Calamity, Conflict and Cash Transfers: How Violence Affects Access to Aid in Pakistan Yashodhan Ghorpade 5 th December 2014.

Calamity, Conflict and Cash Transfers: How Violence Affects

Access to Aid in Pakistan

Yashodhan Ghorpade

5th December 2014

Aid and Violent Conflict• Addressing grievances that give rise to conflict

• “Winning hearts and minds” Berman, Shapiro and Felter 2009

• Political Science literature: does aid reduce conflict?• Yes – Justino - India (2011, forthcoming); Berman, Shapiro and Felter (2009) – Iraq• No – Crost and Johnston (2014) – Phillipines • Mixed evidence - Beath, Cristea and Enikolopov (2012) NSP, Afghanistan

• Development Economics literature: Does conflict affect programme impact?• Higher conflict makes people use transfer to migrate out – Mesnard (2009) –

Familias en Accion, Colombia• Higher impact on school enrolment in conflict areas – Wald and Bozzoli (2009),

Familias en Accion, Colombia

Why is the link between conflict and aid important?• Assumptions by both, political actors and development community

critical to success• Pathway to understanding political landscape of areas affected by

conflict – actors, incentives, capabilities – and therefore options• Better understand considerations while extending aid to conflict-

affected areas

Structure

• Case study setting and context• Data and Identification Strategy• IV Estimation and 1st Stage Results• Does conflict reduce access to aid? YES

• Mechanisms: How does conflict reduce access to aid? ARMED GROUPS

• Conclusions and Policy Implications

Case Study Setting and Context

• Pakistan – several forms of violence• Marked increase in violence

from 2001 – present• Terrorist, insurgency, sectarian

motives for internal political violence• Post 2010 flood period. Soon

after democratic transition, marked by high violence

Fatalities due to Conflict in Pakistan 2001 –2010: BFRS and SATP data

2001 2002 2003 2004 2005 2006 2007 2008 2009 20100

2000

4000

6000

8000

10000

12000

BFRS SATP

Aid Programmes in Pakistan

• Reforming Social Protection landscape: democratic transition in 2008• Benazir Income Support Programme (BISP) – • UCC based on poverty score card• Regular monthly payments to chronic poor

households, women recipients

• CDCP (Watan) - Cash transfer-based flood relief programme in 2010• Universal except in KPK – based on house damage

Data and Methodology

• CDCP Impact Evaluation Baseline data; collected after the 2010 floods• 7802 households, representative for flooded areas at the Province,

National levels• Key modules: Shocks and Coping Strategies, Demographics, Assets,

Flood Exposure, SP receipts• Conflict data: sub-district level conflict exposure measures using SATP

data for 2001 – June 2010 (pre-floods)• Measured as Log (1+n) killings between 2001-10 at sub-district level

Identification Strategy• Conflict not random, not neatly demarcated by an exogenous break:

contagion over time • Violence over 2000s direct spill-over from War on Terror in

Afghanistan, history and politics studied by several authors (Gul 2009; Rashid 2012; Shapiro and Gulzar 2012; Gunaratna and Iqbal 2010; Jones and Fair 2010; Hussian 2008)

• IV estimation: Distance to Afghanistan• Control for confounding factors

Mapping district-level conflict intensity in Pakistan

IV Estimation

Causal Relationship of Interest:

IV First Stage:

2nd Stage:

IV Estimation: Potential Threats and MitigationAssociation between proximity to Afghanistan and Access to Aid• Remoteness: Province Dummies, Distance from Administrative Centres (Provincial,

District)

• Topography: Dummy for topography type (Mountain, Hill, Plateau, Valley, Coastal Plains, Inland Plains)

• State Presence: Community-level Index (Govt. schools, health facilities, immunisation camps, community health workers, post offices administrative HQs)

• Infrastructure: Community-level Index (Bus/wagon stop, Motorable Road, Railway Station, Retail Shop, Wholesale market, Bank, Flour mill, Tractor rental centre, Fertiliser depot, Electricity, Gas, Drainage)

• Linguistic Fractionalisation: Index a la Alesina et al. (2003)• Army Control: Distance to Army Cantonments • Province-specific features, capacity: Province Dummies

IV First Stage Results (1) (2) (3)

Distance to Afghan Border -0.877*** -0.494*** -0.266***

(-12.61) (-10.09) (-12.13)

Controls NO YES YES

Province dummies NO NO YES

N 7802 7786 7786

F-statistic (1, 496) 160.93 101.82 14.99

Prob. > F 0.0000 0.0001 0.0001

Adjusted R-squared 0.3373 0.5411 0.5683

Causal Impact of conflict on likelihood of receiving transfers?

