Who’s the alien?

Xenophobia in post-apartheid South Africa

Marguerite Duponchel∗†

May 2009, draft

Abstract

In May 2008, South Africa became the theatre of widespread violent attacks on migrants. 62
were killed, hundreds wounded and tens of thousands displaced. Fourteen years after the official
end of apartheid and about two years before the long-awaited FIFA world cup, Mandela’s dream
of a rainbow nation finished to vanish.
Xenophobia can certainly be traced to the dawn of human kind. As the Western world observes
a nationalist radicalisation, South Africa is rather the norm than the exception. Yet in a country
which has justly become a model of reconciliation, it shocked. Xenophobia has long been present
in democratic South Africa, but hatred seems to have intensified recently. In a country eroded
by huge inequalities and tremendous levels of crime, the outcome has the potential to turn into
a humanitarian catastrophe. And it did in May 2008. Yet one year following the violent riots,
what has been done? A lot has been said, but there is still no consensus on the root causes of the
violence. It is way past time to identify and tackle them with an iron hand to avoid a new May
2008.
The originality of the paper lies in the fact that it uses a quantitative method to explore this long
disregarded question. In February 2009, the South African Red Cross (SARCS), in partnership
with Wits University and United Nations Office for the Coordination of Humanitarian Affairs
(UNOCHA) conducted a household survey in the Johannesburg inner city and in Alexandra town-
ship. It covers both South African citizens and foreigners who were the primary, but not only,
target. Questions relating to violence and displacements were incorporated to a larger survey on
living conditions. Using a victimisation measure, I thus investigate the characteristics of those
undesirable ”outsiders”. The results show that poorer individuals are most at risk.
Results suggest that poorer people face a higher probability of being victimised on the ground of
xenophobia. In the inner city especially, the incidence of poverty of the risk of being attacked is
very high. Moreover the results contradict the hypothesis that the outbreak of violence was due
to the influx of foreign nationals fleeing Zimbabwe. The longer one has spent in South Africa, (or
Gauteng, for the internal migrants) the higher the risk. The effect is non-linear with a maximum
reached around a decade. Zimbabweans do also face a lower chance of victimisation than other
foreign nationals in particular than Mozambicans and Congolese. Furthermore, economic migrants
appear more at risk in Alexandra.
Finally, running a logistic model on the violence variable that depends on the period of analysis,
the results show a change in trends since the May 2008 attacks. The incidence of violence is higher
and foreign nationals are relatively more at risk today than before the outbreak.

∗ University Paris 1 Pantheon-Sorbonne, CES and CNRS, and Wits University, Forced Migration Studies Programme
† Many thanks to UNDP South Africa, to UNOCHA-ROSA, to the South African Red Cross and to the Forced
Migration Studies Programme at Wits University. I also want to thank Lopa Barjenee, Laura Brown, Veronique Gindrey,
Pete Manfield, JP Misago, Loren Landau, Pinky Vilakazi, Hein Zeelie and all the volunteers of the Red Cross.

1
1 Introduction
Some call it xenophobia, literally in ancient Greek ”fear of the other”, others call it simply racism.
On May 11 2008, Alexandra township witnessed the start of pogroms against outsiders who some
perceived were not South African enough. The extreme violence quickly spread to other provinces
and eventually lasted until the end of that month. Those events are remembered as the May 2008
xenophobic riots. Hundreds of homes and shops were looted and burnt. 62 people died across the
country, among which 21 were South African citizens. Many were injured. An estimated 100,000 fled
their homes and found refuge wherever they could.

My colleague at UNDP (United Nations Development Programme) recalled her story. When the
xenophobic violence erupted in May 2008 and started to spread around the country, she got worried
that Mike, her Mozambican gardener, might be in danger. Mike, who previously stayed in Pretoria,
had moved to Thembisa three years before to live with his South African girlfriend. As the violence
quickly reached the township1 , she phoned to make sure that he was alright. Mike confirmed that
there had been violent attacks against migrants around the neighbourhood. When he was asked if he
was safe his answer was: ”Yes, they would not attack me”. Mike’s story raised questions. Why, in
the same communities, some foreigners were attacked and some were not? What made some residents
South African enough to the eyes of the community when some became aliens that needed to be chased
away at any cost? What made someone a likely target? The violence might not have been as randomly
conducted as it seemed after all. These questions were completely ignored at the start of this research.
For good reasons: no data existed. Nevertheless, finding answers is essential for the design of policies
and programmes targeting both xenophobia and reintegration of the victims. How can one effectively
intervene if one does not know who is in need, who is vulnerable? Luckily, the South African Red
Cross was just starting to design a survey to identify pockets of vulnerability when I arrived. Some
minor additions to the questionnaire enabled us to capture experiences of violence at the household
level, and eventually led to some answers.
While I believe that quantitative analysis has the potential to reveal important features and draw new
lines of research, qualitative research should not been ignored. Xenophobia is a multi dimensional
phenomenon and pluri-disciplinary is analysis undoubtedly necessary for a wider understanding of
xenophobia in South Africa, its roots, its consequences and the way forward.

Part 2 recounts the xenophobic attacks of May 2008, the origins and aftermath and presents a short
literature review of the subsequent analysis on the causes of the events. Part 3 describes the database
1 Thembisa is located in the Gauteng province. It is one of the townships the most affected during the riots

2
and the econometric model. Part 4 reports the results and finally Part 5 concludes.

2 Xenophobia in South Africa: facts and analysis

South Africa, and more particularly the Gauteng province hosting Johannesburg and Pretoria, has
certainly become today’s economic centre of Sub Saharan Africa. Its success has attracted a wide range
of migrants from all over the continent looking for economic opportunities or refuge from oppression
and a better life. Yet the number of immigrants is unclear, a significant number of them crossing the
border illegally. May 11 2008 will surely be remembered by all of them. Violent attacks on people
considered as outsiders started in Alexandra township, on the outskirts of Johannesburg. The levels
of violence observed during this month of pogroms, where some were literally burnt alive, definitely
shattered Mandela’s dream of a rainbow nation.

