«Abstract During World War I the British military condemned over 3,000 soldiers to death, but executed only approximately 12% of these soldiers; the ...»
The Deterrent Eﬀect of the Death Penalty?
Evidence from British Commutations During World War I
Daniel L. Chen⇤
During World War I the British military condemned over 3,000 soldiers to death, but executed only approximately 12% of these soldiers; the others received commuted sentences. Many
historians believe that the military command conﬁrmed or commuted sentences for reasons unrelated to the circumstances of a particular case and that the application of the death penalty was essentially a random, “pitiless lottery.” Using a dataset on all capital cases during World War I, I statistically investigate this claim and ﬁnd that the data are consistent with an essentially random process. Using this result, I exploit variation in commutations and executions within military units to identify the deterrent eﬀect of executions, with deterrence measured by the elapsed time within a unit between the resolution of a death sentence (i.e., a commutation or execution) and subsequent absences within that unit. Absences are measured via “wanted” lists prepared by British military police units searching for deserters and preserved in war diaries. I ﬁnd limited evidence that executing deserters deterred absences, while executing non-deserters and Irish soldiers, regardless of the crime, spurred absences in general, and Irish absences in particular. These ﬁndings suggest that legitimacy may play an important role in why people obey the law.
Keywords: Compliance, Legitimacy, Deterrence JEL codes: N44, K14, K42, P48 ⇤ Chair of Law and Economics, ETH Zurich, email@example.com. I would like to thank Gerard Oram for providing the death sentence data, numerous colleagues for helpful comments and discussions, and several dedicated research assistants. Work on this project was conducted while I received ﬁnancial support from the Ewing Marion Kauﬀman Foundation, Institute for Humane Studies, John M. Olin Foundation, Petrie-Flom Center, and Templeton Foundation.
1 Introduction 1.1 Motivation After decades of empirical research (Ehrlich 1975), there is little convincing evidence that the death penalty deters any form of misbehavior (Donohue and Wolfers 2005). What makes this absence of evidence so intriguing to some observers is that economic theory makes an unambiguous prediction: raising the cost of some activity will cause a decrease in its incidence, be it illegal parking, homicide, or military desertion. The great econometric challenge of death penalty research is that the death penalty is applied in way that makes deﬁnitive conclusions hard. In the U.S., states that allow the death penalty diﬀer from states that do not in important ways that probably have independent eﬀects on the level of crime. Further, assessing the eﬀects of the death penalty requires the examination of crime rates in the future, but since crime has multiple causes, disentangling the eﬀect of the death penalty from other confounding socio-economic or cultural factors is challenging.
Despite these empirical diﬃculties, whether the death penalty deters crime seems in principle to be an answerable question. In an interview with the New York Times1 regarding the state of empirical death penalty research, Professor Justin Wolfers, a skeptic of existing empirical death penalty research, said, “If I was allowed 1,000 executions and 1,000 exonerations, and I was allowed to do it in a random, focused way... I could probably give you an answer.” Mr. Wolfers’ scenario is (thankfully) unlikely to come to pass, but the British Army experience during World War I may be an approximation: a large number of soldiers were executed or commuted for seemingly arbitrary reasons despite having committed essentially identical crimes. In this paper, using the quasi-random application of the death penalty during World War I, I test whether the death penalty deterred desertion. Although this paper answers a question diﬀerent from that addressed in the usual death penalty research, it has the advantage of a relatively clear source variation that allows identiﬁcation of any eﬀects. Further, this study focuses on the more basic and timeless question of whether the threat of death by execution inﬂuences individual decision-making, albeit in a very particular setting.
