Facing displacement and a global pandemic: evidence from a fragile state

Abstract

We use novel survey data to assess the impact of the COVID-19 pandemic on internally displaced persons (IDPs) in Libya. Our analysis compares the effects of the pandemic for displaced and non-displaced citizens, controlling for individual and household characteristics and geo-localized measures of economic activity and conflict intensity. In our sample, 9.5% of respondents report that a household member has been infected by COVID-19, while 24.7% of them have suffered economic damages and 14.6% have experienced negative health effects due to the pandemic. IDPs do not display higher incidence of COVID-19 relative to comparable non-displaced individuals, but are about 60% more likely to report negative economic and health impacts caused by the pandemic. We provide suggestive evidence that the larger damages suffered by IDPs can be explained by their weaker economic status—which leads to more food insecurity and indebtedness—and by the discrimination they face in accessing health care.

1. Introduction

The presence of internally displaced persons (IDPs)—i.e., individuals forced to leave their homes and relocate to a different area in their own country—is a pervasive phenomenon in developing countries. As of end of 2020, UNHCR estimates that 48 million people in the world are internally displaced due to armed conflict, violence, or human rights violations. These individuals have escaped the most immediate life threats but are still residing in extremely hazardous countries and remain highly vulnerable to violence, social exclusion, and destitution.¹

The outbreak of the COVID-19 pandemic has posed a major challenge for economies and societies across the globe. While the core of academic and policy debate on the consequences of—and on the responses to—the COVID-19 pandemic revolved around the experience of wealthier countries, far less attention has been devoted to low- and middle-income countries (LMICs). We know even less about the effects of the pandemic in fragile and conflict-affected countries in which violence and insecurity impede accurate data collection, making the measurement of health and socio-economic outcomes extremely complicated. The World Bank estimates that about 90 million individuals have entered extreme poverty since the onset of the pandemic, leading to the first increase in global poverty over the last two decades [Lakner et al. (2021)]. The pandemic negatively affected the living standards in developing countries by provoking major income losses, drops in employment probability, and widespread food insecurity [Bundervoet et al. (2022), Egger et al. (2021), Hoogeveen and Lopez-Acevedo (2021)].² Learning about the experience of LMICs and of their most vulnerable populations in the midst of the pandemic is essential to conceive policy interventions that are tailored to their specific needs and challenges [Orcutt et al. (2020)]. Extrapolating from the experience of wealthier countries, instead, may lead to the implementation of measures that are ineffective and even self-defeating [Ma et al. (2021), Miguel and Mobarak (2021)]. The effects of COVID-19 are likely to be even more negative in conflict-affected countries where the pandemic shock adds to, and interacts with, pre-existing fragilities. In these settings, often characterized by failed governments and weak public health care systems, the population endures the health hazard and the economic hardship caused by the pandemic with very little public support. For population groups—such as the IDPs—who already start from a vulnerable status, the COVID-19 pandemic shock can lead to even direr consequences.

In this paper, we contribute to the global effort of documenting the effects of the pandemic on marginalized groups in fragile and conflict-affected countries by presenting the first evidence of the impact of COVID-19 on IDPs in Libya. Since the fall of the Gaddafi's regime in 2011, Libya has been experiencing political instability and violence. Each phase of the conflict resulted in an increasing number of IDPs [NRC (2021), UNHCR (2017)]. The outbreak of the conflict in 2011 displaced an estimated half a million Libyan citizens (almost 10% of the resident population), while the resurgence of military confrontations in 2014 triggered a second wave of displacement which was even larger and more persistent than the first one [IOM (2021c)]. The COVID-19 pandemic exacerbated the socio-economic weaknesses caused by political instability that characterize the country, adding another layer of hardship for the Libyan population [IOM (2021a)] and bringing an already fragile health care system to the brink of collapse. In such a setting, IDPs have been exposed—with little or null support—to both the health and economic hazards of the pandemic [ICMPD (2020)]. Yet, the possibility to target this marginalized group is limited by the lack of information on their location and needs. Information on the internal displacement of Libyans is fragmented and incomplete for two main reasons. First, any type of data collection in Libya is extremely difficult due to the ongoing conflict [Rahman and Di Maio (2020)]. Second, shifting patterns of displacement and return, together with cases of multiple displacements, make it particularly difficult to include IDPs in survey samples [El Taraboulsi-McCarthy et al. (2019)].

In our study, we assess the impact of COVID-19 pandemic on the Libyan population by using novel data from a phone survey conducted in Libya in 2021. This survey—whose module on migration and internal displacement we designed—represents the first household data collection since the onset of the conflict in 2011. In our sample, 9.5% of respondents report that a household member has been infected by COVID-19 over the last 12 months while 24.7% of them have suffered economic damages and 14.6% have experienced negative health effects due to the pandemic. Our analysis compares the effects of the pandemic for IDPs and non-displaced individuals, controlling for individual and household characteristics, as well as for geo-localized measures of economic activity and conflict intensity in the area of residence. In our data, displaced individuals report a similar incidence of COVID-19 infections to non-displaced individuals. Nevertheless, they are about 60% more likely to report negative economic and health impacts caused by the pandemic. Our results suggest that the larger damage suffered by IDPs cannot be explained by individual and household characteristics, nor by higher probability of contagion, but rather by their weaker economic status—which leads to more food insecurity and indebtedness—and by the discrimination they face in health care access.

