Peace Dividends? Effect of conflict reduction on activity choices 1

Peace Dividends? Effect of conflict reduction on activity choices 1 Carlos Bozzoli, Tilman Brück and Tony Muhumuza DIW Berlin, and Households in Conflict Network (HiCN) Very preliminary Paper Prepared for Presentation at the Chronic Poverty Research Center International Conference Hosted by the Brooks World Poverty Institute, University of Manchester. 8-10 September, 2010. Abstract Northern Uganda experienced a protracted conflict between the government and rebel groups between 1986 and 2006. Over 90% of the population in the Acholi (and later Lango) region was displaced, and insecurity constrained the number and types of activities that individuals could resort to in order to survive. In this paper, we study the effect of conflict reduction on household activity choices. We argue that there are two ways in which conflict de-escalation eased the constraints on household activity. First, given location (family lives in an Internally Displaced Person -IDP- camp or returned to their place of origin) conflict may affect activity choice. Second, households may choose their location (i.e. leave the camps) and this may also have an impact on activity choice. We separate the effect of IDP-camp residence and the effect of conflict reduction on household activity choice. Because households relocating away from camps may be different from those staying in them, we use a recursive bivariate probit procedure to control for selection on unobservables. Our empirical procedure relies on merging household survey data with a micro-level dataset on conflict "events". This allows us to incorporate conflict in the framework, but also to use a bivariate probit procedure to control for endogeneity (selection of households out from camps). Preliminary results indicate that being a camp-resident has varied effects on economic activities. Households in IDP camps are more likely to cultivate and engage in petty trading, but returnees are in a better position to engage in any activity. Our findings point to the need to emphasize targeting household livelihoods even when they are still in camps rather than wait for complete recovery. Keywords: conflict, household welfare, activity choice. 1 The authors would like to thank MICROCON project for financial support. All opinions expressed in this paper are for the authors, and should not necessarily be ascribed to MICROCON. Correspondence: tmuhumuza@diw.de 1

Introduction Mass violent conflict poses a host of challenges to governments and communities. It is often associated with destruction of the productive sector, distortion of the social fabric and adverse psychological effects which result from loss of lives, property and livelihoods (Collier and Hoeffler 2006). Its effects are often persistent and transmitted across generations. During violent conflict, the most affected category is often the local communities especially the low income groups that are faced with limited livelihood options and subjected to many forms of human rights abuses. Violent conflict often results in wide spread internal displacement which is associated with enormous threats to safety, marginalisation and limited capacity for households to adopt potential livelihood options in order to sustain themselves. Households in displaced communities at times give up or reduce participation in certain income or welfare enhancing activities due to fear of insecurity, pessimism about the end of the conflict, and a number of individual and household -level characteristics (some of which may be conflict-related). With recovery manifested by reduced conflict events and camp decongestion, the livelihood situation is likely to be different. Individuals may be in position to take advantage of peace and recovery initiatives to enhance their capacities to adapt by engaging in a number of income generation activities. Previous contribution to literature in this perspective (Brück, 2004; Brück and Schindler, 2008; Deininger, 2003) has analysed household coping strategies and activity adoption in communities affected by conflict, although much of the existing analysis has cantered on post-conflict scenarios. Due to absence of surveys reflecting war and post-war periods, little attempt has been made to understand how living under conflict stress and recovery from violent conflict may affect household activity options. In the absence of a panel survey for northern Uganda we take advantage of a unique cross sectional dataset collected in IDP camps and return communities to address this gap. The wide spread internal displacement and unstable security situation in the region posed a severed welfare challenge to affected communities. Notable among them was constraints on the number and types of activities that individuals could resort to in order to ensure survival. The recent improvement in security has been followed by a growing number of households leaving camps and reintegrating into their original communities or areas they regard as more secure. The question we attempt to answer here is: How has conflict de escalation (manifested by camp decongestion) influenced activity 2

choices? 2 We posit that there are two ways in which conflict de-escalation eased the constraints on activity adoption. First, given residence status (individual resides an IDP camp or returned home) conflict may affect activity choices. Second, households may choose their location (i.e. leave the camps) and this may also have an impact on activity choice. A bivariate probit model is used to examine the joint decision to stay in or leave the camp and to adopt a certain livelihood activity. Studying activity choices can be important for policies targeting welfare improvement in communities. Empirical evidence suggests that household welfare is in part influenced by the nature of activities that individuals engage in. Households that are able to adopt high valued activities can be in a better position to enhance their wellbeing. Grootaert (1997) for instance notes that self employment can help households in rural areas to escape poverty, but this is only to the extent that households make sizeable investment in their activities. Understanding activity adoptions can also provide insights into possible reasons why income inequalities may exist among households and why some households may be able to catch up with the rest in higher welfare brackets. Adams (2002) findings reveal that income inequality-reducing income sources mostly in the non-farm sector can lead to a decline in the gini coeffient of overall income. On the contrary, an increase in agricultural income may raise income inequality. Understanding adjustment mechanisms among communities during war and its aftermath can go a long way in aiding policy makers and other stakeholders in designing programmes that specifically target these communities, as they ought to differ from other communities living under different conditions. It can be generally argued that reconstruction efforts in communities affected by war warrant specific policy and programme interventions. The conclusions about the most viable interventions can certainly be reached with knowledge about how households cope as well as drivers of activity choices during violent conflict and its aftermath. Our findings underscore the role displacement plays in influencing individual involvement in welfare enhancing activities. We find that living in an IDP camp by far reduces the likelihood of an individual engaging in activities such as crafting. Surprisingly the effect on cultivation and trading is 2 We do not have access to data collected during conflict. We are however able to exploit the uniqueness o the dataset we use here (described later) to shed light on the recovery scenario. We argue that living in an IDP camp may manifest the continued existence of conflict and return to communities may reflect recovery. By combining this data with a conflict events dataset, our analysis brings out the picture. 3

