Opening Up Pandora’s Box: The Effect of Gender Targeting and Conditionality on Household Spending...

This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/copyright

http://www.elsevier.com/copyright

Author's personal copy

Opening Up Pandora’s Box: The Effect of Gender Targeting

and Conditionality on Household Spending Behavior in Mexico’s

Progresa Program

SUDHANSHU HANDA, AMBER PETERMAN *

University of North Carolina at Chapel Hill, NC, USA

BENJAMIN DAVISUNICEF Eastern and Southern Africa Regional Office, Nairobi, Kenya

and

MARCO STAMPINI *

Research Department, African Development Bank, Belvedere, Tunisia

Summary. — This paper evaluates the behavioral impact of conditionality and gender targeting on spending behavior in the Progresaconditional cash transfer (CCT) program from rural Mexico. Results indicate that transfer income is not spent differently from generalincome, suggesting that transfers exert only an income effect. In addition, women who have increased control over their extra cash arenot significantly more likely to spend it in a ‘‘family-friendly” way than they do household earned income. Both features entail significantcosts to beneficiaries and program budgets; our results indicate that further evidence is needed to confidently advocate for their inclusionin cash transfer (CT) programs.� 2008 Elsevier Ltd. All rights reserved.

Key words — Latin America, Mexico, cash transfer, conditionality, gender targeting

1. INTRODUCTION

In late 1997, the Mexican government launched Progresa(now Oportunidades), a conditional cash transfer (CCT) pro-gram, designed to alleviate short-term poverty and to increaselong-term human capital among the extreme poor in the ruralareas. 1 Progresa differed from a typical poverty alleviationprogram in Latin America in at least four ways: (1) it providedbenefits in cash instead of in-kind as was typically done indeveloping countries at the time; (2) benefits were made condi-tional on a set of human capital enhancing conditions includ-ing attendance at monthly health seminars, school enrollmentfor children aged 8–16, and routine preventive health check-ups for all family members; (3) beneficiaries were selectedaccording to a rigorous and transparent targeting mechanism;and (4) the program was externally evaluated through a socialexperiment. 2 Each of these characteristics represented a majorchange in the design and approach to poverty alleviation inthe region, and the positive and credible results of the externalimpact evaluation have made Progresa a widely cited successstory among the international development agencies and argu-ably the most influential poverty alleviation program in thedeveloping world at this time. In the last decade, Progresa typeprograms have spread rapidly through the region and beyond(see Bouillion and Tejerina (2006) for comparative review ofcash transfer (CT) programs and evaluations). Similar pro-grams currently operate in Honduras, Nicaragua, Jamaica,Colombia, Argentina, Brazil, Ecuador, the Dominican Repub-lic, and Kenya, and discussions are underway in several othercountries including Ghana, Peru, Panama, and Tanzania. 3

Two design features of Progresa are presumed to be criticalto its success, and have formed the core of the programsadopted in the other countries. The first is the conditioningof transfers on beneficiary ‘‘co-responsibilities” such as men-tioned above. These responsibilities typically include schoolenrollment, health check-ups, or attendance at seminars thatteach about health, nutrition, and sanitation. The second de-sign feature is the delivery of the cash benefit directly to wo-men. The motivation for the first design feature is to ensurethat families invest in their children’s human capital develop-ment and have the money to support such investments; themotivation for the second design feature is the presumptionthat money in the hands of a responsible female householdmember will more likely be spent in a ‘‘family-friendly” wayand thus be consistent with the objectives of the program.Each of these design features makes assumptions about theeconomic behavior of households, and each has importantimplications for administrative and take-up costs of theprogram. For example, the notion that resources deliveredto women will lead to different spending outcomes contradictsthe standard neoclassical model of household behavior thatassumes income pooling. The conditioning of benefits onschooling and health assumes that households will otherwiseunder-invest, or demand less of these goods than is sociallydesirable. However, both conditionality and the strict require-ment that only one designated person in the household can re-ceive the cash payment adds significantly to the administrative

* Final revision accepted: October 27, 2008.

World Development Vol. 37, No. 6, pp. 1129–1142, 2009� 2008 Elsevier Ltd. All rights reserved

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevdoi:10.1016/j.worlddev.2008.10.005

1129


costs of the program, as well as the time cost of participationfor beneficiaries (Caldes, Coady, & Maluccio, 2006). Althoughthere may be political motivation for these design features,and there is clear evidence of increases in school enrollmentand health facility attendance among the participant house-holds, there is no hard evidence that these increases are dueto conditionality per se, versus the income effect of receivingthe cash grant. Indeed, in the context of developing countriesthat have weak institutional and monitoring structures, ques-tions have been raised about whether these requirements arereally the best method to increase human capital (see, e.g.,Das, Do, & Ozler, 2005; de Janvry & Sadoulet, 2005; Handa& Davis, 2006; Regalia, 2006; Samson, 2006; Szekely, 2006).Similarly, the targeting of transfers to mothers has raised ques-tions about whether this effectively makes them responsible forfulfilling the program conditionality requirements, furtherburdening women while perpetuating traditional notions ofgender roles within the family (Molyneux, 2006).

The purpose of this paper is to assess the behavioral impactof these two design features using the Progresa evaluationdata. The empirical strategy uses multivariate regression tocompare the marginal propensity to spend income from theprogram (transfer income) to the marginal propensity to spendhousehold earned income (general income). We expect that ifprogram conditions are binding, and transfer income is usedto support investment in human capital, then transfer incomewill be spent at a higher rate on goods such as education andfood relative to general income. We also test whether programparticipation affects women’s decision-making within thehousehold, and how differences in women’s decision-makingauthority affect the marginal propensity to spend transfer in-come. If giving Progresa to women enhances their bargainingpower, we expect their decision-making authority to increase.And if women indeed spend money in more ‘‘family-friendly”ways, then women who have greater decision-making author-ity will spend the transfer at a higher rate on ‘‘family-friendly”goods in comparison to general income.

This paper adds value to the existing research on the impactof CCTs in several ways. First, it is the first study to testwhether the marginal propensity to consume out of CCT in-come is different from general income, a key piece of informa-tion to understand whether these programs have both incomeand substitution effects in consumption. It is also the first totest whether the increased decision-making authority that gen-der targeting attempts to increase has an impact on spendingpatterns. 4 This is especially relevant in low income rural envi-ronments where women’s rate of paid employment is low andtheir main source of cash is from their husbands. In thesecircumstances, transfer income to women may crowd out in-tra-household transfers from husband to wife and essentiallycancel out the effect of the transfer, leaving no net effect onbargaining power and influence over spending.

In general, our results show that conditioning requirementsdo not necessarily lead to significantly higher propensities tospend transfer income on education, food, and clothing. Theschool subsidy in particular does not have a substitution effecton school spending, which is consistent with the notion thatthe payments are regarded as simply a replacement for incomeforegone from sending the child to school and not necessarilyto be spent to further support human capital development.Our results also establish that households where women haveincreased control over their extra cash are not significantlymore likely to spend transfers in a more ‘‘family-friendly”way than they do regular earned income. This may be becauseof income pooling or common preferences, or because Progre-sa transfers crowd out intra-household transfers from men to

women and leave no net impact. Overall, these results indicatethat the ‘‘black box” evaluations of Progresa and other CCTsdo not begin to tell the full story of the impact of conditional-ity and gender targeting on intra-household decision-making,women’s bargaining power, and household preferences amongrecipient households.

2. PROGRESA

Mexico’s Progresa was initiated in 1997 as a mechanism todevelop the human capital of poor, rural households byimproving education, health, and nutrition outcomes througha CT. The transfer comprised two parts, a flat food transferand a schooling transfer (beca escolar), which varies by gradeand sex of the child. 5 The food support is given in bi-monthlypayments. To avoid perverse fertility incentives, the schoolingtransfer was available only to children over the age of sevenand the entire transfer was capped at 695 pesos (Skoufias,2005). 6 As mentioned previously, with rare exception, trans-fers are provided directly to mothers under the assumptionthat they are more likely to use funds in a manner that willbe beneficial to the development of their children.

