Ministry of Community Development and Social Services. Impact of Social Cash Transfers on Welfare, Investment, and Education in Zambia

Transcription

1 DRAFT 2 Ministry of Community Development and Social Services Impact of Social Cash Transfers on Welfare, Investment, and Education in Zambia Gelson Tembo* Palm Associates Limited, Lusaka, Zambia tembogel@yahoo.com; tembogel@zamnet.zm & Nicholas Freeland Masdar International Consultants, London, UK nicholas.freeland@wanadoo.fr *Corresponding author Palm Associates Limited i MASDAR International Consultants

2 Table of Contents TABLE OF CONTENTS... I LIST OF TABLES... III LIST OF FIGURES... IV LIST OF ACRONYMS... V ACKNOWLEDGEMENTS... VII EXECUTIVE SUMMARY... VIII 1 INTRODUCTION PROBLEM AND MOTIVATION OBJECTIVES SOCIAL CASH TRANSFER SCHEMES IN ZAMBIA Targeting Size of the transfer Monitoring and Evaluation CONCEPTUAL FRAMEWORK METHODS AND PROCEDURES DATA AND DATA SOURCES Sampling and fieldwork DATA ANALYSIS Measuring impact Heterogeneous impact Balancing tests and common support RESULTS KALOMO SCT SCHEME Descriptive characteristics of the Kalomo sample Factors affecting participation in the Kalomo SCT program Propensity score estimation and balancing tests Impact of the Kalomo SCT program CHIPATA SCT SCHEME Descriptive characteristics of the Chipata sample Factors affecting participation in the Chipata SCT program Propensity score estimation and balancing tests Impact of the Chipata SCT program KAZUNGULA SCT SCHEME Descriptive characteristics of the Kazungula sample Factors affecting participation in the Kazungula SCT program Propensity score estimation and balancing tests Impact of the Kazungula SCT program CONCLUSIONS AND RECOMMENDATIONS REFERENCES ANNEX 1: TABLES AND FIGURES i

3 ANNEX 2: TECHNICAL NOTE ON IMPACT EVALUATION AND EFFICIENT PSM ii

4 List of Tables Table 1. Household demographic and human capital attributes, Kalomo, September Table 2. Household land- and asset-holding, agricultural production and welfare, Kalomo, September Table 3. Probit results for a household participation, Kalomo SCT program Table 4. PSM estimates of the impact of the Kalomo SCT programme on immediate welfare, investment and education outcomes, September Table 5. Household demographic and human capital attributes, Chipata, Match Table 6. Selected outcome variables, Chipata SCT programme, March Table 8. Odds ratio weighted PSM estimates of the impact of the Chipata SCT scheme on welfare, investment and education outcomes Table 9. Household demographic and human capital attributes, Kazungula, March Table 10. Selected outcome variables, Kazungula SCT programme, March Table 11. Probit results for the household participation model, Kazungula SCT programme Table 12. Odds ratio weighted PSM estimates of the impact of the Kazungula SCT scheme on welfare, investment and education outcomes Table A - 1. Results of the propensity schore balancing tests, Kalomo SCT programme Table A - 2. Results of the propensity schore balancing tests, Chipata SCT programme Table A - 3. Results of the propensity schore balancing tests, Kazungula SCT programme iii

5 List of Figures Figure 1. Age pyramids for beneficiary and control households, Kalomo, September Figure 2. Distributions of the body-mass index (BMI) for comparison and beneficiary households, Kalomo, September Figure A - 2. Propensity score kernel distributions for comparison and treatment groups, Chipata SCT programme Figure A - 3. Propensity score kernel distributions for comparison and treatment groups, Kazungula SCT programme iv

6 ACC AIDS ASP ATT BMI CDF CIA CSO CWAC DD DfID DSW DSWO DWAC ECF FAO GTZ HIV ILO IV MCDSS MDD MIC ML NFNC NGO OVC PAL PPA PPS PS PSM PSWO PWAS REBA List of Acronyms Area Coordinating Committee Acquired Immune-Deficiency Syndrome Agricultural Support Programme Average Treatment Effect on the Treated Body-mass index Cumulative Distribution Function Conditional Independence Assumption Central Statistical Office Community Welfare Assistance Committee Double Difference Department for International Development District Social Welfare District Social Welfare Officer District Welfare Assistance Committee East Cost Fever Food and Agriculture Organization German Technical Assistance Human Immunodeficiency Syndrome United Nations International Labour Organization Instrumental Variable Ministry of Community Development and Social Services Matched Double Difference Masdar International Consultants Maximum Likelihood National Food and Nutrition Commission Non-Governmental Organization Orphaned and Vulnerable Children Palm Associates Limited Programme Partnership Agreement Probability Proportional to Size Propensity Score Propensity Score Matching Provincial Social Welfare Officer Public Welfare Assistance Scheme Regional Evidence Building Agenda v

7 Acronyms Continued RHVP SADC SCT SIDA SSN TE UN US $ VAC WB WFP ZMK ZVAC Regional Hunger and Vulnerability Programme Southern African Development Community Social Cash Transfer Swedish International Development Agency Social Safety Net Maddala s Treatment Effects United Nations United States Dollar Vulnerability Assessment Committee World Bank World Food Programme Zambian Kwacha (Zambia s currency) Zambia Vulnerability Assessment Committee vi

