Quick or Broad Patents? Evidence from U.S. Startups

Abstract

Patents reward inventors with monopoly rights over their inventions. The extent of the monopoly right depends on the patent’s breadth and timing. To maximize their monopoly rents, inventors prefer broader patents and patents that are granted as early as possible, leaving a longer period of exclusivity.¹ In contrast, society is better off with patents that are short-lived and just broad enough for patent holders to recoup their R&D investments without imposing undue burdens on rival inventors. Not surprisingly, patent scope and timing are considered the two fundamental levers of patent policy that together determine the effects of patents on innovation and economic growth (Gilbert and Shapiro 1990; Klemperer 1990; Chang 1995; Freilich 2015).

Most empirical work studying the effects of patents on firms and industries focuses on the effects of whether or not a patent is granted,² ignoring how broad the patent is or how long it took to issue. What little empirical work there is on scope and timing studies their effects in isolation,³ ignoring the practical and economic trade-offs between scope and timing faced by inventors and policy makers. As a result, we know little about the causal effects on inventors or their rivals of the two most fundamental levers of the patent system.

How do patent scope and timing affect the value of patent rights to their holders and the externalities imposed on their rivals? We answer these questions by estimating the economic value of broader and faster patents for U.S. startups that filed their first patent applications after 2001 and were granted a patent by December 31, 2013. While startups represent a small fraction of all patent applicants in the United States, focusing on them offers several conceptual and practical advantages. The innovative startups we study represent the population of patenting startups which, upon success, become major contributors to net job growth in the economy, generate wealth for their shareholders and employees, and increase their industry’s productivity through large spillovers (Haltiwanger, Jarmin, and Miranda 2013; Wu and Atkinson 2017). For example, the startups in our sample include several firms that have quickly grown to multi-billion-dollar valuations, such as Acceleron Pharma, iRobot, Pandora, Tesla Motors, and Zillow. Our focus on the first patent applications of such innovative startups allows us to assess the impact of patent scope and timing on economically important actors, whose fortunes early on are shaped by their intellectual property (Farre-Mensa, Hegde, and Ljungqvist 2020). As a practical matter, measuring the effects of a startup’s first patent permits identification without conflating the effects on firm performance of previously or simultaneously issued patents.

Estimating the effects of patent scope and timing is empirically challenging. To measure the value of scope, we need to consider what economic rents a startup could earn if its patent had broader or narrower scope. A twofold endogeneity problem makes this task difficult. First, unobserved quality differences across firms or inventions may affect both patent scope and a startup’s future performance, potentially leading to a spurious correlation between scope and performance: startups of unobserved better quality may seek patents for inventions deserving of broader scope, while enjoying better future performance regardless of the number of claims the U.S. Patent and Trademark Office (PTO) grants. Second, given that broader patents tend to take longer to issue, we need to account for the trade-off between patent scope and timing or else estimates of the effects of scope will be biased. Measuring the value of timely patent grants suffers from similar challenges: if the inherent trade-off between scope and timing is ignored, estimates of the effects of timing will be biased, as inventions of unobserved higher quality or by unobserved better applicants may be examined more speedily.

To overcome these challenges, we exploit plausibly exogenous variation in the patent examination process through an instrumental variables approach that leverages examiner-level variation in application review habits. The validity of the approach rests on two features of the patent examination process at the PTO. First, the PTO assigns applications in each technology field (or “art unit”) to examiners randomly with respect to the characteristics of the underlying invention (Lemley and Sampat 2012). Second, examiners vary in their review habits. Previous work has leveraged differences in examiner leniency with regards to approving patent applications to study the effects of patent grant on startup growth (Farre-Mensa, Hegde, and Ljungqvist 2020), the likelihood of a firm going public or being acquired (Gaulé 2018), and follow-on innovation (Sampat and Williams 2019). We study the effects of patent scope and timing by taking advantage of the quasi-random assignment of applications to examiners who differ in their leniency with regards to patent scope and in their examination speeds.

We utilize a rich data set that combines administrative data from the PTO’s internal databases, which cover the population of granted and rejected applications, with data on four types of firm-level outcomes: (a) growth in sales and employment (from Dun & Bradstreet’s NETS database); (b) follow-on patenting and citations (from the PTO’s database); (c) venture funding (from VentureXpert); and (d) fundraising by startups through initial public offerings (IPOs) (from VentureXpert and Thomson-Reuters’ SDC database). Our sample covers all 34,359 first-time patent applications filed by U.S. startups at the PTO since 2001 that received a final decision by December 31, 2013. For our main results considering the effects of patent scope and timing, we focus on the 22,001 of these applications that are ultimately granted.

