Violence and economic activity: evidence from African American patents, 1870–1940

Abstract

Recent studies have examined the effect of political conflict and domestic terrorism on economic and political outcomes. This paper uses the rise in mass violence between 1870 and 1940 as an historical experiment for determining the impact of ethnic and political violence on economic activity, namely patenting. I find that violent acts account for more than 1,100 missing patents compared to 726 actual patents among African American inventors over this period. Valuable patents decline in response to major riots and segregation laws. Absence of the rule of law covaries with declines in patent productivity for white and black inventors, but this decline is significant only for African American inventors. Patenting responds positively to declines in violence. These findings imply that ethnic and political conflict may affect the level, direction, and quality of invention and economic growth over time.

Social instability and political conflict in a country are strongly associated with poor economic outcomes. A large literature has documented their effects on growth and development in both cross-country (Acemoglu and Robinson 2001; Alesina et al. 1996; Barro 1991; Mauro 1995) and within-country settings (Abadie and Gardeazabal 2003). Political unrest has been associated with reduced savings and investment (Alesina and Perotti 1996; Venieris and Gupta 1986), and property values suffer when civic strife breaks out (Besley and Mueller 2012). Likewise, Luigi et al. (2009) have shown that past warfare is negatively associated with international trade volume in the present.

One issue that has gone unexamined is the effect of violent conflict and social instability on inventive activity and the creation of intangible capital. While there is good experimental evidence that socioeconomic stress can lead to lower scores on cognitive tests (Duncan et al. 2011; Duncan and Magnuson 1998), and leads to poor economic decision-making (Shafir and Mullainathan 2012), the link with economically relevant intellectual output has not been made.

I show that patenting rates by African Americans in the 19th and 20th centuries systematically declined in areas affected by race riots and lynchings. Although violence might seem unrelated to the inventive process, it affected inventors and other economic agents. Between 1870 and 1940, race-related violence in the United States increased dramatically. Major race riots peaked in 1919 and 1921, lynchings in 1892 and 1893, and passage of state segregation laws in 1908, 1928, and 1933. These trends provide a natural experiment for testing the extent to which a shock to personal security and property rights can affect individual creativity and the production of economically intangible capital.

Using a novel data set and exploiting an historical experiment, I find that extrajudicial killings and loss of personal security depressed patent activity among blacks by more than 15 % annually between 1882 and 1940. In addition, productive activity increased after violence ceased. I also find that patenting was lower in states with more riots and laws promoting segregation than in other states and that these factors accounted for more than 1,100 “missing” patents (that is, patents that would have been applied for and obtained absent these factors) over the period compared to 726 actually obtained by African Americans. Finally, using a “placebo study,” I find that a similar shock of increased hate-related violence to white inventors would have depressed U.S. patenting activity by nearly 40 % and ostensibly would have resulted in significantly greater volatility in technological change.¹ The economic significance of the findings in this paper implies that, then and now, conflict and hate-related violence, and the resulting uncertainty in property-rights enforcement, may substantially affect the level, direction, and quality of inventive activity and economic growth.

1 Violence and inventive activity, 1870–1940

1.1 Violence and segregation

Following the emancipation of slaves after the Civil War, race-related violence escalated in the South in the 1870s and spread to other parts of the country by the end of the nineteenth century. Such conflict was often related to the absence or diminished enforcement of the rule of law.²

Major race riots are one indicator of hate-related violence. As reported in Table 1, these events were occasionally politically motivated and were sometimes associated with mob violence and election disputes; blacks were usually, but not always, the targets of race riots. This paper includes only riots that resulted in major violence and loss of life and property and that received national media coverage. There were many smaller riots. Yet, data have not been systematically collected on minor riots, and I exclude them from this paper. In the historical literature there is no universally-accepted set of sufficient conditions that would predict race riots during this period. Race riots frequently had legal and political consequences, such as the imposition of martial law and the ousting of democratically elected black and moderate white officials, along with economic consequences, such as looting of black business districts and destruction of entire black farms, firms, and residential neighborhoods.³ Riots were largely concentrated in the South before 1900 and in the North after 1900. The effects of violence on black economic activity would have been both direct—for example, black inventors’ workshops were located in the affected business districts—and indirect—for example, riots lower the value of commercial and residential property (Collins and Margo 2003), which would reduce financing opportunities and increase operating costs.

Table 1 Conflict, rule of law, and segregation laws, 1870 to 1940

Riots often had consequences far beyond their cities and states of origin. The East St. Louis race riots in May and July 1917 involved a mob of nearly 3,000 white men, several lynchings, as many as 150 black deaths, and extensive damage to black homes and white firms, including a warehouse of the Southern Railway Company (Garvey 1917/1983; New York Times 1917). In support of the victims and in protest of the failure of East St. Louis and other authorities to protect their citizens, the National Association for the Advancement of Colored People (NAACP) organized a “silent march” of 15,000 people down Fifth Avenue in New York City in late July 1917. Historian John Hope Franklin writes, “It was the epidemic of race riots that swept the country early in the century that aroused the greatest anxiety and discomfort among the African-American population.... Riots were perceptibly increasing, and their dramatic nature had the effect of emphasizing the insecurity of blacks throughout the country” (Franklin and Moss 1994, p. 313).

Lynching also may be considered a proxy for absence of the rule of law.⁴ Whereas race riots involved opposing groups, lynchings typically involved a group taking action against a specific individual or individuals. In addition to killing the victim, often a secondary objective was the externality a lynching produced—to intimidate the victim’s family, community, or ethnic or racial group.⁵ A lynching signaled that personal security—and with it the freedom to work and innovate—was not guaranteed.⁶

Table 1 shows that lynchings peaked for black and white victims in the 1890s.⁷ Most lynchings occurred in the South, and most victims were African American. As the table indicates, the average number of lynchings with African American victims each year varied greatly. Although data on lynching are recorded through 1968, the practice had largely stopped by 1930. Media coverage of lynchings spread awareness of the violence. For much of the nineteenth century, lynchings received local coverage in black- and white-owned newspapers, and nationally through newspapers in major urban areas and publications of the NAACP, other national civil rights organizations, and nascent anti-lynching movements.⁸ International attention grew through newspapers and organizations, including the British Anti-Lynching Committee formed in 1894 to protest the lynchings of southern blacks. Although the direct effect of lynching was likely primarily local, its indirect effect—a growing and general sense of diminishing protection in the courts and among law-enforcement bodies—was national.⁹

Other measures associated with violence are the timing and extent of Jim Crow legislation, which often legitimated acts of violence and created or reflected the social and political environment of the day. A flurry of laws promoting segregation followed the period of Reconstruction. As Table 1 shows, laws advancing segregation were primarily related to education and public facilities.¹⁰ Segregation laws had two direct effects. First, and most important, they were proxies for latent violence. Jim Crow legislation formalized customary practices and allowed few legal safeguards for minorities. Residents of a given state understood that violence would occur if the laws were not obeyed. Litwack (1998) argued that lynch mobs and the courts were the de facto enforcers of Jim Crow laws.¹¹ In explaining “racial cleansings” in which blacks were abruptly driven out of counties both North and South, Jaspin (2007) offered an additional direct effect of segregation laws. He argued that the greater the number of and adherence to Jim Crow laws, the fewer the encounters between African Americans and whites and the greater the degree of anxiety, mistrust, and suspicion between the races, which could lead to spontaneous outbreaks of violence.¹² In this paper, the number of segregation laws was included to capture the taste for and degree of segregation and latent violence across states.