2.1 Facts and figures

The events of May 2008 cannot be analysed without understanding the global post-apartheid context
of xenophobia. The xenophobic riots did not emerge from scratch. Xenophobic violence can be traced
to as early as a few months after the first democratic elections on April 27 1994. In December 1994,
in Alexandra, ”armed youth gangs destroy foreign-owned property and demand that foreigners be
removed from the area” (Misago et al., 2009). Evictions, looting of property and even killings are
reported elsewhere in the country. The frequency gradually increases, eventually leading to the May
2008 outbreak, when an unprecedented level of violence is reached2 .
Between January 2008 and the May 2008 riots, at least thirteen foreigners are killed, a large number
of shops are burnt down, many are made homeless. Violence is reported in the Eastern Cape, in the
Western Cape, in Gauteng, in Kwazulu-Natal and in the Free State.

On May 11, an armed mob starts attacking foreigners in Alexandra, Gauteng. Two people are
killed and two women are raped. Sixty people are injured. Shops and homes are looted and self-
appropriated by the perpetrators. Violence in Alex will continue for several days. Hundreds of people
flee their homes and find refuge at the police station. Meanwhile, the violence spreads across the
province, reaching several townships including Diepsloot, Thembisa and Soweto. By May 16, attacks
reach Cape Town, where one Somali shopkeeper is killed. On May 17, violence grips Jeppestown in
downtown Johannesburg and the following day, inner city Hillbrow is affected. Violence and evictions
continue in most places. Durban is shaken on May 17. Attacks are reported in Maukasi, Dukhatole,
2 See Misago et al. (2009) for details on the violent incidents recorded since the end of the apartheid era and during

May/June 2008

3
Zandspruit, Ramaphosa, Primrose, Reiger Park, Kya Sands, Jerusalem, Zamimplio, Joe Slovo informal
settlement, Mayfair and other places. On May 21 clashes are again reported in Durban, Mpumalanga,
the Free State and the North West province. On the same day, President Mbeki decides to involve
the army to stop the violence. Violence in Gauteng gradually scales down, but incidents are still
reported in the other provinces. Meanwhile, thousands march against xenophobia in different parts of
the country3 .
By May 26, the violence is declared under control: 1384 suspects have been arrested, 342 shops looted
and 213 burnt down. Hundreds are wounded. 62 have lost their lives, 21 of them are South Africans.
Thousands are made homeless, forced to seek shelter at friends, in the government administered camps,
in churches, etc. The last shelters will close in October, five months after the riots started.

Some victims are reintegrated into their communities with insurance from the residents that they
will be safe. In some neighbourhoods, returning would literally have meant committing suicide. Those
victims either resettled in other places where their lives were not threatened or simply decided to return
to their countries of origin. For some, going home was just not an option. However, those movements
post-crisis were not adequately monitored. One year down the road, it has become hard to locate the
victims of the attacks. One thing that is certain is that foreign nationals living in South Africa stay as
quiet and invisible as possible, fearing more violence. As the data collection shows, outsiders are still
harassed. Xenophobic violent acts did not cease on May 26 2008. In September 2008, a Somali mother
was murdered with her three children in her shop in Queenstown, Eastern Cape. In Alexandra, shacks
inhabited by foreigners were burnt down around Christmas. In May 2009, two Somalis were found
burnt in their shops. These are only a few examples.

2.2 Literature review on xenophobia and violence

In the aftermath of the violence, many reasons were given to try to explain the outbreak of violence
- ranging from poverty to ”they steal our women”, from the implication of a ”Third Force” to the
recent influx of Zimbabweans fleeing the political and economic crisis in their home country, from
unfair labour competition to crime. Researchers, notably in social sciences, quickly attempted to find
explanations for the riots. Yet quantitative approaches are up to now absent. Literature on xenophobia
in other countries is also strikingly scarce.

The Human Sciences Research Council (HSRC) based in Pretoria, together with the High Com-
mission of the United Kingdom, produced a report trying to access the causes of violence and make
3 Details on the events are questionable, those facts are gathered from different sources including BBC News, Mail &

Guardian and UNOCHA

4
recommendations on the way forward (”Violence and Xenophobia in South Africa: Developing Con-
sensus, Moving to Action”, 2008). As it points out ”Anti-migrant sentiments are widespread in many
countries, including in Europe and North America, and elsewhere in Africa, but do not all necessarily
end in outright violence on varying scales”. The HSRC report underlines three causes: relative de-
privation; South African exceptionalism coming from a very particular relationship with the African
continent and the West; exclusive citizenship.

Misago et al. (2009) in a study commissioned by the International Organisation for Migration
(IOM) took a very interesting approach. During five months of field work, they interviewed residents,
foreign nationals, government officials, community and civil society representatives. They chose seven
sites: five where xenophobic violence was reported, and two where the presence of foreigners did not
lead to attacks. Their idea was to compare the two. Why in some communities were outsiders targeted
and not in others? They identified four factors that triggered the violence: 1/ the presence of ”insti-
tutionalised practices that exclude foreigners from political participation and justice”; 2/ the absence
of conflict resolution mechanisms; 3/ the presence of ”political vacuums or competition in community
leadership that encourages the emergence of unofficial, illegitimate and often violent forms of local
leadership”; and 4/ a culture of impunity, particularly with regard to xenophobic violence.
Moreover, they found that inadequate border control and the supposedly mass influx of Zimbabweans
were not a valid explanation for the xenophobic riots. Indeed, the study showed that outsiders who
had been attacked had been living there for years. Furthermore, no mass influx was reported from
respondents. Finally, the places where the violence occurred were not necessarily the ones with the
highest proportion of foreign nationals. They also conclude that the rise of food and commodity prices
was not a plausible explanation as foreigners managed to sell at lower prices, in smaller quantities and
even provide credit to their customers. Regarding the involvement of a ”Third Force”, the research
concludes that there was ”no evidence these attacks were planned or orchestrated by a single organi-
sation or individual”. However, media played a key role in transferring the information from one site
to another. Some cooperation between perpetrators was also noticed in some areas.
The report also identifies necessary but not sufficient pre-conditions for the violence, such as high
unemployment; poor service delivery; impunity; and limited knowledge of immigration laws.