positive. Our findings justify the need for programmes to target livelihoods of displaced communities rather than wait for them to return. In the next section of this paper, we highlight a background to conflict and subsequent displacement in northern Uganda. Section three reviews literature on conflict, household coping and activity choices. In section 4 we describe the data sources. Methodological issues are addressed in section 5. The following section presents results. We then discuss the results and end with concluding remarks. Civil conflict in Northern Uganda: Historical Context Since gaining independence, Uganda has experienced a series of conflicts that have disrupted the country s road to development. However, the most widespread disruption, which started in 1987 has taken place as a result of the long period of fighting between the Lord s Resistance Army (LRA) and the government. While this conflict was initially intended as a popular rebellion against National Resistance Movement (NRM) government, it became a profoundly violent war in which civilians in the northern part of the country have been the main victims. The long period of violent conflict which was more pronounced in the Acholi sub region (and later in lango and some parts of Teso) has been marked by displacement of people from their homes since fighting began. 1996 marked the beginning of widespread and systematic internal displacement following a government strategy to protect the civilians and aid the army's counter-insurgency campaign against the LRA by forcing communities into IDP camps while it pursued a military solution against the rebels. In 2002, the government carried out Operation Iron Fist, a military offensive in Sudan which drove the rebels back into the region. The government strategy of encampment continued, with an estimated 825,000 people displaced in Acholi and parts of Lango sub region. The LRA attacks in Teso and Lango sub regions in mid 2003 increased further the number of displaced people. Less than a year later, estimates were 1.6 million people (Médecins Sans Frontières, 2004), over 90 percent of the population in the region. The protection of civilians in the displacement camps was not effective enough in as many of the most serious massacres and waves of abduction occurred during the time when communities were in camps (Stites, 2006). 4

While living in camps the community has been subjected to poverty, political marginalization, healthcare crisis and strained social bonds. In 2003 lack of national and international response to the massive humanitarian needs in IDP camps led the then UN Emergency Relief Coordinator to describe the humanitarian crisis as the biggest forgotten, neglected humanitarian emergency in the world today. In spite of the camps having existed for several years, by 2006 the army had not provided effective defensive perimeters that would allow camp residents freedom to move and access their farm lands. Less than half of the IDPs could access land that was more than two kilometres outside of their camps, which affected their ability to produce their own food (International Crisis Group, 2006). This limited many households to cultivating small plots along the army-patrolled roadside which were by no means sufficient to feed them (Baines, Stover, and Wierda, 2006). The day-to-day reality of the war has had a negative effect on the wellbeing of communities, including their access to livelihood opportunities. Following peace talks and subsequent attacks on rebel camps by government and allied forces since 2006, the security situation dramatically improved and many of the displaced started returning home, though patterns of return varied widely across regions (Bjorkhaug et al., 2007). By 2007 more than 50 percent of the IDP population had resettled in their home villages or in transit camps with the aid of development agencies. Nonetheless, the region still faces several development challenges to bring it to the same level of development as the rest of the country. According to the National Human Development Report of 2007 (UNDP, 2007) the region s Human Development Index was the lowest, at 0.499 with districts like Gulu, Amuru, Kitgum and Pader scoring lowest on the HDI table in 2007. It registered a high Human Poverty Index of 30.54 making it worse off than many other areas in terms of welfare. The region also shows the lowest probability of one living up to 40 years, has the highest level of illiteracy and the highest percentage of children who are underweight (25%). The rural poverty levels are still high at 68 per cent (UBoS, 2006) and have not registered significant decline as observed in other regions. The conflict has resulted in loss of productivity following deterioration in infrastructure including roads and bridges and markets. Displacement has undermined agricultural production as large tracts of land have remained unused or underutilized during the war period (GoU, 2007). Effective participation in income generating activities has been affected by factors such as closure of active markets, difficulties in accessing credit, and loss of skills (DANIDA, 2005; UBoS, 2006). This partly 5