8 Acknowledgements We would like to recognize the contributions of many individuals and organizations without which it would not have been possible to complete this study. Our thanks go to the Ministry of Community Development and Social Services (MCDSS), the German Technical Assistance (GTZ), the Department for International Development (DfID), the United Nations Children Fund (UNICEF), CARE International, and all other member organizations of the Technical Working Group on Social Protection (TWG-SP) for the opportunity to carry out this study, and for the financial and/or technical support that they rendered. Our special thanks go to Ms. Esther Schuring (GTZ), Mr. Rick Goodman (DfID), Ms. Kelley Toole (DfID), and Ms. Charlotte Harland (UNICEF) for their technical and logistical support during the design and field work of the Kalomo study. The value of the logistical support obtained from Ms. Sheila Nkunika (MCDSS, Lusaka) also cannot be over-emphasized. Ms. Jean Hamoonga (MCDSS in Kalomo) provided valuable logistical support during data collection in Kalomo, including provision of programme background information and access to the database, which was a key input into frame development and sampling. CARE International, through Messrs David Hunsberger and Robby Mwiinga, provided access to data for the Chipata and Kazungula studies. During the initial planning stages, the study benefited from the technical input of Dr. Melissa Gonzales-Brenes from the University of Massachusetts, and Drs. Sebastian Martinez and Mirey Ovadiya from the World Bank. Mr. Caglayan Isik of Frekans in Turkey also provided very valuable input during the design and initial data analysis for the Kalomo study. We acknowledge Mr. Stuart Kenward s input as an internal reviewer of the zero draft. We are also particularly indebted to the TWG-SP, the International Poverty Centre (IPC), Dr. Stephen Devereux, and Dr. Ricardo Sabates for their very constructive comments on the first draft of this report. Our acknowledgements would be incomplete without mentioning our team of very able research assistants. Specifically, we acknowledge the input of the team of enumerators and supervisors whose dedication during data collection ensured that the data collected were of high quality. The four field supervisors during the Kalomo survey (Ms. Doreen Goma, Mr. Colby Nyasulu, Ms. Alice Tembo and Mr. Joseph Chanda) all went an extra mile when it came to quality control and questionnaire editing. The highly competent team of data entry personnel, led by Messrs Langson Banda and Liseteli Ndiyoi, is also greatly acknowledged. Ms. Bernadette Chimai, Ms. Constance Ng wane, Ms. Abigail Nswima and Mr. Brian Mulenga were very instrumental in the validation and cleaning of all the data sets. The patience exercised by the households and community leaders and members during long hours of interviews are also greatly acknowledged. It is our hope that the insights from the information that they provided will translate into valuable interventions in their communities. Gelson Tembo Nicholas Freeland vii

9 Executive summary This document presents the results of retrospective impact studies for social cash transfer (SCT) pilot schemes in Chipata, Kalomo, and Kazungula districts of Zambia. The studies main objectives were to i) identify factors that explain household participation in the SCT programmes and, hence, determine the effectiveness of targeting; ii) determine the impact of the SCT interventions on welfare, investment behaviour and educational outcomes; and iii) determine the impact of asset wealth status on the effectiveness of the SCT programmes. Data and sampling The studies used household survey data of sizes 200, 886, and 200 households, respectively, selected through a two-stage cluster sampling scheme. The two strata were programme participants and non-participants, where the latter was carefully constructed to include households that were as similar to the treatment households as possible, differing mainly in the treatment status. In Kalomo, the household survey data were complemented with community survey data collected from knowledgeable community leaders and members in each Community Welfare Assistance Committee (CWAC) area. Methods and procedures Probit models of participation were used to determine targeting effectiveness as well as to estimate the conditional probability of participation, or propensity scores (PSs). In each district, the specification of the PS was subjected to and influenced by a series of balancing tests. In the end, and in every case, the final specification of the PS satisfied all the balancing tests as well as the common support requirement. Hirano, Imbens and Ridder s (2003) version of propensity score matching (PSM), through odds-ratio-weighted regression analysis, was used to estimate the P ˆ x / 1 Pˆ x for impact. That is, the weight was equal to 1 for treatment households and ( ) ( ) comparison households, where ( x ) = E( w = 1 x) viii ( ) P is the PS estimated from the probit participation model referred to above. For binary variables, the weighted regression model was specified as a probit. Unlike standard PSM, whose bootstrapped standard errors are at best consistent, the oddsratio-weighted regression approach leads to fully efficient estimates of impact (Hirano, Imbens and Ridder 2003). We also took advantage of its flexibility to estimate heterogeneous impact across asset wealth strata. A household was taken to be asset poor if it fell in the bottom two quintiles of the distribution an asset wealth index, where the index was constructed from data on asset ownership using principal components analysis (PCA). Sample characteristics Descriptive analysis of the data confirmed that the study population was highly vulnerable, characterized by elderly heads, high dependency ratios (averaging at least 309 percent), and high incidences of female- and widow-headed households. More than half the children 0-16 years old were orphaned. There were marked differences in the degree and specific nature of vulnerability between the three districts. Chipata district had the highest incidences of female- and widow-

10 headed households. The degree of orphanhood was also almost twice as much as the other two districts. Kazungula, on the other hand, had the oldest household heads (averaging 62 years) and highest dependency ratios (averaging 560 percent). Kazungula also had the lowest numbers of meals eaten by children (1.5 per day), compared to about 2.4 meals per day in the other two districts. Although the Chipata SCT scheme was in a peri-urban setting and the Kazungula scheme in a rural area, Kazungula households cultivated only 0.3 hectares of land, compared to Chipata s 0.8 hectares. Kalomo households had the largest cultivated land area of 1.5 hectares. Kazungula also had the least non-sct income levels, averaging US $93 per annum, compared US $101 and US $245 in Kalomo and Chipata, respectively. Determinants of participation The probit results of the participation models were largely consistent with the postulated targeting criteria. Widowhood was consistently significant in all the three districts. In Kalomo and Chipata, widow-headed households were significantly more likely to be in the programme if they were also female-headed. Being female-headed raised the widowed households probability of participation by 27.8 percent. In Kazungula, widowhood was statistically important regardless of the sex of the household head and whether or not the household hosted orphans, raising the probability of participation by 147 percent. However, the effect of widowhood was inversely related to the household s asset wealth. One would expect that a higher dependency ratio would enhance the chances of participation among widow-headed households. However, it had an opposite effect, reducing the probability of participation by 7.8 percentage points. The age of the household head was significant only in Kalomo and Chipata. An additional year to the age of the head would raise the probability of being in the Kalomo SCT programme by 1.3 percent. In Chipata, the effect of age was negative at the lower end of the age distribution and positive towards the upper end. This means that child-headed and elderly-headed households had the highest chances of being included in the scheme, which is perfectly consistent with the targeting criteria. The status of the household, as measured by the value of the main house, was important in explaining participation in the Kalomo SCT programme. Holding all other things equal, a one million Zambian Kwacha (ZMK1 million) increase in the value of the main house reduced the probability of participation by 40 percent. Household composition was a factor in Kazungula, where various sub-categories of active age groups were inversely and significantly related to the probability of participation. Impact estimates The impact of the SCTs on consumption expenditure was unambiguously positive and statistically significant in all the three districts and regardless of the asset wealth status. However, the consumption effect was greatest on non-food items. In Chipata, the effect of SCTs on non-food consumption expenditure was 9.6 percent greater in the asset poor group than it was in the asset non-poor category. ix

