migration, violence and welfare in rural colombia

© Institute for Fiscal Studies, 2006

Migration, Violence and Welfarein rural Colombia

by Alice Mesnard, IFS

Orazio Attanasio, UCL, IFS


Introduction

• Civil conflict has displaced many families and individuals from their villages of origin. 4.3% pop., 14% rural – (Arboleda and Correa, 2003)

• Costs are large : assets, inadequate human capital, poverty…

• Policy makers have shown an increasing interest in building interventions to curb these flows


Objectives

• Understand the determinants of mobility decisions in a violent context

• We embed new motives related to violence, community characteristics, and policy interventions in the framework of economic migration

– Do traditional motives for economic migration apply in a violent context?– How do welfare programmes affect household migration in such context ?

• Our concept of mobility differs from displacement.

• Migration decisions are not necessarily entirely forced but are likely to be affected by high intrinsic violence levels in rural villages: how do these factors interact?

• We also compare ‘stayers’ with ‘movers’


Road map

1. Review of literature on migration, violence and welfare2. Data and samples3. Model of household migration with selection4. Does violence modify migration incentives?5. Understanding better the impact of violence and

welfare programme on migration6. Other migration determinants7. Compare a sample of poor individuals from small

towns with a sample of displaced individuals from similar towns

8. Policy implications and future extensions


1 Motivation of the empirical model

Traditional literatureHarris and Todaro (70) Human capital theory (Sjaastad 62, Becker 64)

or in uncertain environment (Da Vanzo 83, Pessino, 91)New economic of migration (Stark, 91): within householdImportance of networks (Massey and al., Munshi, 2003, Munshi and

Rosenzweig, 2005 )

Literature on violence and migrationSchultz 71, Morrison and May 94 : effects of violence on internal migration

in Colombia and GuatemalaDisplacement and asylum seekers (Azam and Hoeffler, 2002, Hatton...)Engel and Ibanez (2005) : displacement differs from migration.

Literature on welfare programmes and migration: scant !Angelucci (2005): impacts of PROGRESA on international migration


Contribution

Microeconomic underpinnings of household migration

Number of factors : social capital, risk exposure, shocks, liquidity constraints, violence, policy interventions...

We allow the violence to affect not only household well-being directly but also to affect the incentives associated to other migration factors

In particular policy interventions may have different impacts on migration depending on violence level


Potential impacts of welfare programmes ?

• side effect of Conditional Cash Transfer programmes

(-) Benefits deter households to move out of “Treated” town

Mitigate aggregate risk, spill-over effects…

(+) Cash transfers help relax liquidity constraints.

Their conditionality may mitigate this effect.

Heterogeneous impacts of the programme

if violence is low (-)

if violence is high (+)


The Familias en Acción Programme

Familias en Acción (FeA) is a CCT implemented in Colombia in 2002.

The programme is modeled after the Mexican Oportunidades/ PROGRESA intervention

It consists of: a health and nutrition component (46500 monthly pesos conditional on

participation in health component)

an education component 14000/28000 conditional on primary/secundary school enrolment and attendance

The transfer is targeted to mothers

The program started in 627 municipalities (small towns with enough infrastructure) and is now being expanded considerably.

It is projected that about 1.5 million households will be in the programme by the end of next year.


The Familias en Acción Survey

A consortium formed by IFS and two Colombian entities won the contract to evaluate the effects of the program.

For this reason a large data operation was started in 122 towns : 57 treatment and 65 control. The allocation was not random

A sample of 11,500 household was drawn from the SISBEN 1 lists of December 1999 and interviewed in 2002.

To achieve a sample of that size, an initial sample of about 19,000 household was drawn from the same lists.

Of these, 11,500 were still living in the same town (and eligible for the programme)

A large number of households were lost because of high mobility.



Baseline collected between June and November 2002

Very rich and exhaustive household survey (3.5 hours on average It includes information on consumption, income, education, time

use, shocks, attitudes, expectations and so on and so forth.

In some of the treatment towns the programme started before the baseline (TCP and TSP)

Survey was complemented by other smaller surveys Schools, health care centres, community nurseries, local authorities, locality

surveys.

The towns are clearly affected by violence and the civil war: Hard evidence (matching with municipality level panel data on violence – DNP)

anecdotes



The follow up survey was planned (and executed) in 2003.

Given the high level of mobility between 1999 and 2002 we were very worried about attrition.

We also thought that the survey gave us a unique opportunity to study the mobility of a very vulnerable population in the places where they lived.

We obtained funds from the IADB to invest in tracking households down and to study mobility and violence.

