51
Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs Author(s): Niels-Hugo Blunch /Claus C. Pörtner Source: Economic Development and Cultural Change, Vol. 60, No. 1 (October 2011), pp. 17-66 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/10.1086/661219 . Accessed: 21/05/2014 14:32 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to Economic Development and Cultural Change. http://www.jstor.org This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PM All use subject to JSTOR Terms and Conditions

Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Embed Size (px)

Citation preview

Page 1: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy ProgramsAuthor(s): Niels-Hugo Blunch /Claus C. PörtnerSource: Economic Development and Cultural Change, Vol. 60, No. 1 (October 2011), pp. 17-66Published by: The University of Chicago PressStable URL: http://www.jstor.org/stable/10.1086/661219 .

Accessed: 21/05/2014 14:32

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access toEconomic Development and Cultural Change.

http://www.jstor.org

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 2: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

� 2011 by The University of Chicago. All rights reserved. 0013-0079/2011/6001-0002$10.00

Literacy, Skills, and Welfare: Effects of Participationin Adult Literacy Programs

niels-hugo blunchWashington and Lee University, and IZA

claus c. portnerSeattle University, and Center for Studies in Demography and Ecology,University of Washington

I. IntroductionIn a population where many adults cannot read or write, adult literacy programsmay be an effective way of improving people’s livelihoods because numerousstudies have found a positive impact on economic outcomes from literacy andnumeracy.1 Large-scale adult literacy programs have, however, generally notlived up to the expectation that they would be a cheap and effective way toeducate the adult population. Most programs have performed poorly in termsof enrollment, graduation rates, and the long-term literacy of participants.During the 1970s, for example, many programs had efficiency rates of about12.5%, with few graduates having stable literacy skills (Abadzi 2003, 2).These experiences led the World Bank to finance almost no adult literacyprograms during the 1980s and also led to questions in the developmentcommunity about whether literacy is really necessary for development. A case

We thank Aya Aoki, Paul Glewwe, Shelly Lundberg, David Ribar, John Strauss, Christopher Udry,participants at the Ghana at the Half Century Conference in Accra, and two anonymous refereesfor helpful comments and suggestions. All remaining errors and omissions are our own. This workwas financed in part by the World Bank’s Human Development Network Education Department.The views and findings expressed here are those of the authors and should not be attributed tothe World Bank or any of its member countries. Partial support for this work came from a EuniceKennedy Shriver National Institute of Child Health and Human Development research infrastruc-ture grant (5R24HD042828) to the Center for Studies in Demography and Ecology at the Universityof Washington.1 See, e.g., Glewwe (1999), Vijverberg (1999), and Jolliffe (1999) on Ghana; Boissiere, Knight,and Sabot (1985) on Kenya and Tanzania; and Moll (1998) on South Africa. For developed countries,examples are found in Murnane, Willett, and Levy (1995), Tyler, Murname, and Willett (2000),Ishikawa and Ryan (2002), Mitra (2002), Chiswick, Lee, and Miller (2003), Dustmann and Fabbri(2003), and Green and Riddell (2003).

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 3: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

18 economic development and cultural change

in point is Venezuela’s recent attempt to eliminate illiteracy through the“Mision Robinson” campaign; as shown by Ortega and Rodriguez (2008),there are at best small gains in literacy as a result of the program.

Surprisingly, there has been little rigorous analysis of the impact of adultliteracy programs on economic outcomes. Existing studies find a beneficialimpact for participants who became literate, but these studies suffer fromseveral methodological problems that cast doubts on the validity of the results.2

First, the samples are small.3 Second, typically only simple statistics such asthe fraction of completers who are able to read and/or do written calculationsare reported. Finally, and most importantly, they do not take account of possibleendogeneity in program placement and participation decisions.

