishing workers’ incentives to put effort into acquiring any type of human capital (Ichino and Riphahn 2005). However, greater job security still places a premium on the relative incentives to invest in firm-specific rather than general human capital. Hence, the level of job security determines the mix of skills that workers choose to develop. When job security is low, workers are more likely to prioritize general, transferable skills that are not subject to the holdup problem. In contrast, higher job security comparatively encourages greater investment in firm-specific skills (Wasmer 2006, MacLeod and Nakavachara 2007). Acharya et al. (2014), who view innovations as inherently firm-specific investments, theorize that employment protection incentivizes non–R&D-specialized workers to generate innovations rather than perform routine tasks by reducing the risk of employer holdup. The corporate governance literature also suggests that provisions protecting managers against takeovers, such as poison pills or staggered boards, may induce them to make firm-specific human capital investments via increased job security (Wang et al. 2016).

In the case of specialized R&D workers, the firm specificity of their human capital investments depends on how deeply their acquired knowledge is embedded in the firm’s technological background. The intellectual effort required to produce innovations involves dedicating time and energy to acquiring and recombining technological knowledge. In this context, investment in firm-specific human capital takes the form of greater effort devoted to understanding and building upon technologies in which the employer holds distinctive expertise relative to other firms. This, in turn, shapes the inventor’s knowledge-sourcing process toward domains closely aligned with the firm’s existing technological base. Firm-specific technological knowledge is particularly valuable to the inventor’s current employer as it fosters innovations that align closely with the company’s particular needs and resources by leveraging internal complementarities. Such innovations are, therefore, primarily valuable to the focal firm. Alternatively, inventors who allocate more effort toward general human capital accumulation develop the capacity to create innovations that are valuable to a broader range of employers but may not align as closely with the specific capabilities and resources of their current organization.

As a consequence, inventors’ job security influences the types of innovations that they generate, and this is the ultimate object of our study. As inventors perceive an extended expected tenure within their organizations, they are more inclined to engage with firm-specific technological domains in their inventive activity. This leads them to develop innovations that draw more heavily on the company’s existing stock of knowledge. In contrast, inventors who perceive lower job security place relatively greater value on more broadly applicable technological knowledge, which is portable across employers. Consequently, inventors with lower levels of job security tend to develop inventions that draw less heavily on the firm’s existing stock of knowledge.

Importantly, the shift toward greater reliance on firm-specific technological knowledge induced by higher job security also affects other relevant aspects of innovation outcomes. In particular, job security may lead inventors to develop innovations that (i) exhibit a lower degree of novelty and (ii) are less likely to become breakthrough innovations. According to the innovation and organizational learning literatures, novel technologies often emerge from a process of broad search and recombination (Fleming 2001). Inventors who allocate greater effort toward firm-specific human capital may, therefore, be less suited for broad recombination endeavors, which typically require integrating diverse external knowledge (Ahuja and Lampert 2001, Rosenkopf and Nerkar 2001, Phene et al. 2006). The emphasis on firm-specific knowledge tends to induce local search, limiting inventors’ exposure to diverse technological inputs and increasing the risk of path dependence and cognitive lock-in (Thrane et al. 2010). This inward-looking approach can prevent them from leveraging insights from external sources, and lower exposure to new knowledge and perspectives decreases their ability to detect new opportunities for recombination (Arts and Fleming 2018), making them less likely to approach technological projects in fresh, innovative ways.

Moreover, inventors who allocate substantial effort toward firm-specific human capital may become less inclined to adopt radically novel approaches in their R&D projects. Their focus on generating innovations that complement the firm’s existing resources and capabilities naturally leads them to build extensively on the organization’s prior technological advancements. Because they have high stakes in the technological domains in which they have invested, their incentive is to generate innovations that consolidate that trajectory and open new applications for it rather than to pursue radically new technologies that may shift paradigms. Thus, within the classic exploration–exploitation trade-off framework in organizational learning (March 1991), job security is expected to encourage inventors to engage more in exploitative rather than exploratory innovation.

The decline in novelty associated with higher job security may also hinder the generation of breakthrough innovations, that is, innovations with particularly strong technological impact. As previous literature notes, novel innovations tend to exhibit greater variability in technological impact and are, therefore, a significant source of breakthroughs (Fleming 2001, Arts and Veugelers 2015). By encouraging greater reliance on the organization’s existing technological knowledge base, job security narrows the range of elements available for recombination in the innovation process. Reduced experimentation with new technological elements, in turn, lowers the variance in innovation success and, therefore, the probability of an innovation becoming a technological breakthrough (Fleming 2001). Accordingly, we expect that high job security reduces the likelihood of producing highly impactful innovations.

3. Empirical Setting, Data, and Methods

3.1. Wrongful Discharge Laws

We exploit the implementation of employment protection laws to identify plausibly exogenous variation in the job security of employee inventors. Because these laws restrict employers’ ability to terminate employees, we assume that their introduction increased job security for affected workers. In Section 4.3.1, we provide evidence that the adoption of these employment protection laws is associated with decreases in inventor turnover, indicating greater job stability for this particular group of workers.

