The Scars of the Eelam War: Eroded Trust in North-Eastern Sri Lanka

Abstract

This study explores the legacies of the protracted 1983–2009 Sri Lankan civil conflict using original household survey data. By differentiating individual- and household-level war exposure, voluntary and involuntary military service experience, and loss of family members of soldiers and civilians, we evaluated the influence of a wide array of war-time experiences on the trust level of people in war-torn regions. We found that civil conflict undermined political trust, and heightened inter as well as intra-ethnic divisions among the Sinhalese and Tamils in Sri Lanka. Thus, the Sri Lankan government needs to take action not only to restore political trust but also to promote reconciliation between the Sinhala and Tamil communities as well as within the Tamil community in war-torn regions.

This research was supported by the Japan Society for the Promotion of Science KAKENHI Grant-in-Aid for Scientific Research (Grant Number 16KT0043 and 20H01497). All remaining errors are our own. Declarations of interest: none.

1 Introduction

Civil conflict leaves enduring legacies, from destroying physical capital to reshaping the social and political preferences of individuals.¹ Macroeconomic studies have provided evidence of the rapid recovery from conflicts in both developed and developing countries. Existing literature on social and political preferences has found that conflict may increase people’s cooperation and pro-social behavior in several dimensions but not in others. On the one hand, results consistently show a positive and significant influence of civil conflict on participation in social groups, community leadership, and pro-social behavior in experimental games; on the other hand, there are no conclusive results for voting behaviors, knowledge or interest in politics, and trust.²

Social and political preferences matter to society from multiple dimensions. For example, trust is essential to successful market development³ and economic growth.⁴ Distrust of state institutions may lead to reduced compliance with government regulations⁵ and lower willingness to pay taxes.⁶ Previous studies, a large number of which focus on African countries, have failed to generate consistent results concerning the social and political legacies of civil conflict.⁷

This study adds to the understanding of social and political legacies of civil conflict by considering one of the longest civil conflicts in the world, the Eelam War in Sri Lanka. We documented the influence of the civil conflict on trust, using original representative household survey data. More than a decade has passed since the end of the civil conflict and Sri Lanka is still marred by tensions between different ethnic groups. Yet, there is scant evidence of the legacies of this conflict.

Our empirical results reveal that exposure to violence undermines political trust, and deepens both inter and intra-ethnic divisions between the Sinhalese and Tamils. Our analysis makes two important contributions to the literature. First, it contributes to an increasing number of studies addressing the legacies of civil conflict by looking at trust in a war-torn region. Second, a comprehensive list of war-time experiences enables us to differentiate war-time victimization into individual and household-level war exposure, voluntary and involuntary military services, and loss of family members of soldiers and civilians.

The rest of this paper is structured as follows. Section 2 provides a brief account of recent Sri Lankan history. Section 3 reviews the relevant literature on how to measure trust as well as the legacies of civil conflict. Section 4 explains the sampling procedure and data collection of our original household survey. Section 5 presents the empirical strategy. Section 6 offers the empirical results. Section 7 concludes the paper.

2 The Eelam War: A Brief History

Sri Lanka was ravaged by a conflict that started in 1983 and lasted until 2009. It was one of the longest civil conflicts in the world, lasting more than 25 years. It is estimated to have caused between 80,000 and 100,000 deaths.⁸

There are three primary ethnic groups in Sri Lanka: Sinhalese, who are predominantly Sinhala-speaking Buddhists; Sri Lankan Tamils, who are mainly Tamil-speaking Hindus; and Moors, who are Tamil-speaking Muslims. These three ethnic groups made up 74, 13, and 7% of the entire population, respectively, based on the 1981 population census.⁹

Originally, the British ruled Sri Lanka “by communalist representation, whereby each minority group had a say in political matter …”.¹⁰ Sinhalese and Tamil politicians worked as a broad coalition to obtain more concessions from the British, but these two ethnic groups differed on the distribution of power in the legislature.¹¹ The practice of communalist representation was abolished by the 1931 Donoughmore Reforms, ushering in guaranteed universal suffrage and shifting representation to the majority Sinhalese.