CDCP - I (Flood Relief) BISP

probit IV probit probit IV probit

Log (n+1) killings at tehsil level 0.003 -0.438*** 0.002 -0.257**

-0.7 (-3.29) -0.49 (-2.07)Controls Y Y Y Y

Province Dummies Y Y Y Y

N 7802 7802 7802 7802

Conflict and Household-level Access to Aid• Conflict reduces access to aid at the household level for both CDCP

and BISP programmes• Controlling for a range of characteristics that determine eligibility and

enable accessing aid (demand), and factors that facilitate rollout at the community level (supply)• Stronger effect on CDCP than BISP• Effect NOT driven by differential eligibility. In fact, on some indicators

households in conflict-affected areas more eligible for aid

How does conflict reduce access to aid? Mechanisms• Important to note controls – infrastructure, state presence,

remoteness, topography – impact of conflict is over and above the effect of these factors• Test one hypothesis/ plausible mechanisms: Presence/ Control of

Non-State Armed groups • Armed groups resent efforts by the government to win over political support

Gompert et al., 2009; Crost and Johnston 2014• Targeting of state services, development interventions: Attacks on NGOs,

abduction of aid workers, attacks on immunisation camps, BISP office (Turbat) - Gul 2010; Jones and Fair 2011; Gunaratna and Iqbal 2010

Armed non-State Actors in Pakistan

• Astounding increase not only in violence but also in the presence and influence of extremist and militant groups, especially in the NW• Avowed aim to overthrow/ transform the nature of the Pakistani

state, delink from the US, adopt Sharia, enforce worldview• Groups loosely coalescing under the umbrella of the Tehrik-e-Taliban

Pakistan (TTP) since 2007• Several groups view state as enemy

Measuring the presence of armed groups• No published data – NOT straightforward!• Violence not a perfect proxy of armed group presence; violence

erupts where there is contest, not in domain of opponents’ complete dominance (Kalyvas 2006; Justino and Ibanez 2014)

• Proxied attempting to model the influence of TTP groups on girls’ primary schooling• Extensive reportage of TTP opposition to girls’ education in areas of

their control• Residual Analysis from Community-level estimations of Girls’

Schooling

Modelling Community-level Female Primary EnrolmentComplete Model:

Observed Model:

However,

Observed :

• If negative, predicted enrolment higher than observed enrolment• More likely associated with presence of ANSA: • Divide communities by +/- residuals, test effects of conflict on aid over

sub-populations• If ANSA more likely in communities with negative residual, negative

could explain effect on aid receipts?• If is indeed random error, communities with < 0 similar to communities

with > 0 (random error with mean 0); expect similar effects of conflict on aid

Estimating Community-level Female Primary Enrolment Rate

Demand Side

• Adult Education (Female)• Adult Education (Male)• Sex Ratio (Adult)• Receipt of Remittances• Average Income level• Land Ownership Patterns• Wealth Status• Enterprise Ownership• Displacement status (Flood)• Linguistic Identity

Supply Side

• Presence of Govt. and Private Girls’ and Co-Ed Schools in Community• Community Infrastructure• State Presence• Remoteness from Administration• Electricity Access• Former Princely State

Effect of Armed non-State Actors presence on CDCP Receipts

Full Sample Effect of conflict on Receipt of CDCP (Flood Relief Transfers): Over Residuals of Female Primary Enrollment Estimation

> 0 < 0

Taliban likely not present Taliban likely present

Log (1+n) killings -0.541*** -0.267 -0.757***

(-4.08) (-1.25) (-6.31)

N 7786 4254 3532

Effect of Armed non-State Actors presence on BISP Receipts

Full Sample Effect of conflict on Receipt of CDCP (Flood Relief Transfers): Over Residuals of Female Primary Enrolment Estimation

> 0 < 0

Taliban likely not present Taliban likely present

Log (1+n) killings -0.463*** -0.333** -0.645***

(-3.81) (-2.38) (-3.48)

N 7786 4254 3532

Effect of Armed non-State Actors presence on BISP Receipts

Full Sample Residuals of Female Primary Enrollment Estimation

Residuals of Primary Enrollment Gender Gap Estimation

Taliban likely not present

Taliban likely present

Taliban likely not present

Taliban likely present

Panel A

Conflict-affected tehsil (dummy)

-1.890*** -1.457** -2.184*** -0.207 -2.568***

(-3.75) (-2.20) (-3.34) (-1.43) (-4.06)

N 7786 4254 3532 3651 4135

Panel B

Log (1+n) killings -0.463*** -0.333** -0.645*** -0.057 -0.576***

(-3.81) (-2.38) (-3.48) (-1.53) (-3.50)

N 7786 4254 3532 3651 4135

Armed Group Presence and Aid

• Strong and significant results for both programmes• Echoed in qualitative, anecdotal references to the presence/ control

of armed non state actors limiting access to aid• Armed group presence deters aid workers/ bureaucrats to enter/

operate in areas of their control (supply of aid) ?• Inhibits households from overtly seeking state aid (demand for aid) ?

