--------------------------------------- 1

 
 
 
 
 
 
 
 
 
 
 
 
 
 Working Paper No. 138 
 
  
 FROM VICE TO VIRTUE? CIVIL WAR 
 AND SOCIAL CAPITAL IN UGANDA 
 
  
 by Giacomo De Luca and Marijke Verpoorten 
 
  
 
  
  
 
  
 
  
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 



--------------------------------------- 2

AFROBAROMETER WORKING PAPERS 
 
 
 
 
     
 
 
 
 
 
 
Working Paper No. 138 
 
FROM VICE TO VIRTUE? CIVIL WAR AND 
SOCIAL CAPITAL IN UGANDA 
 
by Giacomo De Luca and Marijke Verpoorten 
 
 
 
 
 
 
 
 
 
 
 
June 2012 
 
 
 
 
 
 
 
 
 
 
 
 
 
Giacomo de Luca is Post-Doctoral Fellow at the Centre for Institutions and Economic Performance, 
University of Leuven, Belgium. Email: 
 This email address is being protected from spambots. You need JavaScript enabled to view it.
  
 
Marijke Verpoorten is Assistant Professor, Institute of Development Policy and Management, University of 
Anterwerp, Belgium. Email: 
 This email address is being protected from spambots. You need JavaScript enabled to view it.
 .    
 
 
 
      Copyright Afrobarometer           
 i 
            



--------------------------------------- 3

 
 
 
 
 
 
 
 
 
AFROBAROMETER WORKING PAPERS 
 
 
Editor 
Michael Bratton 
  
Editorial Board 
E. Gyimah-Boadi 
Carolyn Logan 
Robert Mattes 
Leonard Wantchekon 
 
   Afrobarometer publications report the results of national sample surveys on the attitudes of 
citizens in selected African countries towards democracy, markets, civil society, and other aspects of 
development.  The Afrobarometer is a collaborative enterprise of the Centre for Democratic Development 
(CDD, Ghana), the Institute for Democracy in South Africa (IDASA), and the Institute for Empirical 
Research in Political Economy (IREEP) with support from Michigan State University (MSU) and the 
University of Cape Town, Center of Social Science Research (UCT/CSSR).  Afrobarometer papers are 
simultaneously co-published by these partner institutions and the Globalbarometer. 
  
  Working Papers and Briefings Papers can be downloaded in Adobe Acrobat format from 
www.afrobarometer.org. 
   
 
 
                   Idasa          
    
    
 
 
co-published with:  
 
      Copyright Afrobarometer           
 ii 
            




--------------------------------------- 4

 
ABSTRACT 
We show that armed conflict affects social capital as measured by trust and associational membership. 
Using the case of Uganda and two rounds of nationally representative individual-level data bracketing 
a large number of battle events, we find that self-reported generalized trust and associational 
membership decreased during the conflict in districts in which battle events took place. Exploiting the 
different timing of two distinct waves of violence, we provide suggestive evidence for a rapid 
recovery of social capital. Evidence from a variety of identification strategies, including difference-in-
difference and instrumental variable estimates, suggests that these relationships are causal. 
 
Keywords: social capital; trust; civil war; Uganda. 
 
 
 Copyright Afrobarometer           iii 
 



--------------------------------------- 5

1
INTRODUCTION 
This paper aims to increase the understanding of the legacy of civil war by analyzing the impact of the 
Ugandan armed conflict on social capital. What is the impact of civil war on social networks? How does 
violence affect trust? Providing an answer to these questions is important for conflict researchers for a 
number of reasons. 
 
First, it is argued that differences among societies in culture² the prevailing values, attitudes, and beliefs²
contribute to differences in economic outcomes. Social capital, which is a broad characterization of culture, 
is one of the aspects considered. In the seminal work by Putnam (1993), social capital is defined as the 
features of social life, networks, norms and trust that enable participants to act together more effectively to 
pursue shared objectives. Social capital arises when people interact in a number of settings, ranging from 
membership in an organization to attendance at religious services and to dinner with a group of friends. Since 
a person is less likely to cheat someone who is a member of his social network, social capital makes people 
more trustworthy (Coleman 1988). And, vice versa, trust is a prerequisite for building social networks. For 
these reasons, trust is often associated with the value of social networks (Fukuyama 2000). 
 
Following Putnam (1993), several scholars have focused on the role of social capital in shaping economic 
performance (Colletta and Cullen 2000, Sobel 2002, Woolcock and Narayan 2000). For instance, Colletta 
and Cullen (2000) argue that high levels of trust in a society reduce transaction costs and private protection 
costs, thereby providing stronger incentives for investment and innovation. Early cross-country studies 
confirmed the positive association between social capital, trust and indicators of economic performance 
(Knack and Keefer 1995, 1997, Zak and Knack 2001, Guiso et al. 2004). And, in the new growth literature, 
these aspects of culture are considered to be fundamental drivers of economic growth, having the potential to 
lift the economy to a higher level growth path (Weil 2009). 
 
Second, social capital may be particularly relevant for the economic development of countries with weak 
formal institutions, which host most of the present-day civil wars (Blattman and Miguel 2010). In these 
countries, the vacuum left by the absence of formal institutions may be filled in part by informal institutions 
that overcome coordination failure. For example, a well-networked society may provide a solid base for 
mutual aid and informal insurance and facilitate the flow of information and collective action (because 
people who already have a relationship with each other can trust one another to do their part in a joint 
enterprise). Furthermore, social capital may stimulate the accountability of ill-functioning governments. 
People who care about their fellow community members may be more likely to vote. Thus, politicians in an 
2
environment where social capital is high may be less inclined to abuse their constituents for personal gain. 
This functioning of social capital may be welfare enhancing, as is suggested by the findings of several 
empirical studies in developing countries that report a positive impact of social capital on household 
expenditure, access to credit, public-service provision, and the adoption of new technologies in agriculture 
(Grootaert et al. 2002, Isham 2002, Narayan and Pritchett 1999). 
 
Third, although culture is generally perceived as sticky and able to change only slowly, there is a strong prior 
that exposure to violence can affect certain aspects of social capital. A recent study by Nunn and 
Wantchekon (2011) provides some support for this prior as it establishes a link between 400 years of slave 
trade and the development of a culture of mistrust in Africa that persists to this day. Scholars have also 
highlighted the role of interstate wars, e.g. World War II, in shaping national identity and state-building by 
reinforcing social cohesiveness and collective action both within and across states (Hanson 2003). However, 
civil wars, which are fought between opposing factions in a society, are often thought to be disruptive of the 
                                                      
1
 We received helpful comments and support from Michael Bratton, Francis Kibirige, Carolyn Logan and Leonard 
Wantchekon. All errors and opinions expressed remain our own. 
2
 However, if social capital is formed at the subgroup level, for example, according to ethnic or regional origin, then it 
may serve as an instrument of exclusion and polarization rather than an instrument of social gain and cohesion 
(Fafchamps, 2006). 
 
1 
 Copyright Afrobarometer           
 



--------------------------------------- 6

society's social fabric thus endangering its political stability and economic recovery (Colletta and Cullen 
2000, Collier et al. 2003). 
 
This rather pessimistic view has been challenged by recent micro-level studies. Studying the aftermath of the 
1991-2002 civil war in Sierra Leone, Bellows and Miguel (2006, 2009) have shown that victimization during 
the conflict increased political participation. In particular, households directly experiencing displacement or 
more intense violence were more likely to attend community meetings, join political and community groups, 
and vote. Similar findings have been presented by Blattman (2009) who investigated the impact of rebel 
conscription by abduction on post-war social and political participation in Uganda. His analysis suggests that 
the level of violence witnessed during the war as an abductee leads to more participation in political life but 
does not affect membership in non-political organizations. In a related paper, investigating the impact of 
conflict on civic and political participation in Uganda, we show that political discussion and meeting 
attendance increase in districts in which battle events took place, although we do not detect any sizable effect 
of the war on formal electoral participation (De Luca and Verpoorten 2011). 
 
In another recent study, Voors et al. (2010) reported that individuals who experienced violence during the 
1993-2003 civil war in Burundi displayed more altruistic behavior in the post-war period. Similarly, in a 
series of behavioral experiments, Gilligan et al. (2011) find that, three years after the end of Nepal's civil 
war, members of communities with greater exposure to violence exhibit significantly higher levels of social 
capital, measured by subjects' trust-based transactions and collective good contributions. 
 
On the other hand, another set of recent contributions adds to the more pessimistic viewpoint. First, Cassar et 
al. (2011) show how conflict exposure in post-war Tajikistan undermines trust and fairness within local 
communities; decreases the willingness to engage in impersonal exchange, and reinforces kinship-based 
norms of morality. Second, Rohner et al. (2011), studying the aftermath of the conflict in northern Uganda, 
3
reported lower levels of trust and reinforced ethnic identity in the affected districts. 
 