Our paper contributes to the literature on the impact of COVID-19 in developing countries by providing novel primary data on an important phenomenon that is still largely unknown in its dimension and effects, especially in the context of fragile and conflict-affected states. We further add to previous studies by documenting how the pandemic differentially affects IDPs. Our findings show that large gaps in the level of hardship endured during the pandemic may arise even in the absence of a differential contagion risk. These results suggest that policy interventions in a fragile context with large presence of IDPs may need to focus more on preventing damage rather than on containing the spread of COVID-19 among marginalized population groups.

The paper is organized as follows. Section 2 discusses key background information on conflict and internal displacement in Libya and describes broader patterns of COVID-19 infection and mortality in the country. Section 3 provides an overview of our data collection and sampling approach. Section 4 describes our empirical strategy and section 5 presents our empirical findings. Section 6 concludes.

2. Background

2.1 Conflict and internal displacement in Libya

Libya has been in a situation of conflict and political unrest for more than a decade. In February 2011, in the wake of the Arab Spring protests, Libya witnessed a popular uprising against General Muammar Gaddafi, who had uninterruptedly ruled the country since 1969. This revolt marked the beginning of the so-called First Libyan Civil War, a violent conflict between various rebel groups and Gaddafi's loyalist army which ended with the toppling of the regime and the execution of its leader in October 2011. Gaddafi's death generated a power vacuum that led to a period of political instability and weak institutional control over the country [Eriksson (2016)]. This situation created the conditions for the beginning of the Second Libyan Civil War in 2014 [Fitzgerald and Toaldo (2016), Pack (2019)]: the formation of two competing governments—the UN-recognized Government of National Accord (GNA) based in Tripoli and the Benghazi-based Libya National Army (LNA)—with backing from other Arab states and “great powers” on both sides, led to political fragmentation and years of violent clashes [Fitzgerald and Toaldo (2016)].³ Terrorist groups and armed militias have exploited the turmoil and used the country as a base for radicalization and organized crime. In October 2020, all belligerent parties accepted a permanent ceasefire in the whole of Libya and, in March 2021, a Government of National Unity (GNU) was formed, increasing hopes of political and social stability. Yet, the dates of the first presidential and parliamentary elections since the onset of the civil war have been fixed and postponed several times and the political and economic situation remains complex and unpredictable.

Instability and violence in Libya have affected over 1.3 million people out of a 6.7 million population and left more than 450,000 people in need of humanitarian support [NRC (2021)]. The civilian population has been directly harmed because most of the battles and fighting have taken place in urban areas, residential neighborhoods, and even in city centers [Pack (2019)]. Hostilities have also badly damaged hospitals and health care centers [WHO (2021a)].

Each phase of the conflict resulted in an increasing number of IDPs (see Figure 1). The peak in fatalities of the First Libyan Civil war in 2011 forced an estimated half a million Libyan citizens to leave their homes and move elsewhere. The resurgence of violence in 2014, triggered a second wave of displacement which was larger and more persistent than the first one. Approximately 340 thousand citizens were forcedly displaced in 2014 alone, followed by more than 600 thousand IDPs over the years between 2015 and 2020. Unlike in 2011, many of those forcedly displaced from 2014 onward were unable to return home quickly and were displaced again as the front lines of the conflict shifted across the country. At the time of our survey (May 2021), over 640,000 IDPs are estimated to have returned to their homes, with the stock of IDPs in the country to be about 212,000 people [IOM (2021c)].

Figure 1. Conflict-related fatalities and IDPs in Libya (2011–2020). Note: The graph reports yearly data on the number of conflict-related fatalities (red line; source: ACLED data) and the number of new IDPs (in hundreds; green line; source: IOM Displacement Tracking Matrix).

The very little available information on the characteristics of the IDPs in Libya indicates that a substantial share of the households which were displaced at the beginning of the conflict was moderately well-off. Yet, their socio-economic conditions rapidly deteriorated as a consequence of the forced displacement [OCHA (2018)]. Insecurity, financial fragility, limited access to health, and other basic services are all critical issues affecting the well-being of the IDPs in the country [IOM (2021c)]. The majority of IDPs are believed to be located in urban areas, hosted by relatives or friends, or in informal settlements. Those living in informal settlements are considered to have the most acute needs, with limited access to adequate shelter, social protection, and health care [UNHCR (2013, 2018)].

2.2 COVID-19 in Libya

Libya has been significantly impacted by the global COVID-19 pandemic [IOM (2021a)].⁴ According to World Health Organization (WHO) data [WHO (2021b)], Libya recorded a cumulative number of over 390 thousand confirmed cases and 5,750 deaths as per January 2022. These figures correspond to almost 5,700 cases per 100 thousand resident population and 830 deaths per 1 million population. Although COVID-19 data in low-income countries suffer from severe measurement issues that make international comparisons potentially unreliable, WHO data suggest that Libya was the second hardest-hit country in the North African region (see Appendix Figure A.2).

The COVID-19 pandemic in Libya has created huge challenges for the provision of basic services, social protection, and health care. As in several other African countries [Shapira et al. (2021)], the COVID-19 crisis has brought the Libyan health care system to the brink of collapse. The combination of armed conflict, underinvestment in health infrastructure, and strong dependence on private health service providers has drastically reduced the capacity of the health sector in Libya to deal with the COVID-19 emergency [IOM (2021b)]. The response from the Libyan government included a variety of initiatives to cope with the pandemic. Nevertheless, a fragmented health sector and a lack of funding and human resources resulted in poor effectiveness of these interventions [IOM (2021a)]. Appendix Figure A.2 shows that Libya is lagging behind in the vaccination campaign relative to other Northern African countries: in January 2022, Libya display the lowest share of fully vaccinated citizens (12.6 per 100 population) in the region.