To my knowledge, no research focuses on the individual level to try to assess what particular
characteristics affect the likelihood of becoming a target. And there is certainly no quantitative
analysis as no data existed.

5
3 Data and econometric model

3.1 Data

The data was collected during February 2009. For the pilot phase, two places where chosen in the
Gauteng province where the violence started: Alexandra township and the Johannesburg inner city.
Over twenty volunteers from the South African Red Cross were trained on the questionnaire designed
with Wits University Forced Migration Studies Programme, and then sent to the field where coordi-
nation was insured by UNOCHA and Wits.
But before going into the data description, it is essential to give some historical background about
these two surveyed sites. The apartheid era has undoubtedly left deep scars both in the minds of South
Africans and in the geography of the territory. Not taking into account history in post-apartheid South
Africa will certainly lower the accuracy of the analysis. Because Alexandra and the Johannesburg city
centre are so particular, results, at this stage, can not be generalised to the whole country. More data
is needed and townships in Kwa Zulu-Natal, Eastern Cape and Western Cape provinces, in particular,
need ideally to be surveyed.
-subsectionAlexandra is, and has always been, a special place (”Alexandra: A History”, 2008)
Reminder of the apartheid regime and also one of the strongholds of the struggle for freedom during
the war years, Alex is a striking illustration of urban poverty and inequalities in the newly reborn South
Africa. Established in 1904, the overcrowded ghetto, today’s home of nearly half a million people, is
situated in the North-East of Johannesburg. From its top hill, you can observe the 2sq km of houses
and shacks. On the horizon, you are able to distinguish the shining towers of the wealthy Sandton
City only a few blocks away, where billions of rands trade daily. Back in its early days, its location,
along with the ease for new arrivals to secure a pass to work in the city, often made it the first stop
for rural blacks seeking jobs in the semi-industrial suburbs.
But between then and now, history has left its traces. Alexandra’s history has long been tumultuous.
The township struggled to survive apartheid, going from intense threats of removals to numerous
failures of upgrading plans, mainly as a result of a clear disinterest on the side of the government.
The last in date, the Alexandra Renewal Plan (ARP) was launched by President Thabo Mbeki in
2001. It has improved the neighbourhood despite a very difficult first period of implementation. The
R1.3 billion seven-year development plan brought new infrastructures and housing, roads upgrade and
other improvements to Alexandrians’ daily lives. But the highly cosmopolitan and ethno-fractionalized
township is still gravely overcrowded. Alex, once constructed to host 30,000 black workers, has reached
a population estimated at half a million. Since the end of apartheid, the number of its residents has
nearly quintupled. As a result there is lot of pressure for already scarce service delivery. Poverty is

6
striking4 and the level of criminality remains tremendously high5 .
Meanwhile, Alexandra has always been a place whose people have been strongly politically engaged,
leading the fight for freedom from its very early days through multiple ways, such as bus-boycotts and
demonstrations. The residents played a central role in the fight against the Bantu Education Act in
1976 alongside the Soweto students. Alex has always been on the front-line of social movements. It is
by no mean a surprise to see it at the epicentre of the xenophobia events that broke out on May 11
2008.

3.1.1 Johannesburg inner-city

When arriving in Johannesburg, one can distinguish the city centre from a distance. Wherever you
are coming from, by land or by air, you can see the famous television tower and the 53 floor high
Ponte Tower from far. At the time of its construction the Ponte Tower was the highest building of
Africa. The price of a square meter was unaffordable to most South Africans. By the 90’s it had
become a symbol of crime, drugs and decay. The story of this once highly coveted building is a prefect
illustration of Johannesburg inner city.
In 1955, the Native Amendment Act urged the removals of black residents from the city centre. The
inner city was declared a ”White Only area” and the government removed illegal tenants, mainly
coloured and Indian families. It was the economic and cultural centre of Johannesburg, hosting many
firms’ headquarters. Its sky-scrappers were the pride of South Africa.
Between 1978 and 1982, the grand apartheid project started to erode, and a large influx of coloureds
and Indians moving back into the neighbourhood was recorded. The main reason was the shortage of
flats in some areas and the oversupply in Hillbrow (Morris, 94). Black people also started to illegally
penetrate the forbidden area. White families who could afford it started to migrate to the wealthy
suburbs of the North.
By the end of the apartheid era, Hillbrow and Joubert Park had switched from an essential white
population from all classes to an ”overcrowded crime ridden black ghetto” (Olufemi, 1998), hosting
among its residents drug dealers, notorious gangs and prostitutes.

With the upcoming FIFA world cup, the City of Johannesburg launched in 2008 a R171 million
uplifting programme for the inner city. The area has been cleared, but remains one of Johannesburg’s
most feared areas. Its population is essentially black. A lot of migrants who were not attracted by the
township life settled in the inner city, thus creating an intensely cosmopolitan place.
4 In our sample, 87% of the respondents in Alexandra reports not having access to running water in their dwelling; 4

out of 10 households interviewed do not have electricity in their dwelling

5 SAPS reports an average of four murders per month

7
3.1.2 The data collection

The sampling strategy was designed to gather a sample with 50% of foreign nationals6 . Among the
national population, targets were set to 50% of South Africans who have lived in the locations of
interview for at least ten years (i.e. long term residents), and 50% of recent migrants who arrived in
the location less than ten years ago. Interviewers were assigned enumeration areas chosen randomly
and knocked on every fourth door. Only adults were to respond with the exception of over 16 year-olds
if they were head of household (this was mainly to avoid evicting from the sample young mothers who
could be particularly vulnerable).