During the colonial period, American missionaries taught English in the northern part of the island, where a majority of the people were Tamils.¹² Due to the continued use of English as the official language immediately after independence, the Sri Lankan Tamils were perceived as having access to a disproportionate share of power and over-representation in public services, and higher education, as a consequence of educational opportunities during the colonial period.¹³ Sri Lanka gained independence from the United Kingdom in 1948 and power was transferred peacefully to the Sri Lankan English-speaking upper-class elite.

Recognizing the under-representation of the Sinhalese, in the aftermath of independence, the first Prime Minister of Sri Lanka, Don Stephen Senanayake passed legislation to disenfranchise Indian Tamils in 1948, giving the Sinhalese a two-thirds majority in Parliament.¹⁴ The successor of Senanayake, Solomon West Ridgeway Bandaranaike, also made use of the over-representation of Tamils in public services, and then, passed the Sinhala Only Act in 1956 (lasting until 1987), under which Sinhala replaced English as the official language to marginalize the rights of Sri Lankan Tamils and English speakers.¹⁵ At the same time, Bandaranaike gave sympathetic attention to Buddhist discontent, which merged well with the language policy in creating a Sinhala-Buddhist nationalism.¹⁶ Since then, a growing number of people began to perceive the state as bestowing public goods selectively, breeding mistrust between ethnic groups.¹⁷ Worse still, Tamil youth were further alienated due to the policy of standardization in university admission introduced in 1971.¹⁸

In response to a series of unfavorable policies, the Tamils mobilized protests by non-violent means at first.¹⁹ However, after the 1970s, Tamils increasingly expressed discontent in the form of militant groups instead of calling for equal rights through democratic means. Velupillai Prabhakaran established the Tamil New Tigers, later known as the Liberation Tigers of Tamil Eelam (LTTE) in 1976, they aimed to build a separate Tamil state in the Northern and Eastern provinces where the Sri Lankan Tamils predominate.²⁰

The absence of sustained ethnic conflict before the late 1970s and the comparatively successful development record of Sri Lanka make its internal strife among the most puzzling domestic conflicts.²¹ Figure 1 provides a picture of the events from the 1970s to 2009 using the Global Terrorism Database (GTB)²² and The Uppsala Georeferenced Event Dataset Global Version 18.1(UCDP GED).²³ Even though these datasets are not completely accurate, they provide a rough estimation of the overall scale of the violence.

Fig. 1

Intensity of conflict events in Sri Lanka, 1975–2009. (Note This figure was created by the authors based on data from the Global Terrorism Database (GTD) and Uppsala Georeferenced Event Dataset (UGED) Global Version 18.1. GTD recorded 2,932 terrorism incidents from 1975 to 2009, 1,597 of which the LTTE was involved in. The UGED recorded 4573 episodes from 1989 to 2009, among which the LTTE was involved in 4325 episodes. An event is defined as “an incident where armed force was used by an organized actor against another organized actor, or against civilians, resulting in at least 1 direct death at a specific location and a specific date” in the UGED.Footnote

Högbladh Stina, “UCDP GED Codebook Version 20.1”, 2020, https://ucdp.uu.se/downloads/ged/ged201.pdf.

The GTD collected only terrorist attacks and defined it as “the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious or social goal through fear, coercion or intimidation.”Footnote

START, “Global Terrorism Database Codebook: Methodology, Inclusion Criteria, and Variables”, https://www.start.umd.edu/gtd/downloads/Codebook.pdf.

)

In the period prior to 1983, there was a relatively low intensity of violence involving the LTTE (Fig. 1). In the early stage, the LTTE was only one among several militant groups and mainly targeted Tamils in the government or rival groups more than the national security forces.²⁶

In 1983, the LTTE launched a full-scale armed conflict. Throughout the conflict, both the LTTE and the Sri Lankan Armed Forces (SLAF) launched attacks not only against each other but also on indiscriminate targets and carried out targeted attacks on civilians.²⁷ The LTTE had frequently targeted political opponents, including many Tamil politicians and civilians, with bombs and forcibly recruited Sri Lankan Tamils, including women and children, into its forces.²⁸ Multiple international actors, such as the Norwegian government, initiated several rounds of peace talks from 1985 to 2006, but all of them eventually proved futile. With an unprecedented military force, the Sri Lankan government launched a major offensive in January 2009. During the final stage of the conflict, the LTTE continued to forcibly recruit civilians into its ranks, used them as human shields, and shot at Tamil civilians who tried to flee the fighting scene. The government forces finally declared victory over the LTTE in May 2009. The UN estimated that at least 100,000 people were killed in the civil war between 1972 and 2009, 40,000 of them in the last months of fighting.²⁹ A UN panel of enquiry accused both the Sri Lankan government and the LTTE of human rights violations in the process of the conflict,³⁰ but little progress has been made on finding the truth and delivering justice, for those who suffered from the war-time activities, in the aftermath of the war.³¹

