Counterinsurgency Tactics, Rebel Grievances, and

Who Keeps Fighting∗

Connor HuffDepartment of Political Science

Rice University

Abstract

Foundational theories of civil war argue that grievances motivate individuals to rebel. This letterexpands this logic to explain within-conflict rebel behavior, arguing that the counterinsurgencytactics of governments can further increase grievances which in turn shape the likelihood rebelscontinue fighting as conflicts unfold. I test the argument using newly compiled individual-leveldata on over 1,700 members of the Irish Volunteers and Irish Citizen Army who participatedin the 1916 Easter Rising. Rebels varied in whether they evaded capture after the uprising. Ishow that rebels who were interned were more likely to fight throughout the entire Irish Warof Independence. Qualitative evidence corroborates the contention that imprisonment furtherradicalized Irish rebels. The letter elucidates how within-conflict events shape rebel behavior,by documenting how the tactics governments employ as they fight can shape the subsequentactions of the rebels they are combating.

∗Connor Huff (connor.huff@rice.edu, http://connordhuff.com) is an Assistant Professor in the Department ofPolitical Science at Rice University. I thank Muhammet Bas, Matthew Blackwell, Melani Cammett, Thomas Cavanna,Dara Kay Cohen, Jeffry Frieden, Michael Gill, Andrew Hall, Soeren Henn, Jonathan Homola, Hyeran Jo, JoshuaKertzer, Dean Knox, Dominika Kruszewska, Shiro Kuriwaki, Boram Lee, Ashley Leeds, Christopher Lucas, DanielMoskowitz, Robert Schub, Livia Schubiger, Jacob Shapiro, Anton Strezhnev, Kai Thaler, Dustin Tingley, MichelleTorres, Ariel White, Steven Wilkinson, and Thomas Zeitzoff for invaluable feedback throughout the development ofthis project. I thank audience participants at Brown University, the Columbia-Harvard-Princeton-NYU-Yale Politicsand History Conference, Harvard International Relations Workshop, Harvard Experiments Working Group, Harvard-MIT-Tufts-Yale Political Violence Conference, NYU Abu Dhabi, Princeton University, Rice University, RutgersUniversity, UCSD, and Yale University, for helpful comments and suggestions. I am grateful to Cecile Gordon,Michael Keane, and the staff at the Military Archives in Dublin for providing invaluable guidance on the complexitiesof the military pension applications.

1 Introduction

In April 1954, British government forces launched Operation Anvil in the city of Nairobi. More than

4,000 British and African troops swept through the city and arrested approximately 10,000 men

in a large-scale effort to root out the Mau Mau rebels.1 The detainees were subsequently sent to

detention camps throughout Kenya. The camps were hellish: rife with disease, under-supplied food,

with allegations of torture and summary executions circulating widely.2 This 1950s British policy

of arresting and imprisoning suspected rebel combatants has been used by a range of governments

throughout history,3 and is but one of many possible counterinsurgency strategies governments

adopt as they attempt to combat violent rebellions. In addition to imprisoning suspected rebels,

governments execute leaders, torture rebels, and destroy the houses of rebels’ family members. How

do these counterinsurgency tactics shape the subsequent behavior of the rebels they are fighting?

In this letter I build upon a vast body of research for how grievances shape conflict participation

(Cederman, Gleditsch, and Buhaug 2013; Gurr 1970; Paige 1978; Schubiger 2021; Wood 2003)

to argue that counterinsurgency tactics can act as grievance-inducing experiences which further

radicalize rebels. The choice to fight in the first place is shaped by frustration induced by living in

poverty (Gurr 1970), past actions of the government (Wood 2003), and atrocities perpetrated by

the state (Henn and Huff 2021; Schubiger 2021). Other scholars argue that government actions such

as imprisonment, torture, and killing can shape the choice to engage in retaliatory violence and

indiscriminate killing (e.g., Balcells 2017: 119). I extend the logic of this prior work to instead focus

on the decision to continue fighting as conflict unfold. I argue that government counterinsurgency

tactics can further increase rebel grievances. Increasing grievances increases the likelihood rebels

continue fighting. This argument shifts the focus from why individuals rebel, to why rebels refuse

to stop.

While the idea that grievances can shape conflict behavior is pervasive throughout political

science, the individual-level quantitative evidence for this is relatively thin. Many of the quanti-

tative tests of grievance-based accounts of rebellion rely on cross-country analyses and comes to

contradictory conclusions (e.g., Cederman, Gleditsch, and Buhaug 2013; Collier and Hoeffler 2004;

1http://news.bbc.co.uk/onthisday/hi/dates/stories/april/24/newsid_3705000/3705049.stm.2For an overview of the Mau Mau rebellion, see Elkins (2005).3Examples include the 19th century Spanish internment of Cuban rebels and U.S. imprisonment of alleged militantsin Guantanamo Bay.

1

Fearon and Laitin 2003). The studies that do use individual-level data commonly rely on rough

proxies for the concept of grievances such as poverty, a lack of access to education, or living in mud

huts (Humphreys and Weinstein 2008). This letter contributes to this prior work by bringing new

detailed individual-level data to bear on better understanding how the actions that governments

take as they fight shape the conflict behavior of the individual rebels they are combating.

I do so by focusing on one experience within one conflict: the imprisonment of members of the

Irish Volunteers and Irish Citizen Army who participated in the Easter Rising in 1916. Focusing on

a single case in a single conflict allows me to hold contextual factors constant, such as either country

or group level characteristics, and focus on a clear behavioral measure in the decision to continue

fighting. Moreover, focusing on the case of historical Ireland allows me leverage the extensive record

of rebel combatants to compile a quantitative dataset following individual rebels from conflict onset

through conflict resolution. Armed conflict began with the Easter Rising of 1916 and its violent

suppression. It continued through a War of Independence that ended with the Anglo-Irish Treaty of

December 1921, which established an Irish Free State and partitioned Ireland into what is now the

Republic of Ireland and Northern Ireland. Following the 1916 Easter Rising—when roughly 2000

individuals rose up in opposition to British rule in Ireland—some rebels were imprisoned and held

in internment camps while others successfully evaded capture. I compare the behavior of rebels

who were imprisoned with those who were not in terms of whether they fought throughout the

entire Irish War of Independence.