To estimate the effect of job security on inventor behavior, we use a stacked DiD approach that exploits variation in firing costs arising from the staggered adoption of WDLs by U.S. state courts between the 1970s and the 1990s. Under the traditional employment-at-will doctrine, employers could fire workers “for good cause, bad cause, or no cause at all” (Payne v. Western & Atlantic Railroad, Supreme Court of Tennessee, 1884). During the last quarter of the 20th century, however, many state courts recognized three common-law exceptions to employment at will: the good faith exception, the implied contract exception, and the public policy exception. These exceptions reduced employers’ ability to terminate workers and constrained their effective firing authority.³

The good faith exception implies a tacit agreement to terminate employment only in good faith and fair dealing, preventing employers from firing workers for bad cause, such as depriving them of commissions or pension benefits. A broad interpretation of this exception implies a general prohibition against termination without just cause, such as economic necessity or poor performance, and can, therefore, have far-reaching consequences (Dertouzos and Karoly 1992, Autor et al. 2007). Legal scholars consider this exception the most significant departure from employment at will and the most far-reaching of the three exceptions (Kugler and Saint-Paul 2004). It provides employees with both contract and tort causes of action, allowing claims for contractual losses, emotional distress, and potentially punitive damages, all of which can be highly uncertain.

The implied contract exception protects workers from dismissal when an employer has made an explicit or implicit promise not to terminate without good cause. Such promises may arise from employment contracts, personnel policy handbooks, or expectations based on firm practices. Employers can largely shield themselves from implied contract claims by explicitly stating that employment is at will, and survey evidence suggests that most employers take such steps to mitigate litigation risk (Sutton and Dobbin 1996).

The public policy exception protects employees against dismissals that violate an important public policy, such as performing jury duty or reporting employer wrongdoing. Prior work suggests that this exception has limited legal and economic significance because compensation awards are rare and courts apply it narrowly (Edelman et al. 1992, Autor et al. 2006).

Figure 1 illustrates the adoption of all three WDL exceptions in U.S. states from 1970 to 1999. Our primary focus is on the adoption of the good faith exception. Although any of the three exceptions may increase employment protection in principle, prior research emphasizes substantial heterogeneity in their economic consequences across employment relationships. In particular, MacLeod and Nakavachara (2007) highlight that the good faith exception is a mandatory rule that applies broadly to all employment relationships in contrast to the implied contract exception, which applies only in limited circumstances when a long-term understanding between employer and employee can be established. Because of this difference in scope and applicability, the implied contract exception is less likely to affect perceived job security for high-skill workers.

Whereas Autor et al. (2006) find no robust aggregate employment or wage effects of the good faith exception, MacLeod and Nakavachara (2007) show that it has pronounced effects in occupations characterized by high levels of training and relationship-specific human capital, whereas its effects are muted or even negative in jobs requiring lower levels of human capital investment. They point out that estimates that average across heterogeneous workers may understate the relevance of the good faith exception for high-skill employees (such as employee inventors). Consistent with this, prior empirical research indicates that the adoption of this exception has significant implications for innovation, whereas the adoption of the implied contract and public policy exceptions does not (Acharya et al. 2014, Bena et al. 2022, Ekinci and Wehrheim 2024). We, therefore, focus on the good faith exception as our primary source of variation in employment protection.

A key identifying assumption in our analysis is that the adoption of the good faith exception is unlikely to be driven by unobserved changes in a state’s economic or political conditions correlated with innovation activity. Several pieces of evidence support this assumption. First, adoption occurs through state judicial precedents rather than prolonged legislative processes. Second, Acharya et al. (2014) show that adoption is not driven by shifts in political or economic conditions or by other labor regulations and that such decisions are unexpected. Third, we present a set of tests suggesting that the good faith exception generates plausibly exogenous variation in employment protection in our study of inventor behavior (Section 4.2.1).

3.2. Data Sources

Our sample consists of employee inventors at public firms in the United States. We obtain disambiguated inventor data and other basic information on patents from the U.S. Patent and Trademark Office’s (USPTO) PatentsView repository, which includes data on disambiguated inventors and patent filings since 1975. To associate inventors with firms, we use the updated Kogan et al. (2017) data from Stoffman et al. (2022), which assigns Center for Research in Security Prices (CRSP) firm identifiers to patent filings up to 2022. We then match these firms to Compustat using the Compustat-CRSP Merged database. For coding the introduction of WDLs, we follow prior literature and use the data from Autor et al. (2006), which provide state-level information on their adoption for the period 1970–1999 (see Online Table A.1).

To calculate control variables and those used in additional tests, we gather financial information from the Compustat Annual Files and state-level characteristics from the U.S. Bureau of Economic Analysis, the U.S. Bureau of Labor Statistics, and the U.S. Census Bureau as well as data on noncompete enforcement from Bird and Knopf (2009, 2015). We also use data on inventor age from Kaltenberg et al. (2023), text-based patent similarity data from Whalen et al. (2020), the private economic value of patents from Kogan et al. (2017), information on disruptive patented innovations from Funk and Owen-Smith (2017), and patent text novelty indicators from Kalyani (2024). Our working sample covers the years between 1975 and 2008, the period for which we can meaningfully track inventors and their innovations with patent data around the WDL introduction events codified in Autor et al. (2006).⁴

3.3. Stacked DiD Sample

To address the problems associated with staggered treatment timing when using traditional two-way fixed effects DiD estimators (de Chaisemartin and D’Haultfœuille 2020, Callaway and Sant’Anna 2021, Goodman-Bacon 2021), we employ a stacked DiD approach (Gormley and Matsa 2011, Cengiz et al. 2019, Xia et al. 2025) as recommended in recent econometrics literature (Baker et al. 2022, Roth et al. 2023, Wing et al. 2024). We use the coding of the adoption of the good faith exception provided by Autor et al. (2006) as our starting point. This coding is particularly appealing because it assigns a law change to the date of a precedent-setting court decision, ensuring that unexpected legal changes are used to assess their effects on outcome variables. We initially define the 13 U.S. states that adopted the good faith exception at different points in time (adoption events) as treated states and those that never adopted it as control states (see Figure 1 and Online Table A.1).