3 Literature Review

In this study, we assessed legacies of the civil conflict in Sri Lanka from the perspective of trust. Even though it is widely acknowledged that trust plays an essential role in social and political theories, there is no consensus on its meaning.³² Thus, we first define the concept of trust and discuss how we measured it in this study, and then we review the literature on the influence of civil conflict on trust.

3.1 Defining and Measuring Trust

3.1.1 Defining Trust

Following Bauer and Freitag (2017),³³ we consider trust as a situation-specific expectation instead of a behavior. In other words, truster A judges the trustworthiness of trustee B regarding some behavior X in context Y at time Z. For simplification, we define trust as a generalized behavior, context, and time-independent expectation of truster A on the trustworthiness of trustee B. This kind of reduced statement has also been termed generalized trust in some previous studies, reflecting a “stable” starting level of a person’s trust.³⁴ Scholars differentiate the concept of generalized trust into other sub-concepts according to who or what trustee B is. For example, political trust is widely used in cases in which trustee B comes from the political sphere, like a government or a political party. Meanwhile, interpersonal trust is frequently used in cases in which both the truster and trustee are individuals or groups.

3.1.2 Measuring Trust

Previous studies have used self-reported or behavioral measures to evaluate the trust level of people. Stokes (1962)³⁵ was one of the first studies to measure political trust using self-reported measures, which later became known as the trust-in-government question.³⁶ Many surveys still use this measure and list a number of institutions to be rated by the respondents.³⁷

When measuring interpersonal trust, previous studies have used not only self-reported measures but also behavioral measures. Self-reported measures ask attitudinal questions about trust toward most people in general; they were first used in the 1940s in a questionnaire.³⁸Even though some innovations have been introduced to measure interpersonal trust in recent decades, such as the wallet question³⁹ and the Interpersonal Trust Scale,⁴⁰ modified versions of the “most-people” question are still the most widely used to measure interpersonal trust.⁴¹ However, this kind of question has been criticized for limited links to trusting behaviors in the real world and ambiguity in interpretation.⁴² Thus, scholars like Levi and Stoker (2000)⁴³ have called for behavioral measures of trust to overcome the weakness of abstract questions in standard survey-based measures.

The behavioral measure of trust aims to infer trusting expectations of individuals by their decisions, behaviors, and reactions using lab experiments. It started out with the prisoner’s dilemma by Deutsch (1960)⁴⁴ and now mainly relies on the classic “trust game” first introduced by Berg et al. (1995).⁴⁵ In a classic trust game, trust is measured by the amount of money sent by trusters, and trustworthiness is measured by the amount returned by trustees. However, the evidence is mixed concerning what the trust game predicts and what it measures.⁴⁶

Then, how are these two kinds of measures correlated? Glaeser et al. (2000)⁴⁷ provide one of the earliest studies disentangling the relationship between survey-based measures, past trusting behavior, and what the trust game measures using a convenience sample of students. They find that past trusting behaviors, such as lending money, are correlated with trusting behavior in the experiment. Self-reported measures using most-people questions are a proxy for trustworthiness in the experiment. However, due to the convenience sampling of the subjects and other methodological reasons, Glaeser et al. (2000) were criticized for not measuring trust.⁴⁸

In Bangladesh,⁴⁹ Peru,⁵⁰ and Russia,⁵¹ results have revealed that the most-people questions can measure trustworthiness better than trust. Meanwhile, Sapienza et al. (2013)⁵² and Banerjee (2018)⁵³ found that the most-people question measures one’s expectation rather than one’s own trustworthiness in the laboratory. Previous studies have produced mixed evidence about how these two kinds of measures are related. Differences in these two kinds of measures may be due to sample selection, since self-reported measures typically rely on population samples while behavioral measures are restricted in a specific subgroup of the population. Going beyond previous studies, Wilson (2018)⁵⁴ used a population sample from two Russian republics to compare self-reported as well as behavioral measures. In line with previous studies, Wilson (2018) confirmed that a mismatch exists between the self-reported measures and behavioral measures. Most-people questions are uncorrelated or weakly correlated with trust and trustworthiness in trust games.

