Content uploaded by Kristof Gosztonyi
Author content
All content in this area was uploaded by Kristof Gosztonyi on Dec 19, 2015
Content may be subject to copyright.
1
Taming the Unruly
The Integration of informal Northern Afghan Militias into the Afghan Local
Police
Kristóf Gosztonyi, Jan Koehler and Basir Feda
Abstract: Though states frequently attempt to formalise and regulate militias, there has been
very little research whether such a formalisation can, indeed, improve the performance of
militias in terms of greater security provision and reduction of abuses against the population.
Using original quantitative and qualitative data from North-East Afghanistan, we show that
the integration of unregulated anti-Taliban militias into the Afghan Local Police (ALP) in late
2011 and early 2012 has, indeed, led to an improved performance of these militias in the eyes
of the population: fear of militias dropped, while their perceived positive security impact
increased when compared to their assessment prior to the establishment of ALP.
Keywords: Militias, Afghanistan, mixed-method research, counterinsurgency
Schlagworte: Milizen, Afghanistan, Methodenmix, Aufstandsbekämpfung
1. Introduction: Militias a Curse or a Cure?
Public and academic debate on militias is highly controversial and tends to ask in general
terms whether militias do more good than harm or vice versa (e.g. Sanford 2003; Schwarz
2007; Abbas/Gerspacher 2015; Human Rights Watch 2011; Jones 2012; Ahram 2011). In
spite of the frequently negative assessment of militias, in many fragile states of the world
militias continue to be a fact of life that is supported by states, local communities or political
parties fighting insurgents, criminal cartels, or ostensibly protecting against crime or engaging
in conflict with ethnic or political rivals.
According to Seth Jones, the main fear associated with militias is that, instead of providing
security, they in fact contribute to instability by undermining state authority, committing
human rights abuses and that, as a result of lacking control, they are militarily ineffective – or
“unreliable and unwieldy” (2012: 4-12). However, through a comparative analysis of 130
insurgencies since World War II, Jones finds that these negative assessments of militias are
“gross over-generalizations” (p. 11). Instead, Jones concludes, “the emphasis […] should be
This article has been double blind peer-reviewed.
We would like to thank the evaluators for their comments.
2
on the quality of regulation, not on whether a militia is inherently desirable or undesirable” (p.
34).
One possible way to exert greater state control over militias is to integrate them into
formalised para-military structures. Even though states frequently attempt such a regulation,
1
there has been surprisingly little research on whether such formalisation can indeed improve
the performance of militias in terms of greater security provision and the reduction of abuses
against the population. A notable exception is the work of Mitchell, Carey and Butler
(Mitchell et al. 2014) who in a cross-country large-n study found that the presence of informal
militias is associated with a significant increase in torture, disappearances and killings. In
contrast, the presence of formalised militias has no statistically significant impact on such
human rights abuses.
In our article we pose a similar question but examine it in-country. Using quantitative and
qualitative data from 25 districts of North-East Afghanistan (cf. Koehler et al. 2015) we
investigate whether the integration of informal anti-Taliban militias, called arbakees
2
, into the
formal structures of the Afghan Local Police (ALP) improved the (perceived) performance of
these militias. We will thereby concentrate on two dimensions: abuses against the population
and military performance.
Our central hypothesis is that – in line with state expectations – the formalisation of militias
will improve their performance along the two above mentioned dimensions. We will
investigate militia formalisation by comparing two waves (baseline and follow-up data)
analysing how respondents perceived militias before and after the introduction of ALP. We
will do this while differentiating between different levels of exposure to militias: (a) whether
a respondent lives in a village with militias, (b) whether he lives in a district, but not in a
village with militias, and, lastly, (c) whether he lives in a village and district without militias.
We proceed as follows: First, we provide an overview of our research in North-East
Afghanistan. Second, we briefly recount the development of the insurgency and the
1
Examples of the formalisation of militias include e.g. the comités de autodefensa set up to fight the Shining
Path in Peru or the US-driven formalisation of the Anbar Awakening as the Sons of Iraq movement.
2
Originally the term arbakee refers to famed tribal militias in Eastern Afghanistan (in fact one of the authors of
this article also briefly participated in such a militia in his home town in Khost). In order to gain legitimacy by
adopting the name of these respected local self-defence forces, anti-Taliban militias in other parts of Afghanistan
also came to be referred to as arbakees.
3
emergence of anti-Taliban militias in the research region, detailing the key differences
between informal arbakee and the formalised ALP. In the following we will use “militia” as a
generic term referring to both formal and informal militias, i.e. ALP and arbakee. Third, we
proceed to present empirical results from four Logistic Regression models to compare the
presence of arbakees and ALP along the two dimensions of militia performance (fear of
militias due to abuses against population and (perceived) security provision). In the final
section we present our conclusions.