We examine the impact of the good faith exception over a five-year preadoption period and a 10-year postadoption period. We use a five-year preadoption period to minimize data loss because of insufficient observations in earlier years. We consider the 10 years after the adoption event to capture the long-term nature of innovation. If a state reversed the good faith exception, the postevent period ends at the reversal date. We exclude two states with adoption events that occurred before 1980 as we lack data for the entire pre-event period.

To identify treated inventors, we link the treated states to the PatentsView–Compustat data using the inventor’s location as reported in patent documents. Patent applicants are required to report their individual address on the patent application and have no economic incentive to misreport it (Alcácer and Zhao 2012). Whereas exact addresses are often unavailable, inventors’ states of residence are nearly always reported. Employment laws typically apply to the state where an employee works, and an advantage of our data is that the geographical information likely reflects the state where an inventor actually works rather than the state of the firm’s HQ. Because events with very few treated units can lead to unreliable estimates and potential bias (Wing et al. 2024), we drop four states that, after all sample filters are applied, contribute fewer than 25 treated inventors: Alaska (3), Nevada (7), Montana (12), and Wyoming (23). Our final sample includes seven adoption events, each with at least 160 treated inventors.⁵

We require candidate control inventors to have patented at least once within the event window (−5 to +10 years around the adoption decision) in any of the 37 states that never adopted the good faith exception as listed in Autor et al. (2006).⁶ We exclude inventors with patent filings in multiple states. We construct each event-specific data set by combining treated inventors with all control inventors who meet the above criteria for that event. Finally, we stack the seven data sets into a single combined file in which the unit of observation is the inventor–patent pair. Whereas the exact number of inventors varies by specification, our main sample comprises 9,148 unique treated inventors and 72,063 unique control inventors.⁷

3.4. Variables

3.4.1. Main Dependent Variable.

Given that the primary objective of our study is to investigate whether job security leads inventors to produce innovations that are more intensely based on the firm’s existing technological stock, we follow a large literature that relies on patent citation flows to measure this outcome (He and Wang 2009; Wang et al. 2009, 2017; Moeen et al. 2013; Alok and Subramanian 2023). Our main dependent variable, Share firm refs, measures the fraction of an inventor’s patent citations to prior art that reference patents owned by the inventor’s employer. Because patent law requires inventors to disclose the relevant prior art underlying an invention, citation patterns provide a meaningful indicator of the knowledge sources underlying inventive activity (Hall et al. 2001). A higher share of firm references, therefore, indicates greater reliance on the firm’s own prior technologies relative to external technologies. As documented by Moser et al. (2018), such references are an indicator of follow-on inventions within the same firm.

3.4.2. Main Independent Variables.

The most important independent variable in our analysis is the treatment indicator, Good faith, which equals one for all inventor–patent observations in the treatment group regardless of time and zero otherwise. The second key independent variable is the time indicator, PostGF, which equals one for all inventor–patent observations in the 10 years following the corresponding adoption event regardless of treatment and zero otherwise. These two variables allow us to define the interaction term Good faith $\times$ PostGF, which equals one for inventor–patent observations in the treatment group during the 10-year postadoption period and zero otherwise.

3.4.3. Control Variables.

We include inventor-level controls to account for heterogeneity in career stage and experience that may affect incentives to build on firm-specific technological knowledge. Specifically, we control for inventor age (Ln(Age)) and its squared term (Ln(Age)²) as well as tenure with the current employer, measured by years since the inventor’s first patent application at the firm (Ln(1+Tenure)) and its squared term (Ln(1+Tenure)²). We also include firm-level controls capturing innovation-related characteristics, including firm size measured by annual net sales (Ln(Sales)), R&D intensity (Ln(1+R&D/Assets)), and patenting intensity (Ln(Pats/Assets)). Finally, we include state-level controls to account for variation in economic conditions, human capital supply, political environment, and legal institutions. These controls include real state gross domestic product (GDP) (Ln(Real state GDP)), real state GDP growth (Real state GDP growth), the number of degree-granting institutions (Ln(Colleges)), higher education enrollment (Ln(Enrollment)), state population (Ln(Population)), unemployment rate (Unemployment rate), political balance in the U.S. House (Political balance), and the state-year noncompete enforcement index (NCA).

Table 1 provides an overview of the data set. Panel A lists variable definitions, panel B reports summary statistics, and panel C reports for each adoption event the number and share of inventors resident in the adopting state and in never-adopting control states.

Table 1. Variable Definitions, Summary Statistics, and the Number of Treated Inventors by Event

3.5. Stacked DiD Estimation

To examine the effects of the adoption of the good faith exception on inventors’ firm-specific investments, we estimate the following stacked DiD regression:

where i is the inventor who develops the patented innovation in state s (which belongs to region r). The patent p is applied for at date t by the inventor’s employer, j, and belongs to technology c according to the National Bureau of Economic Research (NBER) classification (Hall et al. 2001). The specific good faith adoption event is indexed by e. The unit of analysis is the inventor–patent pair.⁸ Importantly, Y${}_{i,p,j,e}$ is the dependent variable, which is the patent’s Share firm refs in our main specification. As explained above, the interaction term ${\mathit{\text{Goodfaith}}}_{i,j,e}\times {\mathit{\text{PostGF}}}_{t,e}$ captures the treatment. Thus, $\beta$ is our coefficient of interest, and it identifies the impact of the adoption of the good faith exception on the level of firm specificity of the innovations developed by treated inventors relative to control inventors. The vector Z${}_{i,j,t}$ corresponds to the set of time-varying control variables described in Section 3.4.3, ${\delta }_{s,e}$ corresponds to state $\times$ event fixed effects, which account for the effect of unobservable time-invariant factors at the state level,⁹${\theta }_{c,t,e}$ are technology $\times$ date $\times$ event effects that account for technology-specific time trends, and ${\kappa }_{r,t,e}$ are region $\times$ date $\times$ event effects that account for region-specific time trends (Autor et al. 2006).¹⁰ We follow the approach suggested by Wing et al. (2024) for stacked DiD and cluster standard errors at the state level.