explains the highest proportion of inactive working-age population in the region, with households mainly relying on transfers from relief agencies as the main source of earning (UBoS, 2006). The Survey of War Affected Youth (Annan et al., 2008), a study documenting realities and ways forward for communities in northern Uganda, reports that in spite of people increasingly getting involved in income generating activities, more than half of youth work fewer than eight days per month and 21 percent of male and 14 percent of female youth work zero days per month. These impediments have resulted in an enormous loss of economic potential, estimated at around US $100 million annually (GoU, 2007). Conflict, household coping and activity choices Understanding household coping during and after crisis requires an insight into the nature of its catalysts. In times of shocks or any events that constrain household welfare, agents often devise means to cope with scenarios that may act to reduce their welfare status (say consumption levels). Households respond differently to negative events, and this variability depends on the nature of events they face and the respective household characteristics (Rashid, Langworthy, and Aradhula, 2006). Valdivia et al., (1996) for instance argue that households with low incomes may seek to engage in a number of activities to ensure higher and stable income flow but ability to diversify may be affected by labour availability in the households. Household with the different labour types are better placed to engage in a wide range of activities. Blocka and Webb (2001) note that while households at higher income levels are in position to adopt meaningful income sources to smooth consumption, poor households may not be in position to adopt effective income generation activities due to resource constraints (Dercon, 2002). It is evident that the ability to cope with income or consumption shortfalls can be enhanced by the capacity of households to engage in certain welfare enhancing activities. A number of empirical studies have underscored the role of household characteristics in enhancing the capacity and decision of households to engage in certain activities. Montmaequette and Monty (1987) in their model of household choice of activities find that the higher the number of years education of a mother or father, the higher is the likelihood of participating in the labour market. A study on determinants of labour market participation in Tanzania (Mduma and Wobst, 2005) also attested to this finding. DeJanvry and Sadoulet (2001) find that the more educated are engaged in non- 6

agricultural activities. Their results also reveal that the asset position of the household affects participation in off-farm activities for farming households. For instance access to land may have a negative effect on household engagement in these activities. They also note that per capita access to land and market access are important factors. In the presence of civil conflict or during the time when features of past conflict are still evident, household ability and willingness to engage in gainful income generation activities may be constrained. Conflict and related fear of insecurity may for instance have an impact on individual abilities to carry on with their known survival strategies (Justino, 2007). In the event of displacement, there is evidence of human capital depreciation manifested by loss of occupation (agriculture activities) at point of origin and difficulties in generating income(ibáñez and Moya, 2006). Lehler (2008) work on northern Uganda reveals a negative impact of conflict on labour force participation of camp residents, with more evidence pointing to males. Kondylis (2007) on the other hand reveals that although displacement reduces the likelihood of individuals being employed by 15 percentage points, displaced men still remain active relative to their settled counterparts. Kondylis attributes this to; loss of assets which induces men to work harder to regain or revive their status quo and; existence of informal labour markets in displaced communities which makes job search different from that of stayers. Destruction of infrastructure as a result of violent conflict increases transaction cost of exchanges in the market which may drive households into subsistence production. Households heavily affected through for instance illness, loss of members and recruitment may opt to restrict their labour to subsistence farming activities and withdraw from other gainful activities (Justino, 2008). In Rwanda, Justino and Verwimp (2006) found a slight increase in participation of male household heads in cultivation and withdrawal from off-farm activities. Deininger (2003) notes that conflict is associated with a high incidence of enterprise mortality in Uganda and that households continue to feel the effects of war for a long time after it has ended. Deininger s findings indicate that the probability to start up non-farm activities reduces for households that were affected by war. Brück (2004) study in Mozambique analyses the long term effects of war on household farm production choices. Findings reveal that households during post conflict period are able to engage in potential income generation activities, but the decisions to participate in them are flexible across household and seasons. Bundervoet (2009) finds that in war regions, wealthier households are more likely to engage in low risk activities whereas in non-war periods, they reduce investment in these activities. 7

Another channel through which conflict can constrain the households capacity to cope through welfare enhancing activities is through reduction or depletion of household resources (Ibáñez and Moya, 2006). This results from death of productive members, asset losses and household splits during displacement. Household that have resources in return areas are able to improve their welfare by making use of the assets they left. Fiala (2009) on the other hand finds that whereas most income brackets are negatively affected in terms of asset holding, for the lowest income quintile, displacement increases household assets mainly due to the presence of supplies from relief agencies, which enhances household ability to cope. For the case of pre-and post genocide Rwanda, resource depleting income activities were evident in household cattle sales which increased with the presence of shocks(verpoorten, 2009). In times of recovery, households begin to rebuild their livelihoods. Verwimp et al. (2009) reveals that not all households may benefit from relief or recovery interventions. In most cases the most vulnerable individuals are not covered by these programmes (Verwimp, Justino and Brück, 2009). Differences in access to assistance may hinder household adaptation. Modelling IDP status and activity choice Individuals are generally assumed to make decisions based on the objective of utility maximization. In what follows, we base our empirical analysis by using an additive random utility model (Cameron and Trivedi, 2005).The underlying utility function depends of specific attributes x and a disturbance term with zero mean such that: U ( x) = β x + ε for adopting a given activity (1) i1 i i i1 U ( x) = β x + ε, otherwise i0 0 i i0 Since utility is random, the ith individual selects the alternative adoption if the utility associated with it is higher than the utility derived from no adoption. Thus, the probability of adoption is given by: 8