Our estimates show that delays in granting a startup’s first patent have a significant negative effect on its growth. This finding is consistent with a speedier patent grant enabling a startup to more quickly commercialize its invention, while preempting the entry of rivals, and thus to enjoy higher growth. Economically, the effects are large, with a 1-year increase in examination time reducing the average startup’s growth in employment and sales by 12.8 and 20.4 percentage points over 5 years, respectively, equivalent to 13.5 fewer person-years of employment and a cumulative loss in sales of $2.6 million over 5 years.

The scope of a startup’s first patent delivers nuanced benefits: unconditionally, broader scope has little effect on growth in employment or sales, but it marginally reduces a startup’s chances of survival (perhaps because it makes the startup an attractive acquisition candidate). Among startups that survive as independent firms, broader scope significantly boosts long-term growth in employment and sales. This finding is consistent with broader scope enabling the startup to exclude more competitors and thus enjoy higher revenues (through greater sales volumes, higher unit prices, or both). In addition, broader scope increases the likelihood of the startup raising IPO capital: each granted claim roughly doubles the likelihood of an IPO.

Broader scope and speedier grant of a first patent spur innovation by the patent holder. Startups that receive broader patents subsequently innovate more and produce higher-quality inventions. Each additional granted claim in a startup’s first patent increases the subsequent number of patents the startup applies for and is granted by 5.9|$\%$|⁠, the total number of citations to its subsequent applications by 12.7|$\%$|⁠, and per-patent citations by 6|$\%$|⁠. Patent grant delays, on the other hand, have a negative effect on innovation at the startup. A 1-year increase in examination time reduces the numbers of subsequent patent applications and subsequent patent grants by 8.4|$\%$|⁠, the approval rate of subsequent applications by 3 percentage points, and the number of citations to subsequent applications by 12.1|$\%$| in total and 5.3|$\%$| on average.⁴|$^{,}$|⁵

Finally, we examine the externality effects of a startup’s quasi-randomly determined patent scope and timing on other startups that operate in the same narrowly drawn technology area and thus can plausibly be considered rivals. We find evidence that broader scope imposes negative externalities on rivals’ growth, consistent with theoretical work arguing that broader patents increase the cost of entry for competitors (Gilbert and Shapiro 1990), and the quality of their inventions. These externalities are large: each additional claim a startup is granted reduces its rivals’ employment and sales growth by 5.0 and 7.6 percentage points over 5 years, respectively, and the citation impact of their future patents by 4.2|$\%$| in total or 2.5|$\%$| for the average patent. Speedier grants, on the other hand, impose positive externalities on rivals with regards to growth, survival, VC funding, access to the stock market, and innovation. Again, these externalities are large: a 1-year reduction in examination time for the focal startup increases industry employment and sales growth by 4.2 and 6.3 percentage points over 5 years, respectively, the likelihood that a rival obtains VC funding in the next 5 years by 11.3|$\%$|⁠, the chance of an IPO by 6.6|$\%$|⁠, the number of patents its rivals apply for and receive by 16.7|$\%$| and 16.1|$\%$|⁠, respectively, and the count of citations to rivals’ subsequent patent applications by 15.0|$\%$|⁠. We suggest that speedier patent grants resolve uncertainty about property rights and thus about the broader intellectual property landscape, facilitating investments by other startups in the patent holder’s industry.

Our findings are robust to concerns stemming from applicants’ use of the PTO’s continuations procedure, alternative measures of scope, and potential noise in the NETS data that may result from the presence of imputed values for some companies.

Our study contributes to the literatures on innovation, entrepreneurial finance, and economic growth. We provide the first causal estimates of the effects of two key determinants of patent value, namely, scope and timing, on an economically important sample of innovative startups. Prior empirical work on patent characteristics is limited to patent scope. In an important study in this vein, Lerner (1994) uses a sample of 173 venture-backed biotechnology firms and shows that a one-standard-deviation increase in average scope is associated with a 21|$\%$| increase in firm value. Ordinary least squares (OLS) estimates, such as Lerner’s, may capture the positive effects of not only patent scope but also the value of the underlying technology. Our identification strategy disentangles the two and shows that the positive effects of scope in our sample are more nuanced.