Because Progresa targets poor households, criteria weredeveloped for determining eligibility based on householdwell-being. The selection of eligible households was done inthree stages. First, potential recipient communities were iden-tified as poor based on a marginality score developed from thenational population census. The score used community vari-ables including share of illiterate adults, access to water anddrainage, availability of electricity and population workingin the primary sector, among others. Those localities classifiedas ‘‘marginal” were considered potential target locations andsubsequently evaluated based on the location and existenceof health and school facilities. Secondly, scores or ‘‘puntaje”were produced for each household using a baseline census ofeach community, and households below a poverty thresholdwere included as beneficiaries. The third step was to presenta list of potential beneficiary households to community assem-blies for review and discussion (see Skoufias, Davis, & de laVega (2001) for further discussion of targeting).

By the end of 1999, Progresa provided bi-monthly transfersto approximately 2.6 million households or about 40% of allrural families and 11% of all Mexican families (Skoufias,2005). The program operated in almost 50,000 communities,and had a budget of US$777 million or nearly 20% of theMexican government’s budget allocation for poverty allevia-tion (Skoufias & McClafferty, 2001). The Progresa transfer ac-counted for an average of 19.5% of household expenditureswithin non-beneficiary households in control localities(Skoufias, 2005). 7, 8 Evaluations of Progresa have shownsignificant positive impacts on a range of educational- andhealth-related outcomes (see, e.g., Dubois, de Janvry, &Sadoulet, 2002; Hoddinott & Skoufias, 2004; Shultz, 2004;Skoufias & McClafferty, 2001).

3. CONDITIONALITY IN CT PROGRAMS

In theory, conditions are set in an effort to induce a partic-ular response by recipient households. Even if it is assumedthat CCT programs alter recipient behavior in the desiredmanner, from a human rights perspective there is a questionas to whether conditions should be placed on households, par-ticularly when the purpose of the CT scheme is to reduce ormitigate the effects of extreme poverty. From an economic

1130 WORLD DEVELOPMENT

https://www.researchgate.net/publication/227651059_Mothers_at_the_Service_of_the_New_Poverty_Agenda_PROGRESAOportunidades_ProgressOpportunities_Mexico's_Conditional_Cash_Transfer_Programme?el=1_x_8&enrichId=rgreq-4cefa448-3a46-4296-bf90-d110827afb68&enrichSource=Y292ZXJQYWdlOzQ2NTA3MzIxO0FTOjEzMjc2MjMxMjkwODgwMEAxNDA4NjY0NDEwMDM2

https://www.researchgate.net/publication/5056963_PROGRESA_and_its_impacts_on_the_welfare_of_rural_households_in_Mexico?el=1_x_8&enrichId=rgreq-4cefa448-3a46-4296-bf90-d110827afb68&enrichSource=Y292ZXJQYWdlOzQ2NTA3MzIxO0FTOjEzMjc2MjMxMjkwODgwMEAxNDA4NjY0NDEwMDM2


perspective, neoclassical theory predicts that households arebetter off with an unconditional transfer to use as they seefit for discussion of economic theory and transfer payments).In addition, for normal goods demand will rise with increasesin income so providing CTs without conditions may be suffi-cient to induce increased consumption, particularly if thetransfers represent a substantial proportion of income. (Fordiscussion of how a range of modifications in conditionalitymay affect outcomes, see de Janvry & Sadoulet, 2005).

Conceptually, assessing the impact of program conditional-ity in a CT scheme such as Progresa entails estimating the rel-ative weight of the income and substitution effects in theresponse by a given household. 9, 10 In the case of school atten-dance, the full price of schooling is made cheaper to the house-hold by the amount of the transfer, inducing both asubstitution and income effect; in contrast, an unconditionaltransfer only exerts an income effect upon the household, sincethe household can allocate its collective resources such as timeor money to any activity. A substitution effect of zero impliesthat an unconditional scheme would have the same effect onhousehold behavior as the conditional scheme. This might oc-cur if the monetary value of the conditional transfer is lessthan the amount already being spent on the good. In suchcases, the program transfer is infra-marginal, it simply replacesthe previous expenditure on the item and so has the same effectas a general income increase, inducing only an income effecton household economic behavior. 11

Figure 1 illustrates the range of possible responses by thehousehold to a CT that is conditional on children’s schooling.The household incurs both direct monetary costs for children’sschooling as well as a loss in income (opportunity cost),which gives rise to the income-schooling trade-off depicted inFigure 1. The original (pre-program) budget line is AE; theCCT is represented by line segment CD, which is only accessi-ble if the child maintains an 85% attendance rate. The impactof this program is ambiguous and depends on household pref-erences (given by the shape of the indifference curve) and theinitial equilibrium position of the household. We can identifythree general types of responses by the household to theprogram offer: (1) households initially close to point A willnot respond to the program, since indifference curve U2 liesabove all points on line segment CD. These are householdswho currently demand very little schooling, and/or who havelow preference for schooling (flat indifference curves); (2)households initially near point B (e.g., U3) will be induced

to increase children’s schooling and move to a higher indiffer-ence curve such as U1

1 on segment CD. For these households,the program exerts both an income and substitution effect; and(3) households initially on BE are already ‘‘consuming” moreschooling than the program requirement, and so the programexerts only an income effect. Since schooling is a normal good,the quantity demanded for schooling increases, as does thequantity demanded for all other normal goods, and thesehouseholds move to a higher indifference curve on line seg-ment CD (and possibly to a corner solution at point D).

Clearly, the total impact of the program on the beneficiarypopulation depends on the initial distribution of householdsin terms of preferences and demand for schooling. Note alsothat two program parameters in particular are key determi-nants of the total household schooling response. First, the lar-ger the size of the transfer (given by the distance BC) the morelikely households in category one will be induced into the pro-gram, and therefore the net impact on schooling will be posi-tive. Second, the lower the minimum schooling required themore likely households in category one will be induced to par-ticipate, and the more likely households in category two willbe infra-marginal. The net effect on schooling in this case istheoretically ambiguous because although category one house-holds will demand more schooling, category two householdswill demand less schooling than in the case where the mini-mum school requirement is higher (since the program nowonly exerts an income effect on their behavior). The total effecton schooling depends on the distribution of household typesamong the target households.

Insofar as conditional schemes are complicated to executeand therefore administratively costly, the ‘‘value” of the con-dition is an important piece of information for the policy ma-ker. Both the public cost of overall implementing andmonitoring as well as the private costs that conditions imposeon beneficiaries are important considerations for program de-sign. In Progresa, for example, to monitor health-relatedrequirements, a system was set up with the community clinic,in which a doctor or a nurse submits certification of visits forhealth check-ups and for nutritional talks by each beneficiaryevery two months. The receipt of certification in turn triggersthe disbursement of the food support transfer (Skoufias,2005). 12 Schooling is monitored on a monthly basis, wherehouseholds in which a child misses more than 15% of schooldays in a month for unjustified reasons will not receivesupport for that child (Coady & Parker, 2004). In terms of

U11

U3

U1

C

D

B

IncomeA

E

85% 100%

U2

Schooling

Figure 1. Household responses to cash transfer conditional on schooling income.

OPENING UP PANDORA’S BOX 1131


overall program costs, monitoring of conditionality repre-sents about 20% of administrative costs of Progresa.Although these administrative costs are relatively small incomparison to the previous transfer programs, the estimateis still substantial at 8.9 pesos ‘‘absorbed” per 100 pesostransferred (Skoufias, 2005). Further, it is estimated that spe-cifically dropping the conditionality constraint of the pro-gram would reduce ‘‘absorbed” administrative costs from8.9 pesos to 6.6 pesos (Skoufias, 2005). In terms of privatecosts, Coady applies a social cost-benefit framework to Prog-resa, examines costs incurred by households in meeting con-ditionalities, and compares them to other transfer programs(Coady, 2000). One estimate of the transportation costs re-lated to compliance with the health and nutrition require-ments is approximately 6.38 pesos per 100 pesos receivedthrough the food transfer. 13 In addition, there are waitingand time costs as well as costs associated with sending chil-dren to schools, especially secondary schools, which may bea substantial distance outside the locality.

4. GENDER TARGETTING IN CT PROGRAMS

A complicating factor in evaluating household spendingbehavior is the somewhat unique program design of Progresa,which specifically targets mothers. This means that outcomesmay not be entirely due to program conditionality, but dueto the influence Progresa has on decision-making power ofthe recipient female. This design feature is motivated by theempirical literature, which relates household budget outlaysto the asset/income share or relative bargaining power of fe-males within a household. In general, evidence from low in-come settings shows that resources in the hands of mothers(relative to the father) tend to have a larger impact on humancapital investment such as schooling and nutrition. (For exam-ple, see Hoddinott & Haddad, 1995 in Cote d’ Ivore; Thomas(1997) in Brazil; Quisumbing & Maluccio, 2000 in Bangladesh,Indonesia, Ethiopia, and South Africa). These empirical re-sults are consistent with a collective (as opposed to unitary)model of household decision-making (Browning, Bourgui-gnon, & Lechene, 1994), where an individual member’s com-mand over total household resources depends on his or herindividual income or assets, or her potential income or assetsif she left the household.