This paper therefore examines the effect of participation in adult literacyprograms on the standard of living, measured by household expenditures peradult equivalent, using data from a large household survey from Ghana. Asdescribed below, the programs in Ghana are different from other literacyprograms in that they take much longer to complete and teach many othertopics in addition to literacy. After controlling for the possibilities of endog-enous program placement and self-selection into the program, we find a sub-stantial and statistically significant effect on consumption for households whereno adults have completed any formal education. For other households, theeffects are smaller in magnitude and not statistically significant. While thereare significant effects on literacy and numeracy from participation, these arecomparable to what have been found for other literacy programs and henceare likely too small to be the only explanation for the positive effect on welfare.Participation does, however, also lead to households being significantly morelikely to be engaged in market activities such as selling agricultural goods.

Adult literacy programs in Ghana have a long history—going back to theeighteenth century—and were mainly religiously based until after the SecondWorld War.4 The first national literacy program was introduced in 1948 butcollapsed in 1968 after the Nkrumah government’s fall in 1966. The programhad become closely politically associated with the Nkrumah movement, andthe skills taught were often not considered relevant by the participants. In1987, the Non-Formal Education Division was established by the Ministryof Education to organize and coordinate adult literacy programs and othernonformal education. Baseline studies were undertaken to ensure that what

2 See, e.g., Carron, Mwiria, and Righa (1989) and Carr-Hill et al. (1991).3 In Carron et al. (1989), the treatment and control groups are 291 and 66 individuals, respectively.The equivalent numbers from Carr-Hill et al. (1991) are 272 and 59.4 For a more complete description of the history of adult literacy programs and their currentfunction and curricula, see Blunch and Portner (2005).

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 4: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 19

was taught would be of immediate use and relevant to participants. Based onthe baseline studies, 28 main themes covering issues as varied as nutrition,immunization, family planning, and traditional and modern farming wereidentified as important issues and concerns of communities to be addressedin adult literacy programs. The themes can be divided into three broad areas:social and health issues, income generation/occupational skills, and civic aware-ness.5 One of the main features of the program is income generation, whichaims at equipping participants with occupational skills appropriate for thearea they live in. It takes about 21 months to complete the course, and theclasses meet two to three times a week for 2–3 hours each time. For comparison,the Venezulean “Mision Robinson” courses mentioned above last only 7 weeks.In most cases there are 20–30 participants to an instructor/facilitator.

II. Estimation StrategyThis section discusses the estimation strategy and the associated potentialeconometric issues. Denote the outcome of interest for household by , andi yi

let be the outcome with participation and the outcome without partic-y yi1 i0

ipation. Furthermore, let be a binary variable equal to one if householdp ii

participated in the program and zero if not. We can then describe the outcomeof interest by

y p ap � x b � u , (1)i i i i

where are other explanatory variables (including a constant) and is anx ui i

error term. The parameter of interest is , which reflects the effect of partic-a

ipation. The two main econometric issues are the possibility of endogenousprogram placement and self-selection into program participation. We discusseach in turn.

Endogenous program placement refers to the possibility that programs maynot be randomly distributed, spatially and/or temporally, throughout the coun-try.6 While this in itself is not a problem, it does lead to biased estimates ifany of the factors that determine program placement also influence the out-comes of interest and are unobservable to the researcher. Say that a local

5 For social and health issues, topics include family planning, teenage pregnancy, environmentalhygiene, immunization, AIDS, safe motherhood and child care, drug abuse, traditional medicine,safe drinking water. For income generation, topics include cocoa farming, maize cultivation, dryseason farming, basket weaving, animal husbandry, bee-keeping, oil palm cultivation, borrowingmoney for work, hygienic way of preserving and selling fish, farm extension services, pottery andsoap making. Finally, for civic awareness, topics include taxation, bush fires, interstate successionlaw, child labor, chieftaincy, community empowerment, and expensive funerals.6 See, e.g., discussions in Rosenzweig and Wolpin (1986), Pitt, Rosenzweig, and Gibbons (1995,114–49), and Angeles, Guilkey, and Mroz (1998).