Timeliness of patent grant, which has received little attention from empirical scholars, has large positive effects on both the startup and its industry. Patent policy is often characterized as striking a fine balance between rewarding inventors and limiting the negative externalities of exclusionary rights. Our results suggest that speedy patent grants unambiguously improve rewards for first-time startup patentees and may increase overall welfare, through at least three mechanisms: (a) by conferring a longer stream of monopoly rents (as patents are valid for a maximum of 20 years from application but can be effectively enforced only after grant), (b) by allowing the holder to gain a competitive edge over rivals engaged in a patent race (Reinganum 1982; Gilbert and Newbery 1982; Fudenberg et al. 1983), and (c) by facilitating investment in the patent holder’s industry through the resolution of uncertainty about the intellectual property landscape in the industry. In contrast, patents that are broad, while privately valuable for startups, impose negative externalities on growth and innovation among the patent holder’s rivals.

Finally, our identification strategy, based on quasi-random assignment of applications to examiners of varying scope leniency and review speed, highlights the profound impact the luck of the draw can have on the fortunes of U.S. startups. Drawing a slow examiner with a tendency to disallow most claims adversely affects a startup’s future prospects. More broadly, our findings extend the growing literature in finance that unpacks the effect of policy instruments on the financing and growth of innovation (Acharya, Baghai, and Subramanian 2014; Fang, Lerner, and Wu 2017; Heath and Mace 2020).

1. Institutional Setting and Data

1.1 The patent examination process

The PTO assigns each incoming patent application to the relevant “art unit” for review. Each art unit consists of a group of patent examiners who specialize in the same narrowly defined technology field. During our sample period, the PTO employed over 13,000 examiners in more than 900 art units. The median art unit has 13 examiners; the largest, more than 100.

In each art unit, applications are assigned to one of the art unit’s examiners, who is responsible for evaluating whether or not the claims in the application meet the legal standards for novelty, usefulness, and nonobviousness. As we argue and will show below, the assignment of an application to an examiner is orthogonal to the characteristics or quality of the application or of the applicant. This quasi-random assignment is central to our identification strategy.

After being assigned an application, the examiner reviews the application and makes a preliminary decision regarding which, if any, claims in the application will be allowed. This preliminary decision---the “first-action decision”---is communicated to the applicant by letter. Through this letter, the applicant first learns the examiner’s identity. On average, sample applications take 0.7 years to be assigned to an examiner, who then takes an additional 1.1 years to make a first-action decision. In our sample of patent applications that are eventually granted, the final decision to accept is made 1.4 years later (i.e., 3.2 years after the application date).

1.2 Patent data and sample selection

To study the value to a startup of a broader patent and faster examination time, we obtain data on approved patents directly from the PTO’s internal research databases, which contain records of all patent applications, both approved and rejected, from 1976 to the present.⁶ Our sample starts in 2001, though we use data from previous years in the construction of our instruments for patent scope and examination time.

Since the PTO does not identify startup applicants, we follow Farre-Mensa, Hegde, and Ljungqvist (2020) and construct a sample of startups as follows. First, we restrict the sample to applications by incorporated applicants based in the United States. Second, we remove not-for-profit organizations, such as universities, government research labs, hospitals, and charities, based on a manual review of the patent assignee’s name. Third, we construct a mapping from the patent data to Compustat to screen out applicants that are or have previously been listed on a stock market at the time of the application, another mapping to NETS to screen out applicants that are a subsidiary of another firm at the time of the application, and a third mapping to Thomson-Reuters’ SDC database to screen out applicants that have been acquired between filing and first-action (as we cannot disentangle the effects of patent scope or examination time from the effects of the acquisition in this case). In each record linkage, we match on standardized assignee name-stems (being careful to take name changes into account using the name change histories available in Compustat, NETS, and CapitalIQ) and block on location at the state and county levels. Manual disambiguation and deduplication relies on the patented invention and google searches.

These three steps identify patent applications filed by stand-alone for-profit U.S.-based firms. Not all these firms are startups. We apply two further filters to identify startups. First, we only include filers that qualify for reduced filing fees as “small business entities” under Section 3 of the Small Business Act. Second, we exclude applicants that have filed any patent application in the 25 years before our sample period. This step requires identifying each patent’s original applicant (since many patents are reassigned over time) and accounting for name changes.

