Appendix: Supporting Materials
Table S1. Support Level Index
Types
Level
Value
High
3
Safe Haven to Members or Arms
or/and any of the rest of the types
(excluding cases with high level of
support)
Moderate
2
Funds, transport of equipment
or/and any of the rest of the types
(excluding cases with high and
moderate levels of support)
Low
1
No support
None
0
Troops or Training Camps and/or
any of the rest of the types
Coding Rules for State Support
After a preliminary analysis and coding of 20% of the entire NAGs and their
supporters, we specified the following rules with respect to some ambiguous forms of
support referred in the sources used for data coding:
1. If a state provides health services to a NAGs’ members or leaders, it qualifies as
providing safe haven for members or leaders.
2. In some instances, states become hosts to negotiations and meetings between a
NAG’s leaders and the target government. This does not qualify as a form of
support.
3. Some states host the headquarters of NAGs. These headquarters organize
propaganda and fundraising activities of a NAG and provide communication with
the militants at home. This is coded as a form of support.
4. Some states host TV channels and radio stations operated by NAGs and used to
disseminate information about the NAG. This does not qualify as a form of support
for the purposes of this dissertation but can be the subject matter of another
research project.
5. A state may provide one or more of the specified support types.
Types of Support
Type 1. Safe Havens to Members: Providing safe havens to members of a NAG is different than
providing training camps or access to existing camps. More often than not, the neighbors of
a state that experiences civil war or ethnic conflict end up accepting refugees within their
own borders. Opening the borders to refugees does not qualify for providing safe havens to
an armed group that is fighting its target government unless the group is engaging in violent
cross-border attacks. The members of the Karen National Union, which has been fighting
Myanmar’s government for over five decades, frequently escape into neighboring Thailand.
They occasionally organize armed attacks back into Myanmar. The following statement
illustrates the type of evidence used to determine whether a state provides safe havens to a
NAG:
‘Thai television reported that Burma was preparing to attack Karen refugee camps inside
Thailand. Mortars reportedly were fired at one camp across the border in the Teakaplaw
region, forcing thousands of refugees to flee. The fighting comes two weeks after a
Karen splinter group supported by government launched cross-border raids against
three camps of refugees loyal to the Karen National Union inside Thailand. Two camps
were burned to the ground and 8,000 refugees fled into the Thai jungle.’
The above statement indicates that the Karen National Union has safe havens in
Thailand. Whether the support is provided intentionally by the state is discussed more
generally under type 8 in this section. In addition, multiple sources were used to determine
whether the KNU members were engaging in cross-border attacks into Myanmar.
Type 2. Safe Havens to Leaders or Hosting Headquarters: Providing safe havens to leaders of a
NAG is different than providing safe havens to its members. NAG leaders end up living in
other states due to reasons such as being expelled from their target countries or not feeling
safe in the target countries anymore. Of the total years that NAGs’ leaders spent in external
safe havens, 35% were in democratic states, in contrast to the 65% spent in autocratic states.
Despite that, democratic states might be preferred by NAGs’ leaders due to the individual
freedoms and liberties that make their arrest difficult. After the assassination of Indian
president Rajiv Gandhi, the Liberation Tigers of Tamil Eelam (LTTE) lost its support base
and funding from India. As a result, they founded offices in Western European countries,
such as Switzerland, France, and UKG, as well as USA and Canada. There is clear evidence
that the Sri Lankan government put pressure on these countries to stop the fund raising
activities of the group and to return the group’s leaders back to Sri Lanka. The United States
banned the group and its fundraising activities in 1997 by passing an anti-terrorism law and
declaring LTTE a terrorist organization. The United Kingdom and Canada did not ban
fundraising activities of the group until 2001.
Type 3. Training Camps: Providing NAGs with training camps requires extra effort on the part
of the supporters than providing safe havens. Training camps are expected to be equipped
with military equipment to help the members of a group in organizing and implementing
violent attacks against their targets. During the Syrian occupation of Lebanon from 1976 to
2005, various Palestinian groups were trained in Lebanese territories by the assistance of
Syria. For instance, the Palestinian Islamic Jihad members had trained in the camps in
Lebanon. Although the headquarters of the group had been placed in Damascus since its
foundation, the training camps are not in Syrian territories. In coding the support of PIJ by
Syria, providing training camps is not coded among the support types but providing safe
havens to leaders is coded among the types of support.
Type 4. Arms & Logistics: This level is coded if there is clear evidence that the arms originated
from the supporting country. The evidence on whether a state provides arms to NAGs is
not easily attainable. Mere allegations by the target states are not enough to prove that a state
provides arms to a NAG. In the following narration directly cited from the source, it is clear
that Libya’s giving arms to IRA was not a mere allegation by the United Kingdom:
‘Histories of the IRA have identified Mr. Murphy as an IRA weapons smuggler who
helped to procure supplies by travelling to Libya using false passports. In the 1980s,
Libya supplied the Provisional IRA with more than 100 tones of weaponry’.