Three components: Tracking down movers Special module on movers.

Much more in depth interview with local authorities on violence and mobility

Social capital games piloted in 12 villages.



Mobility went down considerably.

Attrition was only 6% 2,026 households changed address between treatment and

follow up. 1,316 within village

710 outside village

114 were tracked down

596 were lost

275 moved for unknown reasons

114+321 moved to different municipalities.

Most of the 321 who were lost moved to big cities.


Reasons for migration

within town(1)

out of town(2)

Violence 1.9 14.9

For job related reasons 16.9 54.4

To find better accommodation 22.8 2.6

To live closer to relatives 8.3 14.0

To live closer to centre 1.0 0.0

To live closer to college 3.8 3.6

Others 45.3 10.5

Total 100 100

Notes: in (%) of answers


Costs of migration

Migration costs are high for very poor households:

median costs = 50,000 pesos, mean costs = 103,037 pesos i.e. 21% and 43% of average monthly income

To finance their migration none relied on credit or loans !(2/3 used own funds, 1/3 was helped by friends).

However, past migration flows are estimated around 10-15% per year.


Why are there potential problems of selection ?

• 40 % of households registered as very poor in 1999 were sampled for the FA survey but are not in the baseline survey in July 2002.

• So, possibly, the households in the baseline sample are selected on unobservable characteristics that make them least mobile.

• In this case, migration determinants may be biased

3 Model of migration with selection


Simultaneous estimation of migration equation :

corr( , )vij ij

: with selection equation 2 ij21 'X 0ij ijY v

1 2 3 j 4 ij1 'Violence 'X 0ij j ijY Treat

ij i j

where 1 if migrated between 2002 and 2003, 0 oth.

X : household village characteristics

Violence : proxies for violence at baseline

=1 if treated village, =0 if control village

ij

j

j

Y

and

Treat

2where 1 if i still in baseline survey in 2002, =0 if has disappeared between 1999 and 2002ijY

Estimation of the model :


Determinants of selection equation

• Use data on all households registered in the municipalities in 1999 for the SISBEN survey

• We need at least one instrument that we can exclude from the migration equation

Instruments : number of victims, kidnappings, displaced individuals per 10,000 inhabitants before 1999

source: National Police data matched at municipality level


Results of selection equation

Notes: Coefficients obtained with a Probit model, number of observations 19148

Coefficients Std. Err. z statistic

affiliated~s 0.112 0.041 2.76

urban 0.046 0.019 2.47

educ2 0.031 0.020 1.55

educ3 0.233 0.038 6.07

female -0.068 0.023 -2.98

persfami 0.143 0.021 6.7

persfami_sq -0.008 0.002 -4.95

no_under17 0.140 0.025 5.67

no_under17~q -0.012 0.003 -3.65

displaced 0.000 0.000 -13.29

victims -0.039 0.019 -2.03

sequest -0.033 0.004 -8.46

_cons -0.698 0.055 -12.77


Results of the migration equation

Variables Non migrants Migrants description

d_desertion0.09 0.17* 1 if taskforce desertion in health

center due to violence, O oth.(0.29) (0.38)

d_strike0.25 0.30* 1 if taskforce strike in health

center, O oth.(0.43) (0.46)

curfew0.12 0.15* presence of curfew in

municipality(0.32) (0.35)

eln_farc_pm0.61 0.73* presence of illegal armed groups

in municipality(0.49) (0.44)

probl_op0.65 0.78* problems of public order in

municipality (0.48) (0.42)

n_dispop5.42

(13.51)10.19*

(20.52)displaced households the year

before baseline

Notes: Standard deviations in parentheses,

* significantly different from column (1)


Main results : Wald test of independence rejects the significance of ρ at 31% level (Chi Square(1) =1.04)


Effects of programme and violence

1 2 3 j 4 ij1 'Violence 'X 0ij j ijY Treat

Notes: Column (2) adds proxies for occupation of household head

Marginal effects associated to (1) (2)

1 if lives in treated municipality -0.836 -0.683

(0.505)* (0.450)

Number of displaced households 0.032 0.028

(0.012)*** (0.012)**

curfew 0.939 0.956

(0.567)* (0.557)**

presence of illegal armed groups 0.969 0.920

(0.530)* (0.491)*

1 if suffered taskforce desertion, O oth. 0.512 0.391

(0.597) (0.570)


1 2 j 3 j 4 5 ijyields 1 Treat ' Treat *Violence ' Violence ' Xij j j ijY

2 20 21 j2 parameterise to gain efficiency '*Violence

22 2

2

(low level of violence)1 (Violence)

(high level of violence)

1 2 3 j 4 ijSo far : 1 'Violence 'X 0ij j ijY Treat

4 Does violence incidence modify migration motives ?

Q1: does the programme impact depend on the level of violence ?