Finally, Bozzoli et al. (2010) studied the impact of conflict on expectations relying on a 2007 survey 
conducted in northern Uganda. They found that timing matters: whereas pessimistic expectations prevailed 
shortly after the experience of conflict, optimistic expectations were positively related to conflict intensity in 
the distant past. 
 
In sum, although most scholars would agree on the importance of understanding the impact of civil war on 
different aspects of culture, the scarce evidence available is mixed. Moreover, little is known about the 
underlying mechanisms, the persistence of the impact, and the possible heterogeneity of the impact related to 
the nature of civil war. This is also stressed by Blattman and Miguel (2010), who, in their recent discussion 
on civil war, argue that ³the social and institutional legacies of conflict are arguably the most important but 
the least understood of all  war impacts.
 
The distinctive features of the present article are threefold. First, it is the first study on the impact of civil war 
on social capital that can rely on two rounds of data compilation, one of which took place before the bulk of 
the violence occurred. This unique data set, bracketing a peak in violence of more than 250 battle days in a 
year in the affected area, allows us to adopt a difference-in-difference estimation by studying the change in 
trust and associational membership upon a continuous treatment equal to the number of district-level battle 
days.  
 
                                                      
3
 Rohner et al. (2011) use the same data sources as we do. However, the two papers are substantially different in focus 
and empirical strategy. While we address the effects of conflict on associational membership and generalized trust 
during and after the conflict adopting a difference-in-difference estimation, they investigate the effects of the conflict on 
ethnic cleavages, trust, and the economic situation in the post-war period. 
 
2 
 Copyright Afrobarometer           
 



--------------------------------------- 7

Second, exploiting the geographic variation as well as the variation in timing of two distinct waves of 
violence, we can assess the impact of violence on social capital during the conflict and provide suggestive 
evidence as regards the duration of the impact. 
 
Third, it is also worth stressing that the treatment effect we study is different from the treatment considered 
in most of the previous studies. Instead of studying the impact of individual-level exposure to violence, we 
compare the average change in social capital in areas more or less affected by conflict. In this case, the 
treatment is to be interpreted as living in a high-risk, insecure environment; with possible direct exposure to 
violence as a witness, victim or perpetrator. Although this community-level perspective may not be helpful in 
uncovering the underlying mechanisms, it is important in order to determine whether the heterogeneous 
individual-level effects sum up to a general change in social capital. In this sense, our study is 
complementary to previous work on Uganda by Blattman (2009). 
 
Although we cannot identify the exact mechanisms underlying our results, it is worthwhile mentioning that 
the literature proposes at least three possible mechanisms that link war or trauma to inter-personal relations. 
First, in war environments, information acquisition may be costly or imperfect. To save the individual from 
the costs of acquiring information, the use of heuristic decis-ofi-on thummakb, ingca stn rategies, or rules
be optimal (Boyd and Richerson 2005). Very concretely, in a highly insecure environment, a general rule of 
mistrust may yield the highest payoff. As such rules become entrenched in culture, they may survive the 
situation in which they came into existence. For example, Nunn and Wantchekon (2011) rely on this 
mechanism to explain the persistent impact of historic slave trade on mistrust in SSA. Second, traumatic 
experiences can result in real psychological change impacting in its turn on self-protective behavior and a 
change in perception of other individuals (Weinstein, 1989), whereby positive change is not excluded. For 
example Collins et al. (1990) show that the ³recognition of one¶s vulnerability´ may enhance the value of 
social networks. Finally, if² as argued above² trust is partly the result of social interactions, then a 
breakdown of social networks, e.g. community groups, due to displacement or inter- and intra-community 
violence, may lead to decreased trust levels (de Jong, 2002). 
 
Our findings indicate that both self-reported trust and associational membership decrease substantially 
during the conflict in the affected districts. However, we also find suggestive evidence of a strong recovery 
process once the violence has ended. Hence, these findings mediate between the optimistic and pessimistic 
viewpoints. 
 
The difference-in-differences estimation method, along with the inclusion of several district- and individual-
level controls, provides a solid base for our empirical analysis. Nevertheless, the potential issue of 
endogeneity of conflict intensity remains in establishing a causal effect of violence on social capital. 
Violence may be the consequence rather than the cause of decreased levels of trust and participation in 
associations, because, for instance, rebel recruitment may be easier in regions plagued by antagonistic 
feelings. Although, as argued by Blattman (2009), this is far-fetched in the northern Ugandan context, where 
a large share of the Lord's Resistance Army was composed of abducted youths, we adopt distance measures 
4
to instrument for conflict intensity following the recent empirical literature on civil war. The IV estimations 
broadly confirm the OLS findings. 
 
In the next section, we present the data and provide relevant background information on the armed conflict in 
Uganda. In section three, we present our empirical strategy. The results are presented in section four. The last 
section concludes. 
                                                      
4
 See, for example, Akresh and De Walque (2008), Miguel and Roland forthcoming, Serneels and Verpoorten (2011), 
Voors et al. (2010). 
 
3 
 Copyright Afrobarometer           
 



--------------------------------------- 8

 
 
 
DATA AND BACKGROUND 
 
Social capital 
The data on social capital is taken from the Afrobarometer (AB), an independent, non-partisan research 
5
project that measures the social, political, and economic atmosphere in Africa. We use two rounds of AB 
survey data compiled in Uganda in 2000 and 2005. We do not use the 2002 and 2008 AB surveys. The 2002 
survey does not cover the districts most affected by the civil war, while the questions on social capital 
included in the 2008 survey are not comparable to those in the 2000 and 2005 surveys and therefore cannot 
6
be analyzed in a difference-in-difference model. 
 
Each survey includes information on approximately 2,400 individuals of voting age. The samples are 
nationally representative and geographically stratified across 33 districts in 4 regions, including both urban 
and rural areas. Figure 1 gives an administrative map of Uganda in 2000, and Table A1 lists the districts by 
region. In Table 1, we give the number of observations per region and per survey year, which show that all 
7
four administrative regions of Uganda are well represented in both survey years. 
 
Figure 1: Administrative map of Uganda 
 
                                                      
5
 The data and survey instruments are available from www.afrobarometer.org. 
6
 Compared to the 2000 and 2005 survey, in 2008 an entirely different set of questions on trust was adopted, and only 
two of the associational membership questions were retained. 
7
 The four administrative regions are denominated: Central, Eastern, Northern, and Western. 
 
4 
 Copyright Afrobarometer           
 



--------------------------------------- 9

 
The AB surveys include questions on two interrelated dimensions of social capital: trust and associational 
membership. We restrict our analysis to those questions that are comparable across the survey rounds in 
terms of question formulation and response categories, which leaves us with five questions - one on trust and 
four on membership. The question on trust concerns the respondent's level of trust of others in general and is 
formulated as folGleownersal:l y ³speaking, would you say that most people can be trusted or that you must 
be very careful in dealing  w ith people?
 
The answer categories are ³You ry mcaustr efbe ul (ve coded asMo 0)st, and  people can be trusted (coded 
as 1). The questions on membership are introdNowuce Id  amas goi fng olto lowread s: ou³t a list of 
voluntary organizations. For each one, could you tell me whether you are an official leader, an active 
member, an inactive member, or not a member of that tThe ype listo fof  ororgganianizatzaionts ion? 
includes (i) a religious organization like a church or a mosque, (ii) a trade union or farmer's organization, (iii) 
a professional or business organization, and (iv) a development association. We code the answer categories 
as follows: (0) Not a member, (1) Inactive member, (2) Active member, and (3) Official leader. In the 
empirical analysis, we check the robustness of our results against different ways of coding the answers. 
 
A summary of the social capital variables and the different coding is provided in Table 2. We find rather low 
levels of trust with less than 20% of the respondents answering that most people can be trusted. Membership 
is highest in religious organizations, with more than 80% of the respondents reporting that they were a 
member of a religious organization (inactive, active or leading). The other types of organizations only 
involve the membership of 20% to 30% of the population. Over time, the reported levels of social capital are 
rather stable. On average, generalized trust increases by two percentage points, and membership of a 
religious organization increases slightly from 80% to 83%, but these changes are not significant. Larger and 
significant changes take place in membership in a trade union/farmer organization (a decrease of 13 
percentage points), a professional/business organization (a decrease of 8 percentage points), and a 
community development organization (an increase of 6 percentage points). It is noteworthy that there are 
large differences across regions despite the relatively small changes in the averages. For example, self-
reported trust increased by 10 percentage points in the western region, while it decreased by 12 percentage 
points in the northern region. 
 
 
 Copyright Afrobarometer           5 
 

bTale 1: Sample observations
Individuals in sampl (Nre)
20002005
rCental751656
East559584
North369544
Wset592616
Total22,712,400



--------------------------------------- 10

 
The Conflict 
The conflict intensity data are taken from the Armed Conflict Locations Events Data (ACLED), which is 
based on the screening of news reports and provides geo-referenced information on approximately 3,921 
8
violent events in Uganda between 1960 and 2010 (Raleigh et al. 2010).  
 
