Sie sind auf Seite 1von 24

Military Executions and Commutations in WWI:

Bias against the Irish or Random Assignment?

Tarun Singh

Abstract

Racial and ethnic tensions continue to dominate the news, whether it is

a Presidential election or cases of perceived preferential treatment in college

admissions. The prevalence of these types of feelings is exemplified in a

2005 Australian AC Nielsen poll in which 75% of those surveyed agree that

there is underlying racism in Australia. Most democratic countries pride

themselves on an independent judicial system, yet there is a continued

concern that these racial and ethnic tensions have influenced the judicial

system’s decision-making process. Previous economic studies in the United

States have found racial and ethnic minorities to be given slightly to

moderately harsher prison sentences by US federal courts. The extent of

asymmetric treatment is not limited to prison sentences, but also includes

the issue of fewer probation sentences being given to ethnic and racial

minorities, when the option is available (Mustard, 2001). This paper looks to

British military executions in WWI to examine whether death sentences and

commutations were randomly issued or whether they were biased against

the Irish or biased on some other observable characteristic. Historians have


argued that the executions and commutations of soldiers were not based on

characteristics of the soldiers and were instead random. In this paper, I argue

that an Irish background seems to have no significant effect on executions

and commutations but more data is needed.

1 Introduction and Literature Review

1.1Motivation

The motivation for this paper is twofold: the first to see if the executions

and commutations were biased against the Irish minority and secondly to

see if the executions and commutations were essentially random. Exploring

these two motivations would allow us to address the question of whether

historians have an accurate claim in saying that the Irish were discriminated

against and secondly, would allow us to use this data set as a random

experiment to address if the death penalty has a deterrent effect on crime.

Studies done on race and prison sentences in United States federal courts

system have shown there to be a slight to moderate bias against minorities.

Even after controlling for socio-economic characteristics, minorities, Blacks

and Hispanics in particular, are more likely to receive a harsher sentence

than their white counterparts (Demuth and Steffensmeir, 2000). Other

studies have suggested that the bias also extends to decisions regarding

probation, where minorities are again less likely to be given probation when

the option is available, and have found that certain minorities are more likely

to receive harsh sentences for particular transgression (Mustard, 2001).


Much has also been stated about the deterrence effect of the death

penalty. Proponents of the death penalty cite the potential deterrent effects

of the death penalty, while others argue that the deterrent effect may not be

significant. According to basic economic theory, and basic intuition,

increasing the costs of an activity ought to decrease incentive for that

activity. Thus, it would seem reasonable to conclude that the death penalty

acts a deterrent to criminal activity; however, studies have found no

deterrent effect (Donohue and Wolfers, 2006). However, a problem in

analyzing death penalty data is the difficulty in disentangling the various

causal effects of crime (Chen and Horton, 2008). By showing that the

executions and commutations by the British in WWI were essentially random,

the data can be drawn upon to test the deterrent effect of the death penalty.

2. Historical Information

The British military’s decision to sentence over 3,000 of its own soldiers to

death in WWI, and actually executing about 10% of such cases, was an

unprecedented move. Prior to WWI military executions by the British

government were rare even for serious transgressions; however, 2,007 of the

3,055 soldiers given the death sentence were tried for desertion (McHugh,

1999). Many commanders in the British military viewed the death penalty as

an essential tool in limiting disciplinary issues (French 1998). Since over

3,000 soldiers were sentenced to death but only 355 were actually executed,

some historians have suggested the military employed “an implicit policy on
the confirmation of death sentences that amounted to a form of

‘bureaucratic decimation (Oram, 2003).’” In particular, capital punishment

was used as a means to combat the high desertion rates that the British

were facing. Military commanders were trying to balance executing too few

or too many soldiers while maintaining morale and avoiding public scrutiny

(Oram, 2003, Chen and Horton 2008).

Each execution had to be approved by the Commander-In-Chief of the

military, who was presented a briefing regarding the merits of the case and

information regarding the soldier’s ability and past record. Decisions were

usually made within two weeks, and the soldier’s were typically executed 24

hours after being made aware of the decision (Chen and Horton, 2008). The

process also allowed for an appeals process in which the soldier could

petition the King for a commutation within 24 hours of when the execution

was to be administered but historians have stated that most soldiers were

not aware of the appellate procedure and therefore the death penalty was

more likely to have a deterrent effect (French, 1998).