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 5: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

20 economic development and cultural change

administrator is charged with placing literacy programs in his area and thathe places them according to his perception of how much “entrepreneurialspirit” there is in the different communities. Given that “entrepreneurial spirit”is also likely to be an unobservable determinant of the outcomes of interest,our estimates of the program’s effectiveness will be biased if we do not takeaccount of the endogenous program placement.7

Given the constraints that the available data impose, the best availableoption for dealing with endogenous program placement is to use communityfixed effects. This effectively removes all community characteristics that mightinfluence both placement and outcomes; if the communities are small enough,there will be no bias from endogenous program placement. Using communityfixed effects means, however, that the effect of the program is identified purelyfrom the differences between individuals and households within the com-munity. This is also the main downside of using community fixed effects: itbecomes harder to address the issue of self-selection into the programs becausewe cannot use variation between communities.

The two main methods for dealing with possible self-selection into partic-ipation are the “ignorability of treatment” approach described by Rosenbaumand Rubin (1983) instrumental variables (Wooldridge 2002, 603).8 The “ig-norability of treatment” method assumes that, conditional on a set of explan-atory variables , the participation and the outcomes are independent.x p (y , y )0 1

This is unlikely to hold here, since factors such as the unobservable ability ofindividuals potentially affect both participation and the outcomes of interest.9

Hence, we use instrumental variables to deal with self-selection into partic-ipation. What is needed are instruments that affect the decision to participatenet of the other explanatory variables but are not correlated with the unob-servable heterogeneity. The challenge is to find a set of instruments that varywithin communities and that do not directly affect the outcome(s) of interest.In the absence of community fixed effects, an obvious candidate would be theexposure one has had to the literacy program, measured, for example, by howlong the program has been available, but this clearly does not vary withinthe community. It is, however, likely that different types of individuals and

7 The direction of the bias depends on how the omitted variable(s) affect the placement decisionand the outcome(s) of interest. If the most “entrepreneurial” communities get the programs firstand also do better in terms of the outcome variables, then the estimates of the program’s effectivenesswill be biased upward.8 This is obviously a simplification, as can be seen from the review of methods in Ribar (2003).9 Imagine that someone with higher ability faces a higher return to participation than somebodywith lower ability and that higher ability people also do better in terms of the outcome(s) ofinterest. In this case, the predicted effect of participation will be biased upward if we do not takeaccount of self-selection into the program.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 6: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 21

households will respond differently to the introduction of a program. In thatcase a valid set of instruments is program exposure interacted with exogenousindividual or household characteristics. These interactions should have nodirect effect on the outcome variables and will not be correlated with unob-servables like innate ability. One drawback of the instrumental variables ap-proach is, however, that what we identify is the effect of participation amongthose for whom the instruments affect the decision to participate. We discussour choice of instruments in more detail below.

As discussed by Wooldridge (2002, chap. 18, 621–25), the effect of par-ticipation can be estimated under relatively relaxed conditions using a modifiedtwo-stage method. The first stage estimates the determinants of the partici-pation decision. In the second stage the consumption equation is estimatedby instrumental variables (IV) using the predicted probabilities from the firststage and , which includes all other explanatory variables and an intercept,xas instruments. An attractive feature of this approach is that the results arerobust even if the participation equation is not correctly specified.

III. DataThe Ghana Living Standards Survey (GLSS) is a nationally representative strat-ified multipurpose household survey collected in 1987, 1988, 1991, and 1998,consisting of a household survey, a community survey covering the rural areas,and a price survey. We use the most recent of these surveys (GLSS 4) here.