2 New Individual-Level Data on Rebel Combatants

In order to assess how counterinsurgency tactics affect rebel behavior, we must first start with

an initial pool of rebels who decided to fight at conflict onset and then learn about both the

substance of their experiences and subsequent conflict behavior. I do so by leveraging the extensive

archival documentation held in the Military Service Pension Collection from the Military Archives

of Ireland. In the 1920s and 1930s, the Dail of the newly formed Irish Free State passed legislation

intended to provide a military pension rewarding ex-combatants for their service in the fight for

2

Time Period Organization Claimed Active Service Approved

Easter Week 1916 Irish Volunteers Entire Period

April 1, 1916 – March 31, 1917 Irish Volunteers 343/358

April 1, 1917 – March 31, 1918 Irish Volunteers 1/4

April 1, 1918 – March 31, 1919 Irish Volunteers None

April 1, 1919 – March 31, 1920 Oglaigh na-hEireann None

April 1, 1920 – March 31, 1921 Oglaigh na-hEireann 1/12

April 1, 1921 – July 11, 1921 Oglaigh na-hEireann 1/2

Table 1 – The information collected from the Military Service Certificate. The firstcolumn denotes the time period for which individuals could plausibly fight betweenthe beginning of the Easter Rising and end of the Irish Civil War. The secondcolumn denotes the organization for which they could have claimed service. Thefinal column denotes the fraction of the period for which the military pension boardapproved the individual as being engaged in “active service.” Note that Oglaighna-hEireann was the Irish name for the Irish Republican Army during this timeperiod.

Irish Independence.4 Through the legislation, all rebels who had participated in the Easter Rising

in 1916 were eligible to apply for a pension.

I leverage the extensive paper-trail generated through the pension application process to con-

struct a new individual-level dataset. The pension application process started with an initial

submission in which ex-combatants detailed their involvement in distinct periods of the conflict.

They next acquired supporting materials from fellow rebels who served as references. Finally, the

pension board considered the amount of service claimed, the evidence presented in favor of the

applicant in their sworn testimony and by their references, and decided whether and how much

service to approve for each period. The end result of this process was a Military Service Certificate,

which detailed (1) the organization in which an individual claimed service in each conflict period,

and (2) the fraction of the period for which an individual was approved. An example of a Military

Service Certificate is presented in Table 1.

2.1 Defining the Sample: Easter Rising Participants

Since I am interested in comparing among a group of rebels who decided to fight at conflict onset,

I define my sample to include all individuals who were approved for active service as a member of

either the Irish Volunteers or Irish Citizen Army during the Easter Rising in 1916. Returning to

4For more on the pension application process, see Appendix Sections 1 and 3.

3

the Military Service Certificate presented in Table 1, in order for an individual to be considered

part of my sample, an individual must claim to have been either a member of the “Irish Volunteers”

or “Irish Citizen Army” (first cell of the second column) and also have been approved for service

during Easter Week (first cell of the third column). I use the information conveyed in the remainder

of the Military Service Certificate to learn about rebel behavior throughout the remainder of the

conflict.

Focusing on individuals who participated in the Easter Rising has three main benefits. First,

focusing on participants in the Easter Rising means that all individuals we will be comparing,

regardless of whether they were eventually arrested and imprisoned, were willing to engage in

a violent uprising against the British government with the odds overwhelmingly stacked against

them. Participating in the Easter Rising was a choice, and thus focusing on only individuals who

went down this path helps address the potential concern that there was variation in the baseline

willingness of individuals to participate in violent conflict which affected both the likelihood of being

arrested and of continuing to fight. We are only comparing among individuals who were willing

to fight from the outset. I return to the more general concern of selection into imprisonment in

Section 2.4 and Appendix Section 7. Second, focusing on participants in the Easter Rising helps

ensure that all individuals within the sample could (1) plausibly be arrested following the uprising,

and (2) rejoin the conflict in subsequent periods. This ensures that the analyses exclude individuals

who joined the conflict during a period where it was not possible for them to be interned. Finally,

the requirement that individuals were approved for service helps screen out individuals who might

have over-claimed or misrepresented the extent of their service. In order to be approved for service,

rebels had to successfully document their service, have fellow rebels vouch for them, and have all

of this documentation vetted by the pension board.

2.2 The Dependent Variable: Continuing to Fight Through the Irish War of

Independence

The main dependent variable in this letter is whether individuals who participated in the Easter

Rising subsequently fought throughout the duration of the Irish War of Independence. This is

measured from whether individuals claimed to serve throughout each of the four periods comprising

the Irish War of Independence as measured though the service certificates. Returning to Table 1,

4

this information is conveyed in Rows 4-7 of Column 2. The dependent variable is binary, taking on

a one when individuals claim to have served in all of the periods in the Irish War of Independence,

and a zero otherwise. In Appendix Section 4 I show that the results are substantively similar when

utilizing alternative specifications for the dependent variable, such as whether rebels were approved

for service.

2.3 The Explanatory Variable: Imprisonment After the Rising

The main explanatory variable is whether rebels were interned after the Easter Rising. This in-

formation is collected directly from the pension application in one of two places. The first place

is in the sworn statements of rebels. In these statements rebels provided detailed information on

their involvement in the Easter Rising, whether they were arrested and interned after the Rising,

and their subsequent conflict participation. For individuals for whom a sworn statement is not on

file, the information is then sought in the statements made by applicants’ references. Given that

applicants had an incentive to state that they were arrested and imprisoned since they would be

paid for service during this time period, I assume that if this information is missing, they were not

interned.

2.4 Control Variables Addressing Selection into Imprisonment

Empirically assessing how internment shaped subsequent conflict behavior necessitates first consid-

ering why rebels were imprisoned in the first place. I collect a range of additional individual-level

information to control for the main factors historical research suggests could confound the relation-

ship between imprisonment and continued fighting.

The first factor I control for is whether rebels were rank and file or in the leadership during the

Easter Rising. Rebels who are able to rise in ranks before conflict onset might be on average more

extreme than rebels of lower ranks. Higher ranking individuals might then be more likely to be

captured during the Easter Rising, and also more likely to continue fighting, regardless of whether

they were imprisoned. Given this concern, I collected individual-level information on rebels’ ranks

at the time of the Easter Rising. This variable is binary, taking the form of 1 if rebels are members

5

of the rank and file, and zero otherwise.5 Second, I collected information on whether individuals

joined the rebel organizations in either 1913 or 1914 which is either at, or close to, their foundation.