It is important to note that there have been mixed reviews given to the

Irish and their fighting ability in WWI. Some historians have stated that the

Irish were aware of the poor disciplinary records (Bowman, 2003). Others

have argued that the Irish fought more valiantly than other soldiers as

evidenced by the number of Victoria Crosses received by Irish soldiers

(MacDonagh, 1917). MacDonagh states that 221 Victoria Crosses were

awarded to soldiers by the end of WWI, but had every group received Victoria
Crosses at the same rate as Irish soldiers, the number would have been

around 350.

3. Data

Our data is comprised of a data set of capital cases, data set of absentee

lists recorded by the British government, and a data set of Irish last names.

The capital cases data set has 3,342 observations which are recorded from

October 1914 to September 1923. The capital cases data set includes the

soldier’s name, unit, rank, date of death sentence, offence, final sentence,

location of the theatre of operation, reference number (as in the national

archives), age at time of execution, division, brigade and other information.

The absentee data set contains 2019 observations from 1914 to 1917 which

records details of soldiers who were reported to be absent during roll call.

The data contains the name of soldiers, their military number, rank, unit,

date the soldier was reported absent, age, height, description and who filed

the report.

This paper examines the randomness of military executions and military

commutations by using data from capital cases and data from absentee lists.

Since the motivation for the paper is partially driven by the deterrence effect

of the death penalty, we will be only looking at soldiers who were tried for

desertion, since desertions make up 2,007 of the 3,342 observations in the

capital data set.


The paper attempts to examine randomness by correlating a binary

variable of died (which tells us if the soldier was actually executed) with a

binary variable of whether the soldier is Irish (as determined by the soldier’s

last name). I will also run various regressions to see if other variables such as

the year of the sentence are significant in predicting military executions.

Furthermore, I will look at whether the Irish were more or less likely to

receive a full pardon (meaning that the case was thrown out and no

punishment was given), while again using various regressions which will try

to control for other variables. To use more descriptive variables I merge the

capital case data and the absentee data (on last name first initial and year)

to use descriptive characteristics as control variables. Since the outcome

variables (died and fullpardon) are binary variables the paper relies on using

logit and probit regressions to account for any non-linearity.

Secondly, the paper examines whether the Irish are more like to be

absent during roll call as this would provide the British military with a rational

reason to execute more Irish soldiers as the necessity of a deterrent would

be greater. This will be especially important if it turns out that the Irish were

more likely to be executed for desertion than their colleagues were.

Essentially, if the Irish were more likely to be absent, then it would make

strategic sense to punish the Irish more, and therefore may lead to a non-

random implementation of the death penalty.

There is a concern that if Irish soldiers are being discriminated against

for non-desertion crimes the impact of this discrimination may influence an


Irish soldier’s decision to desert. For instance, if an Irish soldier sees that

smaller crimes such as sleeping are leading to executions, especially for Irish

soldiers, the soldier may sense a bias and therefore be less inclined to

commit a more egregious crime like desertion. Thus, the paper analyzes the

effect of Irish on Died, FullPardon, and PartialCommute for Non-Desertion

data as well. The paper then shifts focus slightly and examines whether

those soldiers who attempted to desert on multiple occasions were more

likely to be executed and if these soldiers were disproportionately Irish. In

addition, the paper breaks down Non-Desertion crimes and analyzes if Irish is

a significant predictor for any of these crimes. Finally, the paper attempts to

disentangle the PartialCommute variable and breaks it down into Prison

Sentences and Labor Sentences to check if the Irish were disproportionately

given a certain type of sentence. We further break down the data and check

if the Prison and Labor sentences received by Irish soldiers were more severe

than the sentences received by their non-Irish colleagues.

