In 2022, the economist Lisa Cook was appointed to the Board of Governors at the Federal Reserve. Paul Romer and Claudia Sahm supported her nomination, citing her article “Violence and Economic Activity: Evidence from African American Patents, 1870 to 1940”, which claimed that racial violence was negatively correlated with patents by Black inventors. This article was the subject of podcast episodes on EconTalk and NPR, and continues to be cited in academic research. To get so much publicity, we would expect the findings in this paper to be robust. 

Cook argues that racial violence reduces patenting by undermining Black inventors’ trust in property rights. The question is not whether that common-sense claim is true. Most likely, it is. The question is whether Cook’s article provides sound evidence to support it. I published a peer-reviewed article “Can We Detect the Effects of Racial Violence on Patenting?” demonstrating that Cook’s article does not provide the required level of evidence. 

On average, it takes 1.4 years for a patent application to be granted. Given this, how would we expect racial violence to affect patenting? Consider the Tulsa race riot of 1921. We should see in 1921 a drop in patent applications, but no decrease in grants, since the corresponding applications were made the year before in 1920. Instead, fewer applications in 1921 means that grants should fall in 1922.  

Since Cook’s data includes measures of patents in both the year of application and year of granting, I can perform this more nuanced test of her theory. None of the predictions holds up: there is no contemporaneous effect of racial violence on patent applications, and no lagged effect on grants. This simple test, using the 1.4-year delay between application and grant, alone undoes Cook’s article. 

But the problems do not stop there. 

One serious issue is that Cook’s results do not hold with a more complete patent variable. Cook’s main results use annual nationwide patent counts, showing that years with more racial violence have fewer patents. But Cook also has a state-level dataset with data on patents by state and year; we should be able to aggregate this data to the nationwide level (adding up patents across states) and get the same results as in Cook’s main analysis. But when I do so, the correlation with racial violence disappears. This is especially puzzling because there are more total patents in the state-level dataset (702) than in the original annual nationwide data (672).  We would expect the correlation to become stronger when including more data, but instead it becomes weaker. 

Moreover, the state-level results are based on a dataset where most observations are missing. Cook’s state-level data uses only 13 percent of all possible state-year observations. For example, Cook reports a negative effect of riots on patenting, but there are only five riots in the state-level data (compared to 35 in the time-series data). This low level of data coverage makes any conclusions unreliable. 

One might say that Cook’s article is arcane academic research, so why make a fuss about it? We should be concerned about how people are being elevated through the ranks of academia and then on into the halls of government. Solid research shows solid character and judgment. Research that is quite otherwise shows otherwise.