This variable is coded as a 1 if a rebel stated that they joined an Irish rebel organization in either of

these years. As with rebels’ ranks, we might expect that rebels who joined the rebel organizations

at their foundation were on average more extreme and thus more likely to be imprisoned and also

more likely to continue fighting.

Third, a range of research documents how individuals who are younger are often more “violent”

or “militant” than otherwise similar members of their organization (e.g., Urdal 2006). Given this,

throughout the analyses that follow I control for the age of combatants. Fourth, I control for whether

rebels were members of the Irish Volunteers or Irish Citizen Army. While these organizations were

largely united in their goal of fighting for a united Irish Republic, they did have important differences

in both ideology and organizational structure which could affect both the likelihood that rebels were

captured during the Easter Rising, and also that they continued fighting as the conflict unfolded.

Fifth, I control for whether rebels left early during Easter Week. Some individuals state that as

the week unfolded they became separated from their unit, and thus returned to their homes early.6

This would then decrease the likelihood that they were arrested at some point during Easter Week.

We might also be concerned that these types of individuals would be less likely to fight in the

future. Given this, I collected information on whether rebels left early during Easter Week and

control for this in the main analyses.

Finally, I include fixed effects for rebels’ last location of fighting during Easter Week. As Easter

Week unfolded, some locations became surrounded by British soldiers while others were locations

of fierce fighting. Moreover, on Saturday April 29th 1916 when Patrick Pearse issued an order to

surrender to avoid further loss of life, there were differences in whether rebels had the opportunity

to evade capture. Given the possibility that (a) rebels who were perceived to be more extreme or

committed might have been sent to more strategically vital positions which made them more likely

to be caught, and (b) the variation in capture that qualitative evidence seems to suggest occurred

is conditional on rebels having been in a particular location, location fixed effects allow me to make

5In Appendix Section 6.1, I consider alternative ways of coding rebels’ ranks. Results are substantively similarregardless of which approach is used.

6For example, in his pension application Francis Brady states that he stopped fighting on April 27th because he was“surrounded and cut off . . . and could not rejoin unit” (Brady 1935: 10).

6

comparisons among individuals who were fighting at the same last location during Easter Week

at the same time. The core assumption for estimating the effect of imprisonment is that after

controlling for these factors, rebels who successfully evaded capture are comparable to those who

were arrested and interned.

Qualitative evidence from sworn statements and witness testimony provides evidence consistent

with this requirement. While some rebels were ordered to surrender, others were instructed by

their commanding officer to evade capture if possible. For example Major MacBride, one of the

commanding officer at Jacob’s Biscuit’s Factory, is reported to have stated that “Liberty is a sweet

thing. . . any of you that sees a chance, take it” (Townshend 2006: 250). In his sworn statement as

part of his military pension application, Peter Dolan—a member of the Irish Volunteers who was

at Jacob’s Biscuit Factory during the Easter Rising—documents the consequences of this order.

When asked whether he surrendered after the rising, Dolan responded that “No, we got orders that

if any man could get away, to get away” (Dolan 1936: 29). Given the qualitative information that

individuals at Jacob’s Biscuit Factory were explicitly instructed to evade capture if they could, a

secondary set of analyses presented below focuses only on individuals in this location. Importantly,

the statements of applicants’ references suggest that those who evaded capture were similarly

committed as those who were captured and subsequently interned. For example, the commanding

officer and pension application reference for William Bruen, Joseph O’Byrne, noted in his reference

that “I wish to make it clear however that his service was complete and loyal during the whole of

the week and that his departure took place at a time 2 hours or so after the fighting was declared

at an end” (Bruen 1936: 23).

3 Main Result: Formerly Imprisoned Rebels Continue Fighting

at Higher Rates

Table 2 presents regression results to investigate the relationship between internment and fight-

ing throughout the Irish War of Independence. All models use OLS. The first column presents

results using a minimal set of controls, including only whether rebels were members of the Irish

Citizen Army or Irish Volunteers. Rebels who were sent to internment camps were approximately

8.3 percentage points more likely to continue fighting throughout the duration of the Irish War of

7

Independence when compared with individuals who were not interned. The second column presents

results using the full set of control variables, and we again observe substantively similar and statis-

tically significant results. Columns 3 and 4 present regression results focusing only on the subset

of individuals who stated they fought at Jacob’s Biscuit Factory—a location we know commanding

officers told rebels to evade capture if they could. Column 3 once again presents results using

a minimal set of controls, and we see that rebels who were sent to internment camps were over

sixteen percentage points more likely to continue fighting when compared within individuals who

evaded capture. When including the full set of control, the results are again substantively similar

and statistically significant.

In Appendix Section 7 I also present sensitivity analyses. I show how a confounder explaining

ten times the residual variance as is explained by Irish rebels’ ages (in internment and the outcome)

would still not reduce the implied effect size to zero. Substantively, this difference is analogous to a

ten year increase in rebels’ ages (i.e., taking a 20 year old and making them 30). This comparison

is helpful since age is a statistically-significant predictor for joining a rebel group (Humphreys and

Weinstein 2008). Similarly, over 95% of Palestinian suicide bombers are under 35 (Berrebi 2007: 21),

while over 86% of ETA members were under 30 at the time of their recruitment (Reinares 2004: 474).

Given the strength of the theorized relationship between age and conflict behavior, this implies that

whether we have fully eliminated confounding or not, a relatively high degree of confounding would

be required to change our estimate substantially.

First-hand accounts of Irish rebel combatants combined with the rich historical literature on

the conflict in Ireland show how Irish rebels’ prison experiences increased grievances toward the

British. Indeed, the radicalizing nature of imprisonment has led the historical literature to dub

the main internment camp “The University of Revolution” (Mahony 1987). Grievance-inducing

factors included the overall poor conditions of the prisons, the quality and quantity of rations, and

also British efforts to conscript members of the Irish Volunteers to fight in British forces in WWI

(Brennan-Whitmore 2013: 68–69). Both the prison conditions and British actions increased anger

among the Irish rebel prisoners. For example, in an autobiography describing his experience as

an Irish Volunteer in the Frongoch internment camp, William Brennan-Whitmore states that the

British efforts to conscript two members of the Irish Volunteers “angered us very much” and the

rebels’ “impotence in the matter but added fuel to our fury” (Brennan-Whitmore 2013: 82–83).

