A Fine Predicament: Conditioning, Compliance and

Consequences in a Labeled Cash Transfer Program

Carolyn J. Heinrich and Matthew T. Knowles

December 29, 2018

Abstract

The Kenya Cash Transfer Programme for Orphans and Vulnerable Children presents avaluable opportunity to examine the effects of imposing monetary penalties for noncom-pliance with conditions in cash transfer programs, in contrast to providing only guidancefor cash transfer use. We take advantage of random assignment to “hard” conditions withinthe CT-OVC treatment locations to estimate the impact of fines imposed on program ben-eficiaries, as well as the marginal effects of being penalized by household wealth. We findthat comparatively wealthier households in the hard conditions treatment arm appearedto be better equipped (and motivated by the potential for fining) to respond to the guid-ance in ways that would help them to avoid penalties. Alternatively, for comparativelypoorer households, being fined was associated with a decrease in consumption of aboutone half the increase in consumption experienced, on average, by CT-OVC beneficiaries,suggesting the potential for unintended, regressive policy effects.

1 Introduction and Background

Cash transfers are one of the most popular forms of aid interventions directed toward re-

ducing poverty and the intergenerational transmission of poverty. More than a fifth of all

countries have implemented a conditional cash transfer (CCT) program, including about one-

third of developing and middle-income countries (Morais de Sá e Silva, 2017). Unconditional

cash transfer programs are profilerating as well and are among some of the largest cash transfer

programs today (e.g., China’s dibao program with about 75 million beneficiaries) (Golan et al.,

2015). One global estimate of the number of beneficiaries of cash transfer programs (Fiszbein

1

et al., 2014) suggests that close to one billion people worldwide are receiving cash transfers as

a form of social protection (i.e., social assistance for poor households). The implementation

of many cash transfer programs has also been accompanied by rigorous evaluation efforts to

identify their impacts, which has contributed to a growing evidence base on a wide range of

potential program effects in education, health, labor, consumption, food security, asset build-

ing, risky behaviors and more (see: https://transfer.cpc.unc.edu/; Hidrobo et al., 2018; Ralston

et al., 2017). In fact, observing the positive findings of cash transfer programs on communities

and households, some governments in poor countries are now implementing them as regular

components of their economic development and social protection efforts (Bastagli et al., 2016).

Most of the inaugural cash transfers programs, and many subsequent program efforts, have

imposed conditions on households’ receipt of cash transfers that prescribe how the monies

should be used (Baird et al., 2013). Among the most common of these conditions are school

enrollment and minimum attendance requirements for the child beneficiaries; regular health

and wellness checks and immunizations for infants and young children, and health and nu-

trition training and information sessions for parents or caregivers of the beneficiaries. For

example, two of the earliest and largest CCT programs, Mexico’s Prospera program (previ-

ously named PROGRESA and Oportunidades), and Brazil’s Bolsa Familia program, require

households to enroll their children in school and the children to maintain 85 percent attendance

rates, ensure that they get preventative healthcare (check-ups) and vaccinations, and participate

in educational activities offered by health teams or attend monthly meetings to access health

and education information, to receive the transfer (Levy, 2006; Fiszbein et al., 2009). While the

marked success of these two CCT programs–including permanent increases in food consump-

tion, reductions in chronic malnutrition, and increased school enrollment rates–galvanized the

replication of this CCT model throughout Latin America and beyond (Fernald et al., 2008;

Handa et al., 2018), the transmission of the conditionalities to other contexts has hit con-

straints.

The implementation and enforcement of conditions requires substantial infrastructure and

administrative capacity. In Brazil, for example, local education departments are responsible for

checking and reporting the school attendance rates of beneficiaries every two months through

the (computerized) School Attendance Surveillance System, and principals are required to re-

2

port the reasons for absences and take appropriate actions when the student attendance report

is returned to the school. A separate computer system managed by the Ministry of Health,

Sistema de Vigilância Alimentar e Nutricional, is used by municipalities for reporting compli-

ance with the health conditions, and municipalities are also required to verify access to quality

health services for program beneficiaries. Furthermore, the direct costs of complying with

conditions can be burdensome for beneficiaries, and may also open the door for corruption

in situations where those verifying conditions charge fees or demand payments for certifying

compliance (de Brauw and Hoddinott, 2011; Heinrich and Brill, 2015). For these and related

reasons, the implementation of unconditional cash transfers (UCTs) has become more com-

monplace in very low-income countries, and intermediate program models, where guidance

for spending the transfer is articulated but not monitored or enforced–sometimes described as

“labeled” cash transfer programs (LCTs)–have also been introduced (Benhassine et al., 2013).

In this research, we focus on a less explored consequence of complying with conditions

for households–the costs to them when financial penalties are incurred because of failure to

comply with conditions. We undertake this analysis in the context of the Kenya Cash Trans-

fer Programme for Orphans and Vulnerable Children (CT-OVC), a LCT that was distinct in

its random assignment of “hard conditions” (conditions with penalties) within locations that

were randomly selected to receive cash transfers. We exploit the random assignment to hard

conditions in the Kenya CT-OVC to explore the implications of fines on household outcomes,

given that the implementation of hard conditions resulted in households in both treatment and

control arms having similar beliefs and incentives regarding program rules and penalties. This

increases confidence that assignment to hard conditions affected outcomes of interest only

through its effect on the probability of being fined. Our research, which shows how particular

features of development interventions can unintentionally harm those most in need of assis-

tance, has clear policy implications for the design and evolution of cash transfer programs and

our understanding of how households respond to unexpected negative income shocks.

In the following section (2), we review the literature on conditional, unconditional and

“labeled” cash transfer programs, focusing on the types of conditions or guidance embodied in

the programs, how they were implemented, and evidence on the relationship of conditions to

program outcomes. We next present background information on the Kenya CT-OVC program

3

and the nature of the conditions, penalities and labeling of the cash transfers (section 3), and

we also describe the design of the experimental evaluation and data collected that we draw on

in this study. In section 4, we introduce the methods we employ in investigating how imposing

hard conditions (vs. providing guidance or labeling alone for cash transfer use) influences

households’ understanding of program rules and their behavioral responses, as well as three

domains of household and children’s outcomes (consumption, dietary diversity and schooling).

We present the findings o f o ur a nalyses i n s ection 5 a nd c onclude w ith a d iscussion o f the

results in section 6. We find that the effect of being subjected to hard conditions (and the risk

of penalty fines that reduce a household’s cash transfer amount) differed greatly by baseline

household consumption. In fact, while comparatively poorer households are more frequently

fined and endure decrements to consumption, comparatively wealthier households appear to

be better equipped to respond to guidance for cash transfer use by increasing spending on

food quantity and variety, perhaps in an attempt to avoid being fined. These findings affirm

conventional wisdom that penalties in cash transfer programs disproportionately harm those

who are least able to respond to them.

2 Literature Review

2.1 Why condition?

As cash transfer programs have expanded to all regions of the world, variation in their

implementation has spread as well, with tinkering typically around the designation and admin-

istration of conditions or rules of cash transfer receipt. Numerous works have articulated the

arguments for and against the imposition of conditions (Ferreira, 2008; Fiszbein et al., 2009;

de Brauw and Hoddinott, 2011), which we briefly review here. As Fiszbein et al. point out, in

ideal circumstances—where individuals are well-informed and make rational choices, govern-

ments are benevolent and operate efficiently, and markets function perfectly—unconditional

cash transfers should be the preferred policy design from both public and private perspectives.

However, if we are concerned that individuals lack information to make the most appropriate

decisions for use of the transfers, the government can play a role in helping them to overcome

4

these informational problems, e.g., conditioning receipt on uses that are believed to increase

their net positive impacts. In other words, the conditions can induce a substitution effect (in

spending) that enhances the overall effect of the cash transfers. Another set of arguments per-

tains to the political feasibility (or political benefits) of offering cash transfers, where public

spending on the programs may be viewed as more palatable or popular if the cash transfers

are conditioned on “good behavior” or if they are delivered as part of a “social contract” with

the state that defines “co-responsibilities” (Fiszbein et al., 2009; Lindert et al., 2007). In ad-

dition, de Brauw and Hoddinott (2011) note that if the conditions serve as a mechanism for

increasing the effectiveness of the transfers and politicians and policy makers can take credit

for the results, the conditions may be a useful tool for helping them to stay in office as well.

Lastly, a third prevailing argument in support of CCTs is that the investments in human capi-

tal encouraged through conditioning generate positive externalities for the public, such as the

benefits associated with immunization, which caregivers would not fully consider in their own

decision making (contributing to underinvestments from a societal perspective).

These potential benefits have to be weighed, however, against the (public and private) costs

of administering and complying with the conditions. There is very limited information avail-

able on the costs associated with implementing and monitoring compliance with conditions,

largely because it is difficult to distinguish these costs from other administrative costs or to

identify those that are imposed on health, education sector and other social welfare staff in-

volved in delivering services. In a study comparing program costs across three Latin American

CCTs, Caldes et al. (2006) estimated the costs of conditions–distributing, collecting, and pro-

cessing registration, attendance, and performance forms to schools and healthcare providers

(distinguishing them from overall program monitoring and evaluation costs)–and found that

the conditions constituted nearly one quarter of the administrative costs in PROGRESA (in

2000). It is also challenging to fully account for the costs of meeting conditions that are

imposed on the program beneficiaries—such as transportation and other transaction costs as-

sociated with accessing required services—and to assess who bears those burdens in the house-

hold. Of course, there are also direct costs to households of any fines or penalties imposed if

they are found not to be in compliance. The research base generally finds that CCTs increase

total household consumption and disproportionately affect food consumption in poor house-

5

holds, and that increases in food expenditures are typically directed at increasing quality (e.g.,

items rich in protein and fruits and vegetables) (Fiszbein et al., 2009; Hoddinott, Skoufias,

and Washburn 2000; Macours, Schady, and Vakis, 2008; Maluccio and Flores 2005). If the

households who find it most challenging to satisfy the conditions are among the poorest of

program eligibles, this could unduly penalize household consumption and constrain dietary

diversity among those most in need (de Brauw and Hoddinott, 2011; Heinrich & Brill, 2015;

Rodríguez-Castelán, 2017). How large and important are the informational problems and ex-

ternalities of CCTs (and the role of conditions in addressing them) relative to the public and

private costs they engender is still an open question, and one where the answer surely varies

considerably across program and country contexts.