In total, 2028 people were interviewed over the course of a month. Enumeration areas were chosen
randomly.
In Alexandra, we quickly switched to snowballing. Indeed, we realised the quota of foreign nationals
would be difficult to reach. Firstly, foreign nationals constitute a minority of Alexandra’s residents,
no more than 3% according to the Alex Renewal Project survey. Secondly, foreign nationals refuse to
answer or try to mask their identity, underlying the accuracy of the assumption that the situation of
foreign nationals was still difficult. Using snowballing methods we managed to successfully interview
277 foreign nationals or 28%. 385 interviewees were long term residents and 326 recent internal mi-
grants.
In the Johannesburg inner-city, we surveyed three neighbourhoods: Hillbrow, Berea and Yeoville.
There, we easily reached the 50% rate of foreign nationals. This confirmed the intuition that the
location was hosting a large proportion of foreigners. Using the ”every fourth door” sampling strategy,
548 foreign nationals were interviewed along with 470 South Africans. 160 of the latter were long
term residents (the inner city happens to be mainly a place of transit) and 310 recent migrants. Over-
all, the data is composed of 59% of South Africans, 25% of Zimbabweans, 6% of Mozambicans, 3%
of Congolese (DRC) and other less represented citizens, including Malawians, Nigerians and Zambians.

Information was collected on a wide range of issues: demographics of the household, migration
patterns, including documentation, livelihoods, access to health and education services, social partici-
pation but also on violence linked to the May attacks. We asked respondents whether they had been
threatened or attacked because of their ethnic/tribal/national identity. We focused on three periods:
before May, during May/June 2008 and after the xenophobic events. We then enquire further about
6 South Africans were to be included because some of them are in many ways vulnerable. They also have been

targeted during the xenophobic attacks. In addition, any programme design, whether it is a research exercise or not,
should include nationals to avoid exacerbating tensions in the communities. The proximity of the national elections
also created a highly volatile environment where looking for a particular group of residents could lead to conflict and
endanger both respondents and interviewers

8
potential subsequent displacements due to the violence, including the length of displacements, the
relocation site, etc. We also asked whether the respondents returned to the place from which they had
been chased away or not and for what reason.

3.2 The econometric model

3.2.1 A potential selections bias

There is undeniably a potential selection bias in our data. People who fled their homes and never
came back are completely absent from our sample. Very little is know about the victims who decided
to leave for the simple reason that there had been no adequate monitoring.
The United Nations High Commissioner for Refugees (UNHCR) assisted the displaced refugees but
very few went home. Quickly, the agency decided that resettlements would not be awarded on the
basis of xenophobic violence. Meanwhile, the United Nations Children’s Fund (UNICEF) gave grants
to support movements home. IOM also assisted vulnerable individuals who expressed the desire to
return to their country of origin. In two successive phases, IOM assisted 578 individuals who resigned
themselves to go back home, mostly originating from the Great Lakes Region7 . The Mozambican gov-
ernment provided its citizens with a bus ticket to return to Maputo. People might also have returned
home by their own means which makes it even more difficult to track movements.
What is even less sure is how many of these people who first left came back when the storm had settled?
Probably a lot of them. Numerous stories from the camps suggested that some would have used the
repatriation process to go home and then come back again. The border crossing statistics from the
Department of Home Affairs can give a vague idea on movements out and back to the country. Figure
2 shows the different trends of arrivals (of African nationals) and departures (by road)8 between 2007
and 2008. The cycles correspond to South African holidays. Apart for Zimbabweans, trends should
follow the approximate same path from one year to the other; the only difference should be in scale.
The first graphic indeed shows very few differences between the two years, apart from a decline in
both arrivals and departures in August/September 2008. Looking at the breakdown in arrivals by
nationalities, the only main difference is an increase, relatively to the preceding year, in the rate of
arrivals of Mozambicans, Namibians and Zimbabweans in July 2008. Possibly, it can be explained by
those who were returning to South Africa after the violence. This is yet just a supposition.
Out of a large number of displaced persons, we can assume that the ones who decided to leave South
7 Phase 1: 170 Burundians, 137 Congolese (DRC), 44 Tanzanians, 15 Somalis (who went back to Ethiopia), 5 Kenyans,1

Congolese, 1 Sudanese, 1 Rwandese. Such details are not available for phase 2
8 For departures Home Affairs only give a breakdown by means of transport and not by nationalities. Presumably,

most victims would have returned home by road, however those statistics also include departures of tourists travelling
in the region.

9
Africa for good constitute a clear minority. IOM, which assisted many of them, seems to believe
that the foreigners who left were the ones who had lost everything and greatly feared for their lives.
Meanwhile, in most of the interviews carried out with humanitarian staff present in the camps at the
time of the crisis revealed the same observation: when it was announced that resettlements would not
occur, people started to leave and return to their communities.
While it is important to keep in mind that we do not observe the characteristics of individuals who
left, I can assume that this will not result in a significant bias in the analysis, as so many came back
to South Africa.

Another potentially and certainly more problematic source of bias resides in the fact that Alexan-
dra was where it all started and people who could afford not to obviously did not come back. They
preferred to resettle somewhere else. The main reason mentioned for returning to Alex was ”we had
no choice”, suggesting that if they had had the means to do so, victims in Alex would have moved
somewhere else. Furthermore, Alexandra was certainly not an asylum location. In our sample, only
12 interviewees found refuge in the township. Victims from other places would hardly have relocated
there. This selection bias will eventually be corrected when the data collection is extended to other ar-
eas in the country where people came back to or found refuge once the xenophobic riots were contained.

3.2.2 Model

The dependent variable is constructed using the answer to the question: ”Have you, or anyone in
your household, ever been threatened or attacked because of your ethnic group, your tribe or your
nationality?”. The ”yes or no” answer results in a dummy variable which analysis requires the use of
a non-linear model. We opted for a probit model. We thus estimated the following equation:

V iolence∗i = α0 + β 0 Xi + εi (1)

1 if V iolence∗ > 0
where V iolence = 
0 if V iolence∗ ≤ 0

Where Xi is a vector of explanatory variables for the household i. It includes the location of
interview, the length of stay in the location and its square to capture a possible non linear effect, the
level of education of the interviewee and proxies for the level of poverty of the household. The analysis
is clustered by enumeration area.

