Violence and the formation of hopelessness: Evidence from internally displaced persons in Colombia

World Development 113 (2019) 100–115

Contents lists available at ScienceDirect

World Development journal homepage: www.elsevier.com/locate/worlddev

Violence and the formation of hopelessness: Evidence from internally displaced persons in Colombia Andrés Moya a,⇑, Michael R. Carter b a b

Economics Department, Universidad de los Andes, Calle 19A # 1-37E, Bogotá 111711, Colombia Agricultural and Resource Economics, University of California, Davis and NBER, United States

a r t i c l e

i n f o

Article history: Accepted 26 August 2018

JEL classification: D1 C9 O1 I1 I3 Keywords: Violence Forced displacement Trauma Beliefs Hopelessness Colombia

a b s t r a c t We explore the impact of violence on beliefs about socioeconomic mobility. For this purpose, we bring together data on the severity of the household-level experience of violence, symptoms of psychological trauma, and subjective probabilities of future positions in a ladder of life for a sample of internally displaced persons in Colombia. After controlling for current socioeconomic circumstances and asset losses, we find that the experience of more severe violence dampens the beliefs about socioeconomic mobility. The estimated impacts are large: a one standard deviation increase in the number of violent events experienced by the household raises the perceived probability of extreme poverty in the following year by 54 percent relative to the mean. In the long run, the expected likelihood of extreme poverty is almost three times higher for victims at the 4th quartile of the distribution of the severity of violence than for victims at the 1st quartile. Additional evidence suggests that psychological trauma explains this result, identifying a channel by which these hopeless beliefs can become self-confirming. Together, the results point to the existence of a behavioral poverty trap and highlight the importance of rethinking the strategies to promote the socioeconomic recovery of victims of violence. Ó 2018 Elsevier Ltd. All rights reserved.

You know, doctor, it’s been a few nights since I do not sleep, I have dreams where I see the heads of my neighbors. I see that they cry, that they supplicate, ask for mercy. I wake up crying. I start thinking about the farm, about my plants in the garden, about our chickens and cattle, and about our dogs that wanted to come with us, but we had to scare them away with rocks so that they would not follow us. I had never felt this way. I had never seen my husband so quiet; I had never seen him cry in silence. [. . .] I do not know what’s going to happen with us, only that we have God and that our life will never be the same since we are now displaced.1

⇑ Corresponding author. E-mail addresses: [email protected] (A. Moya), [email protected] (M.R. Carter). 1 Doctors Without Borders (2010): Testimony of a woman in Florencia, Caquetá, who was displaced from her hometown after an armed group killed and carved up some of her neighbors, and made her bury them [Own translation]. https://doi.org/10.1016/j.worlddev.2018.08.015 0305-750X/Ó 2018 Elsevier Ltd. All rights reserved.

A victim of the tactics of terror that characterized the battle between warring factions in the Colombian countryside, the women quoted above poignantly testifies to both the loss of material assets, as well as the psychological damage created by her experience of violence and forced displacement. Her account of the experience of violence and her current circumstances reveals a sense of hopelessness regarding the future. In this article we explore whether violence dampens the beliefs about future socioeconomic mobility and discuss the implications of this possible behavioral effect for poverty dynamics. A large body of micro-level studies has documented that individuals who experience violence suffer devastating, and sometimes permanent, socioeconomic consequences.2 Violence brings about a severe loss of conventional physical assets that impose severe constraints on the victims’ livelihoods (Ibáñez & Moya, 2010a, 2010b). Such asset losses can thrust victims into chronic poverty and dampen victims’ beliefs about what is possible to achieve.

2 See Blattman and Miguel (2010), Ibáñez and Moya (2016), and Justino (2011) reviews of the literature on the micro-level consequences of violence, war, and civil conflicts.

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

Yet, the experience of violence also has profound psychological consequences that can further damage the beliefs about future economic progress and recovery. In doing so, the psychosocial consequences of violence can create a behavioral poverty trap of its own, akin to that which can result from the loss of conventional assets. Our hypothesis, which we describe in more detail in the following section, builds upon a substantial body of research on the psychological consequences of violence. Studies in clinical and social psychology have documented that victims who experienced more severe violence are at increased risk of suffering from psychological trauma, including major depression, chronic anxiety, and posttraumatic stress disorder among others (Kessler et al., 1996; Mollica et al., 1998; Yehuda, 2002). The nature of the traumatic experiences of violence, in addition, may induce hopelessness; this is, an emotional state of despair and pessimism based on the expectation negative outcomes and the perception of few pathways for progress and recovery (Abramson, Metalsky, & Alloy, 1989; Sympson, 2000; Yehuda, 2002). Therefore, the psychological constraints associated with the experience of violence can reinforce the effect of the more discernible material constraints and lead victims to believe that there are even fewer prospects for socioeconomic mobility.3 In doing so, violence and the resulting psychological trauma can hinder economic behavior and induce a vicious cycle of hopelessness, underachievement, and persistent poverty. The link between victims’ beliefs and poverty dynamics rests on the observation that economic behavior is ultimately driven by what individuals believe is possible to achieve and what they aspire to achieve (Appadurai, 2004; Dalton, Ghosal, & Mani, 2016; Duflo, 2012; Lybbert & Wydick, in press; Ray, 2006). For instance, individuals may cease to aspire and strive for things that do not seem possible to achieve in order to cope with diminished prospects of socioeconomic mobility. If victims believe that there are even fewer prospects for socioeconomic progress, they can become discouraged from making the investments that are required to move out of poverty, from gathering information about possible pathways for progress, and even from modifying existing beliefs and perceptions. Hence, a negative shock that has psychological consequences may diminish human capabilities and economic performance as explored by de Quidt & Haushoffer (2019) and generate a self-sustaining behavioral poverty trap. To analyze the effect of violence on beliefs about socioeconomic mobility, we sampled 344 victims of violence and forced displacement in Colombia. In section 2, we describe our sampling strategy and data, which includes the household-level experience of different violent events, such as murders, threats, and massacres; symptoms of psychological trauma, including depression and anxiety; and beliefs about socioeconomic mobility. For the latter, we built upon the work of Krishna (2001, 2004, 2006) and Narayan, Pritchett, and Soumya (2009) and constructed a ladder of life appropriate for the study population in which each step of the ladder represented a level of wellbeing defined over five dimensions. Then, we followed Delavande, Gine, and McKenzie (2011) to elicit the subjective probabilities of being at each step of the ladder of life the following year—our core measure of victims’ beliefs about socioeconomic mobility. Our empirical strategy exploits the variation in the severity of violence, defined as the number of violent events experienced by each household, conditional on the household’s current position on the ladder of life and previous asset losses. Consistent with our hypothesis, we find that violence dampens the beliefs about 3 Barrett, Carter and Chavas (2019) and Lybbert and Wydick (2018) discuss the coexistence and interactions between external (material) and internal (psychological) constraints and poverty traps.

101

socioeconomic mobility at the intensive margin: an increase of a one standard deviation in the number of violent events experienced by the household raises the perceived probability of being at the bottom step of the ladder of life by 54 percent relative to the mean. Hence, victims who experienced more severe violence believe that there are even fewer prospects for socioeconomic progress, a perception that precisely reveals a sense of hopelessness. Since our empirical strategy controls for the current position on the ladder of life and other material restrictions, the results point to the psychological constraints that result from the experience of violence. In fact, we find suggestive evidence that the effect of violence is explained by the severity of symptoms of depression. We discuss these results in detail in Section 3. The attribution of causality to these findings requires the assumption that the patterns of violence were exogenous to victims’ pre-violence beliefs. Our results would be invalid if the severity of violence was associated to pre-violence characteristics that were correlated with less mobility or more hopeless beliefs. This kind of targeting seems most unlikely; if anything, armed groups in Colombia targeted community activists and leaders who would be expected to have ex ante higher mobility prospects and more optimistic beliefs (Centro Nacional de Memoria Histórica [CNMH], 2013). Furthermore, our data come from a particular time and a place in which armed groups were directly competing for terrain and eschewed targeted for indiscriminate violence towards the civilian population (CNMH, 2013). In fact, we demonstrate that the variation in the severity of violence is not correlated with victim’s pre-violence characteristics. Moreover, we test for the robustness of our results using the subsample of subjects who were victimized en masse with clearly no individual targeting. Finally, the results on the underlying psychological mechanism lend further credence to the notion that the estimated effect of violence is explained by the consequences of the traumatic experience of violence and not by a pattern of violence that targeted those with low mobility. To illustrate the implications of our results, in Section 4 we connect with more conventional analysis of poverty dynamics (e.g., Carter & May, 2001; Carter & Barrett, 2006). We use the estimated effects of the severity of violence to estimate transition matrices that reveal the expected long-run patterns of socioeconomic mobility. On average, we observe that the expected long-run extreme poverty rate will be almost three times higher for victims at the top quartile of the distribution of the severity of violence perceive than for those at the bottom quartile of the distribution. Our study builds upon an emerging body of research on the behavioral consequences of violence. Research in this area has identified that violence brings changes in political behavior and collective action (Bellows & Miguel, 2009; Blattman, 2009), prosocial behavior (Bauer, Cassar, Chytilova, & Henrich, 2014; Bauer et al., 2016; Cassar, Grosjean, & Whitt, 2013, 2014), risk aversion (Callen, Isaqzadeh, Long, & Sprenger, 2014; Moya, 2018; Voors et al., 2012), and time preferences (Voors et al., 2012). We contribute to this literature in the following ways. First, we explore a dimension of behavior that had not been analyzed yet in the context of violence and that can contribute to the persistence of poverty among victims of violence.4 Second, our sampling strategy, which focused on IDPs, and the detailed data on the experiences of violence allows us to move away from dichotomous or aggregate measures of the exposure to violence, and instead explore the effects

4 However, our study is not entirely new. Writing in the 1960’s, a few years after the civil conflict known as La Violencia, Lipman and Havens (1965) compared the degree of personality disorganization between a small sample of victims of forced displacement and a sample of urban poor. Among others, the results indicated that victims did not look forward to the future, which the authors interpreted as indicating the effect of forced displacement on hopelessness.