11 The effect of the SCT schemes on cultivated land area was either significant but with a wrong sign (Kalomo) or totally insignificant (Chipata and Kazungula). However, the impact on value of small livestock owned was positive and significant in the two rural schemes (Kalomo and Kazungula) whereas the impact of the SCTs on investment in micro-enterprises was positive and significant in Chipata. Furthermore, except for small livestock in Kalomo, where the impact was significantly greater among asset poor households, programme effects on all other investments was statistically not different between asset poor and asset non-poor households. The only other exception was in reference to asset-holding, where the impact was positive and significant only among non-poor households in all the three schemes. The impact of the SCTs on educational outcomes was mixed. Enrolment rates improved only among boys in Kalomo. In the other two districts, the impact was either of the wrong sign or insignificant, regardless of the sex of the children. Attendance rates improved for both male and female children in Chipata, regardless of asset wealth category, and for only those in asset-poor households in Kalomo. In Kazungula, attendance rates improved only among girls in non-poor households. Conclusions This study has shown that the survey populations comprised incapacitated and highly vulnerable households. However, the three districts were not uniform in the nature and degree of vulnerability and, from all indications, Kazungula had the most vulnerable households. Obviously, some of the observed differences in the results across the three studies were, at least in part, as a result of such underlying differences in the target populations. The results of the participation models suggest that targeting was effective to some degree in all the three districts. While widowhood was significant in all the districts, either by itself and/or in interaction with other factors, the exact combination of significant factors varied from district to district. This is not surprising as the community-based targeting system entailed a substantial amount of flexibility as to the exact combination of criteria used for targeting. It also seems to identify a possible need to harmonize the participant selecting process across communities and districts. It was clear from the results that the SCT schemes under study did indeed achieve their primary objective of improving immediate welfare levels, as indicated by their consistently positive impacts on consumption expenditure. Given the lack of evidence of impact on autonomous, non- SCT income generating abilities, the observed expenditure effects were directly supported by the SCTs. The greatest consumption effects were observed in Kazungula district, where the vulnerability levels were highest and consumption levels lowest. As consumption draws directly from the additional unearned cash from the scheme (with no need for any intermediate investment), it is not surprising that consumption effects were not influenced by asset wealth status. However, it appears that the increase in food consumption may have been adequate to lead to the observed comparable adult nutrition status between the treatment and comparison households in x

12 Kalomo (as measured by BMI Body Mass Index for adults). While current investments did not result in significant income effects to sustain the adopted livelihoods on a long term basis, the nutritional improvements are a positive indicator and implicitly offer an opportunity for moving out of the poverty trap Beneficiary households will invest in opportunities available to them as dictated by their circumstances and environment. Some investment opportunities are more demanding on the household s human and physical capital than others. The results confirm, as expected, that the easiest enterprises with minimal demands on capital were given preference. Small livestock, such as chickens, do not require much else beyond the money to purchase them. This explains why the impact of the SCTs on the value of livestock was consistently significant in the rural schemes of Kalomo and Kazungula. In the peri-urban scheme of Chipata, where livestock rearing are inherently less appropriate, beneficiary households invested more in micro-enterprises. Trading is usually more common in urban areas than livestock rearing. Investment in micro-enterprises is not as easy as purchasing and rearing small livestock. Although they both require some level of competency, micro-enterprises exert more demands on special business skills, which may not be readily available among incapacitated households. Thus, despite engaging in micro-enterprises significantly more than non-beneficiary households in Chipata, the beneficiary households non- SCT income levels were still unambiguously lower. Despite the unambiguously positive effects on educational expenditures in all the three districts, there were no significant effects on enrolment rates except among male children in Kalomo district. The effect of the scheme on education was more conspicuous with respect to attendance rates. The attendance effects were most consistent and largest in Chipata, where the SCT programme had explicit educational premiums. In Kalomo, only asset poor households exhibited significant attendance effects among both male and female children whereas in Kazungula, the effects were evident only in the asset non-poor stratum and only among male children. The results have exposed interesting evidence of threshold effects for certain expected outcomes and districts. The SCTs effects on asset accumulation exhibited the greatest threshold effects. Whereas beneficiary households had significantly higher asset levels than their non-programme counterparts in the asset non-poor group, the reverse was true in the asset poor category. If asset accumulation and/or prevention of asset depletion is of interest, then poor households may require higher transfer levels than their non-poor counterparts. These considerations may be especially important in Kazungula, where incomes are generally low and levels of vulnerability very high. In Kazungula, similar threshold effects were evident with respect to attendance rates in which they were significantly impacted only in the asset non-poor category. Key words: social cash transfers, impact, welfare, investment, propensity score weighting, Zambia xi

13 1 Introduction Poverty reduction is a major issue of concern in countries that have large proportions of their citizens living in poverty. Most strategies to reduce poverty aim to provide economic opportunities to the poor, which only households with the necessary human and other forms of capital are able to utilize. A portion of the poor fail to benefit because they are incapacitated. This sub-group of the poor has been the target of social protection interventions such as food aid, input support, asset distribution, and social cash transfers (SCTs). SCTs have become increasingly popular because of their inherent flexibility in helping with the target households varied needs. SCTs have also been known to lead to investment in productive enterprises that may lead to permanent welfare improvements (Gertler, Martinez and Rubio-Codina, 2006). 1 SCT programmes have been used and are quite well-established in many Latin American countries. The concept is relatively new to sub-saharan Africa and Africa as a whole. Several countries in the region are still piloting the idea. The specific design of the pilot schemes varies and is a function of parameters such as the target population of interest, the size of the transfer, and the existence and nature of conditionalities. In Zambia, unconditional SCTs have been in existence since With the test phase initiated and the first transfers paid out in November 2003, the first pilot SCT scheme was officially launched in May 2004 in Kalomo. The scheme s main objectives were to: (a) reduce extreme poverty amongst the poorest 10 percent in the district, mostly, but not exclusively, focusing on incapacitated, households, and (b) generate sufficient information on the feasibility, costs and benefits, and impact of SCTs as a viable social protection option in Zambia. Since then, four more pilot schemes have emerged in Chipata, Katete, Kazungula and Monze. 1.1 Problem and motivation Although a number of studies have been conducted on different aspects of the SCT schemes, none have been designed to measure their impact on the recipients welfare. Impact studies could be an invaluable source of lessons for subsequent project design and roll-out. Comparing beneficiary households before and one year into the Kalomo SCT scheme, Schubert (2005) found improvements in a number of indicators, including school enrolment rates (by 3 percent), nutrition (both quantity and quality of meals), asset endowments, ownership of small livestock, and selfperception of social status. Reliance on begging was also found to have reduced as did debt levels among the beneficiaries. However, the Schubert (2005) study did not have a comparison group. Unless there is reason to believe that there was no inherent growth in these variables without the 1 For a comprehensive characterization, advantages and challenges of social cash transfers (SCTs) in both developed and developing countries, the reader is referred to Tabor (2002). Critiques contend that SCTs may encourage dependency (World Food Programme [WFP], 2007), and that they may have inflationary effects on food prices, especially in remote areas where the cost of transporting food supplies from main markets to local areas are high (Regional Hunger and Vulnerability Programme [RHVP], 2007). 2 Social Cash Transfers have been implemented in Southern, Eastern, and Western Provinces by the Ministry of Community Development and Social Services (MCDSS) and with support from donors such as the German Technical Cooperation (GTZ) Social Safety Net (SSN) Project (Kalomo, Southern Province) and DfID through project agreements with CARE International (Chipata, Katete in Eastern Province; Kazungula in Southern Province) and Oxfam GB (Kaoma and Mongu in Western Province). 1

