Forced Migrants as a Percentage of World Population since 1950

       *Data scraped from Google Scholar
       

• Wars are the primary engine of refugee flight
• But not all wars of comparable devastation produce comparable levels of refugees
• The civil war that has plagued Iraq has resulted in an average annual cross-border exodus exceeding 300,000 people with 250,000 displaced internally each year
• Pakistan’s civil war against the Taliban insurgency resulted in an average exodus of only 2,200 people across the border per annum
• But the number of civilians internally displaced at the height of Pakistan's civil war (2009) was well above 1.5 million!

So why do some civil war torn countries produce more refugees relative to their internally displaced population and others displace more of their population internally than across borders?

Does the literature on conflict and forced migration provide any answers?

Surprisingly, research on IDPs as an analytically distinct group has been woefully underexplored in recent years.

Surprisingly, research on IDPs as an analytically distinct group has been woefully underexplored in recent years.

• The only work to quantitatively compare refugee flight and IDP movement to one another is Moore and Shellman (2006)
• In their model, victims of displacement weigh the dangers they perceive at home against those they see in their potential points of destination. When conditions in their points of destination are more favorable than the conditions they face at home, they are more likely to flee across borders
• Their analysis suggests levels of violence in neighboring states increases the proportion of refugees fleeing within borders than across borders because violence in cross-border states renders the choice of internal relocation more attractive

• Moore and Shellman (2006) have identified an exogenous variable that influences where fleeing civilians choise to escape to
• But what factor(s) endogenous to the conflict itself influence that choice?
• The aim of this article is to address this very issue

Central Claim: Civil wars fought along ethnic lines are morely likely to produce a greater IDP to refugee ratio than non-ethnic civil wars

Causal Mechanism: Ethnic markers provide combatants with more effective means of targeting populations loyal to rivals; renders choice to relocate across borders more appealing

1. Rebels and governments want to limit the resources of their rivals and one way to do so is to target populations loyal to them (i.e. population control is an important aim of combatants)

2. Rebels and governments have an incentive to safeguard loyal populations and overzealous targeting of civilian groups that puts their own potential supporters at risk is costly (i.e. indiscriminate oppression is irrational)

3. Rebels and governmnts have an incentive to keep populations loyal to them within their territory of control (i.e. human beings are resources for warfare)

4. Civilians react to the indiscriminate targeting of other civilians by fleeing to regions (i.e. they update their prior beliefs)

5. Civilians prefer to relocate as close to their original location of residency as possible (i.e. prefer internal to external displacement)

Imagine for a moment a head of a household caught in the crossfire between rebel and government forces. She face two decisions:

1. Stay put with her family or to relocate to a safer destination
2. If relocating, whether to relocate her family to a more peaceful region of their country or to flee across international borders

1st choice is clear but 2nd choice depends on the perceived likelihood that the violence will follow her and her family to their choice of destination

That perceived likelihood is influence by patterns of violence against other civilians

Now imagine for a moment a combatant leader in the midst of war. She faces two decisions.

1. Is it beneficial to target civilians from rival populations?
2. If yes, is the fleeing family from a loyal or rival population?

1st choice is clear but 2nd choice depends on the perceived likelihood that she can distinguish between loyal and rival groups.

When the cleavages of war fall along ethnic lines, that perceived likelihood increases!

1. Ethnic civil wars increase civilian deaths by allowing for more discriminate targeting by combatants

2. In turn, civilians learn from the patterns of discriminate violence and have a choice to either relocate internally or externally. If the civilian views the targeting of other civilians as incidental to the conflict, the civilian will flee to regions inside their own country but outside the war zone.

3. If however, civilians view the targting of other civilians as deliberate acts of discriminate violence, then they will be more inclined to not only flee the war zone but also any regions that rival warring combatants can target them -hence movement across international borders.

\(H_{1}\): One-sided violence against civilians increases the probability of forced migration among civil war torn states.

\(H_{2}\): One-sided violence has no independent marginal effect on the proportion of forced migrants that are refugees.

\(H_{3}\): Compared to nonethnic civil wars, civil wars characterized by ethnic cleavages produce more refugees as a share of total forced migrants.

\(H_{4}\): The positive effect of ethnic civil wars on the share of forced migrants that are refugees is conditional on the presence of one-sided violence.