2. Research Background
The research was conducted by Jan Koehler and Kristóf Gosztonyi in 25 districts of North-
East Afghanistan and covers 253 village communities.
3
It included a quantitative survey of
more than 5,000 respondents per wave, complemented by the compilation of demographic,
political, governance-related and historical information on surveyed villages, village clusters
and districts, as well as the conduct of extensive qualitative interviews. We gathered baseline
data in 2010/11 amidst intense fighting in parts of the research region at the height of the US
Surge policy. At this time, arbakees were present in nine of the 25 districts of the survey – all
nine “arbakee-districts” were affected by an intense insurgency. We conducted the follow-up
in late 2012 under significantly improved security conditions as a result of counter-insurgency
operations. By this time, ALP had become the dominant form of militia in seven of the
previously mentioned nine militia districts of our survey.
3. From Informal arbakees to ALP
After a few peaceful years following the collapse of the Taliban regime in 2001, the
insurgency made a comeback by the year 2009 in two provinces of North-East Afghanistan:
Kunduz and Baghlan. Both provinces are strategically important and have significant Pashtun
populations. The Taliban, being a largely Pashtun movement, infiltrated these provinces by
using ethnic networks. The Taliban offensive of 2009 was further facilitated by a power and
security vacuum in the North-East. By this time the militias of the anti-Soviet Jihad and the
subsequent civil war were largely disarmed and demobilised, while the Afghan National
Security Forces (ANSF) had not yet been effectively built up and was not capable of offering
3
We thank the DFG-funded SFB700 C9 for supporting this project; the survey was conducted by Afghan
Human Rights Research and Advocacy Organisation (AHRRAO) under supervision of the authors.
4
serious resistance to the onslaught. Moreover, Germany, the main foreign troop contributor in
the North-East, was not yet willing or able to engage in serious combat with determined
insurgents (Koehler 2014: 70).
The confluence of these factors resulted by mid-2009 in a near collapse of government control
in Kunduz and Baghlan. At this point in time, neither the Central Government in Kabul, nor
international military forces (most importantly the US) could offer relief. Facing impending
collapse, provincial-level Afghan elites turned to dormant Jihadi networks to fight the Taliban
(e.g. Goodhand/Hakimi 2014: 25-37; Human Rights Watch 2011: 27; Gosztonyi/Koehler
2010: 74-75, 121-123). The degree to which this process was approved or even directed by
Kabul is unclear. The task of arming and re-activating former Jihadi groups to fight the
Taliban fell to the provincial branches of the National Directorate of Security (NDS) and
Afghan National Police (ANP). Typically, they relied on networks of mostly non-Pashtun
Jihadi commanders to mobilise their fighters. On occasion, however, former insurgents were
also admitted as arbakees. The arbakees, together with the thinly spread ANSF and an
increasingly assertive German ISAF contingent, succeeded in holding out against the
insurgents until the arrival of the first US surge troops in early 2010. By mid-2010,
international and national pro-government forces were on the offensive and by 2012 managed
to significantly push back the insurgents and pacify the region. Subsequently, as the
withdrawal of international troops progressed, security once again began to deteriorate. In
particular, 2015 saw a dramatic escalation of violence in large parts of the North-East,
reaching or even surpassing the intensity of violence in 2010. This deterioration, however,
took place after our second survey and is thus not considered in the current analysis.
3.1. Arbakee
The arbakee fighting force that emerged to resist the insurgents was a typical militia force. It
was local, with fighters pledging allegiance to their commanders who in turn belonged to
parties and networks going back to the anti-Soviet Jihad and the subsequent civil war. Their
district and provincial state principals had often only tentative control over the groups of
fighters they had helped to set up. Given the ad hoc setup, tentative formalisation and lack of
regular salary, arbakees soon came to be known for abuses, taxation of the population and in-
fighting (Gosztonyi/Koehler 2010: 74-75; Human Rights Watch 2011: 27-42).
4
4
In early 2011 one of the authors was involved in facilitating a meeting between an influential former
commander and community leader from Aliabad and German military officers in Kunduz on militia related
5
3.2. ALP
The emergence of militias was eyed with mistrust by President Hamid Karzai and large parts
of the Kabul elite (Goodhand/Hakimi 2014: 14), while other elite factions and the US military
establishment reportedly supported it (Jones 2012: 31). The formalisation of arbakees as ALP
and their subordination to the Ministry of Interior Affairs (MoIA) was a compromise between
these opposing views on how to ensure local security. The ALP was conceived of as a local
“self-defence force” ensuring the security of communities where they had been set up and
from where its members were recruited. In contrast to arbakees, members of the ALP are
vetted by local elders as well as Afghan government agencies. Approved ALP members
received arms, uniforms, training, and salary, and were integrated into MoI-structures, with
ALP commanders reporting to the district chiefs of ANP. In spite of the implementation of
ALP, arbakees did not completely disappear. In many districts, ALP and arbakees continued
to coexist side by side and in one district, Khanabad, the ALP-programme was never
implemented.