14 programme, the lack of a counterfactual makes it impossible to correctly attribute the observed changes to the programme. The need for rigorous impact evaluation was recognized by the government and donors, leading in 2007 to the sanctioning of a number of such studies by the Ministry of Community Development and Social Services (MCDSS), supported through joint funding from DFID, World Bank, UNICEF, CARE and GTZ. Because of the different stages that the pilots had reached, different approaches were needed to derive full value from the evaluation: a prospective impact study was possible in Monze, but only retrospective impact studies were possible in Chipata, Kalomo, and Kazungula. The studies were conceived just before the Monze scheme, which permitted the prospective impact study to be integrated into the programme design, allowing for longitudinal observation alongside a random phase-in of the Community Welfare Assistance Committees (CWACs). On the other hand, in Chipata, Kalomo, and Kazungula, whose SCT pilot schemes had been running for at least two years, and where no baseline data were available, the studies took a cross-sectional orientation. A fifth SCT pilot, Katete, was excluded because (a) it had already started (in July 2007) and it was therefore not possible (for both practical and ethical reasons) to influence the roll-out of the implementation to allow for a prospective study (as in Monze), and (b) it had not been in operation for long enough to expect significant differences between recipients and non-recipient households to have emerged (as with the retrospective studies in Chipata, Kalomo and Kazungula). This paper reports the results of the retrospective impact studies in Kalomo, Chipata and Kazungula districts. 3 To the best of our knowledge, the studies reported in this document represent the first comprehensive treatment of the impact of SCT programmes in the southern African setting, excluding a number of valuable impact studies undertaken by EPRI and others in a rather different context of South Africa. 1.2 Objectives The three retrospective studies aimed to achieve at least three objectives: i) Identify factors that explain household participation in the SCT programmes and, hence, determine the effectiveness of targeting; ii) Determine the impact of the three SCT programmes on welfare, investment behaviour and educational outcomes; and. iii) Determine the impact of asset wealth status on the effectiveness of the SCT programmes. This document is organized in five sections. The rest of this section presents an overview of the Zambia pilot SCT schemes (Section 1.3). In Section 2, we present the conceptual framework and hypotheses guiding the study, followed by methods and procedures in Section 3. The results are 3 The prospective baseline study for the Monze SCT scheme are presented in a separate report (Tembo and Freeland 2008). 2

15 presented in Section 4 and are organized by district Kalomo (Section 4.1), Chipata (Section 4.2), and Kazungula (Section 4.3). Conclusions and recommendations are presented in Section Social Cash Transfer Schemes in Zambia Zambia has a population of 11.5 million, of which 64 percent and 54 percent are classified as poor and extremely poor, respectively (CSO 2006). These are considered food poor because they consume less than the minimum energy requirement of 1,800 kcal per person (adult equivalent) per day. The high levels of poverty have been the concern of various stakeholders and have resulted in implementation of various poverty reduction strategies. Most of these interventions have been centred on providing economic opportunities to households with the capacity to exploit them. However, such opportunities leave out an estimated 200,000 critically poor and incapacitated households. The latter category of households is characterised by high dependency ratios and often are elderly-headed, and host Orphans and Vulnerable Children (OVCs). Schubert (2005) found that 60 percent of the household members in the incapacitated households were OVCs. SCTs were identified as one of the strategies that could be used to provide social assistance by the SSN Programme of the MCDSS. Initiated in 2003, the scheme is administered through the Public Welfare Assistance Scheme (PWAS), which is widely considered to be a viable channel for delivering the transfers to the targeted group (Schubert, 2003). By December 2004, the Kalomo SCT had covered the targeted 1,027 households. In Kalomo and Monze the SCTs are implemented by the MCDSS and were initially supported by GTZ. In Kazungula, Chipata and Katete, the pilot SCTs are implemented by the MCDSS and were supported by Care International under the Programme Partnership Agreement (PPA). However, at the time of this study, support was provided through DfID, the lead donor. The Chipata the SCT scheme, in existence since 2006, was the first urban-based scheme and included an explicit education allowance. In Kazungula the scheme tested the feasibility of implementing social cash transfers in a sparsely populated, hard-toreach, district. In Kalomo the capacity requirements of implementing a fully scaled up scheme at district level was being tested while soft conditionalities were attached to the Monze scheme. In Katete, the SCT scheme, initiated in July 2007, gave cash transfers to persons aged 60 years or older. The main objective of the pilot SCT in this district was to generate information on the cost effectiveness and acceptability of a social pension Targeting Except for Katete, the SCT pilot schemes used a community-based targeting system, facilitated by the Community Welfare Assistance Committees (CWACs). 5 CWAC members met to identify households in their community that were considered destitute and incapacitated as defined in their training process and in the operations manual (MCDSS/SSN 2007). The identified households were interviewed and the information recorded on official forms which were approved as correct by the village headman. At a second meeting, the households were ranked according to their level of destitution with the poorest household receiving a rank of 1. After ranking, a cut-off point was 4 The baseline survey of the prospective study in Monze is discussed in a separate report (Tembo and Freeland 2008). 5 The Katete SCT programme was the only one that used an age-based targeting system in which all community members 60 years or older were included in the programme. 3