The violent events include battles between armed groups (2,659) or attacks on civilians (1,262). These events 
took place on 1,983 different days within the 50-year period. Hence, on average, a year counts almost 40 
event days, and an event day counts 1.98 violent events. The bulk of these event days - more than 90% - took 
place after 1995. We situate them on a timeline in Figure 2. 
 
 
                                                      
8
 The ACLED event data is compiled by screening news articles with language recognition software, which on average 
yields good results. However, in some cases press accounts may be biased, because very insecure areas may be difficult 
to access or because computer news screening may be sensitive to the language in which events are reported (Restrepo 
et al. 2006, Verpoorten in press). Two separate ACLED datasets for Uganda were released, one recording events 
between 1960 and 2006, and one recording events between 1997 and 2010. We merged the two datasets and removed 
the duplicate observations in the overlapping period 1997-2006. 
 
6 
 Copyright Afrobarometer           
 

blTae 2: De descriptives ofpendeblesnt varia
Distriibuton of 
rseponses across 
Val usueed in aVlue usde in ansrwe caegorites 
main robusstnes (coln)um %
specificaitoncheck20002005
Trust generally
You mus b very cteafurel08583
Mosopt pel cean tru bested11517
bership of:     Mem
tonous oranizaiReligig
 amberNot me002017
Inaberctiemve m102425
mberActi meve214751
Official laeder3197
Trade union or farmter ornizaionga
 amberNot me007083
Inaberctiemve m1084
mberActi meve212011
Official laeder3122
nizationPreional or business orgaofss
 amberNot me007886
Inaberctiemve m1063
mberActi meve211310
Official laeder3121
Community deevlopmeationnt oranizg
 amberNot me007064
Inaberctiemve m1076
mberActi meve211925
Official laeder3145



--------------------------------------- 11

Figure 2: Event days with LRA and ADF involvement 
 
 
Figure 2 indicates that, after a period of relative peace following the power seizure by Museveni in 1986, the 
number of event days started to rise in 1995. This increase occurred on two fronts. First, in northern Uganda, 
the Lord's Resistance Army (LRA), an armed opposition group founded in 1987 by Joseph Kony was able to 
9
intensify its activities in 1995 mainly because of support by the Sudanese government (Dolan 2009). 
 
Second, in western Uganda, another armed group² the Allied Democratic Forces (ADF)² commenced its 
activities in the mid-nineties. The ADF was a fundamentalist Islamic guerrilla group formed by various 
remnant rebels from Uganda, Congo, and Rwanda (Boas 2004). They operated mainly from the Ruwenzori 
Mountains bordering with Congo and received support from both the Congolese and Sudanese governments 
(Behrend 2007). 
 
In Figure 2, we distinguish between the number of event days with LRA and ADF involvement. Whereas 
both groups started to increase their activities in the mid-nineties, the peaks and ends of their activities 
occurred at different times. The activities of the ADF were influenced by conflict trends in the neighboring 
DRC and were mostly concentrated in the period from 1997 to 2001. By 2002, relative degree of peace had 
been established in western Uganda. The bulk of LRA violence, instead, fully unraveled in the period 2002-
2005 following a military operation in southern Sudan by the Ugandan army² the Iron Fis² itn´tended to 
destroy the LRA supply bases (Dolan 2009). LRA bases were, indeed, destroyed and many rebels killed. 
However, the mission was considered a failure (Allen 2006, Dolan 2009). In fact, LRA forces managed to 
outflank the Ugandan army and attacked further south in Ugandan districts until then relatively untouched by 
the conflict (e.g., Apac and Lira). Starting from 2006, however, the area of LRA activities first moved out of 
Uganda into southern Sudan and into the Democratic Republic of Congo (2006-2008) and then further west 
reaching the Central African Republic after 2008 (Accord 2010). This released the pressure on civilians and 
opened the way to peace in northern Uganda. 
                                                      
9
 Kony¶s movement gathered armed groups reluctant to settle with Kampala's new government and was initially called 
the Lord's Salvation Army, then the United Democratic Christian Forces, and eventually, from 1994 onward, the Lord's 
Resistance Army (Allen 2006, Doom and Vlassenroot 1999). The LRA received assistance from Sudan in retaliation for 
Ugandan support of a rebel group operating in southern Sudan. 
 
7 
 Copyright Afrobarometer           
 



--------------------------------------- 12

Linking social capital to the conflict 
To construct our dataset, we merged the AB data with the ACLED data at the district level, which is the 
smallest administrative unit they have in common (the ACLED data is geo-coded but the AB data is not until 
2008). Doing so yields a dataset of approximately 4,500 individual level observations across 33 districts. In 
what follows, we will use the ACLED conflict data as a treatment to study the change in social capital upon 
conflict. Before doing so, three issues have to be discussed. 
 
First, it is evident from Figure 2 that ADF violence peaked before the 2000 AB baseline survey, was still 
fairly high in 2001, and then ceased such that by the time the 2005 survey was carried out, ADF operations 
had come to an end. Therefore, when we single out the effect of ADF violence we can interpret our results as 
the effect of conflict cessation on social capital with respect to the baseline survey collected amidst the 
10
violence. In other words, the ADF treatment captures post-war recovery. 
 
Second, although LRA violence escalated after our baseline year, a non-negligible number of battle days 
took place before 2000. This pre-2000 LRA activity was largely confined to one particular geographic 
region, Acholiland (Kitulgu umdi stand ricts), Gwhile LRA violence outside Acholiland only took off after 
2000. Hence, while the estimated LRA treatment should be interpreted as the impact of continued and 
escalating violence on social capital, a somewhat cleaner treatment effect can be discerned when focusing on 
LRA activities outside Acholiland. This point is illustrated in Figure 3, which gives the number of event days 
11
with LRA involvement inside and outside Acholiland for the period 1986-2010. 
 
Figure 3: LRA event days inside and outside Acholiland 
 
 
Third, that the AB survey was conducted in times of violence as well as amidst a huge refugee crisis 
following violence raises the issue of sample selection bias. For instance, because of insecurity, the surveys 
may have excluded the most affected individuals in certain districts. In order to verify this, we consulted the 
                                                      
10
 In the regression analysis, we evaluate the impact of the ADF battles occurring in the period 2000-2005 on social 
capital. In an alternative specification (not reported) we study the impact of the 1995-2000 ADF battles, which gives 
similar results (because the spatial variation of ADF battles is very similar across the two periods). 
11
 Notice that the average intensity of LRA violence experienced outside Acholi districts before 2000 is fairly low, as 
the violence reported in Figure 3 was spread across 7 districts (see Appendix A1 for a detailed distribution of violence 
by district). 
 
8 
 Copyright Afrobarometer           
 



--------------------------------------- 13

local AB team that conducted the survey. We learned that, when an enumeration area within a district was 
highly insecure, the enumeration area was replaced by a more secure area within the same region. The 
substitution always followed the composition of the original sample in terms of language and ethnicity as 
well as the direction of displacement of the individuals in the original sample, often ending up with a sample 
of within-district internally displaced people (IDP) (correspondence with Francis Kibirige 2011). This 
approach was facilitated by the maintenance of local administrative structures in the IDP camps and also by 
the moving of the IDP within their own district. Given this approach and given the fact that more than 90% 
of the population was living in these local IDP camps, we can be fairly confident that the AB survey is 
12
representative at the district level for each survey year.  
 
Moreover, since substitution was confined to highly insecure areas, it basically would only cause potential 
bias for the impact of LRA violence in Acholiland. Results for other violent events, e.g. the LRA battles 
outside Acholiland, would not be affected. 
 
Finally, to cross-check the ACLED data and verify whether it indeed is in line with individual-level 
perceptions of violence of the respondents, we examine the following question that was included in the 2005 
AB survey: Please tell me if the following things are worse orw  byeaetrs tagero,  nowor  than they were a fe
are they about the same: Safety from crime and violence?. We code the answer to this question as follows: 
(1) Much better, (2) Better, (3) Same, (4) Worse, (5) Much worse. Given the trends in the LRA and ADF 
events described above, we expect respondents in LRA-affected districts to report worsening safety 
conditions, while we expect the reverse for respondents in the ADF-affected districts. This is exactly what 
we find. We find a positive and significant correlation between the LRA treatment (LRA event days in 2000-
5) and the respondents perception of the trend in violence and crime, indicating that, in line with the rising 
trend in ACLED events in LRA districts, the respondents perceived an increase in violence. In contrast, we 
find a negative and significant correlation between the ADF treatment (ADF events in 2000-2005) and the 
respondents¶ perception of the safety situation, indicating that the ADF treatment corresponds with a decline 
in violent events over time. 
 
EMPIRICAL STRATEGY 
To identify the impact of violence on social capital, we use a difference-in-difference estimation that exploits 
variation in the event days across districts and over time. The treatment is a continuous variable equal to the 
event days occurring between the implementation of the 2000 and 2005 AB surveys. The treated group is the 
households located in the districts where the battles and attacks took place. In other words, the empirical 
identification strategy relies on the comparison of the change in social capital in 2000-2005 across areas with 
low violence intensity and areas with high violence intensity. 
 