8

Table 2 – Internment and Fighting Through the Irish War of Indepen-dence. Examines the fighting rates of members of the Irish Volunteers and IrishCitizen Army. Rebels who were interned fought throughout the Irish War of Inde-pendence at higher rates than those who were not interned.

Dependent Variable:

Fighting Through War of Independence

(1) (2) (3) (4)

Interned 0.083∗∗∗ 0.115∗∗∗ 0.167∗∗ 0.155∗

(0.024) (0.032) (0.079) (0.088)

Irish Citizen Army −0.117∗∗ −0.058 −0.287 −0.244(0.047) (0.077) (0.190) (0.207)

Rank and File −0.094∗∗∗ −0.100(0.029) (0.091)

Age −0.005∗∗ 0.0003(0.002) (0.006)

Left Fight Early −0.143 0.427(0.093) (0.358)

Join Org Early 0.048 0.066(0.030) (0.087)

Constant 0.597∗∗∗ 1.075∗∗∗ 0.478∗∗∗ 0.512∗∗∗

(0.019) (0.339) (0.058) (0.175)

Sample Full Full Limited LimitedLocation FE No Yes No NoObservations 1,776 1,537 157 145

Notes: OLS regression with standard errors in parentheses. Outcome is binary, with a 1 indicating a

rebel fought from conflict onset through the Irish War of Independence. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

9

Rebel grievances eventually led to hunger strikes by the Irish rebel prisoners (Murphy 2014: 64).

Ultimately, the qualitative evidence suggests that both the actions of British prison officials and the

overall quality of the prisons increased grievances which in turn shaped subsequent rebel behavior.

4 Implications

In this letter I provided new individual-level data and evidence for how rebels’ within-conflict

experiences shaped their decision to continue fighting. In doing so, the letter contributes to prior

research on the determinants of why individuals rebel (Cederman, Gleditsch, and Buhaug 2013;

Gurr 1970; Paige 1978; Wood 2003) by demonstrating how both government actions and rebel

grievances matter at multiple junctures within violent conflict settings. The finding that rebel

grievances can increase while conflicts unfold highlights how rebel organizations are not static

entities, but instead can become more extreme over time due to increasing individual grievances.

This is important since prior research argues that the relatively more extreme members of rebel

organizations are those most likely to undermine peace settlements and continue fighting (Kydd

and Walter 2002; Stedman 1997). The findings also reveal a trade-off for those charged with fighting

rebellions in that the tactics that might be most effective for combating rebellions might contribute

to the intractability of the conflict in the long-term. At both its theoretical and empirical core,

the successful resolution of violent conflict hinges upon individuals deciding to stop fighting. This

letter seeks to better understand this decision by highlighting how the actions governments take as

they fight can shape the behavior of the individuals they are combating.

10

References

Balcells, Laia. 2017. Rivalry and Revenge: The Politics of Violence During Civil War. CambridgeUniversity Press.

Berrebi, Claude. 2007. “Evidence About the Link Between Education, Poverty and TerrorismAmong Palestinians.” Peace Economics, Peace Science and Public Policy 13(1): 1–36.

Brady, Francis. 1935. “File Reference MSP34REF1512.” Bureau of Military History, 1913–1921.

Brennan-Whitmore, William James. 2013. With the Irish in Frongoch. Mercier Press.

Bruen, William. 1936. “File Reference MSP34REF15319.” Bureau of Military History, 1913–1921.

Cederman, Lars-Erik, Kristian Skrede Gleditsch, and Halvard Buhaug. 2013. Inequality,Grievances, and Civil War. Cambridge University Press.

Collier, Paul, and Anke Hoeffler. 2004. “Greed and Grievance in Civil War.” Oxford EconomicPapers 56(4): 563–595.

Dolan, Peter. 1936. “File Reference MSP34REF20113.” Bureau of Military History, 1913–1921.

Elkins, Caroline. 2005. Imperial Reckoning: The Untold Story of Britain’s Gulag in Kenya. Macmil-lan.

Fearon, James D, and David D Laitin. 2003. “Ethnicity, Insurgency, and Civil War.” AmericanPolitical Science Review 97(01): 75–90.

Gurr, Ted Robert. 1970. Why Men Rebel. Princeton University Press.

Henn, Soeren, and Connor Huff. 2021. “The Legacies of Atrocities and Who Fights.” WorkingPaper.

Humphreys, Macartan, and Jeremy M. Weinstein. 2008. “Who Fights? The Determinants ofParticipation in Civil War.” American Journal of Political Science 52(2): 436–455.

Kydd, Andrew, and Barbara F Walter. 2002. “Sabotaging the Peace: The Politics of ExtremistViolence.” International Organization 56(02): 263–296.

Mahony, Sean O. 1987. Frongoch: University of Revolution. Dublin: FDR Teoranta.

Murphy, William. 2014. Political Imprisonment and the Irish, 1912-1921. Oxford University Press.

Paige, Jeffrey M. 1978. Agrarian Revolution. Simon and Schuster.

Reinares, Fernando. 2004. “Who Are the Terrorists? Analyzing Changes in Sociological ProfileAmong Members of ETA.” Studies in Conflict and Terrorism 27(6): 465–488.

Schubiger, Livia. 2021. “One for All? State Violence and Insurgent Cohesion.” InternationalOrganization Forthcoming.

Stedman, Stephen John. 1997. “Spoiler Problems in Peace Processes.” International Security 22(2):5–53.

Townshend, Charles. 2006. Easter 1916: The Irish Rebellion. Penguin.

11

Urdal, Henrik. 2006. “A Clash of Generations? Youth Bulges and Political Violence.” InternationalStudies Quarterly 50(3): 607–629.

Wood, Elisabeth. 2003. Insurgent Collective Action and Civil War in El Salvador. CambridgeUniversity Press.