Passage of segregation laws also decreased access to patenting institutions and to social networks and institutions that support invention and innovation.¹³ The offices of patent attorneys (all white at the time) were in “white-only” commercial districts, hindering African American inventors from applying for patents.¹⁴ With little recourse to the courts, African Americans would have found it nearly impossible to fight patent infringement, even if they had been represented by white attorneys.¹⁵ Networking opportunities were also limited. Segregation of public buildings led to “Negro Day” during major scientific fairs or “Negro fairs” that were completely separated from major exhibitions.¹⁶ Unequal access to education, which became increasingly important to patenting over the course of the twentieth century, also likely deterred patenting.

1.2 Patenting activity, 1870–1940

Figure 1 shows that, before the early 1900s, patenting rates among African Americans followed a pattern increasingly similar to that of the larger inventor population, albeit at a much lower level. Like overall patent rates, African American patent rates were procyclical, increasing with economic booms.¹⁷ Black patent activity became countercyclical at the turn of the century. As Fig. 2 suggests, black inventors began responding to incentives or conditions that did not affect other inventors. Specifically, a rise in race-related violence coincided with greater divergence in patenting rates between black and white inventors. I test the validity of this apparent correlation statistically using new data on African American inventors as described below.

Fig. 1

Black and white utility patents, per million, 1870–1940

Fig. 2

Conflict and black inventive activity, 1870–1940. SourceCook (2004), EPO, Tolnay and Beck (1995), Tuskegee Institute (2004), USPTO. Note Patent data in Fig. 1 are presented by grant year and in Fig. 2 by application year

2 Data

Race is not recorded in patent records. Therefore, my first task was to identify African American inventors. I collected data from little-known surveys that Henry E. Baker conducted on behalf of the U.S. Patent Office in 1900 and 1913. He sent surveys to 9,000 of the 12,000 patent attorneys and agents in the United States asking if they had African American clients or if they knew of any African American patentees. Data collected from these surveys constitute approximately 65 % of the data set. The Baker data, however, are only partially helpful to my study. He mistakenly identified the first African American known to receive a patent, and the data end in 1914, 26 years short of the period of interest.¹⁸

To fill the data gap, I matched patent records to census data. While matching nearly two million patents to census records is an onerous task, it is more onerous to distinguish African American from non–African American patentees. First, as is evident from Table 2, African Americans obtained patents outside the South, even though only a small percentage of the African American population lived outside the South. Second, with the exception of a few famous inventors, African American inventors’ names were indistinguishable from those of other American, particularly British-born, inventors.¹⁹ Addresses were often limited to just the city or town of residence, which added further difficulties in identifying individuals. Appendix II describes other approaches, including those exploiting the recent literature related to “black names,” such as in Bertrand and Mullainathan (2004) and Fryer and Levitt (2004).

Table 2 Total and African American patentees, 1870–1940

The best strategy was to identify African Americans among the population of inventors and likely inventors from other sources and to match them to patent records. I accomplished this by collecting names from several sources including modern and historical directories of African American scientists, engineers, and medical doctors; archives, including correspondence of the noted African American historian Carter G. Woodson and the Garrett Morgan Papers; obituaries in local newspapers; published biographies and collections of biographies; programs from the “Negro Building” or “Negro Day” at fairs and exhibitions related to science and invention before 1940; census data; and online company-history searches. A more detailed description of these sources appears in Appendix II. I obtained additional patents of inventors appearing in the Baker data by searching USPTO (2004) and EPO (2004) databases.

The data set I constructed extends from 1870 to 1940 and includes 726 utility patents granted to African Americans during this period. The data comprise the patent number; inventor’s full name, full names of co-inventors, and order of appearance of names of inventors; location of the inventor; title of the patent; dates of application and issue; assignment status; assignee’s name and location; current USPTO patent class and subclass; and NBER-Hall, Jaffe, and Trajtenberg two-digit technological class.²⁰

As Fig. 1 shows, despite the difference in levels, trends in inventive activity by race were roughly similar before the early 1900s.²¹ Until 1930, two of the three major fields in which the two groups patented were the same: manufacturing and transportation. Table 2 reports other data on invention by technological category.

Each patent-holder was issued approximately two patents, on average, which is consistent with the findings of Hall et al. (2001) for the entire population of utility-patent-holders from 1963 to 1999, but it is much lower than the average that Khan and Sokoloff (1993) computed for patentees up to 1846.²² Two-thirds of black patentees have one patent. However, 4 % have four or five patents, and 3 % have ten or more patents. Table 3 contains examples of patented inventions in the data set. This sample reflects the significant variation across technical classes and geography among African American inventors of this era. Although patent activity was evident in all regions of the country, the Midwest and Mid-Atlantic states, including New York and New Jersey, accounted for 64 % of this activity, which mirrors general patterns among white inventors at this time. Consistent with the practice of the day, African American inventors were largely individual inventors, but a number of patent-holders were members of well-known research teams.²³

Table 3 Patented inventions by African Americans, selected, 1870–1940

One problem in the data is the potential undercount during the period when African Americans relied heavily on patent intermediaries. Baker’s task of verifying patentees was complicated by a widespread perception that African American patents might be undervalued if the inventor’s race were revealed.²⁴ Truncation due to undercounting would be difficult to measure and to account for in estimation. Nonetheless, the number of “missing blacks” would have to be large to obtain the magnitude of decline apparent in Fig. 1. Further, prolific inventors entered and exited the data set throughout the period of study. Because inventors with one patent dominate the data set (unlike the U.S. data for this period), several prolific inventors would have had to die or retire simultaneously to account for such a large and sustained decline, but the data do not bear this out. Finally, the population of inventors is heterogeneous and extends beyond those who are highly skilled and in the sciences, particularly in the period before the early twentieth century when specialized skills became more useful. As a result, the data set likely underrepresents inventors with fewer or different skills. Upon inspection of related data, such as Sluby (2004), these potential problems appear to be minor and should not significantly affect results from estimation.