10
3.2.3 Constructing a poverty index

Instead of introducing each available proxy of the level of poverty one by one in the regression, it is
attractive to create a ’poverty index’. It is, indeed, a lot simpler for interpretation to have one measure
of poverty instead of a range of proxies. In order to construct that indicator, I was inspired by the
Burnside & Dollar (2000)’s method to construct their indicator of good policies. Similar to weighing
the policies depending on their impact on growth, I thought it was natural to weigh the different
poverty variables according to their impact on whether a household would consider itself poorer than
average. This enables comparing poorest households to the others. Instead of estimating equation 1
directly, I first estimate the following equation using a Probit model.

P overtyi∗ = α0 +0 Pi + εi (2)

where poverty=1 if the respondent of household i describes his household as poorer than average. This
was the case for 35% of the sample and for half of the Alex households. I used clusters by enumeration
areas. All the explanatory variables are significant. The results are presented in Table 2 in annex.
From equation (2) I fixed the coefficients of the different poverty measures.
The poverty index of each household is thus defined as:
P overtyi = 0.046fi + 0.17 ∗ ti + 0.202i ∗ wai + 0.023ri + 0.083hi + 0.15ei
where f is a dummy variable taking the value of 1 if the household ate meat, chicken, fish or fresh
vegetables less than seven times in the past week; t is a dummy variable taking the value of 1 if the
shelter is not equipped with a flush or chemical toilet; wa is a dummy variable taking the value of
1 if the accommodation has no running water; r reports the number of people per sleeping room in
the shelter; h is a dummy variable taking the value of 1 if access to health has proved difficult and e
if access to education is a problem. Hence we obtain a poverty index that varies from 0 to 0.77, the
higher the value the poorer the household. The average index is 0.24 while the median value 0.18.
Interestingly what seems to matter more to households in the surveyed areas is access to water and
sanitation as well as access to school for the children.
As a result equation (1) should be rewritten as follows:

V iolence∗i = α0 + β 0 Xi + γ 0 P overtyi + εi (3)


1 if V iolence∗ > 0
where V iolence = 
0 if V iolence∗ ≤ 0

11
4 Results

4.1 Which characteristics influence the probability of being victimised?

Column 1 of Table 3 is the core regression of the analysis.

The descriptive statistics of the population which has been victimised show a striking difference
between Alexandra and Johannesburg inner city. In the township, 47% of foreign nationals in our
sample reported having been victims of violence, as opposed to only 15% in the inner city. This under-
lines the fact that Alexandra is a particularly rough environment for foreign nationals. One obvious
reason is that foreign-born only constitute a minority of the inhabitants in the township (about 2-3%
according to the estimation of the latest Alexandra Renewal Project’s study). Hence, they become an
easy target. In order to verify this assumption, a dummy for Alexandra was introduced, its coefficient
is however not significant suggesting that rather than the township in itself, it is the characteristics of
those living in the township that can explain why there was more violence in Alex. So, in practice,
what makes people aliens?

Introducing a dummy if the respondent was born outside South Africa, the result is not surprising.
Foreign nationals have a significant 25% higher probability of having been attacked. Looking at a
breakdown of nationalities (column 4), only the coefficient of the South African dummy is significant
and negative, in complete accord with the results of column 1. However the co-significance test of the
four discrete variables (for Congolese, for Mozambicans, for Zimbabweans) reveals that the coefficients
are significant together. Congolese and Mozambicans have a higher likelihood of being victimised
(around 6% for both), while Zimbabweans have a lower probability ceteris paribus. This confirms
the observation by Misago et al. (2009) that the influx of migrants fleeing Zimbabwe is not a good
explanation for the xenophobic riots. Mixed households, i.e. households where at least one member is
South African appear more at risk than others, although the coefficient is not always significant.

Poverty does matter. The poorer the household, the higher the probability of being victimised.
The coefficient of the poverty index is significant and its value relatively high (0.25). To illustrate,
between two very similar households, one whose poverty index is 0.1 and one whose poverty index
reaches 0.5, the poorest household faces a 10% higher chance of having people attacked or threatened
for xenophobic motives.
However, looking at the scatter plot between the predicted violence probability and the poverty level,
we can see that people in the city centre tend to be less poor and less victimised (see Figure 2 in
Annex). Therefore, the same regression is run on the two sub-samples to verify that the relationship

12
between the level of poverty and the violence is similar. The results are reported in columns (2) and
(3). The coefficients of the poverty index looses significance on the Alex sub sample. However, the
coefficient of the dummy for foreigners increases from 26 to 39%, suggesting that it is nationality
rather than poverty that defines targets in Alexandra. In the inner city, the coefficient of the poverty
index is still significant and rises to a large 0.6. Meanwhile, foreigners in the city ”only” face a 12%
higher chance of being persecuted. The huge difference between the two sites in the probability of
victimisation of foreign nationals might be explained by the difference in the proportion of foreigners
in the inner city, where they constitute a majority, and in Alex, where they represent a minority made
easy to target.
Looking at the same scatter plot, but this time highlighting the difference between foreigners and
South Africans, the correlation between poverty and violence follows two different trends. I thus run
the regression on the two sub groups. For South Africans, poverty level does not seem to be correlated
with the probability of victimisation; for foreigners (column 6), the coefficient of the poverty index
increases to a very high 0.53. Let’s use the same two households as in the previous illustration, now
both respondents are foreign nationals. The probability of a member of the poorer household being
attacked or threatened on the grounds of xenophobia is almost 30% higher.