Type 5. Financial Aid: While in some cases, such as Iran and Hizballah, governments provide
funds to a NAG, in many others, the groups themselves manage to raise funds within the
borders of another state, such as the Irish Republican Army (IRA) raising funds within the
USA. When this is the case, the support is assumed to be de facto, i.e. NAGs select certain
states as supporters without necessarily any intentional effort on the part of the supporter. It
is possible to argue that the USA had the capacity to control the IRA’s activities, in which
case the support of the group would have been intentional. However, making this judgment
requires a more extensive analysis of each case in the dataset, which is not an attainable goal
within the time frame of a dissertation. The specific type of support the IRA obtained from
the US is also called ‘passive support’ by Byman 2005c. I coded it as de facto support since the
US government was not intentionally creating any channels to help the group.
Type 6. Transport of Military Equipment & Supplies: If a state ends up being a transport point for
a NAG, it is coded separately from providing arms and military supplies. Cambodia has for
years become a de facto transport point for arms smuggling for many non-state armed groups
in Asia. Zaire was the major transport point for the weapons sent by the USA to UNITA,
which was fighting the communist regime in Angola.
Type 7. Troop Support: In some cases, states allowed their troops to fight on the side of the
NAGs against their targets. When civil wars or ethnic conflict crossed the borders of other
states, there is a risk that the latter acts to protect its borders. This leads to the accusations
for providing troop support to NAGs. Myanmar accused the Thai army multiple times of
providing the KNU with troops during the counter-border operations of Myanmar’s
government into the Karen National Union camps in Thailand. The following illustrates a
statement or narrative used to code troop support:
‘Angola, allied to Sassou Nguesso's Cobra militia, staged a weekend attack along the
border between its oil-producing Cabinda enclave and southwestern Congo, sending
some 1,000 troops into Congo, according to diplomats.’
Only three percent of the binary support years involve states that provide troops to
NAGs. This is normal if we consider that troop support is a very risky strategy, since it
means directly engaging with the target of a NAG. The purpose of supporting a NAG is to
avoid direct confrontation with the adversary, besides trying to undermine the power of an
adversary.
Type 8. Involuntary Support. Involuntary support includes any of the above mentioned types,
except that the supporter does not intentionally provide channels to accommodate the
NAGs seeking support. Some cases, such as Lebanon and the Palestinian groups in that
country, require an extensive examination of how the state turned into a safe haven for these
groups. Initial operations of the members of the Palestinian umbrella organization, the PLO
(Palestinian Liberation Organization), go as far back as the late 1960s. After being expelled
from Jordan, the members of the PLO found refuge in Lebanon. During 1968 and 1969, the
Lebanese security forces themselves had their own confrontations with the Palestinian
guerrilla (Maoz 2006a: 85-102; 174-183). Maoz 2006b, pp.174-183. This provides evidence
that Lebanon was turned into a de facto supporter by the Palestinian groups. Ultimately, the
Southern Lebanon turned into a de facto safe haven, where Hizballah members train and
organize attacks across the border with Israel. For each case in the NAGs dataset, both the
support type and whether the support was de facto or not were coded.
Table S2. Descriptive Statistics
Non-Rivalries
Mean
Std.
Dev.
Strategic Rivalries
Variable
Obs
Min
Max
Obs
Mean
Std.
Dev.
Min
Max
Binary Support
1,118,170
0.001
0.034
0.000
1.000
4,828
0.093
0.291
0.000
1.000
Support Level
1,118,170
0.000
0.003
0.000
3.000
4,828
0.236
0.800
0.000
4.000
Joint Democracy
1,118,170
0.064
0.245
0.000
1.000
4,828
0.021
0.144
0.000
1.000
Trade
1,118,170
0.002
0.019
0.000
1.000
4,828
0.003
0.016
0.000
0.957
Joint IGO
1,118,144
0.562
0.213
0.000
1.000
4,828
0.726
0.235
0.000
1.000
Cultural Similarity
1,118,144
0.082
0.141
0.000
1.000
4,828
0.277
0.279
0.000
0.947
Capability Ratio
1,112,635
113.784
1,476.07
0.000
186,658
4,828
3.292
10.295
0.008
133.256
Rivalry Intensity
1,118,170
0.000
0.000
0.000
0.000
4,828
55.338
0.000
3,696.