2.2 Nature, role and effects of conditions in cash transfer programs

In the growing evidence base on CCTs, UCTs, and their program variants, researchers have

sought to characterize the nature and role of conditions in implementation and to understand

how they relate to program effectiveness (Morais de Sá e Silva, 2017). In their 2013 meta-

analytic review of 35 studies of cash transfer programs focused on CCTs with at least one

condition tied to schooling, Baird et al. conceded that the binary classification of CCTs vs.

UCTs disregarded considerable variation in the nature and intensity of the conditions. In their

analysis, they further categorized the cash transfer programs as having: (i) no schooling con-

ditions, (ii) some schooling conditions with no enforcement or monitoring, and (iii) explicit

schooling conditions that were monitored and enforced; within each of these categories, they

attempted to capture variation in nature and intensity of the conditions. For example, Baird

et al. describe both Bolsa Familia and PROGRESA as having “explicit conditions,” but with

imperfect monitoring and minimal enforcement. Other research similarly suggests that the dis-

tinction between the second and third categories may not always be precise; that is, there may

be more of a gradation from monitoring and enforcement to no monitoring and enforcement in

many programs, where the degree of “softness” is realized in implementation of the cash trans-

fer programs (Fizbein et al., 2009; Ralston et al., 2017; Hidrobo et al., 2018). Silva (2007), for

instance, describes the Bolsa Familia conditions as a “soft type of conditionalities,” where the

6

sanctions imposed for not complying with conditions are moderate and implemented at differ-

ent levels, ranging from a simple warning to temporary suspension of payments or definitive

removal (following a progression of non-compliance), and take into consideration the reasons

for non-compliance.

The more flexible approach to the implementation of conditions in Bolsa Familia reflects

concerns that some families with a greater likelihood of non-compliance may be more econom-

ically vulnerable (and harmed by a financial penalty), and that weaknesses in infrastructure,

such as resources and staff for meeting demand for education and health services (as well as in

the administrative and financial capacities for managing the program), may limit the support

families receive in attempting to meet the conditions. Prospera (in Mexico) likewise applies

a multi-stage approach to fines or sanctions, with suspension of payments as a first step, in-

definite suspension with the option of re-admittance as a second step, followed by permanent

suspension. Other programs also allow exceptions or exemptions to the conditions and sanc-

tions they impose, such as forgiving absences on grounds of illness, or in the case of Jamaica,

granting waivers from attendance requirements for disabled children (Fiszbein & Shady, 2009;

Mont, 2006). In contrast, the Chile Solidario program does not begin paying cash transfers un-

til families have complied with the first criterion, and noncompliance results in an immediate

termination of the transfers (Palma & Urzúa, 2005).

Somewhat distinct from cash transfer programs with a continuum of hard to soft conditions

is the concept of a “labeled” cash transfer program (LCT), where the cash transfer is distributed

to households with a “nudge” or “label” indicating its intended use, in contrast to a monetary

carrot or stick to ensure compliance with specified uses (Benhassine et al., 2013). For example,

if an LCT is to be spent exclusively on more nutritious food, program administrators would

convey this through “loose guidance” to recipients when the cash transfer is received. Like

Baird et al.’s first category (conditions with no enforcement or monitoring), no monitoring

takes place to determine whether the recipients are following the guidance on how the money

is to be spent. In Benhassine et al.’s (2013) evaluation of the Tayssir cash transfer program

in Morocco, a CCT version of the program was compared with the LCT, where the LCT arm

portrayed the cash tranfers as an educational intervention. Enrollment for the Tayssir LCT

was conducted at schools and by headmasters, thereby tying receipt of the cash transfers to

7

an education goal, albeit without formal requirements for attendance or enrollment. Both the

CCT and LCT had two variants: in one, the cash was transferred to the father, and in the

other, the cash transfer went to the mother. More than 320 school sectors (with at least two

communities in each) were randomly assigned to either a control group or one of these four

program variants.

Benhassine et al.’s (2013) analysis of over 44,000 children in more than 4,000 households

found significant impacts of the Tayssir cash transfers on school participation for each program

variant they tested. Interestingly, they saw little difference between the LCT and CCT in how

the program’s intended uses were perceived, and parents’ beliefs about the returns to education

increased in both the LCT and CCT treatment arms. Benhassine et al. (2013) suggested

that this is consistent with parents interpreting the intervention as a pro-education government

program, regardless of whether they formally required regular school participation (through

conditioning). They also found that dropouts related to the “child not wanting to attend school”

and to “poor school quality” declined significantly in the LCT and UCT.

Similarly, Baird et al. (2013) found in their analysis–including 26 CCTs, five UCTs, and

four studies that compared CCTs to UCTs–that both CCTs and UCTs significantly increased

school enrollment, with the odds of a child being enrolled in school 41% higher in the CCTs

and 23% higher in the UCTs (compared to no cash transfers). These differences in effects

between the CCTs and UCTs were not statistically significant. However, they also compared

cash transfer program effects across the three categories that included the middle design alter-

native (some schooling conditions with no enforcement or monitoring). When distinguishing

between whether or not the schooling conditions were monitored and enforced, they did find

that programs where the conditions were monitored and enforced had significantly higher odds

of increasing children’s enrollment than those with no conditions. The implementation of pro-

gram conditions (i.e., intensity of conditions) was the only measured design feature of the 35

cash transfer programs that significantly moderated the overall effect size of the programs.

We expand on this research in our analysis of the Kenya CT-OVC program, in which cash

transfers were explicitly earmarked or “labeled” for spending on education and healthcare

for orphans and vulnerable children in the household, but “hard” conditions (with monitoring

and penalties for noncompliance) were assigned randomly to some districts and a sublocation

8

within the treatment group (Hurrell, Ward & Merttens, 2008). Because the implementation of

hard conditions (i.e., the potential for financial penalties) following random assignment was

plagued with problems and delays in the districts (Ward et al., 2010)1, we approach the analysis

in an “intent to treat” framework. We use detailed information on cash transfer recipients’

understanding of the program rules, guidance and consequences of failure to comply with

conditions to understand the extent to which the imposition of “hard” conditions and associated

penalties (vs. labeling only of cash transfers) influenced household responses and program

outcomes. Based on the research evidence discussed above, we expect the costs of monetary

penalties to be felt most immediately in terms of household consumption, and for households

with less wealth, for the loss of transfer income to constrain their ability to respond to the

LCT guidance (e.g., by increasing dietary diversity). Thus, our analysis focuses primarily on

estimating the impact of fines on households’ total, food and non-food consumption, as well

as their dietary diversity. At the same time, given the emphasis on schooling for children in the

“labeling” of the Kenya CT-OVC program, we also examine whether the imposition of fines

affects children’s (OVC) school attendance (absences).

3 Program Background, Study Design, Data and Measures

The Kenya CT-OVC program is the government’s primary intervention for social protec-

tion in Kenya. The program provides a flat transfer equal to approximately 20 USD per month

(in 2007 dollars, exchange rate: US$1: KSh 75) that is paid bi-monthly to the caregiver for the

care and support of the OVC (Handa et al., 2014). In terms of the average (per adult equiv-

alent) consumption levels at baseline (2007), the monthly cash transfers represent about 22

percent of average monthly consumption. The CT-OVC began as a pilot program in 2004, and

following a three-year demonstration period, the government formally approved its integration

into the national budget and began rapidly expanding the program in 2007. By the end of the

1As the Kenya CT-OVC program evaluation baseline (Hurrell, Ward & Merttens, 2008) and final (Ward etal., 2010) reports are inconsistent in the discussion of the random assignment of hard conditions, we consultedwith the authors and study principal investigators to determine how we could treat the random assignment ofhard conditions in this analysis. The random assignment of hard conditions did take place at baseline in 2007,but communication, administration and monitoring of the conditions and associated penalties was slowly andunevenly rolled out in the districts and sublocations where they were intended to be in effect.

9

impact evaluation in 2011, the CT-OVC program was providing cash transfers to more than

130,000 households and 250,000 OVCs, with the aim to scale up coverage to 300,000 house-

holds (900,000 OVCs). As of fiscal year 2015-2016, approximately 246,000 households and

nearly half a million children were benefitting from the cash transfer.2

We use data from an experimental evaluation of the Kenya CT-OVC program, mandated by

the Government of Kenya, Department of Children’s Services (in the Ministry of Gender, Chil-

dren and Social Development), and undertaken by Oxford Policy Management with financial

assistance from UNICEF. The baseline quantitative survey was conducted between March and

August 2007 using questionnaires in Swahili, Luo and Somali, and follow-up surveys were

administered in 2009 and 2011. The surveys collected information on household consump-

tion expenditures, education and employment of adults, assets owned, housing conditions and

other socio-economic characteristics, as well as information on child welfare measures such

as anthropometric status, immunizations, illness, health-care seeking behaviour, school enroll-

ment and attendance, child work and birth registration. As many of the outcome indicators of

interest for the children are only available in the 2007 and 2009 data collections, we restrict

our analysis to these two years. A total of 2,759 households were included in the 2007 baseline

sample, and of these, 2,255 were interviewed at follow-up in 2009. As Handa et al. (2014)

explain, the 17 percent attrition between baseline and the first follow-up was concentrated in

Kisumu and Nairobi, where the most unrest was experienced following the turmoil of disputed

national elections that occurred in December 2007.

The evaluation of the Kenya CT-OVC was designed as a clustered randomized controlled

trial (RCT) and took place in seven of 70 districts in the country (see Figure 1 that illustrates

the design). Within each of the seven districts, two sub-locations out of four were randomly as-

signed to be treatment locations and two were randomly assigned to the control state (no cash

transfer distribution). Households in treatment locations were eligible to receive cash transfers

if at least one OVC resided in them, they met the designated poverty criteria, and the OVC(s)

were not benefitting from any other cash transfer program. In every treatment location, ben-

eficiary households were expected to comply with program guidance or expectations for how

2See the Kenyan government website: https://www.socialprotection.or.ke/social-protection-components/social-assistance/national-safety-net-program/cash-transfer-for-orphans-and-vulnerable-children-ct-ovc.