12

Online Appendix for “CounterinsurgencyTactics, Rebel Grievances, and Who Keeps

Fighting”

Contents

1 Pension Waves and Constructing the Sample A2

2 The Empirical Implications of the Sample Being Defined Post-Conflict A4

3 Measuring Conflict Behavior From the Military Pensions Collection A73.1 Measuring the Dependent Variable: Military Pension Certificates . . . . . . . A73.2 Measuring the Independent Variable: Imprisonment Status . . . . . . . . . . A9

4 Estimating Effects with Dependent Variable Approved for Service A11

5 Logit Regression Results A13

6 Alternative Covariate Specifications A146.1 Alternative Coding for Individual Rank . . . . . . . . . . . . . . . . . . . . . A146.2 Alternative Coding for Where Individuals Fought During the Easter Rising . A17

7 Selection into Imprisonment and Sensitivity Analyses A20

A1

1 Pension Waves and Constructing the Sample

The initial sample is defined as all individuals who were approved for service in the Easter

Rising by the military pension board. The unique application series, the total number

of individuals approved through each process who fought in the Easter Rising, and the

substantive meaning of the pension series is documented in Table A1.1 As is demonstrated

in the top two rows, the overwhelming bulk of these individuals were approved for service

through either the 1924 or 1934 military pension application processes. This set of individuals

forms the main sample used for analysis, and comprises 2,242 individuals of the 2,577 total

number of individuals approved for service in the Easter Rising. The 335 individuals who

did not receive a military pension through either the 1924 or 1934 pension waves were then

dropped from the sample used for the main analyses (those in row 3 and all subsequent

rows). This is done because according to the military archives, many of these applications

processes were ‘open to greater abuse” than that of the pension applications, the latter of

whose applicants had to “undergo a rigorous and time-consuming investigation procedure”

(Coleman 2016: 20–21). Additionally, these alternative application processes did not generate

the same paper trail necessary for collecting the individual-level data to learn about conflict

participation.

1The most comprehensive details of the respective pension files is presented online here:http://www.militaryarchives.ie/collections/online-collections/military-service-pensions

-collection-1916-1923/about-the-collection/content-of-the-collection.

A2

Pension Series Individuals Approved Substantive Meaning

MSP34 1797 1934 Pension24SP 445 1924 Pension

MD 130 Medal (1941), No PensionD 99 Killed in Action

DP 65 Disability Pension Including Dependents Allowance1924A 19 Military Service Board (Proof of Service for Wound)

P 8 Personal Claims for Wound Pension2RB 1 Military Service Board (Proof of Service for Wound)34E 434D 2

49SP 2340E 134A 134C 1

34SP 1NF1 1

Table A1: The total number of individuals approved for service in the 1916 Easter Rising.The analysis presented in this paper focuses on the 2242 individuals approved for service ineither 1924 or 1934 (the top two rows).

A3

2 The Empirical Implications of the Sample Being De-

fined Post-Conflict

Astute readers will note that since the pension application process did not occur until af-

ter the conflict ended, the sample is defined post-conflict. This raises at least two major

challenges for empirical analyses. The first challenge is driven by the potential that those

who were approved for military service in the Easter Rising through the pension application

system are different than the “true” population of those who fought. If these differences are

correlated with both treatment and outcome, then we might have a biased estimate of the

effect of imprisonment. Similarly, the second challenge is driven by the potential peculiarities

of the pension system itself. If there were differences in the application processes, such as

the information collected across pension waves, then it could be the case that the results we

observe are due to these differences rather than the actual effect of imprisonment. I now

turn to addressing each of these challenges in turn.

On the first point, the differences between combatants who were approved for service

through the pension system and the true population of conflict participants is a potential

empirical concern if (1) the amount of “missingness” is large, and (2) not applying is corre-

lated with treatment. Historical evidence and the statements of archivists suggest that the

sample of individuals approved for military service in the Easter Rising, as was documented

in Table A1, comprises a nearly exhaustive list of those who fought in the 1916 Easter Rising.

As Pat Brennan a senior military archivist stated, “we’re pretty sure this is the definitive

figure.”2 Likewise, when writing about the release of the pension files, The Irish Times

writes that “the only known absentee from the list is Cathal Brugha.”3 These statements

suggest that we should be quite skeptical that the amount of missingness, if there is any, is

2Brennan is quoted in the newspaper the The Irish Times in an article discussing how many individualswere “out” during the Easter Rising. For more, see https://www.irishtimes.com/culture/heritage/

how-many-were-really-out-for-easter-rising-1916-1.2261362.3This statement comes from the same article discussed above. Again, see https://www.irishtimes

.com/culture/heritage/how-many-were-really-out-for-easter-rising-1916-1.2261362.

A4

so large as to substantively affect the results.

On the second point, we might be concerned that differences in the amount or quality

of information available across the pensions waves is substantively affecting results. If,

for example, individuals who applied for military pensions under the 1924 pension act are

simply less likely to mention that they were imprisoned and continued fighting, then this

could potentially lead to biases in our estimates. To partially assuage this concern, I re-

estimate the main results from the paper with a dummy variable for whether individuals

applied under the 1934 pension wave. Table A2 presents results substantively similar to

those presented in the main text of the paper.

A5

Table A2: Internment and Fighting in the Irish War of Independence Among 1934Pension Applicants. Examines the fighting rates of members of the Irish Volunteers andIrish Citizen Army among individuals who applied for pensions under the 1934 act. Rebelswho were interned fought throughout the Irish War of Independence at higher rates thanthose who were not interned.

Fighting Through War of Independence

Interned 0.080∗∗∗ 0.109∗∗∗

(0.024) (0.032)

Irish Citizen Army −0.100∗∗ −0.026(0.047) (0.076)

Rank and File −0.090∗∗∗

(0.029)

Age −0.004∗

(0.002)

Left Fight Early −0.134(0.092)

Join Org Early 0.066∗∗

(0.030)

1934 Pension Applicant −0.156∗∗∗ −0.173∗∗∗

(0.027) (0.036)

Constant 0.718∗∗∗ 1.215∗∗∗

(0.028) (0.337)

Location FE No YesObservations 1,776 1,537

+p<0.1; ∗p<0.05; ∗∗p<0.01

Notes: OLS regression with standard errors in parentheses. Out-come is binary, with a 1 indicating a rebel fought from conflictonset through the Irish War of Independence.