P(1) = P( U > U ) i1 i0 P(1) = P( β x + ε > β x + ε ) 1 i i1 0 i i0 P(1) = P( ε ε < β x β x ) P(1) = P( ε < β x ) P(1) = Φ( β x ) i1 i0 1 i 0 i i i i Where Φ is the cumulative distribution function. In the case of normal distribution function, a probit model can be estimated. (2) Our main interest is to examine the effect of residence status on household choice of coping strategies. Ideally we would expect the explanatory variable to be exogenous in nature. In reality however the influence might take two directions. On one hand, the existence of potential livelihood options outside camps might encourage households to leave camps in search for better welfare. It can also be argued that households with better skills, more resources might find it easier to leave camps and reintegrate into the community, leaving behind households with limited attributes. Bjorkhaug et al (2007) study points out the importance of assets such as land on camp decongestion in the region. Those with access to land are therefore able to return to their original homes and cultivate. Evidence in the study suggests that a large proportion of returnees claim inheritance to land compared to their counterparts in camps. In this case therefore, presence of activities and attributes that enhance coping in the community may influence household location. On the other hand, the decision of whether to stay in camps might act to influence the choice of activities by households. Households living outside camps might be in a better position to engage in more valuable coping strategies in the event of better performing markets outside camps. Stites et al., (2006) study in Kitgum district for instance stresses that social capital is higher among households in semi-settled communities than those residing in camps. They are able to participate in collective farming and are able to share proceeds from communal land. On the contrary this is not quite evident within camps. Ignoring this relationship can lead to substantially biased parameter estimates. To take into account this correlation we jointly model IDP status and activity choice. Since the two dependent variables are dichotomous, we adopt a bivariate probit model. The process of generating outcomes for this question involves a simple recursive model (for details see Maddala, 1983). This model is useful 9

when two dependent variables are interdependent (which is the case in our study) or when they depend on a common set of explanatory variables. The basic model can be specified as a set of structural equations involving a dummy endogenous variable. We present a case of two binary related variables. Our endogenous variable y 1i (residence status indicating whether one resides in an IDP camp) and our other dependent variable of y 2i (activity choice) 3 can each be viewed as being generated by unobserved respective latent variables indicated as * y ki. The latent variable assumes a positive value when the underlying observable indicator is equal to one and a negative value when the indicator is equal to zero. That is: y ki ki = > * 1 y ki 0 y = o otherwise In our case, y 1i = 1if the individual is observed to be residing in an IDP camp and 0 otherwise, y = if the individual is observed to be participating in activity i and 0 otherwise. 2i 1 These two variables are linked through the following structural model. y = β X + u * 1i 1i 1i y = γ y + δ X u * 2i 1i 2i 21 ( u, u x, x ) N(0,0,1,1, ρ) 1 2 1 2 (3) (4) Note that our parameter of concern is the effect of the endogenous dummy variable (IDP status) on the discrete outcome. We control for exogenous variables X 1i and X 2i in both equations, and these are assumed to be independent of the error terms u 1i and u 2i. N(..,..,..,.., ρ) indicate the standard bivariate normal distribution with correlation coefficient ρ. When ρ = 0, the model for activity choice is a standard probit model. The recursive probit model estimates are consistent provided that u1 and u2 are bivariate normal. The diagnostic tests for H : 0 0 ρ = would suggest estimating the two equations separately (implying no evidence of endogeneity) in case the null hypothesis cannot be 3 The question in the survey is whether an individual has skills and is actually participating in a certain activity. We were however unable to track individual activities in the past periods since respondents currently participating in certain activities were not asked whether they were involved in the same activities in the preceding periods. 10

rejected. Otherwise the equations should be modelled jointly, since a standard probit model delivers inconsistent estimates. Given the potential correlation of residence status and the decision to adopt certain coping mechanisms, we obtain an instrument that is correlated with settlement decision but not correlated with the error in the livelihood choice equation. It can be deduced that conflict occurrences in the place of birth of household head 4 are exogenous to the households and strongly influence settlement decision. A household may opt to stay in the camp because its safety is not guaranteed given a range of spontaneous attacks outside camps 5. Conflict occurrences in the place of origin of the household head is assumed to have no direct influence (conditional on control variables) on choice of activities but their effects can only be channelled through the impact on IDP status. We therefore use the conflict intensity index at the place of birth of the household head to instrument for IDP status. Because conflict intensity at the place of birth may be a good proxy for conflict at place of current residence, we use the latter as a control variable (in the bivariate probit it appears in both first and second stage equations). Introducing instruments will remove the endogenous element of settlement selection, leaving us with the exogenous portion with which we can analyse effects on activity choices. Construction of the conflict intensity index Here we provide an insight into the derivation of our conflict intensity index (Bozzoli and Brück, 2010). Our starting point is the definition of a conflict event (subscript i ) which in this case includes individual battles (commonly between government and rebels as well as rebel attacks on 4 The household s decision to leave the camp is highly influenced by the head. Ideally a household would prefer to return to their place of origin (Bjorkhaug et al, 2007), and this is most certainly the place of birth of the household head (in the case of male headed households). A number of factors may influence return to the place of origin. Most striking is the fact that land is communally owned. Under this arrangement land is managed by individuals elected by the clan and it consists of grazing land and other land for community facilities such as markets. The clan also allocates land to families for exclusive use. Therefore, for easy access to land (and other family linkages), a household is better of returning to the ancestral home. 5 These may not necessarily be orchestrated by rebels. Deaths for instance might occur as a result of robberies and conflicts among households although these are less profound in the data. 11