The negative impact of the COVID-19 pandemic on the Libyan population is not limited to the public health area. Due to the instability and insecurity which characterize the country since 2011, the Libyan economy has been struggling for years with rising unemployment, growing inflation and supply shortages [Rahman and Di Maio (2020)]. The COVID-19 outbreak has compounded these economic weaknesses. In addition to rising prices movement restrictions have led to difficulties in securing food and other basic needs [REACH (2020)]. While the government has not been able to put in place a general income support scheme for the Libyan citizens, the fact that a large share of Libya's active labor force is employed in the public sector—which continued to pay salaries to its employees—prevented the direst consequences of the pandemic crisis and avoided wide exposure to risk of destitution.⁵ Still, marginalized groups, including IDPs, were left totally exposed to both the economic and health effects of the pandemic, exacerbating their pre-existing fragilities [ICMPD (2020), IOM (2021a)].

3. Data, sample, and descriptive statistics

3.1 Data

Survey data

Our main source of data comes from the 2021 Libya High-Frequency Phone Survey-Social Protection (HFS-SP). This a novel ad-hoc survey that we contributed to design. The survey was conducted in Libya between April and May 2021 and it is part of the Social Protection Study, a project of the World Food Programme (WFP), the World Bank, and the Libyan Bureau of Statistics (LBSC) to assess household vulnerability, shocks, and coping mechanisms of Libyan households. Notably, this is the first official household survey since the beginning of the conflict in 2011.⁶ The questionnaire—administered to the head of the household or any respondent older than 17 years old—included detailed questions on socio-demographic characteristics of respondents and their households, employment status, income, health, housing, displacement, experience during the COVID-19 pandemic, etc. We participated in the overall design of the survey and we drafted all the questions relative to migration history, internal displacement, and exposure to conflict. In Appendix Section B, we provide more information about the survey methodology and present a validation of our sample using the few other existing data sources on the Libyan population and the IDPs in the country.

Data have been collected by the LBSC using a phone-based data collection method. Calls were made to respondents resident in all the 22 Mantikas (regions) in Libya and the sample was constructed to resemble the population share of each Mantika in the total population. Respondents phone numbers were randomly selected from a database provided by phone companies operating in the country.⁷ Due to movement restrictions and inaccessibility to some areas, phone survey have rapidly become the primary method for tracking economic conditions in developing countries during the pandemic [see for instance, Egger et al. (2021) and Miguel and Mobarak (2021)]. Indeed, the possibility of collecting data without the physical presence of a field enumerator makes phone surveys a highly valuable tool in dangerous and emergency situations, even more so if the target population is geographically mobile as in our case [Hoogeveen and Lopez-Acevedo (2021)]. Still, an important shortcoming of phone surveys is the risk of systematically undersampling individuals who have no access to phones or limited network coverage in their areas of residence [Bundervoet et al. (2022)]. Although the widespread mobile phone ownership among the Libyan population reduces the concern about this potential barrier [REACH (2021)], we may still worry that marginal sub-groups of the population—and poor IDPs in particular—may be underrepresented in our survey. In the absence of official records on the Libyan (and IDPs) population we cannot rule this possibility and we need to briefly discuss the consequences this would have for the empirical findings we present in this paper. Since our focus is on gaps in outcomes and perceptions between displaced and non-displaced Libyan citizens, undersampling poor IDPs would lead to estimate a lower bound of the real effect of being displaced, implying that the actual displacement gaps are even larger than those we report in the paper.

Other data

In our analysis, we also use additional sources of data to measure the local level of economic activity and of conflict intensity. Data on economic activity are extremely scarce in the Libyan context. The last official economic data were collected in 2011. After that, statistics on the Libyan economy have been largely unreliable due to the limited capacity of government services [Rahman and Di Maio (2020)]. To proxy for local economic activity, we thus use geo-localized information on nightlights. Nightlights data are often used as a reliable measure of economic activity when data are missing or badly measured, as in our case. In our analysis, we built our measure of local-level economic activity using nightlight data from the Visible Infrared Imaging Radiometer Suite (VIIRS). Data on conflict events are from the PRIO/Uppsala Armed Conflict and Location Event (ACLED) dataset. ACLED covers conflict events worldwide providing geo-localization, date, and characteristics of the event. Event records are derived from various sources, including reports from war zones, humanitarian agencies, and research publications [Raleigh et al. (2010)]. Using the ACLED dataset, we geo-localized all conflict-related events occurred in Libya during the period 2018–2021 and use them to build a measure of conflict intensity at the local level.⁸

3.2 Sample and descriptive statistics

Our sample includes 2,257 respondents. Descriptive statistics are presented in Table 1. For each variable, we report mean and standard deviation for the full sample and for the sub-sample of IDPs and a t-test of the difference in characteristics between the two samples. The description of each variable construction can be found in Appendix Table A.1.

Table 1. Descriptive statistics

Source: 2021 Libya High-Frequency Phone Survey Social Protection (HFS-SP).

In our sample, women account for 30.6% of the sample, the average age is around 40 years, 65.5% of the interviewees identify themselves as head of household and 42.7% report having completed higher education. Households have an average size of 5.7 members—of which almost a third are children aged 5 years or less and 5% are adults over 60—and 16% of them live in rented accommodation. The average household monthly income is 1352.54 dinars (301.90 USD), 54.9% of the households have experienced lack of food over the last 12 months, 28.5% incurred new debt in the last 3 months, 4% have received transfers from the government, NGOs, or UN agencies.⁹ As far as health outcomes are concerned, 9.5% of respondents report that someone in the household has suffered from COVID-19 over the last 12 months, 16.8% mention a chronic disease affecting themselves or other household members during the same period, while smaller shares report other infectious diseases (2.7%) and mental health issues (1.7%).