Few studies directly test whether unearned income in theform of a transfer is spent differently by males and female; thismay be important if decisions are made collectively in thesense of Browning and colleagues (1994), or if allocations froman external source are viewed differently by the householdthan actual income earned by women. For example, menmay view Progresa money as belonging to the household, orthe household head and not the woman herself. There are afew studies which explore questions of intra-household alloca-tion and gender in the context of various transfer programs.Using South African data, Duflo shows that pensions receivedby women had a large impact on the anthropometric status ofgirls, specifically on granddaughters of recipients (Duflo,2003). In contrast, no effect on the nutrition status of childrenis found for pensions received by men. Conclusions about gen-der targeting of transfers are even more complicated in con-texts such as rural Mexico, where a woman’s paidemployment is very low so that her only source of cash is fromher spouse. Transfer income to women in this context maysimply crowd out intra-household transfers and leave themwith a net gain of zero in terms of cash under their control,thus neutralizing the objectives of the program.

Qualitative studies conducted in Progresa communities areconsistent with the hypothesis that women are more likely tospend income on health and education relative to men. Fieldworkers reported that almost 70% of all comments aboutmoney management specifically mention that men drinkalcohol with the money they receive, rather than giving itto the family (Adato, de la Briere, Mindek, & Quisumbing,2000):

‘‘. . .if they give it [transfer] to the man, he goes out and finds somefriends and they drink. He drinks a few glasses, he finishes themoney that he earns. He goes home and he screams that he wantsto eat but he doesn’t give the money to women to buy somethingto eat. (PG-18)” (promotora in Guerrero, Adato et al., 2000, p. 58)”

As well, field workers noted that women appear to be moreconscientious, when it comes to money and needs of the fam-ily:

‘‘Today I don’t have, what am I going to do for eating, for tortillas? Iwill always look for the manner how to do it. If I have to, I will ask toborrow the money. Nevertheless, the men do not take the risk to goand ask to borrow money. Men go around and say, I’m going to drinka soda, but the women don’t, because they think if I drink it, maybe Iwon’t have enough money for shoes, clothes. Women don’t eat (PG-18)”

In the case of Progresa, analysis of decision-making poweris a necessary component in explaining household spendingbehavior. There are four ways in which respondents suggestthat Progresa can increase women’s autonomy with respectto household expenditure decisions 14: (1) she does not needto ask her husband for money whenever she needs some-thing, she can purchase it independently with transfermoney; (2) because she receives additional income, she hasmore confidence in her ability to determine whether thereis enough money to buy things that she thinks they need;whereas when her husband receives money, he is in a betterposition to judge; (3) if Progresa increases the amount avail-able for food expenditure, which is a part of her domain,women can now make more decisions about such expendi-tures even if the types of decisions have not changed; and(4) transfers potentially increase the domain of women’sdecision-making; for example, if Progresa money is spenton clothing for herself or her children, she may make moredecisions on clothing without her husband’s involvementthan in the past. If any of these pathways translate into ac-tual increased expenditures for women and children, the un-ique program construction will be essential in re-creation ofsuccessful transfer programs such as Progresa. In addition,due to the additional benefits associated with Progresa, itwould be expected that changes in decision-making will bea lower bound on the program effects of targeting womenbecause, as noted above, many benefits are not expected tobe directly reflected in consumption outcomes. 15 Resultsspecifically examining the effect of Progresa on decision-making indicate that certain realms are influenced, althoughoverwhelming empirical evidence of direct effects of Progresaon women’s decision-making are not supported (Adatoet al., 2000). 16

5. DATA

The data we use were collected by the Government of Mex-ico as part of an evaluation of Progresa based on an experi-mental design, and included six survey waves and 24,077households. The 506 Progresa localities selected were ran-domly allocated into treatment and control groups with




https://www.researchgate.net/publication/268356969_'The_application_of_cost-benefit_analysis_to_the_evaluation_of_PROGRESA'?el=1_x_8&enrichId=rgreq-4cefa448-3a46-4296-bf90-d110827afb68&enrichSource=Y292ZXJQYWdlOzQ2NTA3MzIxO0FTOjEzMjc2MjMxMjkwODgwMEAxNDA4NjY0NDEwMDM2


probabilities of 60% and 40% (320 treatments and 186 controllocalities). The random assignment of localities, as well as aphased introduction of the program, allows for a more rigor-ous evaluation of Progresa (see Behrman and Todd (1999) fordiscussion of randomization process). The data are longitudi-nal, with surveys taking place approximately every six monthsfrom 1997 until the year 2000. At different times, question-naires included modules most relevant to program evaluation,including demographics, health, education, income-generatingactivities, and expenditure, as well as matching locality ques-tionnaires. We use data from wave one (ENCEL 98M), whichtook place in March 1998, before program initiation, andwave three (ENCEL 99M), which took place in May/June of1999, because in these rounds the modules on decision-makingand expenditure were included in the survey. 17

6. METHODOLOGY

(a) The effect of program conditionality

We employ an intuitively appealing approach to test thestrength of the program conditionality on household behavior.If conditionality is binding, program transfers will exert an in-come and substitution effect on household spending behavior,while general income only exerts an income effect on suchbehavior. If the substitution effect is big, the marginal propen-sity to consume (MPC) (for items such as schooling and food)out of transfer income will be larger than the MPC out of gen-eral income; if the substitution effect is small or zero, then theprogram only exerts an income effect and the two MPCs willbe statistically equal. This approach has been used extensivelyto test the impact of the US Food Stamp Program on pur-chases of food (Breunig & Dasgupta, 2005; Fraker, Martini,& Ohls, 1995).

We estimate an equation such as (1) which predicts total percapita spending as well as per capita spending on each of thefour other items (food, adult and children’s clothing, school-ing) 18

Cij ¼ b0 þ b1ðProgresaijÞ þ b2ðIncomeijÞ þ b3ðX iÞþ b4ðX jÞ þ uij; ð1Þ

where Cij represents the logarithm of monthly per capitaexpenditure of the ith household (either total expenditure orexpenditure on each of the four items listed above); Progresaij

and Incomeij are the logarithm of monthly per capita transferand earned income, Xi is a vector of socioeconomic and demo-graphic characteristics of household i, and Xj is a vector ofcharacteristics of locality j, and uij is the error term. To deter-mine if the impact of a Progresa peso is different from anearned income peso, we test the following null and alternativehypotheses

H0 : b1 ¼ b2; Ha : b1 – b2: ð1aÞWe expect the substitution effect of the transfer income to

be strongest for schooling and children’s clothing because thelargest component of the total transfer for most families isthe schooling subsidy, and this is directly conditional onschool attendance which has monetary costs in terms of fees,transportation, and uniforms. 19 A flat ‘‘nutrition” subsidy isgiven to each family independent of household size providedthat health check-up regimes are adhered to and the recipientattends a monthly talk on health, nutrition, and diet. Healthcare services are provided free of charge so no increase inhealth expenditures is necessarily expected. Similarly, the

food portion of the transfer is not explicitly conditional,but the health talks are designed to influence preferencesand change behavior towards purchase of more healthyfoods.

In Eqn. (1), regional differences are accounted for through avector of village level prices and state indicators. 20 Total con-sumption expenditure includes both actual expenditure andself-production of goods. 21 The income variable includes thefollowing components: wages and related benefits, agriculturalactivities, livestock, home production of food, non-agricul-tural business, public transfers (including pensions), interests,other forms of return to capital, and incoming transfers frommigration. School expenditures include tuition and other con-tributions, transportation, and utilities.