communities). We then define c i, a two-dimensional vector representing a GPS coordinate of these individual events (described later). With this information we can calculate a conflict intensity index for the location of the household (expressed in degrees). Aggregating events in a given year (in our case, 2006), the index for location (1) can be defined as: C(1) = i g ( d( c i, l)) (5) where d is the distance between the event and the location of the household at certain point in time, given as: d( c, l) = i c i l. (6) Function g(.) which can be defined as g( x) = exp( α x) discounts events by their distance from a given household. These events are weighted depending on how close they are from the respective individuals or households. The parameter α, which can be interpreted as a distance-discount factor, is chosen by evaluating different values and choosing one with the best fit (joint log-likelihood) in the models. We calculated the conflict indices for discrete choices of α = 5,10,15, 25,30, and the log likelihood function was maximized (over this set of values) forα = 10. We therefore considered this value for every model in our analysis. We consider four activities for our analysis. The choice of these activities is justified by the proportion of the sample engaged in them. The questionnaire provided for a wide range of activities but very few had a sizeable number of participants. We therefore selected activities with more than two percentage points of the sample participating. In this regard, our analysis focuses on cultivation, handcrafts, petty trade and any activity. 12

Data sources and sample description Data for this paper comes primarily from the Northern Uganda Livelihood Survey (2007) 6. The survey was jointly administered by Uganda Bureau of Statistics (UBoS) and the Norwegian FAFO Institute for Applied International Studies. A detailed description of the survey can be found in Bjorkhaug et al., (2007). Here we provide a brief description of the sample. The unique feature of this survey is that it is the first ever comprehensive survey conducted in the region right after the end of war and therefore captures features of both war time (camp situation) and recovery (return/camp decongestion). It covered a sample of 5000 households in six districts (Amuru, Gulu, Pader, Kitgum, Lira and Oyam) using a two-stage cluster design. In the first stage a list of IDP camps, IDP residents and returnees was obtained to determine the number of selection areas in each community (IDP or returnee). The second stage involved determining the location of the selection areas using Global positioning system. Four households in return areas and five in IDP camps were then randomly selected. The questionnaire collected a variety of information regarding characteristics such as demographics, camp situation, return, and household economy for each member of the households sampled. The survey provides information on activities that individuals were currently involved in. This enables us to define our variable of interest, whether an individual currently participates in a certain activity. The key drawback is that no questions were administered to investigate whether the respective individuals were involved in the same activities in the preceding period. Unfortunately, we cannot ascertain income from these activities because information about local prices and total production is missing. It is therefore not possible to tell the share of respective activities to household income. The survey instruments were geo-referenced to the household level and therefore facilitate linking households to conflict events. We also use the Armed Conflict Location and Events Data-ACLED (Raleigh, Hegre and Carlsen, 2009) to obtain information on conflict episodes in the region. The data were obtained from press reports, humanitarian reports, periodicals, books written on particular conflicts and information obtained from the Uppsala Armed Conflict Project achives. Effort was taken to document conflict events (mostly battle events between governments and rebel groups), their dates and actual geographic locations. For Uganda the survey provides information on more than 1,000 individual battle events from 1962 to 2006 about 546 of which were in northern Uganda. This makes it 13

possible to analyse this disaggregated data with geo-referenced household surveys. It is therefore possible to link households and communities to violent conflict occurrences in order to observe their behaviour and responses. In the next section we provide justification for incorporating conflicts events data in the analysis. Table 1 presents the definitions and summary statistics for the variables used in the analysis. These can be classified into individual and household level characteristics, employment sources and location-level variables. Our analysis excludes individuals below 15 years as we do not intend to investigate issues of child labour. We also exclude individuals above 64 years since this category is generally considered inactive (MGLSD, 2006). The sample is representative of both the IDP camp population and the return population. Overall, 65 percent of these individuals are camp residents 7. The sub-region consists of a young population averaging 29 years, and the gender distribution of the population is balanced. As we would expect, statistics indicate the presence of more female headed households in camps than in return communities. The challenge often associated with these households in any community may not be different in times of civil conflict. Difficulties of access to land and livelihood opportunities outside camps may hinder faster reintegration of these households. We also see that individuals who were affected much later by the war spent relatively a short time in camps compared to those who were displaced earlier. This probably explains the higher number of returnees residing in Lango subregion who were displaced between 2002 and 2003, compared to residents in Acholi sub-region who started experiencing displacement as early as 1997 8. A report by Bjorkhaug et al. (2007) indicates faster rate of camp decongestion in the Lango sub-region. 7 This category also includes commuters. These are household members who occasionally resided away from their respective households in camps in order to cultivate land in their original homes or carry out other livelihood activities. They were included because they shared services provided in the camps. 8 Fiala (2009) notes ethnic differences associated with language barrier as one of the factors why rebels concentrated their activities more in Acholi land where the language was familiar and infiltrated the Lango area at later at the peak of the war. 14