The share of IDPs in our sample is 7%, and slightly more than half of them (56%) have been displaced for more than 5 years. IDPs do not differ compared to the overall population in our sample in terms of individual characteristics such as gender, age, and education.¹⁰ They also have a very similar household size and composition to the overall population, apart from having a significantly larger fraction of members under 5 years of age. Relative to the general population, IDPs’ socioeconomic status is weak: internally displaced individuals report a significantly lower income, a substantially larger share of households experiencing lack of food and borrowing money (69.1% and 41.5%, respectively) and a five times larger probability (15.7%) of having received income support.¹¹ IDPs are also relatively more likely to suffer from a chronic disease (25.1%), while they do not report a significantly higher incidence of COVID-19 (10%), nor of other infectious diseases (5%) or mental issues (3%).

The sharpest differences between IDPs and the rest of the population emerge when respondents are asked about the economic and health impacts of the pandemic. While 24.7% of respondents say that they suffered a negative economic impact from COVID-19 and 14.5% mention a negative health impact, these shares increase to 40.2% and 23.3% among IDPs, respectively, with both differences being strongly statistically significant. This perception of a largely more negative effect of the pandemic on the IDPs seems at odd with the fact they do not report a higher incidence of COVID-19 infections. Also, it cannot be explained by the local conditions of the area where the respondents live. As the last two rows of Table 1 show, localities hosting IDPs are not different in terms of the level of economic activity and are significantly less exposed to conflict. On average, the IDPs in our sample have been exposed to two episodes of conflict in the last 12 month compared to almost three events for the full sample, a finding which confirms the fact that IDPs effectively moved away from locations where conflict intensity is higher.

4. Regression analysis

We estimate the following cross-sectional equation to compare COVID-19-related outcomes of displaced and non-displaced individuals:

(1)$$Y_{ibm} = \alpha + \beta \ Displaced_{i} + \delta \ {\bf X}_{i} + \mu \ Nightlight_{b} + \gamma \ Conflict \ Events_{b} + \theta_{m} + \varepsilon_{i}$$

where Y _ibm is the individual-level outcome of interest (e.g., having contracted COVID-19, reported impact of COVID-19 on economic and health conditions, etc.) for an individual i, living in Baladiya (province) b, located in Mantika (region) m. Displaced _i is a dummy variable which takes value 1 if the individual is an internally displaced person and 0 otherwise.¹² The main coefficient of interest β identifies the difference in average outcomes between displaced and non-displaced respondents. The equation further includes a vector X_i of individual characteristics of the respondent (gender, age, age-squared, whether she is the household head or not, whether she has a higher education or not) and of the household (number of members, share of children under 5 years, share of adults over 60 years, and whether the house is rented or not). The variables Nightlight _b and Conflict Events _b are constructed at the Baladiya level to capture differences across locations in economic conditions and conflict intensity, respectively. Nightlight _b proxies for the economic activity and is computed as the average intensity of nightlights over the last 12 months before the interview in the 20 km radius of the centroid of the Baladiya of residence of the respondent. Similarly, Conflict Events _b proxies for the level of conflict intensity and counts the number of conflict-related events for the same radius and the same span of time of the economic activity measure. Finally, we include Mantika (region) fixed effects (θ _m) to capture any time-invariant unobservable regional difference that may determine variation in the dependent variable as well as in the probability of observing displaced individuals in the sample. $\varepsilon _{i}$ is the error term. For all our estimates, we report heteroskedasticity-consistent standard errors (HC3).

5. Results

5.1 Incidence of COVID-19

We begin our analysis by testing whether IDPs are more likely to experience COVID-19. Table 2 reports the coefficients from estimating equation (1) with a linear probability model for a binary outcome which takes value one if a member of the household has been infected by COVID-19 (columns 1–4). Column 1 shows the results from estimating the baseline specification that conditions only on Mantika of residence fixed effects, the specification in column 2 additionally controls for individual and household characteristics, while the proxies for economic activity and conflict intensity in the Baladiya are added in column 3. Finally, Baladiya fixed effects are included in column 4, absorbing any potential local determinant of COVID-19 intensity. In all specifications, our estimates indicate that there is no statistical difference in the probability of reporting a COVID-19 case in the household between IDPs and the host population: the estimated β coefficients are fairly precisely estimated zeros. Notably, the estimates reported in column 3 suggest that the incidence of COVID-19 cases does not seem to be related to the local level of economic activity or conflict intensity. Additional results reported in the Appendix show no evidence that the pandemic has a differential effect across types of occupation and sector of employment of the individual (see Tables A.2 and A.3).

Table 2. Displacement status, COVID-19, and other diseases in the household

Note: Estimated coefficients are reported with robust standard errors (in parentheses). Standard errors are clustered by Mantika of residence using the wild cluster bootstrap-t procedure proposed by Cameron et al. (2008), number of bootstraps: 1000. *^, **^, *** indicate statistical significance at the 10%, 5%, and 1% levels. “Displaced” is a dummy variable that identifies IDPs in our sample. The variables “Nightlights per km²” and “Number of conflict events in Baladiya” register respectively the average value of nightlights and the number of conflict events that occurred in a distance radius of 20 km from the centroid of the Baladiya of residence of the respondent in the 12 months before the interview took place, and they are expressed in logs. “Individual controls” include: respondent's gender, age, age-squared, whether is the household head or not, and whether has a higher education or not. “HH controls” include: number of members in the household of the respondent, share of children under 5 years, share of adults over 60 years, and whether the house is rented or not.