4. Empirical Strategy

The empirical strategy is simple and straightforward, I look at the

impact variables like Irish have on the binary variables of Died, FullPardon

and PartialCommute. (Died=0 if not executed and =1 if executed,

FullPardon=0 if a sentences was received and =1 if no sentence was

received, and PartialCommute=1 if a sentence other than execution was

received and =0 if no sentence was received.) I do this by running OLS

regressions, followed by Logit and Probit regressions due to the dependent


variables being binary. The OLS regressions are intended to simply provide a

general idea of the effect of our explanatory variables while the Logit and

Probit models should do a better job of measuring the significance of each

explanatory variable. When exploring the impact of a variable with multiple

categories, for instance, sentence length isn’t entirely continuous but is more

categorical since one can receive a 6 month sentence, 1 yr sentence, but

rarely receives a 272 day sentence; in these cases I use an ordered probit

specification.

4.1 Specifications

OLS Regression Specifications:

Died i = β 0 + β1 Irishi + ... + β nYrsi + u i , Fullpardon = β + β Irish + ... + β Yrs + u ,


i 0 1 i n i i

Partialcommutei = β 0 + β 1 Irishi + ... + β nYrsi + u i

Logit Regression Specifications:

Pr(Diedi = 1|Irishi,…,Yrsi) = 1/(1+e-(β0 + β 1Irishi + βnYrsi + ui)),

Pr(Fullpardoni = 1|Irishi,…,Yrsi) = 1/(1+e-(β0 + β 1Irishi + βnYrsi + ui)),

Pr(Partialcommutei = 1|Irishi,…,Yrsi) = 1/(1+e-(β0 + β 1Irishi + βnYrsi + ui))

Where 1/(1+e-(β0 + β 1X1i+… + β nXni)) represents the cumulative standard logistic

distribution function

Probit Regression Specifications:


Pr(Diedi = 1|Irishi,…,Yrsi) = Φ(β0 + β 1Irishi + βnYrsi + ui), Pr(Fullpardoni = 1|

Irishi,…,Yrsi) = Φ(β0 + β 1Irishi + βnYrsi + ui) , Pr(Partialcommutei = 1|Irishi,…,

Yrsi) = Φ(β0 + β 1Irishi + βnYrsi + ui)

Where Φ is the cumulative normal distribution function

Ordered Probit Predicted Probabilities:

Pr(Offencenumberi = 1|Irishi) = Φ(c1 – β1Irishi)

Pr(Offencenumberi = 2|Irishi) = Φ(c2 – β1Irishi) - Φ(c1 – β1Irishi)

…. Pr(Offencenumberi = 6|Irishi) = 1- Φ(c6 – β1Irishi)

Where Φ is the cumulative normal distribution function

Note: In order to avoid redundancy the specific specification for

each regression is not listed, but the above specifications cover all

types of regressions used in this paper, the remaining regressions

are variations that can be written by replacing/adding variables to

the ones above.

4.2 Choice of Controls

Historians have argued that the capital cases suggest that there was

bias based upon soldier characteristics such as race, ethnicity and physical

characteristics that made the decision to execute or commute non-random.

The racial and ethnic characteristic that historians have thought to be a

significant is whether if the soldier is of Irish decent while the physical

characteristics have been thought to include the soldier’s height, age, etc.
Furthermore, running a simple correlation test between Died and Yr reveals a

negative relationship, leading to some additional concern over whether

executions were in fact completely random. We see a similar relationship

when using the variable monthssincestart, which measures the number of

months since the start of the war and monthsss2 which is just

(monthssincestart)^2.

We are able to address the concerns regarding racial, ethnic and year

bias using the data that is provided in the capital cases data set which allows

for these control variables. (The capital cases data set does include the age

of soldiers that were executed, however it does not not include the age for

any of the soldiers that were partially commuted or fully pardoned, meaning

90% of the data on desertion would not include soldier age. Using any data

from age from this data set would yield high standard errors.) In order to

address the lack of descriptive variables available in the capital data set I

chose to merge the capital cases data set with the absentees data set. The

data was merged on the soldier’s last name, first initial and year of absence/

year of trial. After merging the two data sets, I proceeded to drop those

observations in which the soldiers’ whose units did not match in both the

capital cases and absentees data sets. Doing so we are left with only 37

observations, but the merged dataset allows us to control for variables such

as age and height in inches.