3 Estimation

This section assesses the economic impact of conflict and violence on innovative activity, as measured by patents. Specifically, it explores whether economic activity, namely the level of innovative output, changes in response to changes in hate-related violence, or whether the quality and direction of economic activity change in response to changes in property rights resulting from hate-related violence.

Table 4 reports values of variables typically associated with innovative activity at the beginning of the period to establish a baseline. Table 5 reports the means and standard deviations of variables for blacks and whites jointly and separately.²⁵ Much of the increase in productivity in the mid-nineteenth century occurred in the manufacturing and transportation sectors. According to Margo (1990), African Americans were represented in greater proportions in durable manufacturing and in transportation employment relative to whites, as was the case in agriculture. Illiteracy and school-attendance gaps were large, a fact that may have become more relevant when patenting began to require better education and more specialized skills in the early 1900s. Literacy improved (as measured in the data by illiteracy rates) between 1870 and 1940. In 1870, 68 % more African Americans were illiterate than whites, while in 1940, the gap was only 12 % (Ruggles et al. 2007). In general, patenting activity occurs in regions with relatively more robust economic activity and with significant urban populations.²⁶ Blacks were concentrated in rural areas and the South, which were the least productive areas for patented innovation. As Table 2 shows, however, three-quarters of black patent activity took place outside the South.

Table 4 Baseline characteristics, 1870

Table 5 Descriptive statistics

My analysis employs three different empirical strategies. First, with time-series data, I use state and federal practices promoting segregation and condoning violence as a natural experiment to estimate the effects of changes in hate-related violence on patenting outcomes.²⁷ Second, I account for regional heterogeneity by estimating the effects of increased violence on state-level data for African American inventors. Third, a placebo of white inventors is randomly drawn, and I examine the “counterfactual” effect on patenting among whites.

3.1 Difference-in-differences estimation

Did the difference in patent productivity between blacks and whites change over time? The first part of the estimation strategy uses the implementation of state and federal policy as a natural experiment to estimate the effects of racial conflict. We can use patent data to explore more precisely changes in patenting rates by race following acts of violence that would differentially affect inventors, if not all economic agents. That is, significant changes arise not as a result of violence per se but out of a sense that these hate-related acts could be carried out with impunity.

To capture the direct and indirect effects of hate-related acts, I include two specific years, 1900 and 1921, in which state and federal changes in policy would have increased the indirect effects of violence—that is, years in which policy changes would have signaled that local, rather than state or federal, legal remedies would be final. I include the year 1900 to capture the Plessy v. Ferguson decision, which was implemented with a lag. I include 1921 because this was the year of the Tulsa, Oklahoma, race riot, the largest race riot in American history, and the response to which signaled a major change in government policy.²⁸ This event was considered so grave and alarming that it was the first time the head of the NAACP, the nation’s largest and oldest civil rights organization, appealed directly to and met the President of the United States to request that the federal government intervene.²⁹ However, this appeal to the President was rebuffed, and there was no federal intervention. The Oklahoma Commission to Study the Tulsa Race Riot of 1921 roundly criticized responses to the riot: “Stand back and look at those deeds now. ... In none did government prevent the deed. In none did government punish the deed” (Oklahoma Commission to Study the Tulsa Race Riot of 1921 2001, p. 20). The Tulsa riot of 1921 followed a rash of major (and smaller) race riots throughout the country in 1919.

According to the historical literature, before 1921 potential victims implicitly believed that, if implored, the federal government would act. The response to the Tulsa riot was considered a major policy shift in favor of nonintervention by federal and state governments. Accounts of the Tulsa riot suggest that many at the time believed that government failed at all levels, that this was a turning point in federal policy and national practice related to property-rights protection, and that the country was likely headed toward racial warfare.³⁰ Consistent with historical analysis and the sense of heightened national anxiety concerning racial conflict, a search of the terms “race riot” and “race war” in Google Books Search between 1880 and 1930 shows that these terms peaked in 1921.³¹

To begin statistical inference on the impact of hate-related violence on inventive activity, the basic equation follows Gilfillan (1930), Griliches (1957), Schmookler (1962, 1966), and Sokoloff (1988), all of whom find a relation between economic activity and innovation, and includes positive and negative correlates of innovation—industrial production and unemployment.³² To this model I add the following conflict-related indicators:

$$\begin{aligned} \Delta \hbox {log}(patents_{it})&= \delta _1+\beta _1 \Delta \hbox {log}\left( {lynch_{it}}\right) + \beta _2 \Delta riot_t + \beta _3 \Delta seglaw_t + \delta _2 race_i\nonumber \\&+ \alpha _2 d1921_t + \alpha _3 race *d1921_t +\Delta z_{it} \gamma +\Delta u_{it}, \end{aligned}$$

(1)

where the observation $patents_{it}$ is total utility patents per capita applied for in year $t$ and granted to individuals of race $i$; $lynch_{it}$ is lynchings per capita by race of victim in year $t$; $riot_{t}$ is number of major riots in year $t$; $seglaw_{t}$ is total new state segregation laws passed by year $t$ or total new state and federal segregation laws related to education, housing, and public accommodations passed by year $t$; $z_{it}$ is a vector of controls; and $u_{it}$ is a stochastic error term.³³ The elements of $z_{it}$ are year dummies for peaks and troughs of economic activity, a year dummy for the structural break that occurs in 1900 (“year $\ge$ 1899”), $unem_{t}$ is the Lebergott (1964) unemployment series in year $t$, and $indprod_{t}$ is the Miron–Romer index of industrial production in year $t$.³⁴ Aggregate and race-specific time effects are included in estimation. Because of persistence in the patent series, the basic model is estimated in first differences.³⁵ Table 6 reports the estimated effects from this estimation, along with Driscoll–Kraay standard errors.³⁶

Table 6 Difference-in-differences regressions dependent variable: log patents per capita

The overall effect of the year 1921 is mixed in these regressions, but the interaction for 1921 is negative and significant in the full sample.³⁷ Adjusting for other observables, annual patenting by African Americans was lower by a factor of 2.2 on average than for whites because of events in 1921. Coupled with the introduction of federal anti-lynching legislation in 1921, there was a heightened sense among African Americans that personal security and other property-rights protections were being markedly eroded, and this was the basis upon which Secretary Johnson of the NAACP met President Harding. This suggests, like the Plessy v. Ferguson, U.S. Supreme Court. (1896) decision, that federal action or inaction with respect to hate-related violence may generate significant declines in security and economic activity, as measured by inventive activity.