A6

3 Measuring Conflict Behavior From the Military Pen-

sions Collection

The data used in this paper is gathered from the Military Archives of the Defence Forces

Ireland.4 There are two main sources of individual-level information. The first source entails

the individual pension applications themselves. Each of these applications range in length

from only a few pages to several hundred.5 Within each pension application, I collect infor-

mation for my dependent variable from the Military Service Certificate, and information for

my main independent variable from the sworn statements of applicants and their references.

I discuss each of these sources in more detail in Sections 3.1 and 3.2. The second source

entails the metadata collected by the archivists from the pension applications. This meta-

data provides a host of background covariate information such as the organization, rank, and

brigade of individuals. This material was collected from the website of the Defence Forces

Ireland.

3.1 Measuring the Dependent Variable: Military Pension Certifi-

cates

Figures A1 and A2 present the Military Service Certificates under the 1934 and 1924 pension

acts. Three points are worth noting. First, across both the 1924 and 1934 pension acts, the

conflict is discretized into nearly identical time periods. This is invaluable, as it means that

I am able to pool across the two pension acts in a way that allows for direct comparability

in the fighting rates between distinct pools of individuals. Second, recall that the first main

dependent variable used in this paper is whether individuals “claimed” to fight throughout

the entirety of the Irish War of Independence. This “claimed” variable corresponds with the

4See the website here: http://www.militaryarchives.ie/en/home/.5These longer applications are generally either the result of individuals claiming for disability due to

injuries incurred during the conflict or their dissatisfaction with the amount of service for which they wereapproved (leading them to appeal).

A7

second column across both the 1924 and 1934 military pensions. Third, we can leverage the

fact that, while recorded slightly differently, the time periods corresponding with fighting

throughout the entirety of the Irish War of Independence are identical across the pension

waves. In the 1934 service certificate, we will code an individual as a 1 if they claimed to

fight in periods 3, 4, 5, 6, and 7, which substantively corresponds with rejoining the Irish

Volunteers (or Irish Citizen Army) in the period immediately following the release of the

prisoners, up through the beginning of the ceasefire on July 11, 1921. Readers will note that

the day the ceasefire began corresponds with the last day of period 7. In the 1924 service

certificate, we will code an individual as a 1 of they claimed to fight in periods b, c, d, e,

and f, which corresponds to the identical date windows in the 1934 service certificate.

Figure A1: Military Pension Certificate under the 1934 Military Service Act. Individualrebels are coded as a 1 of they claimed to fight in periods 3, 4, 5, 6, and 7.

A8

Figure A2: Military Pension Certificate under the 1924 Military Service Act. Individualrebels are coded as a 1 of they claimed to fight in periods b, c, d, e, and f.

3.2 Measuring the Independent Variable: Imprisonment Status

As noted in the main text of the paper, the main explanatory variable is individuals being

interned after the 1916 Easter Rising. This information is collected directly from the pension

application in one of two places. The first source of information is in rebels’ sworn statements

about their involvement in the conflict. The second source is information provided by appli-

cants’ references. An example of a rebel’s sworn statement and the statement of a reference,

is provided from the pension application of Louis McEvatt.6 These are depicted visually in

Figures A3 and A4. We can see that McEvatt explicitly states that he was imprisoned and

deported following the Easter Rising in 1916. He then provides three references who can

testify to that effect. We go on to see that an individual named Michael Laffan vouches for

6See the military pension file for Louis McEvatt (MSP34REF4041).

A9

McEvatt’s statement, telling us that (1) he was indeed interned in Frongoch, and (2) that

he rejoined his company and continued fighting upon his release.

Figure A3: Statement of Louis McEvatt about his imprisonment status. Note that he statesthat he was imprisoned following his participation in the Easter Rising in 1916, was deportedto Stafford and Frongoch jails, and was released in December 1916. He also provides thenames and addresses of three individuals who can testify to his statements.

Figure A4: Sworn statement from Michael Laffan about the service of Louis McEvatt abouthis imprisonment status. We can see that Laffan affirms that McEvatt was indeed impris-oned, and rejoined his Company upon his release.

A10

4 Estimating Effects with Dependent Variable Approved

for Service

The main outcome variable used in the paper is whether individuals claimed to serve through-

out the entirety of the Irish War of Independence. An alternative outcome measure is whether

individuals were approved for service by the military pension boards. While the effect size

is attenuated, the results are substantively similar to those presented in the manuscript.

A11

Table A3: Approved for Service for Fighting Throughout the Irish War of Inde-pendence. Examines the fighting rates of members of the Irish Volunteers and Irish CitizenArmy using the alternative dependent variable of whether they were approved for serve.Rebels who were interned fought throughout the Irish War of Independence at higher ratesthan those who were not interned.

Fighting Through War of Independence

Interned 0.045∗∗ 0.036+

(0.015) (0.019)

Irish Citizen Army −0.081∗∗ −0.150∗∗

(0.030) (0.045)

Rank and File −0.147∗∗

(0.017)

Age 0.001(0.001)

Left Fight Early −0.051(0.054)

Join Org Early −0.038∗

(0.018)

Constant 0.075∗∗ 0.542∗∗

(0.012) (0.197)

Location FE No YesObservations 1,776 1,537

+p<0.1; ∗p<0.05; ∗∗p<0.01

Notes: OLS regression with standard errors in parentheses. Out-come is binary, with a 1 indicating a rebel was approved for ser-vice for fighting from conflict onset through the Irish War of In-dependence.

A12

5 Logit Regression Results

Some analysts might prefer to use Logit rather than OLS given the binary nature of the

outcome variable used throughout the manuscript. Table A4 presents this, showing that

results are substantively and statistically similar.

Table A4: Internment and Fighting Through the Irish War of Independence.Examines the fighting rates of members of the Irish Volunteers and Irish Citizen Army.Rebels that were interned fought throughout the Irish War of Independence at higher rateswhen compared to individuals who were not interned.

Fighting Through War of Independence

Interned 0.358∗∗ 0.545∗∗

(0.103) (0.145)

Irish Citizen Army −0.492∗ −0.233(0.200) (0.340)

Rank and File −0.467∗∗

(0.141)

Age −0.023∗∗

(0.009)

Left Fight Early −0.638(0.405)

Join Org Early 0.231+

(0.139)

Constant 0.392∗∗ 18.978(0.083) (4,546.167)

Location Fe No YesObservations 1,776 1,537

+p<0.1; ∗p<0.05; ∗∗p<0.01

Notes: Logit regression with standard errors in parentheses. Out-come is binary, with a 1 indicating a rebel fought throughout theduration of the Irish War of Independence.

