Segregation and Sentiment: Estimating Refugee Segregation and Its Effects Using Digital Trace Data

Abstract

In light of the ongoing events of the Syrian Civil War, many governments have shifted the focus of their hospitality efforts from providing temporary shelter to sustaining this new long-term population. In Turkey, a heightened focus has been placed on the encouragement of integration of Syrian refugees into Turkish culture, through the dismantling of Syrian refugee-only schools in Turkey and attempts to grant refugees permanent citizenship, among other strategies. Most of the existing literature on the integration and assimilation of Syrian refugees in Turkey has taken the form of surveys assessing the degree to which Syrian refugees feel they are part of Turkish culture and the way Turkish natives view the refugee population. Our analysis leverages call detail record data, made available by the Data for Refugees (D4R) Challenge, to assess how communication and segregation vary between Turkish natives and Syrian refugees over time and space. In addition, we test how communication and segregation vary with measures of hostility from Turkish natives using data from the social media platform Twitter. We find that measures of segregation vary significantly over time and space. We also find that measures of intergroup communication positively correlate with measures of public sentiment toward refugees. Attempts to address the concerns of Turkish natives in order to minimize the traction of online hate movements may help to improve the integration process.

14.8 Appendix

Twitter Collection Keywords

Tweets were collected from the Twitter Archives for the period between January 1, 2017 and December 3, 2017. Any tweets that contained the following words which pertain to Syrian refugees were included in our analysis.

Suriye	mültecileri	Suriye Makedonya
Suriyeli	göç dalgasi	şişme bot göçmen
suriyeli	Suriye Yunanistan	sahil güvenlik göçmen
mülteci	Suriye Macaristan	düzensiz göçmen
mülteciler	Yunanistan’a göç	göçmen iadesi
mültecilere	Yunanistan göçmen	ÜlkemdeSuriyeliİstemiyorum

Covariate Abbreviations

Covariate	Description
sentiment	Weekly sentiment score derived from tweets about Syrian refugees in Turkey.
lrPop	Natural log population of district derived from 2014 census.
urban	The percentage of man-made land coverage from CORINE Land Coverage Database.
metroTRUE	Dummy variable where True indicates a district is in one of the top 5 urban provinces.
borderTRUE	Dummy variable indicating whether a district is in a province that borders Syria.
diss	Activity space dissimilarity at the district level calculated from a single week of data.

Model Specifications

$$\begin{aligned}&{{} \mathbf Model 1 }\\ \hat{p}_{ij} = \text {logit}(&\beta _0 + \beta _1~\text {sentiment} + \zeta _i)\\ \\&{{} \mathbf Model 2 }\end{aligned}$$

$$ \hat{p}_{ij} = \text {logit}(\beta _0 + \beta _1~\text {sentiment} + \beta _2~\text {lrpop} + \beta _3~\text {urban} + \zeta _i) $$

$$\begin{aligned}&\mathbf{Model 3 }\\ \hat{p}_{ij} = \text {logit}(\beta _0 + \beta _1~\text {sentiment} + \beta _2~&\text {lrpop} + \beta _3~\text {metroTRUE} + \beta _4~\text {borderTRUE} + \zeta _i)\\ \\&\mathbf{Model 4 } \end{aligned}$$

$$ \hat{p}_{ij} = \text {logit}(\beta _0 + \beta _1~\text {sentiment} + \beta _2~\text {lrpop} + \beta _3~\text {urban} + \beta _4~\text {diss} + \zeta _i) $$

Model Results Table

Model	Covariate	Estimate	Std. Error	Pr( $|$z|)
Model 1	sentiment	0.07	0.03	<0.05*
Model 2	sentiment	0.07	0.03	<0.05*
Model 2	lrPop	−0.02	0.01	0.11
Model 2	urban	−0.16	0.06	<0.05*
Model 3	sentiment	0.07	0.03	<0.05*
Model 3	lrPop	−0.11	0.01	<0.05*
Model 3	metroTRUE	0.23	0.03	<0.05*
Model 3	borderTRUE	−0.01	0.02	0.64
Model 4	sentiment	0.06	0.03	<0.05*
Model 4	lrPop	0.05	0.01	<0.05*
Model 4	diss	1.79	0.08	<0.05*
Model 4	urban	−0.30	0.06	<0.05*

1. *p< .05 level