102

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

of the experience of violence on victims’ beliefs at the intensive margin. In doing so, we are able to highlight considerable heterogeneity in the victims’ behavioral responses that emerges as a function of the severity of the experience of violence in conformity with the doseresponse relationship between violence and psychological trauma (Mollica et al., 1998). Third, the data on symptoms of psychological trauma represents an improvement over previous studies since it allows us to provide an initial assessment of the underlying psychological mechanism through which violence affects victims’ beliefs. Our study also speaks to an interdisciplinary literature that has analyzed the nature of beliefs and aspirations, and their relation to poverty (Appadurai, 2004; Dalton, Ghosal, & Mani, 2016; Duflo, 2012; Lybbert & Wydick, in press; Ray, 2006). The set of empirical studies in this area provides evidence on the way in which the relief of information and material constraints has positive impacts on the aspirations of the poor, and on the existence of psychological multipliers that can create virtuous cycles (Beaman, Duflo, Pande, & Topalova, 2012; Bernard, Dercon, Orkin, & Taffesse, 2015; Chiapa, Prina, & Parker, 2016; Glewwe, Ross, & Wydick, 2017; Jensen & Oster, 2009; Macours & Vakis, 2014). Our study, while not focused on aspirations per se, speaks to this body of work by exploring the opposite scenario of a negative shock and understanding how it brings about psychological constraints that can create vicious cycles. psychological multipliers that can create virtuous cycles. The study more closely related to ours is the work of Kosec and Hyunjung (2017), who analyze how a natural disaster affected aspirations in rural Pakistan and suggest that that the adverse effects on aspirations operated through psychological constraints. In Section 6, we conclude by discussing the policy framework for victims of violence in Colombia and illustrating how the standard asset-transfers may be unable to alter long-run poverty dynamics. We discuss how our results emphasize the importance of rethinking the strategies to promote the recovery of the victims of violence and other populations exposed to traumatic shocks.

1. Conceptual framework: violence, poverty, and psychological trauma in Colombia Colombia has endured decades of violence and civil conflict. In the late 1940s, political disputes and decades of tension between landlords and peasants led to a period of civil conflict known as La Violencia (1948–1958). Despite the signing in 1958 of a peace agreement, violence persisted, and leftist guerrillas and rightwinged paramilitary groups emerged soon after. Starting in the 1980s, illegal armed groups became heavily involved in the illicit drug production and trade, leading to the escalation of violence and increasing patterns of civilian victimization (CNMH, 2013). In the last decade, conflicts dynamics were altered by three major events: the demobilization of paramilitary groups in 2006; the emergence of neo-paramilitary factions that clashed for the control of regions previously under paramilitary control; and the demobilization in 2017 of the Revolutionary Armed Forces of Colombia (FARC), the largest and oldest guerrilla group in the hemisphere. Our data was collected prior to the 2017 peace agreement between the Colombian government and the FARC, during the time period in which FARC and neo-paramilitary forces were contesting for new terrain. Throughout the civil conflict, violence towards civilians has not been accidental. Instead, it has been a deliberate strategy of armed groups who rely on vicious and indiscriminate violence to spread fear and gain control of contested territories (CNMH, 2013). As a result, more than 8.5 million civilians have been victimized since 1985, including 7.4 million internally displaced persons (IDPs) (National Victims Unit, n.d). The latter figure represents 15 percent

of the country’s population and is the highest in the world (United Nations High Commissioner for Refugees [UNHCR], 2016). Violence has severe socioeconomic and psychological consequences. A substantial body of work highlights that violence can drive current and future generations into chronic poverty through income and asset losses, forced migration, disruption of local markets, child malnutrition, and school interruption, among other channels (Blattman & Miguel; 2010; Ibáñez & Moya, 2016; Justino, 2011). In Colombia, for instance, victims have been largely displaced from rural to urban areas. As a result, they were forced abandoned their lands and productive assets, lost their social networks, and are unable to find suitable employment opportunities since their agricultural skills are not well suited for urban labor markets. This constitutes a massive loss of assets that hinders their livelihoods, leading to lower levels of income and consumption and to the adoption of costly coping strategies that jeopardize their future wellbeing (Ibáñez & Moya, 2010a, 2011b). Not surprisingly, 65 per cent of IDP are below the official poverty line. In addition, the literature in social and clinical psychology finds that individuals who experience violence are at increased risk for psychological trauma (Briere & Spinazzola, 2005; Kessler, Sonnega, Bromet, Hughes, & Nelson, 1995; Mollica, McInnes, Poole, & Tor, 1998; Yehuda, 2002). The susceptibility to psychological trauma varies across individuals, but, in general, follows a dose-response relationship: this is, more severe and recent episodes of violence increase the likelihood and severity of psychological trauma (Mollica et al., 1998). Recent studies have documented that victims of violence and forced displacement in Colombia suffer an array of mood disturbances and psychopathologies including anxiety, major depression, complex trauma, and posttraumatic stress disorder (Doctors Without Borders, 2006; Richards et al., 2011; Shultz et al., 2014). The material and psychological consequences of violence bring about two mechanisms through which violence can dampen victims’ beliefs about socioeconomic mobility. The first mechanism corresponds to the victims’ cognitive assessments of their new circumstances and how they hinder actual mobility prospects. For instance, victims can recognize that the loss of assets pushed them down the endowment space, imposed considerable obstacles for upward mobility, and maybe even destined them to a lower level of wellbeing. This process can be reinforced by the observation of persistent poverty among victimized peers, which following Ray’s (2006) concept of an aspiration window, provides information on what is possible to achieve.5 The recognition of the material constrains brought forth by violence can therefore lead victims to update their beliefs and perceive (perhaps correctly) that there is a higher likelihood of remaining in poverty. The second mechanism is driven by the traumatic episodes of violence and the resulting psychological consequences. Psychological trauma, in particular, overwhelms the individual’s coping resources, depletes cognitive resources, and perpetuates states of avoidance (Yehuda, 2002). Trauma, in addition, restricts an individual’s goals to surviving physically and psychologically. These factors lead to pessimistic explanatory styles and to an emotional state of hopelessness that is characterized by a loss of agency and by perceptions of few pathways for recovery and progress (Abramson et al., 1989; Sympson, 2000).6 Taken together, the nature of the traumatic experiences of violence can lead victims to

5 Ray (2006) does not refer explicitly to beliefs or expectation, but rather to aspirations. However, his notion of an aspiration window connects with beliefs, as individuals take experiences of others as a basis for what they themselves may be able to achieve (beliefs) which determines what they can aspire to achieve. 6 The loss of agency is also related to a shift towards an external locus of control: the perception that the individual is unable to control the factors that shape her life (Rotter, 1966).

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

magnify the obstacles for upward mobility and perceive (perhaps incorrectly) that there are even fewer prospects for social mobility. In doing so, violence can inhibit economic behavior and discourage individuals from making the necessary investments that are required to move out of poverty. This creates a mechanism through which the pessimistic beliefs about socioeconomic mobility can become self-confirming and speaks directly to the learned helplessness theory of depression (Seligman, 1975; Garber & Seligman, 1980). This theory suggests that the exposure to incontrollable and traumatic events conditions individuals, leading them to believe that they cannot alter their circumstances. As a result, individuals become unable to respond to external stimuli and to undertake actions towards improving their circumstances. Such conditioning persists, even when the external constraints are not binding anymore and when there are actual pathways for recovery and progress. Importantly, we do not expect that psychological trauma and hopelessness will be prevalent among all victims of violence. On the contrary, some victims may instead experience ‘‘posttraumatic growth”, a phenomenon characterized by a greater sense of personal strength and the recognition of new possibilities or paths for one’s life, among other domains of personal growth (Tedeschi & Calhoun, 2004). Posttraumatic growth has been precisely suggested as a reason behind the positive effects of violence on political engagement and risk aversion (e.g. Blattman, 2009; Voors et al., 2012). Yet, to the extent that violence and psychological trauma follow a dose-response relationship, we expect that the psychological mechanism will operate at the intensive margin, as a function of the severity of violence. In other words, that victims who experienced more severe violence will exhibit hopeless beliefs about the future and their socioeconomic mobility. 2. Sample design and data We employ a complex array of data, including the householdlevel experience of different violent events, psychometric measures of symptoms of psychological trauma, beliefs about socioeconomic mobility, as well as conventional living standards data. After reviewing the sample design, we detail the instruments employed to collect these data and report descriptive statistics in each of these core areas. 2.1. Sample design In 2011, we conducted fieldwork in the departments of Córdoba and Sucre, in Colombia’s Atlantic region, and Tolima, in the Central region. We chose these departments as a first step to address the concerns of violence having been targeted. The three departments were the scenario of struggles between different armed groups for the control of three geographical corridors.7 Consistent with the dynamics of civilian victimization in civil conflicts (Kalyvas, 2006), anecdotal evidence suggests that armed groups relied on vicious and indiscriminate violence towards civilians to spread fear among the population and gain control of these strategic regions (Human Rights Watch, 2010; CNMH, 2013). This pattern of violence makes it likely that the number of violent events experienced by any household was orthogonal to individual characteristics, creating quasiexperimental variation in the severity violence. 7 In the Atlantic region, neo-paramilitary groups emerged after the paramilitary peace demobilization in 2006 and clashed over control of the Nudo del Paramillo and Montes the María—two corridors with favorable conditions for the illegal drug trade (Human Rights Watch, 2010). In the Central region, the FARC retreated to the Cañon de las Hermosas—a corridor in the Central Andes that facilitates the movement of troops and the trafficking of illegal drugs to the Pacific Ocean, after the Colombian military intensified its operations against this group in 2004 (National Ombudsman’s Office, 2009).

103

In each department, we used administrative data from the National Victims Unit to identify the main municipalities of residence of internally displaced persons (IDP) who had been victimized and displaced from the vicinity of the above corridors.8 We visited these municipalities and organized community meetings where we explained the projects’ objectives, including the recall of experiences of violence, and highlighted that participation was voluntary and that participants could opt-out at any time.9 In each meeting, we invited one-third of interested individuals to participate and 94 percent of them accepted. Those who declined had been victimized recently—less than 6 months before—and it is likely that they were suffering from severe symptoms of psychological trauma. We argue that this source of selection should would work against our hypothesis, since we expect that more severe symptoms of trauma are associated with more hopeless beliefs. This argument holds unless we indeed missed on victims experiencing severe symptoms of trauma and the effect of trauma follows an inverted U-shape. While we admittedly cannot rule out the latter, we were successful in sampling victims with critical levels of psychological trauma. For instance, the extent of symptoms of anxiety and depression observed in our sample far outweighs those of Colombian general population, as measured by the 2015 National Mental Health Survey. We sampled 344 IDPs who had been victimized in, and displaced from, rural areas of 34 different municipalities and who were residing in the urban locations where fieldwork was conducted.10 Fig. A1 in the Appendix illustrates the geographical distribution of the municipalities from which the victims in our sample had been displaced. The figure also illustrates the intensity of displacement across Colombian municipalities to highlight how the regions where we conducted fieldwork had been severely torn by violence. Importantly, the sample includes 132 individuals from nine different communities who had been victimized and displaced en masse by cross fire from armed groups. In the following section, we will argue that individual targeting was likely non-existent in the context of these mass displacements.