The ratio of men and children in the household also matter in terms of victimisation. The more
men and the more children, the most likely members are to have been victims of xenophobia. What
does not seem to matter is whether or not the respondent has achieved a higher level of education -
although the coefficient is negative, going on the lines of the correlation between violence and poverty.
Whether the respondent originally grew up in a rural area does not make his/her household more
likely to be victimised. Interestingly, whether you speak Zulu or Ndebele does not significantly affect
the probability of being threatened or attacked9 .
The results on the length of stay in the Gauteng for the internal migrants and in the country of
the immigrants also confirm the observation by Misago et al. (2009). The relationship is non linear:
the probability of being victimised increases with time spent in the province/country. The maximum
is reached at around eleven years. For foreign nationals who have been there for about a decade, the
probability of being targeted will be about 50% higher than for the newly arrived. After eleven years,
the effect decreases again (see Figure 3 in annex). Foreign nationals who have been in South Africa
for a long time are more at risk than those who have arrived more recently.
9 Alex is completely heterogeneous in terms of languages spoken. There is no clear domination of one mother tongue
over the others, reflecting a high level of ethno-fractionalization in the township. In the sample, respondents mentioned
no less than 20 as mother tongues. The most represented languages are in order IsiZulu, Tsonga, Sepedi and Sotho.
The city centre however is more homogeneous in terms of languages. Berea is dominated by Zulu and Ndebele, which
account for 47% of the interviewees mother languages; Zulu and Ndebele represent about 60% of the mother tongues in
Hillbrow and 50% in Yeoville.

13
Note that only the length of stay in the province or in the country has a significant impact. The length
of stay in the location of surveyed does not have an impact on the violence and therefore dropped.

4.2 Do the reasons for migration matter?

Migrants do not all converge to South Africa, or anywhere else, for the same reasons. Some do escape
political oppression or conflict in their home countries. Some migrate for economic opportunities, hop-
ing that the receiving country will give them a better life. Some reunite with family members already
living in the host country. Some come to South Africa from all over the continent to study (South
Africa has a wide range of educational institutions of high quality). Some migrate for a combinaison of
those reasons and others. Xenophobic violence might be directed differently depending on the reasons
that brought the migrant to the province or to the country. Integration in the new community can
prove more or less successful depending on why one left home in the first place.

In order to verify this assumption, I ran the core regressions while adding discrete variables for
whether the individual moved for economic reasons, for educational reasons, because of war or for
family reunion. Respondents were asked to give the two main reasons for coming to Gauteng/South
Africa so, for a particular individual, up to two of these variables can equal one. Column 1, Table
4 shows that those reasons do not seem to influence the probability of being a target of violence. In
column 2, I introduced interacted terms between the reasons for migration and the location. The
results suggest that economic migrants are particularly vulnerable in Alexandra. Compared to a long-
term resident, an economic migrant in Alexandra will have 12% more chance of being attacked. In the
inner city, those migrants will have a lower probability (5%) of victimisation. Individuals who moved to
Gauteng for educational opportunities face a relatively higher chance of being a victim of xenophobia
in the inner city. Familial reunion in Alexandra also seem to have a negative effect everything else
kept equal.

4.3 Were the May 2008 events a different story?

Overall the violence incidence reaches 15% in the sample collected throughout February 2009. How-
ever, there are reasons to believe that this is certainly largely underestimating the national violence
rate as a lot of victims have left since the attacks. 7.6% of the respondents reported having been a
victim of threats before May 2008, 10.9% over the period May-June 2008 and 10.3% since then. If the
fact that many left the neighbourhood during the attacks prevents us to draw any solid conclusion,
the results yet emphasise that the incidence of violence has not decreased between before and after
the May attacks. Indeed, since the May pogroms, people still report having been victims of threats

14
or attacks linked to their ethnicity or nationality. If many undesirable people left after being brutally
chased away, the percentage of respondents being victimised should have decreased or at least stayed
the same. It seems that more people are being victimised after the attacks than before.

Hence I inquire whether the probability of becoming a victim of xenophobia depends on the period
of focus. The dependent variable was thus desegregated in three: whether the respondent or someone
from his/her household was targeted before May 2008, during May and June 2008 and since June 2008.
The results are presented in Table 5.
Most of the coefficients do not vary excessively from one regression to another. Nevertheless, the
differences observed in the coefficients of the foreign born binary variable is striking. The probability
of being victimised for a foreign national was 11% higher than for South Africans before the riots.
During the riots, it almost doubled to over 20%. This underlines a clear refocus on targeting foreigners
during the May attacks. The impact of being a foreign national on probability of being a victim
remains higher after June than before the xenophobic events. Xenophobic violence is not decreasing,
if anything foreigners are more at risk than before.
The threshold in terms of years spent in Gauteng/ South Africa also keep increasing. The maximum
risk was reached at 9 years before the attacks, 11 years during the attacks, and 12.5 years since.

5 Conclusions
The results cannot be generalised to the whole of South Africa, yet some interesting conclusions can be
drawn from the analysis on what make people more likely victims of xenophobia. More data collection
in different areas across the country is definitely needed.

One important information that came out of the analysis, is that violence against ”outsiders” is still
alive in South Africa. The observed probability of being victimised in our sample does not decrease
between before and after the attacks. In fact, it seems that the people who were interviewed are more
likelily to be threatened since the xenophobic riots than before. This is very worrying for the future:
tensions obviously are even more exacerbated then they used to be in the period preceding the attacks.

During the May 2008 xenophobic attacks, not every migrant was attacked. This was the observation
at the root of this analysis. Who is the alien? Is it possible to identify what make some individuals the
targets? or is violence directed at random? Using a household data collected in Alexandra township
and Johannesburg inner city, the results show that particular characteristics of the household affect
significantly the probability of victimisation on the grounds of xenophobia. First and not surprisingly,

15
foreign nationals are more at risk. However, nationalitity does matter. Zimbabweans are less likely to
be targeted. On the contrary, Mozambicans and Congolese are more likely to have been attacked in
our sample. Secondly, the alien is not a new arrival, it is someone who has been leaving in South Africa
or the Gauteng province, for internal migrants, for about a decade. Indeed, the length of stay has
a U-inverted shaped effect on the probability of being a victim. The maximum is reached at around
eleven years. Regarding those conclusions, the influx of Zimbabweans in early 2008 can be ruled out
as a valid explanation for the riots. However, the perceived influx by residents still remains a possible
trigger. Thirdly, the level of poverty has a huge effect on the probability of victimisation. The poorest
are the most likely targets, at least in our sample. Xenophobia can hardly be fought against without
providing better living conditions to the poorest of this country.