A unique feature of our empirical design is that it allows us to include inventor $\times$ firm $\times$ event fixed effects, represented by ${\lambda }_{i,j,e}$. This set of employer–employee fixed effects accounts for the inventor’s innate ability, the employer’s unobserved innovation capabilities, and other time-invariant unobserved factors specific to the inventor, the employer, or the match between them (e.g., assortative matching). Furthermore, including these fixed effects rules out the possibility that the estimated effect of changes in employment protection on inventors’ behavior is driven by inventors who enter or leave employment or change employers.

4. Empirical Results

4.1. Main Results

We begin by investigating whether the increase in job security resulting from greater employment protection leads to a change in inventors’ reliance on their employer’s prior technologies. Figure 2 illustrates the dynamic effects of adopting the good faith exception on the share of backward references to firm-owned patents, comparing treated inventors with control inventors. The preadoption trends appear similar for the treatment and control groups, consistent with the parallel-trends assumption underlying our DiD approach. Specifically, we observe no significant differences in the share of firm references between treated and control inventors during the years preceding the adoption of the good faith exception. Furthermore, we observe no significant change in the adoption year or in the three subsequent years. A statistically significant increase emerges in the fourth year after adoption and persists thereafter. This pattern is intuitive as changes in inventor behavior typically require time to be reflected in their patents.

Table 2 presents the results from estimating Equation (1) with varying fixed effects schemes. Column (1) includes event-specific fixed effects for time, inventor, firm, state, and technology along with the good faith treatment indicator. Columns (2)–(4) progressively incorporate event-specific technology and regional trends. Columns (5)–(8) replicate these specifications but replace the separate event-specific fixed effects for inventor and firm with event-specific inventor–firm paired fixed effects, which represent our preferred choice. Across all specifications in Table 2, the treatment coefficient is positive, statistically significant, and ranges from 0.016 to 0.021. A coefficient of 0.020 in column (8) suggests that the good faith exception increases the inventor’s share of references to the inventor’s employer by approximately 2.0 percentage points compared with control inventors, corresponding to a 10.0% increase over the sample mean. This result is both economically and statistically significant, consistent with the visual evidence documented in Figure 2.

Table 2. Effect of the Good Faith Exception on the Types of Innovations Developed by Inventors

Columns (1) and (2) of Table 3 present models comparable to column (8) of Table 2, but they focus on the implied contract and public policy exceptions instead of the good faith exception. Specifically, we construct two new data sets centered on the adoption of these exceptions, following the same procedure applied to the analysis of the good faith exception as detailed in Section 3.3. The results indicate that the treatment coefficients associated with the adoption of the implied contract and public policy exceptions are very small and statistically insignificant. This finding aligns with the legal literature on the topic and recent studies on innovation as discussed in Section 3.1. Columns (3)–(5) of Table 3 return to our main data set to estimate the impact of the good faith exception, progressively adding two variables that indicate whether the implied contract and public policy exceptions were active (i.e., in effect) in a given state at a given time. The good faith treatment coefficient remains positive with a slightly larger marginal effect compared with column (8) of Table 2. By contrast, the coefficients for the implied contract and public policy exception indicator variables remain very small and statistically insignificant.

Table 3. Effect of the Good Faith Exception and Other WDLs on the Types of Innovations Developed by Inventors

Finally, columns (6)–(8) of Table 3 replicate the specifications of the previous three columns, also incorporating the control variables described in Section 3.4.3. The inclusion of these control variables has a limited impact on the estimated treatment effect. Column (8) of Table 3 includes the full set of control variables and constitutes our preferred specification. The treatment estimate indicates that inventors experiencing enhanced job security via the adoption of the good faith exception increase their reliance on firm-specific technologies by 2.2 percentage points with respect to the control group. This effect represents an 11.0% increase relative to the sample mean.¹¹

4.2. Robustness Checks

4.2.1. Accounting for Changes in Local Economic Conditions.

The source of variation in job security exploited in our analysis occurs at the state level. It is, therefore, important to assess whether our results reflect unobserved changes in local economic conditions that might simultaneously influence both the adoption of the good faith exception and inventors’ reliance on firm-specific technologies. We address this concern with several robustness tests reported in Table 4.

