1. society
2. work
3. unemployment

Opportunity and Resilience: Do Public Works Have it All?
Evidence from a randomized evaluation in Sierra Leone
Nina Rosas1 and Shwetlena Sabarwal2
Abstract
Given the increased reliance on Public Works (PW) programs, especially in fragile and postconflict settings, this paper examines whether such programs can go beyond mere income
stabilization and provide households with pathways out of poverty and underemployment. To
do this we assess the very short-term impacts of a PW program in Sierra Leone targeted at
youth in response to the global food, fuel, and financial crises. Using a randomized phased-in
approach we find that the program successfully reached youth with low levels of education who
were mostly working in the agricultural sector. As a result of program participation, household
monthly income increases by 26 percent. Further, the program appears to have been a highly
productive safety net. Program participation significantly increased the likelihood of enterprise
creation and investments in homes and, in some cases, existing businesses. Positive impacts are
also observed on beneficiary households’ asset accumulation in terms of small livestock assets.
In addition, program participation impacted human capital through increased utilization of
health services. Impacts were much stronger for households facing greater constraints to
economic opportunity at baseline, i.e. households in rural areas and households with low
education levels. These results demonstrate that public works have considerable potential as
‘productive’ safety nets in post-conflict settings; they can provide not just immediate income
support during adverse economic shocks but also open avenues for investments in the
productive capacity of poor households.
JEL Classification: H53, I31, I38, O15
Keywords: Public works, safety nets, social protection, impact evaluation, Sierra Leone
Acknowledgements: This work has been done in collaboration with Sierra Leone’s National
Commission for Social Action and the Ministry of Finance and Economic Development’s
Integrated Project Administration Unit. We thank Statistics Sierra Leone, who conducted the
data collection, and the impact evaluation field coordinators Samantha Zaldívar and Joshua
McCann, for ensuring high quality data. The paper benefited from support and guidance from
Suleiman Namara and John Van Dyck. We thank Deon Filmer and Patrick Premand for
discussions and comments that improved the paper.
1
2
Social Protection Specialist, World Bank, email: [email protected]
Senior Economist, World Bank, email: [email protected]
1
I.
Introduction
Within the anti-poverty toolkit, public works (PW) are currently in fashion. The World Bank
funded public-works programs in 24 countries between 2007 and 2009, and a number of
governments have introduced their own initiatives (Zimmerman 2013). This popularity reflects
a growing recognition of the development potential of such programs. In particular, PWs seem
to have evolved into flexible instruments that can be fashioned to serve multiple objectives
including providing safety nets while also expanding economic opportunities. Often, the choice
in favor PWs as opposed to other, cheaper, anti-poverty instruments rests on precisely the
justification that they can both reduce poverty while contributing to growth. But is this
optimism justified? Can PWs really do it all?
A recent review (Subbarao et al 2013) lists four major ways in which PWs are being used: (i)
mitigation of covariate shocks; (ii) mitigation of idiosyncratic shocks in response to a temporary
or structural job crisis; (iii) as a bridge to more permanent employment; and (iv) poverty relief.
As this list suggests, PWs can, at least theoretically, be deployed both for creating resilience
against shocks and for generating opportunities to escape poverty and underemployment.
The use of PWs as something more ambitious than mere crisis response marks a shift in their
perceived role in public policy3. The theoretical underpinnings of this shift are straightforward.
To the extent that PW programs provide additional sources of employment and income for the
underemployed, even in the absence of adverse shocks, they help generate larger and more
predictable incomes for the poor. By doing this, such programs can directly and positively
impact household welfare through increased consumption, increased investment in human
capital, increased investment in productive assets, and expanded opportunities to engage in
higher risk, higher return activities. These programs therefore have the potential to facilitate
transitions from low to higher productivity activities among poor households.
However, empirical evidence around these theorized multi-faceted impacts is limited. While
there are a handful of studies that show positive impacts of PWs on household income and
expenditures, direct evidence on their productive potential is extremely limited4. Given signs of
growing adoption and the ambitious scope of some recent programs,5 this absence of clear
empirical evidence – on PWs hypothesized productive impacts beyond mere income support –
is a matter of urgent concern, especially since these programs are costly. Not only does their
execution require significant administrative effort which can be a strain on government capacity
especially in developing countries, they can also potentially have distortionary impacts on local
labor markets. Further, the credibility of such programs can be low given the scope for
corruption and mismanagement if not designed and implemented properly. Therefore, now
more than ever, it is important to understand whether PWs are a good public policy choice for
fiscally strapped developing countries. The use of PWs demands caution – used well they have
the potential to transform lives but used badly they can potentially lead to tremendous waste of
public resources.
See Zimmerman (2013) for broader discussion.
See Subbarao et al (2013) for a comprehensive review of evaluations of Public Works programs.
5 One example is India's Mahatma Gandhi National Rural Employment Guarantee scheme which is
reaching approximately 56 million households.
3
4
2
This paper contributes to this critical area of inquiry. We provide rigorous evidence on how
precisely PWs impact household welfare on a number of dimensions in the very short run.
Specifically, we examine the short-term causal impacts of a PW program in Sierra Leone on
household welfare – with a particular focus on how participation impacts household
consumption, access to services, and investments in productive assets.
Evaluating the causal impacts of PWs in a post-conflict setting like Sierra Leone is another
important contribution of this paper. PWs are considered particularly suited for post-conflict
and other fragile contexts. This is because they can provide immediate short-term employment
to poor households, which have most likely faced tremendous deprivation during the conflict.
This employment also helps address youth unemployment and ex-combatant reintegration,
which represent pressing concerns for post-conflict recovery. An added bonus is that PW
projects can be designed to rebuild infrastructure damaged during the conflict. Due to these
features, PW programs have been quickly launched and scaled up following conflicts in GuineaBissau, Liberia, Rwanda, and Sudan. In Nepal, which recently emerged from a decade long
internal conflict, a national public works program is now being designed (Subbarao et al 2013).
However, once again, the evidence of the effectiveness of PW programs in post-conflict settings
is extremely limited This is an issue given that such programs are a costly gamble for countries
in the middle of post-conflict reconstruction. Also, these programs could overstretch the already
weak institutional capacity in these countries with effective delivery being an important
concern against a backdrop of poor governance, political instability, and potential for a return to
civil unrest.
Another major contribution of this work is the use – for the first time to our knowledge – of a
randomized control trial approach to rigorously measure the causal impacts of a large PW
program. It has been argued that evaluations of PWs have often lacked credible identification of
causal impacts (Zimmerman 2013). Most existing evaluations rely on non-experimental – and in
some cases – quasi experimental methods. In contrast, we exploit the phase-in design of a large
PW program to randomize targeted communities into control and treatment groups to measure
the causal impacts of the program in the 3-4 months following program delivery in treatment
communities and before program initiation in control communities.
We find that in the very short run PW programs implemented in a post-conflict setting can have
dramatic welfare impacts for households, especially those with constrained economic
opportunities. As expected, the impacts stem from strong increases in household economic
activity, but these increases go beyond mere program participation effects. Participation in the
PWs appears to have a multiplier effect within the household and crowds-in labor market work
for non-participating household members. These enhanced levels of economic activity are also
reflected in higher rates of female labor force participation and migration (both in and out) in
treated households. All of this translates into higher household incomes for treated households
– on average, the total value of reported cash and in-kind payments received by household for
work in the previous month increases by 26 percent.
This increased income has implications for household consumption, savings, and investment
patterns. Households use at least part of the increased income to improve their quality of life by
spending more on welfare enhancing goods and services. Treated households increase their
consumption in important categories like food and hygiene products. Expenditure on welfare
3
enhancing services, particularly health services also increases. Treated households reported
more frequent visits to health facilities and increased spending on drugs and medications.
A key result of the paper relates to evidence on the productive potential of PW programs. These
households exhibit significantly higher investments in home improvement and existing
businesses. They also invest more in small livestock assets. However, the biggest boost to poor
households’ productive potential comes from new businesses. Treated households are nearly
four times more likely to set up a new household enterprise than control households. Taken
together, these results show that PWs can successfully boost economic opportunities for
beneficiary households.
One critical finding of the paper is that PW impacts are almost entirely confined to households
in rural areas and with low education. In many ways this is intuitive – these programs are
typically designed to benefit households that rely predominantly on casual labor for income and
experience seasonal shocks.
The remaining paper is organized as follows. Section II provides an overview of the Sierra Leone
Cash for Work (CfW) program; Section III describes the evaluation methodology, including a
description of data sources. Section IV provides descriptive analysis, Section V presents the
main results of the evaluation, and Section VI concludes.
II.
Program Design and Implementation
In 2010, the Government of Sierra Leone launched the Youth Employment Support Project
(YESP) with support from the World Bank. The project included a labor-intensive public works
component, known as the Cash-for-Work (CfW) program, whose objective was to provide
additional income and temporary employment opportunities to vulnerable youth in the country.
The CfW program was targeted at individuals in the age group 15 through 35 in poor and
vulnerable communities. Program beneficiaries were selected through a three-stage process: (i)
geographical targeting6 to identify the beneficiary communities, (ii) submission of requests by
communities to receive program funds for a sub-project7, and (iii) community-based targeting
to identify beneficiary households within selected communities.
The National Commission for Social Action (NaCSA), which is a semi-autonomous government
agency, had the overall responsibility for implementation of the program. At the local level, the
program was implemented by independent contractors hired by NaCSA. These contractors were
responsible for managing the day-to-day implementation of the sub-project, including procuring
necessary materials and other inputs, recording attendance of workers, and making payments
to beneficiaries for days worked.