2.2. Data In each municipality, enumerators first administered a household survey during weekdays. After all surveys were completed, we organized weekend sessions where enumerators administered the victimization questionnaire and psychometric scale in private, and we then led a group activity to elicit subjects’ beliefs about socioeconomic mobility. Since we collected sensitive data and conducted fieldwork in municipalities torn by violence, data were collected at the local church to guarantee safe and private environments for respondents and enumerators.11 8 Departmental capitals: Sincelejo and Ibagué in the departments of Sucre and Tolima, respectively. Urban centers: Tierralta and Montelíbano in the department of Córdoba. 9 These meetings were organized with the support of local officials, ombudsmen, and Catholic priest, all of whom were recognized and trusted by victimized and displaced communities. Their support was instrumental to overcome some of the challenges of conducting fieldwork in contexts of civil conflict, such as interacting with victims, obtaining their trust, ensuring the safety of participants, and collecting sensitive information on the experiences of violence. 10 Moya (2018) uses this sample and data to analyze the effect of violence on risk attitudes. To rule out endogenous geographic sorting, he further restricts the analysis to 284 victims who had also resided in the region for more than 10 years. In our paper and in Moya (2018) results are robust if we use the full sample, or the more restrictive sample. 11 Moya (2018) describes the ethical considerations in collecting sensitive data from victims, and the strategies we implemented to mitigate negative effects on subjects. These include, stressing that participation was voluntary and that subjects could skip specific questions or entire modules, defining a protocol for treatment of special cases, and special training prior to fieldwork on strategies for emotional containment.

104

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

Table 1 Experience of violence.

A. Violence Victim: experienced at least one violence event (=1) Severity: number of violent events experienced Temporal proximity: years since the experience of violence Hh member was caught in the middle of armed combat (=1) Hh member received a threat (=1) Hh member was murdered (=1) Hh member experienced one attack (=1) Hh member witnessed and survived a massacre (=1) Hh member was ordered to migrate (=1) Hh member experienced a different violent event (=1) B. Psychological Trauma Depression – % Above Clinical Cutoff (T > 63) Anxiety – % Above Clinical Cutoff (T > 63) GSI – % Above Clinical Cutoff (T > 63) Observations

Total

Number of Violent Events Experienced

[1]

Below the median [2]

Above the median [3]

0.93 [0.253] 6.64 [8.293] 2.52 [3.318] 0.50 [0.501] 0.55 [0.498] 0.24 [0.430] 0.15 [0.359] 0.08 [0.272] 0.42 [0.494] 0.26 [0.439]

0.86 [0.345] 1.69 [0.985] 2.53 [3.470] 0.30 [0.459] 0.40 [0.491] 0.11 [0.310] 0.036 [0.186] 0.01 [0.0772] 0.35 [0.477] 0.12 [0.325]

1.00 [0.000] 11.60 [9.361] 2.51 [3.170] 0.71 [0.456] 0.71 [0.456] 0.38 [0.487] 0.27 [0.444] 0.16 [0.363] 0.49 [0.501] 0.40 [0.491]

0.38 [0.486] 0.27 [0.442] 0.21 [0.405] 336

0.35 [0.477] 0.23 [0.423] 0.17 [0.379] 168

0.41 [0.493] 0.30 [0.459] 0.24 [0.427] 168

Notes: Summary statistics on household-level experience of violence and symptoms of psychological trauma. Column 1 reports sample statistics for the full sample, while columns 2 and 3 report statistics according to an arbitrary stratification of the data: whether the household experienced a number of violent events below or above the sample median. Standard deviations are reported in brackets.

We describe each instrument below and report sample statistics for an arbitrary stratification of the sample: subjects whose households experienced a number of violent events below or above the sample median. This stratification allows us to foreshadow our hypothesis and later empirical strategy, in which we exploit the variation in the number of violent events experienced by each household—our measure of the severity of violence.12 2.2.1. Household-level experience of violence At the beginning of the weekend session, enumerators privately administered a victimization questionnaire that measured whether a household member experienced different types of violent events, and the number of times that each event had occurred in the last 10 years. We also recorded the number of years since the displacement, which we use as a proxy for the temporal proximity of violence.13 Panel A of Table 1 presents sample statistics on the nature and severity of violence. The data highlights three important features: First, all subjects in the sample had been displaced and 93 percent of them experienced at least one violent event in the previous 10 years.14 The most frequent events included being threatened by an armed group (55%), being caught in the cross fire of armed combat (50%), suffering the murder of a household member (24%), suffer12

We can also measure the severity of violence through a victimization score constructed through principal component analysis. 13 Although violence did not necessarily occur at the same time, subjects stated that they were displaced following a peak in violence. The number of years since the displacement therefore captures the temporal proximity of the moment when violence was at its highest. 14 In Table 1 and in the rest of the analysis, we drop 8 outliers who reported a number of violent events more than 5 standard deviations above the mean. Results are robust if we keep these observations.

ing a violent attack (15%), and/or experiencing and surviving a massacre (8%). Second, subjects experienced an average of 6.6 violent events and the victimization occurred 2.5 years on average before the data collection. Third, there is considerable variation in the severity and temporal proximity of violence. 2.2.2. Symptoms of psychological trauma Next, enumerators privately administered a psychometric scale based on a local adaptation of the Symptoms Checklist 90 R (SCL 90-R). This scale measured the experience of symptoms that are associated with different manifestations of psychological trauma over the previous three months.15 The responses to different subsets of symptoms measure the extent and severity of a global severity index (GSI) and of nine different psychopathologies, including depression and anxiety.16 For each dimension, these measures include a continuous standardized T-score and an indicator variable that denotes whether the subject scores above a critical threshold (Ti > 63) and is at risk of developing a clinical psychopathology.17 Fig. 1 illustrates the box-plot distributions of the depression, anxiety, and GSI T-scores according to the stratification of the data described above; a number of violent events below or above the

15 The adapted scale has reliable psychometric properties and has been widely implemented in developing countries and in conflict scenarios. For this sample, the Cronbach alpha of 0.94 indicates an excellent internal consistency—the extent to which all items measure the same constructs. 16 The GSI measures the overall severity of symptoms of psychological trauma. 17 Responses for each question are scaled from 0 to 4, indicating a range of no symptoms to daily symptoms in the last three months. Scores on the relevant questions for each psychopathology are added and divided by the total number of questions answered. Then, a T-score is standardized with mean 50 and standard deviation 10: (Ti = 10 + 50 score).

105

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

80

80

80

Violence > Median

60 50

50

60

70

T Score 70

T Score 70 60 50 Violence < Median

Global Severity Index 90

Anxiety 90

90

Depresion

Violence < Median

Violence > Median

Violence < Median

Violence > Median

Fig. 1. Psychological trauma. Notes: Box-plot distribution of depression and anxiety, and the Global Severity Index. Sample statistics are reported according to an arbitrary stratification of the data: whether the household experienced a number of violent events below or above the sample median. The dotted line depicts the level above which individuals are at risk of developing clinical cases.

sample median.18 The dotted line in each plot indicates the threshold above which a subject is considered at risk. In conformity with the studies in clinical and social psychology, the data in the figure suggests that the psychological consequences of violence follow a dose–response relationship: the experience of more violence brings about more symptoms of depression and anxiety and higher levels of psychological distress—that is, a higher GSI score. For instance, among the group of subjects who experienced a number of violent events below the median, 35 percent of them were at risk of developing clinical depression. This figure increases to 41 percent for the group of subjects who experienced more violence (see Table 1, Panel B).19 As we mentioned earlier, both figures are considerably higher than those observed for the Colombian population.

2.2.3. Beliefs about socioeconomic mobility At the end of the weekend session, we conducted a group activity to elicit subjects’ pre-violence and current living standards, and their beliefs about socioeconomic mobility. For this purpose, we built upon the work of Krishna (2004, 2006, 2001) and Narayan, Pritchett, and Kapoor (2009), and designed a six-step ladder of life that portrayed different levels of living standards or ‘‘stages of progress” among victimized communities.20 We characterized each step of the ladder over five dimensions: housing, land ownership, labor income, children’s schooling, and consumption. 18 Appendix Table A1 provides a more detailed characterization of the scores of the 9 psychopathologies captured by the psychometric scale for the full sample and the sample of subjects who were victimized en masse. 19 The proportion of subjects experiencing critical levels of depression is not statistically different across the arbitrary stratification of the data. Yet, Kolmogorov Smirnov tests indicate that the distributions of depression, anxiety, and GSI are statistically different between the two groups. 20 The ‘‘stages of progress” approach of Krishna (2004, 2006, 2001) and the ‘‘ladder of life” approach of Narayan, Pritchett, and Kapoor (2009) have been used to identify how the poor understand poverty and the strategies, pathways, and major life events that have affected poverty dynamics. This is often carried out during focus groups where participants define the different levels of wellbeing (stages of progress or steps in the ladder of life) within a community. Participants then describe the characteristics of each level, identify where the community poverty line would be located, and discuss the pathways in which households move out or into poverty.

To ensure that our ladder of life provided an accurate representation of the living standards of victims of violence, we characterized each step based on the World Bank’s Moving Out of Poverty Colombia case study, which constructed ladders of life for victimized and non-victimized communities across the country (Matijasevic, Velásquez, Villada, & Ramírez, 2007; Narayan & Petesch, 2010). In our ladder, the bottom two steps illustrated the more salient characteristics of victims living in extreme poverty, whereas the top three steps portrayed the characteristics of non-poor victims (see Fig. 2).21 We measured subjects’ living standards and beliefs as follows: First, we explained that we wanted to understand the socioeconomic changes that occurred to each subject after their experience of violence and forced displacement. To this end, we introduced the ladder of life, used visual aids to describe the characteristics at each step, and provided examples of upward, stagnant, and downward mobility not related to violence. Subjects received a booklet with the illustration of the ladder and were asked to place a stone at the step that best resembled their household’s previolence living standards, and another stone for the post-violence (current) living standards. This was carried out in private, with the assistance of enumerators. Second, we explained that we also wanted to understand how participants perceived their future. For this purpose, we built upon recent methods to elicit of subjective expectations (see Delavande et al., 2011). We handed out 12 stones to each subject and asked them to place them in the steps of the ladder of life where they believed they could be in the following year. To explain this activity, we only mentioned that they should place more (less) stones in a given step if they believed that it was more (less) likely that they would be able to achieve those living standards in a year. After this explanation, subjects placed the 12 stones over the ladder of life 21 Instead of defining a ladder of life for each community or group, as it is the common practice in previous work, we defined a single ladder of life. This allows us to: (1) compare victims’ beliefs regardless of the municipality and community in which they were residing at the time of fieldwork; and (2) resemble comparable pre and post-violence living standards to identify movements up or down the ladder of life over this period.