From the pooled regression, I tested for differences between coefficients on lynching and riots for blacks and whites. The difference in coefficients is significant at the 0.01 level.³⁸ These differences are evident in results from estimation of the baseline regression in the black and white subsamples reported in Table 6. For whites, major riots are correlated with a decline in patenting of 2 % per year, and there is no correlation between the year 1921 and patenting activity.³⁹ For blacks, the estimated effect of the policy shift in 1921 is negative and significant. A 1 % increase in the growth rate of lynchings per capita is associated with 0.9 % lower growth rate in black patent activity, and major riots are associated with 13–14 % lower rate of growth in black patent activity.

Aggregate data in this instance may bias the coefficients on segregation laws toward zero. While laws instituting racial segregation are not significant, the structural break in the black patent series, foreshadowed by Fig. 1, may suggest otherwise. Implemented with a lag in nonsouthern states, Plessy v. Ferguson allowed states to adopt rules that would disrupt previously integrated economic ties and activities.⁴⁰ Nonsouthern states, including Illinois, Ohio, New Jersey, and New York where much of the inventive activity among blacks was taking place, adopted 145 new Jim Crow laws between 1896 and 1940, more than two and a half times the number passed by these states between 1870 and 1895. As I mentioned above, increasing formal race-based restrictions in the workplace and in everyday life may have limited blacks’ access to patent agents and attorneys and to patent-related resources such as patent journals at public libraries. Therefore, inventors’ ability to collaborate, register patents, conduct patent searches, and defend their patents against infringement would have become a binding constraint on patenting activity.⁴¹ Other economic ties were broken, such as property ownership and whites patronizing black-owned businesses.⁴² This evidence supports the view that the aforementioned congressional apologies reflect the intuition of black inventors, and other economic agents of the time, that the federal government had been tacitly condoning race-related violence or actively promoting blockage of federal anti-lynching legislation and erosion of legal protection generally.⁴³ The implication of this finding in the patent data is that official legitimation of hate-related acts can permit their proliferation and produce long-term declines in inventive and economic activity.

When variation across states and technological category are exploited below, estimated coefficients related to Jim Crow laws not only have the predicted sign, they are significant.

3.2 State regressions

Do these results vary geographically? Variation in the institutions and opportunities related to patent activity and in patent activity itself, in the rule of law, and in violence was significant across regions between 1882 and 1940. Without state controls, parameter estimates may be biased, picking up the influence of omitted region variables that are not explicitly included. Another advantage of more refined data on state patents is that they may be evaluated along technological, geographic, and economic dimensions. Therefore, the second prong of the empirical strategy is to estimate a model using panel data containing state-level characteristics of patents (and inventors) that therefore allow me to account better for observed heterogeneity than if I were to rely on the aggregate data. I organize patent data by state-year and fit them to a random-effects model.⁴⁴

In this model, $patents_{st}$ is the total number of utility patents granted to African American inventors in state $s$ in year $t$. Applying random effects to Eq. (1) implies:

$$\begin{aligned} patents_{st}&= \beta _1 lynch_{st} +\beta _2 riot_{st} +\beta _3 seglaw_{st} + \beta _4 firms_{st}\nonumber \\&+\beta _5 illit_{st} + \beta _6 particip_{st} + z_{st} \gamma +\varepsilon _\mathrm{s} + u_{st}, \end{aligned}$$

(2)

where $\varepsilon _{s}$ is the state-specific error component of the composite error term.

In Eq. (2), lynchings are per 100,000 residents in state $s$ in year $t$. Additional covariates of patenting can be included in the state regressions: $illit_{st}$ is the illiteracy rate in state $s$ in year $t$, and $particip_{st}$ is an average over state $s$ and year $t$ of the percentage of blacks represented in the industry with which patents are associated from Margo (1990). Instead of industrial production and unemployment, the number of firms per capita in state $s$ in year $t$ taken from the Census of Manufactures (1883, 1895, 1933, 1942) will approximate the level of economic activity in each state. Standard errors reported are clustered by state and year.⁴⁵ In this model, $z_{st}$ contains a dummy for the region of the country in which state $s$ is located, share of the total black population in the United States residing in state $s$ in year $t$, year dummies for peaks and troughs of economic activity, and share of patents granted to prolific or “great” inventors in state $s$ in year $t$.⁴⁶

The findings employing state-level data and a random-effects specification reported in Table 8 generally support those estimated by pooled OLS models for lynchings.⁴⁷ The correlation remains significant between lynchings and patent activity in the black samples, but the size of the estimated coefficient is smaller. When controlling for state effects, the magnitude, direction, and significance of the estimated coefficients on riots and Jim Crow laws change. The riot estimates become larger, negative, and more significant. This is intuitive, given that state data allow for more precise measurement of the effects of subnational events. One additional riot in a given state in a given year would diminish the state total by an average of nearly half a patent or by 17 patents in a given year for all states. Being in a relatively more segregated state depresses the expected number of patents, but this relation is not significant. Lynchings and riots are associated with an average decline of $-0.4$ per state per year or 1,132 patents between 1882 and 1940, which is roughly equivalent to total patents granted in 1853 or 1854 in the United States.

These patterns precede large-scale black migration. To account for the rapid increase in black migration from the South that begins after 1917, the sample is split accordingly and reported in columns 5 (1882–1917) and 6 (1918–1940) of Table 7. The estimated effect of lynchings is larger in the second period, and the estimated effects of riots and segregation laws are smaller and not significantly different from zero in the post-1917 period.

Table 7 State regressions dependent variable: patents per state per year

Another question is, does hate-related violence covary with economic activity uniformly? These findings also reveal that the effect differs across economic, technological, and regional categories. Among the most economically important inventions at the time were patents assigned at issue, an approximate indication of early commercial viability and, to a lesser extent, mechanical and electrical patents.⁴⁸ As is reported in Table 8, overtly violent acts are negatively and significantly correlated with lower patent activity for assigned patents. For mechanical and electrical patents, the presence of latent violence, as proxied by segregation laws, is negative and depresses mechanical patents by 0.2 per state year, or 579 patents during the period.⁴⁹ Again, this finding related to segregation is intuitive, given the ease of mobility required for inventors to be productive.⁵⁰

Table 8 State regressions—assigned, technological category, region dependent variable: patents per state per year

Further, the results suggest that violence-related factors are particularly important in the South. Similar to the case of mechanical and electrical patents, lynching and riots are negatively but not significantly correlated, but the threat of violence is more negatively and significantly correlated with patent outcomes in the South than in these other groups. This result is not surprising given that the threat of violence that made Jim Crow laws credible likely forged near-convergence between violent acts and latent violent acts, given the persistence and prevalence of hate-related violence in the South during this period.