Similarly, when analyzing the Absentee data set to check for whether

the Irish were more likely to report absent in roll call, I attempted to control
for yrs and height in inches to help limit any possible bias. The Non

Desertion, repeat offences, specific crimes, prison sentence and labor

sentence regressions all use the same controls as the pre-merge Desertion

data. These controls are used for the same reasons as before, and the reason

for not attempting to merge with Absentee data to establish more control

variables is due to the small number of observations we were left with after

merging the Capital Cases and Absentee datasets.

Controlling for these variables in the regressions should do two things;

it should help omit any significant omitted variable bias that would otherwise

be captured by Irish and should allow me to evaluate whether the effects of

these control variables are significant enough to conclude non-randomness.

5.1 Desertion Results

Of the 2,007 soldiers who were sentenced to death for desertion, 405

or about 20.2 percent of these soldiers were Irish. Of those who were actually

executed (274 soldiers total) about 20.1 percent or 55 soldiers were Irish.

Similarly, a total of 123 of the 2,007 soldiers received a full pardon and were

ultimately given no sentence. Of these 123 soldiers, about 24.4 percent or

30 soldiers were Irish. These summary statistics do not provide us with

overwhelming evidence alone, but do suggest that the perceived bias in

treatment of Irish versus non-Irish soldiers was not as prevalent as some

historians have suggested. In order to address whether these differences are

statistically significant the paper looks at the following regressions.


5.1a Desertion Executions

The results of this paper lead us to believe that there is likely little bias

for executions that was based on race. The preliminary regressions in which

Died was regressed on Irish showed a small negative coefficient on Irish

(contrary to what should have been seen according to historians) that was

far from significant. However, when controlling for the time in which the

sentence was given, we see that although the p-value on Irish does not

change too much, Yrs and Monthssincestart turns out to be highly significant

(see Table 1 regressions 4-9). Yrs and Monthssincestart are shown to have a

negative effect on executions and are significant at 1 percent level. After

merging the data with the absentee list, we see that the effect on Yrs

becomes positive and is no longer statistically significant; this indicates

either that the merge data points (only 37) are not representative of the

2,007 data points from capital cases data set or that the standard errors are

extremely high due to the lack of data. Thus, although Irish is shown to be

significant when adding other control variables there is too little, and most

likely a non-random sample of data after the merge occurred. These post-

merge regressions can be seen in Table 1 in the Appendix.

5.1b Desertion Full Pardons

The summary statistics alone suggest that the Irish were more likely to

be given full pardons than the rest of their colleagues, however on further
examination this seems to be inaccurate. Although all regressions showed a

positive effect of being Irish on the likelihood of being give a full pardon, no

regression showed this effect to be statistically significant even at the 10

percent level. Like the case with executions, controlling for Yrs did not make

the effect of Irish statistically significant but did show the effect of Yrs to be

significant at even the 1 percent level (see Table 1 regressions 10-15). The

regressions show a negative effect of Yrs on full pardons, meaning as the war

went on a soldier was significantly less likely to receive a full pardon than

earlier in the war. This is supported by the summary statistics, which show

an initial increase in full pardons followed by a large decrease in full pardons

(see appendix). Although the post-merge data for full pardons showed the

sign on the coefficient of Yrs to be the same as in pre-merge data, I feel

uncomfortable drawing any conclusions from the data due to the same

concerns of it being not representative of the overall data.

5.1c Desertion Partial Commutations

In regressions 16-18 of Table 1 we see that more soldiers are partially

pardoned as Yrs increases. The coefficient on Yrs is significant at the .1

percent level. This information, combined with 5.1a and 5.1b tells us that as

the war progressed the British military shifted away from executions and full

pardons and towards partial pardons. So although fewer soldiers were being

executed as time went on, fewer were also being fully pardoned. However,

the interpretation of partial commutations is slightly nuanced; most soldiers


who received partial commutations were sentenced to either a prison

sentence or a labor sentence of varying time periods- so two questions still

remain, were the Irish more likely to receive a prison sentence or labor

sentence than their non-Irish counterparts and within these sentences were

the Irish subjected to longer sentences? Both questions are addressed in

section 5.6 and 5.7.