The establishment of ALP did not eliminate all human rights abuses and infighting between
militias, and there is an ongoing debate whether there is any difference between the
performance of arbakees and ALP. A number of authors and institutions are highly critical of
ALP (e.g. AIHRC 2012: 20-36, Human Rights Watch 2011: 3-4), while a few reports found
that “many authorities and communities ... considered the deployment of ALP to be a more
desirable alternative for provision of security than armed groups” [i.e. arbakees] (UNAMA
2014: 50).
3.3. Operationalising the research question
We formulate the following hypothesis to test whether the introduction of the ALP
programme has indeed led to improved performance by ALP as compared to arbakees:
abuses and insecurity. The community leader lobbied for the formalisation of the militias under the command
and authority of the district chief of police. When the German officers highlighted their distrust in the
performance of the police in that district, the community leader acknowledged the problems. He replied,
however, that, despite of the existing problems, under a formalised arrangement communities would at least have
a clearly designated person responsible for the militias to whom they could turn to with their complaints and
suggestions.
6
H1: The integration of informal arbakees into ALP has improved the performance of
these militias by (a) reducing fear of ALP as compared to arbakees, and (b) by
increasing the perceived positive security contribution of ALP as compared to
arbakees.
However, as both arbakees and ALP are mainly local, village-level armed groups, their
impact on perceptions (fear and perceived security contribution) will vary depending on the
level of exposure to militias. The greater the exposure of a respondent to militias, the more he
might suffer from their unruly behaviour, but the more he will also benefit from their
protection against insurgents. Hence we formulate our second hypothesis to account for
exposure:
H2: Respondents with more exposure to militias (those living in militia villages) will
be more afraid of militias, but will value their security contribution at the district level
more strongly than respondents with less exposure (those who live in non-militia
villages located in militia districts or in non-militia districts).
4. Data and Empirical Analysis
4.1. Dependent variables
We use Logistic Regression analysis with standard errors clustered at the village level to
estimate the effect of exposure to militias on the perception of fear of, and contribution to
security by militias. Our primary consideration in deciding the models’ specifications are
driven by theory. We report our most preferred model specifications in Tables 1 and 2. A
comprehensive list and description of variables is in Annex 2.
Our key dependent variables (DVs) “fear” of militias and “perceived contribution to district
security” by militias, are two survey questions that ask respondents to indicate their degree of
fear of, and evaluate the extent to which militias (ALP and arbakees) have contributed to
security/insecurity of their district on a 3 Likert-type scale. The first is a proxy for the human
rights record of militias, while the latter is a proxy for military effectiveness as perceived by
rural Afghans. We collapse the three category variables into two categories to construct the
binary DVs suitable for Logistic Regression.
5
5
The main purpose of the analysis is to assess whether formalization leads to mitigating the negative effects
associated with militias, which is more relevant to the lowest or highest outcomes of the variables rather than the
middle category. Despite the loss of some information, we collapse the “somewhat afraid” category to “not
afraid” in the “fear of militias” DV. We also collapse the middle neutral category of “contribution to district
7
Two questions arise with regard to these DVs. First, whether respondents truthfully respond to
these potentially sensitive questions, and second, whether they are good proxies for human
rights abuses and military effectiveness. Regarding the first, so far we have no reason to doubt
respondents’ willingness to answer questions regarding their fear of different armed actors
and their assessment of the security contribution of these same actors. As an example, the
baseline carried out in a context of intense fighting in some parts of the survey region shows
high levels of fear regarding a large number of armed actors with fear being generally higher
in frontline districts than in stable ones. In contrast, fear significantly dropped during the 2012
follow-up survey in a context of substantially improved security. Moreover, even if responses
were partially biased towards giving “expected” answers, it is likely that this bias would be
similar in both waves and would thus still provide valid information on trends over time.