Our analysis examines how scope and timing affect a startup’s ability to grow, fundraise, and innovate over a period of up to 5 years from the evaluation of its first patent application. To this end, we require firms to receive a first-action decision by the end of 2009 and a final decision by the end of 2013.⁷ Because scope is a feature of a granted patent, we focus on the 22,001 applicants whose first application was ultimately approved (referred to as sample startups). Of these, 7,437 (33.8|$\%$|⁠) are granted biochemistry patents (PTO technology centers 16 or 17) and 3,723 (16.9|$\%$|⁠) are granted IT patents (PTO technology centers 21, 24, 26, or 28).

1.3 Timing considerations

Outcomes could be measured from three different starting points: the filing date, the first-action date, or the final-decision date. The appropriate starting point in our setting is the first-action date. The first-action decision resolves a substantial amount of uncertainty regarding the scope (and patentability) of an invention.⁸ After first-action, the applicant can take actions that endogenously affect the remaining time it takes the examiner to reach a final decision.⁹ Resolution of uncertainty is a necessary, but not sufficient, condition for an application to affect outcomes, while the endogenous timing of the final decision could confound our estimates.

1.4 Measuring scope and examination time

Following Marco, Sarnoff, and deGrazia (2019), we measure patent scope as the number of independent claims in a patent grant. The intellectual property protected by a patent is defined by a set of claims made in the application. The broadest of these are called independent claims. These stand independently and do not refer to any other claim in the patent application, while dependent claims reference independent claims and qualify them (Harhoff 2016). Together, the set of claims represents the breadth of the intellectual property covered by the patent. In robustness tests, we consider alternative measures based on the total claim count (Lanjouw and Schankerman 2004) and the word count of the first claim (Kuhn and Thompson 2019).

We measure examination time from the application filing date to the first-action date. As mentioned earlier, subsequent delays are inherently endogenous, as applicants’ actions in response to the first-action letter affect the remaining timing of the patent evaluation process.

1.5 Data on firm outcomes

Because the startups in our sample are privately held, they are not covered in standard financial databases, such as Compustat. We collect data on firm outcomes from four sources.

• Dun and Bradstreet’s National Establishment Time Series (NETS) database. NETS is similar to the U.S. Census Bureau’s Longitudinal Business Database in that it aims to cover the universe of business establishments in the United States but offers the advantage of not requiring special permission for access. The NETS data used in this project cover the period through December 2016, providing us with 5 years of post-first-action sales and employment data for all firms that we are able to match. To match startups to NETS, we utilize a “fuzzy” matching algorithm (with each candidate match verified manually) based on standardized firm name-stems and locations, in conjunction with information on name changes obtained from NETS and CapitalIQ and location moves obtained from the PTO’s firm name and address register. We match 81.5|$\%$| of sample startups to firms in NETS---a higher match rate than that achieved by studies using Census Bureau data.¹⁰

• The PTO’s patent database. This database provides data on sample startups’ subsequent patent applications as well as citations to their patents through December 2016.

• VentureXpert. This database contains VC funding events. We use it to identify which sample firms go on to raise VC funding after the first-action date through February 2020.

• The Thomson Reuters Securities Data Company (SDC) database. We use data from SDC (and VentureXpert) to identify firms that raise capital from public investors via an initial public offering (IPO) of equity on a stock market through November 2018.

Table 1 provides summary statistics for our sample. Panel A shows that at the time of application, the median startup is 2 years old, has 8 employees, and $0.8 million in sales. Following the PTO’s first-action decision, the average startup experiences 21.1|$\%$| growth in employment and 44.4|$\%$| growth in sales over 5 years (panel B) and produces 2.6 subsequently approved patents (panel C). In the 5 years following first-action, 9.4|$\%$| of sample startups raise VC funding and 0.8|$\%$| of them complete an IPO.

2. Empirical Strategy

We focus on how two key examination characteristics---the scope of a granted patent and the length of time an application takes to be examined---affect a startup’s subsequent growth in employment and sales, ability to raise external capital, and follow-on innovation. In this section, we outline our empirical strategy, review the main challenge to identification we must overcome, and outline our identifying assumptions. In the next section, we present our findings on the effects of patent scope and examination time on U.S. startups.

2.1 Empirical setup

2.2 Empirical strategy and identifying assumptions

The coefficients of interest in Equation (1), |$\beta _1$| and |$\beta _2 $|⁠, capture the average treatment effects of patent scope and examination time, respectively. These average treatment effects capture the conditional average difference in outcomes between a startup that receives a patent of given scope in a given time frame compared to a startup subject to identical demand and technology conditions that is granted a patent of different scope in a different time frame. The key challenge to identification is to ensure that differences in scope and timing do not reflect differences in the quality or characteristics of the underlying invention, the applicant, or the application.