106

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

Fig. 2. Ladder of life. Notes: Graphical depiction of the ladder of life. Each step of the ladder was characterized over 5 dimensions (housing, lands, income, schooling, and consumption). Subjects were first asked to place one stone at the step that resembled their living standards before the episodes of violence, and a different one for the current living standards. Then, they were asked to distribute 12 stones in the steps of the ladder where they believed they could end up in the following year.

depicted in the booklet and enumerators recorded their answers. The relative number of stones at each step provides a measure of the subjective probability of reaching that position on the ladder in the following year.22 Fig. 3 illustrates the distributions of pre-violence and current positions on the ladder of life, and the distributions for the subjective probabilities of being at each step in the following year for the same stratification of the data described before.23 The data in the first panel indicates that prior to the episodes of violence, 37 percent 22 As Delavande, Gine, and McKenzie (2011) discuss, this method is well suited to elicit subjective expectations in developing countries and contexts of high illiteracy, because it avoids referring explicitly to the concept of probabilities. Instead, the physical representation of the probabilities is straightforward. 23 Table A2 in the Appendix reports the averages for pre and post-violence locations in the ladder, as well as the average probabilities for being at each step of the ladder in the next year, and two-sample mean differences.

of the subjects were in extreme poverty—at the first two steps of the ladder—, 44 percent were at the 3rd step, and 18 percent were at or above the 4th step. For the pre-violence locations, we observe minor differences in the distributions between subjects who experienced a number of violent events below or above the median. Consistent with the work of Ibáñez and Moya (2010a,b), the data in the second panel indicates that violence and displacement drove victims into poverty and condensed the asset and income distributions downwards. Overall, 91 percent of the subjects reported that they were currently in extreme poverty, while less than two percent were at or above the 4th step. In this case, we observe differences across the stratification of our data: among the group of subjects who experienced a number of violent events below the median, 86 percent reported that they were currently at the bottom two steps of the ladder. This figure increases to 95 for subjects who experienced more violence.

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

107

Fig. 3. Past and current positions and beliefs about mobility. Notes: Box-plot distribution of the pre-violence and current positions on the ladder, and on the subjective probabilities of reaching each step of the ladder within a year. The data is stratified according to whether the household experienced a number of violent events below or above the sample median.

Finally, the bottom panel of Fig. 3 illustrates subjects’ beliefs; this is the subjective probability of reaching each step of the ladder in the following year. The data in the figure can be first interpreted with optimism. Although most victims were at the bottom of the ladder at the time the data was collected, the distributions of subjective probabilities for future positions resemble the distributions of pre-violence positions. Nevertheless, we observe that the experience of more violence is associated with an increase in the subjective probability of being at the bottom two steps of the ladder. Whereas the average probability of extreme poverty is 27 percent for subjects who experienced violence below the sample median,

this figure increases to 40 percent for the group of subjects who experienced more violence; this corresponds to a 46 percent increase.

2.2.4. Household and individual characteristics Enumerators administered a household survey prior to collection of the data described above. The survey captured information on current and retrospective individual and household socioeconomic and demographic characteristics. This information allows us to analyze whether the number of violent events experienced

108

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

Table 2 Sample balance: pre and post-violence characteristics.

A. Post-violence characteristics Age Male (=1) Household head (=1) Household size Literate (=1) Log yearly per capita consumption ($COP) B. Pre-violence characteristics Household head was male (=1) Highest level of education in the household (years) Household devoted to off-farm labor (=1) Household devoted to agriculture (=1) Lands owned (Ha) Household head participated in a social organization (=1) C. Hopelessness - Survey Measures Looks towards the following year with hopelessness and despair (=1) Highly unlikely that economic circumstances will improve (=1) Observations

Total

Number of Violent Events Experienced

[1]

Below the median [2]

Above the median [3]

41.05 [13.322] 0.39 [0.488] 0.78 [0.415] 4.47 [2.035] 0.82 [0.381] 3.80 [0.864]

38.84 [12.524] 0.36 [0.481] 0.81 [0.394] 4.54 [1.982] 0.84 [0.368] 3.80 [0.839]

43.26*** [13.760] 0.42 [0.495] 0.75 [0.434] 4.40 [2.091] 0.81 [0.394] 3.80 [0.890]

0.66 [0.474] 8.49 [3.697] 0.47 [0.500] 0.59 [0.494] 7.22 [23.05] 0.44 [0.498]

0.62 [0.486] 8.61 [3.651] 0.47 [0.501] 0.52 [0.501] 5.73 [14.98] 0.35 [0.480]

0.70 [0.459] 8.36 [3.749] 0.47 [0.501] 0.65** [0.480] 8.72 [28.93] 0.53*** [0.501]

0.55 [0.499] 0.31 [0.463] 336

0.50 [0.501] 0.26 [0.438] 168

0.59 [0.493] 0.36** [0.482] 168

Notes: Summary statistics on individual and household characteristics. Panel A and B report data on subjects’ current and pre-violence characteristics, respectively. Panel C reports data on the survey-based measures of hopelessness and beliefs. Column 1 reports sample statistics for the full sample, while columns 2 and 3 report statistics according to an arbitrary stratification of the data: whether the household experienced a number of violent events below or above the sample median. Asterisks in column 3 indicate the statistical significance of the mean-difference test between the two groups. Standard deviations reported in brackets. *p < 0.1. **p < 0.05. ***p < 0.01.

by the household was associated with pre-violence observable characteristics. Table 2, Panels A and B reports sample statistics on the pre and post-violence characteristics for the full sample and the stratification of the data described above. The data indicates that subjects in the two groups—those who experienced a number of violent events below or above the median—differ on their age, and on the pre-violence main economic activity and participation in community organizations. The latter difference is consistent with qualitative reports of conflict dynamics, which suggest that armed groups targeted social leaders with a higher likelihood than other subsets of the population (CNMH, 2013). Although this statistic is based on the arbitrary stratification of the data, it suggests that the experience of more violence could have been correlated with pre-violence household characteristics. Nevertheless, as mentioned earlier, the targeting of community leaders and activists would be expected to work against the hypothesis that a more severe experience of violence dampens psychological health and the beliefs about socioeconomic mobility. We will return to this discussion in the following section when we describe our empirical strategy. The survey also included a question on whether subjects looked towards the following year with hope and optimism or with hopelessness and despair, and a Likert scale question on whether they believed that it was likely, unlikely, or very unlikely that their economic circumstances will improve in a year. These questions provide alternative measures of subjects’ beliefs. Panel C in Table 2

indicates that subjects who experienced more violence are more likely to look forward to the following year with hopelessness and to believe that it is more unlikely that their economic circumstances will improve. These patterns are consistent with the data in Fig. 2 and suggest that the subjective probabilities for future positions on the ladder of life provide an internally valid measure of the beliefs about socioeconomic mobility. 3. Empirical strategy Our empirical analysis exploits the variation in the severity of violence—the number of violent events experienced by the household—, controlling for the current step of the ladder of life and other individual and household characteristics. This strategy compares subjects who were at the same step of the ladder, and therefore face similar material constraints, but who experienced a different number of violent events. In doing so, we explore the effect of the severity of violence on victim’s beliefs, which hypothesize is driven by the psychological constraints created by such traumatic experiences. For each future step of the ladder k, we estimate model 1, where
 we regress p Stþ1 hr ¼ k —the subjective probability of being in step k in the following year for household h in region r—on V hr , the standardized number of violent events experienced by the household h,   its quadratic term, and I Sthr ¼ j a binary indicator for the current position on the ladder. Since few subjects were at or above the 4th

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

step of the ladder (see Fig. 3), we condensed the top 3 steps into a single step that characterizes the non-poor victims.24 4
 X   2 p Stþ1 dj I Sthr ¼ j hr ¼ k ¼ c0 þ c1 V hr þ c2 V hr þ j¼1

þ C01 X th þ C02 X ht1 þ nr þ ehr

8k 2 ½1; 4

ð1Þ

In model 1, we also control for a vector X thr of current individual covariates, such as each subject’s age, gender, and years of education, which provide a measure of the subject’s human capital.25 The matrix of current covariates also includes whether the subject is the household head, and whether the household experienced an economic shock or the death of a member for reasons not related to violence. We also control for a matrix X t1 of pre-violence househ hold characteristics, including whether agriculture was the households’ main economic activity, the pre-violence organizational membership variable, and the size of the household’s landholdings, which provides a measure of the extent of asset losses. Finally, the model includes a region-specific fixed effect nr , and a White-robust error term ehr . The validity of our empirical strategy hinges on the assumption that the severity of violence was exogenous to pre-violence beliefs, conditional on the observable characteristics in matrix X t1 h . If the severity of violence was associated to pre-violence characteristics that were related to individual beliefs, the resulting omitted variable bias would lead us to overstate or understate the effects of the severity of violence. While the information available suggests that violence was indiscriminate for the time and place from which our data was collected (see Section 1), we follow the strategies used by Moya (2018) to support our assumptions for identification. First, we assess the importance of pre-violence characteristics in explaining the patterns of violence. In Table 2, we had observed that the households’ main economic activity and the participation in local organizations differed across the arbitrary stratification of the data that we employed in the previous section. To provide a more thorough analysis of selection on observables that is not based on the arbitrary stratification of the data, we follow Bellows and Miguel (2009) and regress the number of violent experienced by each household—or the victimization score—on a set of pre-violence household characteristics. The results indicate that the severity of violence was neither driven by specific observable characteristics—including the pre-violence group membership— nor was it jointly determined by the set of observables (see Appendix Table A3). Nevertheless, to be more conservative, in model 1 we control for key pre-violence characteristics in matrix X t1 hr . Under the assumption of conditional unconfoundedness (Imbens, 2003), we remove the bias that stems from the possible targeting of violence based on such observable characteristics. Second, we argue that our results are not invalidated by selection on unobservables; this is, by characteristics that we were unable to measure, but that the armed groups could somehow observe. On the one hand, our study is set in two regions which were newly contested terrain, armed groups relied on indiscriminate violence against civilians as a strategy to obtain territorial supremacy (CNMH, 2013).26 Therefore, the severity of violence would be associated with unobservable characteristics only if civil24 Results are robust if we estimate the system of equations on the six steps, as well as if we drop the controls for the current location on the ladder. Results are available upon request. 25 We do not control for physical capital since few subject reported ownership or access to productive assets. 26 Armed groups used indiscriminate violence to spread fear and undermine the support for their opponents. This allowed them to control the activities and preferences of the population. In the words or former combatants, exerting vicious and indiscriminate violence was the most effective mechanism to achieve territorial dominance (CNMH, 2013).