The differences in activity choices between camp residents and returnees are not highly marked 9. This could partly be explained by close characteristics between them both at household level and individual level. Close to 87 percent of camp residents were in involved in cultivation, 5 percent in handicrafts and 22 percent were active in petty trading. More individuals in return communities were involved in cultivation(88 percent), crafting (7 percent) and in any activity (96 percent), the proportion of camp residents involved in petty trading was about 5 percentage points higher than their counterparts in return communities. Test for mean differences (table 2) reveal statistical difference in all activities between the two groups except cultivation. We find statistical difference between men and women for crafting and petty trading, with more women involved in both activities. A comparison of women in both groups (not presented here) indicates that more women in return communities are involved in handicrafts (10 percent) compared to their counterparts in camps (7 percent), but the latter dominate in petty trading (7 percent difference). The activity choices for men follow the same pattern as women. Preliminary results Test for exogeneity of IDP status and activity choices? The Wald test for exogeneity of IDP status is rejected for all the activities, implying that the error terms in both equations are not independent. The results and corresponding p-values are shown in table 3. Consequently we have reason to believe that the decision to stay in the camp is an endogenous regressor in the activity choice decision. If this is true, then including this regressor in the probit equation would yield inconsistent (and therefore unreliable) estimates, as stated above. Comparison of the recursive bivariate probit model with probit results (not included) reveals some substantial differences. Whereas the sign of the IDP status variable is negative in the latter, it turns out positive when we control for possible endogeneity. The respective marginal effects (table 5) indicate that ignoring potential correlation results in underestimation of the effect of IDP status on 9 Note that here we concentrate on a few activities most given with a highest proportion of individuals. There is a wide range of activities households participated in but the proportion of these households was less than 2 percent. 15

cultivation, making handicrafts and participating in any activity, whereas the effect on trading is overstated. We also notice changes in significance levels of some coefficients. This further confirms that disregarding possible endogeneity could lead us to drawing wrong policy conclusions. We therefore base our evaluation on bivariate probit results. First stage results: Correlates of IDP status In table 3, we provide results of the first stage regression of the model for all the activities. The instrument (confbirth) appears positive and highly statistically significant at 1 percent level. High conflict intensity at the place of expected return increases the likelihood of individuals residing in the camp. In conformity with the descriptive statistics in table 1, results indicate a high likelihood of camp residence if the head of the household is female. We also find that individuals who lived in communities that were displaced much later (Lango06) are more likely to leave the camps compared to their counterparts in Acholi land. This result is further confirmed by the camp duration variable (campdurat), which indicates that living longer in the camp is associated with higher probability of staying in the camp. Second stage: Determinants of activity choices. Table 4 shows results of the bivariate probit analyses of the probability of participating in the selected activities. The models fit the data well with chi 2=3800.06 (p<.000); 4051.05 (p<.000), 5652.52 (p<.000) and; 4167.98, p<.000 for cultivation, crafting, trading and any activity respectively. We find a strong effect of camp residence on activity choices. The camp residence variable is significant at 1 percent level for all activities. As pointed out earlier, this variable has a positive effect on the decision to cultivate and a corresponding marginal effect of 0.276. If we were to use a model that does not take into account the correlation in error terms, the effect would have been interpreted as negative. Similarly camp residence also has a positive and significant effect on the decision to engage in petty trade. The effect is negative and expected for handicrafts and participating in any activity. Whereas the effect of IDP status is positive for cultivation and trading, living longer in the camp significantly reduces probability of participating in the two activities. Participation in activities is less likely for older people (above 45 years) compared to middle aged people. Households that were previously engaged in trading and herding are highly likely to engage in activities. 16

Discussion of results A positive relationship between IDP status and cultivation may generally be unexpected given the usual challenges of land access faced by displaced communities. However, in the context of northern Uganda, the results may be plausible. First, they might conform to Kondylis (2007) view that individuals living in camps may work harder to reinstate their status quo. Second, individuals living in IDP camps may have limited livelihood options available and therefore opt to cultivate. In the absence of active labour markets farming may be the most obvious fallback position to keep individuals active. Reports indicate that households had access to small plots of land around the camps and produced merely for subsistence to supplement on food aid (Bjorkhaug etal., 2007). It could be that individuals in camps may be more inclined to cultivate but produce less than returnees. However we cannot ascertain much output each category produced. Contrary to recent literature that highlights the impediments of displacement to markets, we find that individuals living in IDP camps are more likely to engage in petty trading. Two possible reasons could be at play. First, given limited income generation sources and inadequate aid to provides for all basic requirements of displaced households, individuals might be engaging in sale of food and other aid to generate income 10. An IDP profiling study for Uganda conducted in 2005 (Bøås and Hatløy 2005) reports that about 14 percent of households sold food aid. Second, sparse population in return areas may discourage trading due to inactive markets. For communities in IDP camps, a large population may provide market for products however meager proceeds might be. Evidence of economic opportunities related to petty trading in IDP camps has been cited as one of the major hindrances to return internal (IDMC and NRC, 2010). The negative effect of IDP status on participation in any activity is not surprising. Overall, displacement may limit household activity options. A number of activities require active markets and peaceful environment and often collapse in times of crisis. During crisis, both the private and public sector become less functional and linkages often break down. Agents redirect their attention to areas with relative calm, which reduces labour opportunities. When conflict deescalates (manifested 10 Therefore any analysis that may be based on household incomes as a proxy of welfare may require careful interpretation as incomes may be derived at the expense of consumption. 17