In the remaining columns of Table 2, we explore the possible differential exposure of IDPs to: chronic diseases (columns 5 and 6), infectious diseases (other than COVID-19; columns 7 and 8) and mental diseases (columns 9 and 10). Columns 5 and 6 show that IDPs are significantly more likely to report an infectious diseases (ca. 10 percentage points, that is 60% relative to the sample mean) than comparable Libyan citizens who live in the same areas and were not subject to displacement. This finding is in line with studies from different fragile countries showing that conflict-driven displacement impacts physical and mental health [Thomas and Thomas (2004), Miller and Rasmussen (2010), Daoud et al. (2012)]. It also confirms evidence on the prevalence of non-infectious chronic diseases among displaced populations, which is caused by the exacerbation of existing diseases due to malnutrition and lack of adequate health care [Spiegel et al. (2010)]. Finally, columns 7–10 of Table 2 show that IDPs do not have a higher probability of reporting cases of infectious diseases (other than COVID-19) or mental health issues in the household.

5.2 The (perceived) impact of COVID-19

In the second step of our analysis, we test whether IDPs are differentially impacted by the COVID-19 pandemic. While the results presented in the previous section indicate that IDPs in Libya are not more likely to get COVID-19 relative to comparable non-displaced co-nationals, the descriptive statistics discussed in section 3.2 suggest that IDPs tend to more frequently report negative effects of the pandemic. We now investigate whether these gaps are statistically significant once we condition on the comprehensive set of controls and fixed effects described above.¹³

Table 3 displays results from estimating equation (1) with a binary outcome for reporting economic damage caused by the pandemic. According to the estimates reported in column 1 of Table 3, IDPs are 15 percentage points (pp) (60%, relative to the sample mean) more likely to report a negative economic impact. This substantial gap is barely affected when conditioning on the full set of individual and household-level controls, and on the local-level measures for economic conditions and conflict intensity (column 2). The inclusion of the latter two variables uncovers a strongly significant negative relationship whereby respondents are less likely to report a negative economic impact if they live in areas with higher economic activity (the coefficient on conflict intensity displays the expected sign but it is not significant). Remarkably, the IDPs’ propensity to report a negative economic impact from the pandemic does not seem to be related to having directly experienced COVID in the household. When we condition on having had a COVID-19 case in the household and on its interaction with the displacement status, we estimate non-significant coefficients for both these variables (column 3), while our main coefficient of interest remains virtually unaffected (in this specification, we also condition on dummies for reporting other diseases). Finally, in column 4 we include Baladiya fixed effects to capture any local factor—other than economic activity and conflict—that may explain the probability of having endured economic damage from COVID-19: the estimated coefficient on the displacement status dummy is only marginally affected by the inclusion of this set of controls and remains strongly significant.

Table 3. Displacement status and (perceived) economic impact of COVID-19

Note: Estimated coefficients are reported along with robust standard errors (in parentheses). Standard errors are clustered by Mantika of residence using the wild cluster bootstrap-t procedure proposed by Cameron et al. (2008), number of bootstraps: 1000. *^, **^, *** indicate statistical significance at the 10%, 5%, and 1% levels. “Someone in the HH had COVID 19” is a dummy variable. It takes one if the respondent answered yes to the question “Did you or anyone in your household experienced COVID 19 since March 2020?”. It takes zero otherwise. All other variables are defined as in Table 2.

A similar analysis is performed in Table 4 for respondents who have reported a negative health effect from the pandemic. Column 1 shows that IDPs are significantly more likely to have suffered a health-related negative impact of COVID-19: this gap holds in both the baseline (column 1) and the full specification (in which we control for individual, household and Baladiya controls; column 2), pointing at IDPs being 8 percentage points (53%, relative to the sample mean) more likely to report a negative health impact from the pandemic. Further, estimated coefficients reported in column 3 show that—as expected—all respondents who had a COVID-19 case in the household are 18 percentage points more likely to report a negative health impact of the pandemic, although this effect does not differentially affect the IDPs (as the not statistically significant coefficient of the interaction term shows). Still, the estimated coefficient of the displaced status remains stable, implying the existence of a health damage for the IDPs additional to that caused by direct contagion. Finally, column 4 in Table 4 tests the robustness of this estimated gap to the inclusion of fixed effects for Baladiya of residence.

Table 4. Displacement status and (perceived) health impact of COVID-19

Note: Estimated coefficients are reported along with robust standard errors (in parentheses). Standard errors are clustered by Mantika of residence using the wild cluster bootstrap-t procedure proposed by Cameron et al. (2008), number of bootstraps: 1000. *^, **^, *** indicate statistical significance at the 10%, 5%, and 1% levels. “Someone in the HH had COVID 19” is a dummy variable. It takes one if the respondent answered yes to the question “Did you or anyone in your household experienced COVID 19 since March 2020?”. It takes zero otherwise. All other variables are defined as in Table 2.