Notes: “high level” defined Left : by presence of ELN, FARC, paramilitaries in municipality Right: number of displaced households >5 (most violent quartile)

High level Low level High level Low level

(3) (6) (3) (6)

treat -0.737 -1.739 0.605 -1.375

(0.537) (0.532)*** (1.836) (0.463)***

Obs. 5099 2661 1673 5771

Results of specification 1 :

heterogeneous impacts of programme


Probit estimates (coefficients)

(1) (2) (3)

Programme effect -26.097 -40.390 -26.264

(11.455)** (10.455)*** (12.071)**

Interaction effect ofProgramme*violence

0.780 0.674 0.967

(0.260)*** (0.282)** (0.320)***

Violence effect 0.382 0.346 0.401

(0.152)** (0.141)** (0.159)**

observations 8837 7078 8837

Notes :“Violence” is measured by number of displaced households before the survey

Column (2) adds the controls for occupations of household heads column (3) adds controls for social capital

Results of specification 2 :


Note: points represent municipalities

-.0

2-.

01

0.0

1

effe

ct of F

A w

ith m

np

io c

ontr

ols

0 .2 .4 .6 .8 1displaced households in the past/100

control towns treated towns

Programme impact depends on violence level…


Adding the direct effect of displacement…-.

02

0.0

2.0

4.0

6to

tal effe

ct w

ith

mn

pio

con

tro

ls

0 .2 .4 .6 .8 1displaced households in the past/100

control towns treated towns


Robustness checks

Heterogeneous impact of programme along violence measured by presence of illegal forces (dropping 5 extreme values)

Direct impacts of violence (Presence of a curfew) : 1.186** (0.454)

Direct impact of programme : -2.135***(0.580)

Interaction impact of programme*violence: 1.367** (0.628)

Dropping the municipalities with extreme levels of violence


Q2. Do other migration motives depend on violence?

No significant heterogeneous impacts along:

• Household social position in village (edu. levels, social capital)

• Working in agriculture

• Living in rural, more isolated parts of municipalities

But households with larger size, smaller proportion of children, whose head is

older respond more strongly to violence.

1 2 j 3 j 4 5 ij 6 ij 1 Treat ' Treat *Violence ' Violence ' X ' X *Violenceij j j j ijY


Understanding better the impact of violence and welfare programme

• Is the impact of violence similar to other negative shocks on household income ?

• Is there more evidence for liquidity constraints ?

• Do they affect differently household migration depending on violence incidence ?


Households may respond more strongly to aggregate (village) shocks than to idiosyncratic shocks

Description of negative shocks on household income :

Dummy =1 if hhd income affected by, 0 otherwise:

Shocks in 2002

Shocks occurring in 2000/2001/2002

Obs Mean Std. Mean Std.

death 8837 0.025 0.156 0.054 0.226

illness 8837 0.104 0.305 0.169 0.375

Crop loss 8837 0.161 0.368 0.268 0.443

Business loss 8837 0.014 0.119 0.025 0.156

Fire, flood natural disaster 8837 0.016 0.126 0.036 0.187

Violence, robbery or displacement

8837 0.013 0.112 0.031 0.173

Is the impact of violence similar to other negative shocks on household income ?


(1) (2)

household shock: death02 2.137 2.442

(0.929)** (0.924)***

household shock: illness02 0.194

(0.594)

household shock: croploss02 -0.508

(0.452)

household shock: busloss02 0.460

(1.474)

household shock: fireflood02 -1.457

(1.793)

household shock: violence02 3.067 3.218

(1.173)*** (1.160)***

% of households in village with death 02 -0.344

(0.136)**

%hhs in vill. with income losses due to viol 02 -0.250

(0.144)*

Observations 8837 8837


Is there more evidence for liquidity constraints ?

• Household wealth measured by lots of variables : quality of walls, education of household head and spouse, owning a house, phone, sewage system ...