16
References
[1] City of johannesburg.

[2] Violence and xenophobia in south africa: Developing concensus, moving to action. Technical
report, Human Sciences Research Council, 2008.

[3] Tourism and migration. Technical Report P0351, Statistics South Africa, April 2007 to December
2008.

[4] Protecting refugees, asylum seekers and immigrants in south africa. Technical report, Consortium
for Refugees and Migrants in South Africa, June 2008.

[5] Phillip Bonner and Noor Nieftagodien. Alexandra: a History. Witwatersrand University Press,
2008.

[6] Craig Burnside and David Dollar. Aid, policies, and growth. The American Economic Review,
90(4):847–868, Sep. 2000.

[7] Belinda Dodson and Catherine Oelofse. Shades of xenophobia: In-migrants and immigrants in
mizamoyethu, cape town. Canadian Journal of African Studies / Revue Canadienne des tudes
Africaines, 34(1, Special Issue: Transnationalism, African Immigration, and New Migrant Spaces
in South Africa):125–148, 2000.

[8] Centre for Development and Entreprise. Immigrants in johannesburg, estimating numbers and
assessing impacts. Technical Report CDE In Depth no9, August 2008.

[9] Loren Landau. Attacks on foreigners in south africa: more than just xenophobia?, 2008.

[10] Mail and Guardian.

[11] Mendy Marsh. A rapid inter-agency assessment of gender-based violence and the attacks on
non-nationals in south africa. Technical report, UNICEF, July 2008.

[12] J. P. Misago, Loren Landau, and Tamlin Monson. Towards tolerance, law and dignity: Adressing
violence against foreign nationals in south africa. Technical report, IOM Regional Office for
Southern Africa, 2009.

[13] Alan Morris. The desegregation of hillbrow, johannesburg, 1978-82. Urban studies, 31(6):821,
1994.

[14] Michael Neocosmos. The politics of fear and the fear of politics: Reflections on xenophobic violence
in south africa. Journal of Asian and African Studies, 43(6):586–594, December 1 2008.

17
[15] BBC News. Rainbow nation’s outsiders live in fear, 2009.

[16] Olusola Olumefi. Street homelessness in johannesburg inner-city: a preliminary survey. Environ-
ment and Urbanization, 10(2), 1998.

[17] Lena Opfermann. Xenophobia crisis in south africa, Dec 2008.

[18] Reuters. Factbox: Anti-foreigner riots in south africa.

[19] Reuters. Factbox: What impact will south african violence have?, 2008.

[20] Jonny Steinberg. South africa’s xenophobic eruption. ISS paper 169, November 2008.

[21] Time. Johannesburg is burning, 2008.

[22] New York Times. Anti-immigrants violence in johannesburg, 2008.

18
Table 1: Violence patterns: Have you ever been or someone from the
household- a victim of threat because of nationality, ethnicity or tribe?

Alexandra National born and long term resident 6%

National born and recent migrant (less 8%
than 10 years)
Foreign-born 47%
Johannesburg inner-city National born and long term resident 4%
National born and recent migrant (less
than 10 years) 5 %
Foreign-born 15%
Country of birth, % of population threatened South Africa 6%
Zimbabwe 21%
Mozambique 49%
DRC 26%
Malawi 32%

Table 2: Logit: poverty index

Is your household poorer than average?

Meat, fish or fresh vegetables, less than 7 times a week 0.046
(0.027)∗

No flush/ chemical toilet inside the dwelling 0.170

(0.047)∗∗∗

No running water 0.202

(0.034)∗∗∗

Persons per room 0.023

((0.007)∗∗∗

Difficult access to healthcare 0.083

(0.042)∗∗

Difficult access to school 0.150

(0.063)∗∗∗

Log pseudolikelihood -1090.86

Observed p 0.337
Predicted p 0.319
Observations 1918

Robust standard errors in parentheses

* significant at 10%; ** significant at 5%; *** significant at 1%

19
Figure 1: Poverty index distribution, by sub groups

20
Figure 2: Border crossing

source: Statistics South Africa

21
 Figure 3: Scatter: violence predicted and poverty index
in blue: Alex, in red: innercity

.8
.6
Pr(violence)
.4
.2
0

0 .2 .4 .6 .8
poverty

Pr(violence) Pr(violence)

in blue: South Africans, in red: foreign nationals

.8
.6
Pr(violence)
.4
.2
0

0 .2 .4 .6 .8
poverty

Pr(violence) Pr(violence)

Figure 4: Impact of the length of stay in Gauteng (for South Africans) in South
Africa (for foreign nationals) on the probability of being victimised

22
Figure 5: Table 3: logit, characteristics of the victims

All Alex Inner South Foreign All

city Africans nationals
Alexandra=1 0.024 -0.002 0.129 0.018
(0.035) (0.023) (0.091) (0.035)
Foreign national=1 0.256 0.390 0.120
(0.031)*** (0.064)*** (0.022)***