Formally, the empirical model can be written as follows: 
 
ññ
5Üáçá×
LÙ5$×
EÙ6UAç=N
EÙ7-:$×ÛUAç-;=N
E:Üáçá×Ö
E:çá×Æ
Eßå
EÝÜáçá× 
 
where i indexes individuals, d districts, r regions and t survey years. The variable 5Üáçá× denotes individual-
level social capital. $× denotes logged event days per district in the period 2000-2005; and UAç =Nis an 
indicator variable taking one for respondents in the 2005 survey. Thus, the coefficient of interest is Ù7, which 
is the coefficient of the interaction term between $× and UAç. =NTo reduce heterogeneity across the 
observations on social capital, we control for a number of relevant individual-level and district-level 
covariates. The vector : denotes a set of individual-level covariates, including the respondent's age, age 
Üáçá×
squared, a gender indicator variable, an indicator variable that equals one if the respondent lives in an urban 
location, ten fixed effects for the respondent's ethnicity, and nine fixed effects for the educational attainment 
                                                      
12
 For instance, figures released by the Internal Displacement Monitoring Centre reveal that, by 2005, 98% of the entire 
population of Gulu district - as reported by the 2002 population census - was living in IDP camps located within Gulu 
(www.internal-displacement.org). 
 
9 
 Copyright Afrobarometer           
 



--------------------------------------- 14

of the respondent (all recorded in the AB). The vector :çá× denotes a set of district level covariates, which 
include historical battle days experienced in the period 1960-2000 (taken from ACLED) and ethnic 
13
fractionalization. 
 
Both variables may affect the level of trust in the baseline year. All explanatory variables are summarized in 
Table 3. In a series of robustness checks reported at the end of this paper, we show that these district-level 
controls are not critical for our results. 
 
 
                                                      
13
 Ethnic fractionalization is taken from the 1991 Ugandan population census accessed through IPUMS at the Minnesota 
Population Center. 
 
10 
 Copyright Afrobarometer           
 

blTae 3: detrablesscription of explanaory vai
20002005
Dependn viaetarbles
nrGeeal trust0.150.17
Reliigous organization1.461.48
Trade union or farer ormganiaztion0.530.32
Professional or busisnes organization0.390.26
Communivln orty deeopmetganization0.560.70
Explanatory vaiarbles
Balvents tte e'01-'0519.822.9
 i   LRAn Acholiland7.638.20
 ous   LRAticde Aholiland7.329.27
   ADF2.302.30
eAg33.432.6
2e^Ag12751207
Male0.500.50
Urban0.140.30
Education lvele 3.143.35
hEtnicity (%)
steoA4.767.25
ugaLnda253.6186.3
ubLgara2.914.92
Lumasaba5.154
uoL (incl. Acholi)128.6129.2
Lusoga9.919.88
Rukiga6.436.25
Runyankole2.95110.4
Rutooro119.33.63
hrote177.521.5
Historic batltes24.528.4
hiEtno-Lnguisitc fracitonaliaztion04.04.
Education lvele is a caegoritcal variablae tking valen ueswe bet
0 and 8, wiht 0 denoting no schooling and 8 post-university 
educatih Acholon; teithni ecr goup is a subgrouph br of teoarde 
coh iategry of Luo; tendex othno-lf einugisitc fractionalization is 
calculad frhe 19teom t91 population cnesus.



--------------------------------------- 15

Finally, ßå denotes regional fixed effects, which are included to capture region-specific unobserved factors 
that may affect social capital and ÝÜáçá× is the standardized error term. In order to account for a potential 
correlation of these errors within districts and within years, we adjust the standard errors for two-way 
clustering as suggested by Cameron et al. (2006). 
 
Since the answer categories for the social capital questions in the AB surveys are categorical, we have the 
option between two different estimation strategies. First, we can maintain the categorical nature of the 
answers and estimate an ordered probit model. Second, we can estimate our empirical model by OLS, 
treating the categorical answer as if they were part of a continuous scale. We use the latter approach in the 
14
baseline result and report the former as a robustness check. 
 
To account for the different timing of the violent events, we estimate an expanded model in which we 
distinguish three different types of violence: LRA violence in Acholi districts (LA), LRA violence in non-
Ý
Acholi districts (LN), and ADF violence (AD). Formally, we replace $× in the previous equation by $×, with 
j=(LA,LN,AD) denoting the type of violence, obtaining: 
 
źÅǺ½ÅºÅǺ½
5Üáçá×
LÙ5$×
EÙ6$×
EÙ7$×
EÙ8UAç=N
EÙ9
k$×ÛUAç
o=N
EÙ:
k$×ÛUAç
o=N
EÙ;
k$×ÛUAç
o=N
ññ

E:Ö
E:Æ
Eßå
EÝÜáçá× 
Üáçá×çá×
 
where Ù9, Ù:and Ù; are the coefficients of interest. In this specification, the vector :çá× also includes the 
interaction term of ethnic fractionalization in 1991 and the 2005 year dummy. By doing so, we can rule out 
that the differential impacts we may find across the three types of violence are due to different degrees of 
ethnic heterogeneity across the affected districts. Again, this district-level control is not critical for our 
results. 
 
As noted above, LRA activities exploded in 2002 but affected northern districts already before 2000, albeit to 
ź
a lesser degree. Hence, we interpret the coefficient Ù9 on the interaction term $×ÛUAç =Nas the effect of 
additional and escalating violence on social capital. The LRA activities reached further south only after 2002, 
ÅÇ
which allows us to interpret Ù: on the interaction term $×ÛUAç =Nas the impact of violence on social 
capital relative to a situation without a direct confrontation with violence. Finally, the coefficient Ù on the 
:
º½
interaction term $ÛUAç =Ncaptures the change in social capital when moving from a situation amidst 
×
violence into a post-war phase. 
 
With respect to these coefficients of interest, we formulate two intuitively appealing hypotheses: 
 
Hypothesis 1: Both the start and the escalation of LRA violence reduce social capital. Ù9
O   and Ù:
O . 
 
Hypothesis 2:  The ending of ADF violence is associated with an increase in social capital. Ù:
P . 
 
Thus, we hypothesize that violence reduces social capital and that, once the violence ends, social capital 
recovers. It is much less intuitive to conjecture about the relative magnitude of the coefficients. For instance, 
if the start of violence reduces social capital more than the escalation of violence, then Ù9
OÙ:. On the 
other hand, if the escalating violence reaches a very high intensity, the reverse may be true: Ù9
PÙ:. 
 
As already mentioned, the AB survey was conducted when violence was on going as well as amidst a huge 
refugee crisis. In the districts most heavily affected by the violence were districts the vast majority of the 
population was living in local IDP camps. When interpreting our results we are unable to disentangle the 
effect of facing violence from the effect of living in an IDP camp. However, since both experiences are often 
                                                      
14
 One advantage of using OLS is that it allows us to estimate the standard IV model. This is also the approach taken by 
Nunn and Wantchekon (2011) in their analysis of the AB trust data. 
 
11 
 Copyright Afrobarometer           
 



--------------------------------------- 16

emerging together in civil wars, we believe that identifying their joint effect is valuable and enhances our 
knowledge on the impact of civil wars. 
 
EMPIRICAL RESULTS 
 
Baseline results 
Table 4 shows the results of estimating our first equation, i.e., when all types of violence (LRA and ADF 
event days) are pooled together. There are no sizable effects of violence on the level of generalized trust and 
on the associational membership, except for a significantly negative impact on membership in religious 
organizations. However, as argued above, it is more appropriate to separate ADF- and LRA- related violence 
as well as LRA violence inside and outside Acholiland. 
 
 
 
Table 5 shows the estimation results of our second equation when the three types of violence are 
distinguished. Consider first the treatment effect associated with LRA violence in Acholiland. The estimated 
coefficient Ù9 indicates a decrease in associational membership (Columns 2-5), but no significant impact on 
generalized trust (Column 1). The estimated treatment effect of LRA violence outside Acholiland, Ù:, 
indicates a significant decrease in generalized trust as well as associational membership of religious groups 
and community/development organizations but is insignificant in explaining the change in membership in 
economic associations (trade/farmer & professional/business). The magnitude of the impact is substantial: for 
instance, according to our estimate in column (1) a 1% increase of LRA battle days outside Acholiland leads 
to a decrease in generalized trust of 0.025, which is a sizable effect considering the average generalized trust 
of 0.15 for the whole sample in 2000 (see Table 3). Similarly, a 1% increase of LRA violence outside 
 
 Copyright Afrobarometer           12 
 

blTae 4: OLS results
TrustbersMemhip
Trade/     Prsofesional/  muniComty/   
ndnt vaDepeeriable:nrGeeal trustReligiousFarrmeBusinsesvelntDeopme
(1)(2)(3)(4)(5)
0.005-0.055*00.060.005-0.014
(0.011)(0.032)(00.27)(0.014)(00.28)Batlte days '01-'05*2005 
round
Batltes riraent pein tetmod-0.0190.083-0.06*2**-0.076***-0.1*56**
(0.030)(0.052)(00.21)(0.019)(00.30)
aPont yerst-treatme-0.102**0.061-0.36*8**-0.248***02.35*
(0.046)(0.150)(01.05)(0.069)(01.35)
Distric-llteve controlssYesYeesYsYeesY
Individual-level controlssYesYeesYsYeesY
iRegon fiffectsxed esYesYeesYsYeesY
brNume of clusrtes63/263/263/263/363/2
Obsrevations4,4814,52945,074,5004,506
R-squarde0.0640.06400.760.1030.123
bat eh p<0.05 * p<0.1; Tl rpors* p<0.01, **,eet OS esies  **Ltmaes The uniNot:t.t of observation is an 
individual The rob.us sttandard errors are adjusd forwo-wate ty clustering wihtin rural and urban 
sapmlse in each district anid wthin years aen brnd aweacsh ir repored betket. Tendietvidual-lvele and 
distric-ltevel controls ar thosee scpeifid eian Tble 3.