To minimize possible bias due to reporting error, selectiveadministrative failures, or the decision to not participate inthe program, we instrument the actual Progresa transferamount using the intent to treat (ITT). ITT is the theoreticalamount that should be paid to households that are entitledto participate in the program (details and motivation behinduse of this measure are available from the authors upon re-quest), as such it can be viewed as the program offer. Allprices, wages, transfers, and expenditures are deflated to1997 pesos and models use huber-white robust standarderrors. 22 Note also that expenditures, income, and transferamounts are logged to normalize values and account forskewed distributions. Hence, our equations are estimated indouble logarithmic form, and so our hypothesis test translatesinto a test of the equality of elasticities, or spending responses,of transfer and general income, though we will refer to thesecoefficients as MPCs for convenience. 23

(b) Decision-making and effects on spending behavior

To test the determinants of decision-making, a model simi-lar to (1) is run, which predicts an indicator of decision-mak-ing power

DM ij ¼ b0 þ b1ðProgresaijÞ þ b2ðIncomeijÞ þ b3ðX iÞþ b4ðX jÞ þ uij: ð2Þ

The decision-making index (DM) comprised five questionswhich were asked in both survey rounds capturing householdbehavior and the power of the female head to make decisionsregarding household spending, her children and her own in-come. The responses to these typically included either: (a)the husband, (b) she herself, or (c) both spouses would makethe decision together (see Table 3 for list of questions asked).Each of the three responses was assigned a value 0, 0.5, or 1,where a higher value indicates the woman has more bargainingpower. The questions were then summed to create a decision-making index. In addition to this composite index, we sepa-rately analyze the specific question on the degree to which awoman has control over how to spend her extra income sincethe program in fact specifically delivers extra income to thewoman householder.

Finally, to determine what impact decision-making powerhas on household spending of the Progresa transfer andhousehold earned income, the specification below is run usingthe same outcome expenditure categories as in (1)

C ij ¼ b0 þ b1ðProgresaijÞ þ b2ðIncomeijÞ þ b3ðDM ijÞþ b4ðProgresaji �DM ijÞ þ b5ðIncomeij �DMijÞþ b6ðX iÞ þ b7ðX jÞ þ uij; ð3Þ



where (Progresaij � DMij) and (Incomeij � DMij) are interac-tion terms allowing each type of income to vary by decision-making power of the female head of household. The followingnull and alternative hypothesis is tested:

H0 : b4 ¼ b5; Ha : b4–b5: ð3aÞ

If the marginal propensity to spend the Progresa transferinteracted with women’s decision-making is not significantlydifferent from that out of earned income, this indicates thatbargaining power matters equally for both income sources.However, if the difference is significant then women with high-er decision-making authority who have control over their ownincome spend it on different sources relative to householdearned income. Because decision-making is determined in partby other observables and unobservables which may also becorrelated with spending outcomes, the final specification isrun only using the fixed-effects models in order to controlfor factors like taste for gender equity or initial bargainingpower.

(c) Endogeneity of household income

Use of income presents a problem of endogeneity since in-come and consumption are jointly determined by the house-hold through the allocation of time between work andleisure. Unobserved ability or tastes may determine both in-come and the allocation of that income to consumption andschooling. To minimize the potential bias due to suchunobserved heterogeneity we estimate household fixed-effectsmodels which sweep away the fixed unobserved householdlevel error component. The underlying assumption is thatthe unobserved component affecting both earnings capacityand expenditure decisions is fixed over time, a plausi-ble assumption given the 1 year time frame used in ouranalysis.

7. RESULTS

(a) Summary statistics

Table 1 (top panel) summarizes the data on outcome vari-ables from ENCEL 99M, collected roughly 1 year after pro-gram initiation, for households with at least one programeligible child (less than 18 years of age). In the first four col-umns, data from the entire sample is presented, followed bytreated and control samples. Approximately, 60% of the en-tire sample is comprised of treatment households, while theremaining 40% are control households. On average, house-holds spend approximately 124 pesos per capita each month.Households that receive Progresa have a higher total house-hold spending of nearly 16 pesos than those who do not re-ceive Progresa. Food expenditures are the single highestexpenditure representing approximately 75% of total expen-ditures followed by child and adult clothing expendituresat 4% and 3% respectively, and then schooling expendituresaccounting for 2% share of expenditures. The averageschooling transfer among all sample households is 32 pesosper capita each month (53 for the treated sample). Ninety-nine percent of households are infra-marginal with respectto food expenditures, while 40% is infra-marginal with re-spect to schooling expenditures. In other words, all house-holds are already spending more than the transferearmarked for food, and just over a third of the householdsare already spending more than the transfer allotted forschooling in the baseline. For both food and school spend-ing, the ‘‘conditional” aspect of the subsidy is thus likelyto be quite weak.

Table 2 reports descriptive statistics of the control variablesfrom ENCEL 99M for the working sample of households withat least one program eligible child. We expect Progresa andnon-Progresa households to have similar characteristics, astreatment and control communities were chosen randomly.This is found to be a reasonable assumption, as all control

Table 1. Descriptive statistics of dependent variables using ENCEL 99M

Pooled sample Treated sample Control sample(N = 10,188) (N = 6,189) (N = 3,999)

Mean SD Min Max Mean SD Min Max Mean SD Min Max

Full eligible sample (with any children age 6 17 years)

Expenditure outcomesPer capita consumption 123.8 96.38 4.45 2599.85 135.11 96.49 4.41 1568.46 119.06 95.38 5.95 2599.85Per capita food expenditure 90.9 76.23 1.46 2597.5 95.2 75.45 1.45 1541.77 84.23 76.97 2.77 2597.5Per capita adult clothing expenditure 4.23 8.79 0 343.76 4.58 9.39 0 343.76 3.67 7.74 0 196.13Per capita child clothing expenditure 5.04 6.75 0 174.64 5.79 7.34 0 174.64 3.87 5.5 0 55.54Per capita schooling expenditure 2.59 9.34 0 395.88 2.77 10.21 0 395.88 2.31 7.8 0 135.59

Decision-making outcomesFull index 10.04 1.71 5 15 10.06 1.71 5 15 10.01 1.7 5 15Women’s spending indicator (yes = 1) 0.4 0.49 0 1 0.41 0.49 0 1 0.38 0.48 0 1Reduced index 7.66 1.61 4 12 7.66 1.61 4 12 7.66 1.61 4 12


Mean SD Min Max Mean SD Min Max Mean SD Min Max

Reduced eligible sample (with any children age 12–17)

Expenditure outcomesPer capita consumption 113.86 84.65 5.44 2599.85 120.17 79.75 5.44 1267.54 103.92 90.97 5.95 2599.85Per capita food expenditure 76.73 67.97 2.18 2597.5 83.91 60.73 2.18 1227.56 73.13 77.58 2.77 2597.5Per capita adult clothing expenditure 3.81 8.08 0 343.76 4.23 9.09 0 343.76 3.15 6.1 0 130.26Per capita child clothing expenditure 4.79 6.78 0 174.64 5.61 7.51 0 174.64 3,50 5.17 0 44.92Per capita schooling expenditure 3.16 9.88 0 193.32 3.46 10.68 0 193.32 2.69 8.44 0 135.59



variables are very similar in magnitude except possibly for theschooling of the head, where those in the treated group havean additional 0.2 years of schooling relative to heads in controlhouseholds. On average, households in the sample have justunder six members, of which there are comparable numbersof males and females. Average village population is 356 indi-viduals, of which approximately 57.5% are over 15 years ofage. The regional distribution of the sample includes all sevenintervention states, with the largest percentage living in VeraCruz (31%) followed by Hidalgo (17%), San Luis (14%),Michoacan (12%), and finally Guerrero and Queretaro bothaccounting for 4% of the sample.