For targeting of individuals within communities, Community Oversight Committees (COCs)
were set up. These COCs were responsible for identifying the poorest households with at least
one member willing and able to work between the ages of 15-35 years. COCs relied on local
Geographic targeting was undertaken based on poverty and food-security estimates
The communities are given a list, or positive menu, of eligible sub-project types, namely: (i) feeder road
rehabilitation and maintenance; (ii) agriculture; and (iii) renewable energy and environmental
mitigation. A subset of communities submitting requests are then selected based on whether the subproject requested conforms to several requirements, including being on the positive list, suitability of the
sub-project for the locality, and community endorsement.
6
7
4
definitions of poverty for selecting beneficiaries. The COCs, were also tasked with monitoring
the progress of works and payments and resolving CfW-related disputes.
The wage rate within the program was set lower than the market wage to discourage non-poor
applicants from participating. The wage rate was 7,500 Leones, which was equivalent to
approximately US$1.8 at the time of the evaluation in 2012. On average, 60 percent of subprojects costs were allocated toward payment of wages, while the remaining 40 percent
covered contractor fees, materials to implement the works, and administrative costs. In addition
to these costs, NaCSA incurred operational costs of 10 to 15 percent of sub-project costs for
implementation support and monitoring.
Program roll-out took place through a randomized phase-in strategy. In early 2012, 276
communities were identified as potential recipients of the CfW intervention8. These were
randomly divided into two groups: a treatment group (143 communities) which was scheduled
to receive the CfW program during the evaluation period (April-Aug, 2012) and a control group
(133 communities) which receive the program during the evaluation period. As per the phasein, the control group was scheduled to receive the program about four months after the
treatment group. The design is summarized in Figure 1.
One key feature of program implementation was the practice of rotation. It is not uncommon for
PW programs in Sierra Leone (and elsewhere) to impose ‘rotation’ informally. This is based on
the principle that every eligible and willing individual in the community should have an
opportunity to participate in the program. According to Subbarao et al (2013), rotation systems
are extremely common in PW implementation across the world, wherever demand for
employment exceeds the opportunities created, to give the largest number of poor people a
chance to work. This system plays an especially crucial role in post-conflict settings, where
considerations of fairness are central concerns within program implementation. However,
rotation is often imposed in an informal and ad-hoc fashion – which makes it difficult to
measure and document. Also, if not incorporated adequately into operational procedures,
rotation can increase the risk of leakages.
III.
Empirical strategy and Data
A. Empirical Strategy
Given the randomized phase-in of the program, we rely on a phased-in randomized control trial
(RCT) methodology to establish causal impacts of the CfW program. The advantage of phased-in
RCT design is that a simple comparison of outcomes for the two randomly created groups yields
an unbiased estimate of the impact of the CfW program.
As mentioned earlier, the overall evaluation sample comes from 276 communities – 143
treatment and 133 control. The population for the study includes 17,608 beneficiary households
(8,944 in the treatment group and 8,664 in the control). Note however that because control
communities received the program approximately four months after the treatment
communities, therefore, the evaluation allows us to examine the impacts of the program only in
the very short-run (approximately three-four months).
The overall CfW program included 470 sub-projects over four wavess, with 108, 143, 133, and 86 subprojects in the first, second, third, and fourth waves, respectively. Randomized phase-in was carried out
over the second and third waves of the program; as such, the discussion in this paper refers to those two
rounds only.
8
5
In analyzing program impacts we undertake two sets of analysis. First, since all the surveyed
households had been selected to participate in the program, we conducted intent-to-treat (ITT)
analysis. We estimated the impacts of the program on various household-level outcomes of
interest using ordinary least squares regression. The regression to obtain the ITT estimates is
the following:
Yi = α + βTreat + ϵi, ,
where Treat is a dummy for assignment to the treatment group (i.e., equals 1 if a household
belongs to a community that was randomly assigned to the treatment group and 0 if assigned to
the control group).
Despite careful efforts to ensure adherence to randomized assignment, there is a possibility of
non-compliance, i.e. households that may have initially signed up for the program subsequently
did not participate; or some of the control beneficiaries might have participated in the program.
Household surveys show very low levels of non-compliance - around 1.8 percent of treatment
households report not participating and 7.4 percent of control households report participating.
Despite very low levels of non-compliance, for robustness, we also estimate the effect of
treatment on those whose treatment status was affected by the random assignment (i.e.,
compliers), which is known as the local average treatment effect (LATE)9. These estimates
the instrumental variable (IV estimates) of β in the equation above, using the dummy
random assignment as an instrument for treatment.
the
the
are
for
B. Data
One challenge in collecting data for the evaluation was that PW beneficiaries within
communities – people who would be working on specific CfW sub-projects –were not preidentified. Instead they were selected on the first day of the sub-project, largely on a first-come
first-serve basis subject to eligibility criteria being met. This precluded the possibility of
simultaneous baseline data collection in control and treatment households.
Data was collected in three phases – (i) data collection for beneficiary tracking; (ii) data
collection through unannounced site visits at CfW projects in treatment communities; and (iii)
endline household survey.
To enable documentation and follow-up of treatment and control households, a beneficiary
tracker survey was administered on the first day of implementation for treatment (beneficiary
tracker administered in April 2012) and control (beneficiary administered in July 2012) CfW
sub-projects. This tracker was the basis of which treatment and control households were
identified and interviewed for the endline survey (July-Aug 2012).
The endline household survey coincided with the start of CfW implementation in control subprojects, so that even though the households in control sub-projects were identified they still
had not started working. Endline household survey was implemented simultaneously in
treatment and control households before control households received treatment.
The content of survey data is summarized below:
9
See Angrist and Imbens (1994, 1995).
6
• The beneficiary tracker surveys collected basic demographic information from
beneficiaries, contact details for tracking them in subsequent survey rounds, and
information on their perceptions of the program and its processes prior to beginning work
on the sites. This survey was administered at 276 sites (April 2012 for treatment subprojects; July 2012 for control sub-projects) to a total of 17,670 beneficiaries on the first
day of sub-project implementation.
• The unannounced observational visits to treatment sub-project sites midway through the
physical implementation of the sub-projects (May-June 2012). The unannounced site visit
survey was administered at 141 sites of the 143 original treatment sites10. This survey was
a two-part survey designed to collect information on overall program implementation and
specific processes, as well as beneficiary knowledge of and satisfaction with the program.
Part 1 consisted of observations (e.g., worker roll call, checks of attendance and other
records) and an interview with a contractor representative on site; Part 2 consisted of an
interview of two (one male and one female) randomly selected beneficiaries. A total of 279
beneficiaries (3 percent percent of the 8,883 working on the 141 sub-projects) were
interviewed, around half of whom were female.
• The endline household survey was administered concurrently to treatment and control
households (July-August 2012) - at the end of implementation for the treatment group and
the start of implementation for the control group. This survey was administered to an
average of 20 beneficiary households from each of 275 sub-projects, as one sub-project in
the Western area was cancelled during the IE implementation, for a total of 5,506
beneficiary households. The survey covered a range of topics, but focused on measuring
program effects along the following dimensions: (a) labor market outcomes and economic
activity; (b) household assets, consumption and savings levels; and (c) utilization of
education and health services.
All data collection was carried out by Sierra Leone’s national statistical agency, Statistics Sierra
Leone. The administrative data was collected and maintained by NaCSA for operational and
monitoring purposes.
C. Measurement issues and possible threats to identification
We do not consider the absence of detailed baseline data across treatment and control
communities to be a serious limitation to causal identification. Power calculations suggest that a
randomized cluster design (clustered at the community (CfW sub-project) level) with 276
communities is sufficient to ensure balance between treatment and control households thereby
ensuring comparability of the two groups for causal attribution.
Nonetheless we are able to present some – admittedly limited – evidence of pre-intervention
balance between treatment and control using those variables collected during endline surveys
that were not expected to change over the program’s short duration. As shown in Table 1, the
overall treatment group is statistically comparable to the control group for most of these
variables tested.
Another measurement issue relates to the practice of rotation (described in Section II) which
was applied in an informal and largely undocumented way in treatment communities. This has
10
Two treatment sites were found to be inactive at the time of unannounced visit.
7
implications for measurement of program impacts because we measure program impacts
through ‘registered beneficiaries’ who are identified on the first day of sub-project
implementation. If rotation occurs, the impacts captured by the IE may be under- or overestimated. For instance, registered beneficiaries may be more informed and better-networked
than non-registered beneficiaries, which may lead them to have better outcomes even in the
absence of the program, leading the IE to over-estimate the program effects. Conversely, nonregistered beneficiaries may be poorer and may have a higher return to receiving the transfer
(for instance through lower substitution effects), in which case the IE may be under-estimating
the program’s effects.
Data collected during the unannounced site visits indicates that on average, 13.1 percent of
beneficiaries on site at the time of the visits were not registered beneficiaries. Similarly,
timesheet records show that on average, 13.6 percent of the workers listed on the timesheets
were not registered beneficiaries. On average there are 65 registered beneficiaries per subproject and 10 non-registered beneficiaries listed on timesheets. While this is not definitive
evidence of rotation, as these non-registered beneficiaries may have simply been alternates (i.e.,
replacements from the same household who are nominated to work in the event of absence by
the beneficiary), at least a portion of these additional workers are likely to come from
households not captured by the IE. Discussions with administrative staff confirm this practice
took place at least in some sites, despite attempts to prevent it. Beyond the potentially over- or
under-stated program impacts captured by the IE, the existence of rotational practices would
also imply that the actual program impacts may have been diluted, as the same total transfer
amount was distributed across a larger number of beneficiaries.