--------------------------------------- 17

Acholiland reduces participation in community/development organizations by 0.047, relative to the average 
15
of 0.56 for the whole sample in 2000. 
 
 
 
Finally, for ADF violence, we find a positive and significant impact for all social capital variables considered 
(Ù;
P ), indicating that individuals living in the ADF-targeted districts reported higher levels of trust and 
greater involvement in all types of association in peaceful 2005 than in war-torn 2000. 
 
Overall, these results are in line with Hypotheses 1 and 2 formulated above. Hypothesis 1 is, however, only 
partially confirmed since Ù9, although negative, is not estimated significantly different from zero in 
explaining the change in generalized trust. In contrast, Ù: is estimated negative and significant, indicating 
                                                      
15
 Ordered probit estimations reported in the robustness checks produce very similar figures. 
 
13 
 Copyright Afrobarometer           
 

blTae 5: OLhing beS restinguistwesults, dien three tpeiolenceys of v
TrustmbersMehip
Tra/  de   Prosfesional/  Comunimty/   
ndnt vaDepeeriable:neraGel trustlReigiousFarrmeBusinesslDeveoentpm
(1)(2)(3)(4)(5)
-0.005-0.09*6**-0.056***-0.014*-0.07*5** iLRAn Acholiland Batl:te days 
(0.006)(00.12)(0.018)(0.007)(00.16)'01-'05x2005 round
-00.25***-00.77**0.0170.005-0.047** outsLRAichde Aolila Balnd:tte 
(0.007)(00.36)(0.026)(0.023)(00.23)days '01-'05x2005 round
 BatlADF:te days '01-'05x2005 
0.058***0.1*59**0.13**0*00.55***0.15*5**round
(0.013)(00.16)(0.023)(0.020)(00.30)
 iLRAn Acholiland Batl:te days 
-0.0110.000-0.008-0.033-0.066**in perin treametodt
(0.007)(00.37)(0.015)(0.027)(00.26)
 outsLRAichde Aolila Balnd:tte 
0.0080.032-0.076***-0.063*-0.10*5**days in perin treametodt
(0.009)(00.37)(0.018)(0.033)(00.28)
 BatlADF:te days iraent n tetm-00.72***0.016-0.10*8*-0.064-0.19*3**
period(0.009)(00.52)(0.046)(0.068)(00.58)
aPont yerst-treatme-0.0440.061-0.399***-02*.53**0.28*7**
(0.032)(00.79)(0.063)(0.058)(01.08)
Distric-llteve controlssYeesYsYesYeesY
Individual-level controlssYeesYsYesYeesY
iRegon fiffectsxed esYeesYsYesYeesY
brNume of clusrtes63/263/263/263/363/2
Obsrevations4,4814,5294,5074,50045,06
R-squarde0.0740.0810.0840.10501.35
bat eh p<0.05 * p<0.1; Tl rpors* p<0.01, **,eet OS esies  **Ltmaes The uniNot:t.t of observation is an 
individual The rob.us sttandard errors are adjusd forwo-wate ty clustering wihtin rural and urban samplse 
in each district anid wthin yaers aen brnd aweash ir rpored betcket. Tendieetvidual-lvele and distric-llteve 
controls ahre tos sepecified in Tabl 3 ae,sl wel as thno-l theieraen eintcitwenuton term begisitc 
fracitonaliaztion and the 2005 year dummy.



--------------------------------------- 18

that trust is negatively affected by LRA violence outside Acholi districts. Thus, in the case of generalized 
16
trust Ù9
OÙ:, indicating that the start of LRA violence was more detrimental to trust than its escalation. 
The pattern is reversed for associational membership, where we find that Ù9
PÙ:. As pointed out above, 
one explanation may be that the escalating LRA violence inside Acholiland reached epic proportions. In fact, 
in the period 2000-2005, the number of event days in an Acholiland district was more than five times as high 
than the number of event days in an LRA affected district outside Acholiland (see Table 1). The 
concentration of intense fighting in Acholiland triggered a large refugee crisis, which may have disrupted the 
associational life of its residents. In contrast, associational life outside the Acholi districts may have been less 
affected given the lower exposure to intense violence and the lower degree of population displacement. 
 
Robustness checks 
We perform three types of robustness checks. First, we test if our results hold in subsamples of the AB data. 
Second, we check whether our results are robust with respect to the use of alternative estimation models and 
alternative definitions of the main variables of interest. All these results are condensed in Table 6, in which 
we report only the coefficients for the interaction terms of interest. Third, to tackle potential endogeneity 
issues (for instance, conflict events may be measured with error), we repeat our estimates instrumenting for 
the different types of violence. 
 
                                                      
16
 Beside the different timing of the violence, the three types of violence were also characterized by differences along 
ethnic lines. In particular, the LRA was composed by Acholi individuals, and perpetrated violence against people 
belonging to the same ethnic group when operating in Acholi districts, and against other groups when operating 
elsewhere. The ADF was more heterogeneous in terms of ethnic composition as were their targeted victims. Rohner et 
al. (2011) dig further into these issues. 
 
14 
 Copyright Afrobarometer           
 



--------------------------------------- 19

 
 
 
 Copyright Afrobarometer           15 
 

0.00000.000.00000.000.000 BalADF:tte days '01-'05x2005 round
(00.00)(0.000)(0.000)(00.00)(0.000)
Only Banuat-ares (South)
 iLRAn Acholila Balnd:tte days '01-
0.00000.000.00000.000.000
'05x2005 round
(00.00)(0.000)(0.000)(00.00)(0.000)
 outsLRAicde Aholiland: Battl deays 
0.033-0.096**-01.63***-00.34-0.081*
'01-'05x2005 round
(00.28)(0.047)(0.025)(00.28)(0.044)
 BalADF:tte days '01-'05x2005 round
0.0*59**0.14*3**01.23***0.05*5**0.196***
(00.18)(0.023)(0.021)(00.13)(0.028)
onrolsExcludinict level ctg distr
 iLRAn Acholila Balnd:tte days '01-
-00.10**-0.08**7*-00.67***-0.017**-0.050***
'05x2005 round
(00.05)(0.009)(0.017)(00.07)(0.010)
 outsLRAicde Aholiland: Battl deays 
-0.0*33**-0.062**0.002-00.02-0.003
'01-'05x2005 round
(00.05)(0.030)(0.026)(00.24)(0.016)
 BalADF:tte days '01-'05x2005 round
0.0*56**0.16*2**01.27***0.05*4**0.157***
(00.13)(0.014)(0.021)(00.18)(0.029)
a
Ooprbit/probit
 iLRAn Acholila Balnd:tte days '01-
-0.010-0.12**1*-0.109*-00.50-0.132**
'05x2005 round
(00.40)(0.037)(0.062)(00.46)(0.051)
 outsLRAicde Aholiland: Battl deays 
-00.78*-0.09**7*0.007-00.14-0.059
(00.40)(0.033)(0.038)(00.35)(0.040)'01-'05x2005 round
 BalADF:tte days '01-'05x2005 round
0.2*36**0.21*6**02.00***01.18*0.222***
(00.61)(0.046)(0.054)(00.60)(0.056)
a
oAernative cdinglt of anerssw
 iLRAn Acholila Balnd:tte days '01-
-0.19**2*-01.04***-00.25-0.116***
'05x2005 round
(0.037)(0.037)(00.57)(0.040)
-0.17**4*0.02000.08-0.078 outsLRAicde Aholiland: Battl deays 
(0.035)(0.041)(00.44)(0.048)'01-'05x2005 round
 BalADF:tte days '01-'05x2005 round
0.099*02.58***01.37*0.162***
(0.052)(0.064)(00.74)(0.050)
Event dayeas instd of logged even dtays
 iLRAn Acholila Balnd:tte days '01-
-0.000-0.00**2*-00.01***-00.00-0.002***
'05x2005 round
(00.00)(0.000)(0.000)(00.00)(0.000)
-0.0*02**-0.00**7*0.000-00.02-0.003** outsLRAicde Aholiland: Battl deays 
'01-'05x2005 round
(00.01)(0.002)(0.002)(00.02)(0.001)
 BalADF:tte days '01-'05x2005 round
0.0*14**0.03*9**00.28***0.01*3**0.040***
(00.03)(0.009)(0.008)(00.04)(0.007)
Loggds ineade eventst of logged event days
 iLRAn Acholila Balnd:tte days '01-
-00.08*-0.08**4*-00.53***-0.016**-0.074***
'05x2005 round
(00.04)(0.010)(0.015)(00.06)(0.012)
 outsLRAicde Aholiland: Battl deays 
-0.0*23**-0.068**0.01800.03-0.040**
'01-'05x2005 round
(00.06)(0.029)(0.022)(00.19)(0.019)
 BalADF:tte days '01-'05x2005 round
0.0*48**0.13*4**01.14***0.05*0**0.131***
(00.11)(0.015)(0.020)(00.19)(0.028)
a0001 The tbl p<0.01 * p<.5, * p<.;e reporses  **,*t OS esiNot:*Ltmaesh unit. Tet of obsrevaiton is an individual. 
 Thestandardr erors are adjused forwo-wat ty clustering within rural and urban saplmes in each district and 
a
wihtin years aen brnd aweasre repored betcket. t Bootsrapped standard errors i.  Thenlcudd ceontrol variablse 
are as in Tabl 5 ue,nlses indicad otherwites.e