Regarding the second question, whether popular perceptions are an adequate proxy for the
human rights record and military effectiveness of militias, one has to consider the counter-
insurgency (COIN) context of the establishment of these militias. In this context, “winning the
hearts and minds” of the population is a key objective. Admittedly, popular perceptions do not
capture the full dimension of the issue areas “human rights” and “military effectiveness”. For
example, “fear of militias” will likely not capture the targeted killing of Taliban informers or
the torture or execution of captured insurgent fighters by militias; but it gives an accurate
assessment of militia abuses against the population such as summary retaliations, violent
intimidation and financial or sexual extortion. In a COIN context, these are all highly relevant
and counterproductive crimes committed against the civilian population. Thus, militias that
evoke little fear and are perceived as contributing positively to security are, from a COIN
perspective, successful.
4.2. Independent variables
Arbakee and ALP in village: Our key independent variables (IVs) used to estimate the effect
of militia exposure on our DVs are the dummy coded “arbakee in village” (1 for arbakee in
village and 0 otherwise) and “ALP in village” (1 for ALP in village and 0 otherwise). This
reflects our assumption that the presence of militias at the local level is a key determinant of
security” into “positive contribution”. Results from a multinomial regression where all categories are included
indicate very similar results for most of the variables across the four models with no significant difference in
relation to key independent variables.
8
perceptions of fear of, and perceived security contribution by the two different forms of
militias. (For an overview of districts and villages with and without militias for the two survey
waves see Annex 1).
In order to test Hypothesis 1, we estimate the effect of local arbakee presence on the two DVs
using baseline data from 2010/11 around the time when arbakees were active in nine out of 25
districts of our survey. We then estimate the effect of local ALP presence on the two DVs
using the follow-up data from 2012 when ALP was introduced in seven out of the nine
arbakee districts.
Since Hypothesis 2 is specifically interested in village-level exposure to militias, we introduce
dummy coded variables for ALP/arbakee presence on the village level into our models in
interaction with militia district dummies, “arbakee district” and “ALP district” (1 for
districts with arbakees, 2 for districts with ALP and 0 otherwise).
6
For the analysis of ALP performance with the follow-up data, we introduce two additional
IVs: ANSF in village (1 for presence, 0 otherwise) and ALP and arbakee in village (1 for
presence, 0 otherwise).
Fear of, and perceived district security contribution of militias: We use “perceived district
security contribution by militias” as a control in estimating the “fear of militias” as a DV and
vice versa, “fear” as a control for estimating our “perceived district security contribution by
militias” as a DV. We assume a strong mutual relationship between these two variables:
Somebody who is afraid of a certain type of militia is unlikely to believe that they
nevertheless contribute positively to security; in reverse, somebody who feels protected by a
militia group, is less likely to be very afraid of them. In the Afghan context this assumption is
not self-evident. Earlier research showed for example that fear of local commanders does not
always mean that people also consider them detrimental to local security; this relationship
seems to depend on the local security context (Koehler et al. 2011: 36-37).
6
The interaction terms introduced in the models allow us to a) conduct the analysis using the full sample of 25
districts, including those districts that have no militia activity and hence capture the variation associated with
respondents living in both militia and non-militia districts; b) test Hypothesis 2 by taking into consideration the
two-level clustering of militia presence effect on the DVs at the village as well as district levels. The interaction
between arbakee/ALP districts and villages result in the generation of three valid categories of: 1) non-militia
villages in militia districts; 2) militia villages in militia districts and; 3) non-militia villages in non-militia
districts.
9
Ethnicity: In particular, we consider Pashtun and Tajik ethnicities. As mentioned in the
historical overview, the Taliban movement tends to be dominated by Pashtuns, while the
militias show a heavy Tajik bias. We thus assume that Pashtuns would evaluate the
performance of militias more negatively along both DVs than Tajiks. Moreover, we expect
that this effect will be less pronounced with regard to the more disciplined ALP than for
arbakees.
Fear of Taliban: Further we expect that persons who fear the Taliban would be less afraid of
militias as they see them as their protectors and thus would also view their contribution to
security more positively.
District security rating: Security improved significantly between the baseline and the follow-
up. Our main control for this changed security environment is the subjective assessment of
district security by respondents.
Governance zones: Our previous research shows that the governance context is not uniform
and homogenous in the research region. Rather, it is a patchwork of different governance-
providing institutions and actors, showing great geographic variance and change over time.
We approach this diversity of governance forms through the concept of “governance zones”,
which we define as geographic areas with distinct modes of governance. For the North-East
we identified five different governance zones, each defined by the varying presence of
statehood, social control, and use of force by powerful actors (Koehler/Gosztonyi 2014).
In addition to accounting for differences in the provision of governance that are likely to have
an impact on the performance of militias, the inclusion of governance zones into the statistical
analysis is also important for a further reason: together with “district security rating”,
governance zones are our main controls for changes in the security, political and governance
context between baseline and follow-up. While during the baseline we noted significant zones
of Taliban control, during the follow-up such areas largely disappeared, while zones of
“contested governance”, “commander rule” and “governance by government” expanded.