109

ians exposed themselves in ways that would increase the risk of victimization. In a context of widespread violence, it is unlikely that the more hopeless subjects would behave in this way. On the other hand, we provide further support the validity of our empirical strategy, by analyzing the robustness of the results on the sample of subjects who were victimized en masse. The importance of this analysis relies on the assumption that this sub-sample was victimized en masse in massacres or in the crossfire of combat between illegal armed groups, where the patterns of violence were close to random.27

4. Results We find that the experience of more severe violence dampens victims’ beliefs, leading them to perceive that there are few opportunities for upward mobility. Our results are robust when we conduct the analysis on the sub-sample of subjects who were victimized en masse and under alternative specifications of the econometric model, thus lending credence to our empirical strategy. Finally, we find suggestive evidence that the experience of symptoms of depression may explain the effect of the severity of violence. Together, the results highlight the way in which the psychological consequences of violence reinforce the effect of material constraints and induce hopeless beliefs.

4.1. Violence and hopelessness Table 3 reports the results of model 1. Each column on the table reports the estimated effects of the severity of violence and the current position on the ladder on the subjective probabilities of future positions. For expositional purposes, we omit the coefficients for the set of individual and household covariates. As expected, the results first indicate that the current position on the ladder has a strong and significant effect on the beliefs about socioeconomic mobility. As subjects move up on the ladder of life and their circumstances improve, their beliefs about the likelihood of future poverty fall, whereas those of the likelihood of upward mobility increase. For example, relative to subjects at the second step of the ladder, those at the bottom step believe that the likelihood of remaining in the bottom of the ladder is 4 percentage points higher, while the likelihood of moving up to the top step is 13 percentage points lower. These results are predictable: objectively, it is more difficult to a subject at the bottom of the ladder in the span of a year, than for subjects further up in the ladder. Hence, the results are consistent with a cognitive assessment of the way in which current circumstances shape the prospects of socioeconomic mobility. Consistent with our main hypothesis, we also find that violence dampens victims’ beliefs. The positive statistically significant coefficient of the severity of violence in Column 1, indicates that an increase of one standard deviation in the severity of violence raises the perceived probability of being at the bottom step of the ladder by 3.6 percentage points. This entails a considerable 54 percent effect relative to the mean. Likewise, the positive and significant effect of the severity of violence in Column 4 indicates that an increase of one standard deviation in the severity of the victimization lowers the subjective probability of reaching the top of the 27 The characteristics of massive victimizations can be portrayed by the emblematic massacre of the municipality of El Salado in the department of Bolivar, where some of the subjects in our sample were victimized. In February 2002, over 300 paramilitaries arrived to El Salado and gathered the inhabitants in the central plaza. Paramilitaries then literally selected the victims at random and killed them in front of the community. Paramilitaries abandoned the town days later, after murdering 70 civilians, while all the survivors migrated soon after (CNMH, 2013).

110

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

Table 3 Violence & beliefs about mobility. p(Step t + 1 = 1) # of violent events (standardized) # of violent events squared (standardized) Step t = 2 Step t = 3 Step t = 4 Constant Hh pre and post-violence controls R-squared Observations Mean value of dependent variable

***

0.036 [0.013] 0.005 [0.005] 0.038** [0.017] 0.054** [0.023] 0.062*** [0.017] 0.104* [0.053] Yes 0.10 311 0.067

p(Step t + 1 = 2)

p(Step t + 1 = 3)

p(Step t + 1 = 4)

0.037 [0.024] 0.009 [0.008] 0.174*** [0.025] 0.255*** [0.036] 0.285*** [0.050] 0.382*** [0.075] Yes 0.25 311 0.267

0.016 [0.019] 0.008 [0.005] 0.080*** [0.029] 0.033 [0.044] 0.043 [0.087] 0.361*** [0.070] Yes 0.08 311 0.337

0.057** [0.028] 0.005 [0.011] 0.132*** [0.037] 0.342*** [0.069] 0.389*** [0.126] 0.153 [0.095] Yes 0.21 311 0.329

Notes: Each column reports the results of estimating model 1 on the subjective probability of being in each step of the ladder of life in the following year. The table reports the coefficients for the severity of violence, measured by the standardized number of the violent events experienced by the household, its quadratic term, and a binary indicator for the current position on the ladder. The bottom step of the ladder is the omitted category. As described in Eq. (1), each model includes a set of pre and post-violence covariates and a regional fixed effect. Post-violence covariates include the subject’s age, gender, and years of education, whether he or she is the household head, and whether the household experienced an economic shock or the death of a household member for reasons not related to violence. Pre-violence covariates include the household size, size of land holdings, participation of a household member in local organizations, and participation of the household in agricultural work. Estimated coefficients of the covariates and fixed effect are not reported but are available upon request. White-robust standard errors are reported in brackets. *p < 0.10. **p < 0.05. ***p < 0.01.

ladder by 5.7 percentage points—a 17 percent effect relative to the mean.28 We also assess whether the effect of the severity of violence is robust when we control for the temporal proximity of violence. This allows us to understand whether the results in Table 3 are temporary and driven by subjects who had been victimized and displaced recently. In Appendix Table A4, we find that the temporal proximity of violence does not have a significant effect on victim’s beliefs, whereas the estimated effect of the severity of violence is robust to controlling for the temporal proximity of violence. This result suggests that the psychological constraints created by the experience of violence persist in time Taken together, the results above indicate that subjects who experienced more severe violence believe that they live in a different world, one with diminished prospects for social mobility. Since we control for current circumstances, material capacities, and the time since the episodes of violence, the differences between two otherwise similar victims who experienced different levels of violence, point to the existence of psychological constraints by which violence begets hopelessness. 4.2. Robustness—subsample of subjects victimized en masse Appendix Table A5 reports the results of model 1 for the sample of subjects who were victimized en masse. The results indicate that the severity of violence brings about a qualitatively robust effect on victims’ beliefs.29 A one standard deviation increase in the severity of violence brings about a 5 percentage point increase in the perceived probability of remaining at the bottom of the ladder. While this effect is not statistically significant, since the standard error increases considerably due to the smaller sample, the point estimate is stronger in magnitude than what we observed in Table 3. Likewise, we find a negative and statistically significant effect on the perceived likelihood of upward mobility. In this case, the point estimate indicates an even stronger effect: a 9 percentage points reduction. These results provide further support to the 28

The results are robust if we use the victimization score in the empirical analysis. In this analysis, we do not control for the current position on the ladder, since there is even fewer variation in this characteristic for this subsample. 29

underlying assumption of our empirical strategy and to the validity of our results. 4.3. Alternative specifications We address three concerns regarding the specification of model 1. First, in model 1 we ignored the pre-violence positions on the ladder of life, which may have been correlated with a host of skills and assets and that could act as a reference point for the victims’ beliefs. For this reason, in Appendix Table A6 we estimate model 1 including a set of binary indicators of the pre-violence position on the ladder of life. We again find that the severity of violence and the current position on the ladder have a robust effect on victims’ beliefs, while the pre-violence positions do not influence victims’ beliefs. Second, the results in Table 3 may hinge on whether the characteristics of each step of the ladder of life depicted the living standards of our sample, and whether there is considerable error in the subjects’ assessments of their current positions on the ladder. An alternative is then to control for an objective measure of the current living standards. In Appendix Table A7 we estimate model 1 controlling for the per capita consumption, which was calculated using data from a standard consumption module included in the household survey. Again, the results indicate a robust and significant effect of the severity of violence on victims’ beliefs. In addition, the households’ per capita consumption portrays the effect of the current step of the ladder of life that we had observed in Table 3: higher levels of consumption have a positive and significant effect on the belief of moving to the top step of the ladder of life. Finally, we address whether the probabilities that were elicited using the ladder of life are a valid representation of subjects’ actual beliefs. For this reason, we analyze the robustness of our results using the survey-based measures of hopelessness as our dependent variables in model 1. In Appendix Table A8, we find that a more severe experience of violence has a positive and statistically significant effect on the likelihood that subjects look towards the future with hopelessness and despair and on the beliefs that socioeconomic circumstances will not improve within a year. To the extent that we find consistent results under these alternative

111

A. Moya, M.R. Carter / World Development 113 (2019) 100–115 Table 4 Violence and psychological trauma. Depression

# of violent events (standardized) Years since violence # of violent events* years since violence Observations

Anxiety

Global Severity Index

T-score

Ti > 63

T-score

Ti > 63

T-score

Ti > 63

0.699** [0.346] 0.236** [0.100] 0.124 [0.090] 336

0.074** [0.029] 0.011 [0.009] 0.022** [0.011] 336

0.994* [0.562] 0.523*** [0.115] 0.111 [0.112] 336

0.052** [0.024] 0.025*** [0.009] 0.012 [0.007] 336

1.093*** [0.360] 0.318*** [0.083] 0.152* [0.082] 336

0.072*** [0.021] 0.020** [0.009] 0.006 [0.008] 336

Notes: Estimates for the effect of violence on depression. Column 1 reports the impact on the continuous Ti measure. Column 2 shows the estimated effect on the probability of being above the clinical threshold (Ti > 63). White-robust standard errors in brackets. *p < 0.1, ** p < 0.05, ***p < 0.01.

measures, we argue for the validity of our core measure of victims’ beliefs about socioeconomic mobility.30 4.4. Exploring the psychological mechanism Finally, we follow a reduced-form procedure to explore whether psychological trauma is a channel through which violence affects the beliefs about socioeconomic mobility. We proceed as follows: First, we document how symptoms of psychological trauma vary according to the severity and temporal proximity of violence. For this purpose, we estimate model 2, where we regress Traumahr , the continuous T-score or the dichotomic threshold measure for depression, anxiety, and the GSI, on V hr , the standardized number of violent events, thr , the years since the episodes of violence, and their interaction.31