IV.
Descriptive analysis
A. Sub-project characteristics
The CfW sub-projects have national coverage and are spread fairly evenly across urban and
rural areas. Figure a shows the geographical distribution of all the sub-projects covered under
the evaluation; Figure b shows the precise location of the treatment sites based on GPS data.11
The most common types of sub-projects are feeder road rehabilitation projects (67 percent),
inland valley swamp rice (9 percent), and other agriculture (13 percent). The average labor
intensity of the sub-projects is 60 percent, but there is some variance, with roads sub-projects
typically lower in labor intensity.
B. Beneficiary characteristics
The data indicate that CfW program primarily reaches youth with low levels of education
working in the agricultural sector. The program also met its gender targets and did not induce
negative impacts on schooling for its young participants.
The average age among beneficiaries is 27 and 92 percent of beneficiaries are between the ages
of 15 and 35, and hence fall within the eligible age group for the CfW program. The program
design emphasized adequate participation (minimum 30 percent) of female beneficiaries and
this goal seems to have been met – the average female share of beneficiaries is 33 percent.
Female participation varies by sub-project type, with the share of female beneficiaries is highest
among IVS rice sub-projects (43 percent) and lowest among roads sub-projects (30 percent).
11
GPS data was only collected for treatment sites.
8
Nearly half of beneficiaries (49 percent) worked in the agricultural sector as their main
occupation outside of the CfW program. Other top occupation categories outside of the program
included: students (15 percent), street and related sales and services (10 percent), and building
and related trades workers (4 percent). More than half (56 percent) of beneficiaries had not
engaged in any remunerated work in the month prior to participating in the program. Among
the 44 percent who had paid work outside the program, average daily earnings were $2.5 and
the majority (75 percent) was self-employed. Almost half (46 percent) of the beneficiaries
reported that they engaged in unpaid family farm work.
Education levels are fairly low among CfW beneficiaries – 52 percent have less than primary
education (i.e., incomplete primary or no schooling), 35 percent have completed primary, and
only 12 percent have completed secondary or above. The average level of education is lower
among female beneficiaries. Nearly 60 percent of females have no education compared to about
35 percent for males.
One potential concern with public works programs targeted to youth is that they may attract
youth that would otherwise be attending school, leading them to drop out. Our data suggest that
for the most part, the program does not lead youth to drop out of school to enter the program:
only 3 percent of beneficiaries who had ever attended school reported that they had stopped
attending school to enter the CfW program.
V.
Results
Our results suggest the program induced significant welfare impacts in the very short run. For
households with limited economic opportunities at baseline, these impacts manifest along so
many different dimensions that CfW participation appears to be fairly transformative – at least
in the immediate aftermath. Specifically, we find strong impacts not just on short-run economic
activity, income, and consumption patterns but also on the overall productive potential with
implications for household resilience and opportunity in the longer run. The impacts vary
systematically along key household characteristics like rural/urban location and education level
of household head. Less impact heterogeneity is seen across gender of participant within CfW.
As mentioned above (see Section III.C), data from unannounced site visits as well as discussions
with administrative staff indicate that rotation took place in at least some of CfW sites. This
implies that in that there is low likelihood that any given household participated in the program
for its entire duration, indicating that estimated program impacts might be lower bounds.
A. Impacts on economic activity
As anticipated, the CfW program has significant impacts on overall household economic activity
(see Table 2.1). Also, even in the very short run, we observe strong spillover economic activity
impacts on non-participating household members. Overall, we find that treatment households
are significantly more likely to be engaging in remunerated work. Household members in
treatment households are 34 percent more likely to have had paid work in the last 12 months.
Participation in CfW leads to a net increase in the household labor market participation - share
of household members working for cash is significantly higher for treatment households.
9
Increase in levels of household labor market activity is more pronounced among rural
households and households where the head has lower levels of education (see Table 2.2 and
2.4); suggesting these effects are stronger for households whose labor market opportunities
were more constrained at baseline. This most likely signals that the program especially benefits
households that rely on casual labor. The relatively higher impacts in these households also
supports the notion that more effective targeting would enhance program impacts. It also
signals that relying heavily on self-targeting for such programs may not be effective in contexts
with widespread underemployment even among the better off.
Given the net increase in household economic activity, program participation by one household
member is not crowding out other household members from the labor market. This is true for
both male and female beneficiary households (see Table 2.3). On the whole, employment of nonparticipating household members doesn’t just remain constant, it increases – much more so
when participating household member is male (see Table 2.3). However, labor market impacts
on non-participating household members in beneficiary households are substantively different
depending on household characteristics – location and education level of household head. In
rural areas both male and female non-participating members register more employment; in
urban areas its mostly non-participating males (and not females) who register more
employment.
Therefore one noteworthy impact of the CfW program is greater female labor market
participation in participating households – both as primary beneficiaries and also as indirect
beneficiaries within participating households. As above, these impacts are more pronounced in
households with constrained economic opportunities and higher reliance on casual labor –
those in rural areas and those where household heads have low levels of education (see Table
2.4). It is worth noting that there is no evidence of short term impacts on incidence of child
labor. This result is robust to rural/urban interactions and interactions by education level of
household head.
One interesting auxiliary impact of program participation is increase in reported migration for
treated households (see Table 2a). An increase in both in (individuals moving into the
household from other town/city) and out (individuals moving out of the household to other
town/city) is observed. These impacts are almost exclusively concentrated in rural households
and households with low education levels of household heads. Intuitively, increased migration
seems like a natural outcome of increased economic potential in areas where job opportunities
are scarce.
B. Impacts on household income, expenditure, savings, and investment
B.1 Income effects
Given the strong impacts CfW participation has on household economic activity, it is not
surprising to find clear income effects. Treated households have higher reported income - on
average, the total value of reported cash and in-kind payments received by household for work
in the previous month increases by 26 percent (see Table 3.1).
In line with the results on economic activity, rural households exhibit stronger income effects
than their urban counterparts. On the other hand, income effects are stronger for households
where the head has at least some education (see Table 3.4).
10
Interestingly, the average increase in cash income reported by treatment households
(approximately 41,100 Leones) is less than one-third of what these households were entitled to
receive under the program over the period under analysis. Based on impacts on household
economic activity, it is clear that CfW participation does not crowd-out employment of nonparticipating household members – in fact in some cases crowds-in more employment. As such,
it is clear that this discrepancy cannot be attributed to intra-household substitution effects in
terms of labor market participation (i.e., non-participants within the household reducing the
amount of paid work outside the program upon entering the program). And given the low
percentage of beneficiaries who were doing any paid work before the program was introduced,
individual substitution effects among CfW participants are likely to be very low.12 Even among
those who were not working prior to the program – those for whom the substitution effect
should have been zero – the average increase represented only 40 percent of the intended
transfer amount, suggesting this discrepancy cannot be explained by substitution effects alone.
Instead, this finding is more likely to be linked to the practice of rotation or to leakages. As
mentioned above, rotation is a practice in which a sub-project accepts more beneficiaries
working fewer days than it is designed to, typically because socially it is perceived as more fair.
This had been a common practice in previous rounds, and although the data does not allow us to
ascertain the exact extent to which it occurred, it does provide an indication that it likely
rotation occurred. As with any program providing cash or in-kind transfers, there is also the
possibility of leakages, or money reaching individuals who are not the intended beneficiaries.
While there is no concrete evidence of payment leakages found by the IE, the program
implementers have acknowledged that the payment arrangement in place during the IE left a
high risk of leakage13.
B.2 Impacts on consumption
B.2.1 Consumption goods
The efficacy of Public Works programs as anti-poverty instruments depends in part on how
their short-run income effects impact household spending and consumption patterns. We
examine this question in the very short run, but unfortunately, the time-frame of the evaluation
prohibits us from looking at these patterns in the longer term. Nonetheless, the very short run
impacts are suggestive and can provide important insights on potentially longer term dynamics.
For consumptions goods, a short consumption module was administered to capture program
impacts on key expenditure categories, which included: utilities (e.g., water, electricity, fuel,
phones), food, children’s schooling, hygiene, home improvements, transfers out to nonhousehold members, and some temptation goods. Results shows that program has positive
impacts on beneficiary households’ spending on food, in line with the CfW design as a
mechanism to support beneficiary households in meeting their food needs, particularly in the
face of rising food prices (see Table 4.1). Treatment households reported spending 8 percent
more on food in the past month than control households.14 Not surprisingly, the most significant
increases in food consumption are registered by households where the household head has less
These cannot be accurately measured due to rotation practices and its implications for attendance
tracking.
13 Wherein contractors were responsible for making payments to beneficiaries.
14 Sierra Leone Youth Employment Support Project Emergency Project Paper (June 2010).
12
11
than primary education. Household expenditures on hygiene products also increases by 15
percent.15
As expected, impacts on household consumption patterns appear to vary systematically across
key baseline characteristics – location, gender of the primary participant, and education level of
household head. In general, rural households register stronger consumption impacts. This is
expected given that compared to other urban counterparts; rural households register stronger
labor market participation impacts (greater paid employment for both male and female nonparticipating members in beneficiary households) and in turn stronger income effects.