10

the cash transfers would be used. These included visits to health facilities for immunizations,

growth monitoring and nutrition supplements, school enrollment and basic education institu-

tion attendence, and caregiver “awareness” session (see Appendix Table A.1 (A)). However,

in four locations—Homa Bay, Kisumu and Kwale districts and one sub-location in Nairobi

(Kirigu)—households were randomly assigned to the “hard conditions” CCT treatment arm,

where the expected penalty for not following the program conditions was a deduction of KSh

500 from the transfer amount per infraction, and multiple infractions could result in ejection

from the program. The other districts and one sub-location—Garissa, Migori, Suba and the

other Nairobi location (Dandora B)—were assigned to the LCT arm where non-compliance

was not supposed to be penalized. More than a third of households subject to hard conditions

were fined within the first two years of cash transfer receipt, although in practice, there was

considerable variation in the implementation and enforcement of the conditions within and

across locations (which we discuss further below). In addition, attendance requirements were

waived for children deemed to be without access to schools or clinics (Government of Kenya,

2006).

In treatment locations, a list was compiled containing the households eligible to receive

the cash transfer, and households on the list were prioritized for treatment by several “vul-

nerability” criteria. These include the age of the caretakers of the OVCs, and the number of

OVCs and chronically ill living in the household, in that order. Thus, within treatment loca-

tions, there was an intent to prioritize somewhat poorer households for cash transfer receipt,

but this contributes to only one systematic difference in household characteristics between the

study treatment and control groups at baseline once standard errors are clustered at the level of

treatment (sub-location) (see Appendix Table A.2). We account for these selective differences

between the treatment and control groups in our estimation of program impacts (discussed in

Section 4).

3.1 Treatment measures

Following the baseline data collection and implementation of the cash transfer program,

household surveys were conducted in 2009 to assess the receipt of cash transfers and how

11

Figure 1: RCT Design

households used them. For all households that received the transfer, household members were

asked about their perceptions of any conditions or obligations they faced in receiving the cash

transfers and about any consequences they faced for noncompliance, as well as how they used

the cash transfers. In addition, the household members were asked if they “have to follow any

rules in order to continue receiving the program,” and they were prompted to list the rules that

they thought they had to follow “in order to receive the full payment from the OVC program.”

Furthermore, household members were asked if they knew which members of the household

the rules applied to, if they knew what would happen if they did not follow the rules, and if

they believed that anyone was checking on the conditions.

In regard to the penalties associated with hard conditions, the 2009 household survey asked

respondents if they had ever gone to the Post Office to collect their payment and “received less

than 3000KSh for the payment cycle3”. The interviewer was instructed to look at all of the

receipts the respondent provided and to identify cash transfer amounts of less than KSh 3000 to

3Payment cycles were two months in length. Since households were to receive 1500 KSh per month (if nofines had been applied), this translates to a transfer of 3000 Ksh each cycle.

12

determine if a monetary penalty had been applied. Household respondents identified as having

been fined were also asked if they knew why the payment was less than the full amount and if

they were aware of an appeal /complaints process they could pursue if they ever received less

than 3000 KSh in a payment cycle. In our study sample, about 37 percent of the households

subject to the hard conditions were reported to have received a fine in the two years since

becoming CT-OVC beneficiaries. Appendix Table A.1 (B) and (C) show all of the survey

questions that we used in constructing measures of the treatment as implemented, perceived

and used.

Because the implementation of “hard conditions” was intended to impose concrete expec-

tations for how households would spend the cash transfers and penalties for their failure to

comply, we hypothesized that households in districts and sublocations randomly assigned to

hard conditions might differ in their perceptions, responses to, and uses of the cash transfer

from those randomly assigned to the status quo of labeling only , i.e., instructions for how to

use the cash transfers but without penalities. Furthermore, we also expect there to be hetero-

geneity in responses to the hard conditions among those randomly assigned to this treatment

arm, given the variation observed in how those conditions were implemented within sites. The

final operational and impact evaluation report (Ward et al., 2010) indicated that 84 percent of

the beneficiaries believed that they had to follow some sort of rules to continue receiving the

cash transfers, but the report also noted that most beneficiaries were not aware of the full set

of conditions with which they were expected to comply. Monitoring and enforcement of the

hard conditions within and across locations was hindered by onerous forms and logistical chal-

lenges, which the literature suggests can impact poorer families disproportionately (Heinrich

& Brill, 2015; Heinrich, 2016). In addition, community representatives charged with the role

of communicating and checking on conditions were typically informally appointed and lacked

remuneration, and implementation of that role was highly dependent on a given community

representative’s knowledge, interpretation of their obligations, and activism. Two years af-

ter random assignment, many beneficiaries had not been reached with communications about

the penalities, and where penalties were imposed, those affected often did not understand the

reason for the decrement in their transfer (Ward et al., 2010; FAO, 2014).

The literature on CCTs suggests that these types of program capacity constraints in imple-

13

menting conditions and verifying compliance are relatively common. Fiszbein et al. (2009)

point out that these constraints can delay actions to sanction noncompliance, even in estab-

lished programs such as Prospera in Mexico. They also argue (p. 89) that longer lag times

between household noncompliance and the reduction of cash transfer program benefits are

likely to weaken the “positive quid pro quo” effects of the conditions on program outcomes.

Furthermore, because it is well-documented that taking a “hard line” on compliance with CCT

conditions is likely to impose higher costs on the poorest and most vulnerable among those

targeted for cash transfers—who, because of their greater need, also have less budetary capac-

ity to absorb the monetary loss—we expect there to be differential effects of being penalized

or fined for noncompliance by household baseline need and consumption levels.

3.2 Outcome measures

We evaluate the impact of being fined (penalized for noncompliance) in the Kenya CT-OVC

program on the following dimensions of household and child wellbeing: consumption (food

and non-food), nutrition and dietary diversity, and schooling. The sample sizes in our analysis

vary by outcome, primarily because the outcomes we focus on are measured for distinct groups

receiving the cash transfers: households for consumption and the dietary diversity score, and

school-aged children (6-17 years) for absences from school.

We follow the Kenya CT-OVC Evaluation Team (2012) in adjusting consumption (reported

at baseline in 2007) for household adult equivalents; children under age 15 were counted as

three-quarters of an adult, and individuals aged 15 and over were counted as one adult. Con-

sumption measured at follow-up (in 2009) was deflated to 2007 Kenya Shillings (KSh), follow-

ing Ward et al. (2010), with separate price deflators for food and non-food items. These price

adjustments were critical, given that the Kenyan post-election violence and world food crisis

that occurred between baseline and follow-up each engendered upward pressures on the rela-

tive price of food and increased poverty among the beneficiary population as a whole (Kenya

CT-OVC Evaluation Team, 2012). Household expenditures (by broad household item groups)

were combined into three main categories for our analysis: total household consumption, food

consumption, and nonfood consumption. Analyses by the Kenya CT-OVC Evaluation Team

14

showed that none of the nine separate categories of household (food and non-food) expendi-

tures were significantly different at baseline between CT-OVC treatment and control house-

holds, in spending levels, shares, or proportion of households reporting positive spending.

The second dimension that we examine reflects the broader program goal of increasing

food security and dietary diversity in OVC households. A highly consistent finding among

cash transfer program evaluations is their effectiveness in reducing hunger and food insecurity,

given the monetary resources newly made available to households for meeting their basic con-

sumption needs (Devereux & Coll-Black, 2007; Fernald et al., 2008). The final impact evalu-

ation report (Ward et al., 2010) described increases in food expenditure and dietary diversity

associated with cash transfer receipt, with significantly increased frequency of consumption

within five food groups: meat, fish, milk, sugar and fats. Ward et al. (2010) also reported an

increase of 15 percent (from baseline) in the dietary diversity score; this is consistent with the

findings of Lopez-Arana et al. (2016), who found a 16.5 percent increase in the purchase of

protein-rich foods among families benefitting from Colombia’s CCT program. Asfaw et al.

(2012) also evaluated the average difference between the treatment and control households in

the Kenya CT-OVC program in terms of different components of food consumption expendi-

ture and found positive and statistically significant impacts of the program on consumption of

animal products (e.g., dairy, eggs, meat and fish) and fruits. In our analysis, we use as an index

of dietary diversity that mirrors that of Hurrell et al. (2008), tallying the number of different

food groups from which the household ate in the past week.

The third outcome we investigate, school attendance, was one element of the Kenya CT-

OVC program’s explicit goal to increase schooling (enrollment, attendance and retention) of

children aged six to 17 years. At baseline (2007), about 95 percent of children aged 6-17 years

in both treated and control households were enrolled in school, and the final impact evaluation

report (Ward et al., 2010) did not find statistically significant impacts of the cash transfers

on enrollment or attendance of basic schooling (although it did report statistically significant

increases of 6-7 percentage points in enrollment in secondary schooling). The baseline (2007)

data also show that children in our sample missed an average of 1.5 days of school in last

month, and 10 percent of these children missed over five days in one month. We therefore

focus our analysis on school attendance, which we measure as days missed from school during

15

the school year (in 2007 and 2009). The education literature has also increasingly looked to

attendance as a more informative measure of children’s progress in schooling. Attendance rates

have been linked to the development of important sociobehavioral skills such as motivation

and self-discipline (Gernshenson, 2016; Heckman, Stixrud & Urzua, 2006) and to improved

cognitive development (Gottfried, 2009), as well as to retention rates and increased educational

attainment (Gershenson et al., 2017; Nield & Balfanz, 2006; Rumberger & Thomas, 2000).

In addition, existing research finds that the harm of absences, in terms of reduced academic

achievement, is greater among low-income students (Gershenson et al., 2017; Gottfried, 2011),

and that non-school factors, such as poverty, family emergencies and work obligations, are the

primary determinants of attendance rates (Balfanz & Byrnes, 2012; Ladd, 2012). If being fined

reduces resources for poor families that enable them to overcome these non-school barriers

to school attendance, we would expect being fined to potentially diminish the cash transfer

program’s impact on reducing student absences.