We look at the effect of • Net value of property• Net stock of savings• Net wealth=net value of ppty + Net stock of savings


Marginal impact on migration (1) (2) (3) (4)

net wealth /10e+08 0.033

(0.021)

household shock: violence02 2.978 2.972 2.939 3.727

(1.208)** (1.208)** (1.205)** (1.286)***

household shock: death02 2.122 2.102 2.248 1.812

(0.933)** (0.936)** (0.932)** (0.970)*

Net value of property /10e+08 0.039 0.014

(0.019)** (0.027)

Value_ppty interacted with death02 0.076

(0.030)**

Value_ppty interacted with viol02 -0.289

(0.142)**

Net savings /10e+08 -0.159

(0.097)*

Observations 8837 8837 8837 8837


Magnitude?

• Median net wealth : 1+e6

• Mean : 3.52325 +e6

• Std. Dev.: 7.42107 +e6

• Increasing net wealth by median net wealth would increase probability to migration by 0.04 percentage points

• Increasing net wealth by one standard deviation would increase it by less than 0.3 percentage points


6 other determinants of migration

Marginal effect (per 1000 pesos)

(1) (2) (3) (4) (5)

hourly wage in urban part of municipality

0.007 0.010 0.007 0.006 0.008

(0.010) (0.008) (0.010) (0.008) (0.010)

hourly wage in rural part of municipality

-0.005 -0.004 -0.004 -0.003 -0.005

(0.002)** (0.002) (0.002)** (0.002) (0.002)**

Effect of wages:

Notes : Marginal effects of a Probit model ,1,000 pesos represents more than 1.5 standard deviations from mean hourly wage100,000 pesos = 40% of monthly income of very poor households in treated municipality Standard errors in parenthesis. *** significant at 1%, ** significant at 5%, * signi at 10 % (1) household and municipality characteristics (2) dropping some municipality level characteristics (3) more education levels for household head and spouse (4) adding occupation of household heads and spouse (5) adding social capital


Other Municipality factors

• Altitude increases the probability to migrate• Regional characteristics have significant impacts• No significant impact of social capital• Weak impacts of infrastructure: health, education, sewage and

water

Household characteristics

• (-) Size of households (quadratric effect)• (+) household head is single (0.8)

• Education levels have no significant impactsBut occupations have strong impact: (-) agriculture

(-) self-employed, employed and employer


Strong effects associated to property rights…

Marginal effectin percentage pts

(1) (2) (3) (4) (5)

1 if house is rented or in mortgage, 0

1.970 2.183 2.012 1.254 2.021

(0.554)*** (0.614)*** (0.599)*** (0.673)* (0.600)***

1 if house is occupied without legal agreement, 0

-2.998 -3.378 -3.374 -3.364 -3.363

(1.681)* (1.598)** (1.592)** (1.363)** (1.591)**

1 if house is in usufruct, 0 otherwi

0.865 1.054 1.020 1.092 1.016

(0.374)** (0.390)*** (0.383)*** (0.370)*** (0.385)***

Notes: Omitted category = house is owned


Strong effects associated to type of insurance

Notes : in Column (2) we add proxies for occupation of the household head and spouse4% households have type 1 insurance69% households have type 2, 10 % are not insured.

Marginal effect (percentage points) (1) (2)

1 if EPS =unsubsidized health insurance, “best” type -3.553(1.090)***

-2.844(1.153)**

1 if ARS (2nd best type of insurance) -1.282 -1.049

(0.575)** (0.627)*

1 if Vinculado (3rd best type) -0.696 -0.365

(0.648) (0.677)


Comparing ‘stayers’ and ‘movers’

• So far we have analyzed mainly the FeA sample and focused on the features of households who are in their villages in 2002

• We now compare these households with a sample of households that were displaced.

• This data is taken from a survey of displaced individuals contacted in several large cities in Colombia by Econometria within a study of food security.

• We restrict the sample to displaced individuals coming from the same regions and type of municipalities in the FeA sample.


Table 8: household composition of displaced and non displaced households

Number of people in HH

Kids 0-6

Kids 7-11

kids 12-17

kids 7-17

Adults

Female older than 18

DISPLACED HOUSEHOLDS

Percentiles

1% 1 0 0 0 0 1 0

25% 4 0 0 0 1 2 1

50% 6 1 1 1 2 2 1

75% 7 2 2 2 3 3 2

99% 15 5 4 4 6 8 4

Mean 5.99 1.39 1.01 0.93 1.94 2.66 1.35

Std. Dev 2.75 1.25 1.04 1.07 1.61 1.43 0.78

NON-DISPLACED HOUSEHOLDS

Percentiles

1% 2 0 0 0 0 1 0

25% 4 0 0 0 1 2 1

50% 6 1 1 1 2 2 1

75% 7 2 2 2 3 3 2

99% 14 4 4 4 6 7 4

Mean 5.98 1.15 1.04 1.03 2.07 2.76 1.36

Std. Dev 2.42 1.15 0.95 1.02 1.39 1.36 0.72

Note: the non-displaced households are from the FA survey.