Table 3 shows the percentage distribution responses to thedecision-making module in both points in time for our work-ing sample. 24 When these indicators are compiled in a score,the values range from 5 to 15; the average score of the indexis 9.4 in the baseline and 10.0 in the follow-up (not shown inTable 3). Consistent with the findings in prior reports, overallindicators did not exhibit large changes, though moving in apositive direction towards increased women’s decision-mak-ing. The exception to this is the question regarding expendi-ture of a woman’s extra income, which changes significantlybetween the 2 years. While initially only 5.4% of women hadexclusive control over their extra income, the percentage

Table 2. Descriptive statistics of control variables within full eligible sample with any children age 6 17) using ENCEL 99M


Mean SD Mean SD Mean SD

Progresa transfer

Per capita total transfer (log pesos) 42.31 44.09 69.66 35.99 0.00 0.00Per capita total schooling transfer (log pesos) 31.91 40.67 52.52 40.49 0.00 0.00

Household demographic characteristics

Household size (log) 1.85 0.37 1.84 0.37 1.85 0.38Number of children under 2 years 0.43 0.61 0.43 0.61 0.43 0.61Number of children 3–7 years 1.05 0.94 1.05 0.94 1.05 0.95Number of males 8–11 years 0.44 0.63 0.44 0.64 0.43 0.62Number of females 8–11 years 0.41 0.62 0.41 0.62 0.41 0.62Number of males 12–17 years 0.56 0.75 0.57 0,76 0.54 0.74Number of females 12–17 years 0.51 0.72 0.51 0.73 0.51 0.72Number of males 18–24 years 0.29 0.56 0.29 0.54 0.29 0.56Number of females 18–24 years 0.31 0.54 0.3 0.54 0.31 0.55Number of males 25–59 years 0.91 0.46 0.92 0.47 0.91 0.48Number of females 25–59 years 0.94 0.47 0.94 0.46 0.95 0.48Number of males 60 years and over 0.12 0.33 0.11 0.32 0.13 0.34Number of females 60 years and over 0.11 0.37 0.11 0.32 0.12 0.35

Household socioeconomic characteristics

Age of household head (years) 43.26 13.11 42.87 12.86 43.86 13.47Age of household head squared (years2) 2043.95 1264.48 2003.87 1228.74 2105.07 1315.69Missing age of household head (=1) 0 0.02 0 0.02 0 0.02Household head indigenous (=1) 0.4 0.49 0.41 0.49 0.4 0.49Education of household head (years) 2.97 2.71 3.03 2.76 2.87 2.62Household head literate (=1) 0.72 0.45 0.72 0.45 0.71 0.45Maximum female education (years)a

Household income (log pesos)Average adult female education (years) 0.98 1.23 0.98 1.23 0.97 1.23Female assets brought to marriage (log pesos)b

Land holdings (in hectares) 1.24 2.5 1.26 2.55 1.21 2.42

Locality variables

Average salary (pesos/day) 31.21 9.58 31.13 9.66 31.33 9.45Population (in 1,000’s) 0.36 0.25 0.36 0.24 0.36 0.25Population above 15 years (in 1,000’s) 0.21 0.14 0.2 0.13 0.21 0.15Distance to health center (log meters) 6.89 2.7 6.96 2.61 6.78 2.82Distance to state highway (log meters) 8.64 1.37 8.62 1.38 8.67 1.37Distance to secondary school (log meters) 5.61 3.47 5.7 3.43 5.48 3.53

State indicators (omitted Vera Cruz = l)Guerrero (=1) 0.07 0.26 0.09 0.29 0.05 0.21Michoacan(=l) 0.12 0.31 0.12 0.32 0.11 0.32Queretaro (=1) 0.04 0.2 0.04 0.19 0.05 0.21San Luis (=1) 0.14 0.34 0.13 0.34 0.14 0.34Hidalgo (=1) 0.17 0.38 0.2 0.4 0.13 0.33

a Only in the decision-making estimations.b In addition, all regressions contain the following locality prices: (1) tomatoe, (2) plantain, (3) banana, (4) masa, (5) corn tortillas, (6) chicken, (7) pork, (8)beef, (9) cornmeal, (10) beans, and (11) fresh cheese.



jumps to 39.7% in the subsequent year. In addition, the per-centage of men who had exclusive control over their wife’s in-come drops from 19.4% to 2.0%. From these results, it isapparent that while indicators are changing slightly over thepanel, any significant change in the overall index is being dri-ven by this specific question.

(b) Comparison of MPCs

Table 4 presents summary results of the spending responsesout of transfer and general income for each of the expenditureitems. The OLS results in the left panel, estimated over theENCEL 99M cross-section, indicate that the MPC out of gen-eral income tends to be significantly higher than that of trans-fer income for total, food, and adult clothing expenditures, butthere is no statistical difference in MPCs for children’s clothingand school spending. The fixed-effects results in the right-handcolumn provide even stronger evidence of the absence of asubstitution effect on transfer income; now the MPC forschool spending is significantly larger for general income rela-tive to transfer income. The school expenditure results suggestthat there is household level unobserved heterogeneity thatbiases the MPC of transfer income upwards, possibly due tostrong preferences for schooling.

We provide several extensions to assess the robustness ofour main results. First, we look for a substitution effect byfocusing only on the school subsidy part of transfer income,the component which should have the strongest effect onschool expenditure and children’s clothing. The OLS results

for school expenditure indicate a marginally (at 10% confi-dence) significant and larger MPC of transfer income relativeto general income, but this difference is actually reversed inthe fixed-effects estimates. These estimates also do not sup-port a strong impact of program conditionality on householdspending behavior (results are reported in Table A1 inAppendix).

Shultz, 2004 reports that the impact of Progresa on school-ing is greatest at older ages, when drop-out rates increase sig-nificantly, and when the school subsidy also increases toaccommodate the increased opportunity cost of schoolingfor these older children. Consequently, we might expect verylittle impact on school spending for households with childrenunder 11 (primary age), but higher impacts in householdswith older children (middle and secondary school age). Totest this proposition we repeat the estimation on householdswith at least one child between ages 12 and 17 (summarystatistics for this sample are shown in the bottom panel ofTable 1, and summary results are reported in Table A2 inAppendix). We focus on children’s clothing and schoolexpenditures, and show results for full transfer income (toppanel) and the school subsidy portion of transfer income(bottom panel). The OLS estimates (left panel) show no sig-nificant differences in MPCs for either total transfer incomeor the schooling portion of transfer income. The fixed-effectsestimates on the right-hand side are consistent with the re-sults from Tables 4 and A1; the MPC for schooling out ofgeneral income continues to be larger than that for transferincome.

Table 3. Patterns of decision-making within households (percent distribution by year in eligible sample)

Ql: Who decides totake children to thedoctor when sick?

Q2: Who tells thechildren to go toschool when theydo not want to?

Q3: Who decideshow to spend awomen’s extra

income?

Q4: Who decideson house repairs?

Q5: Who decideson children’sclothes/shoesexpenditures?

98M 99M 98M 99M 98M 99M 98M 99M 98M 99M

Husband only 10.42 9.83 11.05 10.75 19.35 1.95 31.13 30.91 25.10 20.62Wife only 11.97 11.31 11.59 11.06 5.39 39.66 3.71 5.24 7.16 6.77Both spouses 77.61 78.86 77.36 78.19 75.26 58.4 65.16 63.85 67.73 72.61

Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00

Table 4. Conditionality results for full transfer (among eligible households with any children age 6 17 years)

OLS regression Fixed-effects regression99M (N = 10,188) 98M–99M (N = 8,259)

Per capita transfer(log) (SE)

Per capita income(log) (SE)

P-value fordifference



P-valuefor difference

Total expenditure 0.034*** 0.084*** 0.000*** 0.026*** 0.049*** 0.158(0.005) (0.011) (0.010) (0.013)

Food expenditure 0.035*** 0.076*** 0.003*** 0.026*** 0.046*** 0.191(0.005) (0.012) (0.009) (0.013)

Adult clothing expenditure 0.031*** 0.133*** 0.000*** 0.022 0.091*** 0.028**

(0.009) (0.016) (0.019) (0.025)Child clothing expenditure 0.071*** 0.064*** 0.684 0.056*** 0.042* 0.655

(0.009) (0.015) (0.018) (0.025)Schooling expenditure 0.023*** 0.011 0.430 �0.003 0.044** 0.053*

(0.007) (0.014) (0.014) (0.021)

Reported SE is robust, clustered on the locality level.* It denotes 10% statistical significance.** It denotes 5% statistical significance.*** It denotes statistical significance at the 1% level.



As mentioned earlier, many treatment households are actu-ally infra-marginal in terms of the school subsidy portion ofthe transfer income that they are eligible to receive. For thesehouseholds, the program has only an income effect, and so wewould expect no difference in the MPC out of transfer andgeneral income. We should thus have a better chance of find-ing a difference in MPCs among those households who are notinfra-marginal, and so we repeat our previous estimation butnow we keep only the treated households that spend less onschooling (prior to the program) than they are eligible to re-ceive—for these households, the transfer income theoreticallyhas both an income and substitution effect. It is inappropriateto compare this select sample with all households in the con-trol group, so we select a matched sample of control house-holds using the puntaje or composite welfare index createdby Progresa to determine eligibility as our score. We applynearest neighbor matching based on this score, keep all con-trol households that are selected in the matching, and thenestimate our regression using this matched sample and thenon-infra-marginal sample of treated households. Results ofthis exercise are shown in Table A3 in Appendix for both def-initions of transfer income and the two outcomes of particularinterest. These results also do not show evidence of a signifi-cant substitution effect associated with transfer income. Infact, as we saw before, the fixed-effects estimates of the MPC

are actually larger out of general income relative to transfer in-come for schooling expenditures.