16 established separating the bottom 10 percent of the households from the rest. All households within the bottom 10 percent bracket, together with destitute households just above the 10 percent cut off point, were presented to the rest of the community members at community meetings chaired by Area Coordinating Committee (ACC) officials. The ACC officials checked the forms for completeness, correctness and consistency before submitting them to the District Social Welfare (DSW) Office. In the presence of CWAC representatives, the District Social Welfare Officer (DSWO) and the District Welfare Assistance Committee (DWAC) representative scrutinized each application for eligibility and approved or rejected the applications and amounts of money to be received accordingly. Cases requiring further scrutiny and investigations were reviewed at a DWAC meeting. Once the decisions had been made, a list of the applicants was made with the approval status and reasons, in the cases of rejection, indicated. Applicants were informed of the final decision by the CWAC, who also provided counselling to rejected households. Information on the approved households was entered in a database for records and use in the development of summaries, payment orders and receipts. Approved households were informed of the payment procedures by the CWAC. The DSWO submitted the payment orders to the district Finance Bank branch while payments were made bi-monthly at designated pay points. Retargeting was done every three years to include new households and/or remove others from the list of beneficiaries based on eligibility criteria. Those who were removed after retargeting were, however, expected to benefit from other on-going assistance programs such as the Fertilizer Support Program and the Food Security Pack Size of the transfer The sizes of the monthly transfers that the beneficiary households received depended on household composition and on the scheme or district in question. In Kalomo, Kazungula and Monze, households without children received a transfer of ZMK40,000 (US $10) per month while households with children were given ZMK50,000 (US $12.5). 6 The transfers for Chipata were similar to Kalomo and Monze except for the school allowance given to households with schoolgoing children. Each child enrolled in a primary school received an allowance of ZMK10,000 (US $2.5) while a child in secondary school received ZMK20,000 up to a maximum of four children (CARE International, 2008). The transfers given to the households were not meant to get them out of poverty, but merely to get them out of extreme poverty by allowing them to get an extra meal each day. A transfer of ZMK40,000 (US $10) was equivalent in value to a 50 kg bag of maize. However, the households were not restricted on what to spend the money on, but were simply advised to use it wisely. 7 6 Assuming an exchange rate of US $1 = ZMK4, Although different amounts were disbursed in different districts per household, there has not been a deliberate effort to estimate the effects of these variations on the programme s ability to achieve its intended objectives; nor has the design been explicitly structured for this purpose. For empirical examples on how different amounts may impact on programme effectiveness see the evaluation by the Institute of Fiscal Studies on the Educational Maintenance Allowance Programme in the UK. 4

17 1.3.3 Monitoring and Evaluation The SCT pilot schemes were designed with an internal monitoring system by the CWACs, ACCs, DSWO, Provincial Social Welfare Officer (PSWO) and the Social Welfare Headquarters (SWHQ) to ensure that the operations of the scheme were in accordance with the set guidelines. In addition to the internal monitoring system, an external monitoring and evaluation system was put in place to monitor the operations and to help discern the impact and performance of the scheme. Prior research has shown that beneficiary households were relatively more vulnerable than the national average (see, for example, Tembo and Freeland 2008). 2 Conceptual framework In Zambia, SCTs are targeted at the poor and incapacitated households (Schubert 2006). These dire conditions, the resultant liquidity constraints, and the inherently imperfect credit markets that characterize most developing countries are expected to have kept the eligible households in perpetual poverty traps. The SCT programmes are seen as a means to relax these extreme liquidity constraints by providing access to unearned additional cash (Gertler, Martinez and Rubio-Codina 2006). Although the cash comes with flexibility as to how it can be spent, evidence has shown that most of the beneficiary households use it largely for purposes that improve their wellbeing. The most anticipated effect of the SCT programmes in Zambia is increased welfare levels in the immediate and short terms but with long term implications. The idea is that better nourishment could improve the general condition and productivity of the members of the incapacitated households and, hence, reactivate their dormant potential to further improve their welfare both in the immediate and long terms. For example, school-age children may be encouraged to enrol in school and, for those already enrolled, to improve their attendance rates. For most of the eligible households, their children s education is one of the surest ways to break out of the poverty cycle in the long term. By providing some of the daily requirements, the SCT will also free up time previously used to look for food and to survive, which could then be applied towards improved agricultural and non-agricultural production. These processes could be further enhanced if part of the SCT is actually directly invested in those areas. Although not a deliberate objective, the SCTs have been shown to have led to increased investment in agriculture (seasonal inputs, equipment, animals), human capital (health, education), and/or micro-enterprises. A recent qualitative study in Kalomo showed that the SCTs, though not large enough to discourage informal transfers from better-off members of the communities, made it possible for SCT recipients to pay for services (Wietler, 2007). The study also argued that the SCT beneficiaries had changed their livelihood strategies from unsustainable activities like piece work to hiring labour and investing in livestock. Schubert (2005) and RHVP (2007) also present evidence of programme-driven investment in (small) production animals, children s education, and assets. 8 The educational effects were expected to be particularly large for the Chipata scheme, where school allowances were more explicit. 8 These studies were not designed to elicit impact per se but nonetheless present the observed trends and differences in the outcome variables of interest. 5

18 Evidence from Latin America shows similar results. Gertler, Martinez and Rubio-Codina (2006) found that beneficiary households of Mexico s OPORTUNIDADES programme (formerly called PROGRESA) were able to increase their consumption levels by as much as 34 percent through investment in productive activities. 9 Angelucci and De Giorgi (forthcoming) provide a concise description of the design and structure of OPORTUNIDADES and demonstrate evidence of positive indirect effects on consumption levels of ineligible households, operating through informal insurance and credit markets. Sadoulet, de Janvry and Davis (2001) estimate an income multiplier effect of One of the over-arching objectives of SCT programmes is to achieve positive welfare effects for the poorest segments of the target communities. Possible threshold effects may prevent investment in certain areas by low-wealth beneficiary households. 10 Capturing such differential effects calls for estimation of heterogeneous impacts across different wealth groups. Asset endowment is one of the key wealth indicators often used in empirical studies (see, for example, Langyintuo et al. 2008). In summary, if we ignore any general equilibrium effects, the following hypotheses seem reasonable for the Zambia SCT schemes: i) Beneficiary households have increased welfare levels with respect to: a) household income through investment in income-generating activities b) consumption expenditure (food, non-food, school-related, etc) ii) Beneficiary households have enhanced investments in: a) micro-enterprises b) small production farm animals, such as goats, pigs, and, especially, poultry c) consumer durables such as television (TV) sets, radios, bicycles, etc d) their children s education, leading to higher enrolment and attendance rates iii) Due to threshold effects, the impact of the SCTs on investment is most visible in the asset nonpoor segment of the target population. Ultimately, one hopes that the SCTs have resulted in significant impacts on i) and ii), with the latter helping to ensure significant and sustained impacts on the former in the long term. If raising investment levels among the least wealthy is of importance, failure to reject iii) may imply the need to help the asset-poor cross the thresholds through greater transfers. This study tests these assertions for all the three pilot SCT schemes in Zambia Chipata, Kalomo and Kazungula. 9 In addition to immediate consumption and investment, evidence of increased savings has been found in some SCT programmes (Ravallion and Chen 2005). 10 Although the sample comprises the poorest and incapacitated households, they are not all expected to have uniform wealth levels. 6