Table 4. Effect of the Good Faith Exception on the Types of Innovations Developed by Inventors: Robustness Tests

First, Hall (2004) documents an explosive growth in patenting after 1990, particularly in industries that had not previously relied heavily on patents as an appropriability mechanism. If geographic specialization across industries is important, this trend could differentially affect our measure of firm-specific innovations across states. In column (1) of Table 4, we restrict the sample to the period from 1975 to 1990. The estimated treatment effect remains positive and statistically significant. In column (2), we exclude inventor–patent observations in the semiconductor industry, which experienced especially rapid patent growth during the sample period and accounts for a substantial share of U.S. patenting activity (Hall and Ziedonis 2001). Specifically, we exclude patents classified in NBER technology subcategory 46. The estimated coefficient remains similar in magnitude and statistically significant, suggesting that our results are not driven by developments in this sector.¹² In column (3), we further account for heterogeneity in innovation regimes across state–technology combinations by including event–state–technology fixed effects. These fixed effects absorb all time-invariant differences in patenting intensity and innovation practices specific to a given state–technology pair within each stacked event window. The estimated treatment effect remains positive and statistically significant, indicating that our results are unlikely to be driven by persistent differences in state–technology innovation regimes.¹³

Second, Autor et al. (2004) observe that Southern states were slower to adopt the good faith exception and also experienced slower employment growth, whereas adoption was particularly prevalent in the Western states. To account for such regional differences, our baseline specification includes event-specific regional trends. In addition, column (4) of Table 4 includes an indicator for whether the good faith exception was active in a neighboring state. The coefficient on the neighboring state’s good faith exception is small and statistically insignificant, whereas the estimated treatment effect for the inventor’s own state remains positive and statistically significant. These findings suggest that regional economic conditions and spillovers are unlikely to explain our results.

4.2.2. Alternative (Text-Based) Measure for the Use of Firm-Specific Technologies.

We also assess the sensitivity of our results to the measure of inventors’ firm-specific innovations. Whereas the share of firm references is widely used in innovation studies for this purpose, concerns remain about the extent to which patent citations accurately capture knowledge flows (Kuhn et al. 2020, Whalen et al. 2020). For instance, Kuhn et al. (2020) highlight the increasing inclusion of potentially irrelevant references in patents over time. To address potential noise arising from this issue, we construct an alternative measure based on the text similarity between a patent and the firm-owned patents it cites. Text similarity between patent documents serves as a proxy for the technological proximity of the underlying innovations.

Specifically, following Whalen et al. (2020), we compute cosine similarity between vectorized representations of patent texts and use these similarity scores as weights on cited patents. Firm references that are more textually similar to the focal patent, therefore, receive greater weight. The resulting similarity-weighted share of firm references captures the extent to which cited firm-owned prior art is technologically proximate to the focal invention rather than more distant or generic. Accordingly, in column (5) of Table 4, we use this similarity-weighted share of firm references as an alternative dependent variable. The estimated treatment coefficient remains qualitatively similar to our baseline results, indicating that measurement error in patent citations is unlikely to drive our findings.

4.3. Evidence of the Mechanism

4.3.1. Employment Protection and Inventor Turnover.

To provide evidence that the effect of increased employment protection on inventor behavior operates through enhanced job security, we examine whether the adoption of the good faith exception leads to meaningful increases in job stability. Our analysis hinges on the premise that this legal shift extends expected employment duration, thereby strengthening the incentives for inventors to invest in firm-specific human capital. Whereas prior work indicates that the employment effects of the good faith exception are heterogeneous, they are most likely to take place in high-skill employment relationships characterized by idiosyncratic investments (MacLeod and Nakavachara 2007). As inventors exemplify this high-skill labor category, a reduction in their turnover serves as an indicator that the job security mechanism is active in our empirical setting.

In column (1) of Table 5, we estimate the impact of the good faith exception on inventor turnover. Following established approaches in the innovation literature (e.g., Marx et al. 2009, Melero et al. 2020), we identify turnover based on the assignment patterns of successive patents by an inventor, focusing on changes in the firms to which patents are assigned. Whereas we cannot distinguish between the effects of the good faith exception on voluntary and involuntary turnover, we find that its adoption leads to a 5.1-percentage-point decrease in overall separation, corresponding to a 30.7% reduction relative to the sample mean. This indicates that the legal change translates into a substantial increase in expected job stability for inventors, thereby providing an empirical foundation for our main argument: that longer expected horizons enable a shift toward investing in firm-specific technological knowledge.

Table 5. Effect of the Good Faith Exception on the Types of Innovations Developed by Inventors: Evidence on the Mechanism

4.3.2. Heterogeneity Across Inventors with Different Career Concerns.

To further examine whether increased job security influences inventors’ incentives to allocate effort toward firm-specific human capital, we explore heterogeneity in our results across inventors with different levels of career concerns. Because employment protection increases expected tenure, its impact on incentives should be particularly pronounced for inventors whose career concerns make them more sensitive to job loss. If the observed rise in reliance on firm-specific technological knowledge reflects a reallocation of inventive effort induced by job security, the good faith exception should have a larger impact on these inventors.

We use two proxies to capture differences in career concerns among inventors. The first proxy is the inventor’s age. Evidence from specialized labor markets, such as mutual fund managers (Chevalier and Ellison 1999), suggests that younger employees are more likely to face termination than their older counterparts. In addition, younger employees’ promotion prospects are more adversely affected by poor performance outcomes resulting from bold actions. Consequently, career concerns are likely to be stronger for younger inventors, and the impact of increased employment protection should be more pronounced for them. Following Baginski et al. (2018), we classify inventors in the top quartile of age as old inventors and those in the bottom quartile as young inventors, using inventor birth year data from Kaltenberg et al. (2023).

The second proxy is whether the inventor is a star inventor. The career concerns of star and nonstar inventors differ systematically. Compared with nonstar inventors, star inventors contribute more to innovation success and firm performance, exert greater control over research resources, and generate knowledge spillovers to peers (Zucker et al. 1998, Azoulay et al. 2010). They are also more attractive to other firms, which gives them broader external job opportunities (Ganco et al. 2015). As a result, increased employment protection is likely less relevant for their career concerns. We define nonstar (star) inventors as those in the bottom 90% (top 10%) of the citation distribution in the five years preceding the adoption of the good faith exception (see, e.g., Tandon and Toh 2022).