Both self-reported and behavioral measures of trust have methodological drawbacks.⁵⁵ Self-reported measures are superior to behavioral measures in the following aspects. First, self-reported measures can elaborate many dimensions of trust by changing the category of the trustee B into political institutions, specific groups, or people.⁵⁶ Second, self-reported measures can be used in large-scale representative surveys at a low cost. Third, self-reported measures can adapt to a trade-off between generality and specificity by specifying a particular trustee B and behavior X.⁵⁷

In post-conflict settings, like Sri Lanka, a comprehensive measure of trust is urgently needed. That is because it is difficult to realize post-conflict reconstruction without trust in government. Meanwhile, it is difficult to restore economic activities and reconciliation without trust-based cooperative actions and transactions. Therefore, in our study, we use a wide array of survey questions to measure trust in post-conflict Sri Lanka. We overcame the weakness of the most-people question by using more specific questions to differentiate who or what trustee B is and to measure the behavior, context, and time-independent expectation of truster A on the trustworthiness of trustee B.

3.2 Influence of Civil Conflict on Trust

An increasing number of previous studies have examined the legacies of civil conflict in both real-life and experimental settings, focusing on post-war outcomes.⁵⁸ As discussed in the former subsection, we use self-report questions to measure trust to obtain a relatively complete picture of the war-torn area in Sri Lanka using a representative sample. The literature reviewed in this subsection concerns only observational studies that also used self-reported measures.

Using a nationally representative survey in Sierra Leone 3–5 years after the end of the civil conflict, Bellows and Miguel (2009)⁵⁹ found that individuals whose households had direct war exposure were more likely to trust people from outside of their community.⁶⁰ Relying on three rounds of nationally representative survey data from Afrobarometer in Uganda, De Luca and Verpoorten (2015) found that district-level war exposure derived from the Armed Conflict Location Events Dataset (ACLED) undermined generalized trust amidst violence.⁶¹ However, this outcome was strongly increased in the affected areas only a few years after the end of the war.

By contrast, exploiting two waves of survey data from Afrobarometer and ACLED in Uganda, Rohner et al. (2013) found that more intense fighting at the county-level decreased generalized trust in both OLS and 2SLS estimates but had no significant effect on trust toward known people and relatives.⁶² Similarly, combining household-level and municipal-level war exposure data in Kosovo, Kijewski and Freitag (2018) revealed that both these kinds of war exposures are related to a lower level of trust in people in the neighborhood.⁶³ For Burundi, Voors and Bulte (2014) fail to find a significant effect of war exposure on villagers’ trust toward fellow villagers.⁶⁴

Most previous studies focus on interpersonal trust and pay little attention to trust toward institutions albeit with some exceptions. For instance, Grosjean (2014) used nationally representative surveys in 35 countries to explore the influence of World War II and civil conflict in Western Europe on trust toward central institutions and people in general.⁶⁵ She finds that victimization in conflict, especially in civil conflict, is negatively related to trust in central institutions, but fails to provide evidence on the relationship between any type of conflict victimization and generalized trust. Using geo-referenced survey data and village-level information on conflict, De Juan and Pierskalla (2016) provide evidence that village-level war exposure to civil conflict undermines trust toward the national government in Nepal.⁶⁶

In summary, there is a consensus that civil conflict is negatively associated with trust toward institutions, although previous studies disagree on the direction of influence on interpersonal trust. The first possible reason is that countries usually have different pre-war situations and every civil conflict is different. Another reason may be that there are different ways of measuring war exposure; most of the previous studies deal only with a limited set of war-time experiences. Even though it is well established that traumatic experiences during the war may profoundly change individual beliefs and preferences,⁶⁷ there is no consensus about which kind of war experience is the most transformative.⁶⁸ Some previous studies directly focus on former soldiers and compare people with soldier experience and those without soldier experience. Other studies measure war victimization at the household level without differentiating the war-time experience of the respondents themselves and their family members. It is reasonable to expect that the influences of war experience differ between the general population and war veterans, and war-time experience may influence members differently even in the same household. Similarly, households with war veterans may be systematically different from households with no war veterans.