10
5. Results
7
5.1. Arbakee performance before formalisation (baseline survey in 2010/11)
In this section we investigate the performance of arbakees some 1.5 years after their initial
emergence in 2009. Models 1 and 2 in table 1 illustrate results related to our DVs, “fear of
arbakees” and “perceived contribution to district security by arbakees”. The results show that
militias evoke significantly more fear in districts and villages where arbakees have been
introduced than in “non-arbakee districts”. However, within these “arbakee districts” we see
no meaningful difference in the opinions of respondents living in villages with or without
arbakees.
Continuing with the same three-way comparison, we get a similarly negative result for the
perceived contribution of arbakees to district security. As compared to districts without
arbakees, “non-arbakee villages in arbakee districts” and “arbakee villages” (in arbakee
districts) are on average associated with 26 and 25 percentage points (pps) decrease in the
probability of perceiving the arbakees as “positively contributing” to district security. Thus
once again we see no significant difference in the effect of arbakee exposure between the two
types of communities.
As expected, ethnicity plays a role in assessing arbakees. Likely due to an ethnic Tajik bias of
arbakees, Pashtuns tend to fear arbakees more and are more likely to negatively assess their
performance than non-Pashtuns. Fear of the Taliban appears to be a key driver of respondent
assessments of arbakees: respondents who are afraid of the Taliban are likely to be also afraid
of arbakees (a strong and statistically highly significant result), but they also tend to assess the
security contribution of arbakees more positively A further statistically significant finding
shows a strong link between our two DVs: respondents who are afraid of arbakees are very
likely to also negatively assess their contribution to district security, while those who perceive
them as contributing positively to district security will likely not be afraid of them. Lastly, the
more secure a respondent perceives his district to be, the more he is also likely to fear
arbakees.
7
Note that we only report results from post-hoc analysis. We present average marginal effects because we view
these as most easily understood and substantively meaningful in making sense of the regression outputs.
11
Table 1 Average Marginal Effects (Arbakees)
12
5.2. ALP performance after formalisation (2012)
Our follow-up survey in 2012 allows us to investigate the effect of the introduction of ALP.
Our key DVs are fear of ALP and contribution to security by ALP. We proceed with the same
three-way comparison as previously applied for arbakees. Respondents in “ALP villages” and
respondents in “non-ALP villages” but in ALP districts are on average only 5 and 6 pps more
likely to be “very afraid” of ALP than respondents in villages in non-ALP districts. This is a
more than 50% drop in fear as compared to arbakees in 2010/11.
We note a similar positive change regarding perceived contribution to district security by ALP
as compared to arbakees. As compared to villages in “non-ALP districts”, respondents in
“ALP villages” (in ALP districts) are 14 pps more likely to see the ALP as contributing
positively to district security, while we note no statistically significant result for “non-ALP
villages in ALP districts”. Thus, in 2012 ALP fares better on both DVs, fear and contribution
to district security, compared to arbakees in 2010/11. Moreover, within ALP districts, villages
with ALP fare better than non-ALP villages on both DVs. This difference is very weak for
fear (merely 1 pp), but quite strong for perceived district security contribution (14 pps).
Further significant results pertain to ethnicity: Pashtuns, on average, are more likely to be
afraid of ALP (we saw similar results for arbakees), but are also more likely to perceive ALP
as contributing positively to district security (this was less likely for arbakees in 2010/11).
Tajiks are also somewhat more likely to be afraid of ALP (we had no significant results for
arbakees in 2010/11). This effect might be the result of a conscious policy to also integrate
Pashtuns into the ALP. Fear of Taliban is less strongly associated with an increase in the
probability of fear of ALP than was the case for arbakees. In contrast, fear of Taliban is
strongly associated with an increase in the probability of seeing ALP as positively
contributing to district security (this effect was markedly weaker for arbakees). Lastly, we see
a link between our two DVs: respondents who are afraid of ALP are also less likely to assess
their contribution to district security as positive, while respondents who assess the
contribution of ALP to district security as positive are less likely to be afraid of them. This
correlation between our two DVs resembles what we observed for arbakees, but it is
significantly less pronounced.