The household part of the GLSS contains modules ranging from educationand health to migration, credit, and assets. For our purposes, the most im-portant modules are the education and household expenditure modules. Theeducation module includes information on educational attainment, partici-pation in adult literacy courses, and literacy and numeracy proficiency (readingand writing in a Ghanaian language, English reading and writing, and abilityto do written calculations). The module on household expenditures includesinformation on food and nonfood expenditures, where the former is additionallybroken down into frequently and less frequently purchased items. From thecommunity questionnaire we use information whether there is or has been anadult literacy program in the community and, if so, for how long.

GLSS 4 employed a two-stage sampling design. Of the available enumerationareas, 300 were chosen in the first stage with a probability based on the sizeof the enumeration area in the 1984 census. In the second stage, 20 householdswere selected from each of the chosen enumeration areas, yielding a total of5,998 households and 24,611 individuals. We dropped 13,145 individualswho are less than 18 years of age, because our focus is on adult literacy andparticipation in adult literacy programs. Furthermore, we dropped 529 adults

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 7: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

22 economic development and cultural change

who are either not family members (hired help or tenants) or are householdmembers who do not fulfill the GLSS’s definition of being present and thereforehad no information collected about them, leaving 12,737 adults. Of these,253 have missing information about their schooling and/or their participationin adult literacy programs. Since there is only information on adult literacyprograms in the community survey and the community survey was only donein rural areas, we drop all 4,613 individuals who live in urban areas. Anadditional 38 individuals are dropped because there was no information aboutthe household head in the survey, and 124 individuals are dropped because ofinconsistent or unusable age and/or education data.

A number of enumeration areas had multiple communities within them,and the enumerators were supposed to note which households belonged towhich communities. Unfortunately, this did not happen for the vast majorityof the areas.10 As mentioned above, we rely on community fixed effects tocounter any potential bias from endogenous program placement; to avoidproblems with this approach we drop all enumeration areas that covered morethan one community, which leads to 997 individuals being dropped.11 Thisensures that enumeration area fixed effects are exactly the same as communityfixed effects. On average an enumeration area covers less than 18 square ki-lometers (or just less than 7 square miles), which means that it is, in principle,possible to walk from one end to the other in a little over one hour withouttoo much effort. The 1984 population census had a total of 13,012 enumerationareas, and even though enumeration areas were likely bigger in rural areasthan in urban areas, most houses in rural areas tend to be clustered in com-munities close to roads, paths, or water sources.12 Hence, we are confident thatany bias from possible endogenous program placement within an enumerationarea will be negligible. After dropping these observations, our base sampleconsists of 6,713 adults from 3,219 households living in 166 communities.

10 While it is, in principle, possible to deduce where the individual households reside from theraw data, Ghana Statistical Service has not done this and will not allow nonemployees access tothe relevant raw data.11 Of the 997 individuals dropped because of this, 233 live in enumeration areas where somecommunities have a literacy program and some do not, 360 live in enumeration areas where theprogram has not been available the same number of years in different communities, and theremainder are dropped simply because the community survey states that there are multiple com-munities within the enumeration area.12 The enumeration areas are described in more detail in Central Bureau of Statistics, Ghana (1984)and had to be small enough to be handled by an enumerator covering all households in one day.In 1984, the average population of an enumeration area was 1,000 in urban areas and 700 in ruralareas (Central Bureau of Statistics [Ghana] 1984, 7).

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 8: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 23

TABLE 1DESCRIPTIVE STATISTICS: INDIVIDUALS

Variable Mean SD Min Max

Dependent:Attended literacy program .11 .32 .00 1.00Log of per adult equivalent expenditure 13.76 .68 11.68 17.09Literate .41 .49 .00 1.00Numerate .45 .50 .00 1.00

Individual:Female .55 .50 .00 1.00Age

30–44 .31 .46 .00 1.0045–59 .20 .40 .00 1.0060–99 .16 .37 .00 1.00

Primary education .12 .33 .00 1.00Middle education .29 .45 .00 1.00Post-middle education .06 .23 .00 1.00