In addition, there is little evidence on how the civil conflict affected antagonistic ethnic groups differently given that a high proportion of minority ethnic groups engaged in ethnicity-based rebellions against states, especially from the 1940s to the 1990s.⁶⁹ The wide array of data on war-time experience makes this study among the most comprehensive datasets in a post-conflict context. We not only differentiate war exposure between individual and household levels but we also differentiate household-level war victimization based on whether there were war veterans in the household, and we evaluate their influences by ethnic groups.

4 Sampling Design, Data, and Measurement

4.1 Sampling Design

The Survey of Conflict-Affected Regions in Sri Lanka (SCARS 2018) dataset used in this study was collected by the authors from March 2018 to May 2018 collaborating with the Kandy Consulting Group (KCG) based in Kandy, Sri Lanka. Given that most battles during the civil conflict were fought in Eastern and Northern provinces, subjects were sampled from eight districts across these two provinces using a multi-stage stratified cluster sampling method to obtain a representative sample. A detailed description of this survey was described elsewhere.⁷⁰

In short, in the first stage, we classified the Grama Niladhari (GN) divisions⁷¹ into four strata based on the population share of three ethnic groups; namely, Sri Lankan Tamil dominant, Sinhalese dominant, Moor dominant, or mixed-ethnicity divisions according to the 2012 population census. A dominant ethnic group is defined as consisting of more than 90% of the population in each GN division. We then randomly chose a certain number of GN from each stratum, with the number of GNs in each stratum being proportional to the population share of each stratum in a certain district. A total of eight GN divisions were randomly selected from each district. In the last stage, within each GN sampled above, 25 households were randomly chosen from the list of voters by the KCG. A pilot survey was conducted in November 2017 in the Trincomalee district to finalize the survey questionnaire, which was then translated into Tamil and Sinhalese. The translated questionnaires were used for the interviews with 1600 households conducted by enumerators of the same ethnicity in their own language.

In the early stage of data collection in March 2018, an ethnic riot between radical Buddhists and Muslims broke out and a curfew was imposed by the government. Moreover, this civil conflict was between the majority Sinhalese and the minority Sri Lankan Tamils. We finally limited our sample to 1,308 Sinhalese and Tamil households, of which 220 were Sinhalese and 1,088 were Sri Lankan Tamils.⁷² For the empirical analysis and comparison, we further limited the sample to respondents who answered all trust-related questions, leaving 976 observations.⁷³

4.2 Data and Measurement

To check the representativeness of our sample, we compared basic demographic characteristics with the latest Household Income and Expenditure Survey 2012 (HIES 2012) in Northern and Eastern provinces in Table 1. Table 1 confirms that the household characteristics, ethnic composition, and religious composition of the two surveys are quite similar.

Table 1 Representativeness of the SCARS 2018 sample: A comparison with HIES 2012

In the sampled households, we asked the following questions to measure the trust level of the main respondents: “In general, the government/military/most Sri Lankan Tamils/most Sinhalese are trustworthy?” Questions were answered using a 5-point Likert scale: 1 (strongly disagree), 2 (disagree), 3 (neither agree nor disagree), 4 (agree), and 5 (strongly agree). For simplicity of the empirical analysis, we combined answers of trusting Sinhalese and Tamils into two categories, namely co-ethnic trust and non-co-ethnic trust.

5 Empirical Strategy

The literature discussed in Sect. 3 suggests that both individual- and household-level war victimization may affect post-war behaviors. We included these two kinds of victimization and compared the outcome variables of respondents who suffered from different degrees and types of war victimization during the civil conflict. We devised the following equation:

$$ Y_{{\text{ijk}}} = \alpha + {{\varvec{Individual}}}_{{{\varvec{ij}}}}^{\prime} \;{{\varvec{\varGamma}}} + {{\varvec{Household}}}_{{{\varvec{ij}}}}^{\prime} \;{{\varvec{\varPhi}}} + {{\varvec{X}}}_{{{\varvec{ij}}}}^{\prime} \Psi + {{\varvec{Z}}}_{{\varvec{j}}}^{ {\prime}} {{\varvec{B}}} + \delta_k + \varepsilon_{{\text{ijk}}} $$

(1)

where Y_ijk denotes dummy variables of outcomes for individual i in household j living in GN division k at the time of the interview: Y_ijk equals 1 if the respondents chose “strongly agree” or “agree” in the trust-related questions, and 0 otherwise.