3.3 Estimating the effect on productivity

What would productivity have been absent hate-related violence? I use state-level data to execute this counterfactual exercise. By collecting a random sample of patents of white inventors similar to those of African American inventors, I constructed a “placebo” study and compared the productivity of inventors subject to hate-related violence with that of inventors not (or less) subject to hate-related violence. Specifically, I draw a random sample of 714 patents by application year of patents by African American inventors from the USPTO database using Google Patents.⁵¹ The inventors are similar in most respects, such as field of invention.

For this estimation, I organize and pool patent counts by state, and I fit a negative binomial model to the count data.⁵² In this model, ${ TP}_{s}$ is the number of utility patents granted to individuals in state $s$ which is the count variable. I assume that the number of utility patents that can occur follows a Poisson distribution. Moreover, the Poisson parameter is allowed to vary across states and is assumed to follow a gamma distribution. The number of utility patents granted to individuals in state $s$ follows a negative binomial distribution:

$$\begin{aligned} \Pr (\textit{TP}_s |L_s ) = \frac{\Gamma (\textit{TP}_s +\alpha ^{-1})}{\textit{TP}_s !\Gamma (\alpha ^{-1})}\left( {\frac{\alpha ^{-1}}{\alpha ^{-1}+\mu _s }} \right) ^{\alpha ^{-1}}\left( {\frac{\mu ^{-1}}{\alpha ^{-1}+\mu _s }} \right) ^{\textit{TP}_s },\quad \textit{TP}_s =0,1,2,\ldots ,49,\nonumber \\ \end{aligned}$$

(3)

where $\mu _{s} = \hbox {exp}(\mathbf{L}_{s}{\varvec{\lambda }}), \Gamma$ is a gamma function, $\alpha$ is degree of dispersion, and $L_{s}$ is the $(K \times S)$ matrix of conflict-related and other explanatory variables as in earlier state regressions.⁵³ The model in Eq. (3) is estimated in both samples. Table 9 reports results from the placebo-study regressions.

Table 9 State regressions, African American and white control group inventors dependent variable: patents per state

The marginal effect of lynchings is negative for both groups but both negative and significant in the African American regressions. Absent race-related violence, the most significant marginal effect is that derived from economic activity, as theory would predict. In sum, the placebo study suggests that differences between African American and white inventors are largely explained by hate-related violence rather than other factors. Given that estimated effects of black lynchings are biased toward zero because of underreporting, the evidence likely represents a lower bound on the size and significance of the relation between violence and economic activity.⁵⁴

The empirical analysis shows the threat of violence or actual violence likely altered incentives and outcomes for black inventors. The story of Garrett Morgan, the inventor of the modern traffic light (1912) and gas mask (1914), and inductee into the Inventors Hall of Fame, offers an example of these transactions costs. In 1904, Springfield, Ohio, was one of the first cities north of the Mason–Dixon Line to record a lynching, which was the cause of a major riot (see Table 1). As lynching began to spread across northern states and personal insecurity increased, black groups formed to protect their neighborhoods, families, and property. Morgan joined one of these societies in Cleveland and purchased a gun (Fryer and Levitt 2007). The papers suggest that segregation laws and customs constrained Morgan’s market opportunities (Cook 2012). Rather than he himself promoting the masks, advertisements for his gas (and fire-safety) helmets depicted white or racially ambiguous figures wearing the helmets. Further hiding his identity as a black inventor, when demonstrating his helmet across the country, he posed as a Native American chief, the “real” inventor of the mask, and claimed that Garrett Morgan was his assistant. After fire chiefs in southern cities learned his true identity, orders for his mask in the South fell precipitously.⁵⁵

4 Alternative hypotheses and robustness

To check the robustness of the above results, I test whether participation in certain industries (being in the right place at the right time) and literacy explain observed economic outcomes.

4.1 Alternative hypothesis—“Right Place, Right Time”

At the time of the Second Industrial Revolution, invention-intensive firms such as AT&T and General Electric were increasingly internalizing their research activities.⁵⁶ Simultaneously, Margo (1990) finds that employment became more racially segregated between 1900 and 1950, particularly among skilled blue-collar workers and in manufacturing. Although his results are focused on the South, the evidence suggests that outside the South the labor market, through union rules, state legislation, federal legislation, or custom, was becoming more racially divided.

From this change in industrial organization, there are at least two outcomes of interest to this study. First, the move by firms to incorporate patentees into newly established research departments may have lifted the veil on anonymity, thus raising uncertainty and diminishing the incentive to patent for African Americans, who had worked through intermediaries in the past.⁵⁷ Second, even if we assume that black and white inventors had roughly equal access to scientific and invention-related resources, including apprenticeships, before this change, the gap between insider-inventors’ and outsider-inventors’ access to resources should have diverged significantly, particularly if externalities from industrial research groups are captured by the firm.

To test the employment hypothesis, I control for the share of employment in patent-intensive industries using (Margo (1990)) industry-participation variable. I matched this variable to the technological category of the patent in estimation. Industry participation is significantly different from zero in the regressions using mechanical patents. In general, the effect of black representation by industry on innovation is ambiguous.

4.2 Alternative hypotheses—literacy

Did the increasing requirement of specialized skills for patent activity at the end of the nineteenth and start of the twentieth centuries affect patent outcomes? If so, differences in literacy, education, and training might explain the “patent gap.” I estimate that 79 % of blacks were illiterate in 1870. High illiteracy rates are related to low levels of schooling in the post-Civil War era, as is consistent with the findings of Card and Krueger (1992), who show a high but declining racial gap in school quality from 1915 and those of Collins and Margo (2003), who find significant but narrowing racial differences in literacy, school attendance, spending per pupil, and other education variables. If patenting activity were increasingly a function of tertiary education in the sciences, blacks would have been at a disadvantage, because it was not until the 1920s that blacks began earning PhDs in the sciences in earnest. Consistent with the historical literature on patenting, I find that illiteracy is negatively correlated with patent activity in more specialized fields—in the electrical regressions—and is increasingly negatively correlated with patenting over time (Table 8). Nonetheless, the effect of illiteracy is ambiguous across the category regressions. In sum, the evidence supports neither industry participation nor education as a significant determinant of patenting activity across models and subsamples.