Traumahr ¼ a0 þ a1 V hr þ a2 thr þ a2 V hr  thr þ ehr

ð2Þ

The results in Table 4 indicate that a more severe and recent violence produces more severe symptoms of depression and anxiety, and a higher GSI score. For instance, an increase of one standard deviation in the severity of violence increases the probability that the symptoms of depression and anxiety disorders are above the critical threshold by 7 and 5 percentage points, respectively. These two effects account for an 18-percentage increase relative to the mean. Likewise, we observe that a more recent experience of violence brings about higher symptoms of depression, anxiety, and GSI, and a higher likelihood of experiencing symptoms of anxiety and GSI above the critical threshold. These results, in addition, are qualitatively similar when we focus on the continuous T-score as our dependent variable. Overall, these results are consistent with the data in Fig. 2 and with the studies in psychology outlined before on the dose-response relationship between violence and psychological trauma (Mollica et al., 1998). Having established the effects of violence on different psychopathologies, we assess whether psychological trauma has a similar effect on the beliefs about socioeconomic mobility as the one estimated for the severity of violence. For this purpose, we estimate
 model 3 where we regress phr Stþ1 ¼ k —the perceived probability of being in step k in the following year—on Depressionhr , the T-score for the symptoms of depression, and its quadratic term. Again, we 30 We also explored heterogeneous effects based on key characteristics, such as the subject’s gender, age, years of education, and the pre-violence participation in social organizations. Contrary to what we could have expected, we do not observe significant heterogeneous effects of the severity of violence over these characteristics. Yet, our inability to observe heterogeneous effects may be related to our small sample. 31 In addition, we control for the subjects age and gender to account for well-known differences in the susceptibility to psychological trauma among men and women and across age groups. We also include the regional fixed effect to control for differences in the prevalence of symptoms of trauma between regions. The results are stronger in magnitude and significance if we do not control for these characteristics or for the regional fixed effect.

include binary indicators for the current position of the household   on the ladder I Sthr ¼ j , control for the vectors X thr and X t1 hr of current individual and pre-violence household covariates and include a regional fixed effect nr . We focus on depression, rather than on other manifestations of psychological trauma because depression is often accompanied with feelings of pessimism and hopelessness and conceptually it is directly tied to future-oriented perceptions (American Psychiatric Association, 2000).32


 tþ1 p Shr ¼ k ¼ b0 þ b1 Depressionhr þ b2 Depression2hr þ

4 X

  dj I Sthr ¼ j þ C01 X th þ C02 X t1 h

j¼1

þ nr þ ehr

8k 2 ½1; 4

ð3Þ

Table 5 reports the results of model 2, which confirm our hypothesis regarding the effect of depression on the victims’ beliefs about socioeconomic mobility. An increase of one point in the depression T-score raises the perceived probability of being at the bottom two steps of the ladder by 3 and 7 percentage points, respectively. In addition, it lowers the perceived probability of reaching the top of the ladder by 9 percentage points. These effects are remarkably similar in magnitude to the effects of the severity of violence. The data in the table indicate, in addition, that the effect of the current step of the ladder persists when we control for the extent of psychological trauma. Together, the results in Tables 4 and 5 provide suggestive evidence on the role of psychological trauma in explaining why a more severe experience of violence induces pessimistic beliefs about socioeconomic mobility. Admittedly, this reduced-form analysis does not constitute a proper mediation analysis and we are unable to assert that psychological trauma is the only mechanism through which the severity of violence affects victims’ beliefs. In theory, a more rigorous strategy would be to follow Acharya, Blackwell, and Sen (2016) and estimate the average controlled direct effect of the severity of violence; this is, the causal effect of the severity of violence when the effect of a potential mediator—psychological trauma – is taken into account. In practice, however, this method is not suitable since we likely violate the strict sequential unconfoundedness assumption.33 Yet, to the extent that the results of Tables 4 and 5 are consistent with the evidence from studies in psychology outlined in Section 1, we make a plausible case that the relationships that we document are informative of a psychological mechanism that dampens victims’ beliefs. Still, more work is required in this area. 32 Our results are robust results if we instead focus on two anxiety disorders (general anxiety and phobic anxiety). This is not surprising since depression often cooccurs with anxiety (Kessler et al., 1996) 33 Likewise, we abstain from the conventional approach of simultaneously controlling for the treatment and potential mediating variables. As Acharya et al. (2016) discuss, this latter method often leads to biased and inconsistent estimates as a result of M-bias or posttreatment bias.

112

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

Table 5 Depression and beliefs about mobility.

A. Reduced Form: Depression T-score Depression T-score Depression T-score squared Step t = 2 Step t = 3 Step t = 4 Constant Hh pre and post-violence controls R2 Observations Mean value of dependent variable

p(Step t + 1 = 1)

p(Step t + 1 = 2)

p(Step t + 1 = 3)

p(Step t + 1 = 4)

0.032** [0.016] 0.000** [0.000] 0.000 [0.000] 0.047*** [0.018] 0.068*** [0.024] 0.836 [0.522] Yes 0.08 307 0.067

0.072** [0.033] 0.001** [0.000] 0.000 [0.000] 0.184*** [0.025] 0.266*** [0.035] 1.906* [1.055] Yes 0.25 307 0.267

0.013 [0.028] 0.000 [0.000] 0.000 [0.000] 0.087*** [0.028] 0.029 [0.044] 0.775 [0.896] Yes 0.08 307 0.337

0.090* [0.046] 0.001** [0.000] 0.000 [0.000] 0.144*** [0.037] 0.362*** [0.067] 2.966** [1.500] Yes 0.21 307 0.329

Notes: Each column reports the estimated coefficients for the severity of symptoms of depression, its quadratic term, and the current position on the ladder on the subjective probability for future positions on the ladder. The models include a regional fixed effect and pre and post-violence covariates. Estimated coefficients for these covariates are available upon request. White-robust standard errors are reported in brackets. Robust standard errors are reported in brackets. *p < 0.10. **p < 0.05. ***p < 0.01.

Table 6 Transition matrices and long-run beliefs about mobility. Experience of Violence – 1st quartile Step in Period t

Step in period t-1

1 2 3 4

ke

1

2

3

4

0.06 [0.025] 0.03 [0.025] 0.01 [0.025] 0.00 [0.025]

0.32 [0.066] 0.14 [0.066] 0.06 [0.066] 0.03 [0.066]

0.33 [0.027] 0.41 [0.027] 0.29 [0.027] 0.28 [0.027]

0.30 [0.078] 0.43 [0.078] 0.64 [0.078] 0.69 [0.078]

1

2

3

4

0.12 [0.029] 0.08 [0.029] 0.06 [0.029] 0.05 [0.029]

0.37 [0.050] 0.20 [0.050] 0.12 [0.050] 0.09 [0.050]

0.31 [0.039] 0.39 [0.039] 0.28 [0.039] 0.27 [0.039]

0.21 [0.070] 0.34 [0.070] 0.55 [0.070] 0.59 [0.070]

0.004 [0.028] 0.046 [0.080] 0.292 [0.026] 0.658 [0.106]

Experience of Violence - 4th quartile Step in Period t

Step in period t-1

1 2 3 4

ke

0.063 [0.114] 0.127 [0.062] 0.287 [0.038] 0.523 [0.093]

Notes: Transition matrices that define the perceived probabilities of transitioning from one step of the ladder to another one over each period and associated long-run distributions for subjects in the 1st and 4th quartiles of the distribution of the severity of violence. Each element of the transition matrix and of the long-run population distribution corresponds to the average values for subjects in these two groups. Standard deviations are reported in brackets.

5. Violence and long-run poverty dynamics The implications of the results above can be better understood by simulating the patterns of social mobility and long-run distributions that are associated with the victims’ beliefs. For this purpose, we use the estimated coefficients from Table 3, to construct subject-specific transition matrices that define the subjective probabilities of transitioning from one step of the ladder to another one time. This allows us to illustrate how the experience of more severe violence implies a less favorable long-term distribution. Define the one period transition matrix P that defines the perceived probability of transitioning from current ladder step j to ladder step k within a year:

2

p11 6 . P¼ 6 4 .. p41

3    p14 .. .. 7 7 . . 5    p44

where element pjk is the perceived probability that an individual at step j in period t will be at step k in period t þ 1. This structure can accommodate a wide variety of probability processes ranging from convergent to divergent processes. Furthermore, let kt be the 4x1 vector that denotes the population distribution across the 4 steps of the ladder of life in period t. Given P, the expected distribution of the population in period t þ 1 will be ktþ1 =P0 kt . If we further assume that the transition process is governed by a stable Markovian process in which transition probabilities only depend on the current position, then ktþ2 ¼ P0 ½P0 kt : For a well-defined probability matrix, the population distribution will converge in the long-run to the stable equilibrium distribution given by the eigenvector ke ¼ P0 ke : Note that we can construct the subject-specific transition matrix using the results from the previous section. Model 1 corresponds to a set of equations for the subjective probabilities associated with each future step k on the ladder of life, conditional on the

113

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

current step j in the ladder. Hence, for a subject currently at step j, the four equations in model 1 provide the information necessary to

1

th

0.8 Extreme poverty headcount

estimate the elements of the j row of the transition matrix. To calculate the elements of the other rows, we estimate the equation for the probability of being at step k in the following period, and then calculate the predicted probability for step k, exogenously varying the current step of the ladder j—this is, we simulate the subjective probabilities as if we observed the same subject at the different steps of the ladder of life. Formally, the predicted conditional probability, will be given by:

0.6

0.4

0.2 t

b 1X þ C b 2 0 X t1 þ b b jk ¼ b c 0 þ bc 2 V hr þ bc 2 V hr þ dbj þ C nr p i h

ð4Þ

8j; k 2 ½1; 4

ð4Þ

0

In Table 5, we report the average predicted probabilities—the elements of the transition matrices—and the average associated with population long-run distributions for subjects in the 1st and 4th quartiles of the distribution of the severity of violence. We illustrate the results of the simulations for these two quartiles because moving from the median value of the 1st quartile to the median value of the 4th quartile roughly corresponds to an increase of one standard deviation in the severity of violence. Hence, this stratification allows us to connect with the results from the previous section.34 The effect of the severity of violence can be observed by noting that the elements in the lower triangle of the transition matrix are higher on average for subjects who experienced more severe violence than for who experienced less violence. For instance, the average subjective probability of being in the bottom two steps of the ladder of life—the elements in the first two columns of each Markov matrix—are considerably higher for subjects who experienced more severe violence than for those who experienced less violence. These differences, in addition, are statistically significant as follows from the results in Table 3. Moreover, the long-run distributions indicate that the average extreme poverty headcount— the percentage of subjects at the bottom two steps of the ladder of life—is 14 percentage points higher for subjects in the top quartile of the distribution of the severity of violence than for those at the bottom quartile. This entails a sizeable and statistically significant 280 percent effect. Together, these differences in the average elements of the two Markov matrices and in the long-run eigenvectors are consistent with hopeless beliefs of socioeconomic mobility among victims who experienced more severe violence. To further illustrate this point, Fig. 4 plots the evolution of the average extreme poverty headcount and the associated 95 percent confidence interval for subjects in the 1st and 4th quartiles of the distribution of the severity of violence. The figure illustrates how the extreme poverty headcount falls relatively quickly, in conformity with the victims’ beliefs outlined in Fig. 3. However, the figure also illustrates how the transitional dynamics diverge signaling a large increment in the number of victims that expect to be in extreme poverty because of the experience of more severe violence. The results above sharply illustrate how violence operates as an additional force that dampens victims’ beliefs about future social mobility and the way in which these beliefs can become selfconfirming. Of course, these long-run estimates are subject to the proviso that the actual socioeconomic dynamics can be characterized by a Markov process, and that the transitional matrices remained unaltered over time even as the socioeconomic circumstances of the household improve. Yet, this is a good characterization of victims’ beliefs since the results from the previous section 34 Table A10 in the Appendix reports transition matrix and long-run population distributions for the four quartiles of the distributions of the severity of violence.