We are interested in the influences of both individual and household-level war exposure on social and political outcomes. We measured individual-level war exposure Individual_ij of individual i in household j using two variables: (1) war-related health difficulty index, which counts the total number of physical and mental health difficulties due to the civil conflict and (2) previous military service during civil conflict, which is a dummy variable for self-reported solider experience. Physical and mental health difficulties included difficulties in seeing, hearing, walking, cognition, day-to-day self-care, and voice communication in the respondents’ own language. We coded previous military service 1 if the respondent had ever joined the SLAF or the LTTE and 0 otherwise. The largest concern in identifying the causal effect of individual-level war exposure is that the correlation between war experience and the outcomes of interest is driven by reverse causality. In the case of trust, whether this reverse causality holds depends on how persistent trust is; if the current level of trust is highly correlated with the past level of trust, people with higher pre-war trust levels might have been exposed to violence due to self-selection. Previous studies disclosed that trust is a relatively stable human trait⁷⁴; thus, we aimed to alleviate these problems using the following approaches. First, we differentiated previous military service into voluntary and involuntary military service relying on self-reported answers. People who volunteered to join the army might have done so due to unobservable pre-war traits: Sinhalese people might have volunteered to join the SLAF due to a higher level of trust in the government and the military; Sri Lankan Tamils might have volunteered to join the LTTE due to mistrust of the government or non-co-ethnics. For example, anecdotal evidence showed that some Tamils volunteered to join the LTTE after witnessing harassment of their family members by Sinhalese officials.⁷⁵ In that case, the causality between individual-level war experience and trust levels may be reversed. Human rights organizations have recorded that many civilians, including children, were recruited into the LTTE forcibly.⁷⁶ Respondents with involuntary military service were less likely to be exposed to violence due to self-selection. Second, we control for observable individual characteristics X_ij, which includes age, female dummy, and years of schooling of the respondents, which may predict voluntary participation in the armed forces.

We measured household-level war exposure Household_ij of individual i in household j by constructing the following three variables: (1) total number of current household members who were former soldiers; (2) total number of family member losses; and (3) property loss index. The total number of family losses was constructed using the total number of family members who were killed or declared missing during the conflict. This variable includes family losses of both soldiers and civilians. The property loss index was constructed using three war experience questions, which aggregated the total number of times one experienced asset loss, land confiscation, and household damage.Due to the absence of pre-war measures of household characteristics, the correlation between household-level war exposure and post-war outcomes of a certain household member might have been driven by omitted variable bias arising from some pre-war traits of their household members who suffered from violence personally. For example, more trusting people might have been more likely to join the armed forces and thus, their family members would have been more likely to live in a family with household-level war exposure. Considering that household members usually share attitudes and values to some extent, the trust level of respondents whose family members had military service experience may differ from those whose family members had no experience of military service. We overcame this problem using the following approaches. First, we included the total number of former soldiers currently living in the household. Second, we differentiated the victimization of family members into loss due to military service and loss of civilians. Third, we controlled for observable household characteristics Z_j, including dummies for female-headed households and recipients of the Samurdhi program, wealth quantiles, as well as age and years of schooling of the household head considering that poverty and lack of educational opportunities often fueled Tamils, especially those in the Eastern province, to join the LTTE.⁷⁷ Wealth quantiles are calculated using principle component analysis based on asset ownership.⁷⁸ The Samurdhi program is a major poverty alleviation program in Sri Lanka initiated by the government in 1995. Fourth, we eliminated possible bias caused by pre-existing spatial variations in trust level by the inclusion of GN division-level fixed effects δ_k. The GN division in Sri Lanka consists of either a collection of small villages or a part of a larger village. The GN fixed effects allowed us to isolate the variation in war exposure across neighbors within the same village where families were relatively homogeneous. Table 2 provides summary statistics for empirical analysis.

Table 2 Summary statistics

Even after controlling for individual and household characteristics, and GN division fixed effects, we could not fully rule out the possibility of selection into war victimization. We dealt with this by restricting our attention to sub-samples for which targeted violence was less likely to occur. The first sub-sample contains respondents with no war-related health difficulties, voluntary military service experience, and no self-reported family members who were former soldiers, or who were killed or declared missing due to military service. The first sub-sample excludes people who were most likely to be exposed to the conflict by self-selection. The second sub-sample further excluded respondents without involuntary military service experience and focused on people born after the war started in 1983. The second sub-sample reduced the possibility of falsely claiming using their involuntary military service experience and limited the sample to respondents who were too young to be involved in the violence personally.