4 Methods and Estimation Strategy

The primary objective of our analysis is to estimate how implementing conditions with

penalties in a labeled cash transfer program affects households differently across the wealth

distribution (proxied by baseline consumption levels)4. In particular, we are interested in

how assignment to hard conditions affects per adult-equivalent consumption, dietary diver-

sity, and children’s schooling within households. The Kenya CT-OVC is an ideal program

for identifying these effects for two reasons. First, hard conditions were randomly assigned

(by district/sub-location) within the randomized treatment group, which aids in thwarting

many typical threats to the identification of causal effects. Second, our data include an

extensive battery of questions evaluating households’ knowledge about program conditions

and expectations. This allows us to disentangle the most likely channels through which hard

conditions (monetary penalties) affect outcomes.

4Measurements of consumption are often considered the “gold standard” for approximating a household’slevel of wealth in low-income countries (Deaton & Zaidi 2002)

16

4.1 Potential mechanisms for impacts of hard conditions

The Kenya CT-OVC Impact and Operational Evaluation Team produced two reports on the

Kenya CT-OVC program evaluation, the second of which includes a qualitative assessment of

the implementation of the labeling and hard conditions in multiple program regions (primar-

ily focus group discussions with program participants). The fieldwork also included “semi-

structured” interviews with officials who were responsible for completing compliance forms at

schools and clinics as part of the process of monitoring households with hard conditions (Ward

et al. 2010). As discussed above, conditions are imposed in cash transfer programs with the

expectation that they will alter a household’s incentives and decisions regarding how to spend

the transfer. In the case of the Kenya CT-OVC program, we have compelling evidence

from the Operational Evaluation Team reports and household surveys that both households

randomly assigned to the hard conditions treatment arm and those in the “labeling only” arm

had similar beliefs about the program guidance and expectations for the use of the cash trans-

fers. In fact, similar proportions of households in the hard conditions and labeling only arms

believed that they could be ejected from the program (lose the cash transfer entirely) if they

did not respond to the conditions. These findings suggest that households in both treatment

arms took the guidance (labeling) seriously, similar to the findings of Benhassine et al. (2013),

who likewise found that parents across their CCT and LCT arms did not have very different

understandings of the Tayssir program requirements. This also raises our expectation that any

differences in these treatment households’ behavioral responses to cash transfer receipt would

come primarily throughthe channel of receiving a fine in the hard conditions arm.

We accordingly implemented several tests to further examine the channels through which

assignment to hard conditions would affect household behavioral responses. The results from

these tests are displayed in Table 1, which we divide into panels by the category of survey

questions. Each variable in the left-most column of Table 1 is a binary indicator of what the

household believed about the program operations. Columns (1) and (2) are the mean affirma-

tive response rates for these beliefs, divided by treatment arm (assignment to hard conditions

versus labeling only). Columns (3) and (4) contain the p-values (estimated by different meth-

ods) on the “controlled” difference in the rates of beliefs between the treatment households

17

with and without hard conditions5. These results generally confirm the implications of the

qualitative research. On the whole, assignment to hard conditions is not significantly (dif-

ferentially) related to households’ understanding of the program rules, their beliefs about the

checking of conditions, or their understanding of the criteria for suspension or expulsion from

the program6. The largest (and statistically significant) difference in beliefs between the hard

conditions vs. no conditions groups is for the item “Believes Fining is a Penalty” (see Panel

C). There is also a minimally significant difference (at α <0.10) in the rate at which house-

holds believed that they needed to follow rules to continue receiving payments (see Panel B).

Apart from another small difference in households’ understanding about growth monitoring

requirements (which is unlikely to meaningfully affect household consumption or children’s

schooling), these are the only two notable differences in beliefs about the program between

these treatment arms.

In addition, we also created a summary index of household perceptions and understanding

of program rules based on of all of these variables to test whether assignment to hard conditions

was significantly related to households’ understanding of the program rules. The point estimate

of the average difference in this scalar measure (between households with hard conditions vs.

labeling only) is small in magnitude and statistically insignificant, and the index’s distribution

is visibly similar between treatment arms (see Figure 2). We argue that this confirms that any

differences in beliefs about the potential to be fined in the Kenya CT-OVC program likely arose

mechanically from random assignment to hard conditions, since by design, only households in

that treatment arm could be fined.5We estimate standard errors for column (3) using the regular formula for clustering from White (1984) and

p-values for column (4) using the wild cluster bootstrap method of Cameron, Gelbach, and Miller (2008) sincethe number of sub-locations (15) and districts (7) are both low by asymptotic standards. This bootstrap methodhas been shown to perform well even when the number of clusters is as few as 6. Additionally, we include thecontrol variables from equation (2) when estimating the differences between columns (3) and (4)

6A marginal analysis reveals that the statistical significance of these results does not vary by baseline con-sumptions levels, either.

18

Table 1: Household Program Knowledge and Beliefs by Treatment Arm(Labeling Only vs. Hard Conditions)

(1) (2) (3) (4) (5)Labeling Only Hard Conditions P-Value Boot P-Value N

Panel A: Understanding of Program RulesEnrollment/Attendance in Primary 0.294 0.313 0.781 0.802 1092or Secondary School

Visit Health Facility for Immunizations 0.158 0.226 0.146 0.237 1092

Visit Health Facility for Growth Monitoring 0.092 0.151 0.056∗ 0.065∗ 1092

Visit Health Facility for Vitamin A Supplement 0.059 0.057 0.837 0.847 1092

Adequate Food and Nutrition for Children 0.596 0.723 0.179 0.199 1092

Attendance at Program Awareness Sessions 0.043 0.077 0.067* 0.137 1092

Panel B: Perceived Likelihood of Being PenalizedBelieves HH Must Follow Rules to 0.733 0.909 0.060∗ 0.068∗ 1092Receive Payments

Believes No One is Checking if HHs 0.424 0.492 0.268 0.338 882are Following Rules

Panel C: Understanding of PenaltiesBelieves Fining is a Penalty 0.048 0.217 0.000† 0.000† 1092

Believes HHs can be Ejected from 0.434 0.457 0.877 0.880 1092Program for Incompliance

Claims to Know Specific Criteria for 0.466 0.509 0.583 0.611 1092Ejection from Program

Panel D: Understanding of Ejection CriteriaHH has no OVCs Below 18 Years Old 0.143 0.119 0.496 0.522 1092

At Least One Program Rule is Ignored for 0.190 0.289 0.179 0.250 1092Three Consecutive Pay Periods

HH Moves to Non-Program District 0.019 0.004 0.183 0.234 1092

HH Does Not Collect Transfer for Three 0.011 0.013 0.952 0.952 1092Consecutive Pay Periods

Panel E: Summary TestIndex of Knowledge and Understanding3 4.033 4.260 0.213 0.283 882

1 P-values on differences in column (3) are estimated using the typical clustered standard error formula,clustered at the sub-location level. * p < .01, ** p < .05, *** p < .01, † p < .001. P-values in column(4) are estimated using the wild cluster bootstrap. If clustered at district level, all differences becomeinsignificance except for that of "Knows About Fining as Punishment".

2 Index of Knowledge and Understanding is an unweighted linear combination of all of the abovevariables, except for "Adequate Food and Nutrition for Children", which means its support is from 0to 14.

To summarize, our analyses demonstrate that, for the most part, households facing hard

conditions had similar beliefs about program rules and penalties as households with only la-

beling. We therefore hypothesize that any changes in household consumption, dietary diversity

19

Figure 2

or school attendance through assignment to hard conditions would be primarily caused by the

financial burden associated with receiving penalty fines. Based on the prior literature discussed

above (Fiszbein et al., 2009), we also hypothesize that being fined will disproportionately af-

fect outcomes in poorer households. At the time of the 2009 follow-up survey, almost 16 per-

cent of recipient households reported ever having their cash transfer reduced by a penalty fine

(and nearly all were in the hard conditions treatment arm). However, while households from

across the baseline consumption distribution received fines, our analysis reported below shows

that households from the lower end of the distribution reported being fined more frequently. If

poorer households were less capable of complying with conditions and avoiding being fined,

they may also be in a weaker position to avert the negative consequences of reductions in their

cash transfers.

In the remainder of this section, we first lay out our empirical specification for estimating

the impact of assignment to hard conditions on household per adult-equivalent consumption,

20

dietary diversity, and school absences (for children aged 6-17 years). We also present two

tables that assess balance among baseline characteristics of study participants: one comparing

households in the hard conditions arm to those in the labeling only arm, and one comparing

cash transfer recipients to control households (who did not receive transfers). These serve as

checks that randomization was implemented correctly at both stages.

4.2 Empirical specifications

As stated above, we are primarily interested in how assignment to hard conditions affects

household outcomes, based on the households’ location in the baseline wealth distribution. We

specify the following linear model in order to estimate these impacts: .

yi jk,2009 = α +δ1yi jk,2007+δ2hardi jk+δ3totalconsi jk,2007

+δ4hardi jk× totalconsi jk,2007+X′i jk,2007β1+ei jk

(1)

Our sample for this estimation is comprised entirely of households i in sub-location j and

district k that received the cash transfer in the Kenya CT-OVC program7. The variable yi jk,2009

represents the follow-up (2009) survey value of our outcome variables, yi jk,2007 represents the

baseline outcome values, and hardi jkis a binary indicator for whether the household was lo-

cated in a district (or sub-location) assigned to hard conditions. The existing evidence base

(discussed above) suggests that we should pay special attention to the heterogeneous effects

of being fined, particularly according to household wealth. For this reason, we also examine

the marginal effects of assignment to hard conditions on our outcomes by including an in-

teraction term between a household’s assignment status and its baseline per adult-equivalent

consumption. Lastly, Xi jk,2007 is a vector of baseline household-level control variables includ-

ing household size, the sex of the household head, whether or not a member works for wage

employment, whether the household receives a transfer that is external to the program, whether

the household owns livestock, the number of agricultural acres the household owns, measures

used by program officials to target households for the transfer, and an indicator for whether

the household was in a sub-location assigned to the transfer. The model specifications with the

7Note that there were a few housedholds that received the transfer despite not being assigned to treatment inthe CT-OVC program. They are retained in these regressions and we control for their presence.