Household:Owns land .31 .46 .00 1.00Nonfarming household .10 .30 .00 1.00Percent with primary education .12 .23 .00 1.00Percent with middle education .29 .33 .00 1.00Percent with post-middle education .06 .17 .00 1.00Household size 5.29 2.77 1.00 21.00Number of adults in household 2.64 1.30 1.00 8.00

Instruments:Female exposure# 2.94 4.07 .00 15.00Age # exposure

30–44 1.63 3.36 .00 15.0045–59 1.11 2.90 .00 15.0060–99 .88 2.58 .00 15.00

Owns land exposure# 1.54 3.18 .00 15.00Nonfarming exposure# .44 1.89 .00 15.00Primary education (%) exposure# .69 1.70 .00 15.00Middle education (%) exposure# 1.63 2.64 .00 15.00Post-middle education (%) exposure# .30 1.09 .00 15.00Household size exposure# 29.18 31.09 .00 225.00Number of adults exposure# 14.57 15.52 .00 105.00

Number of observations 6,713Number of communities 166

Hence, there are just over 40 individuals and just below 20 households withineach community.

Tables 1 and 2 present the descriptive statistics for the sample of individualsand the sample of households, respectively. The variables are divided intodependent, individual, household, and instrumental variables. The four vari-ables that are treated as dependent or endogenous variables are attendance inan adult literacy program, log of per adult equivalent expenditure, literacy,and numeracy. This section presents the variables for the base sample and theassociated basic descriptive statistics. The following section examines the de-

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 9: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

24 economic development and cultural change

TABLE 2DESCRIPTIVE STATISTICS: HOUSEHOLDS

Variable Mean SD Min Max

Dependent:Attended literacy program .20 .40 .00 1.00

Log of per adult equivalent expenditure 13.87 .71 11.68 17.09Household:

Owns land .30 .46 .00 1.00Nonfarming household .13 .34 .00 1.00Average age:

30–34 .18 .39 .00 1.0035–39 .17 .38 .00 1.0040–49 .21 .40 .00 1.0050–59 .12 .32 .00 1.0060–99 .14 .34 .00 1.00

Percent with primary education .12 .26 .00 1.00Percent with middle education .28 .36 .00 1.00Percent with post-middle education .05 .18 .00 1.00Household size 4.37 2.54 1.00 21.00Number of adults in household 2.09 1.07 1.00 8.00

Instruments:Average age # exposure:

30–34 .91 2.59 .00 15.0035–39 .90 2.63 .00 15.0040–49 1.14 2.93 .00 15.0050–59 .66 2.32 .00 15.0060–99 .64 2.11 .00 15.00

Owns land exposure# 1.46 3.08 .00 15.00Nonfarming exposure# .58 2.11 .00 15.00Primary education (%) exposure# .66 1.83 .00 15.00Middle education (%) exposure# 1.54 2.75 .00 15.00Post-middle education (%) exposure# .25 1.06 .00 15.00Household size exposure# 23.34 26.05 .00 225.00Number of adults exposure# 11.18 12.10 .00 105.00

Number of observations 3,219Number of communities 166

scriptive statistics for program participation, welfare, literacy, and numeracyin more detail.

Information on participation in adult literacy programs comes from theEducation section of the GLSS 4, which includes the question: “Has [NAME]ever attended a literacy course?”13 All individuals who answer yes are codedone and the rest zero. Households where at least one adult member answersyes are coded one and those where nobody has participated are coded zero.For the base sample, about 11% of all individuals have participated in anadult literacy program, while 20% of the households have at least one memberwho has attended.

13 Note that there is unfortunately no information about whether the individual has finished thecourse or when participation took place.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 10: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 25

We use household expenditures per year as our measure of household welfare.As is often discussed in poverty analyses, not all household members requirethe same level of consumption; children and adults, for example, have differentcaloric requirements. We therefore use an adult-equivalent corrected measureof consumption, where different types of household members are assigneddifferent weights depending on their age and sex, and total consumption isdivided by the sum of these weights.14 Henceforth, when we refer to welfare,consumption, or expenditures, this should be taken to mean per adult equiv-alent expenditures per year.