In an ideal experiment, we would randomize patent scope and examination time to ensure that unobserved quality differences do not confound the effects of patent scope and examination time. While this ideal experiment is not feasible, we exploit two lottery-like features of the PTO’s review process that have been employed in previous research:¹⁴ patent applications are assigned to examiners within an art-unit in a quasi-random fashion; and patent examiners differ systematically in their review habits. This second feature has been used to argue that more lenient examiners are more likely to grant a patent than are stricter examiners, holding the quality of the invention constant (Sampat and Williams 2019; Farre-Mensa, Hegde, and Ljungqvist 2020). Under these assumptions, previous research has used examiner leniency with respect to patent approval as an instrument for the approval of a patent, allowing for causal estimation using 2SLS. We extend this logic from patent approval to patent scope and examination time, on the assumption that randomly assigned patent examiners vary systematically in their propensity to allow more or fewer claims and in how long it takes them to arrive at a first-action decision.¹⁵

Figure 1

Joint distribution of examiner scope leniency and examiner review speed

The figure shows the joint sample distribution of examiner scope leniency and examiner review speed. Examiner scope leniency is defined as the average count of independent claims in patents previously allowed by an examiner, estimated within art unit and year using a regression of the count of independent claims on a full set of art-unit-by-application-year fixed effects. Examiner review speed is defined as the review speed from application date to first-action date, estimated within an art unit and year using a regression of examiner review speed on a full set of art-unit-by-application-year fixed effects. In panel B, data points are grouped into 100 equally sized bins and an aggregate statistic is used to summarize each bin. The OLS line drawn in panel B represents a slope of.006 with a |$p$|-value of.217.

Open in new tabDownload slide

The use of two examiner characteristics as instruments raises the question whether these characteristics are sufficiently distinct from each other to permit causal inference of both patent scope and timing, or whether they reflect unobserved characteristics that drive both (say, “attention to detail”). Figure 1 graphs the joint distribution of scope leniency and review speed (residualized against a full set of art-unit-by-application-year fixed effects). The figure shows that examiners vary independently in their scope leniency and review speed. Independence is stronger than necessary for causal inference of both scope and timing, which only requires that the two instruments not be perfectly multicollinear (Stock and Watson 2015, ch. 12).

Figures 2 and 3 show the marginal distributions of residualized examiner scope leniency and review speed. Both characteristics vary substantially across examiners (even within art unit and year), which bodes well for our ability to use these examiner characteristics to identify Equation (1). To illustrate, compare an examiner at the 25th percentile to an examiner at the 75th percentile. This corresponds to a difference of 0.4 claims allowed and 6.4 months to first-action. To put these numbers in perspective, 0.4 claims represent a 14.3|$\%$| increase relative to the median number of claims. Based on Kuhn and Thompson (2019), the median patented invention is worth around $6.4 million in 2012 dollars. Assuming value increases linearly in the number of claims, a 14.3|$\%$| broader patent might then be worth around $0.5 million more. A time saving of 6.4 months represents a 38.2|$\%$| reduction in the median first-action examination time in our sample.

Figure 2

Distribution of examiner scope leniency

The figure shows the sample distribution of examiner scope leniency, defined as the average count of independent claims in patents previously allowed by an examiner, estimated within an art unit and year using a regression of the count of independent claims on a full set of art-unit-by-application-year fixed effects.

Open in new tabDownload slide

Figure 3

Distribution of examiner review speed

The figure shows the sample distribution of patent examiners’ historic review speeds measured from application date to first-action date, estimated within an art unit and year using a regression of examiner review speed on a full set of art-unit-by-application-year fixed effects.

Open in new tabDownload slide

2.3 First-stage estimates

The first-stage estimates of equations (4) and (5) are reported in Table 2, panels A and B, respectively. Both first-stage estimates suggest that our instruments satisfy the relevance condition for identification in a 2SLS framework. The estimates of |$\theta $| and |$\delta $| confirm that an examiner’s past scope leniency is a strong predictor of the number of claims she will grant in a given application and that her average past review speed combined with the docket date lag is a strong predictor of how long she will take to reach a first-action decision. The coefficient estimate for |$\theta $| in column 1 in panel A suggests that an increase in scope leniency of one independent claim leads to a 0.52 increase in the number of independent claims in the patent granted to the startup (⁠|$p <.001$|⁠). Similarly, a 1-year increase in the review speed in panel B leads to an increase of 0.53 years (6.3 months) in first-action examination time (⁠|$p <.001$|⁠). Both instruments are statistically strong, with |$F$|-statistics well above the rule-of-thumb value of 10. This ensures that our 2SLS estimates are not likely subject to weak-instrument bias.