21

number of days missed from school as the outcome are not estimated at the household level;

rather, the unit of analysis is a child aged 6-17 in the household.

Lastly, any change in outcomes due to hard conditions will only make sense if household

outcomes were affected by assignment to treatment (cash transfer receipt). Indeed, the Kenya

CT-OVC Evaluation Team (2012) found in their differences-in-differences impact analysis that

being in a sub-location randomly assigned to the CT-OVC cash transfer was associated with

increases in household consumption of both food and non-food items. We first replicate their

findings and then extend the analysis to examine the impacts of hard conditions on consump-

tion, dietary diversity and schooling outcomes. Our specification for the impact of the transfer

is given below, where CTi j is a binary indicator of whether or not household i was assigned

to treatment due to its residence in sublocation j. We also look at the marginal effects of

assignment to treatment (cash transfer receipt) by baseline levels of per adult-equivalent con-

sumption.

yi j,2009 = µ +γ1yi j,2007+γ2CTi j +γ3totalconsi j,2007

+γ4CTi j × totalconsi j,2007+X′i j,2007β2+vi j

(2)

4.3 Balance tests

In the Kenya CT-OVC program, assignment to hard conditions within the treatment arm

was done randomly at the district level, with the exception of the Nairobi District, where it

was conducted at the sub-location level. If random assignment to hard conditions worked as

intended, we would expect it to produce two statistically equivalent groups of program bene-

ficiaries (treated with and without hard conditions) at baseline. Table 2 presents the results of

our tests for equality of means of various household characteristics between the two treatment

arms at the outset of the experiment. We also display the results from a balance test between

cash transfer andcontrol (nonbeneficiary) groups in Table A.2 to confirm that the initial ran-

dom assignment was implemented correctly as well. The household characteristics we test are

based on the variables used in Annex F of Ward et al. (2010) to test for balance across the treat-

ment arms. The sample used for these comparisons is the same sample we use in estimating

22

our main specifications8.

The results in Table 2 indicate that balance between these groups is achieved, that is, there

are no statistically significant differences in means (at the 5% level) in their observable charac-

teristics at baseline9. We cluster standard errors and p-values at the sub-location level, which

is recommended by Oxford Policy Management based on their sampling methodology for the

CT-OVC evaluation data collection. However, results are very similar if we cluster by dis-

trict instead. Some believe that a more appropriate test for balance involves comparing the

standardized differences in means between the treatment and control groups to some threshold

value. Imbens and Wooldridge (2009) suggest ±0.25 as heuristic values for this threshold, be-

low which linear regression should not be sensitive to the inclusion of the variable in question.

Of course, this also depends on the degree of the correlation between said variable and the

outcomes. Table 2 includes a column of the standardized differences in the means of baseline

characteristics between each treatment arm10. Only three differences are marginally larger in

magnitude than 0.25, and only one of these is greater than 0.30. We control for these variables

in our specifications, though they alter the results negligibly. While this gives us confidence

that random assignment to hard conditions achieved the intended result, we still adjust (con-

trol) for additional characteristics such as rural location and agricultural land ownership in our

main specifications to improve the efficiency of our estimation (Gennetian et al., 2006).

Table A.2 indicates that the initial randomization between the CT-OVC treatment and non-

beneficiary control group was also successful, as none of the differences in means are statis-

tically significant at the 5% level. The standardized differences are less informative in this

case, because households in the treatment sub-locations were explicitly prioritized to receive

the transfer based on the three baseline vulnerability criteria mentioned in Section 3, the bal-

ance of which we assess in Table A.2 as well. This implies that tests of balance or evaluations

of the program should control for these criteria in estimations, which is what we do in Table

8Some variables have missing values for a subset of the observations in our estimation sample. In these cases,we cannot include the variables as controls in the specifications and display them here for illustrative purposes.

9Since our paper’s main results deal with heterogenous treatment effects, we also run balance tests afterseparating households into bins by baseline consumption quintile. Across all five bins and 29 variables, only twodifferences are significant at the 5% level: poultry owned at baseline in the 40th-60th percentile bin and receiptof an external transfer in the 60th-80th percentile bin. This is no more than one might expect from chance.

10The formula for a standardized difference is µ1−µ0√(σ2

1−σ20 )/2

where µ1 and µ0 are the means of the variable and

σ21 and σ

20 are its variances, seperated by treatment arm.

23

Table 2: Balance Table for Hard Conditions Versus Labeling Only

(1) (2) (3) (4) (5) (6)Labeling Only Hard Conditions P-Value on Diff. Boot P-Value Standardized. Diff. N

Years of Edu. of HH Head 5.808 6.057 0.577 0.617 -0.084 511

Sex of HH Head 0.357 0.347 0.695 0.702 0.021 1092

HH Receives Labor Wages 0.043 0.026 0.544 0.637 0.098 1092

HH Receives Outside Transfer 0.349 0.219 0.155 0.184 0.290 1092

Chronically Ill Members of HH 0.117 0.085 0.321 0.340 0.095 1092

Carer Age Index 0.468 0.516 0.262 0.283 -0.278 1092

Number of OVCs in HH 2.714 2.617 0.481 0.516 0.062 1092

Poor Quality Walls 0.683 0.783 0.443 0.476 -0.227 1092

Poor Quality Floor 0.717 0.740 0.857 0.868 -0.053 1092

HH Owns Livestock 0.825 0.755 0.567 0.592 0.171 1092

Cattle Owned 1.177 1.445 0.247 0.311 -0.133 868

Poultry Owned 3.903 5.144 0.188 0.193 -0.212 868

Owns Telephone 0.108 0.104 0.989 0.983 0.011 1092

Owns Blanket 0.823 0.866 0.744 0.748 -0.118 1092

Owns Mosquito Net 0.648 0.551 0.216 0.224 0.198 1092

Acres of Land Owned 1.387 1.853 0.326 0.312 -0.236 1092

Household in Rural Location 0.879 0.762 0.513 0.536 0.310 1092

HH Cons. 2007 1.651 1.525 0.362 0.382 0.132 1092

HH Food Cons. 2007 0.963 0.927 0.550 0.593 0.057 1092

HH Non-food Cons. 2007 0.688 0.598 0.400 0.442 0.168 1092

Dietary Diversity Score 2007 4.976 5.291 0.296 0.309 -0.210 1092

Size of the HH 5.410 5.513 0.852 0.856 -0.039 1092

Age of HH Head 56.540 59.87 0.331 0.345 -0.244 1085

People Aged 0-5 in HH 1.600 1.702 0.505 0.517 -0.099 448

People Aged 6-11 in HH 1.648 1.755 0.266 0.269 0.118 790

People Aged 12-17 in HH 1.732 1.736 0.956 0.957 -0.004 861

People Aged 18-45 in HH 1.855 2.023 0.469 0.499 -0.142 648

People Aged 46-64 in HH 1.111 1.100 0.576 0.575 0.036 641

People Aged 65+ in HH 1.132 1.103 0.435 0.441 0.086 389

1 P-values on differences in column (3) are estimated using the typical clustered standard error formula, clusteredat the sub-location level. * p < .01, ** p < .05, *** p < .01, † p < .001. P-values in column (4) are estimatedusing the wild cluster bootstrap. Differences remain insignificant if clustering is done at the district level.

2 The Carer Age Index is a scale from 0 to 1, values closer to 1 indicating that the age of the main caregiver inthe household is very far from 18 (in either direction). This was an important criterion used in targeting for thetransfer.

3 Consumption is in terms of 1000 KSh per adult-equivalent.

24

A.2 (emphasizing the t-test results over the standardized differences). As one can see in the

table, the Carer Age Index (the main vulnerability criterion) is imbalanced between groups

(according to the standarized differences). Controlling for it not only achieves balance for the

Carer Age Index itself, but also balances several other characteristics, especially those related

to caregivers’ ages, according to the t-tests. We also control for the number of chronically ill

individuals and OVCs in the households, the two other vulnerability criteria, in our regressions

that generate the p-values.

5 Impact Estimation Results

In this section, we first undertake an intent-to-treat (ITT) analysis of assignment to the

CT-OVC program and its effect on our outcome measures. Next, we estimate the relationship

between assignment to hard conditions and our outcomes among transfer recipients, according

to each household’s place in the baseline wealth (proxied by consumption) distribution. Lastly,

we further discuss the mechanisms by which hard conditions likely impacted household out-

comes.

5.1 Impacts of labeled cash transfers

In investigating whether assignment to hard conditions affected household and child out-

comes, we first estimate equation (2) to confirm (as shown in prior studies) that being ran-

domly assigned to receive cash transfers had an impact on households. As assignment to the

cash transfer program was random at the sub-location level, we compare households in sub-

locations randomly assigned to receive the cash transfer to those in sub-locations randomized

to serve as controls. Like the Kenya CT-OVC Evaluation Team (2012), we find that cash

transfer receipt was associated with significant increases in consumption. Specifically, we find

that assignment to the transfer increased total per adult-equivalent consumption by around 340

KSh per month on average. This represents a 19-22% increase over households’ monthly per

adult-equivalent consumption at baseline. Scaling this effect by 4.42, the average number of

adult-equivalent individuals in the household, implies that the transfer typically increased total

household consumption by about 1511 KSh per month, which is almost exactly the size of the

25

Table 3: Impacts of Assignment to Cash Transfer on Outcomes

(1) (2) (3) (4) (5)Dep. Variable: HH Cons. 2009 HH Food Cons. 2009 HH Non-food Cons. 2009 Dietary Div. Score 2009 Days Missed 2009

Panel A: Average EffectsAssigned Transfer 0.378∗∗ 0.249∗∗ 0.119 0.778∗∗ -0.218

(0.170) (0.097) (0.100) (0.336) (0.344)[0.050] [0.032] [0.278] [0.066] [0.573]

HH Cons. 2007 0.196∗∗∗ 0.028 -0.019 0.213∗∗ 0.063(0.068) (0.050) (0.048) (0.103) (0.182)[0.030] [0.611] [0.725] [0.103] [0.762]