5.2 Absentee Data Results

The absentee data did not reveal much in terms of showing that the

Irish were more likely to be absent for roll call. Of the 1563, total soldiers that

were reported absent 20.86 percent, or 326 of the soldiers were Irish. This

number is similar to the percentage of soldiers sentenced to death for

desertion who were Irish (20.2 percent). In regressions 1, 2 and 3 of Table 2,

we check to see if yrs had any effect on the number of Irish soldiers

reporting absent. Doing so, we were left with 1405 observations in which the

variable yrs had a negative impact on Irish that was significant at the 5

percent level. However, when running the same regressions but also

controlling for height in inches in regressions 4, 5 and 6 (N=1069) we see

that the negative coefficient on yrs is still there but it is no longer significant

at the 5 percent level.

5.3 Non-Desertion Results


Of the 1,335 soldiers who were sentenced to death for non desertion

crimes, 233 or about 17.45 percent of these soldiers were Irish. Of those who

were actually executed (166 soldiers total) about 12.7 percent or 21 soldiers

were Irish. Similarly, a total of 109 of the 1,335 soldiers received a full

pardon and were ultimately given no sentence. Of these 109 soldiers, about

13.8 percent or 15 soldiers were Irish. Of the 1,059 soldiers receiving partial

commutations, 18.6 percent or 197 soldiers were Irish. These summary

statistics by themselves suggest little about whether there is significant

evidence to show that the Irish were treated differently than other soldiers,

but the following regressions attempt to show if this is the case.

5.3a Non-Desertion Executions

Looking at the summary statistics alone leads one to believe that the

Irish were actually treated preferentially when it came to being executed. If

there was a bias against Irish soldiers we would’ve expected to see more

than 17.45 percent of the soldiers executed to be Irish, but Instead we found

a figure almost 5 percent lower. Our suspicion that the Irish are being treated

favorably with respect to executions is somewhat confirmed in regressions 1-

12 of Table 3. All the regressions show that Irish has a negative effect on the

likelihood of execution, but this coefficient is insignificant at the 5 percent

level in all 12 regressions. What are significant are the time variables. The

variable Yrs, has a positive effect on the chance of being executed which is

significant even at the .1 percent level as seen in regressions 4-6. In


regressions 7-9 we see that this positive coefficient on time and this level of

significance is maintained when we replace Yrs with Monthssincestart.

Monthssincestart is still significant at the 1 percent level in both the probit

and logit regressions when adding the squared term Monthsss2. These

results show that contrary to desertion executions, non-desertion executions

rose as the war went on. Fortunately, Yrs and Monthssincestart are variable

that are easy to control for when attempting to analyze the deterrent effect

of the death penalty.

5.3b Non-Desertion Full pardons

The summary data for non-desertion full pardons suggests the

opposite of the data for non-desertion executions; it seems as if the Irish

were discriminated against by being fully pardoned at a lower proportion

than 17.45 percent. This negative relationship is upheld in Table 3

regressions 13-18 of the non-desertion data, however, like before, the

negative coefficient on Irish is insignificant in all of these regressions.

Another departure from the results we found for non-desertion executions is

seen with the fact that Yrs had an insignificant relationship with being fully

pardoned (regressions 16-18). Although the coefficient on Yrs was negative

in the desertion data as well as non-desertion data the fact that Yrs is

insignificant for non-desertion data is surprising considering Yrs had been

shown to be extremely significant (.1 percent level) in the desertion data.

This suggests that there may be some unobservable difference between


desertion and non-desertion crimes, or a specific trend due to the nature of

the fighting that is causing these somewhat inconsistent results.

5.3c Non-Desertion Partial Commutations

Like the case with the execution data, the numbers suggest that the

Irish were more likely to be partially commuted compared to the proportion

of Irish soldiers, although the difference is only about 1 percent. Again our

first inclination is upheld in Table 3 regressions 19-21 where we see a

positive coefficient on Irish, but as in 5.3b neither Irish nor Yrs had a

significant effect on the dependent variable. Like in the full pardon data, we

see a difference in desertion and non-desertion crimes since the coefficient

on Yrs is not significant in our non-desertion data but is in the desertion data.