13
Table 2 Average Marginal Effects (ALP)
14
6. Conclusions
In this article we set out to investigate whether the formalisation and regulation of informal
arbakee militias has led to a better performance by these militias (Hypothesis 1), and whether
direct exposure to militias (militias in village) leads to more fear, but also a higher perceived
security contribution by militias for the district (Hypothesis 2). Regarding Hypothesis 1, we
clearly see a better performance by ALP than arbakees in villages with and without militias:
respondents are less likely to be very afraid of ALP than of arbakees and more likely to see
them as contributing positively to district security than arbakees (who are seen as contributing
negatively to district security). In contrast, our results only partly confirm Hypothesis 2:
within militia districts, villages with or without militias show only negligible differences
regarding their fear of militias. However, while we note no security dividend associated with
living in a village with arbakees, we noted a marked and statistically highly significant
positive impact of ALP on security perceptions in their “home” villages.
It is, however, not only our main IVs that are associated with changes in our DVs, but also
some of our control variables. Respondents who perceive their district as secure are much
more likely to strongly fear arbakees. It seems that once the pressing need for security
provision is lower (a secure district), arbakees are perceived as a dangerous nuisance. We find
no comparable result for ALP suggesting a more disciplined, less abusive force from the
perspective of the local population.
ALP also appears to be ethnically somewhat less polarising. While Pashtuns tend to fear both
arbakee and ALP, they see the ALP as positively contributing to district security, whereas
they evaluate arbakees as negatively contributing to district security. The less polarising
nature of ALP is also evident when we consider “contribution to district security” as an IV. If
respondents feel that arbakees contribute positively to security, they are much less likely to be
afraid of them and vice versa, if they feel that arbakees contribute negatively to district
security, they are also more likely to be afraid of them. To some extent, people are either on
the right side or wrong side of arbakees. This relationship is there for ALP as well but is
weaker as compared to arbakees.
While arbakees strongly polarise, they seem to be doing a poorer job at protecting people
against the Taliban. Respondents, who are very afraid of the Taliban, are only 9 pps more
likely to perceive arbakees as contributing positively to district security. In contrast,
15
respondents who are afraid of the Taliban are 26 pps more likely to perceive the ALP’s
contribution to district security as positive.
Our statistical results show the ALP to be – despite documented abuses (see e.g. Human
Rights Watch 2011: 3-4, UNAMA 2014: 9) – from the perspective of the population a
significantly more benevolent and more effective militia force than the informal arbakees. The
key question then is, whether the noted improvements are the result of formalisation or
whether they were caused by other factors such as changes in the security and governance
situation between baseline and follow-up. Statistically two variables control for the changed
security and governance situation between the two surveys, “district security rating” and
“governance zones” leaving formalisation as the likely remaining reason for observed
changes.
This interpretation is further supported by our qualitative data. In particular the training and
integration of militias into official chains of command emerges as a key factor in this change
for the better. In a series of interviews we conducted with district and provincial police and
government officials as part of our survey, district chiefs of police in militia districts
confirmed good coordination with ALP, but complained about arbakees who allegedly
followed unofficial and informal chains of command and over whom they only had little or no
control.
8
The effective integration into an official command structure does not only instil greater
discipline through establishing clear chains of command; crucially it also allows for a more
strategic deployment of and more concerted military action by militias under the command of
the district chief of police.
9
This more strategic approach also led to a more ethnically
balanced recruitment to the ALP as compared to arbakees. The resulting stronger inclusion of
Pashtuns is very likely the reason why Pashtuns tend to evaluate ALP performance better than
that of arbakees.
Interviewed village representatives also tended to view formalisation as leading to a better
behaviour of militias: “In the past, [those who are now] ALP had committed many illegal
acts; but now they have been trained and do not commit any crimes. In general, people are
8
Interviews with the Deputy Chief of Police and the acting District Manager of Khanabad on 26 September
2012; interview with Provincial Deputy Chief of Police of Kunduz on 21 February 2013.
9
Discussion with a US officer involved in setting up ALP in Wardak Province on 21 May 2015.
16
afraid of those who act illegally in society”.
10
Lastly, the simple fact that ALP members
receive salary appears to have contributed to a steep drop in illegal taxation by militias in
2012 in districts where arbakees had previously levied such taxes (e.g. in Dashti Archi, Imam
Sahib and Aliabad). In conclusion, with regard to the arbakee-ALP debate in Afghanistan, our
results support a relative minority of voices who see the ALP as an improvement on arbakees
(see e.g. UNAMA 2014: 9-10, 50).
But what do these results tell us about the general social scientific debate on informal and
formal militias? So far research has shown (see Mitchell et al. 2014) that in a cross-country
comparison informal militias show a strong association with human rights abuses whereas no
such relationship exists in the case of official militias. Even though an important hint, these
results do not conclusively respond to our question regarding the impact of regulation. First,
this research only considers human rights abuses, while the two other negative outcomes
frequently associated with militias – lacking effectiveness and undermining state legitimacy –
are not addressed. Moreover, Mitchell et al. do not provide us with an unequivocal answer
whether regulating and formalising pre-existing informal militias can indeed improve their
performance.