--------------------------------------- 20

Subsample analysis 
Our empirical strategy relies on the comparison of the change in social capital between individuals living in 
heavily war-affected districts and individuals living in less affected districts. Thus, these latter districts are 
used to proxy the counterfactual: what would have happened if violence would not have taken place? This is 
a valid approach if both groups of districts are broadly comparable in terms of other potential determinants of 
the change in social capital. To put this approach to a test, we estimate our empirical model for two 
subsamples of broadly comparable districts: (1) a sample including only the northern area to test the impact 
of the LRA violence and (2) a sample only including the southern area for testing the impact of the ADF 
violence. The former is a rather radical test since it leaves us only with one third of the sample observations. 
 
This north-south division of the sample follows the division of Uganda along ethnolinguistic lines: the 
southern part is exclusively Bantu, whereas the northern part is almost exclusively of Nilotic origin. In Table 
A1, we indicate the districts with Bantu origin based on Lewis (2009). The results, reported in the first two 
panels of Table 6, are qualitatively the same as our baseline results with one exception. For non-Acholi 
districts, the impact on generalized trust loses significance, which may be due to the drastic reduction in the 
sample size. 
 
Alternative estimation models and variable definitions 
We first estimate a more parsimonious model that excludes the district level controls (historical battles, 
ethnic fractionalization and the interaction term between the latter and the 2005 survey dummy). The results, 
reported in the third panel of Table 6, are qualitatively the same as the baseline results except for the impact 
of LRA violence in Acholiland on general trust, which is now estimated to be significantly negative (instead 
of insignificant). 
 
Second, since the responses to the AB questions on trust and membership are categorical in nature, a sensible 
robustness check consists in replicating the estimations using the original categorical nature provided by the 
AB. Using an (ordered) probit model produces estimates that are qualitatively identical to our baseline OLS 
estimates. 
 
We also estimate our empirical model using probit with an alternative binary coding for our dependent 
variables (see Table 2 for details on the codes). The results are given in the fifth panel of Table 6: 13 out of 
the 15 coefficients of interest remain qualitatively identical, and the remaining two coefficients lose 
significance but do not change sign. 
 
Finally, we repeat our main specification measuring conflict intensity in two different ways (instead of the 
logged number of event days): (1) by the number of event days, and (2) by the logged number of events. The 
results ± displayed in the last two panels of Table 6 ± do not change qualitatively except for the impact of 
LRA violence in Acholiland on the generalized trust level, which is now weakly significant (instead of 
insignificant). 
 
IV estimates 
Ý
We instrument for the three types of violence as well as the three interaction terms of interest, i.e., $ and 
×
Ý
$ÛUAç. =NThus, we need to instrument for six different variables, using at least six excluded instruments. 
×
In order to do so, we follow the three-step procedure described in Wooldridge (2002, p.236) which will give 
Ý
us six different instruments. In the first step, we predict conflict intensity by regressing $ on the set of 
×
Ý
included instruments as well as three excluded instruments, with the latter denoted by <. Next, three 
×
Ý
predicted conflict intensity variables are interacted with the post-treatment year, 
k$
àÛUAç
o=N, which yields 
×
ÝÝ
three additional excluded instruments. Finally, both 
k$
àÛUA
o=Nç and < are used as instruments in a 
××
ÝÝ
conventional 2SLS procedure, instrumenting for $ and the interaction terms $ÛUAç.=N 
××
 
 
16 
 Copyright Afrobarometer           
 



--------------------------------------- 21

Formally, the Wooldridge procedure is given by the following set of equations: 
 
(Step 1) 
 
ÝÝÝÝñÝñññÝññÝññÝññ
$
L<Á
EÜUAç=N
E:Ô
E:Æ
Eß
EÝ 
××Üáçá×çá×åÜáçá×
 
(Step 2) 
 
ÝÝñÝñÝñÝñÝñÝññññÝñññÝñññÝñññ
$×
L<×Á
EÜ5UAç=N
EÜ6
k$
à×ÛUAç
o=N
E:Üáçá×Ô
E:çá×Æ
Eßå
EÝÜáçá× 
 
ÝÝññÝññÝññÝññÝñÝñññññÝññññÝññññÝññññ
$ÛUA=N
L<Á
EÜUA=N
EÜ
k$
àÛUA
o=N
E:Ô
E:Æ
Eß
EÝ 
×ç×5ç6×çÜáçá×çá×åÜáçá×
 
(Step 3) 
 

àן
à×ÅÇ
à׺½Åº
ãÅÇ
㺽
ã
5Üáçá×
LÚ5$
à
EÚ6$
à
EÚ7$
à
EÚ8UAç=N
EÚ9-@$×ÛUAç-A=N
EÚ:-@$×ÛUAç-A=N
EÚ;-@$×ÛUAç=N-A
ññññññ

E:Ô
E:Æ
Eßå
EÝ 
Üáçá×çá×Üáçá×
 
The coefficients of interest are Ú, Úand Ú, which capture the treatment effect of the predicted battle days 
9:;
on social capital. 
 
Ý
As excluded instruments, <, we use the 1991 district-level population share of Acholi, the distance to 
×
Sudan, and the logged distance to the Ruwenzori Mountains. The first of these instruments captures LRA 
17
violence, which was directed mainly against the Acholi.  
 
The second instrument is relevant because, as part of the Sudanese support for the LRA rebels, the LRA was 
provided with logistics and bases on Sudanese territory from where they organized raids. Finally, since ADF 
bases were located in the Ruwenzori Mountains, where rebels could easily hide and be supplied from the 
DRC, we expect the distance to these mountains to be highly correlated with the location of ADF operations. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
                                                      
17
 The LRA was constituted by people of Acholi origin. Nevertheless, LRA received little support among the Acholi 
population as it resorted to looting and youth abduction to sustain itself. The situation worsened further when the 
government started to organize self-defense militias in Acholi districts. The LRA leadership tagged this decision as 
betrayal and launched a campaign of killing and mutilation of Acholi civilians to dissuade further collaboration with the 
government army (Behrend 1999, Branch 2005). 
 
17 
 Copyright Afrobarometer           
 



--------------------------------------- 22

 
 
 
 
 
 