A13

6 Alternative Covariate Specifications

Given the structure of the pension application system and nature of the data, there are

alternative approaches for coding two of my control variables. In the remainder of this

section I discuss possible alternatives for coding information about rebels’ (1) ranks, and (2)

location of fighting.

6.1 Alternative Coding for Individual Rank

There are at least two main alternative approaches for coding the rank of rebel combatants.

In the first and simplest approach, we can instead use rebel rank fixed effects which simply

leverage the stated rank of rebels. Results for each of the main outcome variables with

rank fixed effects are presented in Table A5. While I prefer the binary approach since it is

not clear that there are theoretically meaningful differences among the more highly-ranked

individuals (where we also have a small number of individuals per category), results are

substantively similar to those presented in the main body of the manuscript.

The second approach considers alternative ways of dealing with the individuals whose

ranks are listed as “Unknown.” Rebels’ ranks are specified in such a way either because

(a) information about rebels’ ranks was not provided in the pension file, or (b) rebels were

“between” ranks, and thus the archivists were unable to clearly categorize them of being

in one rank or the other.7 Given this, Table A6 presents the results using an alternative

specification including a dummy variable for if the rank is listed as “Unknown.” Again,

results are substantively similar to those presented in the manuscript.

7This coding information was learned from direct conversations with archivists at the Military Archives.

A14

Table A5: Internment and Fighting Through the Irish War of Independence.Examines the fighting rates of members of the Irish Volunteers and Irish Citizen Army.Rebels that were interned for longer periods of times fought throughout the Irish War ofIndependence at higher rates when compared to individuals who not interned.

Fighting Through War of Independence

Interned 0.072∗∗ 0.116∗∗

(0.024) (0.032)

Irish Citizen Army −0.083 0.026(0.065) (0.092)

Age −0.004∗

(0.002)

Left Fight Early −0.128(0.093)

Join Org Early 0.043(0.031)

Constant 0.928+ 1.109(0.476) (0.750)

Rank FE Yes YesLocation FE No YesObservations 1,758 1,522

+p<0.1; ∗p<0.05; ∗∗p<0.01

Notes: OLS regression with standard errors in parentheses. Out-come is binary, with a 1 indicating a rebel fought throughout theduration of the Irish War of Independence.

A15

Table A6: Internment and Fighting Through the Irish War of Independence.Examines the fighting rates of members of the Irish Volunteers and Irish Citizen Army.Rebels who were interned fought through the Irish War of Independence at higher rateswhen compared to individuals who were not interned.

Fighting Through War of Independence

Interned 0.076∗∗ 0.115∗∗

(0.024) (0.032)

Irish Citizen Army −0.121∗ −0.059(0.047) (0.077)

Rank and File −0.105∗∗ −0.110∗

(0.037) (0.045)

Rank Unknown 0.002 −0.023(0.043) (0.051)

Age −0.005∗

(0.002)

Left Fight Early −0.143(0.093)

Join Org Early 0.047(0.030)

Constant 0.673∗∗ 1.096∗∗

(0.039) (0.342)

Location FE No YesObservations 1,776 1,537

+p<0.1; ∗p<0.05; ∗∗p<0.01

Notes: OLS regression with standard errors in parentheses. Out-come is binary, with a 1 indicating a rebel fought throughout theduration of the Irish War of Independence.

A16

6.2 Alternative Coding for Where Individuals Fought During the

Easter Rising

There are at least two main alternative approaches for considering how to code the location

of fighting during the Easter Rising. In the first approach, we could use the starting location

rather than ending location for rebels during Easter Week. We might imagine that rebels’

initial location assignments were in part a function of their baseline levels of “extremism”

and thus, some analysts might be more likely to think that this will better capture the

selection process leading some rebels to be imprisoned rather than others. However, there

is an empirical downside with this approach driven by the fact that rebels listed a wide

range of starting locations (over 700 in total). This large number means that when we

include starting location fixed effects, there are an extremely small number of individuals

within each location. Results for this approach are presented in Table A7. Results are

substantively similar and statistically significant.

In the second approach, we could use whether rebels were located within the city of

Dublin or outside of it. As Easter Week unfolded, a larger number of individuals would rise

up, and the fighting would be most intense within the city of Dublin. We would thus expect

individuals who fought in Dublin to have a much higher likelihood of being captured than

individuals located outside the city.8 Moreover, given that a smaller number of individuals

from outside of Dublin decided to participate, we might expect those that did to be on

average more extreme than individuals located within the city itself. The downside of this

approach is we are no longer able to capture the process through which individuals were

able to evade capture or not (which occurred at a lower geographical unit). Results for this

specification are presented in Table A8. Again, results are substantively and statistically

similar to those presented in the manuscript.

8Individuals who state they fought in the Dublin Brigade are approximately 12% more likely to beinterned abroad.

A17

Table A7: Internment and Fighting Through the Irish War of Independence.Examines the fighting rates of members of the Irish Volunteers and Irish Citizen Army.Rebels that were interned fought in the Irish War of Independence at higher rates thanthose who were not interned.

Fighting Through War of Independence

Interned 0.091∗∗ 0.077∗

(0.027) (0.031)

Irish Citizen Army −0.083 −0.066(0.065) (0.070)

Rank and File −0.103∗∗

(0.030)

Age −0.004∗

(0.002)

Left Fight Early −0.148+

(0.081)

Join Org Early 0.083∗∗

(0.030)

Constant 0.909+ 1.037∗

(0.478) (0.479)

Starting Location FE No YesObservations 1,772 1,537

+p<0.1; ∗p<0.05; ∗∗p<0.01

Notes: OLS regression with standard errors in parentheses. Out-come is binary, with a 1 indicating a rebel fought from conflictonset through the Irish War of Independence.

A18

Table A8: Internment and Fighting Through the Irish War of Independence.Examines the fighting rates of members of the Irish Volunteers and Irish Citizen Army.Rebels that were interned fought in the Irish War of Independence at higher rates thanthose who were not interned.