(c) Determinants of decision-making and effects on spendingbehavior

Table 5 presents the determinants of the decision-making in-dex using the wave three (ENCEL 99M) cross-section. Threedifferent variations on the decision-making index are pre-sented. The first is simply the sum of all questions as previ-ously discussed; the alternate two specifications break theindex up into the women’s income indicator (Table 3, Ques-tion 3) and the remaining ‘‘modified index”, which is the fullindex excluding the women’s income question. Table 6 usesthe household fixed-effects model to examine changes in deci-sion-making using the five expenditure outcome variablesincluding the interaction terms for both Progresa and house-hold earned income.

When examining a cross-section of the data and the fulldecision-making index, neither Progresa income nor regularearned income contributes to an increase in female decision-making (Table 5 left-hand panel). However, when the indexis broken into an indicator of women’s control overspending extra cash (Q3) and the remaining index (linetwo of Table 5), Progresa income is only significant in the

Table 5. Decision-making results for full transfer (among eligible households with any children age 6 17 years)








Full decision-making index 0.020 0.016 0.858 0074*** �0.018** 0.000***

(Q1–Q5) (0.016) (0.026) (0.18) (0.37)Reduced decision-making index 0.006 �0.007 0.628 0.006 �0.044 0.137(Ql, Q2, Q4, Q5) (0.013) (0.024) (0.017) (0.034)Women’s income indicator (yes = l) 0.012** 0.011** 0.709 0.047*** �0.021** 0.000***

(Q3) (0.006) (0.007) (0.006) (0.010)

Reported SE is robust, clustered on the locality level.** It denotes 5% statistical significance.*** It denotes statistical significance at the 1% level.

Table 6. Interactions for transfer and decision-making (among eligible households with any children age 6 17 years)

Fixed-effects regression98M–99M (N = 9,114)


DM � transfer(SE)


DM � income(SE)

P-value for difference

(1) (2) (3) (4) (2) = (4)

Total expenditure 0.019* 0.020* 0.047*** 0.009 0.216(0.011) (0.011) (0.014) (0.026)

Food expenditure 0.018* 0.017 0.046*** �0.011 0.304(0.011) (0.012) (0.014) (0.026)

Adult clothing expenditure 0.010 0.036 0.086*** 0.040 0.272(0.020) (0.024) (0.027) (0.049)

Child clothing expenditure 0.044** 0.044** 0.029 0.036 0.141(0.019) (0.023) (0.026) (0.047)

Schooling expenditure �0.009 0.027 0.025 0.043 0.264(0.014) (0.018) (0.021) (0.039)

Note: Decision-making is defined as women’s income spending indicator.* It denotes 10% statistical significance.** It denotes 5% statistical significance.*** It denotes statistical significance at the 1% level. Reported SE is robust, clustered on the locality level.



former and not in the latter, but none of the incomecoefficients is different from each other. Results are some-what different in the fixed-effects model presented in theright-hand panel of Table 5. After controlling for householdlevel heterogeneity, Progresa income significantly increasesthe overall decision-making power as well as decision-mak-ing over female income (Q3), while general income decreasesthis decision-making power. In addition, the effects are sta-tistically significantly different. Note that general incomeoverwhelmingly accrues to men as the labor force participa-tion rate for women is extremely low (less than 5%) in thispopulation.

Table 6 shows the fixed-effects results for the impact ofinteractions between decision-making and income sourceson the five expenditure outcomes analyzed previously. Wefind no statistically significant difference between the mar-ginal propensity to spend transfer pesos and the marginalpropensity to spend general income by variation in deci-sion-making, although in two cases (total and children’sclothing expenditure) the interaction is significant for transferincome. 25

8. DISCUSSION

(a) Conditionality and income and substitution effects

The impact evaluation of Progresa has demonstrated in-creases in the amount spent on the food and total caloricavailability among intervention households relative to thecontrol group (Skoufias, 2005). Our results indicate that thisincrease is due solely to an income effect rather than anysubstitution effect associated with program rules. This resultis plausible, since virtually all households are infra-marginalwith respect to food expenditures and are poor enough, sothat food is still a normal good for these households. Thisraises the question of whether unconditional CTs will raisehealth and nutrition of recipient households. Evaluationsof an unconditional CT program in Ecuador have foundthat children in the program showed improvements in therange of health outcomes including anthropometrics, hemo-globin levels, motor control, and cognitive outcomes (Leon& Younger, 2007; Paxton & Schady, 2007). In addition,Aguero and colleagues found that the Child Support Grant,a CT given without conditions in South Africa increasesheight-for-age of recipient children (Aguero, Carter, & Woo-lard, 2006). However, since Progresa is conditional onhealth-related behavior, it cannot be assumed that a parallelunconditional program would have resulted in significantlybetter health and nutritional outcomes beyond spendingon food.

The lack of a substitution effect for school expenditures,even among households that are not infra-marginal with re-spect to school expenditures, seems surprising since theschool subsidy portion of the transfer is explicitly linkedto the program condition of school enrollment andminimum attendance, and attendance can be costly in termsof transportation, lunch, uniforms and supplies. Further,evaluation results reported in Shultz (2004) show clear in-creases in aggregate school enrollment of children aged 8–16 among program households. Our results indicate thatthis additional schooling is not supported by greaterschool-related expenditures than is predicted by general in-creases in income. This is consistent with the idea that thelargest cost associated with schooling is foregone income

and households view the school subsidy as a replacementfor (general) income that is lost when children are sent toschool.

(b) Determinants of decision-making and effects on spendingbehavior

We found that Progresa significantly effects woman’s deci-sion-making in one realm, the ability to spend her own extracash, but not in other household decision-making categories.Moreover, increases in general income, which is essentially in-come under the control of men, tend to decrease women’sdecision-making authority within the household, and these re-sults are robust to household level heterogeneity (right-handpanel of Table 5). This result is consistent with intra-house-hold differences in bargaining power related to incomebrought to the household. However, we find differences inthe MPC out of transfer and general income in only two cases(schooling and adult clothing, Table 4). We also do not seeany significant change in spending behavior related to deci-sion-making authority (Table 6). These results are consistentwith two theories. The first is that husbands and wives havecommon preferences with respect to spending, so that whileeach one has authority over his/her income, there is no differ-ence in the pattern of expenditure. This is consistent with thefindings from Ecuador, which suggest that an unconditionaltransfer given to women has the same effect as regular house-hold income on child nutritional outcomes (Leon & Younger,2007). On the other hand, it could be that the woman simplysubstitutes transfer money for that money, which she previ-ously requested from her husband. If, for example, a woman’shusband would have provided her money for a given expendi-ture and now simply uses the money for some other purpose,the Progresa transfer may simply ‘‘crowd out” intra-house-hold transfers from men to women, and women and childrenmay not actually be made directly better off by the transfer.Alternatively, if the power relationship is such that the hus-band can simply take the money from his wife, then gendertargeting may be ineffective and lead to lack of significantlydifferent spending as we have found. 26 This result is consistentwith the hypotheses by Molyneux suggesting that althoughProgresa transfers improve overall household welfare, theydo little to truly empower women and that such programsshould seek to build skills which will provide longer term eco-nomic autonomy and security for women. (Molyneux, 2006).Our results and other forthcoming research on effects of Prog-resa on divorce rates and domestic violence suggest that thereis clearly a need for further gender specific research surround-ing effects of CT programs (Bobonis, 2007; Bobonis, Gon-zalez-Brenes, & Castro, 2006).