As a further step in this analysis, we explore heterogeneity among IDPs in the exposure to COVID-19 contagion and in its perceived impact. Figure 2 shows the estimated coefficients (and 95% confidence intervals) on interaction terms between the Displaced variable and gender and education dummies when the outcomes are, alternatively, the probability of having experienced COVID-19 in the household (as in Table 2) and the probability of reporting a negative economic or health impact (as in Tables 3 and 4, respectively). These additional estimates confirm that neither IDPs as a whole nor any subgroup of IDPs display a significantly higher likelihood of reporting a direct contagion from COVID-19 (black dots) than comparable non-displaced individuals. Nevertheless, we observe significant differences when we turn to the perceived impacts of the pandemic. Displaced men and women show a similarly higher probability of reporting a negative economic effect (gray dots) compared to non-displaced individuals, although a significant differential in the health impact is present only for men (light gray dots). The pattern for education, instead, clearly points at low educated IDPs having endured significantly worse economic and health consequences from the pandemic than non-displaced interviewees and highly educated IDPs.

Figure 2. Heterogeneity in incidence and (perceived) impact of COVID 19. Note: We augment the model specification in Table 2 (column 3) and Table 3 (column 3), and Table 4 (column 3) by alternatively interacting the dummy variable Displaced with the following individual characteristics: (i) gender (male/female); (ii) education (high/low education). For each of these interaction terms, we report the estimated coefficient and a 95% confidence interval for COVID-19 incidence (black dots), economic impact (gray dots), and health impact (light gray dots).

5.3 Further results and robustness checks

Returnees

So far in our analysis we have considered as IDPs those individuals who were displaced by the conflict and who are still displaced at the moment of the interview (see footnote 12). We thus group together in the non-displaced population both individuals who have not been displaced (never-displaced) and those who were forced to move away from home but eventually returned there (returnees). Since our survey data allow to identify this latter subpopulation, we can empirically investigate to which extent their outcomes resemble those of the displaced rather than non-displaced populations. In order to do so, we re-estimate our main regression equation including a dummy that identifies returnees as a separate category (we have 215 returnees in our sample). Panel A in Table 5 reports estimation results for our main outcomes of interest: the probability of having suffered from different diseases (COVID-19, chronic diseases, other infectious diseases, mental illness; columns 1–4) and the probability of reporting a negative economic or health damage from the pandemic (columns 5 and 6, respectively). Results indicate that the inclusion of the Returnee dummy does not affect the estimated coefficients on the Displaced dummy which maintain sign, size and significance of our main estimates (see Tables 2–4). At the same time, there is no significant correlation between returnee status and our outcomes of interest: all estimated coefficients for the dummy Returnee are not significantly different from zero, implying that returnees do not display systematic differences in outcomes relative to the non-displaced population. These findings suggest that the negative effects we document are associated with being currently displaced rather than having experienced displacement in the past.

Table 5. Further results: displaced, returnees, and duration of displacement status

Note: Estimated coefficients are reported along with robust standard errors. Standard errors are clustered by Mantika of residence using the wild cluster bootstrap-t procedure proposed by Cameron et al. (2008), number of bootstraps: 1000. *^, **^, *** indicate statistical significance at the 10%, 5%, and 1% levels. All other variables are defined as in Table 2.

Duration of displacement status

In panel B of Table 5 we investigate whether the duration of displacement status plays a systematic role in determining exposure to COVID-19 and its consequences. We do so by splitting the IDPs population into two subgroups, those who were displaced less than 5 years before the interview and those who have been IDPs for 5 years or more. IDPs with longer durations of displacement status may display better outcomes than recently displaced individuals because they had more time to integrate into receiving communities and develop a social network there. A longer duration of the displacement status, however, may exacerbate any adverse consequence associated to being forcedly moved. In addition, we may expect to have some negative selection in those who remain displaced for long periods: these are possibly the individuals whose displacement was caused by major losses (extreme violence, house destruction, forced seizing of the land, etc.) and who are unable to return home. In line with these ambiguous theoretical predictions, estimation results in panel B of Table 5 are fairly inconclusive. As far as health outcomes are concerned (columns 1–4), we do not find any significant difference between either group of displaced and the non-displaced population in the incidence of COVID-19 (column 1), other infectious diseases (column 3) and mental health (column 4). The gap in chronic diseases observed in our main results is instead driven by IDPs who have been for 5 years or more in this status (column 2). As discussed in section 5.1, the prevalence of chronic diseases among displaced populations has been documented in several other contexts and it is attributed to the exacerbation of existing diseases due to malnutrition and the lack of adequate health care. It is thus not surprising that this higher prevalence becomes visible among individuals who have endured the displacement status for longer. In columns 5 and 6, we look at the economic and health impact of the pandemic, respectively. For the economic impact (column 5), we estimate positive coefficients for both groups of IDPs, but the coefficient on the group with longer duration is larger and statistically significant, suggesting that it is this latter group that suffered more economic hardship during the pandemic. The pattern is somehow reverted when we look at the health impact (column 6): for both groups of IDPs we now estimate positive and statistically significant coefficients, but the effect is larger for IDPS with less than 5 years of duration of residence.

Alternative estimators

As an additional robustness check, we re-estimate our main regressions for disease incidence (Table 2) and COVID-19 impact (Tables 3 and 4) using both inverse probability weighting and propensity score matching. Estimation results obtained from both estimators have the same sign, same level of significance and very similar size to our main OLS estimates, confirming the robustness of our findings to the use of alternative estimators.¹⁴

5.4 Mechanisms

Our results indicate that, although IDPs are not more likely to get infected by COVID-19, they display a higher propensity to report negative economic and health impacts from the pandemic. In this section, we explore potential explanations for these findings.

5.4.1 Economic fragility

The pandemic has worsened the living conditions of the overall Libyan population [REACH (2021)]. Even if—as our results indicate—IDPs do not have a higher risk of COVID-19 contagion, their weaker socio-economic status compared to the host population [World Bank (2019)] may have left them more vulnerable to detrimental consequences of the pandemic [IOM (2021a)]. We explore this hypothesis by assessing differences between displaced and non-displaced respondents in income, food security, and degree of indebtedness (see Appendix Table A.1 for the definitions of these variables).