by camp decongestion), regeneration of the economy may be expected and communities may start to expand their scope of activities. Conclusions In this paper, we provide evidence of the effect of conflict reduction on adoption of livelihood activities. We find that residing in an IDP camp poses both negative and positive effects on decisions to engage in income generation activities. The high likelihood to engage in activities such as petty trading among camp residents may probably be explained by opportunities within or around these settlements which returnees may not have access to especially at the start of recovery. In general however, return may enhance individual capacities to engage in a range of activities which may not have well functioning markets around camps. In spite of evidence of loss of skills among displaced communities (Kondylis, 2007), our results reveal that individuals in camps are active. This observation underscores the need for livelihood interventions and other recovery programmes to target return communities but also create opportunities for individuals and households still in displacement. Programmes that by-pass displaced communities may instead constrain their capacity to recover after return. References Adams, R. H, 2002. Nonfarm income, inequality, and land in rural Egypt.. Economic Development and Culture Change 50 (2 ):339-63. Annan, J., C. Blattman, K. Karlson, and D. Mazurana, 2008. The state of female youth in northern Uganda. Findings from Phase II of the Survey of War Affected Youth (SWAY). Baines, E., E. Stover, and M. Wierda. 2006. War affected children and youth in northern Uganda: Toward a brighter future. An assessment report. Bjorkhaug, I., B. Morten, A. Hatloy, and K. M. Jennings. 2007. Returning to Uncertainity? Addressing Vulnerabilities in Northern Uganda. United Nations Development Programme Blocka, S., and P. Webb, 2001. The dynamics of livelihood diversification in post-famine Ethiopia. Food Policy 26:333 350. Bøås, M. and A. Hatløy, 2005. Northern Uganda internally displaced persons profiling study: Office of the Prime Minister, Department of Disaster Preparedness and Refugees, Kampala. Bozzoli, C., and T. Brück, 2010. Child morbidity and camp decongestion in post-war Uganda. MICROCON Research Working Paper 24 18

Brück, T., 2004. Coping Strategies in Post-War Rural Mozambique. HiCN Working Paper 02 Brück, T., and K. Schindler, 2008. The Impact of conflict and fragility on households. A conceptual framework with reference to widows. UNU-WIDER. Research Paper No. 2008/83 Bundervoet, T, 2009. Assets, activity choices and civil war: Evidence from Burundi. Worl Development doi:10.1016/j.worlddev.2009.12.007. Cameron, A.C., and P.K. Trivedi, 2005. Microeconometric methods and applications.: Cambridge University Press. Collier, P., and A. Hoeffler, 2006. Civil war. Draft chapter for the Handbook of Defence Economics. : Department of Economics, University of Oxford. DANIDA, 2005. Restoration of Agricultural Livelihoods in Northern Uganda Component (RALNUC): Agricultural Sector Programme Support. Ministry of Foreign affairs, Denmark. Draft Report. De Janvry, A. and E. Sadoulet, 2001. Income strategies of rural off-farm households in Mexico: The role of off-farm activities. World Development Vol. 29, (3):467-480. Deininger, K., 2003. Causes and consequences of civil strife: Micro-level evidence from Uganda. Oxford Economic Papers 55:579-606. Dercon, S., 2002. Income risk, coping strategies, and safety nets. World Bank Research Observer 17 (2):141-166. Fiala, N., 2009. Consequences of forced displacement in Northern Uganda: Department of Economics, University of California. GoU., 2007. The Peace Recovery and Development Plan for Northern Uganda. 2007-2010: Office of the Prime Minister. Kampala. Grootaert, C., 1997. Determinants of poverty in Côte D"Ivoire in the 1980s. Journal of African Economies. 6 (2):169-206. Ibáñez, A. M., and A. Moya, 2006. The impact of intra-state conflict on economic welfare and consumption smoothing: Empirical evidence for the displaced population in Colombia. HiCN Working Paper 23 Internal Displacement Monitoring Center and Nowegian Refugee Council, 2010. Peace, Recovery and Development: Challenges in Northern Uganda. March 2010. International Crisis Group, 2006. A strategy for ending northern Uganda s crisis. Policy Briefing N 35 Justino, P., 2007. On the links between violent conflict ad household poverty: How much do we really know?, edited by B. M. MICROCON Research Working Paper 1.., 2008. The impact of armed conflict on household welfare and policy responses. Background paper, World Economic and social survey 2008, United Nations 19