Recent studies use micro evidence from experiments to find a relationship between violence and economic outcomes of interest and propose alternative mechanisms. Using experimental and administrative data from Afghanistan, Callen et al. (2012) found a relationship between violent trauma and risk preferences, a finding that could also apply to the United States at the turn of the twentieth century. Others have found similar relationships internationally (Becchetti et al. 2011; Voors et al. 2012; Bauer et al. 2011; Gilligan et al. 2011; Bertrand and Mullainathan 2004; Callen et al. 2012; Blattman 2009; Rohner et al. 2013). The direction of causality in these studies is also from violence to economic activity.

Another plausible mechanism arises from displacement. Inventive activity, which may require periods of concentrated, uninterrupted work and thought, would likely have been disrupted and fallen as a result of displacement. Likewise, social networks would be disrupted. Cook (2011), for example, finds that inventors, like researchers and other economic agents, were displaced by riots and segregation laws, which would lead to ruptured social networks. Likewise, using county-level census and other geographic data, Tolnay et al. (1996), Loewen (2005), and Jaspin (2007) identify displacement due to lynching and other violent acts. Recent work by Shafir and Mullainathan (2012) also points to an alternative mechanism— scarcity. Using data from a randomized experiment involving sugar-cane harvesters in India, Shafir and Mullainathan (2012) find that in periods of relative scarcity, or pre-harvest periods, harvesters perform worse on IQ and attention tests and have shorter time horizons than in post-harvest periods. Similarly, Wang et al. (2010) find that those making more difficult decisions involving more conflicting tradeoffs—personal security or livelihood in this instance—make decisions with worse outcomes than those making easier decisions, e.g., those without extraordinary threats to personal security. Both of these situations could apply to black innovators at the turn of the century.

4.3 Reverse causality

Can we entirely rule out causality that runs in the other direction? It should be acknowledged that earlier studies have found support for the direction of causality going from economic growth to violence, such as Miguel et al. (2004). Yet, the evidence from recent experiments is consistent with a direction of causality going from violence to economic outcomes. In the analysis in this paper, four empirical reasons suggest that violence causes changes in economic activity. First, the best evidence on the economic factors related to lynching is the empirical relation between cotton prices and lynching, an association that breaks down after 1905 and is valid for only a fraction of the period of interest. Specifically, several studies have looked at the causal relation between labor-market competition between blacks and whites and lynchings; Raper (1933), Hovland and Sears (1940), and Tolnay and Beck (1995) find an inverse relation between cotton prices, and therefore competition for jobs in agriculture, and black lynchings. This relation breaks down in the early twentieth century. Darity and Price (2003) examine the relation between racial stigma, or status as a former slave, and lynching activity. Their findings suggest that racial stigma is a relatively less important determinant of lynching activity than labor-market competition. Evidence on the determinants of lynchings after 1905 is inconclusive in this study as well.⁵⁸ Further, these causal factors are unrelated to traditional determinants of patenting activity.

Second, the violent or violence-related acts are not confined to economically depressed regions. To recall, 60 % of riots between 1900 and 1940 did not occur in the South—that is, they occurred in areas that were relatively less economically depressed than the South.

Third, in a systematic review of recorded motives for riots and lynchings, neither type of violent act had a direct economic motive. Contemporaneous newspaper reports (e.g., Ginzburg 1962/1988), case studies (e.g., Cecelski and Tyson 1998 on the Wilmington riots and Crowe (1968, 1969) on the Atlanta riot), and official government investigations, such as the Final Report of the Oklahoma Commission to Study the Tulsa Race Riot of 1921 (2001), rarely cite economic motives for riots. Of the 27 major riots between 1870 and 1940, only two—in New Orleans in 1895 and in East St. Louis in 1917—were documented in the literature as having an economic motive. In the HAL lynching data set that includes offenses ostensibly leading to lynching, of 2,806 victims of all races listed, only 98 were lynched for offenses related to possible commercial factors, including two strikebreakers, two men suspected of being foreign workers, one brothel owner, one moonshine producer, and one horse thief (HAL 2004).⁵⁹

Finally, I execute a Granger causality test on the riot variable, which is the best-measured violence variable. In the black sample I find that major riots Granger-cause patent activity, whereas I cannot reject the null hypothesis that patent activity has no useful predictive content with respect to riots at the 10 % level of significance. Simultaneously, in the white sample I cannot reject the null hypotheses that patent activity has no useful predictive content with respect to major riots and that major riots have no useful predictive content with respect to patent activity. Results from the Granger causality test support our intuition that violent acts can predict patent outcomes and not vice versa.

To be sure, the correlation between the variables proxying for violence and the error term will not be zero. For example, the magnitude and full extent of informal segregation and deep psychological factors, such as degree of racial mistrust, are difficult to measure and cannot be included in estimation. Data on minor riots, which may be correlated with major riots and lynchings (and Jim Crow laws), are neither systematically reported nor available and also cannot be included in the regressions. However, the data, historical literature, and empirical tests suggest that the direction of causation from violence to economic activity is the one more consistent with the evidence available.

Finally, the quantitative measure of legal segregation—number of new segregation laws passed in a given year—will not fully capture the depth and scope of informal segregation, such as the extent of discriminatory informal customs and practices, the quality of legal enforcement, and laws overturned after 4 years. It is reasonable to assume that informal Jim Crow customs and practices for which there was significant political consensus became embedded in law; however, many such practices did not rise to this level of agreement but remained embedded in society. For example, Margo (1990) finds that southern apprenticeship and employment opportunities were considerably restricted by discrimination, though not necessarily by formal laws on the books, prior to 1950. As I mentioned above, customary segregation in the North and South led Garrett Morgan to dress as a Native American or hire a white person to demonstrate his gas mask to white audiences. The segregation variable can measure some, but not all, informal segregation.

4.4 A final observation

Are these results unique to patents? Some readers may suggest that the data and inventors are unique and the results difficult to generalize. However, the data are not special. The data I constructed on the establishment of black newspapers, for example, are quite similar to the patent data.⁶⁰ Newspapers have some of the same features of patents, including reliance on the protection of property rights. However, they differ in the sense that newspaper publication is an obviously public act, whereas patenting is not. Newspapers owned by or sympathetic to African Americans may attract attention from mobs and individuals, as would any retail firm or independent communications outlet. As Fig. 3 shows, these data follow largely the same pattern as the patent data. The series increases significantly up to 1899 and falls to a permanently lower rate of increase after 1900. This implies that the findings from the current study may be more generally applicable to productive activity.