0

1

2

3

4

5

Periods Experience of Violence in 1st quartile 95% confidence interval Experience of Violence in 4th quartile 95% confidence interval

Fig. 4. Extreme poverty headcount. Notes: Simulated evolution of the extreme poverty headcount based on the transition matrices depicted in Table 6.

indicate that the effect of violence is independent of the level of wellbeing and persists over time. 6. Discussion and policy implications Using unique data for a sample of IDP’s in Colombia, we find that the experience of more severe violence beliefs about socioeconomic mobility and leads victims to become increasingly hopeless. Importantly, this effect persists even after we account for the effect of current levels of wellbeing, consumption, education, and asset losses. The effect of violence therefore signals to the existence of psychological constraints, that go beyond the ‘‘true” obstacles or material constraints imposed by violence and forced displacement. In fact, we provide suggestive evidence that the psychological consequences of violence are one mechanism through which violence begets hopelessness. Taken together, our results echo the testimony of the victim at the beginning of this paper, which portrays how violence diminished her ability to hope for a better future. More generally, our findings are consistent with those of psychological studies that show that the experience of trauma triggers depressive explanatory styles and can induce learned helplessness. Violence can therefore hinder the victims’ willingness to try to make the best out of what they have and contribute to the persistence of poverty. Precisely, the simulation analysis of the previous section speaks to the work of Sen (1999), who suggested that internal (psychological) constraints can be more binding than the more discernible material constraints and can create a behavioral poverty trap. We admittedly recognize that our analysis is limited by our small and non-representative sample of victims of violence and by the lack of data to test whether victims’ beliefs map into actual behavior and economics transitions. Yet, the results on the underlying role of psychological trauma, which are consistent with those of studies in clinical and social psychology, serve as a first indication that the that the results are externally valid and that the underlying psychological principles can be applied to other groups of victims and to survivors of other traumatic hocks. Moreover, the work of Cuartas and Moya (2016), who used the same approach to collect data on beliefs about socioeconomic mobility from a subsample of the Colombian Longitudinal Survey, offers support to the argument that violence can alter poverty dynamics through a behavioral channel. In particular, they find that subjective beliefs, which were measured in 2011, have a strong and significant impact on households’ economic trajectories between 2011 and

114

A. Moya, M.R. Carter / World Development 113 (2019) 100–115

2013. Nevertheless, the generalizability of these results requires more work and points to important avenues for future research. Our paper has important policy implications and suggests reconsidering the strategies to assist the victims of violence and other negative shocks. In the case of Colombia, for example, the Government has set a progressive set of laws and implemented comprehensive programs to assist the victims. These include humanitarian and conditional cash transfers and access to subsidized education and health, which are thought to provide a safety net to minimize the negative consequences of violence and forced displacement. In addition, the Government has laid out a strategy to promote the socioeconomic recovery of victims through asset and land transfers, job training programs, and indemnities up to US$8,000, among others. Unfortunately, the psychological consequences of violence have been largely neglected and mental health programs are scarce and poorly funded. According to data from the Colombian Ministry of Health, between 2013 and 2016, less than 4 percent of the victims in the country have received psychological assistance, while less than 1% of the funds allocated by the Government for the victims’ recovery are invested in mental health programs. This is unfortunate since our results suggest that the psychological consequences of violence may hinder the effect of standard interventions and set seeds for persistent poverty. We conclude by conducting a simple thought experiment to illustrate how psychological trauma may hinder the effectiveness of other, more traditional types of programs. For this purpose, we return to the simulation analysis of the previous section and assess how the long-run poverty dynamics depicted in Fig. 4 would change if we provided an asset transfer to the victims who experienced more severe violence, but without implementing any psychological assistance that could alleviate the psychological constraints. We simulate the poverty headcount across time using the initial population distribution and the transition matrices depicted in Table 5. After period 1, we provide an asset transfer to victims with a moat the top quartile of the distribution of the severity of violence. This need-based asset transfer lifts the more psychologically vulnerable victims one step above their current position in the ladder of life. Fig. 5 illustrates the evolution of the extreme poverty headcount for the victims in the first quartile of the distribution of the severity of violence, and for those in the 4th quartile with and without the asset transfer. The figure indicates that in the medium run, the asset transfer lowers the extreme poverty head-

count for the victims who experienced more severe violence. Nevertheless, in the long run, the extreme poverty head count for victims who experienced more severe violence converges to the previous level without the asset transfer. The dynamics in Fig. 5 therefore indicate that even if victims live a world of convergent socioeconomic mobility, the psychological consequences of violence can render standard interventions ineffective and alter the long run distribution of well-being. Our paper thus provides additional motives for a better understanding the psychological consequences of violence and how they influence poverty dynamics, and for designing and implementing psychological programs for victims of violence. Acknowledgements This paper previously circulated under the title ‘‘Violence and the Formation of Hopelessness and Pessimistic Prospects of Upward Mobility in Colombia.” We are grateful to Steve Boucher, Arnold Cecile, Hilary Hoynes, Miriam Golden, Guy Grossman, Rachid Laajaj, Evan Lieberman, Travis Lybbert, Juan Fernando Vargas, and three anonymous referees for useful comments. We would also like to thank participants at the Northeast Development Conference at Harvard University; the Households in Conflict Network Workshop at UC Berkeley; EGAP at Cape Town; and seminars at CIDE; UC Davis; University of Cape Town; and Universidad de los Andes for useful comments. Gabriela Paredes and María José Torres provided outstanding research assistance. This paper would have not been possible without the support of community leaders and local priests, and without the victims’ willingness to participate and revisit their painful experiences of violence. We gratefully acknowledge financial support from the Harry Frank Guggenheim Foundation, the Henry A. Jastro Award, the Pacific Rim Research Dissertation Grant, and RIMISP. IRB Approval for the collection of human subjects data was awarded by the University of California, Davis. The usual disclaimers apply. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.worlddev.2018.08. 015. References

1

Extreme poverty headcount

0.8

0.6

0.4

0.2

0 0

1

2

3

4

5

Periods Experience of Violence in 1st quartile Experience of Violence in 4th quartile Experience of Violence in 4th quartile with asset transfer

Fig. 5. Extreme poverty headcount under a need-based asset transfer. Notes: Simulated evolution of the extreme poverty headcount based on the transition matrices depicted in Table 6, and an asset transfer for the victims who experienced more severe violence that elevates them one step in period 1.

Abramson, L. Y., Metalsky, G. I., & Alloy, L. B. (1989). Hopelessness depression: A theory-based subtype of depression. Psychological Review, 96(2), 358–372. https://doi.org/10.1037/0033-295X.96.2.358. Acharya, A., Blackwell, M., & Sen, M. (2016). Explaining causal findings without bias: Detecting and assessing direct effects. American Political Science Review, 110(3), 512–529. https://doi.org/10.1017/S0003055416000216. American Psychiatric Association (2000). Diagnostic and statistical manual of mental disorders DSM-IV-TR. Washington, DC: American Psychiatric Publishing Inc. Appadurai, A. (2004). The capacity to aspire: Culture and the terms of recognition. In V. Rao & M. Walton (Eds.), Culture and public action (pp. 59–84). Stanford, CA: Stanford University Press. Barrett, C., Carter, M. R., & Chavas, J.-P. (2019). Introduction. In C. Barrett, M. R. Carter, & J.-P. Chavas (Eds.), The economics of poverty traps. Chicago: University of Chicago Press. Bernard, T., Dercon, S., Orkin, K., & Taffesse, A. S. (2015). Will video kill the radio star? Assessing the potential of targeted exposure to role models through video. The World Bank Economic Review, 29(1), 226–237. https://doi.org/10.1093/wber/ lhv014. Bauer, M., Cassar, A., Chytilova, J., & Henrich, J. (2014). War’s enduring effects on the development of egalitarian motivations and in-group biases. Psychological Science, 25(1), 47–57. https://doi.org/10.1177/0956797613493444. Bauer, M., Blattman, C., Chytilová, J., Henrich, J., Miguel, E., & Mitts, T. (2016). Can war foster cooperation? Journal of Economic Perspectives, 30(3), 249–274. https://doi.org/10.1257/jep.30.3.249. Beaman, L., Duflo, E., Pande, R., & Topalova, P. (2012). Female leadership raises aspirations and educational attainment for girls: A policy experiment in India. Science, 335(6068), 582–586. https://doi.org/10.1126/science.1212382.