Specifically, categories in which rural households register positive impacts, but urban
households do not, include: public transport (potentially linked to migration, see Section V.A)
and hygiene products.
On average, we do not see a systematic increase in spending on cigarettes/ tobacco, festivities,
or inter-household transfers. The overall patterns suggest that in the short run increased
income leads, in part, to increased spending on welfare enhancing goods and services (see
section B.2.2 and B.2.3 below).
B.2.2 Health Services
Program participation produces positive impacts on beneficiary households’ utilization of
health services, particularly for young male children. These results are summarized in Table 5.
On average, treated households reported 12 percent more visits to health facilities than the
control group and the proportion of boys aged 0 to 5 who were taken to a health facility when
sick was 9 percent higher in treatment households. If we consider all boys ages 0 to 5
irrespective of their health status at the time they were taken to the doctor, the increase is even
higher (23 percent).
In addition, program participation leads to increased spending on drugs and medications.
Treated households reported spending 18 percent more on drugs and medications in the
previous month than the control group. These results are in line with the finding of increased
utilization of health services. Increase in health-seeking behavior manifests for both rural and
urban households and irrespective of whether the participating household member is male or
female. However, these impacts are stronger for households where household head has less
than primary education, due most likely to lower baseline values.
These findings are particularly interesting as the intervention did not have any specific design
aspects intended to encourage more health-seeking behavior. Since part of the transfer was
spent on health, this suggests that the transfer may have relieved a financial constraint that
beneficiary households faced in accessing health services.
B.2.3 Education services
In contrast to program impacts on access to health services, CfW participation has no systematic
impact on access to education (see Table 6). In fact, student absenteeism appears to increase in
treatment households. Increase in student absenteeism is stronger when program participant is
male household member, for rural households, and for households where head has less than
secondary education. Clearly, student absenteeism increases most significantly for households
15 We also find some puzzling decreases in the household spending on fuel (19 percent) for lighting,
heating, and cooking.
12
that experience greatest increases in economic activity among adults. One possible explanation
for these results is that for households where labor market participation of adults increases
significantly, including for non-participating household members (see section V.A), school-going
children might be pulled into helping with household chores or taking care of younger siblings,
leading to increased absenteeism. Note however that at least in the short run, treated
households do not report lower school enrollment for children; just lower school attendance. It
is difficult to speculate on the longer term implications of these short-term effects.
B.3 Impacts on savings and investment
Based on the section above, it is clear that part of the income gains from CfW participation are
being used for welfare-enhancing consumption (e.g. food, hygiene products) especially for rural
households and households with low education. However, the longer-term poverty alleviation
potential of this instrument is directly linked to the extent to which CfW income is used to
enhance the productive potential and overall resilience of targeted households. It is on this
dimension that we see some of the most interesting program impacts. This is despite the fact
that the CfW program’s emphasis was on temporary employment rather than on boosting
productive capacity of poor households.
First, in terms of savings we see some positive impacts. For treated households participation in
informal savings groups (osusus) increased by 16 percent (see Table 7.1), however there are no
discernable impacts for formal savings (as in likelihood of having a savings account).
Participation in informal savings groups increases more markedly for rural households and for
households where the head has less than primary education. In fact for the latter group, there
appears to be some substitution with households opting out of formal savings accounts and into
informal savings groups. However, the total monetary value of reported savings does not appear
to be systematically higher in treated and control households, although we find marginal
increases in rural households.
Various types of investments are examined. Treated households have significantly higher
investments in home improvement (by 33 percent). They also invest more in small livestock
assets.16 Treatment households are 34 percent more likely to own goats or pigs and the number
of poultry owned is 26 percent higher than control households. In addition, in rural households
we also see higher investments in existing businesses. These findings validate the argument of
PW as promoting resilience against shocks. Not surprisingly, these impacts are largely confined
to rural households and households with low levels of education, which experience stronger
employment and income effects.
One of the starkest impacts is in terms of new businesses. Treatment households are nearly 4
times more likely to set up a new enterprise than control households. Only 8.9 percent of
control group households reported that someone in the household had set up a new enterprise
in the last 3 months, compared to 33.6 percent in the treatment group. Interestingly, likelihood
of starting a new business is stronger for households where participating household member in
the CfW program is female. Once again, the likelihood of starting new business is stronger for
rural households and households with low levels of education.
Households in the treatment group are also more likely to own a motorbike – 15% in treatment
households compared to 10% control households. This points to the fact that the program is not very
well-targeted, as nationally representative data from the 2011 SLIHS less than 5% of poor households
own motorbikes.
16
13
C. Impacts on social cohesion
Safety net programs are sometimes thought of as providing a mechanism through which social
cohesion can be promoted by including groups that tend to be marginalized (such as the poor
and vulnerable), but a recent review of public works programs indicates that the evidence of
this is quite limited (Andrews et al 2013). In the context of Sierra Leone, disenfranchised youth
as considered to be a particular threat to social cohesion, as many youth were engaged in the
violent civil conflict which ended in 2002. With this in mind and given the CfW program’s focus
on youth from poor households, we try to capture the program’s effects on social cohesion.
We find that the program had positive impacts on family trust and cohesion (see Table 8).
Treated households were more likely to report having high trust in other household members
and extended family (10% and 40% higher than the control group, respectively).
The effects of the program on social cohesion are more ambiguous. Treated households were
also more likely to report high trust in people of the same religion (24% higher).17 However,
they were less likely than the control group to report high trust in someone from the same or a
different ethnicity (30% and 27%, respectively). Factors behind these results are being
explored.
VI.
Conclusions
Public works have become a popular policy instrument for protecting the poor from income
shocks. They have the added perceived benefit of creating useful public goods or services for the
communities, making them particularly attractive for post-conflict societies. However,
increasingly their development effectiveness is being judged, not just by the degree of
stabilization they provide during adverse economic shocks but also by their ability to improve
the overall productivity of beneficiary households through investments in productive assets and
human capital development.
Using this lens we provide evidence around the very short term effectiveness a Cash for Work
program in Sierra Leone targeted at unemployed youth in poor and vulnerable communities to
mitigate impacts from the global food, fuel, and financial crises. The phase-in implementation of
the program was exploited to implement a community-level randomized control trial which
helps measure the causal impacts of the program on household outcomes over a period of three
to four months.
We find that, as per design, the program succeeded in attracting young people (ages 15-35) with
low levels of education who were predominantly working in the agricultural sector. Further, it
was successful in impacting poor households through the anticipated employment channels. We
find that treatment households are significantly more likely to be engaging in remunerated
work. In fact, participation in the PWs appears to have a multiplier effect within the household
and crowds-in labor market work for non-participating household members.
These employment impacts manifest in household income, which increases by 26 percent for
treatment households. Part of the increased income is spent on food, health services, and
hygiene products. We also find positive impacts on beneficiary households’ utilization of health
services, particularly for young male children. However, we do not see any corresponding
17
There were no effects found on trust in people of a different religion or from the same community.
14
increases in access to education. In fact, within the short time frame of the study, rates of school
absenteeism are higher among children in treatment households. Note however that we do not
see an increase in children’s participation in the labor market, suggesting that increased student
absenteeism rates are most likely linked to household chores and child-minding. This, along
with results on non-participating adult household members above, suggests that PWs can
significantly alter household time allocations.
Beyond these immediate welfare impacts, the program appears to have been a highly
productive safety net for beneficiary households. Participation in the CfW program significantly
increased the likelihood of enterprise creation for households. Further, program participation
also boosted participation in informal savings groups. However, we do not find a corresponding
increase in the amount of household savings reported. We also find that the CfW program has
positive impacts on the beneficiary households’ asset accumulation in terms of small livestock
assets. Investment in homes and existing businesses also increase.
However, almost all documented impacts are strongly mediated by household characteristics at
baseline; specifically, impacts appear to be strongest for households that at baseline were most
constrained in terms of economic opportunities. These include rural households and
households where household heads have little or no education.
Clearly therefore PWs have the potential to unleash dramatic, perhaps even transformative,
impacts on households. Even in the fragile, post-conflict context of Sierra Leone, the CfW
program catalyzed strong impacts on nearly all key aspects of household welfare and decisionmaking – employment, income, consumption, savings, investment, health and education. In
addition, if targeted properly, PWs can be highly pro-poor - expanding opportunities where they
are most constrained and the margins are greatest.
This paper shows that PWs can in fact do it all; even in a fragile setting they have the potential to
not just build resilience, but also provide opportunity to households caught in the trap of
underemployment and poverty. Even when they are designed as crisis response, PWs can
nonetheless provide households pathways towards higher productivity. While it is critical to get
design and implementation right, the optimism surrounding PWs for reconstruction of postconflict communities is not entirely unjustified.
15
References
Adams, L. and E. Kebede. 2005. “Breaking the Poverty Cycle: A Case Study of Cash Interventions
in Ethiopia.” HPG Background Paper. London, UK: Overseas Development Institute (ODI) and
Humanitarian Policy Group (HPG).
Alderman, H., and R. Yemtsov, 2012. ‘ Productive role of safety nets : background paper for the
World Bank 2012-2022 social protection and labor strategy’ Social Protection Discussion
Papers 67609. Washington DC: World Bank.
Andrabi, T. and Jishnu Das (2010). In Aid we Trust.
Andrews, C., & Kryeziu, A. (2013). Public Works and the Jobs Agenda: Pathways for Social
Cohesion. Background paper for the WDR.