\(H_{5}\): One-sided violence against civilians during non-ethnic civil wars result in a smaller share of forced migrants that are refugees relative to one-sided violence against civilians during ethnic civil wars

Unit of analysis: Country-year (panel data)

Population: All civil war torn countries experiencing forced migration across all available years

Sample: 361 civil war-torn country-years between 1993-2011

Estimation Technique: two-stage heckit selection model (Probit;OLS)

\[ \begin{equation} \label{eq:star0} P(D_{it} = 1 | Z_{it}) = \phi(Z_{it} \gamma_{it}) \end{equation} \]

\[ \begin{equation} \label{eq:star} f_{it}*= X_{it}\beta + u_{it} \end{equation} \] \[ \begin{equation} \label{eq:star3} E[ f_{it} | X_{it}, D_{it} = 1] = X_{it} \beta + E[u_{it} | X_{it}, D_{it} = 1] \end{equation} \]

\[ \begin{equation} \label{eq:star2} E[ f_{it} | X_{it}, D_{it} = 1] = X_{it} \beta + \rho \sigma_{u} \lambda (Z_{it} \gamma) \end{equation} \]

Due to non-random assignment of the "treatment variable" (i.e. endogeneity).

• There exists at least some benefit for leaders to mobilize along ethnic lines.
• If leaders can choose what type of war to wage then the effect of civil war type on forced migration patterns will be difficult to untangle. Bias? No!

Due to truncated data at two stages

• Stage #1: This study is interested in the pressures that civilians face in the midst of civil war. Data subsetted to only include observations recorded with an ongoing civil war. Bias? No!
• Stage #2: Not all civil war-torn states experience forced migration. Data subsetted to only include observations w/ forced migration. Bias? Yes!
  • Solution: Two-stage Heckman Correction

\[P(D_{it} = 1 | Z_{it}) = \phi(Z_{it} \gamma_{it})\] 1st Stage: Probit model identifying factors that predict binary forced migration in any given country-year observation

\(Z_{it}\) refers to the list of explanatory {Civil War Intensity (more than 1000 vs 25-1000), lag of Forced Migration, Interstate War (binary), Conflict Duration (no of years), Intercommunal Violence (binary), Natural Disasters (count), Civilian Targeting (binary)}

\(D_{it}\) Forced migration in country \(i\) at time \(t\) {1=Forced Migration; 0=No Forced Migration, mean: 0.64}

\[f_{it}*= X_{it}\beta + u_{it}\] 2nd Stage: OLS regression using Inverse Mill's ratio, generated from the probit model, as an additional regressor.

\(X_{it}\) refers to the list of explanatory variables {Ethnic War, Civilian Target X Ethnic War, Population Density, Island, neighboring Civil War, neighboring Intercommunal Violence, neighboring Civilian Targeting, Borders}

\(f_{it}\) ratio of refugee flows to total migrant flows in country \(i\) in time \(t\) {0:1.00, mean: 0.49}

\(f_{it} = \frac{R_{it}}{R_{it}+I_{it}}\)

• Categorization of ethnic vs non-Ethnic civil wars and intensity of war: Categorically Disaggregated Civil Wars (CDC) (Henrikas 2016)

• Forced Migration (IDP and Refugges): UNHCR’s (2016) database of persons of concern

• Intercommuncal Violence and Civilian Targeting: Uppsala Conflict Data Program (UCDP 2016)

• Natural Disastrers: The International Emergency Disasters Database (EMDAT 2014)

*The marginal effect of one-sided violence on the composition of forced migrants during ethnic civil wars is negative and statistically significant (not shown)

1. Civilian targeting (i.e. one-sided violence) increases the probability of forced migration during civil war

2. The only other predictor of forced migration them model found evidence for is past experience of forced migration

3. Civilian targeting has no independent effect on the share of force migrants who are refugees.

4. Ethnic civil wars result in a greater share of force migrants who are refugees

5. Compared to non-ethnic wars, ethnic civil wars characterized by one-sided violence result in a greater share of forced migrants fleeing across borders

6. Geography (island) and history (time lag of ratio of refugees:forced migrants) are the only other predictors of refugee composition that the model found evidence for