Fig. 3

African American newspapers and patents, 1870–1940. Source Author’s calculations; see data appendix

5 Conclusion and future research

This research contributes to the literature on the economic effects of conflict and political instability. By using an historical experiment, I examine the effects of hate-related violence on innovation, and, by extension, real economic activity and living standards. This paper introduces and analyzes a new data set on patents obtained by African Americans between 1870 and 1940. The evidence from time-series and cross-section estimation suggests that hate-related violence, the reporting of which began nationally during this period, was by itself important. More important was the sense among African American inventors and other economic agents that hate-related violence would likely not be adjudicated and that the rule of law, typically through federal government intervention, would likely not prevail.

The increase in scope and intensity of hate-related violence in the late nineteenth and early twentieth centuries depressed patent activity among African Americans by 1 % per year, or the equivalent of a year’s worth of African American patent activity. The gap between white and black patents per million was maximized in years of heightened violence, such as 1889 (536.7) and 1909 (547.1), and minimized in years of diminished violence, such as 1937 (341.5) and 1940 (357.1). In general, by 1936, the effect of conflict indicators on patenting by blacks falls as conflict itself wanes. This violence would have implied a decline of roughly 40 % in patenting and greater volatility in output among most U.S. inventors during that period. The most valuable patents—assigned, electrical, and mechanical—were sensitive to acts of hate-related violence and to laws promoting racial segregation. I tested alternative theories against my main hypothesis that hate-related violence reduced patent activity, but I find mixed or no support for these theories. Using patents as an example, the results suggest that changes in personal security and the rule of law can shift the scale, quality, and direction of technological progress and economic activity. This evidence is consistent with existing research on conflict and economic outcomes, particularly recent research from randomized experiments.

The import of this data set goes beyond patenting outcomes. A comparison to newspapers founded by African Americans implies that my results may reflect more general effects on economic activity. These findings would be particularly relevant for countries that are experiencing violence and ethnic conflict and are characterized by weak protection of property rights, but aspiring to catch up to rich countries in economic growth and development.

It is important to recognize the limitations of my argument and the data. Given data constraints, the decline in patenting activity with increased violence can be attributed to several factors. These include the direct effect of diminished personal security owing to riots, lynchings, and segregation laws as well as the indirect effect of mistrust of institutions that results from these acts. In addition, the direct effect of declines in property values owing to the lack of the rule of law (diminishing resources to finance innovation) can also influence patenting activity. The direct effect of informal (or formal but not legislated) segregation could also have an effect, particularly segregation that placed physical constraints on movement. This could serve to limit protection of intellectual property because of limited access to patent agents, attorneys, and information. More detailed data on individual characteristics of all inventors, such as property ownership, and on informal or more localized segregation would be required to disentangle these effects.

Appendices

Appendix I: Data

1.1 Data sources

Data on patents obtained by African Americans between 1870 and 1940 come from the author’s data set, which extends the Baker (1913/1969, 1917, 1921) data set. Total patent data are from the U.S. Patent and Trademark Office database. Patents held by white inventors are derived by subtracting patents obtained by black inventors from the total. Data on lynchings by race of victim per year per state in southern states are from Beck and Tolnay (1995). Data on lynchings in other states are from the Tuskegee Institute data set and are averages for the period 1882 to 1930; annual data are not available by state. Data for blacks and whites after 1930 are from the Tuskegee data set. I find that the Tuskegee data underestimate lynchings among blacks and whites in nonsouthern states. Data from Ginzburg (1988) were added for blacks in nonsouthern states. Data on major riots and segregation laws are from the U.S. House of Representatives Select Committee (1979), Library of Congress (2004), Higgs (2007), “The History of Jim Crow” (www.jimcrowhistory.org), and the Final Report of the Oklahoma Commission to Study the Tulsa Race Riot of 1921 (2001). Aggregate and state illiteracy data are extracted from Integrated Public Use Microdata Series (2004), approximately 50,000 individuals over 10 years old, and from the University of Virginia Library (2004), full sample, individuals over 10 years old. Data on aggregate illiteracy rates in 1890 are taken from Collins and Margo (2003). These data are derived from the population of 10- to 69-year-olds using the full count. Population data are extracted from Gibson and Jung (2002). Regions do not conform exactly to census divisions: Delaware and Maryland are considered Mid-Atlantic states in this paper and are considered South Atlantic states by the U.S. Census Bureau. Data on black-owned banks are from Ammons (1996). Industry and occupation data are from Margo (1990). The industry-participation variable is available only for 1910 and 1940 and is only for the South. Patents obtained up to and including 1900 are assigned industry-segregation values for 1900, and patents obtained after 1900 are assigned values for 1910. Technological categories are taken from NBER-Hall et al. (2001). Economic peak and trough data are from the NBER Business Cycle Dating Committee (2007). As many years of abnormal economic activity are controlled for in estimation as possible. Data on unemployment rates are taken from Lebergott (1964) and are available from 1890 to 1940. Data on industrial production are taken from the Miron and Romer (1990) aggregate index of industrial production, which has 13 components. Data on African American newspapers founded in a given year were collected from the University of Georgia (2007), Harvard University (2007), and the Library of Congress (2007). Firm data for each state are collected from U.S. Census Bureau’s Census of Manufactures (1883; 1895; 1933; 1942).

1.2 Variable definitions

Variable	Definition
Lynchings	Lynchings per million in a given year
Riots	Major riot in a given year
Segregation laws	Laws promoting segregation between races passed in a given year and not overturned within 3 years
Newspaper	Newspaper established for or by African Americans in a given year
Illiteracy rate	Person can neither read nor write; over 14 male population, by race
Bank	African American banks founded in state by a given year
Great inventor	Prolific inventor as defined in Cook (2011)
Population, south	Proportion of U.S. population living in the South, by race
Industry participation rate	Proportion of employment in given industry, southern blacks only, 1900
Industrial production	Miron–Romer index of industrial production
Unemployment	Annual national rate