6 Results

6.1 Baseline Results

We begin by analyzing the relationship between war victimization and the perceived trustworthiness of the government, the military, co-ethnics, and non-co-ethnics using the full sample. In all the specifications, we control for individual and household characteristics, and GN fixed effects. All estimates are weighted by inverse sampling probabilities. Standard errors are clustered at the GN division level.

We differentiate war exposure into individual and household-level war exposure. The estimated results reveal that respondents who suffered from war-related health difficulties were less likely to trust the government, the military, and non-co-ethnics. The point estimates on the war-related health difficulty index show that an increase from 0 to 1 in war-related health difficulties is associated with a decrease of approximately 37.1, 29.6, and 25.8 percentage points in the probability of trusting the government, the military, and non-co-ethnics, respectively.

Personal military service is not significantly related to the trust level of the respondents, as shown in columns (1)–(4) of Table 3. This is possibly because we combine respondents with both voluntary and involuntary military service experience from different ethnic groups. Given that the main objective of the LTTE was to build a separate Tamil state in the Northern and Eastern provinces of Sri Lanka, Tamil soldiers with voluntary military service experience might have had different levels of trust in the government compared to Sinhalese or Tamil soldiers who were forcibly recruited into the LTTE. Thus, we differentiate military service into voluntary and involuntary military service. We include two interaction terms between Sri Lankan Tamils, involuntary and voluntary military service, in columns (5)–(8) of Table 3 to check whether the influence of military service differs by ethnic group and voluntarism. Since the analysis focuses only on two ethnic groups, Sinhalese and Sri Lankan Tamils, the estimated coefficients for involuntary or voluntary military service naturally indicate Sinhalese respondents in columns (5)–(8) of Table 3. As expected, Sri Lankan Tamil and Sinhalese former soldiers show different probabilities of trusting the government, co-ethnics, and non-co-ethnics, and the signs of the estimated coefficients depend on whether they served in the military forcibly or not. Sinhalese soldiers who joined the army voluntarily had a higher likelihood of trusting the government while those who were forced to join the army were less likely to trust the government and non-co-ethnics and were more likely to trust the co-ethnics. By contrast, Sri Lankan Tamil soldiers who volunteered to join the LTTE were less likely to trust the government and non-co-ethnics. Involuntary military service was associated with a higher probability of Tamil soldiers trusting non-co-ethnics than voluntary military service. The estimated results on voluntary military service are highly likely to be driven by reverse causalities, since they might have had different levels of pre-war trust in the government and other ethnicities compared to other groups and might have self-selected into the conflict.

Table 3 War exposure and trust, baseline

For household-level war exposure, we find that both the number of former soldiers among current family members and family member losses are positively associated with increases in the probability of trusting the co-ethnics; the magnitude of the coefficient for the number of former soldiers is five times larger than that for the number of family losses. Property loss is negatively associated with perceived trustworthiness of non-co-ethnics. In addition, recipients of the Samurdhi programs and respondents with higher education levels have a higher likelihood of trusting non-co-ethnics.

We compare the magnitude of the estimated coefficients of individual and household-level war exposure and find that personal and direct exposure to violence are more significantly correlated with the trust level of respondents than household-level war exposure. This result is quite intuitive and possibly because direct war experiences have a significantly larger effect on war-related distress, as suggested by previous studies.⁷⁹

The small magnitude of the estimated coefficient on family member loss may be because we do not differentiate between family loss due to military service or otherwise. In the following regression analysis, we further classify the number of family losses into family losses of soldiers or civilians and display the estimated results in Table 4.

Table 4 War exposure and trust, subcategory of household-level war exposure

The estimated results in columns (1)–(4) of Table 4 show that losses of household members due to military services are negatively associated with trusting the government, while losses of civilian household members are positively associated with trusting the co-ethnics. Similar to Table 3, to examine whether the influence of family losses on trust levels differs by ethnic group, we further include two interaction terms between Sri Lankan Tamils, and the number of family losses that were soldiers or civilians in columns (5)–(8) of Table 4. We find that the family loss of soldiers is not significantly associated with trust level, while family loss of civilians is positively associated with trusting the co-ethnics and negatively associated with trusting non-co-ethnics among Sinhalese respondents. By contrast, the estimated results reveal that family loss of soldiers is negatively associated with trusting the government or the military among Tamil respondents.