19 3 Methods and Procedures 3.1 Data and data sources This study used three cross-sectional data sets from the three districts. It is important to emphasise that the Chipata and Kazungula fieldwork was undertaken quite independently of the retrospective survey in Kalomo that formed part of this contract between MCDSS on the one hand, and Masdar International Consultants (MIC) and Palm Associates Limited (PAL) on the other. Indeed the design of the CARE surveys in Chipata and Kazungula pre-dated discussions around the Kalomo (and Monze) studies, and the fieldwork for them was completed even before final contracts had been signed for the Kalomo retrospective (and Monze prospective) study. As a result, the sample sizes for Chipata and Kazungula were much smaller, the sampling itself less scientific, the questionnaire less sophisticated (excluding, for example, anthropometric data and value of assets), supervision of the fieldwork less comprehensive, and validation of the data less rigorous. MIC and PAL had no influence over, nor responsibility for, these processes; but they did subsequently invest significant time in checking and cleaning the Chipata and Kazungula datasets, and converting them to a format that made possible some degree of consistent analysis across the three datasets Sampling and fieldwork In Kalomo, an optimal target sample size of 1,200 was determined using power calculations (power = 0.80). Sampling was done through a two stage stratified cluster sampling scheme. The first stage involved the selection of eight Area Coordinating Committees (ACCs) using Probability Proportional to Size (PPS), which ensured that ACCs with larger populations had larger probabilities of being selected. 11 The second stage involved the selection of program and comparison households from each selected ACC using stratified random sampling with equal allocation. Comparison households were selected from among households that had qualified to be in the programme based on the selection criteria (see Section 1.2.1) but were left out only on account of the 10 percent cut-off point. In the final analysis, the sample was spread over 60 CWACs and 18 wards. At the household level, interviews were successfully conducted with 886 households, representing an overall response rate of 74 percent, and information received on 3,994 household members. The household data were complemented by community-level data collected using a community questionnaire administered to formal gatherings of community leaders in each CWAC. The community questionnaire collected information about community infrastructure, access to facilities, the efficacy of ACCs and CWACs, and participation in community meetings, among other things. 11 Different administrative and geographic units are used to divide a district for the purpose of efficient and effective targeting. The CWACs and ACCs are the grassroots structures of the Public Welfare Assistance Scheme (PWAS) which is a national social welfare scheme responsible for coordinating all social welfare interventions down to the grassroots levels. Both the CWACs and ACCs are committees elected by members of their respective communities. As geographic units, the CWACs are used to define specific areas with a certain number of villages, and in turn the CWACs define the ACCs, that is, ACCs are made up of a certain number of CWACs. 7

20 In Chipata and Kazungula, data were collected from 400 households (200 in each district). 12 The questionnaire was administered by CWAC members under CARE s supervision. 13 In each district, the sample households were selected randomly from the programme CWACs and equally allocated between the two strata (treatment and comparison groups). The control group was drawn from the list of households that had been interviewed by the CWACs but had not been selected into the programme only because of the 10 percent cut-off point. Unlike the Kalomo study, whose data were collected just after harvest (in September 2007), in Chipata and Kazungula data were collected over the period March through May 2007, which was inherently a lean season (just before the next harvest). From the aforesaid, it is clear that the three studies were not entirely uniform in design and implementation. However, they were similar in a few important respects, including (and most importantly) random selection of sample households through a two-stage stratified random sampling procedure. To minimize the effects of the timing differences between the Chipata/Kazungula survey and the Kalomo survey, and hence increase comparability across the three districts, we adopt an annual reference period in all computations. That is, computing estimates of incomes, consumption and other variables over a period of 12 months prior to the survey helped to even out the seasonal spikes and troughs that are common with most livelihood activities. 3.2 Data analysis For each of the SCT schemes studied, descriptive statistics on key household and, where possible, community attributes were presented first, comparing beneficiary and non-beneficiary strata. Mean differences between the two strata were measured and tested using unequal variance t tests. In all three surveys, data were collected on number of meals per day and the extent to which different categories of foods were eaten by adults and children. In the Kalomo survey, additional data were collected on height, weight and age of all household members. These allowed the computation of the body mass index (BMI) for adults above 19 years of age and a number of anthropometric measures for children 0-59 months old. For these, Z scores for weight-for-height (under-weight), weight-for-age (wasting), and height-for-age (stunting) were computed and compared with a standard population (WHO, 2000) using a user-written ado programme, Zanthro, in Stata. As per standard procedure, a child 0-59 months old was taken to be under-weight, wasted and/or stunted if the corresponding Z score fell below 2 standard deviations. 12 A ball-pack figure thought to be adequate but without any power calculations. However, power calculations done after collecting the data taking consumption expenditure as the main variable of interest indicate that the samples could support tests of power 0.94 and 0.62 in Chipata and Kazungula, respectively. This implies that the sample of 200 was adequate in Chipata for consumption expenditure but inadequate in Kazungula. Perhaps the urban setting in Chipata leads to more even consumption levels among the poor than rural areas. 13 A key rationale for using CWAC members in Chipata and Kazungula was to use the exercise to enhance the latter s capacities in monitoring and evaluation, which will be required as the SCT programmes evolve and are scaled up. However, using the CWAC members to administer the questionnaires among themselves could be prone to personal biases. 8