9. CONCLUSION

Our investigation of the behavioral impact of two key de-sign elements of the well-known Progresa CT program re-veals several key results. First, neither the total transfernor the school subsidy components of the transfer has asubstitution effect on spending. Hence, the transfer is treatedlike general income, a result which is particularly importantfor the school subsidy, as it indicates that the subsidy pri-marily functions as a replacement for the foregone incomefrom sending the child to school. It also indicates that trans-fer income is not used to support human capital investmentat a higher rate than from general (unconditional) income.Second, gender targeting of the transfer has not appreciably



increased the overall decision-making power of women,although women are reported to have more control overcash income as a result of the program. Third, this addi-tional control over cash income does not significantlychange spending behavior. This may be because husbandsand wives have common preferences or because transfer in-come crowds out intra-household transfers from husbandsto wives and so have no net impact on the composition

of spending. 27 Although we only examine one potential out-come, spending behavior, in one particular region, our re-sults indicate the need for more concrete evidence on thebenefit of these two design features as well as their behav-ioral consequences, before confidently recommending theirinclusion in CT programs where budgets are tight, capacityto enforce conditionalities is low and the time cost of com-pliance is significant.

NOTES

1. Progresa (Programa de Educacion, Salud y Alimentacion) was renamedOportunidades by the Fox administration after the change in governmentin 2000. As part of the updated program in 2000, Oportunidades madedesign improvements based on evaluations of Progresa (see Skoufias(2005) for details of program modifications).

2. The International Food Policy Research Institute (IFPRI) wascontracted to conduct the evaluation. The social experiment was madepossible through a randomly phased expansion; see <http://www.if-pri.org> for data access and documentation.

3. Cited programs include the Programa de Asignacion Familiar II

(PRAF II) in Honduras, Red de Proteccion Social (RPS) in Nicaragua,Program for Advancement through Health and Education (PATH) inJamaica, Familias en Accion in Colombia, Programa Familias in Argen-tina, Bolsa Escola in Brazil, Kenya OVC-CT Program in Kenya, Bono de

Desarrollo Humano in Ecuador and Solidaridad in the DominicanRepublic.

4. This hypothesis has been tested in an unconditional CT program inEcuador; findings suggest that income given to the mother has nodifferential effect on child nutrition as compared to household income(Leon & Younger, 2007).

5. Larger transfers were given to female children. In addition, an in-kindtransfer in the form of a nutritional supplement (‘‘papilla”) was given topregnant/lactating women and children between the ages of two monthsand four years of age. Support was also given to the local health clinicsand schools to improve services and facilities. For the purpose of thisanalysis, the in-kind and community-level support will be ignored andcontrolled for using demographic and community-level indicators inmultivariate analysis.

6. The cap increased as the program expanded. At the initiation fromJanuary–June of 1998 the cap was set at 585 pesos, by July–December of1999 the cap was approximately 750. This included 95–125 pesos for theflat food transfer. See Skoufias (2005), Table 1.1 for breakdown of pesotransfer by sex and grade of eligible children.

7. This calculation includes households which did not receive anytransfer because of non-adherence to program conditions, as well as delaysdue to verifications or delivery of transfers.

8. Progresa households were prohibited from receiving benefits fromsimilar welfare programs.

9. There is a large body of literature, which attempts to evaluate therelative substitution and income effects of a ‘‘conditional” in-kind transfer,the US Food Stamp Program (FSP). The strategy is to compare themarginal propensity to buy food out of food stamps and earned income(see, e.g., Devaney & Moffitt, 1991; Moffitt, 1989).

10. Note that the concepts of income and substitution effects are relevantfor small changes only, and not large discontinuous changes like schoolenrollment or attendance at health clinics. Behavioral aspects such asthese—while relevant and interesting—are not appropriate for this type ofanalysis.

11. A conditional transfer that is less than the amount already spent bythe household on the conditioned item is termed an infra-marginal transfer(see Senaur & Young, 1986).

12. In addition to the nutrition talks or ‘‘platicas” which must beattended by the mother, health check-ups are required of every householdmember. For example, adults require annual visits while children aged 5–16 require bi-annual visits. See Skoufias (2005), Table 1.3 for annualfrequency of visits for each age group.

13. This is based on an average of 25 trips per year, mostly by themothers. The average distance to the clinic is 3.98 km, and the averagecost of a trip is 3.95 pesos. The total annual travel cost on average perfamily is 95.7 pesos (see Coady (2000), Gertler (2000) for details.)

14. In addition to specific decision-making, Progresa aimed at influenc-ing overall woman’s status in five realms: (1) increasing resources in thehands of women; (2) helping women learn through health education; (3)creating a network of co-beneficiaries with whom women meet on aregular basis; (4) improving confidence and self-esteem; and (5) throughpromoting the education of girls, and thus improving the position offuture women (Adato et al., 2000).

15. Adato et al. (2000) present a comprehensive mixed methods report ofthe effect of Progresa on woman’s empowerment in both treatment andcontrol communities.

16. The authors use multi-nomial logits to predict each decision-making question as a function of being in a treatment community aswell as Progresa transfer. They find residing in a treatment communitydoes not significantly effect decision-making, although receiving Prog-

resa transfer does decrease incidence of husband’s sole decision-makingin five out of eight realms. In addition, spending of extra income isspecifically highlighted as a category in which Progresa transfer doeshave a significant effect. Simply residing in a locality with Progresa isnot shown to affect decision-making (see Adato et al. (2000) for furtherdiscussion).

17. In addition, wave one is supplemented by data from the baselineevaluation survey (Encuesta de Evaluation de los Hogares [ENCEL]),which took place in November of 1997.

18. Original specifications included both expenditure on alcohol/tobaccoand indicator of any spending on alcohol/tobacco, however, due to thelow percentage of households reporting expenditure (within the pooled



sample, 4% reported any spending and mean expenditure was approxi-mately 1.04 pesos) these outcomes were excluded.

19. As previously mentioned, concepts of income and substitution effectsare valid for small changes only, so we cannot relate these concepts todiscrete change such as school enrollment. School enrollment incurs directcosts which should show up in school expenditures, but households mayminimize these costs while still maintaining sufficient attendance to remaineligible for benefits.

20. Market prices per kilogram are representative of staple consumption(tomato, plantain, banana, masa, corn tortilla, chicken, pork, beef,cornmeal, beans, and fresh cheese).

21. To obtain per capita values, a per-adult equivalent is used. Thenumber of equivalent adults is calculated by assigning different weights tothe different members of the household, according to their age and gender.A minimum of 0.41 is given to children aged 0–4, and a maximum of 1.38to men aged 15–19.

22. In 1997, 1 peso was equivalent to approximately 0.11 US$.

23. The double-log functional form is the one that is most commonlyused in the literature to compare MPCs. The other approach is theWorking-Leser model which uses the share of total food or food groups

out of total spending as the dependent variable. Levadahl (1995) showsthat the double-log model imposes fewer restrictions on the relationshipbetween the MPCs out of different sources of income. Note that we can‘‘back-out” the actual MPCs from the estimated elasticities by multiplyingthem by the ratio of the sample mean of transfer or general income to thesample mean of the dependent variable.

24. Approximately 300 households were dropped from all analyzesconcerning the decision-making variables since they did not have asignificant female head over 18 years of age to which the module wasasked or the female head was not available for interviewing.

25. These results are robust to breaking the women’s spending of incomeinto three categories (wife only, husband only, both spouses).

26. Qualitative evidence suggests that this is happening; however, it isthought to be on a small scale (Adato et al., 2000).

27. Our results are not inconsistent with the evaluation results ofProgresa which show increases in caloric availability or total foodexpenditure. They simply indicate that these increases are due to anincome effect only.

REFERENCES

Adato, M., de la Briere, B., Mindek, D., & Quisumbing, A. (2000).Final Report: The impact of PROGRESA on women’s status andintra-household relations. Washington, DC: International FoodPolicy Research Institute, Food Consumption and Nutrition Divi-sion.

Aguero, J. M., Carter, M. R., & Woolard, I. (2006). The impact ofunconditional cash transfers on nutrition: The South Africa ChildSupport Grant. Working Paper. Riverside: University of California.

Behrman, J. R., & Todd, P. E. (1999). Randomness in the experimentalsamples of PROGRESA (Education, Health and Nutrition Program).Research Report. Washington, DC: IFPRI, Food Consumption andNutrition Division.

Bobonis, G. J. (2007). The impact of conditional cash transfers on marriageand divorce. Working Paper. University of Toronto.

Bobonis, G. J., Gonzalez-Brenes, M., & Castro, R. (2006). Public transfersand spousal violence. Working Paper. University of Toronto.

Bouillion, C. P., & Tejerina, L. (2006). Do we know what works? Asystematic review of impact evaluations of social programs in LatinAmerica and the Caribbean. Working Paper. Inter-American Develop-ment Bank, Poverty and Inequality Unit.