6.2 Alternative Sub-Samples

One concern of the OLS estimates displayed in the former subsection is that the correlation between war victimization and unobservable confounding factors may affect war victimization and post-conflict outcomes at the same time. Due to the absence of pre-war measures, it is difficult to isolate the effect of bias from self-selection into war victimization. We next estimate these relationships for sub-samples that are less likely to be affected by selection into violence. We first limit our attention to respondents who had no war-related health difficulties, no voluntary military service experience, and no household members were former soldiers or died or were declared missing due to military service. In this way, we can reduce bias arising from reverse causality and omitted variables. We further limit the sample to respondents who had no involuntary military service experience and who were born after 1983, since they were at least far less likely to self-select into violence. The sample sizes for these two sub-samples are 810 and 215, accounting for 83% and 22% of the full sample, respectively.

Columns (1)–(4) of Table 5 show the estimated results for the first sub-sample. We find that for Sinhalese respondents, involuntary military service is related to a lower likelihood of trusting the government and non-co-ethnics, and a higher likelihood of trusting the co-ethnics in both the full sample and the first sub-sample. Consistent results for the full sample and the first sub-sample on involuntary military service experience give us more confidence that this group of respondents is less likely to be exposed to the war by self-selection and the characteristics of household members. Similarly, we find that family loss of civilians is related to a lower likelihood of trusting the government and non-co-ethnics, and a higher likelihood of trusting the same ethnicity for Sinhalese respondents. These results imply that civil conflict not only has resulted in mistrust of the government but also has reduced ethnic mistrust among Sinhalese toward Tamils.

Table 5 War exposure and trust, results of sub-samples

By contrast, Tamils who were forced to join the LTTE showed a higher probability of trusting non-co-ethnics; family loss of civilians is positively associated with trusting non-co-ethnics as well. A higher likelihood of trusting other ethnicity may be due to social desirability bias among Tamil respondents. This is because, after the defeat of the LTTE, its former cadres received frequent military surveillance visits⁸⁰ and they might have been afraid of further questioning or interrogation by the security force. Meanwhile, mistrust within the Tamil community may be due to the following two reasons. First, many Tamils suffered from forced recruitment practices and coercion by the LTTE during the war, even though they did not fully believe in its cause and ideological goals. Second, the security force in the north-east of Sri Lanka made use of civilians for surveillance, which subsequently sowed the seeds of suspicion and distrust among Tamil communities.

7 Conclusions

Civil conflict can reshape people’s social and political perceptions. Using originally representative data for the Northern and Eastern provinces of Sri Lanka, we provided evidence on the possible consequences of the protracted civil conflict on mistrust of government, inter and intra-ethnic divisions by ethnicity. People included in this study were victimized by the civil conflict in different ways. Policy interventions may need to target different groups of people in different ways based on their victimization and experience during the war. The findings revealed in this study provide implications for post-war recovery policies in Sri Lanka as follows.

First and foremost, restoring political trust is extremely urgent especially among people with war-related health difficulties, Sinhalese who were forced to serve the military, and Tamils who volunteered to join the LTTE. Second, reconciliation is needed not only between the Sinhala and Tamil communities but also within the Tamil community. We found evidence of salient ethnic divisions between different ethnic groups after the brutality of the civil war especially among Sinhalese. Worse still, mistrust was prevalent within the Tamil community among those who lost their families during the war.

Even though we adopted several strategies to control for possible confounders, we might not have fully eliminated concerns about self-selection and omitted variable bias. However, remarkably robust evidence on the relationship between forced recruitment, family loss of civilians, and social and political outcomes give us confidence that self-selection and omitted variable bias might not be the main drivers of our results, at least among the first sub-sample of respondents. Due to data limitations, we could not provide channels through which civil conflict affects trust levels in war-torn regions of Sri Lanka, or why the influence of civil conflict differs by ethnicity. More research is needed to identify the specific mechanisms and to explain the reasons behind ethnic differences in post-war trust levels.