21 3.2.1 Measuring impact Identifying programme-induced changes is often confounded by the fact that communities and individual households are inherently heterogeneous and project direct beneficiaries are often not representative of the population of interest. This is further complicated by the fact that programmes use specific criteria to locate their interventions and to identify the beneficiary households. The resultant non-random selection of participants is expected to introduce what is commonly referred to, in econometric circles, as selection bias. Under these circumstances, the mean difference in the outcome variable between the treatment households and comparison households does not necessarily represent program impact. Ravallion (2001; 2003) characterizes the various methods used to estimate impact under these quasi-experimental conditions. Although the SCT schemes targeted households in the top 10 percent on an implicit index of vulnerability, there were no consistent household ranks available at the project offices. The schemes started ranking at the beginning but later abandoned the process when it became increasingly difficult to differentiate the levels of vulnerability among the eligible households. The inconsistent rank information made it impossible to account for the ranking through appropriate statistical frameworks such as the regression-discontinuity design (RDD). However, due to the fuzziness of the cutoff point (as the criteria were not completely discreet), it was possible to find matches between those in the programme and those just above the cutoff. The MCDSS offices had lists of beneficiaries and households that had qualified (based on targeting criteria) but were not included in the programme due to the 10 percent requirement. With this background, this study used propensity score matching (PSM) to identify and estimate programme impacts. PSM presents a unique set of techniques for reconstructing an experimental environment out of a non-random, quasi-experimental design. The treatment or participation relationship was modeled through a probit framework with the aim to identify the factors that explained participation as well as to estimate the conditional probabilities of participation (given the observed characteristics), also known as the propensity score (PS). The probit model was specified as: / (1) Prob ( = 1 x) = Φ( β + δ x + ε ) w, where w is a dichotomous variable equal to one if the household is an SCT beneficiary and zero otherwise; Φ is a normal cumulative distribution function (CDF), ε is the error term, and β and δ are parameter and vector of parameters to be estimated. x is a vector of household and community covariates or attributes/criteria used in the targeting process, Beneficiary households in the Kalomo, Chipata and Kazungula SCT programmes were identified and targeted using a community-based approach. Watkins lists the criteria used by the communities and the numbers of CWACs and communities that indicated having used each (Watkins 2008). 14 This list and knowledge gathered by talking to MCDSS staff were used to specify x. Equation (1) was estimated using maximum likelihood (ML) procedures in Stata. 14 Notice that there were differences in the combinations of variables used for targeting in the different districts. Watkins (2008) recognizes this and argues for greater harmonization in future. 9

22 The weighted regression approach of implementing the PSM that results in efficient estimates, credited to Hirano, Imbens and Ridder (2003), was used in all PSM models. Under this framework, impact is the estimated slope coefficient 1 ˆβ in the simple regression model (2) y β 0 + β w + e, = 1 ( ) but with the observations weighted by 1 and P( x) / 1 Pˆ ( x) households, respectively, where ˆ ( x ) = E( w = 1 x) ˆ for treatment and comparison P is the conditional probability of participation, or PS, estimated from Equation (1) (see Annex 2 for a technical note); β 0 and β1 are parameters to be estimated; and e is a zero mean error term. The outcome variable, y, was specified as the natural logarithm of per capita values for all continuous outcome variables (income, expenditure) and as levels for dichotomous outcomes (whether the household invested in and/or received income from micro-enterprises), education outcomes (enrolment and attendance), productive asset wealth index and consumer durable wealth index. For dichotomous outcomes, Equation (2) was estimated as a weighted probit Heterogeneous impact In addition to leading to fully efficient estimates of impact, estimating the PSM using the Hirano, Imbens and Ridder (2003) framework also has the advantage of permitting estimation of heterogeneous impact. In this study, we recognize the fact that the impact of the SCT programmes is likely to vary by the households existing wealth status. Wealthier households are expected to be more able to take advantage of the cash transfer and improve their immediate welfare, as well as to invest in future welfare. Disaggregating impact by asset poverty helped to test for the presence of threshold effects for some of the expected impacts (such as investment), which tends to be a common phenomenon in these communities. We estimated an asset wealth index using principal components analysis (PCA) as in Filmer and Pritchett (2001) and categorized all households falling in the bottom two quintiles of the distribution of this variable as asset poor and the rest as asset non-poor. 15 Thus, heterogeneous impact was estimated as (3) y = γ + γ w + w* D + u, 0 1 γ 2 where D is a dummy variable equal to 1 if the household was categorized as asset-poor and zero otherwise; γ 0, γ 1, and γ 2 are parameters to be estimated; and u is a zero-mean error term. Here, as in Equation (2), the weighting scheme based on the propensity score was applied. Based on Eq (3), the impact of the programme on an outcome variable is equal to ˆ γ 1 for asset non-poor households 15 In Chipata and Kazungula, the respondent was presented 26 durable items (11 productive assets and 15 consumer durables) and asked to state which they owned. No data were collected on the number or value of such assets. In the Kalomo study, however, such additional information was collected. We use the 0-1 asset dummy variables and the numbers of assets to compute the asset wealth indices by PCA in Chipata/Kazungula and Kalomo, respectively. We avoided using the value of assets in Kalomo to assure comparability with Chipata and Kazungula. 10

23 (D = 0) and ˆ γ (1) ˆ 1 + γ 2 (1) = ˆ γ ˆ 1 + γ 2 for asset poor households (D = 1). Thus ˆ γ 2 is the additional impact that an asset-poor household would experience relative to its asset non-poor counterparts Balancing tests and common support Propensity score matching (PSM) leads to unbiased estimates of the impact of a programme by matching the participants and non-participants on their observed characteristics. However, PSM is only as good as the quality of the matching and is valid only under certain identifying assumptions. The balancing effects of the PSM models were tested using a number of procedures, including t tests for the differences in covariate means between the two groups (participants and nonparticipants) before and after the matching (Rosenbaum and Rubin 1985), effectiveness in reducing standardized bias to within acceptable levels (no more than 5 percent), and ability to drive the overall probit relationship to insignificance as measured by a joint likelihood ratio (LR) test and pseudo R 2 (Caliendo and Kopeinig 2008). 16 The estimated PS was regarded as satisfactory and used in further analysis if and only if its specification was certified as having satisfied all the balancing tests. In all the three districts, balance was only achieved when a number of interaction terms were introduced. The estimated PS was also inspected for the common support requirement, which was found to be satisfied in every case. PSM is prone to what has come to be known as hidden bias if there are unobserved factors that influence both participation and the outcome variable. 17 Although long recognized as a source of concern in impact evaluation, it is only in recent years that researchers have increasingly paid attention to this problem (Rosenbaum and Rubin 1983; Caliendo and Kopeinig 2008; Jalan and Ravallion 2003). Hidden bias arises from two sources, omitted variables and/or unobserved and factors (Cameron and Trivedi 2005). The standard procedure of using interaction terms and higher order polynomials in the specification of the PS (as explained above) helps to deal with the former. The latter problem is non-testable and is in practice assumed away through the conditional independence assumption (CIA). That is, we assume that the unobserved factors are equally distributed between the two groups. 16 A well-balanced propensity score is necessary for artificially constructing an experimental environment from a quasi-experimental situation. In all the three districts/schemes, the propensity score estimated with a rich set of covariates that included interaction terms (see Tables A 1 through A 3 in Annex 1) satisfied the balancing property within an optimally determined number of blocks. In each of these blocks, the null hypothesis that the mean propensity score was not different between treated and control households could not be rejected. The idea is that there should be no association between treatment status and each covariate once the observations have been restricted to the region of common support. Estimation of the propensity score and generation of balancing tests were achieved through a combination of psmatch2, pscore and pstest procedures in Stata. 17 A key identifying assumption for the PSM is that there should be no unobserved factors (not explicitly accounted for in Equation 1) that influence both participation and the outcome variable. This is variantly called in the literature as the conditional independence assumption (CIA), matching on observables, unconfoundedness, etc. 11