Angrist, J. D., and G. Imbens (1994). “Identification and Estimation of Local Average Treatment
Effects," Econometrica, 62(2), 467{475.
Angrist, J. D., and G. Imbens (1995). “Two Stage Least Squares Estimation of Average Causal
Effects in Models with Variable Treatment Intensity," Journal of the American Statistical
Association, 90(430), 431-442.
Angrist, J. D., G. W. Imbens, and D. B. Rubin (1996). Identification of Causal Effects Using
Instrumental Variables," Journal of the American Statistical Association, 91, 444-455.
Azam, Mehtabul, and Basab Dasgupta. 2011. “The Impact of the Indian Job Guarantee Scheme on
Labor Market Outcomes: Evidence from a Natural Experiment.” Working Paper Series 6548,
World Bank, Washington, DC, and IZA, Bonn, Germany.
Berhane, Guush, John Hoddinott, Neha Kumar and Alemayehu Seyoum Taffesse. 2011. The
Impact of Ethiopia’s Productive Safety Nets and Household Asset Building Programme: 20062010. IFPRI. Processed. Bertrand Marianne, Sendhil Mullainathan and Douglas Miller. 2003.
Public Policy and Extended Families: Evidence from Pensions in South Africa World Bank
Economic Review 17(1): 27-50.
Departamento Nacional de Planeación (DNP). 2007. “Evaluación de impactos del Programa
Empleo
en
Acción”.
Available
at:
https://www.dnp.gov.co/Portals/0/archivos/
documentos/DEPP/Evaluacion_Politicas_Publicas/Serie_Evauacionl_Politicas_5_Evaluacion_Em
pleo_en_accion.pdf
Dutta, Puja, Rinku Murgai, Martin Ravallion, and Dominique van de Walle. 2012. “Does India’s
Employment Guarantee Scheme Guarantee Employment?” Policy Research Working Paper
6003, World Bank, Washington, DC.
Galasso, Emanuela and Martin Ravallion (2004). “Social Protection in a Crisis: Argentina’s Plan
Jefes y Jefas”, World Bank Economic Review, 18(3): 367-399.
Gilligan, D., J. Hoddinott, and A. Taffesse. 2008. "The Impact of Ethiopia's Productive Safety Net
Programme and its Linkages." Discussion Papers 839. Washington, DC: International Food
Policy Institute.
Hirway I., M. Saluja, and B. Yadav. 2009. Analysing Multiplier Impact of NREGA Works through
Village SAM Modelling. Annandale-on-Hudson, NY: The Levy Economics Institute of Bard
College. Draft.
McCord, A. and D. van Seventer. 2004. The Economy-wide Impacts of the Labour Intensification
of Infrastructure Expenditure in South Africa. Conference on African Development and Poverty
Reduction, the Macro--Micro- Linkages 13-15 October 2004, South Africa.
16
Rosas and Solbes (2013). “Assessing the Targeting of Sierra Leone’s Cash for Work Program:
Current Performance and Proposed Improvements.” Draft.
Subbarao, Kalanidhi, del Ninno, C., Andrews, C. and Rodriguez-Alas, C. 2012. “Public Works as a
Safety Net: Design, Evidence, and Implementation.” World Bank.
Zimmermann, L. (2013). Why guarantee employment? evidence from a large indian publicworks program. Working Paper.
World Bank Independent Evaluation Group. 2011. “Social Safety Nets: An Evaluation of World
Bank Support.”
17
Tables
Table 1: Balance tests
Variable
Urban, percent of households
District code
Number of children ages 6-14
Age of head
Head is female, percent of
households
Head ever attended school, percent
of households
Highest degree head attained
(categorical, 1-11)
Head did not complete primary,
percent of households
Distance to main water source,
minutes
*** p<0.01, ** p<0.05, * p<0.1
Mean in
treatment
Mean in
control
48.7
26.4
1.1
41.2
24.4
50.1
27.2
1.0
40.9
17.7
Statistically
significant
difference
No
No
No
No
Yes***
41.6
42.4
No
3.3
3.4
No
63.4
62.4
No
9.1
8.7
No
18
Table 2: Impacts on Economic Activity
Table 2.1 Impacts on Economic Activity - Overall
(1)
(2)
% of hh members
% of non-beneficiary
who worked for
hh members who
cash in last 12
worked for cash in
months.
last 12 months.
(3)
% females who
worked for cash in
last 12 months.
(4)
% non-beneficiary
females who
worked for cash in
last 12 months.
(5)
% children aged 614 who worked for
cash in last 12
months.
0.112
0.0544
0.110
[0.0123]***
[0.00932]***
[0.0145]***
Constant
0.332
0.224
0.352
[0.0101]***
[0.00739]***
[0.0116]***
Table 2.2 Impacts on Economic Activity - Urban-Rural
Treatment
0.123
0.0658
0.137
[0.0187]***
[0.0143]***
[0.0197]***
urban
-0.0164
0.00718
0.00850
[0.0201]
[0.0148]
[0.0230]
urban x treatment -0.0234
-0.0231
-0.0558
[0.0243]
[0.0186]
[0.0287]*
Constant
0.340
0.220
0.348
[0.0169]***
[0.0118]***
[0.0170]***
Table 2.3 Impacts on Economic Activity – Gender of Main Beneficiary
treatment
0.101
0.0513
0.0693
[0.0142]***
[0.0105]***
[0.0177]***
female
-0.0259
-0.0197
-0.0267
beneficiary
[0.0130]**
[0.00912]**
[0.0171]
0.0400
[0.0139]***
0.282
[0.0113]***
0.00644
[0.00813]
0.0391
[0.00503]***
0.0729
[0.0200]***
0.0298
[0.0224]
-0.0678
[0.0274]**
0.268
[0.0168]***
0.00816
[0.0125]
-0.0229
[0.00978]**
-0.00674
[0.0155]
0.0501
[0.00815]***
0.0290
[0.0177]
-0.218
[0.0140]***
0.00703
[0.00942]
-0.00234
[0.00934]
treatment x
female
beneficiary
Constant
0.0391
[0.0193]**
0.00239
[0.0140]
0.355
[0.0140]***
0.0395
[0.00571]***
0.0601
[0.0158]***
0.0465
[0.0198]**
0.00658
[0.0105]
-0.0204
[0.0116]*
Treatment
0.0291
[0.0161]*
0.0103
[0.0127]
0.110
[0.0213]***
0.340
0.228
0.360
[0.0118]***
[0.00830]***
[0.0139]***
Table 2.4 Impacts on Economic Activity - Education Level of Household Head
treatment
0.126
0.0649
0.128
[0.0143]***
[0.0111]***
[0.0171]***
primary
0.0528
0.0126
0.0489
completed
[0.0170]***
[0.0131]
[0.0207]**
secondary
completed
treatment x
primary
completed
treatment x
secondary
completed
Constant
Observations (N)
0.0120
[0.0179]
-0.0383
[0.0216]*
0.0310
[0.0143]**
-0.0121
[0.0174]
0.00545
[0.0254]
-0.0498
[0.0287]*
0.0376
[0.0249]
-0.0576
[0.0268]**
-0.0332
[0.0105]***
-0.00449
[0.0170]
-0.0373
[0.0226]
-0.0424
[0.0183]**
-0.0587
[0.0318]*
-0.0625
[0.0312]**
0.00393
[0.0158]
0.320
[0.0120]***
5,323
0.216
[0.00874]***
5,323
0.344
[0.0137]***
5,190
0.269
[0.0127]***
5,190
0.0479
[0.00668]***
3,110
19
Table 2.a: Impacts on Migration
Table 2.5 Impacts on Migration
Table 2.5.1 Impacts on Migration - Overall
(1)
Migration of Household member to other
town/city in last 3 mos.
Treatment
0.0570
[0.0128]***
Constant
0.126
[0.00828]***
Table 2.5.2 Impacts on Migration – Urban - Rural
Treatment
0.0611
[0.0171]***
urban
0.0268
(2)
Migration of Household member from
other town/city in last 3 mos.