2.4 Threats to identification

For our instruments to be valid, they must satisfy two further conditions. First, they must meet the exclusion restriction, which requires that each instrument has no direct effect on the outcome except through the treatment. In this case, quasi-random assignment of patent applications to patent examiners ensures that the exclusion restriction is plausibly justified.

The second condition requires that the instruments must not correlate with omitted variables that could drive a startup’s future success. If this were the case, our instruments would not be “as good as randomly assigned conditional on covariates” (Angrist and Pischke 2009, p. 117). This could happen if the characteristics of the startup or the application influenced assignment of an application to an examiner. We investigate this threat to identification in three ways. First, based on prior literature and institutional grounds, we argue that patent applications are in fact assigned quasi-randomly within art units. Second, we conduct Righi and Simcoe’s (2019) validation test of quasi-random assignment. Third, we examine whether an examiner’s review habits correlate with observable characteristics of the applicant or the application.

As noted above, a large body of literature argues that the PTO assigns applications to examiners quasi-randomly. The precise details and procedures of the assignment process vary across art units,¹⁸ but they have in common that they are consistent with our identifying assumption that applications are assigned randomly with respect to application or applicant quality. Importantly, given our focus on scope, Righi and Simcoe (2019) report that there is no evidence that particularly important or broad applications are assigned to specific examiners.

Next, we implement Righi and Simcoe’s (2019) validation test. Under the null of quasi-random assignment, the first-stage estimates of |$\theta $| and |$\delta $| in equations (4) and (5) should be invariant to the characteristics of the startup, application, and examiner. Accordingly, adding further controls to our first-stage regressions shown in Table 2 should not change |$\hat{\theta}$| and |$\hat{\delta}$|⁠. Specifically, we add size and growth (in sales and employment) at the time of first-action to investigate the possibility of assignment based on applicant characteristics (columns 2 and 3), highly granular technology-subclass-by-year fixed effects to investigate the possibility of assignment based on technological specialization (column 4), and examiner tenure and seniority to investigate the possibility of assignment based on examiner experience (column 5).

Adding controls makes little difference to the first-stage coefficient estimates of |$\theta $| and |$\delta $|⁠. In Table 2, panel A, the coefficients for patent scope vary between 0.49 and 0.55. In Table 2, panel B, the coefficients for review speed vary even less, ranging from 0.52 to 0.53. Both instruments thus pass Righi and Simcoe’s (2019) validation test. None of our measures of applicant quality predicts patent scope or the length of examination time. Examiner tenure and seniority, on the other hand, do influence patent scope and examination time: examiners with longer tenure grant broader patents, and more senior examiners reach first-action decisions more quickly. However, this does not undermine identification, given random assignment, and adding these examiner characteristic does not significantly alter the estimates of |$\theta $| and |$\delta $|⁠. In short, Table 2 is consistent with our assumption that unobserved examiner or applicant characteristics are unlikely to be correlated with both our instruments and our outcomes of interest (i.e., startup success).

To shed further light on the random-assignment assumption, Table 3, panels A and B, test whether our two instruments correlate with observable characteristics of the applicant or the application, in which case selection into the sample might be a concern. Columns 1 and 2 in each panel show that applicant characteristics, such as size, growth, and age, do not predict whether a scope lenient or speedy examiner is assigned. Column 3 shows that application characteristics (including claim count at pregrant publication, average word count across claims, count of backward citations, and count of citations to nonpatent literature) do not predict examiner assignment either. Columns 4 and 5 report a placebo test which exploits the fact that a subset of startups file for patent protection not just in the United States, but also at the European Patent Office and/or Japanese Patent Office. For this subset of applications, we use foreign patent grants as a measure of the quality of the applicant or the underlying invention to validate the quasi-random-assignment assumption. Specifically, we test whether applications granted by a foreign patent office are more likely to have been assigned to more scope-lenient or faster U.S. examiners. Consistent with quasi-random assignment, we find no such evidence.