 
 Copyright Afrobarometer           18 
 

Table 7: First stage IV results
RA iLnside RA outsLi de
 iLRAnside  ousLRAtide  iLRAnside  outsLRAide cAholiland Baltte Acholiland Battl eDF BalAtte days 
Acholiland Batl tecAholiland Batlte ADF Battl daeys Acholiland Battl echAoliland Batl teDF BalAtte days days '01-'05 x days '01-'05 x '01-'05 x '05 
Dependent variable:days '01-'05days '01-'05'01-'05days '01-'05days '01-'05'01-'05'05 round'05 roundround
(1)(2)(3)(4)(5)(6)(7)(8)(9)
Shar of Acholei in disrtict-00.00***-0.00**3*-0.002**-00*.00**-00.03***-0.002**-00.00-0.000-0.000
(0.000)(0.001)(00.01)(0.000)(0.001)(0.001)(00.00)(0.000)(0.000)
Distano Suce tdan-00.16***0.19**0*-0.14*0**-00*.16**0.190***-0.14**0*-00.010.0010.005
(0.004)(0.046)(00.44)(0.004)(0.046)(0.044)(00.01)(0.007)(0.011)
log(Distao Ruenzorince tw mountains )6.486***-5.24**3*-0.91*7**65*.22**-52.88***-0.92**6*00.02-0.004-0.015
(0.036)(0.207)(01.55)(0.012)(0.177)(0.146)(00.02)(0.027)(0.020)
Prdiecti ion of LRAnsicde Aholiland 
-00*.08**0.009-0.0040.9*96**0.007-0.011baltte daysa x posnt yert-treatme
(0.001)(0.011)(0.011)(00.01)(0.005)(0.008)
0.003-0.003-0.0050.0*01**10*.08**-0.021Prdiecti outson of LRAichde Aoliland 
(0.002)(0.035)(0.029)(00.00)(0.012)(0.016)baltte daysa x posnt yert-treatme
0.002-0.008-0.01700.01-0.0000.989***Prdiection of ADF batl dateys x post-
(0.003)(0.046)(0.037)(00.01)(0.018)(0.011)nt yeartreatme
aPont yerst-treatme-0.0020.045-00.56-00.05**0.049-0.038-00.02*-0.0290.087
(0.002)(0.052)(00.56)(0.002)(0.057)(0.086)(00.01)(0.039)(0.066)
Distric-llteve controlssYeYesesYsYesYeYesesYsYesYe
Individual-level controlssYeYesesYsYesYeYesesYsYesYe
iRegon fiffectsxed esYeYesesYsYesYeYesesYsYesYe
brNume of clusrtes63/263/263/263/263/263/263/263/263/2
Obsrevations4,5394,53945,394,5394,53945,394,5394,5394,539
R-squarde0.9990.87607.600.9990.87607.6010.000.9120.829
es RobNot:us sttandard errors adjusd fooater tw-wy cluseritng withiorn gegaphic unist and within yearsh i i0 p<0.05 * p<0.1. Tencn pas* p<.01, **,lrnthees **uded contre;ols ar aes in Tabl he e5. T
irnteactirson tem are insnted atrumes sugesed by Wgtooldrie (20g00 p2,.36); first consrtucting prdiecd vatelues bal of thette days by rgreses iihm on thencng teluded insntstrume a excnd theluded 
insstrument (columnsn 1-3 the us);ihe ir png tnteaen theredid ctirswectebattl daon tem beteys an yeand the pos-trametr atets additional inidetfying instrent ih fiumn ters s 2S (cttage of theSLolumns 4-9).



--------------------------------------- 23

The first stage results, reported in Table 7, indicate that the instruments are relevant, with the estimated 
ÝÝ
coefficients on the instruments < significantly different from zero in predicting $
à (Columns 1-6, with 
××
Ý
Columns 1-3 corresponding to Step 1 of Wooldridge's Procedure), and $
àÛUAç =Nsignificantly different 
×
18
from zero when instrumenting for the interaction terms (columns 7-9). The relevance of our instruments is 
confirmed by the Kleibergen-Paap test for underidentification. The second stage results are qualitatively very 
similar to the OLS results for most of the social capital variables, suggesting that endogeneity is absent and 
our results can be interpreted as causal. The only noteworthy change concerns the impact of LRA violence in 
Acholiland on trust, which is now estimated significantly negative (instead of insignificantly negative). 
                                                      
18
 The F-statistics of all first stage regressions are larger than 100. In interpreting first stage results, notice that each 
instrument is meant to predict one particular type of violence. This explains why, for instance, the share of Acholi 
population displays a negative sign when predicting violence outside Acholiland, and a positive sign when predicting 
violence inside Acholiland. 
 
19 
 Copyright Afrobarometer           
 



--------------------------------------- 24

 
 
 
CONCLUSIONS 
We analyzed the impact of armed conflict in Uganda, making use of two nationwide surveys that not only 
bracket the peak in LRA violence taking place in the north of the country, but also capture the transition from 
violence to peace in the west in the aftermath of ADF violence. In sum, our findings indicate that social 
capital decreases amidst violence. However, in line with several previous findings, we also document that 
social capital strongly increases in the post-conflict phase. 
 
More specifically, we find that both the level of trust and participation in religious and community 
associations decreases when transiting from relative peace to violence. Regarding participation in economic 
associations (farm, trade, business and professional voluntary organizations), the negative impact of violence 
 
 Copyright Afrobarometer           20 
 

blTae 8: IV, 2nd stage
TrustbrsMemehip
Trade/     Prsofesional/  unCommiy/   t
ndnt vaDepeeriable:rGenealr tustReligiousFarmerBusisnesvelntDeopme
(1)(2)(3)(4)(5)
 iLRAn Acholiland Batl:te days 
-0.00*8**-0.092**-0.0*46**-0.001-00*.62**'01-'05x2005 round
(00.01)(0.041)(00.14)(0.024)(0.018)
-0.03*3**-0.071*0.0320.020-0.028 outsLRAichde Aolila Balnd:tte 
(00.03)(0.037)(00.26)(0.017)(0.040)days '01-'05x2005 round
0.05*0**0.202***0.1*81**0.11**5*02.06*** BatlADF:te days '01-'05x2005 
(00.02)(0.038)(00.12)(0.014)(0.061)round
-0.03*5**-0.005-0.0*72**-0.066***-02*.26** iLRAn Acholiland Batl:te days in 
(00.02)(0.054)(00.25)(0.017)(0.042)nt peritreatmeod
-0.01*0**-0.051-0.1*70**-0.08*8*-02*.79** outsLRAichde Aolila Balnd:tte 
(00.02)(0.041)(00.35)(0.036)(0.076)days in perin treametodt
 BatlADF:te days iraent n tetm-0.13*3**0.442***-00.74*-0.191***-04*.16**
period(00.06)(0.039)(00.43)(0.062)(0.097)
aPont yerst-treatme-0.023**0.058-0.4*50**-0.319***0.213
(00.09)(0.128)(00.28)(0.022)(0.183)
Distric-llteve controlsesYsYeesYsYesYe
Individual-level controlsesYsYeesYsYesYe
iRegon fiffectsxed eesYsYeesYsYesYe
brNume of clusrtes63/263/263/263/363/2
Obsrevations44,814,5294,5074,5004,506
R-squarde00.720.0860.0840.1050.136
Klein-Pbergeaap r sk LMtaitstic10.49104.110.310.310.32
p-value(00.01)(0.001)(00.01)(0.001)(0.001)
bat eh p<0.05 * p<0.1; Tl rpors* p<0.01, **,eet OS esies  **Ltmaes The uniNot:t.t of observation is an 
individual Boos.trapd spetandard errors, adjusd foro-wate twy cluseritng wihtin rural and urban samples in 
each distric atndi wthin years aen brnd atweactshe ire repored beke. Tnctldude controls ar aes in Tabl 5e. 
 nTheull hypothesis he Klof teirn-Pbegeaap unrdeinidetficaiton tes itsqua tha the etiton is undreintidefi ed,



--------------------------------------- 25

is confined to Acholiland, where the conflict was most disruptive as the majority of the population was living 
in IDP camps to protect themselves from LRA attacks.  
 
On a more positive note, our results suggest a strong post-violence recovery process. A few years after the 
end of the ADF-related violence, the level of trust dramatically increases in the affected areas, and 
participation flourishes in all the types of voluntary organizations considered. It is difficult to assess to what 
extent this result can be generalized. For instance, a rapid post-conflict recovery of social trust may not occur 
in areas targeted by LRA operations because, contrary to the ADF-affected region, the LRA-affected areas 
remained largely excluded from power in the aftermath of the conflict, a factor that may play against the easy 
restoration of trust. 
 
How do these findings fit with the literature? At this stage, the evidence from existing studies balances 
between a positive and a negative civil war impact on social capital. So does our study, which suggests that 
much may depend on the time span between the measurement of social capital and the end of violence. In 
addition, the overall community-level effect may depend heavily on what proportion of the community was 
directly exposed to violence as a witness, perpetrator or victim, and what part of the community was only 
indirectly affected, e.g. by displacement and fear.  
 