On another level, the data allow me to examine the potential role of legitimacy in why people obey the law. Do people follow the law because they fear punishment or because they think the law and the lawgiver has legitimacy and the law ought to be followed? Minorities in the U.S. are disproportionately sentenced to death (Donohue 2011) and higher rates of crime among disadvantaged groups have been attributed to mistrust of legal institutions (Tyler and Huo 2002) and brutalization (Bowers and Pierce 1980; Bailey 2006). Irish soldiers during World War I were roughly six times more likely to be sentenced to death and executed than British soldiers and Ireland declared independence shortly after the war. Examining whether Irish soldiers were spurred to desert following executions of Irish soldiers thus contributes to a broader debate — whether, as in a classical law and economics framework, deterrence alone drives behavioral responses to the law or whether considerations such as legitimacy explain why people obey the law. While social scientists and political Does Death Penalty Deter? A New Debate, November 18th, 2007.
philosophers have long speculated on the role of legitimacy in organizations (Suchman 1995), courts (Gibson et al. 1998), and democracies (Lipset 1959), no causal ﬁeld evidence examining actual behaviors exists to date.
1.2 Historical Context British Commanders of the era were convinced of the deterrent power of the death penalty (Oram and Putkowski 2005). Over 3,000 soldiers received a death sentence, but British Expeditionary Force (BEF) commanders conﬁrmed the sentences of only a fraction of condemned soldiers, with those not executed receiving commuted sentences. The lower panel of Figure 1 shows a plot of the distribution of death sentences and their resolution over the course of the war (the upward-pointing tick marks indicate a commutation and the downward-pointing tick macks indicate an execution).
Historians believe that there were two reasons for this restraint: (a) commanders were sensitive to political pressure and were concerned about popular anger back home, and (b) commanders were reluctant to execute soldiers who might still make some contribution to the war eﬀort (Oram 2003). These two concerns, balanced against the desire to deter desertions, led to a fairly constant execution rate of around 12% across divisions (see Figure 2) and across time (see Figure 3) — an almost literal decimation — with most soldiers being executed by a ﬁring squad of their fellow soldiers, usually from that soldier’s same unit. Soldiers whose lives were spared normally returned to the trenches and received prison terms or hard labor to be served after the war. A soldier could not get a safe jail sentence that would have allowed him to leave the trenches.
1.3 Basic Empirical Framework
To examine whether executions deterred desertions, I adopt the language of potential outcomes:
I observe what happened in a particular Army division following an execution — I would like to know what would have happened if that same unit had instead experienced a commutation (Rubin 1974). Because I cannot observe the alternate history in which the soldier’s life was spared, I must make an inference. If I believed that the execution and commutation decision was truly random at all times for all Army units, then the logic of the controlled experiment would allow me simply to compare some metric (such as a count of absences in some speciﬁed time period or the duration until some number of absences) in the execution cases with a similar metric in the commutation cases. While some historians do believe this strong randomization occurred, describing the process as a “pitiless lottery,” others are doubtful.
If the commutation decisions were non-random, the non-randomness is likely due to the military commanders’ consideration of several factors: the reputation of the condemned soldier’s unit, the past sequence of executions and commutations within that unit, and the condemned soldier’s individual characteristics. Military historians such as Julian Putowski and Anthony Babington (Babington 1983; Putkowski and Sykes 2007) have argued that the command targeted certain units for execution for their perceived indiscipline but that individual characteristics were irrelevant, while Gerard Oram, a historian of World War I military justice on both the Allies and Central Powers sides, argues that both unit and individual soldier factors mattered. In particular, he argues that Irish soldiers, non-commissioned oﬃcers, and those seen as physically weak or otherwise undesirable were more likely to be executed.
My response to this possibility of non-randomness has three parts. First, I examine whether the information I have about individual condemned soldiers and environmental factors can predict the commutation decision. Second, I restrict my analysis to comparing how executions and commutations within a division inﬂuenced outcomes. Third, I try to detect non-randomness in the sequence of commutation decisions within a division by using a variety of statistical tests to see if commanders targeted units for executions during certain time periods or whether commanders felt certain units were “due” for an execution or became more lenient after an execution.