A. Moya, M.R. Carter / World Development 113 (2019) 100–115 Bellows, J., & Miguel, E. (2009). War and local collective action in Sierra Leone. Journal of Public Economics, 93(11–12), 1144–1157. https://doi.org/10.1016/j. jpubeco.2009.07.012. Blattman, C. (2009). From violence to voting: War and political participation in Uganda. American Political Science Review, 103(2), 231–247. https://doi.org/ 10.1017/S0003055409090212. Blattman, C., & Miguel, E. (2010). Civil war. Journal of Economic Literature, 48(1), 3–57. https://doi.org/10.1257/jel.48.1.3. Briere, J., & Spinazzola, J. (2005). Phenomenology and psychological assessment of complex posttraumatic states. Journal of Traumatic Stress, 18(5), 401–412. https://doi.org/10.1002/jts.20048. Callen, M., Isaqzadeh, M., Long, J. D., & Sprenger, C. (2014). Violence and risk preference: Experimental evidence from Afghanistan. American Economic Review, 104(1), 123–148. https://doi.org/10.1257/aer.104.1.123. Carter, M. R., & May, J. (2001). One kind of freedom: Poverty dynamics in postapartheid South Africa. World Development, 29(12), 1987–2006. https://doi.org/ 10.1016/S0305-750X(01)00089-4. Carter, M. R., & Barrett, C. B. (2006). The economics of poverty traps and persistent poverty: An asset-based approach. Journal of Development Studies, 42(2), 178–199. https://doi.org/10.1080/00220380500405261. Cassar, A., Grosjean, P., & Whitt, S. (2014). Social preferences of ex-combatants: Survey and experimental evidence from postwar Tajikistan. In K. Wärneryd (Ed.), The economics of conflict: Theory and empirical evidence (pp. 231–262). Cambridge MA: MIT Press. https://doi.org/10.7551/mitpress/ 9780262026895.001.0001. Cassar, A., Grosjean, P., & Whitt, S. (2013). Legacies of violence: Trust and market development. Journal of Economic Growth, 18(3), 285–318. https://doi.org/ 10.1007/s10887-013-9091-3. Centro Nacional de Memoria Histórica (CNMH) (2013). ¡Basta ya! Colombia: Memorias de guerra y dignidad. Bogotá: Imprenta Nacional. Retrieved from http://www.centrodememoriahistorica.gov.co/micrositios/informeGeneral/ descargas.html. Chiapa, C., Prina, S., & Parker, A. (2016). The effects of financial inclusion on children’s schooling, and parental aspirations and expectations. Journal of International Development, 28(5), 683–696. http://dio.org/10.1002/jid.3137. Cuartas, J., & Moya, A. (2016). Self-fulfilling prophecies: Expectations, psychological distress, and social mobility in Colombia (Working Paper). Dalton, P. S., Ghosal, S., & Mani, A. (2016). Poverty and aspirations failure. The Economic Journal, 126(590), 165–188. https://doi.org/10.1111/ecoj.12210. Delavande, A., Giné, X., & McKenzie, D. (2011). Measuring subjective expectations in developing countries: A critical review and new evidence. Journal of Development Economics, 94(2), 151–163. https://doi.org/10.1016/j. jdeveco.2010.01.008. De Quidt, J., & Haushofer, H. (2019). Depression through the lens of economics: A research agenda. In C. Barrett, M. R. Carter, & J.-P. Chavas (Eds.), The economics of poverty traps. Chicago: University of Chicago Press. Doctors Without Borders. (2006). Living with fear: The cycle of violence in Colombia. Retrieved from https://issuu.com/msf_australia/docs/colombia_ living_in_fear_0406. Doctors Without Borders (2010). Three-Time Victims: Victims of Violence, Silence, and Neglect. Retrieved from https://doctorswithoutborders.org/sites/ default/files/2018-08/MSF-Colombia-Three-Time-Victims.pdf. Duflo, E., & Duflo, E. (2012). Human values and the design of the fight against poverty. The Tanner lectures on human values. . Harvard University. Glewwe, P., Ross, P. H., & Wydick, B. (2017). Developing hope among impoverished children: Using child self-portraits to measure poverty program impacts. Journal of Human Resources, 53(2), 330–355. https://doi.org/10.3368/ jhr.53.2.0816-8112R1. Garber, J., & Seligman, M. E. P. (1980). Human Helplessness: Theory and applications. New York: Academic Press. Human Rights Watch. (2010). World report 2010: Events of 2009. Retrieved from http://www.hrw.org/sites/default/files/reports/wr2010.pdf. Ibáñez, A. M., & Moya, A. (2010a). Vulnerability of victims of civil conflicts: Empirical evidence for the displaced population in Colombia. World Development, 38(4), 647–663. https://doi.org/10.1016/j.worlddev.2009.11.015. Ibáñez, A. M., & Moya, A. (2010b). Do conflicts create poverty traps? Asset losses and recovery for displaced households in Colombia. In R. Di Tella, S. Edwards, & E. Schargrodsky (Eds.), The economics of crime: Lessons for and from Latin America (pp. 137–172). Chicago: University of Chicago Press. http://doi.org/ 10.1093/ acprof:oso/9780199378296.001.0001. Ibáñez, A. M., & Moya, A. (2016). Who stays and who leaves in genocides and mass killings? In C. H. Anderton & J. Brauer (Eds.), Economics aspects of genocide, mass atrocities, and their prevention (pp. 137–162). New York: Oxford University Press. Imbens, G. (2003). Sensitivity to exogeneity assumptions in program evaluation. American Economic Review, 93(2), 126–132. https://doi.org/10.1257/ 000282803321946921. Jensen, R., & Oster, E. (2009). The power of TV: Cable television and women’s status in India. The Quarterly Journal of Economics, 124(3), 1057–1094. https://doi.org/ 10.1162/qjec.2009.124.3.1057. Justino, P. (2011). War and poverty. In M. R. Garfinkel & S. Skarpedas (Eds.), Oxford Handbook of the Economics of Peace and Security. Oxford: Oxford University Press. Lipman, A., & Havens, E. (1965). The Colombian Violencia: An ex post facto experiment. Social Forces, 44(2), 238–245. https://doi.org/10.1093/sf/44.2.238.

115

Lybbert, T. J., & Wydick, B. (2018). Poverty, aspirations, and the economics of hope. Economic Development and Cultural Change, 66(4), 709–753. https://doi.org/ 10.1086/696968. Kalyvas, S. (2006). The logic of violence in civil war. Cambridge: Cambridge University Press. http://doi.org/10.1017/CBO9780511818462. Kessler, R. C., Sonnega, A., Bromet, E., Hughes, M., & Nelson, C. B. (1995). Posttraumatic stress disorder in the national comorbidity survey. Archives of General Psychiatry, 52(12), 1048–1060. https://doi.org/10.1001/ archpsyc.1995.03950240066012. Kessler, R. C., Nelson, C. B., McGonagle, K. A., Liu, J., Swartz, M., & Blazer, D. G. (1996). Comorbidity of DSM-III-R major depressive disorder in the general population: Results from the US National Comorbidity Survey. The British Journal of Psychiatry, 168(30), 17–30. Kosec, K., & Hyunjung, C. (2017). Aspirations and the role of social protection: Evidence from a natural disaster in rural Pakistan. World Development, 97, 49–66. https://doi.org/10.1016/j.worlddev.2017.03.039. Krishna, A. (2001). Moving from the stock of social capital to the flow of benefits: The role of agency. World Development, 29(6), 925–1243. https://doi.org/ 10.1016/S0305-750X(01)00020-1. Krishna, A. (2004). Escaping poverty and becoming poor: Who gains, who loses, and why? World Development, 32(1), 121–136. https://doi.org/10.1016/ j.worlddev.2003.08.002. Krishna, A. (2006). Pathways out of and into poverty in 36 villages of Andhra Pradesh, India. World Development, 34(2), 271–288. https://doi.org/10.1016/ j.worlddev.2005.08.003. Macours, K., & Vakis, R. (2014). Changing households’ investment behaviors through social interactions with local leaders: Evidence from a randomized transfer programme. Economic Journal, 124(576), 607–633. https://doi.org/ 10.1111/ecoj.12145. Matijasevic, M.T, Velásquez, L., Villada, C., & Ramírez, M. (2007). Moving out of poverty: Understanding freedom, growth, and democracy from the bottom-up. Colombia National Synthesis Report. Manizales, Colombia. Retrieved from http://siteresources.worldbank.org/INTMOVOUTPOV/Resources/21042151187710751691/Colombia_synthesis.pdf. Mollica, R. F., McInnes, K., Poole, C., & Tor, S. (1998). Dose-effect relationships of trauma to symptoms of depression and post-traumatic stress disorder among Cambodian survivors of mass violence. The British Journal of Psychiatry, 173(6), 482–488. http://doi.org/ 10.1192/bjp.173.6.482. Moya, A. (2018). Violence, psychological trauma, and risk attitudes: Evidence from victims of violence in Colombia. Journal of Development Economics, 131, 15–27. https://doi.org/10.1016/j.jdeveco.2017.11.001. Narayan, D., Pritchett, L., & Kapoor, S. (2009). Moving out of poverty: Volume 2. Success from the bottom up. Moving out of poverty. Washington, DC: World Bank and Palgrave Macmillan. Retrieved from https:// openknowledge.worldbank.org/handle/10986/11838. Narayan, D., & Petesch, P. (2010). Moving out of poverty: Volume 4. Rising from the ashes of conflict. Moving out of poverty. Washington, DC: World Bank and Palgrave Macmillan. Retrieved from https://openknowledge.worldbank.org/ handle/10986/11837. National Victims Unit. (n.d.). National victims’ unit official registry. Retrieved from http://rni.unidadvictimas.gov.co. Ray, D. (2006). Aspirations, poverty, and economic change. In A. Banerjee, R. Benabou, & D. Mookherjee (Eds.), Understanding poverty (pp. 409–443). Oxford, UK: Oxford University Press. Richards, A., Ospina-Duque, J., Barrera-Valencia, M., Escobar-Rincón, J., ArdilaGutiérrez, M., Metzler, T., & Marmar, C. (2011). Posttraumatic stress disorder, anxiety and depression symptoms, and psychosocial treatment needs in Colombians internally displaced by armed conflict: A mixed-method evaluation. Psychological Trauma: Theory, Research, Practice, and Policy, 3(4), 384–393. https://doi.org/10.1037/a0022257. Rotter, J. B. (1966). Generalized expectancies for internal versus external control of reinforcement. Psychological Monographs: General and Applied, 80(1), 1–28. https://doi.org/10.1037/h0092976. Seligman, M. E. P. (1975). Helplessness: On depression, development and death. San Francisco: W.H. Freeman. Sen, A. (1999). Development as freedom. Oxford University Press. Shultz, J. M., Garfin, D. R., Espinel, Z., Araya, R., Oquendo, M. A., Wainberg, M. L., ... Neria, Y. (2014). Internally displaced ‘‘Victims of Armed Conflict” in Colombia: The trajectory and trauma signature of forced migration. Current Psychiatry Reports, 16(10), 475. https://doi.org/10.1007/s11920-014-0475-7. Sympson, S. C. (2000). Rediscovering hope: Understanding and working with survivors of trauma. In C. R. Snyder (Ed.), Handbook of hope (pp. 285–300). Elsevier. https://doi.org/10.1016/B978-0-12-654050-5.X5000-3. Tedeschi, R. G., & Calhoun, L. G. (2004). Posttraumatic growth: Conceptual foundations and empirical evidence. Psychological Inquiry, 15(1), 1–18. https:// doi.org/10.1207/s15327965pli1501_01. United Nations High Commissioner for Refugees (UNHCR). (2016). Population statistics. Retrieved from http://popstats.unhcr.org/en/overview. Voors, M. J., Nillesen, E. E., Verwimp, P., Bulte, E. H., Lensink, R., & Van Soest, D. P. (2012). Violent conflict and behavior: A field experiment in Burundi. American Economic Review, 102(2), 941–964. https://doi.org/10.1257/aer.102.2.941. Yehuda, R. (2002). Post-traumatic stress disorder. New England Journal of Medicine, 346, 108–114. https://doi.org/10.1056/NEJMra012941.