Mixed household=1 0.035 0.014 0.036 0.057 -0.004 0.039

(0.022) (0.044) (0.026) (0.039) (0.040) (0.022)*
Secondary school or -0.016 -0.017 -0.012 -0.007 -0.042 -0.014
more=1
(0.016) (0.020) (0.024) (0.015) (0.042) (0.015)
Come from rural -0.003 0.049 -0.059 0.031 -0.094 -0.004
area=1
(0.019) (0.026)* (0.024)** (0.014)** (0.045)** (0.019)
Ratio of men 0.085 0.132 0.010 0.081 0.004 0.084
(0.037)** (0.062)** (0.036) (0.032)** (0.065) (0.037)**
Ratio of children 0.129 0.214 0.055 0.058 0.216 0.127
(0.033)*** (0.057)*** (0.040) (0.031)* (0.077)*** (0.034)***
Ln (household size) -0.019 -0.031 -0.005 -0.005 -0.046 -0.018
(0.012) (0.018)* (0.018) (0.012) (0.027)* (0.012)
Length of stay in 0.045 0.075 0.020 0.017 0.091 0.045
GP/SA
(0.017)*** (0.031)** (0.019) (0.022) (0.035)** (0.017)***
Length of stay in -0.002 -0.003 -0.001 -0.001 -0.004 -0.002
GP/SA, sq
(0.001)** (0.001)** (0.001) (0.001) (0.002)** (0.001)**
Poverty index 0.246 0.091 0.603 0.068 0.533 0.252
(0.079)*** (0.095) (0.137)*** (0.057) (0.194)*** (0.078)***
Speak zulu=1 0.019 0.046 -0.020 0.020 0.005 0.029
(0.017) (0.025)* (0.023) (0.016) (0.037) (0.018)
Speak Ndebele=1 -0.031 -0.004 -0.015 -0.028 -0.030 -0.006
(0.019) (0.039) (0.024) (0.033) (0.037) (0.026)
DRC=1 0.059
(0.058)

MOZ=1 0.058
(0.053)

ZIM=1 -0.019
(0.032)

SA=1 -0.238
(0.041)***

Log pseudo - -317.77 -304.77 -243.74 -385.11 -

likelihood 646.78 643.86
Observed P 0.159 0.190 0.129 0.069 0.284 0.159
Predicted P 0.121 0.134 0.102 0.058 0.257 0.121
Observations 1770 869 901 1031 739 1770

23
 Figure 6: Table 4: logit, reasons to migrate

(1) (3) (2) (4)

Alexandra=1 0.025 -0.071
(0.035) (0.044)

Foreign national=1 0.241 0.232 0.376 0.117

(0.032)*** (0.032)*** (0.051)*** (0.024)***
Mixed household=1 0.037 0.042 0.013 0.039
(0.022)* (0.022)* (0.041) (0.026)
Secondary school or more=1 -0.020 -0.014 -0.017 -0.017
(0.015) (0.015) (0.022) (0.025)
Come from rural area=1 -0.012 -0.025 0.010 -0.063
(0.020) (0.022) (0.032) (0.024)***
Ratio of men 0.087 0.082
(0.037)** (0.038)**

Ratio of children 0.133 0.141

(0.034)*** (0.033)***

Ln (household size) -0.020 -0.016 -0.005 -0.005

(0.012)* (0.012) (0.016) (0.018)
Length of stay in GP/SA 0.044 0.042 0.078 0.016
(0.017)*** (0.017)** (0.031)** (0.019)
Length of stay in GP/SA, sq -0.002 -0.002 -0.003 -0.000
(0.001)** (0.001)** (0.001)** (0.001)
Poverty index 0.251 0.228 0.071 0.631
(0.078)*** (0.077)*** (0.087) (0.123)***
Speak zulu=1 0.021 0.022
(0.017) (0.017)

Speak Ndebele=1 -0.030 -0.023

(0.019) (0.019)

Economic migrant=1 0.016 -0.050 0.068 -0.018

(0.019) (0.029)* (0.026)*** (0.025)
Education=1 0.039 0.060 -0.029 0.070
(0.028) (0.036)* (0.031) (0.035)**
Escape conflict=1 0.046 0.019 0.020 0.045
(0.029) (0.031) (0.045) (0.031)
Familial reunion=1 0.020 -0.034 0.049 -0.017
(0.026) (0.035) (0.035) (0.033)
Alex x economic migrant 0.170
(0.047)***

Alex x education -0.062

(0.026)**

Alex x conflict 0.044

(0.048)

Alex x Family reunion 0.124

(0.074)*

Logpseudolikelihood -644.54 -633.85 -325.85 -302.98

Observed p 0.159 0.159 0.189 0.129
Estimated p 0.120 0.117 0.139 0.100
Observations 1770 1770 873 904
Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%

24
Figure 7: Table 5: logit, different measures of violence

Violence Violence in Violence Ordered

before May/June after the logit
May riots
Alexandra=1 -0.038 0.027 0.018 0.342
(0.024) (0.024) (0.027) (0.348)
Foreign national=1 0.116 0.204 0.198 2.221
(0.018)*** (0.026)*** (0.030)*** (0.219)***
Mixed household=1 0.022 0.008 0.016 0.204
(0.014) (0.014) (0.020) (0.194)
Secondary school or more=1 0.006 0.002 -0.012 -0.042
(0.011) (0.013) (0.013) (0.158)
Come from rural area=1 -0.005 0.002 0.002 -0.051
(0.014) (0.014) (0.013) (0.180)
Ratio of men 0.017 0.055 0.043 0.683
(0.024) (0.024)** (0.026) (0.319)**
Ratio of children 0.049 0.087 0.056 1.141
(0.026)* (0.030)*** (0.026)** (0.341)***
Ln (household size) -0.011 -0.022 -0.011 -0.196
(0.009) (0.009)** (0.009) (0.115)*
Length of stay in GP/SA 0.018 0.043 0.025 0.494
(0.010)* (0.014)*** (0.010)** (0.173)***
Length of stay in GP/SA, sq -0.001 -0.002 -0.001 -0.022
(0.000) (0.001)*** (0.001)** (0.008)***
Poverty index 0.165 0.146 0.154 2.012
(0.059)*** (0.056)*** (0.059)*** (0.725)***
Speak zulu=1 0.001 0.017 0.001 0.143
(0.011) (0.012) (0.013) (0.157)
Speak Ndebele=1 -0.029 -0.029 -0.029 -0.430
(0.011)*** (0.013)** (0.012)** (0.186)**
Logpseudolikelihood -419.11 -498.05 -476.04 -864.68
Observed p 0.075 0.110 0.104
Estimated p 0.055 0.070 0.065
Observations 1767 1768 1768 1767
Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%

25