Table 6 shows that IDPs’ average income is 19–21% lower than the income of the non-displaced population, even after controlling for a large set of controls (columns 1 and 2). In column 3, we control for an indicator taking value one if the respondent is employed in the public sector. Our estimates show that public sector workers’ income is substantially higher (around 20% higher) than the rest of the population—confirming the existence of a large wage premium in the public sector in Libya [see World Bank (2015)]. Still, the disadvantage of IDPs remains virtually unaffected. In columns 4–6 of Table 6, we further explore the economic fragility of IDPs by looking at their food security over the last 12 months. Our estimates show that IDPs face higher food insecurity than non-displaced individuals. In particular, they are 13 percentage points—or 24%—more likely to report instances of lack of food than comparable non-displaced individuals living in the same areas.

Table 6. Displacement status, income, and food insecurity

Note: Estimated coefficients are reported with robust standard errors (in parentheses). Standard errors are clustered by Mantika of residence using the wild cluster bootstrap-t procedure proposed by Cameron et al. (2008), number of bootstraps: 1000. *^, **^, *** indicate statistical significance at the 10%, 5%, and 1% levels. The variable “(log) Monthly income” registers the respondent's (log) monthly income from any source. “Lack of food” is a dummy variable. It takes one if the respondent declares that “someone in the household is currently engaging in any following behaviors due to a lack of food or have engaged in any of these behaviors within the last 12 months before the interview, but he/she cannot continue to do it: Spent savings; Reduced non-food expenditures; Borrowed money; Sold productive assets; Took an additional job; Reduced Health Expenditure; Begged; Engaged in illegal work; Sold house or land.” It takes zero otherwise. “Works in the public sector” is a dummy variable. It takes one if the respondent indicated the option “Public Administration and Defence; Compulsory Social Security” when answering to the question “What is the sector you currently work in?’.’ It takes zero otherwise. All other variables are defined as in Table 2.

During the pandemic, there has been a significant rise in prices of basic goods and services [REACH (2020)]. Moreover, IOM (2021a) reports anecdotal evidence suggesting that the COVID-19-related mobility restrictions have contributed to an economic slowdown. Survey evidence from the same report indicates that, to cope with the worsening economic situation, IDPs had to resort to savings in order to buy food, with the majority of them indicating that their savings are not sufficient to sustain them for more than 3 months. In Table 7, we test this possibility by looking at the likelihood of having incurred new debt to cover the household's basic needs over the last 3 months before the interview. Column 1 indicates that this probability is 11 percentage points (39% relative to the sample mean) higher for IDPs than for comparable non-displaced respondents. This finding holds also controlling for labor income (column 2) which—as expected—is negatively correlated with the accumulation of new debt. Interestingly, column 3 shows that incurring new debt is not directly related to having someone in the household who experienced COVID-19. In all three specifications, the probability of accumulating a new debt is decreasing in the local level of economic activity (proxied by nightlights), a finding which is consistent with lower economic damage due to COVID-19 reported by individuals living in such localities.

Table 7. Displacement status and new debt accumulation

Note: Estimated coefficients are reported with robust standard errors (in parentheses). Standard errors are clustered by Mantika of residence using the wild cluster bootstrap-t procedure proposed by Cameron et al. (2008), number of bootstraps: 1000. *^, **^, *** indicate statistical significance at the 10%, 5%, and 1% levels. “Incurred in a new debt in last three months” is a dummy variable that takes one if the respondent answers yes to the question “Has your household incurred new debts in the last 3 months to cover your basic needs?,” and zero otherwise. All other variables are defined as in Table 2.

Taken together, these results point at IDPs being characterized by economic fragility and by worse economic conditions than non-displaced Libyan citizens. Since all the outcome variables discussed in this section are measured during the pandemic, we are unable to fully disentangle gaps that pre-date the COVID-19 shock from those that are potentially generated—or widened—by differential exposure to its broader impact on the Libyan economy and society. Nevertheless, the estimates reported in Table 7 strongly suggest that IDPs must have faced harsher economic consequences from the pandemic than comparable non-displaced co-nationals as they have been, ceteris paribus, substantially more likely to be recently forced to borrow money to cope with economic difficulties.¹⁵ Insofar as human capital shielded individuals from economic hardship, these findings are also consistent with the heterogeneity in the perceived impact of the pandemic that we uncover in Figure 2, whereby low-educated IDPs were far more likely to report having been economically damaged by the pandemic than IDPs with a higher level of education.

5.4.2 Health care access

Despite the fact that IDPs do not report a differential risk of COVID-19 contagion (see Table 2), IDPs are more likely to report that they experienced a negative health impact because of the pandemic (Table 4). Notably, this difference adds to the more negative health impact associated with the individual herself, or any other member of the household, having had COVID-19 (see Table 4, column 3).

One potential way to rationalize these findings is that IDPs face additional hurdles in accessing the health care system relative to the resident population and are therefore more likely to develop more serious medical complications from the same level of exposure to health hazard. As a matter of fact, limited access to health care has been documented for IDPs in general [Cantor et al. (2021)] and for those in Libya in particular [ICMPD (2020), IOM (2021a)].¹⁶ In this latter context, IDPs have suffered discrimination and stigmatization as they are perceived by the local community as potential carriers of the virus [Amnesty International (2020)].¹⁷ In the following analysis, we provide suggestive evidence supporting this mechanism as a possible explanation for our results.