Our empirical analysis coupled with extensive qualitative research support the hypothesis that
formalisation can improve the performance of unruly, previously unregulated informal
militias along a number of parameters often seen as problematic with regard to militias. Our
analysis also shows that formalisation does not just affect one area of militia performance, e.g.
abuses against the population. Instead, we observe consistent positive change over a number
of areas relevant for assessing the performance of militias. Formalised militias thus evoke less
fear, are perceived as contributing more positively to security, are less polarising and are less
threatening for non-co-ethnics.
Literature
Abbas, Hassan/Gerspacher, Nadia 2015: The Irregulars. Arming vigilantes in places like Iraq and Afghanistan
to work alongside struggling police forces isn't a solution -- it's a time bomb, in: Foreign Policy,
30.03.2015, online edition at: http://foreignpolicy.com/2015/03/30/the-irregulars-vigilantes-police-
security-iraq-afghanistan/; 26.06.2015.
Ahram, Ariel 2011: Learning to Live with Militias, in: Journal of Intervention and Statebuilding 5: 2, 175-192.
10
Interview with Head of Cluster Shura Sufi Ameer Mohammad Khan on 16 October 2012; This finding is in
line with the already quoted UNAMA report (2014: 50).
17
Goodhand, Jonathan/Hakimi, Aziz 2014: Counterinsurgency, local militias, and statebuilding in Afghanistan
(Peaceworks), Washingtion/DC, United States Institute of Peace, at:
http://www.usip.org/sites/default/files/PW90-Counterinsurgency-Local-Militias-and-Statebuilding-in-
Afghanistan.pdf; 01.07.2014.
Gosztonyi, Kristóf/Koehler, Jan 2010: PCA Analysis North Afghanistan, Kabul, Berlin, Gesellschaft für
Technische Zusammenarbeit, ARC-Berlin.
Human Rights Watch 2011: "Just Don't Call It a Militia". Impunity, Militias, and the “Afghan Local Police”
(September 2011), Human Rights Watch, at:
http://www.hrw.org/sites/default/files/reports/afghanistan0911webwcover.pdf; 05.01.2014.
Jones, Seth G. 2012: The Strategic Logic of Militia (January 2012), RAND National Defence Research Institute.
Koehler, Jan 2014: The Afghan perspective on ISAF – changes and trends in North-East Afghanistan, in: Chiari,
Bernhard (Ed.): From Venus to Mars? Provincial Reconstruction Teams and the European Military
Experience in Afghanistan, 2001-2014, vol. 3, Freiburg i.Br./Berlin/Wien, Rombach Verlag, 65-86.
Koehler, Jan/Gosztonyi, Kristof 2014: The international intervention and its impact on security governance in
North-East Afghanistan, in: International Peacekeeping 21: 2, 231-250, at:
http://dx.doi.org/10.1080/13533312.2014.910402.
Koehler, Jan/Gosztonyi, Kristof/Böhnke, Jan 2011: Conflict and stability in Afghanistan. Methodological
approaches, (Violence, Drugs and Governance: Mexican Security in Comparative Perspective, Stanford
University, 4 October 2011), at: http://cddrl.stanford.edu/events/6716.
Koehler, Jan/Gosztonyi, Kristóf/Feda, Basir/Child, Keith 2015: Mixed method impact evaluation - making
stabilisation assessments work for development cooperation, in: EPSJ 10: 2, 61-74, at:
http://www.epsjournal.org.uk/index.php/EPSJ/article/view/229; 09.10.2015.
Mitchell, Neil J./Carey, Sabine C./Butler, Christopher K. 2014: The Impact of Pro-Government Militias on
Human Rights Violations, in: International Interactions 40: 5, 812-836, at:
http://dx.doi.org/10.1080/03050629.2014.932783; 26.06.2015.
Sanford, Victoria 2003: Learning to Kill by Proxy: Colombian Paramilitaries and the Legacy of Central
American Death Squads, Contras and Civil Patrols, Washington DC, Robert F. Kennedy Center for
Human Rights.
Schwarz, Anthony J. 2007: Iraq’s Militias: The True Threat to Coalition Success in Iraq, in: Parameters 37, 55-
71.
18
Annex 1
Survey districts and villages grouped according to militia presence
19
Annex 2
Description of Variables
20
Variables
Description
Source
Pashtun Dummy
Dummy variable with 1 representing self-identification as Pashtun and 0 representing
all other ethnicities
Comment: The Taliban movement tends to be dominated by Pashtuns, while militias
show a heavy Tajik bias. We assume that Pashtuns evaluate the performance of militias
more negatively along both DVs.