Consumption or household expenditures are preferable to earnings or wagesas a measure of economic well-being for at least two reasons. First, if householdssmooth their consumption over time, the link between latent household eco-nomic well-being and earnings or wages is weaker than that between latenthousehold economic well-being and consumption expenditures. Second, con-sumption expenditures are likely to be more accurately measured than earningsor income (Appleton 2002). This is especially the case for our sample, whichconsists purely of rural households, most of which have agricultural productionas their main and often only income source.

The average of the log of per adult equivalent expenditure is around 13.8for both the individual and the household samples. This is approximately equalto 985,000 cedis per adult equivalent.15 For comparison, previous work onpoverty in Ghana using the same data set employ two poverty lines: an upperline at 900,000 cedis and a lower at 700,000 cedis (Canagarajah and Portner2003). Consuming above the upper poverty line means that one is able tomeet one’s nutritional requirements and basic nonfood needs. The lower povertyline is the level at which an individual is not able to meet his minimumnutrition requirements, even if he allocates his entire budget to food.

We consider a person literate if he can read a letter in either English or aGhanaian language, while he is considered numerate if he answers yes to beingable to do written calculations.16 Both the ability to read a letter and do writtencalculations are self-reported, although if the respondent was not present duringthe interview the main respondent is asked to supply the information. Of the

14 The results for the per capita consumption measure are available on request from the authors.They are broadly similar to the results for the adult-equivalent consumption measure.15 Note that in 2007 Ghana redenominated its currency with 10,000 old cedis equal to 1 newcedi. All values here are given in old cedis (which are called second cedi).16 The relevant questions in GLSS are “Can [NAME] read a letter in English?” and “In whatGhanaian language can [NAME] read a letter?” The former is answered yes/no, while the latterincludes the option “None.” Numeracy is based on the question: “Can [NAME] do written cal-culations?”

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 11: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

26 economic development and cultural change

individuals in the base sample, 41% report that they can read a letter in eitherEnglish or a Ghanaian language, while 45% can do written calculations. Thedistribution of literacy and numeracy by education levels and program par-ticipation is discussed below.

Slightly more than half of the individuals are female and, as expected giventhe population growth in Ghana, relatively young. To ease interpretation, weuse four age groupings for individuals: 18–29, 30–44, 45–59, and 60–99.These age groups roughly correspond to young adults, prime age adults, mid-dle-aged adults, and older adults and constitute 33%, 31%, 20%, and 16%of the sample, respectively. More than half (53%) of all adults in the samplehave not completed any education, while 12% have completed primary school,29% middle school, and 6% a post–middle school education.

Just over 30% live in households that own land, while around 10% are innonagricultural households, which are households that report no agriculturalproduction at all. The remainder hence live in households where the only landavailable is rented land. The three household level education variables measurethe percentage of individuals within a household that have completed a givenlevel of education. The distribution is approximately the same as for individualeducation and varies little whether one looks at the sample of individuals orthe household sample. It is worth noting that in just over half of all households,none of the adults have completed any formal education. In the individualsample, households consist of on average just over five persons, of which 2.6are adults, while in the household sample the corresponding numbers are 4.4and just over two.17 For the household sample, age is defined as average ageof the adults in the households and split into six groups: 18–29, 30–34, 35–39, 40–49, 50–59, and 60–99. These groups are of approximately the samesize.