Fighting Through War of Independence

Interned 0.087∗∗ 0.091∗∗

(0.024) (0.027)

Irish Citizen Army −0.132∗∗ −0.111∗

(0.049) (0.054)

Rank and File −0.101∗∗

(0.026)

Age −0.003∗

(0.002)

Left Fight Early −0.124+

(0.070)

Join Org Early 0.061∗

(0.027)

Dublin Brigade −0.027 −0.015(0.024) (0.026)

Constant 0.610∗∗ 0.712∗∗

(0.023) (0.055)

Location FE No NoObservations 1,776 1,537

+p<0.1; ∗p<0.05; ∗∗p<0.01

Notes: OLS regression with standard errors in parentheses. Out-come is binary, with a 1 indicating a rebel fought from conflictonset through the Irish War of Independence.

A19

7 Selection into Imprisonment and Sensitivity Analy-

ses

While the qualitative accounts presented in the main letter provide some suggestive evidence

that there was random variation in who successfully evaded capture, we might still be con-

cerned that there were differences in the types of individuals who sought to evade capture.

If the types of individuals who did not evade capture (and thus were interned) are also on

average more likely to continue fighting at all phases of the conflict, then it could be the

case that the observed results are due to these differences in baseline types rather than the

effect of imprisonment. The task at hand is thus to consider how much selection must occur

in order to explain away the main results presented in this paper. I do so using a sensitivity

analysis approach (Blackwell 2014; Cinelli and Hazlett 2018).

Following Cinelli and Hazlett (2018), I focus on two main approaches. The first, presented

in Table A9 shows the Robustness Value. The Robustness Value tells us “the minimum

strength of association unobserved confounding would need to have, both with the treatment

and with the outcome, to change the research conclusions” (Cinelli and Hazlett 2018: 1).

We see that unobserved confounders that explain more than 9.73% of the residual variance

of both the treatment and outcome are enough to reduce the absolute value of the effect

size by 100%. For reference, the size of the Robustness Values presented in this paper

are comparable to recent work in political science adopting a similar sensitivity approach

(Hazlett and Mildenberger 2019: 20).

Table A9: Sensitivity statistics for regression of internment on fighting throughout the en-tirety of the Irish War of Independence. Model uses OLS and the full set of control variables.

Outcome: Fighting Throughout War of Independence

Treatment Est. SE t-value R2Y∼D|X RV df

Internment 0.115 0.032 3.619 1.04% 9.73% 1250

For the second approach, I focus on presenting the proportion of variation in the outcome

A20

explained uniquely by treatment, which tells us “how strongly confounders that explain 100%

of the residual variance of the outcome would have to be associated with the treatment in

order to eliminate the effect” (Cinelli and Hazlett 2018: 2). In order to ease interpretability

for the this second metric, we will benchmark it relative to the residual variance in outcome

explained by rebels’ ages, a variable which will be substantively meaningful to many schol-

ars of conflict and rebellion. A range of prior empirical research documents how younger

individuals are often more violent and risk prone (Berrebi 2007; Humphreys and Weinstein

2008; Reinares 2004). Figure A5 demonstrates that a confounder explaining ten times the

residual variance as is explained by age (in internment and the outcome) would still not

reduce the implied effect size to zero. Substantively, this difference is analogous to a ten

year increase in rebels’ ages (i.e., taking a 20 year old and making them 30). Given the

strength of the theorized relationship between rebels’ ages and their conflict behavior, this

implies that whether we have fully eliminated confounding or not, a relatively high degree

of confounding would be required to change our estimate substantially.

A21

Hypothetical partial R2 of unobserved confounder(s) with the treatment

Hyp

othe

tical

par

tial R

2 of u

nobs

erve

d co

nfou

nder

(s)

with

the

outc

ome

−0.45 −0.4

−0.35

−0.3

−0.25

−0.2

−0.15

−0.1

−0.05

0.05 0.1

0.0 0.1 0.2 0.3 0.4

0.0

0.1

0.2

0.3

0.4

0 Unadjusted

(0.12)

5x age(0.08)

10x age(0.03)

15x age(−0.01)

Figure A5: Sensitivity analysis for the estimated effect of internment on fighting throughoutthe Irish War of Independence, using OLS regression controlling for age, rank, whether rebelsleft early, the organization they fought in, year joined, and last location of fighting. Thehorizontal axis specifies a hypothesized strength of association between confounding andthe treatment (whether rebels are interned following the Easter Rising), in terms of thepartial variance in internment after accounting for covariates. The vertical access showshypothetical values of confounding related to the outcome (fighting throughout the durationof the Irish War of Independence) in terms of the partial variance explained. The countoursdemonstrate the the adjusted effect implied by each hypothesized level of confounding. The“Unadjusted” conventional estimate is depicted in the bottom left corner, and assumes thatthere is no confounding. Let us assume that confounding can explain up to 10 times as muchresidual variance (in both interment and continuing to fight) as is explained by age. Evenif such a strong confounder exists, it would imply that our adjusted effect size is the onemarked by 10x age on the plot.

A22

References

Berrebi, Claude. 2007. “Evidence About the Link Between Education, Poverty and TerrorismAmong Palestinians.” Peace Economics, Peace Science and Public Policy 13(1).

Blackwell, Matthew. 2014. “A Selection Bias Approach to Sensitivity Analysis for CausalEffects.” Political Analysis 22(2): 169–182.

Cinelli, Carlos, and Chad Hazlett. 2018. “Making Sense of Sensitivity: Extending OmittedVariable Bias.” Journal of the Royal Statistical Society: Series B (Statistical Methodology).

Coleman, Marie. 2016. “The Military Service (1916-1923) Pension Collection: The MedalSeries.”. Oglaigh na hEireann chapter Service Medals and Special Allowances.

Hazlett, Chad, and Matto Mildenberger. 2019. “Wildfire Exposure Increases Pro-ClimatePolitical Behaviors.” Working Paper.

Humphreys, Macartan, and Jeremy M. Weinstein. 2008. “Who Fights? The Determinantsof Participation in Civil War.” American Journal of Political Science 52(2): 436–455.

Reinares, Fernando. 2004. “Who Are the Terrorists? Analyzing Changes in SociologicalProfile Among Members of ETA.” Studies in Conflict and Terrorism 27(6): 465–488.

A23