Assigned Transfer × HH Cons. 2007 -0.021 -0.026 0.011 -0.249∗∗ -0.169(0.074) (0.047) (0.042) (0.104) (0.189)[0.786] [0.601] [0.781] [0.040] [0.452]

Panel B: Marginal Effects by HH Cons. 2007Percentile: 10 0.364∗∗ 0.231∗∗ 0.127 0.610∗∗ -0.331

(0.132) (0.073) (0.079) (0.282) (0.194)[0.022] [0.016] [0.131] [0.070] [0.214]

Percentile: 25 0.358∗∗∗ 0.223∗∗∗ 0.130∗ 0.539∗∗ -0.380∗

(0.119) (0.065) (0.071) (0.261) (0.218)[0.015] [0.007] [0.087] [0.079] [0.100]

Percentile: 50 0.348∗∗∗ 0.212∗∗∗ 0.135∗∗ 0.427∗ -0.455∗∗

(0.104) (0.057) (0.062) (0.233) (0.193)[0.009] [0.003] [0.048] [0.109] [0.021]

Percentile: 75 0.336∗∗∗ 0.198∗∗∗ 0.142∗∗ 0.291 -0.547∗∗

(0.098) (0.057) (0.058) (0.207) (0.209)[0.004] [0.003] [0.025] [0.215] [0.017]

Percentile: 90 0.318∗∗ 0.177∗∗ 0.151∗∗ 0.086 -0.685∗∗

(0.117) (0.076) (0.068) (0.195) (0.305)[0.017] [0.033] [0.055] [0.706] [0.057]

Controls X X X X XObservations 1530 1530 1530 1530 2717

1 Clustered standard errors in parentheses, * p < .1, ** p < .05, *** p < .01, † p < .001. P-values generated from the wild cluster bootstrapare in brackets. The standard errors on the marginal effects are estimated by applying the delta-method to the variance-covarianceestimator of the preceding estimation.

3 Household consumption is measured in terms of 1000 Kenyan Shillings (KSh) per adult-equivalent.

26

monthly transfer. If we break total consumption into food and non-food components, we find

that while food consumption increased fairly evenly across the income distribution, the only

statistically significant increases in non-food consumption were seen at the wealthier end of

this distribution. Additionally, we find that the only statistically significant effects of the cash

transfer on dietary diversity are among the poorer households, which become insignificant at

the 5% level when using the wild cluster bootstrap’s p-values. Lastly, we find that cash transfer

receipt among children in households from the upper-middle of the consumption distribution

contributed to their improved school attendance (i.e., missing slightly fewer days of school

per month). We present these estimated impacts in Table 3. Overall, these results indicate that

being assigned to receive the cash transfer was associated with increased levels of consumption

for everyone, although as expected, responses in terms of the types of consumption varied by

baseline wealth level.

5.2 Impacts of hard conditions

With confirmation that the cash transfers benefitted households and information on the

magnitude of these effects, we can use these estimates to bound our expectations for our

estimation of the impacts of assignment to hard conditions. The results from estimating

equation (1) are displayed in Table 4, along with the marginal effects of hard conditions by

consumption percentile. Like Ward et al. (2010), we do not observe any statitically signifi-

cant impacts of hard conditions on our outcomes when estimating average impacts across all

households. This is consistent with the findings of Benhassine et a l. (2013), who compared

the effects of CCTs vs. LCTs in Morroco and likewise found little to no effect of imposing

conditions (with penalties) on a labeled cash transfer. However, a focus on the average impact

ignores consistent findings in the prior literature (discussed above) showing that the impacts of

CCTs (and how households respond to conditions) are notably heterogenous across the wealth

distribution (Fiszbein et al., 2009). The marginal effects in Table 4 show that assignment

to hard conditions is associated with significant d ecreases i n r elatively p oorer households’

non-food consumption per adult-equivalent. Households in the 10th and 25th percentiles of

baseline consumption saw average decreases of 174 and 147 KSh, respectively, per month at

27

Table 4: Impacts of Assignment to Hard Conditions

(1) (2) (3) (4) (5)Dep. Variable: HH Cons. 2009 HH Food Cons. 2009 HH Non-food Cons. 2009 Dietary Div. Score 2009 Days Missed 2009

Panel A: Average EffectsHard Conditions -0.336∗∗ -0.093 -0.239∗∗ -0.026 -0.347

(0.150) (0.097) (0.081) (0.249) (0.273)[0.087] [0.393] [0.024] [0.928] [0.268]

HH Cons. 2007 0.122∗∗∗ -0.018 -0.039 -0.109 -0.107(0.034) (0.053) (0.032) (0.060) (0.076)[0.025] [0.734] [0.295] [0.179] [0.255]

Hard Conditions × HH Cons. 2007 0.194∗∗ 0.097∗ 0.090 0.314∗∗∗ 0.164(0.083) (0.052) (0.038) (0.079) (0.155)[0.132] [0.168] [0.153] [0.002] [0.337]

Panel B: Marginal Effects by HH Cons. 2007Percentile: 10 -0.207∗ -0.029 -0.174∗∗∗ 0.184 -0.237

(0.114) (0.076) (0.058) (0.228) (0.187)[0.130] [0.703] [0.019] [0.473] [0.283]

Percentile: 25 -0.152 -0.003 -0.147∗∗ 0.271 -0.192(0.104) (0.069) (0.052) (0.222) (0.158)[0.213] [0.980] [0.018] [0.306] [0.295]

Percentile: 50 -0.064 0.041 -0.103∗∗ 0.414∗ -0.117(0.099) (0.064) (0.038) (0.216) (0.128)[0.579] [0.569] [0.084] [0.157] [0.402]

Percentile: 75 0.044 0.094 -0.048 0.589∗∗ -0.026(0.111) (0.068) (0.061) (0.217) (0.139)[0.728] [0.229] [0.487] [0.059] [0.847]

Percentile: 90 0.203 0.172∗ 0.031 0.846∗∗∗ 0.109(0.154) (0.090) (0.093) (0.235) (0.225)[0.308] [0.130] [0.797] [0.008] [0.644]

Controls X X X X XObservations 1092 1092 1092 1092 1873

1 Clustered standard errors in parentheses, * p < .1, ** p < .05, *** p < .01, † p < .001. P-values generated from the wild cluster bootstrapare in brackets. The standard errors on the marginal effects are estimated by applying the delta-method to the variance-covarianceestimator of the preceding estimation. Nothing substantial changes if clustering is done by district, except that the increase in foodexpenditure for the 90th percentile becomes significant at the 0.05 level only according to both the normal clustering and wild clusterbootstrap approaches.

2 Household consumption is measured in terms of 1000 Kenyan Shillings (KSh) per adult-equivalent.

28

follow-up. These reductions amount to about half the increase in consumption experienced by

CT-OVC beneficiaries, and accordingly represent substantial negative effects.11

As hypothesized, we believe these observed reductions in consumption for households at

lower levels of baseline household consumption are primarily driven by financial penalties (or

fines) levied against these households. While we do not see a similar impact of hard conditions

on food consumption for poor households, the CCT literature in sub-Saharan Africa (Handa et

al., 2018) suggests that this is most likely because these extremely poor households were

already living on subsistence levels of food consumption and further reductions were untenable.

Instead, it appears that households bear the brunt of the penalities in non-food consumption

reductions, which include more categories of discretionary spending.

Interestingly, we also find that under hard conditions, households in the 90th percentile of

baseline consumption increased the number of food groups consumed by about 0.85, on average

(see again Table 4). Based on the qualitative research conducted in the Kenya CT-OVC

evaluation and the broader cash transfer program literature, we speculate that comparatively

wealthier households were better equipped to respond to the labeling of the cash transfer that

encouraged increased dietary diversity (communicated in informational sessions for CT-OVC

beneficiaries). Additionally, we believe that these households were also more strongly mo-

tivated than those in the “labeling only” arm by the desire to avoid penalty fines on their

transfers if they did not make these efforts. Relatively poorer households in the hard conditions

treatment arm who were closer to subsistence levels of consumption would have been less

capable, however, of responding similarly. It is also important to point out that because the

implementation of conditions varied widely across and also within locations where they were

imposed (Ward et al, 2010), even within the same consumption percentile, households in the

hard conditions treatment arm almost certainly experienced heterogeneous consequences of

being exposed to hard conditions, which we further discuss below.

The third area of impacts of assignment to hard conditions that we investigated was

educational outcomes, recognizing that CCTs have traditionally been used as instruments for

11These results are consistent with estimates generated in our two-stage least squares (2SLS) instrumental variables (IV) impact estimation, reported in an earlier version of the paper. As expected given that assignment to the hard conditions treatment arm is estimated using random assignment as an instrument, the standard errors on the IV impact estimates are larger. These results are available from the authors upon request.

29

incentivizing investment in education. Our results in Table 4 do not identify any effects on

the number of days missed from school in the past month across the consumption distribution.

These findings are consistent with Benhassine et al.’s study of CCTs vs. LCT’s in Morocco

(2013), which failed to find economically meaningful effects on education outcomes of impos-

ing hard conditions. However, this does not imply that the reductions in nonfood consumption

experienced by poorer households were unrelated to educational purchases. In fact, the house-

hold survey was explicitly designed to identify education-related expenses separately from

other nonfood household consumption. We estimated equation (1) again using household ed-

ucational expenditures as the outcome variableto identify the impact (and marginal effects) of

hard conditions on household spending on education. Similar to the patterns of effects reported

in Table 4, our estimates (available from the authors) show that households in the lowest end

of the baseline consumption distribution reduced education spending on average by about 43

Ksh per adult-equivalent per month at follow-up (p-value for the wild cluster bootstrap = 0.049

clustered at sub-location level).

While the heavy program emphasis and labeling of the CT-OVC as a transfer to support the

education and welfare of orphans in the household may have mitigated any effects of fining on

children’s school attendance in the short timeframe between baseline and follow-up (especially

given the slow roll out of hard conditions), we do not claim that there would be no longer-term

effects if these reductions in household expenditures continued over a longer time horizon.