Appendix II: Identifying African American Inventors in Patent Data

As I mentioned, it is very difficult to identify the race of a patentee because patent records did not record it, with only one exception since 1790. The first systematic attempt to identify African American patentees was an effort by the Patent Office, which undertook surveys in 1900 and 1913. The objective of the surveys was to locate African American patentees whose achievements would be featured in the 1900 Paris World’s Fair and to commemorate scientific achievements by African Americans in the 50 years following the end of the Civil War. Directed by one of the lead examiners, Henry E. Baker, surveys were sent to 9,000 of the approximately 12,000 patent attorneys and agents. Responses to the survey were collected and analyzed by Baker and published in several formats (Baker 1913/1969, 1917, 1921). A subset of the original responses were donated to Carter G. Woodson, a noted historian, and, in turn, donated by him to the Library of Congress. The Baker data extend from 1834 to 1917. The investigation in this paper required that the data be extended to 1940. A first strategy to extend the data set was to include patents obtained in 1913 and beyond by inventors already in the data set. These data were collected using the European Patent Office (EPO) search engine, which is searchable by name from 1920. Google Patent Search, which can also conduct historical searches, became available after 2004, when these data were originally collected. Google Patent Search misses some historical patents, and the EPO search is more reliable. One strategy for identifying additional black inventors would be to match patentees from USPTO data to Census data. This method should work for inventors who lived and patented in the same place. However, this procedure fails, because African Americans during this period did not patent where most African Americans live, as Table 2 shows. Before 1940, most African American inventors obtained patents in northern states rather than in southern states. Unlike today, specific addresses were not reported by the Patent Office, just the city or town in which the inventor resided or from which he or she applied for a patent. It is difficult to find a unique first- and last-name match using census data, because of the proximity of first and last names of African American inventors to those of other inventors, especially inventors of British origin. Eight patentees were identified as African American using this method. Only with significant additional biographical data does this method work, and these data are available for few inventors in the data set. And if additional biographical data were introduced, the selection problem would be of greater concern, because biographical information is available for only the most famous and prolific inventors. Another strategy would be to match common names given to African Americans to patent data. A three-pronged strategy in the spirit of Fryer and Levitt (2004) and Bertrand and Mullainathan (2004) was executed but was unsuccessful in identifying black patentees. This mechanism is described below. A second-best method would be to match known black inventors to names in the patent data. This method was significantly more successful in producing matches. The second method and its limitations are described in the text. An index of black names for the period 1870 to 1940 was constructed from census data in two ways. The first strategy answered the question: conditional on being black, which names are most likely to be observed? Random samples of black (“Negro”), “mulatto”, and “colored” heads of households from the 1870, 1900, and 1920 censuses were drawn for the District of Columbia, Georgia, Michigan, and New York. From these samples, frequencies were calculated for first and last names separately. There were 14 first names and 11 last names that appeared more frequently than the median frequency and were included in the index. The second strategy answered the question: conditional on observing a certain name, what is the likelihood that the person is black? First and last names of blacks (“Negroes”) and whites were extracted from the 5% IPUMS sample of the 1870 census. Unlike the above samples, names were not restricted to heads of households. From these samples, frequencies were calculated for first and last names separately and by race for names occurring at least 80 times. Among blacks, there were 27 first names and 20 last names that appeared more frequently than the median frequency for whites or were a larger share of the total names than the black share of the total population and were included in the index. A third approach was an extension of the second approach and answered the question: conditional on having a name widely adopted by African Americans following the end of slavery, what is the probability that the person is black? This strategy was intended to take advantage of a well-known practice among African Americans of adopting the first and last names of presidents, e.g., George Washington, or famous people in the black community, e.g., Booker T. Washington, as first and middle names. The entire 1900 census was used and also was not restricted to heads of households. These approaches yielded largely similar results from which I constructed an index of “black names.” Results were nearly identical with respect to surnames. Yet the composite index was unable to predict matches in the 1880 census sample of the 690 individuals identifying their occupation as “inventor.” I was able to predict a small number of black inventors but, with the exception of George Washington Carver and George Washington Murray, not ones that could be matched to a patent. The index significantly underpredicted matches to black inventors and overpredicted matches to white inventors in New England, particularly those born in England, as was the case with the first census-based approach. Additional location and biographical data would have been required to obtain unique first- and last-name matches. In general, these methods are more suitable for the current rather than historical period. This highlights a problem associated with occupation identification and reporting among inventors. Many identify themselves as machinists or artisans or engineers rather than inventors, irrespective of race. Thomas Edison, among other “great inventors” who are alive and active as inventors, does not appear in the 1880 sample. The final strategy to extend the Baker data set was to construct a broad-based data set of African American inventors, i.e., potential patentees, and to match the resulting data to patent data. Among the historical and contemporary sources used to create a pool of potential patentees were searches of historical newspapers, including obituaries, e.g., from the Ohio Historical Society Newspaper online database and newspaperarchive.com; correspondence from Carter G. Woodson, Henry E. Baker, and patent survey participants (Library of Congress); the Garrett Morgan Papers; historical and contemporary directories of African American medical doctors, scientists, and engineers, e.g., academic journals, including the Journal of Economic History and the Journal of Negro History; historical and contemporary biographies of African American inventors and general biographies, e.g., Great Negroes Past and Present; and programs of exhibitors in the African American sections or exhibitions of historical fairs, including the “Exhibit of American Negroes” at the 1900 Paris World’s Fair, the 1904 “Great Negro Fair” in Raleigh, North Carolina, and the 1933 Chicago World’s Fair “Negro Day.” Newspaper and obituary searches and programs of exhibitions allowed the identification of lesser known inventors. A complete list of sources appears in a companion paper. Not all inventors and others in the pool of potential patentees were matched to patent records and were dropped from the data set. Others were dropped if there was not a unique first- and last-name match, e.g., James Young in the patent data. Ultimately, while second best, this process provides a more systematic and less ad hoc means of recovering black patentees to extend the data set. This summary is adapted from Cook (2011).

Appendix III: Estimating the Effect on Productivity Using Black Parameters

Let us consider a second counterfactual exercise. How much lower would inventive activity in the U.S. have been if all inventors operated under violence-related conditions? We can address this question by taking parameter estimates from estimation of equation 3 in the black subsample and using them in the white subsample. As can be seen in Figure 4, patent output over the period 1882 to 1940 would have been nearly 1% per year, or 40% over the period, lower and would have displayed significantly more volatility for white inventors, who constitute the overwhelming majority of inventors at that time. Like Abadie and Gardeazabal (2003), I find that volatility seems to increase in the presence of greater violence. Of course, the comparison using black parameters with the white sample should be interpreted with caution, because it not only reflects the evolution of violence but also pre-violence differences in determinants related to patent or economic activity. There is an imperfect mapping between technological progress and patent activity, and there are other factors that would change in the white subsample, e.g., illiteracy rates. Nonetheless, the results are suggestive that the rate of technical change in the U.S. may have been substantially lower in the absence of the rule of law affecting both races. Further, concomitant improvements in living standards may have increased much more slowly.

Fig. 4

Predicted White Patent Activity Using African American Estimates