0.0388
[0.0119]***
0.118
[0.00775]***
0.0361
[0.0166]**
-0.00746
urban x treatment
[0.0164]
-0.00774
[0.0155]
0.00537
Constant
[0.0254]
0.113
[0.0238]
0.121
[0.0113]***
Table 2.5.3 Impacts on Migration – Gender of Main Beneficiary
Treatment
0.0507
[0.0169]***
female beneficiary
-0.0311
[0.0162]*
treatment x female beneficiary
0.0207
[0.0237]
Constant
0.137
[0.0111]***
Table 2.5.4 Impacts on Migration – Education Level of Household Head
Treatment
0.0523
[0.0152]***
primary completed
0.0155
[0.0202]
secondary completed
0.0310
[0.0203]
treatment x primary completed
0.00546
[0.0288]
treatment x secondary completed
0.0271
[0.0337]
Constant
0.117
[0.0104]***
Observations (N)
5,311
[0.0113]***
0.0389
[0.0149]***
0.00795
[0.0141]
-0.000913
[0.0214]
0.115
[0.00971]***
0.0408
[0.0143]***
-0.0112
[0.0172]
0.000649
[0.0174]
0.00419
[0.0262]
-0.0127
[0.0298]
0.119
[0.00910]***
5,309
20
Table 3: Income Effects
Table 3.1: Impacts on household income – Overall
(1)
(2)
Total money hh received Total value of inin cash (past month)
kind payments
Treatment
41,114
-24,058
[18,092]**
[8,397]***
Constant
240,701
47,708
[16,146]***
[6,592]***
Table 3.2: Impacts on household income – Urban-Rural
Treatment
47,882
-16,911
[15,863]***
[8,639]*
Urban
58,576
21,974
[30,832]*
[13,021]*
urban x Treatment
-7,185
-14,025
[34,582]
[16,780]
Constant
210,149
36,681
[12,287]***
[8,047]***
Table 3.3: Impacts on household income – Gender of Main Beneficiary
Treatment
32,759
-27,318
[21,796]
[9,187]***
female beneficiary
-18,401
-8,310
[17,970]
[9,209]
Treatment x female beneficiary
20,112
11,113
[21,836]
[11,378]
Constant
247,278
49,799
[19,854]***
[7,981]***
Table 3.4: Impacts on household income – Education of Household Head
Treatment
29,302
-23,852
[23,257]
[9,519]**
primary completed
-5,776
-624.5
[24,643]
[13,050]
secondary completed
53,748
12,063
[24,693]**
[14,210]
Treatment x primary completed
14,633
6,987
[27,790]
[16,340]
Treatment x secondary completed
90,669
-13,104
[35,702]**
[18,630]
Constant
230,008
46,616
[21,783]***
[7,080]***
Observations (N)
4,415
5,298
VARIABLES
(3)
Total value of cash and in-kind
payments (past month)
58,832
[18,229]***
224,204
[15,452]***
84,376
[17,868]***
79,362
[30,089]***
-50,188
[35,763]
184,432
[14,147]***
47,004
[20,933]**
-21,829
[17,088]
30,917
[21,887]
230,993
[18,298]***
56,948
[21,134]***
16,472
[24,156]
82,371
[23,546]***
642.9
[27,848]
29,449
[34,601]
206,296
[18,441]***
5,323
21
Table 4: Impact on Consumption – Part 1
4.1: Impact on Consumption – Part 1 - Overall
(1)
(2)
(3)
(4)
les
Water per
Wood and
Fuel for
Electricity
cooking/
(past
month
similar
products18
lighting
month)
(past
/heating
month)
per month
ent
-806.2
3,684
-6,051
3,471
[1,055]
[3,279]
[3,228]*
[2,691]
nt
4,262
29,911
32,012
12,080
[772.8]*** [2,243]*** [2,589]*** [1,755]***
4.2: Impact on Consumption – Part 1 – Urban – Rural
ent
-278.9
2,854
-2,437
1,396
[647.0]
[2,912]
[2,204]
[1,094]
6,217
32,219
34,234
21,395
[1,443]*** [3,528]*** [4,311]*** [2,979]***
x Treatment
-894.6
2,459
-7,015
5,012
[2,012]
[5,195]
[5,382]
[4,607]
nt
1,150
13,856
15,139
1,424
[504.7]**
[2,314]*** [1,548]*** [350.7]***
4.3: Impact on Consumption - Gender of Main Beneficiary
ent
86.87
5,629
-5,143
3,543
[1,193]
[3,552]
[3,696]
[3,277]
beneficiary
565.2
1,200
-3,030
-2,367
[906.6]
[2,997]
[3,722]
[2,808]
ent x female -2,322
-5,141
-1,998
-823.5
iary
[1,320]*
[4,209]
[4,458]
[4,206]
nt
4,088
[725.9]***
18
29,691
[2,160]***
33,226
[2,722]***
12,992
[2,187]***
(5)
Electricity
per month
(6)
Food (past
one
month)
2,124
[2,035]
7,742
[1,224]***
21,967
[12,381]*
269,641
[9,121]***
460.8
[749.7]
14,710
[2,077]***
3,969
[3,597]
384.5
[301.7]
16,532
[13,982]
92,130
[16,465]**
*
14,941
[21,718]
222,889
[9,361]***
2,669
[2,369]
-2,505
[1,480]*
-959.8
[2,370]
8,537
[1,431]***
(7)
Fuel for
motor
vehicles
(past
month)
-2,862
[1,236]**
5,842
[1,109]***
(8)
Public
transport
(past
month)
(10)
Cigarettes
or tobacco
(past
month)
(11)
Hygiene
products
(past
month)
(12)
Household
supplies
(past 2
months)
(13)
Festivities
(past 2
months)
3,906
[3,872]
45,417
[3,143]***
(9)
Fixed or
mobile
phones
(past
month)
-2,213
[4,304]
36,699
[3,464]***
477.5
[982.5]
6,776
[822.4]***
3,942
[2,192]*
25,437
[1,671]***
-1,747
[1,959]
15,206
[1,446]***
2,363
[3,245]
21,979
[2,184]***
-1,991
[1,161]*
5,190
[2,176]**
10,658
[3,472]***
29,716
[5,837]***
3,718
[2,739]
47,040
[5,680]***
-54.23
[1,290]
-3,991
[1,606]**
5,747
[1,711]***
17,089
[3,028]***
2,011
[1,713]
14,692
[2,600]***
7,604
[3,926]*
10,709
[4,291]**
-1,642
[2,430]
3,266
[1,090]***
-12,588
[7,292]*
30,764
[2,409]***
-8,549
[7,022]
13,402
[2,036]***
955.5
[1,917]
8,778
[1,075]***
-3,109
[3,984]
16,913
[1,155]***
-7,194
[3,707]*
7,868
[1,225]***
-10,548
[6,455]
16,659
[2,370]***
25,289
[13,363]*
-9,616
[10,354]
-7,931
[13,195]
-2,134
[1,469]
975.1
[2,426]
-1,946
[2,554]
6,815
[4,156]
-178.4
[3,408]
-7,029
[4,352]
327.9
[4,753]
-3,448
[4,056]
-6,329
[4,829]
1,372
[1,125]
-416.4
[1,717]
-2,235
[1,864]
5,443
[2,472]**
-1,177
[2,255]
-3,790
[2,788]
-1,996
[2,333]
-2,300
[2,562]
1,323
[3,163]
4,908
[4,074]
-2,376
[3,802]
-5,893
[5,026]
272,813
[9,514]***
5,482
[1,288]***
45,231
[3,157]***
37,500
[3,636]***
6,931
[907.8]***
25,927
[1,797]***
15,905
[1,729]***
22,487
[2,628]***
Wood/charcoal/ kerosene/ paraffin/ candles/ matches
22
4.4: Impact on Consumption - Education Level of Household Head
ent
-1,042
5,785
-4,425
2,845
[799.9]
[3,013]*
[3,794]
[2,474]
y completed
3,069
14,466
7,937
7,036
[919.0]*** [3,177]*** [4,519]*
[3,557]**
ary
ted
2,130
[1,551]
5,262
[1,788]***
26,116
[12,166]**
50,753
[11,336]**
*
79,772
-902.1
[933.6]
1,585
[1,571]
6,282
[3,143]**
19,149
[5,024]***
-1,188
[3,722]
20,129
[4,692]***
605.9
[1,103]
-853.2
[1,125]
3,384
[2,009]*
1,206
[2,021]
-691.4
[1,787]
6,009
[2,184]***
5,870
[3,496]*
4,411
[4,145]
15,719
23,253
40,852
-2,730
15,625
18,635
14,410
[5,054]***
[6,213]***
[6,487]***
[1,909]
[4,974]***
[5,341]***
[6,626]**
14.07
[1,778]
-6,195
[6,343]
-2,436
[5,534]
2,291
[1,808]
6,164
[2,952]**
-1,701
[3,012]
-8,174
[5,543]
3,897
20,592
22,212
23,492
17,768
[1,243]***
[3,703]***
[5,808]***
[5,386]***
[2,795]***
ent x
y completed
1,017
[1,662]
-5,167
[4,579]
-4,645
[5,085]
887.9
[5,145]
1,078
[3,039]
[13,205]**
*
-18,997
[15,882]
ent x
ary complete
246.2
[1,786]
-853.0
[5,371]
-4,912
[7,052]
4,095
[7,359]
-1,363
[3,943]
7,191
[17,228]
-9,541
[5,497]*
2,163
[8,000]
2,653
[8,600]
-1,353
[2,081]
626.2
[6,475]
-3,333
[6,686]
-6,166
[8,933]
nt
2,961
[616.6]***
5,289
22,594
[1,907]***
5,010
27,075
[3,053]***
4,485
6,924
[1,547]***
5,165
3,870
[845.6]***
5,266
244,635
[8,697]***
4,933
2,686
[798.2]***
5,178
36,958
[2,316]***
4,806
25,793
[3,067]***
4,746
7,012
[893.1]***
5,083
21,821
[1,558]***
5,042
10,915
[1,323]***
4,932
18,324
[2,176]***
5,122
ations (N)
23
Table 4: Impact on Consumption – Part 2
Table 4.1: Impacts on Consumption – Part 2- Overall
(1)
(2)
VARIABLES
Boys
Girls
schooling
schooling
(past month)
(past month)
(3)
Clothing for
adults (past 2
months)
(4)
Clothing for
girls (past 2
months)
-6,111
-6,085
-4,952
-3,993
[4,622]
[3,946]
[3,636]
[2,544]
Constant
38,809
34,349
35,653
20,294
[3,644]***
[3,134]***
[3,054]***
[2,186]***
Table 4.2: Impacts on Consumption – Part 2- Rural – Urban
Treatment
3,578
-594.4
154.2
1,825
[3,318]
[2,921]
[3,804]
[2,188]