Conclusions & Policy Recommendations• Conflict reduces access to both cash transfer programmes

• Robustness: Results hold when using dummy for conflict-affected sub district

• Presence of TTP-linked armed groups lowers access to aid• Reduces supply by threats to aid workers, officials, AND demand by threats of retaliation, reduced mobility

of local populations• Robust to estimation of community-level gender gap in primary schooling

• This is worrying: The lack of access to aid may exacerbate development shortfalls, horizontal inequalities – with implications for future peace and development

• Need to question more carefully the ability to implement programmes in conflict-affected areas to achieve development, political goals

• Need for conflict-sensitive analyses to inform development policy and operations in conflict-affected areas. Realities too distinction for “development as usual”.

• Reaching households may require striking a deal with/ paying off armed groups – ultimately a political, tactical decision, e.g. Tsunami aid in Sri Lanka

Discussion: Winning Hearts and Minds or Losing Lands and People?• Clear intent to use aid to win hearts and minds, foster nation-building• Aid as an instrument for achieving state goals – popular support, redress of historic grievances, and

tactical support in fight against militants/ insurgents

“Conflict in NWFP [old acronym of the Khyber Pakhtunkhwa province], FATA and Balochistan has severely challenged the ability of the state as well as the legitimacy of the idea of a functioning state in Pakistan. Social protection must be part of the strategy to reclaim the space and legitimacy for the state in Pakistan, through protection to the basic entitlements of people in the conflict-affected areas…

“Expanded social protection programmes, particularly directed at the conflict-affected areas are essential to protect innocent victims of conflict, and to regain legitimacy for the idea of a functioning state through creating, expanding and ensuring the delivery of citizenship-based entitlements.”

GoP (2010), pp. 145

… Yet, Ability may curtail intent:

“The challenge in NWFP and FATA comes from groups that seek to impose their own vision of society on the majority. They violate democratic norms, actively and violently oppose social policy and development initiatives such as girls’ schooling, immunization campaigns, and even income transfers to women. The armed activities of militants and the state’s armed response has led to large-scale loss of life, displacement, and destruction of infrastructure, particularly social infrastructure. There has been massive disruption to livelihoods and economic activity.”

(GoP 2010), pp. 145

Robustness Tests: Conflict Dummy CDCP – Phase I BISP CDCP – Phase I BISP

probit probit probit probit IV probit IV probit IV probit IV probit

Conflict-affected tehsil (dummy)

0.118*** -0.016 -1.619*** -1.100**

-4.36 (-0.24) (-3.44) (-2.05)Controls Y Y Y Y Y Y Y Y

Province Dummies Y Y Y Y Y Y Y Y

N 7802 7802 7802 7802 7802 7802 7802 7802

Likelihood of entire community being excluded from programme

CDCP - I BISP

Conflict-affected Tehsil Log (1+n) killings - tehsil

Conflict-affected Tehsil Log (1+n) killings - tehsil

Conflict-affected Tehsil (dummy)

1.275 2.457***

(1.45) (7.38)

Log (1+n) killings - tehsil 0.188** 0.613***

(2.16) (3.88)

N 497 497 497 497

Community-level coverage rates of programmes

CDCP - I BISP

Log (1+n) killings - tehsil

Conflict-affected Tehsil Log (1+n) killings - tehsil

Conflict-affected Tehsil

Conflict-affected Tehsil (dummy)

-1.475 -0.847*

(-1.41) (-1.75)

Log (1+n) killings - tehsil -0.256** -0.156**

(-2.30) (-2.48)

N 497 497 497 497

Estimation of Female Primary Enrolment

N 439

R2 0.517

adj. R2 0.472

F 11.59

Prob. F > 0 0.0000

Effect of Armed non-State Actors presence on CDCP Receipts Full Sample Residuals of Female Primary

Enrollment EstimationResiduals of Primary Enrollment

Gender Gap Estimation

Taliban likely not present

Taliban likely present

Taliban likely not present

Taliban likely present

Panel A

Conflict-affected tehsil (dummy)

-1.959*** -1.126 -2.344*** -0.367 -2.575***

(-4.13) (-1.22) (-5.66) (-1.12) (-5.38)

N 7786 4254 3532 3651 4135

Panel B

Log (1+n) killings -0.541*** -0.267 -0.757*** -0.101 -0.696***

(-4.08) (-1.25) (-6.31) (-1.25) (-5.02)

N 7786 4254 3532 3651 4135

Effect of Armed non-State Actors presence on BISP Receipts

Full Sample Residuals of Female Primary Enrollment Estimation

Residuals of Primary Enrollment Gender Gap Estimation

Taliban likely not present

Taliban likely present

Taliban likely not present

Taliban likely present

Panel A

Conflict-affected tehsil (dummy)

-1.890*** -1.457** -2.184*** -0.207 -2.568***

(-3.75) (-2.20) (-3.34) (-1.43) (-4.06)

N 7786 4254 3532 3651 4135

Panel B

Log (1+n) killings -0.463*** -0.333** -0.645*** -0.057 -0.576***

(-3.81) (-2.38) (-3.48) (-1.53) (-3.50)

N 7786 4254 3532 3651 4135