We use as instruments the exposure to the literacy program in the com-munity interacted with individual and household characteristics. Exposure isdefined as the length of time the program has been in the community and iscapped at 15 years because the national program essentially was nonexistentfrom 1968 until 1986 and generally believed to be of little use before 1968.18

17 A possible concern with including variables reflecting household characteristics is that someindividuals might have participated in an adult literacy program before joining the households.Hence, those variables could, in principle, be influenced by the participation decision and subsequentoutcomes rather than the other way around. This could, for example, happen if a person who haslearned to read through an adult literacy program is a more attractive partner than somebody withthe same characteristics who has not. As we show below it appears that relatively few individualsare very young when they participated in the program, making these issues less of a concern.18 The relevant questions in the community questionnaire are: “Is there, or has there been, anadult literacy program in this community?” and “In what year was this program launched?”

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 12: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 27

For the individual sample, the variables interacted with exposure are female,the age groups, ownership of land, nonfarming household, the levels of edu-cation in the household in percent, household size, and number of adults inthe household. For the household sample, the same set is used except thatthere is no female dummy and the household average age groups are usedinstead of individual age groups.

What is important for the performance of the instruments is the role ex-posure plays in the decision to participate. One can think of two oppositeeffects of increasing exposure: one where the return to participation is knownand one where it is not. First, the longer the program has been in an area,the more likely it is that those individuals with higher expected net returnswill have participated. In practice, even with a higher expected return thanother individuals, there are likely factors preventing at least some of the highnet return individuals from participating right away.19 If this is the case, onewould expect increasing exposure to lead to a divergence in the proportionwho have participated between those types of individuals with a high expectednet return and those with a lower expected net return. The second possibleeffect arises from people learning about the return to participation over time.Even provided the program has a positive return, we would expect the mostrisk-averse (and/or credit-constrained) individuals to postpone participationuntil more and better information is available about the effect of participation.In this case, the expected result is a convergence in participation over time.

No matter which of these effects dominate, the important point is whetherthe interactions between exposure and characteristics capture differences in thelikelihood of participation for each of these groups. We will discuss this indetail below when we present the results. Furthermore, while we would expectage and the other variables to have a direct effect on outcomes, it is unlikelythat the interactions of these variables with exposure should have a direct effecton outcomes once we control for observable and unobservable communitycharacteristics through fixed effects and the direct effect of the individual andhousehold characteristics. This is the case even if exposure works mainlythrough learning about the return to participation. Consider as an examplethe difference in participation of men and women. If women are more riskaverse and/or otherwise constrained, that should clearly have an effect on theiraverage consumption. However, once we control for being a woman, the in-teraction between being a woman and exposure should not have an impact onconsumption except through participation in the adult literacy program.

19 This can, for example, be the case if there is not enough space to accommodate all interestedparticipants when they first apply to participate.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 13: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

28 economic development and cultural change

IV. Descriptive AnalysisThis section provides a more detailed look at the descriptive statistics forprogram participation, literacy, numeracy, and welfare. It first looks at partic-ipation, literacy, and numeracy and then at welfare. Table 3 shows participationrates by completed education level and literacy and numeracy rates for par-ticipants and nonparticipants, also by level of completed education. Not sur-prisingly, participation rates decrease with level of education. Of those whohave not completed primary school, 17% have attended an adult literacyprogram, while just over 10% of adults who have completed primary schoolhave. For those with middle school and those with more than middle schooleducation, the participation rates are 3.6% and 1.5%, respectively.20 The dif-ferences in participation rates across education levels do not reflect differencesin access to an adult literacy program. Around 80% of all adults in the samplehave access to an adult literacy program in their community, and there appearsto be little difference in access rate by education level.21

For all education levels participants in adult literacy programs generallyhave higher literacy and numeracy rates than those who have not participated.It is clear, however, that even among participants the literacy and numeracyrates are still low. For those without education, only about 15% of adults whohave participated in a program can read and do written calculations, comparedwith 3% literacy and 6% numeracy among nonparticipants. Among adultswho have completed primary school, literacy increases from 48% withoutparticipation to 57% with participation, while numeracy appears to declinefrom 67% to 61%. Only for graduates of middle school and above do literacyand numeracy rates approach 100%, and even then there are more than 7%of those who have graduated from middle school who report being unable toread a letter in either English or a Ghanaian language.