urban
38,419
29,628
19,288
16,009
[6,499]***
[5,759]***
[5,851]***
[4,168]***
urban x Treatment
-19,165
-10,547
-9,771
-11,532
[8,559]**
[7,388]
[7,047]
[4,925]**
Constant
19,746
19,640
26,007
12,317
[2,454]***
[2,244]***
[3,154]***
[1,783]***
Table 4.3: Impacts on Consumption – Part 2– Gender of Main Beneficiary
Treatment
-5,002
-3,489
-5,881
-1,846
[5,235]
[4,214]
[3,883]
[2,732]
female beneficiary
-7,805
8,459
-5,565
4,078
[4,254]*
[4,874]*
[3,961]
[3,584]
Treatment x female
-1,152
-8,987
5,971
-3,595
beneficiary
[5,104]
[5,566]
[4,818]
[4,013]
-4,687
[1,951]**
17,376
[1,727]***
(6)
Given to nonhousehold
members
(past 2 mos)
-2,371
[2,248]
19,850
[1,913]***
1,046
[1,816]
12,271
[3,302]***
-11,508
[3,772]***
11,275
[1,362]***
3,005
[2,039]
17,311
[3,573]***
-10,702
[4,256]**
11,315
[1,479]***
-4,786
[2,177]**
-936.2
[2,916]
777.8
[3,226]
-2,501
[2,642]
-2,242
[2,849]
1,161
[3,474]
Constant
Treatment
(5)
Clothing for
boys (past 2
months)
41,358
31,731
36,491
17,987
[3,955]***
[3,183]***
[3,199]***
[2,372]***
Table 4.4: Impacts on Consumption – Part 2– Education of Household Head
Treatment
-432.7
-4,610
2,422
-918.8
[4,778]
[3,838]
[3,506]
[2,323]
primary completed
10,040
3,235
20,949
6,372
[5,873]*
[4,612]
[6,030]***
[4,912]
secondary completed
30,884
24,941
30,143
16,620
[8,946]***
[8,751]***
[8,802]***
[6,542]**
Treatment x primary
-15,692
-2,604
-18,600
-4,019
completed
[6,935]**
[6,172]
[6,667]***
[5,560]
17,574
[1,913]***
20,546
[2,196]***
-2,048
[2,176]
4,535
[3,573]
13,728
[6,514]**
-3,564
[4,113]
-917.2
[1,962]
6,413
[2,879]**
22,253
[5,965]***
-1,449
[4,119]
Treatment x secondary
completed
-14,748
[11,415]
869.5
[11,527]
-20,489
[10,019]**
-8,847
[7,544]
-11,669
[6,799]*
-6,350
[6,942]
Constant
31,764
[3,720]***
4,987
28,682
[3,162]***
5,019
26,318
[2,738]***
4,877
15,512
[1,995]***
5,010
14,134
[1,930]***
5,005
15,002
[1,620]***
5046
Observations (N)
24
Table 5: Impacts on Consumption – Health Services
Table 5.1: Impacts on Consumption – Health Services – Overall
(1)
(2)
(3)
(4)
(5)
No of
Household
drugs or
% of boys 0-5 % of boys 0-5
visits
health visits
medication who went to
who were sick
(hhld)
(incl.
in hh
health facility and went to
transport)
health facility
Treatment
0.447
-1,769
11,117
0.0774
0.0773
[0.256]*
[1,836]
[4,577]**
[0.0262]***
[0.0309]**
Constant
3.633
16,624
61,074
0.337
0.846
[0.162]*** [1,290]***
[2,956]*** [0.0191]***
[0.0275]***
Table 5.2: Impacts on Consumption – Health Services – Urban-Rural
Treatment
0.465
-162.2
10,236
0.0632
0.0858
[0.383]
[2,327]
[5,600]*
[0.0317]**
[0.0400]**
Urban
-0.799
4,896
15,804
-0.0319
0.0152
[0.315]**
[2,542]*
[5,751]*** [0.0392]
[0.0532]
urban x Treatment
-0.0588
-3,160
2,243
0.0389
-0.0249
[0.502]
[3,651]
[8,933]
[0.0555]
[0.0609]
Constant
4.034
14,170
53,154
0.349
0.841
[0.261]*** [1,640]***
[3,524]*** [0.0239]***
[0.0360]***
Table 5.3: Impacts on Consumption – Health Services – Gender of Main Beneficiary
Treatment
0.486
-624.2
10,224
0.0598
0.0923
[0.280]*
[2,259]
[5,372]*
[0.0327]*
[0.0400]**
female beneficiary
0.524
1,062
-1,726
-0.0186
0.0116
[0.248]**
[2,192]
[4,709]
[0.0361]
[0.0489]
Treatment x female
-0.267
-3,757
-667.5
0.0517
-0.0325
beneficiary
[0.339]
[2,844]
[6,829]
[0.0526]
[0.0552]
Constant
(6)
% of girls 0-5
who went to
health facility
-0.0368
[0.0240]
0.380
[0.0172]***
(7)
% of girls 0-5
who were sick
and went to
health facility
-0.0237
[0.0268]
0.896
[0.0183]***
-0.0175
[0.0286]
0.00330
[0.0368]
-0.0500
[0.0510]
0.379
[0.0204]***
0.00339
[0.0345]
0.00895
[0.0360]
-0.0738
[0.0543]
0.892
[0.0246]***
-0.0343
[0.0311]
0.0224
[0.0353]
-0.0103
[0.0471]
-0.0111
[0.0323]
0.00554
[0.0354]
-0.0155
[0.0468]
3.475
16,430
62,233
0.343
0.840
[0.182]*** [1,498]***
[3,429]*** [0.0235]***
[0.0368]***
Table 5.4: Impacts on Consumption – Health Services – Education Level of Household Head
Treatment
0.632
1,825
12,889
0.0387
0.0886
[0.295]**
[2,093]
[5,214]**
[0.0324]
[0.0364]**
primary completed
-0.0441
5,177
9,064
-0.0795
0.0109
[0.287]
[2,805]*
[5,491]*
[0.0513]
[0.0673]
Secondary completed 0.213
8,453
18,134
-0.0476
0.166
[0.318]
[4,463]*
[6,382]*** [0.0498]
[0.0325]***
Treatment x primary
-0.273
-5,500
-2,245
0.120
-0.00969
completed
[0.413]
[3,723]
[7,952]
[0.0678]*
[0.0764]
0.373
[0.0234]***
0.891
[0.0242]***
-0.0344
[0.0289]
0.0383
[0.0399]
0.0168
[0.0504]
0.0563
[0.0608]
0.000105
[0.0339]
0.0239
[0.0407]
0.0517
[0.0435]
-0.0540
[0.0641]
Treatment x
secondary completed
-0.639
[0.497]
-11,553
[5,146]**
-109.2
[10,990]
0.0726
[0.0728]
-0.158
[0.0518]***
-0.0293
[0.0735]
-0.0705
[0.0746]
Constant
3.585
[0.194]***
5323
13,870
[1,122]***
5325
55,565
[3,309]***
5325
0.364
[0.0246]***
1,506
0.834
[0.0325]***
684
0.369
[0.0209]***
1,434
0.885
[0.0242]***
637
Observations (N)
25
Table 6: Impacts on Consumption – Education Services
Table 6.1: Impacts on Consumption – Education Services - Overall
(1)
(2)
(3)
% children 6-14 % girls 6-14 in
% boys 6-14 in
in school
school
school
Treatment
Constant
-0.0249
[0.0218]
0.776
[0.0163]***
-0.0312
[0.0235]
0.781
[0.0170]***
-0.0255
[0.0254]
0.781
[0.0182]***
(4)
% children 6-14 in
school who missed
school in last 4 wks
0.0579
[0.0177]***
0.101
[0.00882]***
Table 6.2: Impacts on Consumption – Education Services – Urban- Rural
Treatment
-0.0163
-0.0342
-0.0205
0.0923
[0.0292]
[0.0309]
[0.0344]
[0.0275]***
Urban
0.194
0.170
0.193
-0.0150
[0.0269]***
[0.0301]***
[0.0305]***
[0.0179]
urban x Treatment
0.00348
0.0375
0.00574
-0.0746
[0.0355]
[0.0407]
[0.0420]
[0.0339]**
Constant
0.682
0.699
0.692
0.109
[0.0237]***
[0.0253]***
[0.0263]***
[0.0149]***
Table 6.3: Impacts on Consumption – Education Services – Gender of Main Beneficiary
Treatment
-0.0236
-0.0200
-0.0289
0.0560
[0.0261]
[0.0296]
[0.0304]
[0.0211]***
female beneficiary
0.0321
0.0570
0.0198
-0.0124
[0.0200]
[0.0265]**
[0.0270]
[0.0157]
Treatment x female
-0.00997
-0.0385
0.0121
0.00608
beneficiary
[0.0310]
[0.0392]
[0.0377]
[0.0253]
Constant
0.765
0.760
0.773
0.106
[0.0197]***
[0.0222]***
[0.0221]***
[0.0110]***
Table 6.3: Impacts on Consumption – Education Services – Education Level of Household Head
Treatment
-0.0209
-0.0288
-0.0315
0.0826
[0.0257]
[0.0283]
[0.0303]
[0.0221]***
primary completed
0.143
0.140
0.110s
0.00857
[0.0228]***
[0.0297]***
[0.0300]***
[0.0198]
Secondary completed
0.203
0.195
0.173
0.0556
[0.0273]***
[0.0342]***
[0.0353]***
[0.0220]**
Treatment x primary
-0.0533
-0.0577
-0.00830
-0.0132
completed
[0.0352]
[0.0451]
[0.0444]
[0.0348]
Treatment x secondary 0.0432
0.0475
0.0734
-0.122
completed
[0.0348]
[0.0426]
[0.0448]
[0.0337]***
Constant
0.716
0.722
0.734
0.0915
[0.0198]***
[0.0213]***
[0.0223]***
[0.0105]***
Observations (N)
3,110
2,059
2,013
2543
(5)
Ages 6-14 avg.
school days missed
in the last 4 weeks
0.165
[0.0726]**
0.346
[0.0501]***
0.188
[0.114]*
-0.104
[0.0973]
-0.0661
[0.142]
0.397
[0.0880]***
0.196
[0.0889]**
0.0470
[0.109]
-0.0903
[0.135]
0.328
[0.0603]***
0.224
[0.0779]***
0.158
[0.0973]
0.200
[0.100]**
0.0190
[0.165]
-0.383
[0.133]***
0.286
[0.0464]***
3110
26
Table 7: Impacts on Savings and Investments
Table 7.1: Impacts on Savings and Investments – Overall
Variables
Household
Household
Total
Household
had savings
participated
household
member set
account in
in osusu in
savings
up new
last 3 months last 3
(past 3
enterprise in
months
mos.)
last 3 mos.