A second empirical challenge beyond non-randomness is that my within-unit design means that each division is essentially serving as its own control. This method is problematic if I think that past events in a unit’s history can continue to aﬀect outcomes in later time periods. In other words, the stable unit treatment value assumption (SUTVA) is potentially violated since the “treatment assignment” (i.e., execution or commutation) in one unit can aﬀect the outcomes in another unit. I address this problem in three ways: ﬁrst, I assume a strong form of SUTVA in which I posit that only the most recent event matters and second, I parametrically model the eﬀects of previous events and explore whether or not my results are robust to the inclusion of prior events in the model speciﬁcation. Strong SUTVA can lead to underestimates since it assumes control groups are unaﬀected.
My third approach uses a day-by-day, non-parametric model of absence to ensure identiﬁcation does not come from functional form assumptions on the hazard rate of absence. Because it includes all prior events, the non-parametric dynamic treatment design also incorporates idiosyncratic variation over time in the execution rate.
1.4 Unit of Analysis It is necessary to choose a unit of analysis for the study. Military organizations are obviously hierarchical and there is a great deal of discretion in choice of unit-size. The casualty data and absence data is at the battalion level, so I could in principle choose any unit from this level up to the Corps. While there are exceptions, in general, the sequence of military units listed from lowest to highest was: Battalion ! Regiment ! Brigade ! Division! Corps ! Army ! Army Group. Each higher level of organization contains three or four subordinate units plus headquarters and higher-level assets. According to at least one historical account, the division commander was the highest-level commander whose recommendation was ignored (Putkowski and Sykes 2007). If higher-level commanders did target bad units or show discretion, then the division is the highest level appropriate for analysis.
The thinness of the outcome data compels a fairly high-level of organization, even though the salience of an execution and hence its deterrence eﬀect (if any) would be strongest at lower levels of organization. The Order of Battle, which contains dates when each battalion is associated with a particular division, provides incomplete information on intermediate levels of units. The absence data I am able to collect (thus far) is relatively thin. I am able to identify 1,061 usable matches from the Military Police wanted lists preserved in war diaries from France and Flanders; they are preserved mostly for four months in late 1916 (Figure 1 upper panel), so I restrict the analysis of war diaries data to the 899 absences that occur during this window. The median time between trial and next recorded desertion in the ﬁeld at the level of a division is 15.5 days. For a complete coverage, I turn to the Police Gazettes published in the U.K. for a four-year universe of roughly 2,800 BEF deserters who were absent for at least a month.
1.5 Literature In addition to a large empirical death penalty literature summarized critically by Donohue and Wolfers (2005) as well as a recent U.S. government task force (Nagin and Pepper, eds 2012), my paper builds on a literature examining determinants of desertion (Costa and Kahn 2003), a literature using commuted prison sentences to identify causal eﬀects of prison on subsequent crime (Drago et al. 2009; Kuziemko 2011), and a literature using alternative settings (Kuziemko 2006) and structural models with diﬀerent identiﬁcation assumptions to estimate the deterrent eﬀect of the death penalty (Cohen-Cole et al. 2009; Manski and Pepper 2011).
Finding a deterrence eﬀect in the context of World War I would certainly not be a strong argument, leaving aside moral issues, that the death penalty is good policy. However, the British experience may provide a mechanism experiment (Ludwig et al. 2011) for death penalty policy — are individuals less likely to do some action after seeing someone executed for doing that action — and a low-bar test. A negative result showing no deterrent eﬀect might have more policy salience since if we ever expected to ﬁnd an eﬀect, it would be in the World War I context. The World War I death penalty was designed for maximum deterrence: executions were immediate, brutal, and public, with particulars of the situation promulgated widely. In contrast, the modern application of the death penalty seems to be more about retribution — the trend is toward more “humane” forms of execution, exacted after lengthy appeals, conducted basically in private (Katz et al. 2003).