Columns (2) and (3) of Table 5 present the results of this test. We augment our preferred specification with interactions between the good faith treatment indicator and variables representing inventors with different career concerns. As expected, the effect of the good faith exception on reliance on firm-specific technological knowledge is concentrated among younger and nonstar inventors. The treatment coefficient for young inventors is positive and larger than the corresponding estimate for the full sample, whereas the estimated effect for inventors nearing retirement is small.¹⁴ A similar pattern holds when comparing star versus nonstar inventors, with the estimated effect concentrated among nonstar inventors. These heterogeneity patterns are consistent with the proposed mechanism: increased job security raises expected tenure and thereby reallocates inventive effort toward firm-specific technological domains, particularly among inventors for whom career concerns are most salient.

4.4. Addressing Alternative Explanations

4.4.1. Firms’ Reaction to the Increase in Labor Adjustment Costs and Litigation Risks.

Our findings suggest that increased employment protection encourages inventors to deepen their mastery of firm-specific technologies. Alternatively, because employment protection raises labor adjustment costs (Autor et al. 2007, Bena et al. 2022), firms might proactively concentrate investment in existing technologies, curtailing riskier, novel R&D (Griffith and Macartney 2014). To assess whether such firm-level responses could be driving our results, we exploit the fact that approximately 38% of inventors in our sample are employed in states other than their firm’s HQ.¹⁵ Because geographic distance may limit firms’ ability to monitor and direct detailed R&D practices, inventors located outside HQ are plausibly less exposed to centralized innovation policy (Glaeser et al. 2023).

We first compare inventors working in the firm’s HQ state with those located elsewhere. Our treatment indicator corresponds to activation of the good faith exception in the inventor’s state of employment. If our main results were driven by centralized firm-level responses to adjustment costs, the effect should be most pronounced for HQ-based inventors, who are most directly subject to corporate-wide innovation mandates (Bena et al. 2022). In contrast, if the mechanism operates through inventor-level job security, the effect should be stronger among non-HQ inventors, whose greater autonomy allows for more significant behavioral adjustments (Glaeser et al. 2023). As shown in Table 6, column (1), the effect is large and statistically significant for inventors located outside the HQ state but small and insignificant for those in the HQ state.

Table 6. Effect of the Good Faith Exception on the Types of Innovations Developed by Inventors: Alternative Explanations

We next focus exclusively on the subsample of non-HQ inventors. Column (2) replicates our main findings for this group. In column (3), we augment this specification with an indicator for whether the good faith exception is active in the firm’s HQ state. If centralized firm-level decision making were driving the results, this variable should carry some explanatory power. Instead, its coefficient is small and statistically insignificant, whereas the effect of the law in the inventor’s state of employment remains robust.

Finally, we address the possibility that WDLs represent an increase in labor litigation risk that leads firms to pursue derisking strategies. If the shift toward greater reliance on firm-specific technologies reflected firm-level retrenchment, the effects should be strongest among financially constrained firms, which are more sensitive to risk-induced cost shocks (Campello et al. 2010). Using standard measures of financial constraints measured prior to adoption (Almeida et al. 2004, Hadlock and Pierce 2010), we find no evidence supporting this interpretation (Online Table A.4). If anything, the effects are more pronounced among financially unconstrained firms. This pattern is consistent with our proposed mechanism: inventors respond most strongly to employment protection when the firm’s long-term stability and, thus, the expected return on firm-specific investment is highest. Overall, whereas we cannot fully rule out concurrent firm-level adjustments, the evidence in this section strongly supports the individual-level human capital mechanism as the predominant driver of our findings.

4.4.2. Changes in Collaboration Patterns and Innovation Effort.

We next examine whether our findings reflect, as theorized, inventors’ individual decisions to reallocate learning efforts toward firm-specific technological knowledge or whether they instead reflect changes in inventors’ collaboration behavior or overall innovation effort. Stronger employment protection may reduce inventor turnover and thereby limit the arrival of new collaborators within the firm. A reduced inflow of new colleagues could mechanically constrain exposure to external knowledge and increase reliance on internal technologies. If this mechanism were driving our results rather than a deliberate reallocation of inventors’ learning efforts toward firm-specific technologies, we would expect to observe declines in both the total number of collaborators and, in particular, the number of new collaborators. Columns (4) and (5) of Table 6 report the estimated effects of the good faith exception on the total number of coinventors and the number of new coinventors. The treatment coefficients are small and statistically insignificant in both cases, indicating that changes in collaboration patterns do not explain the observed increase in reliance on firm-specific technological knowledge.

We also investigate whether employment protection affects overall innovation effort. Prior work suggests that employment protection may either increase inventive effort by mitigating holdup problems (Acharya et al. 2014, Ekinci and Wehrheim 2024) or dampen effort by raising the cost of disciplining shirkers (Bradley et al. 2017, Francis et al. 2018). If such effort-based effects were operative in our setting, the increased reliance on firm-specific technological knowledge could reflect changes in innovation intensity rather than a reallocation of learning effort across technologies.