5.4a Repeat Deserters and Execution

If the administration of the death penalty was indeed random, we

would expect that there would be no correlation between execution and the

number of times a soldier had deserted. Whereas, if the commander in chief

acted rationally, we would expect to see repeat deserters punished more

severely. In order to test this I created a categorical variable, offencenumber,

which is distributed from 1 to 6: 1 is a soldier who has deserted once, 2- a

soldier who committed another offence in addition to desertion, 3- a soldier

who deserted twice, 4- a soldier who deserted 3 times, 5- a soldier who

committed another offence in addition to deserting 3 times and 6- a soldier


who deserted 4 times. Regressions 1-9 in Table 4 show that even when

controlling for time using Yrs or Monthssincestart, offencenumber has a

positive effect on execution that is significant at the .1 percent level in all 12

regressions. Like in the 5.1a, we see that the time variables, Yrs and

Monthssincestart have negative coefficients and are significant at the .1

percent level. This is while Irish is insignificant, meaning that repeat

offenders are more likely to be executed regardless of whether they are Irish.

5.4b Repeat Deserters and Irish

A possible concern from 5.4a is that the effect of being Irish is being

picked up by being a repeat offender. Thus I used an ordered probit model to

regress Irish on offencenumber. Doing so in regressions 10-12 of Table 4 we

see that Irish an insignificant predictor of whether a soldier was a repeat

offender. So it is unlikely that the effect of Irish is somehow being picked up

by offencenumber in 5.4a. Furthermore, this finding only provides further

evidence that the commander in chief was looking at case specifics, by

looking at whether the soldier was a repeat offender, rather than randomly

assigning the death penalty as some historians have suggested.

5.5 Individual Crimes and Executions

Having already analyzed desertion and whether Irish was a significant

predictor, and having looked into whether all non-desertion crimes were

significantly influenced by Irish there is still a concern that there may be


particular crimes within the non-desertion crimes for which the Irish are

being asymmetrically executed. As mentioned earlier, if it turns out that the

Irish are being disproportionately executed for smaller crimes such as

sleeping it may provide Irish soldiers with a deterrent effect that carries over

to desertion. Of the crimes listed in the capital cases data set I chose to only

use those which had at least ten observations in order to somewhat minimize

standard errors. Using this method there were 12 non-desertion crimes that I

looked at: Against Inhab., Cowardice, Disobedience, Insubordination, Murder,

Mutiny, Quitting, Rebellion, Sleeping, Striking, Treason and Violence. As

shown in Table 5 the only crimes which yielded any significant results were

Murder, and Sleeping. As seen in regressions 36-38 Yrs had statistically

significant (at the 5 percent level) negative impact on the amount of soldiers

who were executed for Murder. Similarly, regressions 68 and 70 show a

positive effect of Yrs on died for those soldiers prosecuted for sleeping (also

significant at the 5 percent level).

5.6 Were Irish More Likely to Get Prison or Labor Sentences?

To look at whether the Irish were more likely to receive a prison

sentence or a labor sentence I created a new variable, commutetype, which

equals 0 if the soldier was given a prison sentence and equals 1 if the soldier

was given a labor sentence. In Table 6 we see that Irish is insignificant at

predicting what type of sentence was received in all nine regressions.

However, regressions 4-9 show that as time went on the British military
moved away from issuing labor sentences and towards prison sentences.

This effect was significant even at the .1 percent significance level. As

mentioned earlier, this doesn’t tell us whether the Irish were being treated

equally, as it doesn’t address whether the lengths of the prison and labor

sentences being received by Irish soldiers are commensurate with their non-

Irish colleagues- this concern is addressed in the following two sections.

5.7 Irish and Prison Sentences

When disentangling whether the Irish are more likely to receive longer

prison sentences, I created a new variable called sentencelengthordered. The

variable is a categorical variable from 0-10 since there were 11 different

prison sentences that were issued, ranging from 2 years to Life. Thus, a

value of 0 corresponds to a 2 year sentence and 10 corresponds to a life

sentence. Thus, since the dependent variable is a categorical variable and is

not binary, we use ordered probit regressions in Table 7. Doing so we see

that Irish is not a significant predictor, but the time variables Yrs and

Monthssincestart are. Both Yrs and Monthssincestart have positive

coefficients that are significant at the .1 percent level, meaning that as the

war progressed there was increased likelihood of receiving an increased

sentence.