Rivalry Duration
1,118,170
0.000
0.000
0.000
0.000
4,828
17.717
175.78
3
13.898
1.000
57.000
Satisfaction Supporter
1,118,170
-0.300
5.769
-195.000
200.000
4,828
-19.217
44.528
-301.000
200.000
Satisfaction Target
1,118,170
-0.300
5.796
-198.000
200.000
4,828
-11.566
47.145
-200.000
301.000
Satisf. Level Supporter
1,118,170
-0.014
0.187
-1.000
1.000
4,828
-0.332
0.739
-1.000
1.000
Satisf. Level Target
1,118,170
-0.014
0.187
-1.000
1.000
4,828
-0.268
0.765
-1.000
1.000
Frustration game
1,118,170
0.013
0.112
0
1
4,828
Opportunity game
1,118,170
0.009
0.094
0
1
4,828
Common ID Potential
supporter—NAG
1,118,170
0.001
0.022
0
1
4,828
0.264
0.195
0.107
0.441
0.396
0.309
0
0
0
1
1
1
Instrumental Variables Analysis:
Model Estimated:
SUPPORTij (t 1)    b1 AFFIN iNAG (t  2)  b2 FRUSTGAMEi (t 2)  b3OPPGAMEi (t 2)
m
 b4 SUPPORTij (t  2)   CONTROLkij (t  2)  1
k 5
MIDijt    b1 SUPPORTi (t 1)  b2 RIVINTij (t 1)  b3 RIVDURij (t 1) 
m
m 4
k 4
r m 2
  bk CONTROLij (t 1)  bm 1 NONMIDYRS 

SPLINEr   2
Where AFFIN is a binary affinity variable between would be supporter and the NAG, and
SUPPORT is the estimated support variable from the first equation.
Table S3: Instrumental Variable Regression of Target Response
Independent Variable
Joint Democracy
Alliance
Affinity Potential
Supporter↔NAG
Frustration Game
Opportunity Game
Lagged level of support
Constant
Model Statistics
N
Chi-Square
R-Squared
Level of Support for NAGs
Joint Democracy
Trade
IGO Membership
Cultural Similarity
Capability Ratio (Potential)
Supporter-Target
Allies
Strategic Rivals
Rivalry Intensity
All Dyads
Strategic Rivals
Binary
Level of
Level of
Support
Support
Binary Support
Support
Equation 1: Support for NAGs
-0.021**
0.000
0.119**
0.000
(0.001)
(0.000)
(0.042)
(0.011)
-0.007**
0.000
-0.006
-0.002
(0.001)
(0.000)
(0.012)
(0.003)
0.879**
0.026**
0.846**
0.049**
(0.007)
(0.002)
(0.007)
(0.006)
0.002
0.001
-0.036**
-0.004
(0.002)
(0.001)
(0.01)
(0.004)
0.024**
0
-0.046**
0.002
(0.003)
(0.001)
(0.012)
(0.004)
0.907**
0.903**
(0.001)
(0.006)
0.019**
0.000
0.248**
0.002
(0.002)
(0.000)
(0.018)
(0.003)
108,154
32,045.45
0.880
Equation 2: Target Hostility
0.556**
2.192**
(0.06)
(0.225)
-0.442**
-0.519**
(0.048)
(0.088)
0.015
-0.404
(0.381)
(0.904)
0.07
-0.175
(0.043)
(0.167)
-0.04
-0.102
(0.059)
(0.184)
0.001*
0.001
(0.000)
(0.001)
-0.04
-0.181**
(0.024)
(0.072)
0.907**
8.213**
(0.029)
(0.162)
4375
32,045.45
0.880
0.171*
(0.072)
-0.008
(0.15)
5.936**
(2.14)
-0.222*
(0.09)
-0.13
(0.094)
-0.002
(0.002)
-0.232**
(0.051)
4.617**
(1.363)
-1.07
(2.568)
105.374*
(46.911)
-5.465**
(1.879)
0.047
(1.63)
-0.017
(0.035)
-3.021**
(0.977)
0.001**
0.014**
Independent Variable
All Dyads
Binary
Level of
Support
Support
Rivalry Duration
Lagged Hostility
Constant
N
Chi-Square
R-Squared
Wald Test of Exogeneity
-1.176**
(0.032)
100748
4874.96
Chi Sq. = 0.02
p > Chi Sq. = 0.833
0.500**
(0.003)
0.887**
(0.108)
108154
46676.47
0.3168
Strategic Rivals
Level of
Binary Support
Support
(0.000)
(0.003)
0.003*
0.021
(0.001)
(0.028)
0.55**
(0.013)
0.119
11.716**
(0.072)
(1.486)
4375
4375
583.82
2360.26
0.3511
Chi Sq. = 14.85**
p > Chi Sq. = 0.001