We suggest that our findings of reduced educational expenditures are further cause for

concern about the potential longer-term negative impacts of CCTs with hard conditions

(compared to LCTs). This may be particularly true when the cash transfer amount is a

larger fraction of the household budget, as in the Kenya CT-OVC program, but unlike

Morocco’s Tayssir program studied by Benhassine et al. (2013).

5.3 Futher discussion of results

As discussed above, we expected that comparatively wealthier households likely had greater

means to act on their knowledge about potential fines in order to prevent them, or at least mit-

igate their effects. In other words, less resource-constrained households could take greater

30

measures to conform to the perceived rules of the program. In the Kenya CT-OVC program

case, the most salient expectation, communicated in informational sessions for program par-

ticipants, related to providing adequate food and nutrition to OVCs. If relatively wealthier

households had more capacity for attending these sessions and following through on the guid-

ance offered, this could explain why the change in expenditures among wealthier households

was concentrated in food consumption and contributed to increased dietary diversity. As Ta-

ble 1 (Panel A) showed, at least 60 percent of households in both treatment arms believed

they needed to provide adequate nutrition in order to continue receiving the transfer. In fact,

households were over two times more likely to believe that providing adequate nutrition was

a rule than they believed that enrolling their OVCs in school was a condition for continuing

to receive the transfer, even though only the latter was an actual rule for avoiding penalties.

Additionally, we plot in Figure 3 the marginal effects by baseline consumption from a regres-

sion of the belief that the household could be fined for noncompliance on assignment to hard

conditions (with control variables).12 The marginal effect on assignment to hard conditions

increases along the consumption distribution, and the estimate for households in the 90th con-

sumption percentile is almost double that of households in the 10th percentile (although the

difference between the two is not statistically significant at the 5% level). We believe that this

pattern, combined with the greater resources at their disposal to respond behaviorally, provides

a credible explanation for why only relatively wealthier households increased dietary diversity

in response to the belief that they could be fined..

Additionally, while households from across the entire consumption distribution received

fines, households from the lower end reported having received them more frequently. Figure 4

presents results using the same specification as the regression that produced Figure 3, but with

an indicator for whether or not the household was ever fined as the dependent variable. The

marginal effects show that households in the 10th consumption percentile were fined about 50

percent more frequently than households in the 90th percentile. In fact, all of these marginal

effects are (statistically) significantly different from each other at the 0.1% level. These results

appear to confirm that relatively poorer households were less financially capable of responding

12In Figures 3 and 4, the bars plot standard errors according to the White (1984) formula and the stars refer top-values from the wild cluster bootstrap procedure.

31

Figure 3

32

Figure 4

behaviorally to reduce the risk of being fined and avoid additional penalties. Lastly, it is

well-documented that relatively poorer households are less capable of smoothing consumption

after negative income shocks (such as those associated with fines) than relatively wealthier

households in developing countries (Alderman, 1996; Jalan and Ravallion, 1999; Reardon and

Tayler, 1996). This gives further credence to our findings of greater negative effects of hard

conditions on consumption among poorer households in the Kenya CT-OVC program.

6 Conclusion

In a 2013 blog post13, Berk Ozler characterized efforts to describe or define cash transfer

programs as “an unconditional mess,” arguing that the distinctions between CCTs and UCTs

were “too blurry” and that interested stakeholders (donors, policymakers) would be better off

13https://blogs.worldbank.org/impactevaluations/defining-conditional-cash-transfer-programs-unconditional-mess

33

thinking about them along a “continuum from a pure UCT to a heavy-handed CCT”. Our re-

search further suggests that a particular cash transfer program, such as the Kenya CT-OVC

program, may not correspond to a single point along such a continuum. Indeed, our examina-

tion of the the Kenya CT-OVC program shows that where it fits along a continuum from fully

unconditional to “hard” conditions may depend on the implementation of the program as expe-

rienced by households. And as Ozler opined and we found in this research, there are tradeoffs

for household outcomes in terms of how the conditions (or lack thereof) are implemented.

Our findings show that the imposition of hard conditions in the Kenya CT-OVC program–i.e.,

a “heavy-handed” implementation of CCTs that monetarily penalized families for their fail-

ure to comply with program conditions–had tradeoffs for the well-being of targeted families,

depending on their baseline poverty or wealth. These findings are consistent with prior litera-

ture showing that the effects of CCTs on household consumption vary according to household

baseline wealth (or depth of poverty) (Fiszbein et al., 2009; Hoddinott, Skoufias, and Wash-

burn 2000; Macours, Schady, and Vakis, 2008; Maluccio and Flores 2005). Indeed, one of the

more compelling aspects of our estimates showing that the consumption of poorer households

may be harmfully reduced, while that of better-off families may be improved through the im-

position of fines, is that they are largely consistent with what development practitioners and

researchers have long suspected (even if debate in the literature is ongoing).

Having a program where households face penalities for not complying with expectations to

spend cash transfers wisely (or for the benefit of the children) is a potentially promising way to

achieve the broader goals of cash transfers programs, that is, to reduce not only poverty but also

the intergenerational transmission of poverty. Our findings show that comparatively wealthier

households not only have more resources to circumvent any immediate consumptions losses

that a fine might present, but they may also use these resources to respond behaviorally in ways

that help them to avoid being fined. While providing adequate food and nutrition was not an

explicit program rule of the CT-OVC program that was punishable (if violated) by a fine, over

70 percent of households assigned to hard conditions thought that it was a formal requirement.

In fact, far more household respondents believed that it was a program rule than the number

who knew the correct rules (about schooling and health conditions), as we showed in Table

1. We argue that this gives credibility to the noticeable pivot toward greater dietary diversity

34

among households who had the resources to respond to the risk of being fined.

On the other hand, researchers and practitioners have long been concerned about the undue

burdens that conditional cash transfers place on the poorest of the poor. Not only is complying

with rules more challenging for them, but penalizing their transfers may cut them off from pur-

chasing the most basic of necessities that their more meager budgets afford. Regrettably, this

is what appears to have happened in the case of the Kenya CT-OVC program, where poorer

households experienced decreases in consumption after getting fined. We found that, for com-

paratively poor households (in the 10th and 25th percentiles of baseline consumption), getting

fined is associated with a decrease in monthly consumption at follow-up of about 174 and 147

KSh, respectively, which corresponds to about half the increase in consumption experienced

by CT-OVC beneficiaries. In this particular case, it appears that while imposing monetary

penalties for not following program rules provided some additional motivation for compar-

atively wealthier households to use their resources to the greater benefit of their household

members, it came at the cost of regressive impacts on the very poor that may have ultimately

contributed to lasting harms. Our research also suggests that concerns are warranted for po-

tential longer-term negative effects of being fined on household and child outcomes, which we

could not fully explore given the timeframe between baseline and follow-up data collection

and the slow implementation of the hard conditions and associated fines in the Kenya CT-OVC

program.

Surprisingly, given the expansive literature that has emerged over time on CCTs and UCTs

(and now LCTs), we found little empirical exploration of the consequences of experiencing fi-

nancial penalties (or suspension or termination of benefits) for households and children receiv-

ing cash transfers. The Kenya CT-OVC program evaluation design with random assignment to

hard conditions that supported our identification of impacts via the mechanism of being fined

may have allowed us a unique opportunity to examine the consequences of financial penalties

in CCTs in terms of household and children’s outcomes. That said, our study is not without

limitations. As noted above, we are only examining the impact of fines on households within

a two-year window of program implementation, and we do not have detailed data to identify

the frequency or timing of penalties that households experienced in this program. Ideally, we

would have had better data to explore a fuller range of impacts of being fined on household and

35

children’s well-being, but we are constrained by sample sizes within the CT-OVC treatment

group and by the fact that many outcomes were measured only for age-appropriate subgroups.

Lastly, while we believe that we have presented compelling evidence to argue that our iden-

tification of the impacts of being fined in the Kenya CT-OVC program are plausibly causal,

we tip our hat to the statistician, George Box (1976: 792), who articulated the view that “all

models are wrong.”

References

[1] Alderman, H. (1996). Savings and economic shocks in Pakistan. Journal of Development

Economics, 51(2): 343-65.

[2] Baird, S., Ferreira, F.H.G., Ozler, B., Woolcock, M. (2013). “Relative Effective-

ness of Conditional and Unconditional Cash Transfers for Schooling Outcomes in

Developing Countries: A Systemic Review”. Campbell Systematic Reviews. DOI:

10.4073/csr.2013.8

[3] Balfanz, R., & Byrnes, V. (2012). “The importance of being in school: A report on

absenteeism in the nation’s public schools”. Baltimore, MD: Johns Hopkins University.

[4] Bastagli, F., Hagen-Zanker, J., Harman, L., Barca, V., Sturge, G., Schmidt, T.., Peller-

ano, L. (2016). “Cash transfers: what does the evidence say?”. Overseas Development

Institute. https://www.odi.org/sites/odi.org.uk/files/resource-documents/10749.pdf

[5] Benhassine, N., Florencia D., Esther D., Pascaline D., and Victor P. (2015). “Turning a

Shove into a Nudge? A "Labeled Cash Transfer" for Education”. American Economic

Journal: Economic Policy, 7(3): 86-125.

[6] Box, G. E. P. (1976). Science and Statistics. Journal of the American Statistical Associa-

tion, 71: 791–799, doi:10.1080/01621459.1976.10480949.

[7] Cameron, A.C., Gelbach, J.B., and Miller, D.L. (2008). “Boostrap-Based Improvements

for Inference with Clustered Errors”. Review of Economics and Statistics 90 (3): 414-

427.

36

[8] Deaton, A. and Zaidi, S. (2002). “Guidelines for Constructing Consumption Aggregates

for Welfare Analysis”. LSMS Working Paper, No. 135. World Bank.

[9] de Brauw, Alan, and John Hoddinott. (2011). “Must conditional cash transfer programs

be conditioned to be effective? The impact of conditioning transfers on school enrollment

in Mexico.” Journal of Development Economics 96 (2): 359–70.

[10] Fernald, L. C. H., P. J. Gertler, and L. M. Neufeld. (2008). “The Role of Cash in Con-

ditional Cash Transfer Programmes for Child Health, Growth, and Development: An

Analysis of Mexico’s Oportunidades.” The Lancet 371 (9615): 828–37.