Breunig, R., & Dasgupta, I. (2005). Do intra-household effects generatethe food stamp cash-out puzzle?. American Journal of AgriculturalEconomics, 87(3), 552–568.

Browning, M., Bourguignon, F., & Lechene, V. (1994). Income andoutcomes: A structural model of intrahousehold allocation. Journal ofPolitical Economy, 102(6), 1067–1096.

Caldes, N., Coady, D., & Maluccio, J. A. (2006). The cost of povertyalleviation transfer programs: A comparative analysis of threeprograms in Latin America. World Development, 34(5), 818–837.

Coady, D. (2000). Final Report: The application of social cost-benefitanalysis to the evaluation of PROGRESA. Report submitted toPROGRESA. Washington, DC: International Food Policy ResearchInstitute.

Coady, D. P., & Parker, S. W. (2004). Supply-side education interventions:The case of PROGRESA in Mexico. Review of Development Econom-ics, 8(3), 440–451.

Das, J., Do, Q.-T., & Ozler, B. (2005). Reassessing conditional cashtransfer programs. World Bank Research Observer, 20(1), 57–80.

de Janvry, A., & Sadoulet, E. (2005). Conditional cash transfer programsfor child human capital development: Lessons derived from experience inMexico and Brazil. Working Paper. Department of Economics,University of California at Berkeley.

Devaney, B., & Moffitt, R. (1991). Dietary effects of the food stampprogram. American Journal of Agricultural Economics, 73, 202–211.

Dubois, P., de Janvry, A., & Sadoulet, E. (2002). Effects on schoolenrollment and performance of a conditional transfers program inMexico. Working Paper. University of Troulouse.

Duflo, E. (2003). Grandmothers and granddaughters: Old age pension andintra-household allocation in South Africa. World Bank EconomicReview, 17(1), 1–25.

Fraker, T., Martini, A., & Ohls, J. (1995). The effect of food stampcashout on food expenditures. Journal of Human Resources, 30,633–649.

Gertler, P. J. (2000). Final Report: The impact of PROGRESA on health.Report submitted to PROGRESA. Washington, DC: InternationalFood Policy Research Institute.

Handa, S., & Davis, B. (2006). The experience of conditional cashtransfers in Latin America and the Caribbean. Development PolicyReview, 24(5), 513–536.

Hoddinott, J., & Skoufias, E. (2004). The impact of PROGRESA on foodconsumption. Economic Development and Cultural Change, 53(1), 37–61.

Hoddinott, J., & Haddad, L. (1995). Does female income share influencehousehold expenditure? Evidence from Cote d’Ivoire. Oxford Bulletinof Economics and Statistics, 57(1), 77–96.

Levadahl, J. (1995). A theoretical and empirical evaluation of thefunctional forms used to estimate the food expenditure equation offood stamp recipients. American Journal of Agricultural Economics, 77,960–968.

Leon, M., & Younger, S. D. (2007). Transfer payments, mothers’ incomeand child health in Ecuador. Journal of Development Studies, 43(6),1126–1143.

Moffitt, R. (1989). Estimating the value of an in-kind transfer: The case offood stamps. Econometrica, 57(2), 385–409.

Molyneux, M. (2006). Mothers at the service of the new poverty Agenda:Progresa/oportunidades, Mexico’s conditional transfer programme.Social Policy & Administration, 40(4), 425–449.

Paxton, C., & Schady, N. (2007). Does money matter? The effects of cashtransfers on child health and development in rural Ecuador. World BankPolicy Research Working Paper 4226. Washington, DC.

Quisumbing, A., & Maluccio, J. (2000). Intrahousehold allocation andgender relations: New empirical evidence from four developing countries.FCND Discussion Paper No. 84. Washington, DC: InternationalFood Policy Research Institute.



Regalia, F. (2006). Some thoughts about ‘conditionalities’ in cash transferprograms: Lessons from Latin America and the Caribbean. InLearning workshop on orphans and vulnerable children (OVC) andconditional cash transfers (CCT), February. Nairobi, Kenya: Inter-American Development Bank.

Samson, M. (2006). Are conditionalities necessary for human develop-ment? In Presentation at the third international conference on condi-tional cash transfers, June 26–30. Istanbul, Turkey.

Senaur, B., & Young, N. (1986). The impact of food stamps on foodexpenditures: Rejection of the traditional model. American Journal ofAgricultural Economics, 68(1), 37–43.

Shultz, T. P. (2004). School subsidies for the poor: Evaluating the MexicanProgresa poverty program. Journal of Development Economics, 74(1),199–250 [Special Issue June 2004].

Skoufias, E. (2005). PROGRESA and its impact on rural households inMexico. Research Report 139. Washington, DC: International FoodPolicy Research Institute.

Skoufias, E., Davis, B., & de la Vega, S. (2001). Targeting the poor inMexico: An evaluation of the selection of households into PROGRE-SA. World Development, 29(10), 1769–1784.

Skoufias, E., & McClafferty, B. (2001). Is PROGRESA working? Summaryof the results of and evaluation by IFPRI. FCND Discussion Paper No.118. Washington, DC: International Food Policy Research Institute.

Szekely, M. (2006). To condition. . .or not to condition. In Presentation atthe third international conference on conditional cash transfers, June 26–30. Istanbul, Turkey.

Thomas, D. (1997). Incomes, expenditures and health outcomes: Evidenceon intrahousehold allocation. In L. Haddad, J. Hoddinott, & H.Alderman (Eds.), Intrahousehold resource allocation in developingcountries. Baltimore: Johns Hopkins University Press for IFPRI.

APPENDIX A. SUPPORTING TABLES

See Tables A1–A3.

Table A1. Results for schooling transfer (among eligible households with any children age 6 17 years)








Total expenditure 0.032*** 0.084*** 0.000*** 0.029*** 0.049*** 0.188(0.005) (0.011) (0.009) (0.013)

Food expenditure 0.033*** 0.076*** 0.002*** 0.028*** 0.046*** 0.239(0.005) (0.012) (0.009) (0.013)

Adult clothing expenditure 0.026*** 0.132*** 0.000*** 0.026 0.092*** 0.033**

(0.009) (0.016) (0.018) (0.025)Child clothing expenditure 0.070*** 0.063*** 0.683 0.049*** 0.040 0.776

(0.009) (0.015) (0.017) (0.025)Schooling expenditure 0.038*** 0.011 0.076* �0.014 0.043** 0.023**

(0.008) (0.013) (0.014) (0.021)

Reported SE is robust, clustered on the locality level.* It denotes 10% statistical significance.** It denotes 5% statistical significance.*** It denotes statistical significance at the 1% level.

Table A2. Results for full and schooling transfer (among eligible households with any children age 12–17 years)

OLS regressions Fixed-effects regression99M (N = 6,254) 98M–99M (N = 3,279)







Full transfer

Child clothing expenditure 0.076*** 0.067*** 0.678 0.056*** 0.085** 0.520(0.009) (0.019) (0.020) (0.038)

Schooling expenditure 0.027*** 0.010 0.391 �0.006 0.070** 0.039**

(0.008) (0.018) (0.018) (0.050)

Schooling transfer

Child clothing expenditure 0.080*** 0.073*** 0.739 0.060*** 0.085** 0.590(0.009) (0.019) (0’.020) (0.040)

Schooling expenditure 0.030*** 0.015 0.462 �0.008 0.070** 0.036**

(0.009) (0.017) (0.18) (0.032)




Available online at www.sciencedirect.com

Table A3. Results for non-infra-marginal sample (among eligible households with any children age 6 17 years)

OLS regressions Fixed-effects regression99M (N = 6,097) 98M–99M (N = 3,203)







Full transfer

Child clothing expenditure 0.071*** 0.064*** 0.690 0.000 0.011 0.803(0.009) (0.015) (0.030) (0.033)

Schooling expenditure 0.024*** 0.011 0.428 �0.057** 0.025 0.024**

(0.007) (0.014) (0.023) (0.029)

School transfer

Child clothing expenditure 0.070*** 0.063*** 0.690 0.002 0.011 0.826(0.010) (0.015) (0.031) (0.033)

Schooling expenditure 0.039*** 0.011 0.075 �0.059 0.025 0.021**

(0.008) (0.013) (0.024) (0.029)



Opening Up Pandora’s Box: The Effect of Gender Targeting and Conditionality on Household Spending...

Documents

Transcript of Opening Up Pandora’s Box: The Effect of Gender Targeting and Conditionality on Household Spending...