Treatment
-0.0135
0.0408
6,133
0.247
[0.00958]
[0.0206]**
[7,938]
[0.0283]***
Constant
0.0641
0.260
77,636
0.0887
[0.00732]*** [0.0147]*** [6,160]***
[0.00967]***
Table 7.2: Impacts on Savings and Investments – Urban- rural
Treatment
-0.00278
0.0501
14,624
0.289
[0.00658]
[0.0295]*
[7,704]*
[0.0325]***
urban
0.0904
-0.0528
53,085
0.0521
[0.0123]***
[0.0290]*
[11,432]*** [0.0187]***
urban x Treatment
-0.0195
-0.0206
-15,946
-0.0838
[0.0164]
[0.0402]
[14,876]
[0.0567]
Constant
0.0188
0.287
51,032
0.0626
[0.00548]*** [0.0205]*** [4,720]***
[0.00932]***
Table 7.3: Impacts on Savings and Investments – Gender of Main Beneficiary
Treatment
-0.0154
0.0311
1,325
0.213
[0.0110]
[0.0234]
[8,335]
[0.0291]***
female beneficiary
-0.0125
0.0357
856.1
-0.0131
[0.00944]
[0.0200]*
[7,824]
[0.0171]
Treatment x female
0.00630
0.0265
9,764
0.0821
beneficiary
[0.0126]
[0.0285]
[10,432]
[0.0315]***
Constant
0.0675
0.248
77,782
0.0911
[0.00843]*** [0.0171]*** [6,480]***
[0.0109]***
Table 7.4: Impacts on Savings and Investments – Education Level of Household Head
Treatment
-0.0175
0.0500
10,397
0.241
[0.00661]*** [0.0253]**
[7,556]
[0.0255]***
primary completed
0.0317
0.0111
28,939
0.00715
[0.0127]**
[0.0248]
[8,934]***
[0.0158]
Secondary completed
0.148
-0.0360
80,229
0.0871
[0.0241]***
[0.0278]
[14,620]*** [0.0258]***
Treatment x primary
-0.00141
0.0334
-8,198
0.111
completed
[0.0151]
[0.0381]
[12,474]
[0.0439]**
Treatment x secondary
0.0384
-0.0628
-8,189
-0.0832
completed
[0.0349]
[0.0387]
[20,346]
[0.0448]*
Constant
Observations (N)
0.0334
[0.00586]***
5314
0.260
[0.0175]***
5314
58,288
[5,444]***
5323
0.0674
[0.00902]***
5304
Amount spent
on household
improvements
(past 2
months)
7,470
[3,357]**
22,622
[2,143]***
Amount spent
on own
businesses
(past 2
months)
650.6
[4,622]
42,223
[3,670]***
8,030
[3,558]**
12,935
[4,122]***
-621.3
[6,593]
16,124
[2,455]***
15,783
[5,343]***
40,847
[6,451]***
-30,257
[8,521]***
22,011
[3,885]***
7,950
[4,000]**
-4,518
[3,594]
230.9
[5,416]
2,246
[5,535]
191.6
[5,447]
-3,224
[7,203]
23,937
[2,426]***
41,856
[4,307]***
10,118
[3,013]***
9,664
[4,713]**
35,503
[8,946]***
-7,322
[6,085]
-5,807
[12,477]
4,795
[4,841]
22,063
[6,892]***
28,684
[10,440]***
-19,328
[8,864]**
5,976
[13,543]
15,048
[1,726]***
5012
32,532
[3,665]***
5001
27
Table 8: Impacts on Social Cohesion
Table 8.1: Impacts on Social Cohesion - Overall
(1)
(2)
Variables
Dummy:
Dummy:
high trust
high trust
household
extended
member
family
Treatment
0.0450
0.0645
[0.0190]**
[0.0147]***
Constant
0.459
0.158
[0.0119]*** [0.00891]***
(3)
Dummy:
high trust
community
member
0.00533
[0.0104]
0.0999
[0.00792]***
(4)
Dummy:
high trust
same
ethnicity
-0.0386
[0.0118]***
0.128
[0.00939]***
(5)
Dummy:
high trust
same
religion
0.0471
[0.0168]***
0.196
[0.0117]***
(6)
Dummy:
high trust
diff.
ethnicity
-0.0280
[0.00976]***
0.102
[0.00746]***
(7)
Dummy:
high trust
diff. religion
Table 8.2: Impacts on Social Cohesion – Urban-rural
Treatment
0.0465
0.111
[0.0265]*
[0.0188]***
urban
-0.00758
0.0533
[0.0238]
[0.0172]***
urban x Treatment
-0.00322
-0.0948
[0.0379]
[0.0289]***
Constant
0.463
0.132
[0.0165]*** [0.0120]***
-0.00368
[0.0150]
-0.0229
[0.0157]
0.0178
[0.0208]
0.111
[0.0119]***
-0.0682
[0.0160]***
-0.0440
[0.0183]**
0.0594
[0.0233]**
0.150
[0.0133]***
-0.00805
[0.0217]
-0.0477
[0.0229]**
0.112
[0.0329]***
0.220
[0.0161]***
-0.0323
[0.0143]**
-0.0179
[0.0148]
0.00835
[0.0194]
0.111
[0.0110]***
-0.0212
[0.0156]
0.00250
[0.0142]
0.0120
[0.0220]
0.109
[0.0110]***
Table 8.3: Impacts on Social Cohesion – Gender of Main Beneficiary
Treatment
0.0518
0.0662
0.00101
-0.0355
[0.0211]**
[0.0172]***
[0.0122]
[0.0135]***
female beneficiary
0.0120
-0.00258
-0.00263
0.0117
[0.0223]
[0.0152]
[0.0151]
[0.0173]
Treatment x female
-0.0219
-0.00357
0.0133
-0.00976
beneficiary
[0.0300]
[0.0223]
[0.0194]
[0.0212]
Constant
0.456
0.157
0.101
0.124
[0.0133]*** [0.0106]***
[0.00907]*** [0.0110]***
0.0577
[0.0198]***
0.00530
[0.0193]
-0.0252
[0.0274]
0.193
[0.0139]***
-0.0179
[0.0112]
0.0208
[0.0156]
-0.0236
[0.0194]
0.0933
[0.00842]***
0.000449
[0.0129]
0.0233
[0.0163]
-0.0376
[0.0211]*
0.101
[0.00876]***
Table 8.4: Impacts on Social Cohesion – Education Level of Household Head
Treatment
0.00945
0.0695
0.00197
-0.0526
[0.0227]
[0.0177]***
[0.0134]
[0.0144]***
primary completed
0.0262
0.00982
-0.0353
-0.0433
[0.0261]
[0.0212]
[0.0143]**
[0.0172]**
Secondary completed
0.0428
0.0551
0.00866
-0.0173
[0.0307]
[0.0248]**
[0.0176]
[0.0173]
Treatment x primary
0.0845
0.0266
0.0337
0.0402
completed
[0.0381]**
[0.0294]
[0.0227]
[0.0228]*
Treatment x secondary 0.102
-0.0636
-0.0217
0.0472
completed
[0.0444]**
[0.0368]*
[0.0231]
[0.0257]*
Constant
0.446
0.147
0.106
0.139
[0.0135]*** [0.0117]***
[0.0106]***
[0.0122]***
Observations (N)
5,310
5,317
5,318
5,317
0.0120
[0.0191]
-0.0277
[0.0211]
-0.00911
[0.0246]
0.0476
[0.0300]
0.160
[0.0369]***
0.203
[0.0139]***
5,316
-0.0402
[0.0122]***
-0.0355
[0.0152]**
-0.0189
[0.0155]
0.0507
[0.0224]**
0.0207
[0.0200]
0.111
[0.00999]***
5,317
-0.0235
[0.0134]*
-0.000818
[0.0179]
-0.00591
[0.0167]
0.0257
[0.0239]
0.0139
[0.0231]
0.111
[0.00981]***
5,313
-0.0154
[0.0110]
0.111
[0.00709]***
28
Figures
Figure 1: Randomized Phase-In Design
Figure 2a: Geographical distribution
of sub-projects
Figure 1b: Geographical location
of treatment sites
29
Figure 3: Targeting performance of CfW program
30