[11] Ferreira, Francisco H. G. (2008). “The Economic Rationale for Conditional Cash Trans-

fers.” Unpublished manuscript, World Bank, Washington, DC.

[12] Fiszbein, A., Kanbur, R., & Yemtsov, R. (2014). “Social protection and poverty reduc-

tion: Global patterns and some targets”. World Development, 61, 167–177.

[13] Fiszbein, A., and N. Schady, with F. H. G. Ferreira, M. Grosh, N. Kelleher, and others.

(2009). “Conditional Cash Transfers: Reducing Present and Future Poverty”. Washing-

ton, DC: World Bank.

[14] Garcia, M., Moore, C.M.T. (2012) “The Cash Dividend”. The World

Bank. http://documents.worldbank.org/curated/en/435291468006027351/pdf

672080PUB0EPI0020Box367844B09953137.pdf.

[15] Gennetian, L.A., Morris, P.A., Bos, J.M., and Bloom, H.S. (2006). “Constructing in-

strumental variables from experimental data to explore how treatments produce effects”.

Learning More from Social Experiments: Evolving Analytic Approaches, 75-114.

[16] Gershenson, Seth, Alison Jacknowitz, and Andrew Brannegan. (2017). “Are student ab-

sences worth the worry in U.S. primary schools?” Education Finance and Policy, 12(2):

137-165.

[17] Golan, J., Sicular, T & Umapathi, N. (2015). “Unconditional cash transfers in China : an

analysis of the rural minimum living standard guarantee program”. World Bank Policy

Research Working Paper WPS7374.

37

[18] Gottfried, Michael A. (2009). “Excused versus unexcused: How student absences in ele-

mentary school affect academic achievement”. Educational Evaluation and Policy Anal-

ysis 31(4):392–419.

[19] Government of Kenya, Office of the Vice President and Ministry of Home Affairs. (2006).

Program Design, Cash Transfer Pilot Project. Nairobi.

[20] Grosh, M., Del Ninno, C., Tesliuc, E., & Ouerghi, A. (2008). “For protection and pro-

motionr: The design and implementation of effective safety nets”. Washington D. C: The

World Bank.

[21] Jalan, J. and Ravallion, M. (1999). “Are the poor less well insured? Evidence on vulner-

ability”. Journal of Development Economics 58: 61-81.

[22] Handa, S., C. T. Halpern, A. Pettifor, and H. Thirumurthy. (2014). “The Government of

Kenya’s Cash Transfer Program Reduces the Risk of Sexual Debut among Young People

Age 15–25.” PLoS One 9 (1): e85473.

[23] Heckman, James J., Jora Stixrud, and Sergio Urzua. (2006). “The effects of cognitive and

noncognitive abilities on labor market outcomes and social behavior”. Journal of Labor

Economics 24(3):411–482.

[24] Heinrich, C.J. and Brill, R. (2015). “Stopped in the Name of the Law: Administrative

Burden and its Implications for Cash Transfer Program Effectiveness.” World Develop-

ment, Vol. 72: 277–295.

[25] Heinrich, C.J. (2016). “The Bite of Administrative Burden: A Theoretical and Empirical

Investigation. Journal of Public Administration Research and Theory, 26 (3): 403-420.

[26] Hidrobo, M., Hoddinott, J., Kumar, N. & Oliver, M. (2018). “Social Protection, Food

Security, and Asset Formation”. World Development 101: 88–103.

[27] Hurrell, A., Ward, P. & Merttens, F. (2008). “Kenya CT-OVC Programme Operational

and Impact Evaluation Baseline Survey Report: Final Report”. Oxford Policy Manage-

ment, July.

38

[28] Imbens, G. W. and Wooldridge, M. T. (2009). “Recent developments in the econometrics

of program evaluation”. Journal of Economic Literature 47 (1): 5-86

[29] Levy, Santiago. (2006). “Progress Against Poverty: Sustaining Mexico’s PROGRESA-

Oportunidades Program”. Washington, DC: Brookings Institution Press.

[30] Lindert, Kathy, Anja Linder, Jason Hobbs, and Bénédicte de la Brière. (2007). “The Nuts

and Bolts of Brazil’s Bolsa Família Program: Implementing Conditional Cash Trans-

fers in a Decentralized Context.” Social Protection Discussion Paper 0709, World Bank,

Washington, DC.

[31] Mont, Daniel. (2006). “Disability in Conditional Cash Transfer Programs: Drawing on

Experience in LAC.” Report prepared for the Third International Conference on Condi-

tional Cash Transfers, Istanbul, Turkey, June 26–30.

[32] Morais de Sá e Silva, M. (2017). “Poverty Reduction, Education, and the Global Diffu-

sion of Conditional Cash Transfers”. Palgrave Macmillan.

[33] Nield, Ruth C., and Robert Balfanz. (2006). “An extreme degree of difficulty: The ed-

ucational demographics of urban neighborhood high schools”. Journal of Education for

Students Placed at Risk 11(2):123–141.

[34] Palma, J. & Urzúa, R. (2005). “Anti-poverty Policies and Citizenry: The “Chile Soli-

dario” Experience”. United Nations Educational, Scientific and Cultural Organisation,

Policy Papers 12.

[35] Ralston, L., Andrews, C. & Hsiao, A. (2017). “The Impacts of Safety Nets in Africa

What Are We Learning?” Policy Research. World Bank: Social Protection and Labor

Global Practice Group & Africa Region Working Paper 8255.

[36] Reardon, T. & Taylor, J. E. (1996). “Agroclimatic shock, income inequality, and poverty:

Evidence from Burkina Faso”. World Developent 24(5): 901-914.

[37] Rumberger, Russell W., and Scott L. Thomas. (2000). The distribution of dropout

and turnover rates among urban and suburban high schools. Sociology of Education

73(1):39–67.

39

[38] Silva, Maria Ozanira da Silva e. (2007). (O Bolsa Família: problematizando questões

centrais na política de transferência de renda no Brasil”. Ciência & Saúde Coletiva, 12(6),

1429-1439.

[39] The Kenya CT-OVC Evaluation Team (2012). “The impact of the Kenya Cash Trans-

fer Program for Orphans and Vulnerable Children on household spending.” Journal of

Development Effectiveness, 4(1), 9-37.

[40] Ward, P., Hurrell, A., Visram, A., Riemenschneider, N., Pellerano L., O’Brien C.,

MacAuslan, I., and Willis, J. (2010). “Cash Transfer Programme for Orphans and Vulner-

able Children (CT-OVC), Kenya: Operational and Impact Evaluation, 2007-2009: Final

report”. Oxford Policy Management.

[41] White, Halbert. (1984). Asymptotic Theory for Econometricians. San Diego: Academic

Press

[42] World Bank. “Cash transfers”. https://www.unicef.org/esaro/5483_cash_transfers.html

Appendix

40

Table A.1: (A)

41

Table A.1: (B)

42

Table A.1: (C)

43

Table A.2: Covariate Balance Between Households Randomly Assigned to CT-OVCTreatment Group vs. Control Group (Non-beneficiaries)

(1) (2) (3) (4) (5) (6)Control Transfer P-Value on Diff. Boot P-Value Standardized. Diff. N

Years of Edu. of HH Head 6.748 5.902 0.682 0.682 0.301 778

Sex of HH Head 0.411 0.352 0.432 0.448 0.122 1530

HH Receives Labor Wages 0.043 0.026 0.401 0.402 0.098 1530

HH Receives Outside Transfer 0.199 0.294 0.398 0.411 -0.222 1530

Chronically Ill Members of HH 0.130 0.102 . . 0.075 1530

Carer Age Index 0.354 0.490 . . -0.723 1530

Number of OVCs in HH 2.553 2.672 . . -0.076 1530

Poor Quality Walls 0.849 0.726 0.124 0.123 -0.304 1530

Poor Quality Floor 0.824 0.730 0.180 0.182 0.228 1530

HH Owns Livestock 0.797 0.799 0.763 0.757 -0.004 1530

Cattle Owned 1.606 1.289 0.122 0.174 0.133 1220

Poultry Owned 6.08 4.413 0.063∗ 0.054∗ 0.264 1220

Owns Telephone 0.164 0.105 0.577 0.565 0.173 1530

Owns Blanket 0.872 0.842 0.480 0.500 0.086 1529

Owns Mosquito Net 0.703 0.604 0.227 0.254 0.207 1529

Acres of Land Owned 2.324 1.599 0.088∗ 0.089∗ 0.152 1530

Household in Rural Location 0.799 0.832 0.954 0.955 -0.086 1530

HH Cons. 2007 1.640 1.603 0.594 0.590 0.038 1530

HH Food Cons. 2007 0.924 0.952 0.406 0.404 -0.048 1530

HH Non-food Cons. 2007 0.716 0.651 0.884 0.876 0.103 1530

Dietary Diversity Score 2007 5.514 5.114 0.141 0.150 0.271 1530

Size of the HH 5.703 5.444 0.724 0.736 0.102 1530

Age of HH Head 48.25 58.07 0.515 0.660 -0.668 1515

People Aged 0-5 in HH 1.667 1.649 0.666 0.683 0.018 664

People Aged 6-11 in HH 1.784 1.692 0.419 0.429 0.102 1115

People Aged 12-17 in HH 1.696 1.735 0.639 0.669 -0.044 1215

People Aged 18-45 in HH 1.927 1.912 0.954 0.957 0.013 989

People Aged 46-64 in HH 1.147 1.106 0.401 0.407 0.125 807

People Aged 65+ in HH 1.144 1.118 0.389 0.415 0.077 480

1 P-values on differences in column (3) are estimated using the typical clustered standard error for-mula, clustered at the sub-location level. * p < .01, ** p < .05, *** p < .01, † p < .001. P-values incolumn (4) are estimated using the wild cluster bootstrap.

2 All p-values are calculated using linear regressions that control for differences in Carer Age Index,Chronically Ill Members of HH, and Number of OVCs in HH. This is why the p-values are recordedas missing above.

3 Consumption is in terms of 1000 KSh per adult-equivalent.

44