Quantitative surveys
2010/2011 & 2012
Tajik Dummy
Dummy variable with 1 representing self-identification as Tajik and 0 representing all
other ethnicities
Comment: Militias tend to be dominated by Tajiks. We assume that Tajiks evaluate
arbakees and ALP more positively than other ethnic groups, in particular Pashtuns.
Quantitative surveys
2010/2011 & 2012
Arbakee in village
Dummy variable with 1 for Arbakee presence in the village and 0 otherwise
Qualitative survey
2010/2011 (village
profiles)
ALP in village
Dummy variable with 1 for ALP presence in the village and 0 otherwise
Qualitative survey
2012 (village profiles)
ALP and Arbakee
in village
Dummy variable with 1 for both ALP and Arbakee present in a village and 0 otherwise
Qualitative survey
2012 (village profiles)
Arbakee district
Dummy variable with 1 for districts that have Arbakee presence and 0 otherwise
Qualitative survey
2010/2011 (village
profiles)
ALP district
Dummy variable with 1 for districts that have ALP presence and 0 otherwise
Qualitative survey
2010/2011 (village
profiles)
ANSF in village
Dummy variable with 1 for ANSF in the village and 0 otherwise
Qualitative survey
2012 (village profiles)
ALP and arbakee
in village
Dummy variable with 1 for presence of both ALP and Arbakee in the village and 0
otherwise
Qualitative survey
2012 (village profiles)
District Security
Rating
Categorical variable with previously four categories collapsed into two with 2
representing “very secure” rating of the district by the respondent and 2 representing
very “very insecure” rating of the district security by the respondent. categories were
collapsed due to almost no difference of effects between the neighbouring positive-
positive and negative-negative categories
Comment: Security improved significantly between the baseline and the follow-up. Our
main control for this changed security environment is the subjective security
assessment of the respondents’ district.
Quantitative surveys
2010/2011 & 2012
Fear of Taliban
Survey Question asking respondents if they are afraid of the Taliban. Valid response
options include: 1 for “not afraid”, 2 for “somewhat afraid” and, 3 for “very afraid”.
Comment: We assume that persons who fear the Taliban would be less afraid of militias
as they see them as their protectors and thus would also view their contribution to
security more positively.
Quantitative survey
2010/2011 & 2012
Contribution to
security by
Arbakee (DV in
model 2)
Survey Question asking respondents whether Arbakees have contributed to
security/insecurity in their district. Valid response options include: 1 for “contributed
negatively”, 2 for “contributed neither negatively nor positively” and, 3 for “contributed
positively”. Categories 2 and 3 have been collapsed to construct the binary DV variable
Quantitative survey
2010/2011
Contribution to
security by ALP
(DV in model 4)
Survey Question asking respondents whether ALP have contributed to
security/insecurity in their district. Valid response options include: 1 for “contributed
negatively”, 2 for “contributed neither negatively nor positively” and, 3 for “contributed
positively”. Categories 2 and 3 have been collapsed to construct the binary DV variable
Quantitative survey
2012
Fear of Arbakee
(DV inmodel 1)
Survey Question asking respondents if they are afraid of the Arbakee. Valid response
options include: 1 for “not afraid”, 2 for “somewhat afraid” and, 3 for “very afraid”.
Categories 1 and 2 have been collapsed to construct the binary DV
Quantitative survey
2010/2011
Fear of ALP (DV
in model 3)
Survey Question asking respondents if they are afraid of the ALP. Valid response options
include: 1 for “not afraid”, 2 for “somewhat afraid” and, 3 for “very afraid”. Categories
1 and 2 have been collapsed to construct the binary DV
Quantitative survey
2012
Governance
Zones
A 5 category expert coded variable with 1 for “governance by government” 2 for
“Arbitrary rule”, 3 for “Remote self-governance”, 4 for “Contested governance” and, 5
for “Taliban governance”. The variable is coded based on extensive research of North-
east Afghanistan. The different zones reflect the patchwork of different actors providing
different services to different recipients in distinct geographic areas (Author 2012).
Categories 1 – 5 define typical and recurring governance situations in zones
(communities) ranging from areas where the government is the main governance actor
to areas where insurgents have established their own administration.
Comment: In addition to accounting for differences in the provision of governance that
are likely to have an impact on the performance of militias, the inclusion of governance
zones into the statistical analysis is also important control for changes in the security,
political and governance context between baseline and follow-up. While during the
baseline we noted significant zones of Taliban control, during the follow-up such areas
largely disappeared, while zones of “contested governance”, “commander rule” and
Qualitative survey
2012 (village profiles)
21
“governance by government” expanded.