To assess this possibility, we aggregate the data to the inventor–year level and examine the impact of stronger job security on annual patent output.¹⁶ Column (6) of Table 6 shows that the estimated treatment effect is small and statistically insignificant, indicating no meaningful change in overall patenting. This result suggests that employment protection affects the direction rather than the intensity of inventive activity, consistent with a reallocation of effort toward firm-specific technological knowledge.¹⁷ Finally, Online Table A.5 examines additional citation-based measures of knowledge acquisition. Whereas backward citations to firm-owned patents increase following adoption, the total number of backward citations remains unchanged. This pattern also indicates a reallocation of effort toward firm-specific knowledge rather than a change in overall knowledge-search intensity. We also find that inventors place greater weight on high-quality firm references with no equivalent change in the quality of nonfirm references.

4.5. Innovation Novelty and Breakthrough Potential

Consistent with human capital theory, our results suggest that increased job security motivates inventors to allocate effort toward firm-specific human capital, reflected in a greater reliance on the organization’s internal technological knowledge base. At the same time, the innovation literature predicts that such a shift in incentives may lead to less novel technological advances as a stronger emphasis on firm-specific technological domains can result in increased local search and reduced exposure to diverse knowledge inputs (Ahuja and Lampert 2001, Rosenkopf and Nerkar 2001, Phene et al. 2006). This inward-looking search behavior may lead inventors to focus on incremental innovations that refine and extend the firm’s existing technological paradigm, overlooking more distant external knowledge. Because exposure to diverse knowledge sources is a key driver of novelty and breakthrough innovation (Fleming 2001), this reallocation of inventive effort may ultimately limit inventors’ capacity to generate radical and highly impactful technological advances.

To explore the implications of this mechanism for technological newness, we analyze the impact of job security on several widely used patent-level indicators of novelty. We begin by examining innovation disruptiveness, using the CD index and its extended version, the magnitude-weighted mCD index, as proposed by Funk and Owen-Smith (2017). These citation-based measures assess the extent to which an innovation disrupts a technological field by becoming a highly influential advance that renders prior knowledge less relevant. As Funk and Owen-Smith (2017) explain, the CD index captures the direction of a patent’s impact—whether it consolidates or destabilizes the existing technological trajectory—whereas the mCD index incorporates both the direction and magnitude of that impact (see also Azoulay et al. 2019). Columns (1) and (2) of Table 7 present the results, indicating that adoption of the good faith exception significantly reduces innovation disruptiveness. For example, column (1) shows that the CD index declines by 0.02 units, corresponding to a decrease of approximately 0.07 standard deviations.¹⁸

Table 7. Effects of the Good Faith Exception on Innovation Novelty, Breakthrough, and Value

Next, we rely on natural language processing techniques to assess innovation novelty based on patent text. Following Arts et al. (2021) and Kalyani (2024), we measure novelty using the share of two-word combinations (bigrams) that appear for the first time in the title, abstract, description, or claims of a patent. This measure captures the extent to which an innovation introduces novel conceptual elements and distinguishes more groundbreaking patents from more derivative ones. Column (3) of Table 7 reports results consistent with the disruptiveness analysis: adoption of the good faith exception leads inventors to produce innovations that are significantly less novel. Specifically, the estimated treatment effect implies a reduction in text novelty of 0.07 units, corresponding to a decrease of approximately 0.08 standard deviations.

Finally, we examine whether adoption of the good faith exception affects the likelihood of generating breakthrough innovations. Following prior research, we define breakthrough patents as those in the top 5% of the distribution of citations received from subsequent patents. Specifically, we construct an indicator equal to one if a patent ranks in the top 5% of patents filed in the same year and technological class (Ahuja and Lampert 2001, Phene et al. 2006). Column (4) of Table 7 presents the results. We again observe a negative and statistically significant effect: the likelihood of generating a breakthrough innovation is 1.6 percentage points lower for inventors experiencing increased job security, corresponding to a 21.9% decline relative to the sample mean. Taken together, the results in this section indicate that greater reliance on inventors’ firm-specific technological knowledge is associated with a systematic decline in the novelty and breakthrough potential of their innovations, consistent with predictions from the innovation and organizational learning literatures and with the exploration–exploitation trade-off highlighted in our theoretical framework.

4.6. Value of the Innovation to the Focal Firm

To further evaluate the economic implications of the firm-specific human capital mechanism, we examine how increased job security for employee inventors affects the value of their innovations to their employers. Our analysis does not yield an unambiguous prediction for this outcome. On the one hand, human capital theory suggests that greater job security mitigates the holdup problem, encouraging employees to make efficient firm-specific investments that enhance firm performance (Wasmer 2006, MacLeod and Nakavachara 2007). In the context of R&D, such investments should lead to innovations that are particularly valuable to employers as inventors allocate effort toward firm-specific human capital and exploit complementarities within the firm’s existing technological base. From this perspective, adoption of the good faith exception may increase the private, firm-specific value of innovations by better aligning them with the firm’s resources and capabilities.

On the other hand, the innovation and organizational learning literatures emphasize that firms often exhibit an inherent tendency toward exploiting their own technologies (Levinthal and March 1993). Absent countervailing incentives, this tendency can lead to excessive local search and reduced innovative performance (Ahuja and Lampert 2001, Rosenkopf and Nerkar 2001, O’Reilly and Tushman 2013). In this equilibrium context, firms typically design incentive structures that balance inventors’ incentives to invest in both firm-specific and general technologies. An exogenous increase in job security may, therefore, disrupt this balance, inducing an inefficiently high level of exploitation of firm-specific technologies and potentially reducing the value of the resulting innovations. Which of these forces dominates is ultimately an empirical question.