Justino, P., and P. Verwimp, 2006. Poverty dynamics, Violent conflict and convergence in Rwanda: HiCN Working Paper 16. Kondylis, F, 2007. Conflict-induced displacement and labour market outcomes: Evidence from post-war Bosnia and Herzegovina. Lehler, K., 2008. Gender differences in labour larket participation during confict:evidence from displaced people's camps in northern Uganda. Job Market Paper: University of British Columbia Maddala, G.S., 1983. Limited dependent and qualitaative variables in econometrics. Cambridge University Press, Cambridge:242-247. Mduma, J.K., and P. Wobst, 2005. Determinants of rural labor market participation in Tanzania. African Studies Quarterly 8 ( 2). Médecins Sans Frontières, 2004. Life in northern Uganda. All shades of grief and fear. Annual report Montmaequette, C., and L. Monty,1987. An empirical model of household s choice of activities. Journal of Applied Econometrics 2:145-158. Raleigh, C., H. Hegre, and J. Carlsen, 2009. Introducing ACLED: And armed conlflict location and events dtaset. : Peace Research Institute, Oslo. Rashid, D. A., M. Langworthy, and S. Aradhula, 2006. Livelihood shocks and coping strategies: an empirical study of bangladesh households. Paper prepared for presentation at the American Agricultural Economics Association annual meeting, Long Beach, Califonia, July23-26, 2006 Stites, E., 2006. Humanitarian agenda 2015: Northern Uganda country study: Feinstein International center. Briefing Paper. Stites, E., M. Dyan, and C. Khristopher, 2006. Movement on the Margins:Livelihoods and Security in Kitgum District, Northern Uganda: Feinstein International Centre. UBoS, 2006. Uganda National Household Survey 2005/2006: Uganda Bureau of Statistics. Kampala., 2006. Uganda National Household Survey 2005/2006: Socioeconomc module. : Uganda Bureau of Statistics. Kampala. United Nations Development Programme, 2007. Uganda Human Development Report 2007. Rediscovering agriculture for human development. : United Nations Development Programme. Valdivia, C., E.G. Dunn, and C. Jetté, 1996. Diversification as a risk management strategy in an Andean agropastoral Community. American Journal of Agricultural Economics, 78 (5 ):1329-1334. Verpoorten, M., 2009. Household coping in war- and peacetime: Cattle sales in Rwanda, 1991 2001. Journal of Development Economics 88. :67-86. Verwimp, P., P. Justino, and T. Brück, 2009. The analysis of conflict: A micro-level perspective. Journal of Peace Research. 46 (3):307 314. 20

Table 1: Descriptive statistics (age category 15-64 years) Full Sample Residents Returnees Description n= 5329 3396 1933 Individual characteristics Age_15_30 Individuals aged between 15-30 years 0.627 0.625 0.63 Age_46_60 Individuals aged between 31-45 years 0.267 0.271 0.261 Age_46-64 Individuals aged between 46-64 years 0.106 0.104 0.109 Female Individual is female (1=Female, 0, Male 0.478 0.5 0.445 Head Dummy=1 if head, 0, otherwise 0.355 0.352 0.359 Spouse Dummy=1 if spouse 0, otherwise 0.252 0.263 0.234 Single Dummy=1 if single, 0, otherwise 0.339 0.338 0.341 Literate Dummy=1 if individual is literate, 0 otherwise 0.575 0.555 0.606 Household Characteristics Resides in a household headed by a Femhead female; 1=yes, 0, 0therwise 0.207 0.244 0.148 Depratio Dependency ratio 1.263 1.258 1.27 Hssize Number of people in the household 6.758 6.5 7.163 Headeverherded Head has ever herded animals; 1=yes, 0, otherwise 0.661 0.634 0.702 Headevertraded Head has ever traded; 1=yes, 0, otherwise 0.403 0.414 0.386 Activity choices Cultivates currently cultivating; 1=yes, 0, otherwise 0.869 0.862 0.881 currently cmaking handicrafts; 1=yes, 0, Crafts otherwise 0.056 0.047 0.071 Trade currently trading; 1=yes, 0, otherwise 0.197 0.215 0.167 Anyactivity engaged in any activity income gen (including those not listed here) 0.937 0.925 0.956 Location variables Campresid Camp resident ; 1=yes, 0, otherwise 0.654 Campdurat Duration in camp 6.212 7.463 4.247 Confhere Conflict index at current location 10.399 11.738 8.293 Confbirth conflict index t place of birth of head 23.525 23.959 22.842 Lango06 Lived in Lango sub region in 2006 0.413 0.225 0.708 Amuru Amuru District 0.091 0.131 0.028 Gulu Gulu District 0.139 0.196 0.05 Pader Pader District 0.235 0.259 0.197 Kitgum Kitgum District 0.121 0.189 0.015 Kira Lira District 0.16 0.069 0.303 Oyam Oyam District 0.254 0.156 0.407 21