Your Donation Will Be Doubled. Support Today!

Inertia and the Death Penalty

Most of us go to the doctor regularly, or at least use the Internet to identify health information of questionable medical value. Either way, we have heard some variation on the phrase “one of the very best predictors of [medical event X] is a prior instance of [X].” One of the very best predictors of whether you’ll […]

Baumgartner FR, Box-Steffensmeier JM, Campbell BW (2018) Event dependence in U.S. executions.

Most of us go to the doctor regularly, or at least use the Internet to identify health information of questionable medical value. Either way, we have heard some variation on the phrase “one of the very best predictors of [medical event X] is a prior instance of [X].” One of the very best predictors of whether you’ll have a stroke later is whether you’ve had a stroke before. As it turns out, that same statistical relationship describes another, ahem, morbidity risk: executions. In a newly published article, a group led by University of North Carolina political scientist Frank Baumgartner shows that there is extraordinary event dependence in American execution practice. Stated a little more accessibly, the Baumgartner data shows that one of the best predictors of whether a county will execute someone in the future is how many the county killed in the past.

This finding matters, a lot; someone should be capitally punished because their crime and moral blameworthiness merits death, not because a particular county happens to have acquired professional expertise in killing people. In this very forum I wrote a short piece, based on my longer law review article, describing the concentration of American execution activity in a few “outlier” counties — counties that maintain exorbitant levels of capital punishment activity. I presented a theory of “local muscle memory,” positing that the small cohort of outlier counties kill because they’ve gotten used to killing, and the rest largely abstain because they haven’t had the practice. And I argued that the already-concentrated map of American capital punishment activity was still concentrating even more, reflecting a crude bureaucratic path dependence. Institutional stakeholders in a particular locality must invest considerable time and capital to develop the skill — the muscle memory — to produce death sentences and then convert them into executions. As a result, bureaucratic habituation was causing counties to separate into two groups: a small group of outlier counties that maintained elevated capital punishment activity, and a much larger group of abstainers. In documenting the increasing concentration of American execution practice, I relied on some of Professor Baumgartner’s data, and some of his prior work. Nonetheless, because I am not a statistician and because empirical testing of the hypothesis was somewhat beyond the scope of the project, my muscle-memory thesis did not include regressions and other advanced statistical analysis that would determine the fit between the theory and available data. Professor Baumgartner and his collaborators have largely filled that gap.

They worked from a sample of 1,422 executions spanning 1977 to 2014, from 474 counties in 34 states. I’ll leave out most of the major modeling details, but suffice it to say that the analysis was capable of controlling for four phenomena that had typically been assumed to drive execution activity: (1) county-level homicide activity; (2) racial demography; (3) poverty rates; and (4) county-level population. The fourth control is deceptively important insofar as large population centers experience more homicides and have more professionalized prosecution offices, and hence tend to produce more death sentences. (Of course, the study removed states that didn’t use the death penalty from the analysis.)

The result of the Baumgartner analysis is pretty astonishing, both because of its bottom line and, to a lesser extent, because of what it discloses about the influence of several controlled-for variables. The headline finding is that when it comes to executions, event dependence is extraordinary; over an eighteen-month interval, a county with previous executions will be five times as likely to execute a new prisoner as a county with no previous executions. Quite surprisingly, county-level homicide activity and poverty rates have no effect on how many people a county executes. There is a smaller positive correlation between executions and the county’s population of people of color, a variable sometimes called “racial threat.” That is, majority-white communities are more likely to resort to aggressive social control — including ramping up police presence and aggressive prosecutions — in situations where there is a greater fraction of minority membership. Although I am focused on event dependence, and although racial threat is just a control variable, the horror of the concept bears repeating: even controlling for criminality and poverty, predominately white communities that include larger non-white membership tend to use the death penalty more extravagantly.

The authors readily admit that a better test for event dependence might involve capital sentences, rather than executions — most defendants who are sentenced to death are never actually executed, and so the execution data (as opposed to capital sentencing data) is uniquely sensitive to certain types of shocks, including the availability of lethal injection drugs. The execution model is nonetheless powerful evidence of event dependence — and there is also recent county-level sentencing data, assembled by University of Virginia Professor Brandon Garrett, which he has in turn used in at least one articleand one book. (I used an early version of the data set in my muscle-memory article.) Using county-level death sentencing data covering 1990 to 2016, Professor Garrett, along with several co-authors, studied the degree of event dependence. Although the methodology used by Professor Garrett and his coauthors diverges somewhat from the methodology used by Professor Baumgartner and his, the findings converge. As with executions, one of the greatest predictors of whether a county will sentence people to death in the future is whether it sentenced people to death in the past. Moreover, the explanation offered by Professor Garrett et al. lines up with that offered by Professor Baumgartner — as well as with mine. Whether the term is “inertia” or “muscle memory” or “path dependence” or “self-reinforcement,” local, bureaucratic habituation is causing the vast majority of American counties to sentence to death and kill no one, and causing an extremely small number to sentence and kill repeatedly.

To understand the theory writ small, consider the role of the prosecutor. To train its prosecutors to navigate the various institutional hurdles that an execution presents, district attorneys must train their lawyers to investigate aggravating evidence, select a death-qualified jury, try a sentencing case, conduct post-conviction proceedings, obtain an execution date, and defend the judgment in crisis-phase litigation — all unique facets of a death penalty case. The human capital involved in successfully performing all of these tasks is enormous; to realize a reasonable return on that considerable investment, an office must produce a lot of capital punishment. In some ways, it’s a simple economic story of spreading high fixed costs across a more output. But once an office is trained to, well, kill, its existing human resources are bureaucratically conditioned to transmit that lethality to the next generation of attorneys, with the subsequent generation passing the torch to one that comes after it, and so forth. But the bureaucratic habituation that causes certain urban, well-endowed counties to remain active capital punishers actually causes most others to abstain. Once the institutional knowledge necessary to secure executions begins to disappear, it is unlikely to be revived in the absence of unusual political will, substantial financial resources, and caseload of death-eligible cases worthy of renewed institutional investment.

A capital punishment practice operating through muscle memory is quite concerning, because it signals profound arbitrariness. States are meting out death sentences not by systematically isolating the worst of the worst, but by isolating those in certain area codes. When people like Clarence Thomas extol the virtues of geographically differentiated punishment, they generally contemplate things like different states having different punishment schemes, or different localities expressing the collective blaming preference of an affected community. Kent Scheiddeigger, an energetic proponent of vigorous death penalty practice, responded to a famous study showing geographic disparities by snarking: “The study calls the variation by county ‘geographic disparity.’ I call it local government.” What studies like Professor Baumgartner’s show, however, is that maps of capital punishment activity are not capturing the output of good governance or particularly severe punishment preferences; they are capturing a largely arbitrary distribution of death penalty activity, suffused with path dependence.