24 4 Results This section presents and discusses the results of the three retrospective impact studies. The results are presented by district/scheme. In each district, following a brief overview of the district s socioeconomic characteristics, the household characteristics are summarized, comparing the beneficiary households with comparison households. The descriptive analysis covers household demographics and assets; nutrition and mortality; access to services; economic activities; and welfare status. The results of the probit estimation of the participant selection process and its effectiveness at balancing the covariates between the two strata are presented next, followed by empirical estimates and discussion of the impacts of participation on welfare, investment, and education outcomes. 4.1 Kalomo SCT scheme Kalomo district is located in the Southern Province of Zambia, 400 kilometres south of Lusaka and 120 kilometres north of Livingstone. It is serviced by both the national rail line and the Great North Road. Southern Province has a total population of 1.4 million of which 73 percent are categorized as poor. Kalomo district is home to 16 percent of the total provincial population of which 66 percent are categorized as poor. The district is ranked 44 (out of 73) on the district poverty rankings based on the incidence of poverty (CSO, 2006). The main economic activity in Kalomo is agriculture (both crop and livestock), constituting a major source of livelihood for percent of the inhabitants. Major crops grown include maize, groundnuts, beans, cotton, sorghum, millet, and cowpeas. According to the 1990/92 national census of agriculture, Kalomo had 14,209 agricultural holdings, making up 25 percent of all agriculture holdings in the province. The over-reliance on nature-based agriculture, with little irrigation, and inadequate early warning mechanisms or capacity to respond to disasters, makes these people vulnerable to weather changes and other natural disasters. In the last two decades, the district s agricultural and livestock production have suffered as a result of droughts and livestock disease outbreaks, such as East Coast Fever (ECF). The latter have wiped out large numbers of livestock for most households. Both the decimation of livestock, including draught animals, and the recurrent droughts have adversely affected crop production. Small-scale trading is also an important economic activity in the peri-urban areas of the district, including purchase and sale of vegetables, second hand clothes, maize grains, and groceries (VAC, 2003) Descriptive characteristics of the Kalomo sample Of the 886 interviewed households, 404 (46 percent) were SCT programme participants, and the rest were comparison households. Table 1 presents the characteristics of the sample households and also the results of unequal-variance t tests of mean differences between beneficiary households and comparison households. The results suggest high levels of vulnerability among the sample households. 12

25 Table 1. Household demographic and human capital attributes, Kalomo, September 2007 Household stratum Variable Variable label Full sample Comparison households Programme households hage Age of the household head (years) *** hsex Male headed households (%) 36.0% 40.0% 31.0% ** dwido Widow headed households (%) 63.0% 56.0% 71.0% *** hhsize Household size *** nact Number of effectively active members a *** c0to14 Number of children 14 years and below * m15to60 Number of male members years *** f15to60 Number of female members years *** m61plus Number of members 61 years and above *** nmofa Number of children with both parents alive *** nmoth Number of children with only the mother alive nfath Number of children with only the father alive ** nnone Number of double orphans *** dovc Households with orphans (%) 47.0% 42.0% 53.0% *** ddeath Households with a death (%) 5.0% 7.0% 3.0% *** avage Mean age of all household members (years) *** hedu Education level of hh head (years) *** medu Education of male members (years) fedu Education of female members (years) * maxedu Education of the most educated member (years) pschen School enrolment rate, 5 20 year olds (%) 78.0% 76.0% 81.0% ** pschat School children absent at least 1 day (%) b 14.0% 12.0% 16.0% kmroad Distance to nearest main road (km) kmprim Distance to nearest primary school (km) kmsec Distance to nearest secondary school (km) c ** kmhosp Distance to nearest health centre (km) kmu5 Distance to nearest under five post (km) * Treatment-control group mean difference tests based on unequal-variance t tests Significance: * = 10 percent, ** = 5 percent, *** = 1 percent. a nact = Number of members aged years that were fit to work (excluding ill members). b pschat = Proportion of school children that had reported having missed at least one day of the last school week prior to the survey. c kmsec: Most secondary schools are in or near district centres. Thus, the distance to the nearest secondary school could as well proxy for distance to the nearest town. Source: Authors calculations, Data from the Kalomo SCT impact survey, September, Most of the interviewed households had elderly heads (average of 59 years), and were femaleheaded (64 percent) and widow-headed (63 percent). Education levels were also generally low, averaging 2.7 years for household heads, 5.1 years among female members, and 5.5 years among male members. Even the most educated member of the household had on average of 6.3 years of education. The government s target minimum basic education is 9 years. Almost half (47 percent) 13

26 of the households hosted orphaned children. On average, of the 2.2 children below the age of 16 years in the household, 59 percent were orphaned. Eighty-seven (87) percent of the orphaned children had lost their fathers, compared to 65 percent who had lost mothers. Thirty-five (35) percent of the orphaned children had lost both parents. With a household size of 4.5 and an average of 1.1 effectively active members (i.e. those that were years old and fit to work), the dependency ratio averaged about 309 percent across the whole sample. Compared to comparison households, beneficiary households were 15 percent more likely to be female-headed, 27 percent more likely to be widow-headed, and 26 percent more likely to host orphans. Beneficiary households also had older heads, smaller household sizes, and lower education attainment by household heads and members. They were also significantly further from secondary schools but closer to under-five clinic posts than their comparison counterparts. Most of the differences between the two groups were statistically significant at 1 percent level (Table 1). Figure 1. Age pyramids for beneficiary and control households, Kalomo, September 2007 Source: Authors calculations, Data from the Kalomo SCT impact survey, September, Age pyramids for both strata (comparison and beneficiary households) were widest at the base and narrowest in the active age groups (20-60 years) as well as beyond 80 years (Figure 2). The pyramids also confirm the presence of a significant number of elderly members, especially those in the age group. These results further confirm the assertion that in most of the SCT target households it is the grandparents that are looking after the orphaned young children. Comparatively, however, the age pyramid for comparison households reflects relatively larger numbers of prime-age adults than does the one for beneficiary households. 14