To address this question, we capture the private economic value of innovations to the focal firm using the stock market–based patent value measure proposed by Kogan et al. (2017) based on stock price movements following patent grants. As the authors note, this measure leverages the forward-looking nature of stock prices to estimate the private value of patented innovations to their holders based on ex ante information. Whereas other patent-based measures, such as citation counts, reflect broader technological or social value, this approach is designed to capture the private economic value of patented innovations to their owners. As emphasized by Farre-Mensa et al. (2020), stock market–based patent value measures incorporate not only the inventive component of an innovation but also its legal and strategic components, including the value of exclusionary patent rights.

Although the Kogan et al. (2017) measure is widely used in related studies, a concern is that it is inherently noisy at the individual patent level (Krieger et al. 2024, Arora et al. 2026). Beyond the strong distributional assumptions underlying its estimation, the measure is inferred from abnormal firm-day stock returns, which may reflect other contemporaneous firm-specific information in addition to the value of patent grants. More generally, stock market indicators of innovation value based on abnormal returns may be subject to evaluator bias if investors tend to overlook the value generated by exploitative innovations as suggested by Fitzgerald et al. (2021). To address these concerns, we complement the Kogan et al. (2017) measure with information from patent renewal decisions. Renewal decisions involve real economic costs and, therefore, reflect patentees’ private assessments of whether expected private returns exceed the costs of maintaining a patent (Schankerman 1998, Serrano 2010).¹⁹

Columns (5) and (6) of Table 7 show that increased job security for inventors leads, on average, to the development of innovations with higher private economic value for the focal firm. The estimated treatment effects imply that adoption of the good faith exception increases the stock market–based value of patents by approximately 9.3% and raises the likelihood of patent renewal by 4.5 percentage points, corresponding to an increase of about 9.6% relative to the mean renewal rate. These results are consistent with human capital theory, which predicts that increased expected tenure encourages efficient firm-specific human capital investments that firms might otherwise find too costly to incentivize because of the holdup problem.²⁰

5. Discussion and Conclusion

This article examines the impact of job security on inventor behavior. We argue that increased employment protection alters the incentive structure for human capital investment, driving inventors to deepen their mastery of firm-specific technologies. This behavioral shift, particularly prevalent in the context of private-sector employee inventors, reveals what we term the job security paradox: whereas job security provides the safety to fail that should encourage technological exploration (Azoulay et al. 2011, Manso 2011), it concurrently increases the expected returns to firm-specific human capital investments in ways that favor exploitation over novelty.

Exploiting the staggered adoption of the good faith exception to employment at will across U.S. states between 1980 and 1999—a period during which many industries were transitioning toward more open innovation regimes that increased the relative returns to general and transferable knowledge—we find that greater job security leads inventors to rely more heavily on their firm’s established technological base. This shift is associated with less novel innovations that, on average, generate greater private economic value for employers. Our supplementary analyses reinforce the interpretation that these patterns reflect inventors’ behavioral responses to enhanced job security rather than shifts in firm-level innovation policy. Moreover, whereas changes in inventive output could, in principle, reflect altered effort, we find no evidence that increased job security affects either patent productivity or the overall intensity of knowledge search. Instead, the evidence points to a reallocation of inventors’ human capital investments toward firm-specific technologies.

A key feature of our inventor-level analysis is the ability to control for firm–inventor match fixed effects. This approach allows us to examine changes in innovative behavior within employer–employee matches following increases in job security, excluding changes driven by the reallocation of inventors across firms. As such, our design isolates inventors’ direct responses to employment protection from potential composition effects, representing an important methodological advantage over prior work. At the same time, we acknowledge that focusing on inventor-level behavior necessarily provides only a partial view of the broader firm-level consequences of wrongful discharge laws.

Notwithstanding this limitation, our findings offer valuable insights for both researchers and practitioners. Whereas existing work often portrays job security as a cornerstone of human resource policies designed to motivate risk-taking and creativity (Amabile and Conti 1999, Azoulay et al. 2011, Manso 2011, Ederer and Manso 2013), our analysis highlights a more nuanced role in corporate R&D settings. Professional inventors engage in creative work that requires sustained investments in knowledge acquisition, much of which can be idiosyncratic to their employer. Consequently, the prospect of a stable employment relationship encourages inventors to deepen their knowledge of internal technological paradigms rather than explore a broader range of general technologies. Although this pattern appears economically beneficial for firms, it does not necessarily promote highly novel technological outcomes. Given the central role of a specialized R&D workforce in technological progress, future research on innovation incentives should take into account the career-driven motivations that shape inventors’ learning and search behavior.

Our findings also contribute to research on employment protection and innovation by highlighting the importance of worker-level responses. Whereas prior studies primarily examine how employment protection affects firm-level R&D decisions, we focus on how changes in job security shape inventors’ knowledge acquisition and, in turn, the nature of the innovations they produce. More generally, existing work on employment protection and innovation focuses almost exclusively on how firms respond to labor laws in their HQ state (e.g., Acharya et al. 2014, Griffith and Macartney 2014, Keum 2020, Bena et al. 2022). In contrast, our results highlight the importance of considering employee-level responses to local labor regulations.

From a managerial perspective, our study underscores the importance of tailoring innovation incentives to the specific characteristics of the R&D workforce. Unlike the general workforce, professional inventors do not choose between innovating and maintaining routine production; instead, they decide how to conduct innovation, including where to search for knowledge and which technologies to prioritize in their human capital investments. As our analysis shows, job security can meaningfully influence these choices. To effectively implement their technological strategies, firms must, therefore, ensure that employment and human resource policies for R&D workers are aligned with their broader innovation objectives.