A question included in the survey allows us to directly investigate the reasons that prevented IDPs to seek health care—when they needed medical assistance—relative to comparable non-displaced individuals. The questionnaire lists a set of possible reasons regarding both the demand (“too sick to travel,” “lack of money,” and “fear of COVID-19”) and the supply side (“movement restrictions,” “no access to facilities,” “access denied due to IDP status”) of health care services. Figure 3 shows the estimated coefficients for the Displaced dummy obtained when estimating our regression model (1) using an indicator variable for each of these reasons as a separate dependent variable. The only case in which the Displaced dummy is significant is when the outcome of the regression is the reason “access denied due to IDP status,” a result which suggests that IDPs are discriminated against in the access to health care. Interestingly, the other large coefficient (although not significant) is the one for the reason “lack of money,” a finding which speaks directly to the more fragile economic status of IDPs we document in the previous section.

Figure 3. Reasons for not seeking care. Note: The graph shows the estimated coefficient (and the corresponding 95% confidence interval) on the variable Displaced when we alternatively use as dependent variable in equation (1) one of the variables listed on the y-axis.

We further explore the decision to seek health care in Table 8. Column 1 shows that, among those reporting to have at least one disease (796 individuals, 35% of our main estimating sample), internally displaced individuals are not less likely to seek care. Column 2 shows that this finding continues to hold when we control for the full set of individual and household characteristics and also for the type of disease the individual suffers from [COVID-19, chronic disease, infectious disease (non-COVID-19), and mental health]. We dig more into the determinants of the decision to seek care by the IDPs by considering how that choice may be influenced by the fear of discrimination that we discussed in the previous paragraph. To account for this possibility, we add to our regression specification an interaction between the internally displaced status and an indicator variable capturing the degree of discrimination against IDPs which characterizes the Mantika of residence of the individual. We construct this measure as the number of interviewed IDPs (other than the respondent) who have not received assistance in the Mantika because of their status, over the number of interviewed IDPs in the same Mantika. As shown in column 3 of Table 8, the coefficient of this interaction term is large, negative, and highly significant. We interpret this result as indicating that IDPs are significantly less likely to seek health care if they live in locations in which discrimination against them is stronger.¹⁸

Table 8. Displacement status, health care, and discrimination

Note: Estimated coefficients are reported with robust standard errors (in parentheses). Standard errors are clustered by Mantika of residence using the wild cluster bootstrap-t procedure proposed by Cameron et al. (2008), number of bootstraps: 1000. *^, **^, *** indicate statistical significance at the 10%, 5%, and 1% levels. “Seek care” is a dummy variable. It takes one if the respondent answers yes to the question “Have you or any adult member of your household needed medical treatment for any disease since March 2020?.” It takes zero otherwise. “% I% IDPs in the Mantika who were denied care for their status” is equal to the number of interviewed IDPs (other than the respondent) who did not received assistance in the Mantika because of their IDP status, over the number of interviewed IDPs in the same Mantika. “Disease Type” registers whether someone in the household of the respondent reports to suffer from COVID-19, a chronic disease, an infectious disease (non-COVID), or a mental disease. The number of observations drops in column (3) because IDPs are absent from three Mantikas. Results are qualitatively unchanged by using the same sample of individuals across all three columns. All other variables are defined as in Table 2.

Taken together, the evidence presented in this section may explain why IDPs more frequently report negative health impacts from the COVID-19 pandemic relative to the resident population. Indeed, our findings indicate that discrimination against IDPs in access to health care—together with the fact that they suffer from severe financial constraints—reduces their ability to receive health assistance when they need it, increasing their chances of developing lasting health damages.¹⁹ These results speak directly to the heterogeneity of the pandemic health impact for IDPs that we document in Figure 2. There we show that the probability of reporting a health damage is particularly high among male and low-educated IDPs, which are two categories of forcedly displaced migrants that are likely to be subject to more discrimination from hosting communities.

6. Concluding remarks

The spread of the COVID-19 pandemic has severely affected the lives of households in low and lower-middle income countries. A global effort is needed to better document the possibly broad consequences of the COVID-19 shock on individuals living in these countries. Understanding the economic and social impacts of the pandemic in LIMCs has been generally difficult due to a lack of data. This task has been even more complex in the context of countries that are experiencing a conflict and for populations that are inherently hard to capture in survey samples, such as IDPs. Yet, learning about the experience of fragile and conflict-affected countries and their most vulnerable populations is essential to formulate policy interventions that are tailored to the specific needs and challenges of those more likely to suffer from the pandemic.

Our paper contributes to this effort by presenting the first assessment of the COVID-19 impact on the Libyan population and, in particular, on those internally displaced by the conflict. We use novel and unique data from a phone survey conducted in Libya in 2021 to document the differences between displaced and non-displaced individuals facing the COVID-19 pandemic. Our empirical analysis shows that displaced individuals do not report significantly higher incidence of COVID-19 relative to comparable non-displaced individuals, but are largely more exposed than non-displaced respondents to economic and health impacts caused by the pandemic. Our results indicate that the larger damage suffered by IDPs cannot be explained by individual and household characteristics, nor by a higher probability of contagion, but rather by their weaker economic status and the discrimination they face in accessing health care. The finding that in a fragile country like Libya large differences in the level of hardship endured by IDPs during the pandemic are not related to their actual COVID-19 contagion suggests that policy interventions in similar settings may need to focus more on preventing damage (e.g., through vaccination campaigns and income support schemes) rather than on containing the spread of the disease among marginalized population groups.