5.8 Irish and Labor Sentences


The same methodology that was used in section 5.7 was used to

examine potential bias in labor sentences. Again, the categorical variable

sentlenghtordered was created to span from 0 to 7 to correspond to

sentences ranging from 168 days to Life respectively. Like with prison

sentences, Irish was not a significant predictor for labor sentence length

either but the time variables Yr and Monthssincestart were significantly

positive at the .1 percent level (see Table 8). This means that the British

were increasing sentence length for all partial commutations as the war went

on.

6. Possible Explanations

The results from the data provide us with mixed results, when looking

at desertion and non-desertion crimes the general theory put forward by

those who believe the death penalty was used randomly as a deterrent tool.

It makes sense for there to be a negative effect of Yrs on executions if the

executions are being used as a deterrent since a high number of early

executions would signal a serious attitude towards desertion, causing there

to be less need for more executions later due to the lag effect of an

execution. This would also explain why we say prison sentence and labor

sentence lengths increase over time, if fewer people are being executed,

those who may have been on the “borderline” of executions may instead be

given life sentences. This also follows historical accounts that cite the

growing opposition in parliament, especially amongst Liberal Party members,


towards military executions. However, there is a problem with this

explanation: the repeat offenders data shows that the commander in chief

was taking into account the soldier’s priors when making the decision to

execute, which is far from random. However, higher execution rates at the

start of the war does not match the theory of some historians who have

argued that the Irish became more likely to desert after Irish conscription

began in 1918 as it made for lower likelihood of Irish home rule (Gregory and

Peseta 2002).

Similarly, a soldier being less likely to receive full pardons as the war

went on seems reasonable under two possible theories. The first theory is

that of experience; as the war proceeded, the British would have had a

better understanding of which sentences would be upheld and which would

be overturned by the Commander-In-Chief, therefore fewer egregious cases

would be sentenced to death, therefore resulting in fewer full pardons. This

explanation should also manifest itself in increasing execution rates as the

war went on, since hypothetically only legitimate cases were being given the

death sentence. This is actually confirmed when we look at non-desertion

cases where we see that executions increased as the war went on, giving

some credence to this theory. The second possible explanation is that the

military was trying to avoid media and political attention due to its capital

punishment policies and therefore a push away from executions as the war

lingered on was counteracted with fewer full pardons in order to have a

similar overall deterrent effect as earlier in the war.


6.1 Further Analysis

Further analysis is required in order to comfortably conclude that the

issuing of executions and commutations was randomly assigned, especially

given the results regarding repeat offenders. Although Yrs has been shown to

have a significant effect in almost all regressions, it would be appropriate to

try to account for a soldier’s other observable characteristics in order to see

how much of this effect can in fact be attributed to years. The potential for

omitted variable bias is therefore a reasonable threat that needs to be

further explored. In order to address this concern it would be worthwhile to

pursue a more complete absentee data set that extends beyond 1917. This

would allow for better merging between the capital and absences data sets

and would allow testing for historian claims, which have suggested a change

in Irish attitude towards the war after 1918 as the year brought with it forced

Irish conscription and dwindling prospects for Home Rule.

7. Conclusion

The execution of its own soldiers for desertion has been a controversial

issue since a few years after WWI when the Labor Party brought the issue to

Parliament in order to abolish military capital punishment. Since then claims

by historians of discrimination against Irish soldiers have only made the

somewhat tense race relations between Irish and non-Irish worse. To add

another dimension to the issue, the British government recently decided to


pardon all executed soldiers from WWI posthumously due to a lack of

understanding of shell shock by military leaders at the time. My analysis in

this paper leads me to believe that the Irish weren’t discriminated against in

the administration of the death penalty and other sentencing, but it is also

unlikely that the decision making was truly random. There is still potential for

this data to be used as a natural experiment if one is able to control for

repeat offenders, and is somehow able to correctly assign a diminishing

deterrent effect for the execution of a repeat offender. There is definitely

potential for this data to be used effectively to measure the deterrence effect

of the death penalty.

Das könnte Ihnen auch gefallen