Although we control for a large set of individual- and district-level covariates, and although our results are 
stable throughout a number of robustness checks, and after controlling for possible endogeneity and 
attenuation bias, these results remain tentative. First, this is obviously not an experimental setting and the 
econometric techniques used cannot fully substitute for the unobserved counterfactual: what would have 
happened in the absence of violence? Second, many questions remain unanswered. What are the precise 
mechanisms underlying our results? Does social capital bounce back to its pre-war level, fall behind or even 
exceed it? How can these results be generalized to other settings with violence of different forms and 
duration? To answer these questions, more data points are needed from more countries on more forms of 
violence. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Copyright Afrobarometer           21 
 



--------------------------------------- 26

REFERENCES 
 
Accord (2010): Initiatives to end the violence in northern Uganda. Conciliation Resources. 
Akresh, R., and D. de Walque (2008): Armed Conflict and Schooling: Evi dence from the 1994 Rwandan 
Genocide, HiCN Working Papers 47, Households in  Conflict Network.
Allen, T. (2006): Trial justice: the international criminal court and the Lord's Resistance Army. Zed Books. 
Behrend, H. (1999): Alice Lakwena and Holy Spirits. Ohio University Press, Athens, Ohio. 
Behrend, H. (2007): Catholics and Cannibals: Terror and Healing in T ooro, Western Uganda, in The 
practice of war: production, reproduction and communication of armed violence, ed. by M. B.  
Aparna Rao, Michael Bollig. Berghahn Books. 
Bellows, J., and E. Miguel (2006): War and Institutions: New Evidence  from Sierra Leone, American 
Economic Review Papers and Proceedings, 96 (2). 
Bellows, J., and E. Miguel (2009): War and local collective action i n Sierra Leone, Journal of Public 
Economics, 93. 
Blattman, C. (2009): From Violence to Voting: War and Political Part icipation in Uganda, American 
Political Science Review, 103(2). 
Blattman, C., and E. Miguel (2010): Civil War, Journal of  Economic Literature, 48(1).
Boas, M. (2004): Uganda in the Regional- NWaarrr Zatone:ives , PasMetats and Present s, Journal of 
Contemporary African Studies, 22(3), 283-303. 
Boyd, Robert, and Peter J. Richerson. 2005. The Origin and Evolution of Cultures. Oxford:Oxford  
University Press. 
Bozzoli, C., T. Brueck, and T. Muhumuza (2010): Conflict Experiences a nd Household Expectations on 
Recovery: Survey Evidence from Northern Ugscandaussion Pap, DerI.W Di 
Branch, A. (2005): Neither Peace nor Justice: Political Violence and t he Peasantry in Northern Uganda, 
1986-1998, Afran Sictudies Quarterly, 8(2). 
Cameron, A. C., J. B. Gelbach, and D. L. Miller (2006): Robust Inf-werayenc Cle wiusterth i ngMul, ti
NBER Technical Working Papers 0327, National Bureau of Economic Research. 
Cassar, A., P. Grosjean, and S. Whitt (2011): Civil War, Social Capi tal and Market Development: 
Experimental and Survey Evidence on the Negative Consequences of Viol ence, SSRN Working 
Paper 1917111. 
Coleman, J. S. (2000): Social capital in the Creation of Human Capital , American Journal of Sociology, 
94, 95-120. 
Colletta, N. J., and M. L. Cullen (2000): Violent conflict and the transformation of social capital. The World  
Bank, Washington, D.C. 
Collier, P., V. L. Elliot, H. Hegre, A. Hoeffler, M. Reynal-Querol, and N. Sambanis (2003): Breaking the  
Conflict Trap: Civil War and Development Policy. World Bank and Oxford University Press. 
Collins, R. L., Taylor, S. E., & Skokan, L. A. (1990). A better world or a shattered vision? Changes in life  
perspectives following victimization. Social Cognition, 8, 263-285.De Jong, T.V.M., 2002. Trauma,  
War, and Violence: Public Mental Health in Socio-Cultural Context (eds), New York: Kluwer  
Academic Publishers. 
De Luca, G., and M. Verpoorten (2011): Civali Pal Wrtiarcipat aniond Poli: Evidencte ficrom Uganda,  
Working paper, Households in Conflict Network. 
Dolan, C. (2009): Social Torture: The Case of Northern Uganda, 1986-2006. Berghahn Books. 
Doom, R., and K. Vlassenroot (1999): Kony's Message: a New Koine? The L ords Resistance Army in 
Northern Uganda, African Affair-36.s, 98( 390), 5
Fafchamps, M. (2006)vel: Deopment and social capital, The Journal of Developm- ent Studies, 42(7), 1180
1198. 
Fukuyama, F. (2010): Social Capital and Civil Society, IMF Working  Papers No. 74, International 
Monetary Fund. 
Gilligan, M. J., B. J. Pasquale, and C. D. Samii (2011): Civil War and Social Capi-Gtalam:e  B ehavioral
Evidence from Nepal, SSRN Working Paper  1911969.
Grootaert, C., G.-T. Oh, and A. Swamy (2002): Social Capital, Household Welfar e and Poverty in Burkina 
 
 Copyright Afrobarometer           22 
 



--------------------------------------- 27

Faso, Journal of African Econom4-38.ies , 11(1), 
Guiso, L., P. Sapienza, and L. Zingales (2004): The Role of Social Capit al in Financial Development, The 
American Economic Review, 94(3), 526-556. 
Hanson, V. D. (2003): Ripples of Battle: How Wars Fought Long Ago Still Determine How We Fight, How  
We Live, and How We Think. Doubleday. 
Isham, J. (2002): The Effect of Social Capital on Fertiliser Adoption:  Evidence from Rural Tanzania, 
Journal of African Economies, 11(1), 39-60. 
Knack, S., and P. Keefer (1995): Institutions and Ecce: Crosso-nomCountic Perry Testfs Usorimngan  
Alternative Measures, Economics and Po litics, 7.
Knack, S., and P. Keefer (1997): Does social capital have a-countn ecryon  omic payoff? A cross
investigation, Quarterly Journal of  Economics, 112.
Lewis, M. P. (2009): Ethnologue: Languages of the World. Sixteenth edition. SIL International, Dallas, TX. 
Miguel, E., and G. Roland (forthcoming): The Long Run Impact of Bom bing Vietnam, Journal of 
Development Economics. 
Minnesota Population Center (2010): Integrated Public Use Microdata Series, International: Version 6.0  
[Machine-readable database]Minneapolis: University of Minnesota. 
Narayan, D., and L. Pritchett (1999): Cents and Sociability: Household I ncome and Social Capital in Rural 
Tanzania, Economic DevelopmCentultur and al Change, 47(4), pp. 871-897. 
Nunn, N., and L. Wantchekon (forthcoming): The Slave Trade and the Ori gins of Mistrust in Africa, 
American Economic Review. 
Putnam, R., R. Leonardi, and R. Y. Nanetti (1993): Making Democracy Work. Princeton University Press,  
Princeton, NJ. 
Raleigh, C., A. Linke, H. Hegre, and J. Karlsen (2010): Introducing ACLE D: An Armed Conflict Location 
and Event Dataset, Journal of Peace Re-se660.ar ch, 47(5), 651
Restrepo, J. A., M. Spagat, and J. F. Vargas (2006): Special Data Feature; The Se verity of the Colombian 
Conflict: Cross-Country Datasets Versus New Micro-Data, Journal of Peace Resear- ch, 43(1), 99
115. 
Rohner, D., M. Thoenig, and F. Zilibotti (2011): Seect in Uds ofganda D," SSistRNr eLiustbr:a Cry.o nfli
Sobel, J. (2002): Can We Trust Social Capital?, Journal of Econ-154.om ic Literature, 40(1), 139
Verpoorten, M. (2012): Detecting Hidden Violence: The Spaxcesst Miaolrt Dalitiyst in r ibution of E
Rwanda, Political Geog 31(1)ra: 44phy-56. 
Voors, M., E. Nillesen, P. Verwimp, E. Bulte, R. Lensink, and D. van Soest  (forthcoming): Violent Conflict 
and Behavior: A Field Experiment in Burundi, America n Economic Review.
Weil, D. (2009): Economic Growth. Pearson International Edition. 
Weinstein, N. D., 1989. Effects of Personal-Experience on Self-Protective Behavior, Psychological Bulletin, 
105: 31-50. 
Woolcock, M., and D. Narayan (2000): Social Capital: Implications for D evelopment Theory, Research, 
and Policy, The World Beseaarchnk O Rbserver, 15(2), 225-249. 
Wooldridge, J. M. (2002): Econometric analysis of cross section and panel data. MIT Press, Cambridge,  
Massachusetts. 
Zak, P. J., and S. Knack (2001): Trust and Growth, The Econom-321ic  Journal, 111(470), 295
 
 
 
 
 
 
 
 
 
 
 
 
 Copyright Afrobarometer           23 
 



--------------------------------------- 28

APPENDIX  
 
 
 Copyright Afrobarometer           24 
 

Table A1: distribution of event days in '95-'00 & '00-'05 acrosypessh three t of vi teolncee
1995-20002000-2005
RA iLnside RA outsLi de iLRAnsi de osRAutiLd eBanut
DistrictcAholilandcAholilandDF AAcholilandcAholilandDF A
NORTH:
Apac51360
Arua61290
Gulu8302470
Kitgum8522390
Lira130750
bbNei20120
EAST:
Iganga0020X
Jinja0011X
Kamuli0000X
Kataikw10130
Kuim0020
Mbale0030X
Pallisa0010X
Soroti70430
Tororo0400X
RA:CENTL
Kampala1603X
Kiboga0000X
roLuwe0010X
Masaka0001X
Mpigi0303X
Mubende0122X
Mukono0212X
Rakai0000X
WST:E
Bushenyi0719X
Kabale0603X
Kabarole092012X
Kasese060013X
Kibaale0707X
Kisoro0201X
Masindi56311X
Mbarara0502X
uNtngamo0000X
Rukungiri0500X
Sum1684021048622570
es The 20Not:00B s Aury rveound stadn darte i month of Ma evetysn they. The prioro Ma ty 2000 are incldude 
in the 1995-2000 period whin da,l evetyse the occuring larte are includd ienhe 20 t00-2005 period.