Table 9: number of nuclear families living together

DisplacedNon-displaced

Only 1 family lives in the household

88.0 94.1

Only 2 nuclear families live in the household

9.1 4.6

3 or more nuclear families live in the house

3.0 1.2

Total 100 100.0


Table 10: Number of deaths in the household during the last 12 months

Percentiles

Displaced

Non-displaced

1% 0 0

25% 0 0

50% 0 0

75% 0 0

90% 1 0

95 % 1 0

99% 2 1

Mean 0.146 0.020

Std. Dev. 0.433 0.146


Table 11: house propertyDisplaced

Non-displaced

House is owned 27.2 64.3

House is rented or in mortgage 32.4 9.9

House is occupied or borrowed 29.8 4.6

House is in usufruct 10.7 21.3

Total 100.0 100.0


Table12: floor materials

DisplacedNon-displaced

Sand 51.0 40.4

Conglomerate 37.5 50.5

Tiles 6.4 4.4

Wood 5.1 4.6

Total 100.0 100.0


Distribution of total expenses of displaced and non-displaced households

Displaced households

Non displaced households

1% 29,600 45,483

25% 192,600 225,723

50% 308,501 338,342

75% 469,234 490,402

99% 1,380,803 1,174,228

Mean 370,571 385,286

Std. Dev. 273,563 237,109


Displaced Non displaced

head spouse head Spouse

None 24.08 19.76 26.04 20.94

Incomplete primary 42.5 46.85 45.9 47.03

Primary 17.96 17.07 14.64 16.75

Incomplete secondary 10.63 12.99 9.47 10.69

Secondary or more} 4.84 3.34 3.94 4.58

TOTAL 100 100 100 100


Occupation before displacement

HEAD DisplacedNon-displaced

Work 79.6 82.2

Farmer 58.4 46.0

Family worker 2.7 1.1

Employer 1.1 2.7

Self employed 41.3 39.7

Domestic 2.1 2.3

Employed 32.3 35.7


Distribution of total

expenditures

Non-displaced households

(FA)

Displaced households by duration since displacement

months after displacemen

t t<=6 6<t<=1212<t<=20

t> 20

1% 45,483 32,800 8,000 29,000 50,200

25% 225,723 181,350 192,000 196,600 207,300

50% 338,342 280,380 307,300 321,350 330,600

75% 490,402 431,925 466,400 479,996 484,600

Mean 385,286 341,638 374,885 380,312 382,446

Std. Dev. 237,109 255,393 295,934 278,660 255,467


Age groups Displaced households (WFP)

Non migrant households (FA) at baseline

7-13 years old children

82 % 91 %


46 % 59 %


83 % 94 %


59 % 68 %

Children enrolment


Differences between displaced and migrant households

displaced households migrant households

head Spouse head spouse

None 24.08 19.75 24.56 19.14

Incomplete primary 42.5 46.85 47.15 46.86

Primary completed 17.96 17.07 13.65 16.5

Incomplete secondary and more 15.46 16.33 14.64 17.5

TOTAL 100 100 100 100


7 CONCLUSIONS

• Violence incidence and adverse income shocks affect strongly and positively migration

• Receiving welfare benefits decreases migration only if violence is not unduly high

• Other strong impacts are associated to property rights and type of insurance


Policy implications:

• Our paper does not take any normative stand.

« If policy makers want to curb migration », then:

• Policy measures oriented towards rural development and better insurance could be very effective

• Welfare Programmes too• However cash transfers may also help households to leave very

violent areas

Is this a bad or a good thing in such a context?


Future research

Experimental risk-sharing games : effect of social capital and risk.

Investments in physical assets, human capital and migration under extreme violence.

Ambiguous effects of violence:

child labour may serve as « buffer » / human capital is the most mobile asset.Impact of Conditional Cash Transfers under uncertainty ?

Intra-household risk diversification mechanisms : individual migration, time uses allocation, transfersHow does the CCT programme affect these mechanisms ?


Effect of social capital using survey and games

Marginal effect in percentage points

group 1.824 1.138

(1.078)* (1.129)

number of risk sharing groups (from the games) 0.254

(0.085)***

dum_game 1.721

(1.273)

Notes: the dummy variable “dum_game” = 1 for pilot municipalities/0 oth.

Other proxies for social capital are not significant (group size, proportion of people who join a group…).

migration, violence and welfare in rural colombia

Documents