We can learn three important things from these results. First, it appearsthat the adult literacy programs in Ghana do poorly in terms of teachingpeople literacy and numeracy, although the efficiency rates are in line withthose discussed in Abadzi (2003). Second, that the literacy and numeracy ratesare so low even after 6 years of education indicates that the quality of therural primary schools in Ghana is indeed low. Finally, it appears that over-estimation of respondents’ ability to read and do math is not a serious issue.The latter is important since there is often a perception that self-reporting ofliteracy will lead to literacy rates that are too high. Ortega and Rodriguez

20 The low-participation rates for the two highest education levels mean that the sample of par-ticipants from these groups consists of only 75 adults (69 for middle and 6 for post-middle).21 The possible exception is among those with post-middle education, where just over 71% currentlyhave access.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 14: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

29

TAB

LE3

PAR

TIC

IPA

TIO

N,

LITE

RA

CY,

AN

DN

UM

ER

AC

YR

ATE

SB

YP

RO

GR

AM

AV

AIL

AB

ILIT

YA

ND

ED

UC

ATI

ON

Par

tici

pan

tsN

onp

arti

cip

ants

Co

mp

lete

dE

duc

atio

nP

arti

cip

ated

Lite

rate

Num

erat

eLi

tera

teN

umer

ate

Num

ber

of

Ob

serv

atio

ns

Co

mp

lete

sam

ple

:N

one

/Ko

rani

c.1

70.1

48.1

49.0

31.0

6235

92P

rim

ary

.103

.571

.607

.480

.667

816

Mid

dle

.036

.986

.971

.926

.955

1914

Po

st-m

idd

le.0

151.

000

1.00

0.9

92.9

9239

1N

op

rog

ram

avai

lab

le:

No

ne/K

ora

nic

.119

.177

.167

.027

.065

809

Pri

mar

y.0

72.8

18.8

18.4

33.6

8815

2M

idd

le.0

371.

000

1.00

0.9

10.9

6738

2P

ost

-mid

dle

.009

1.00

01.

000

.991

.982

111

Pro

gra

mav

aila

ble

:N

one

/Ko

rani

c.1

84.1

42.1

46.0

32.0

6127

83P

rim

ary

.110

.534

.575

.491

.662

664

Mid

dle

.036

.982

.964

.930

.952

1532

Po

st-m

idd

le.0

181.

000

1.00

0.9

93.9

9628

0

No

te.

Ap

erso

nis

cons

ider

edlit

erat

eif

heca

nre

ada

lett

erin

eith

erE

nglis

ho

ra

Gha

naia

nla

ngua

ge

and

num

erat

eif

hean

swer

sye

sto

bei

ngab

leto

do

wri

tten

calc

ulat

ions

.P

rog

ram

avai

lab

ility

isb

ased

on

the

que

stio

nin

the

com

mun

ity

que

stio

nnai

re:

“Is

ther

e,o

rha

sth

ere

bee

n,an

adul

tlit

erac

yp

rog

ram

inth

isco

mm

unity

?”

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 15: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

30 economic development and cultural change

(2008), for example, argue that semiliterate persons might claim that they areliterate after participation in a “Mision Robinson” course. It is, however, alsopossible that participation in an adult literacy program will lead to participantslearning that they really cannot read.

A noticeable feature of the data is that a substantial number of people reporthaving participated in an adult literacy program but live in communitieswithout a program. Not surprisingly, the participation rate is generally largerin areas with a program, although the differences are not as substantial as onemight expect: the largest difference is for individuals without any completededucation, where the participation rate is about 19% in areas with a programand 12% in communities without. The differences in literacy and numeracyrates among nonparticipants in communities with and without an adult literacyprogram are small, with literacy rates lower among those who live in areaswithout a program. There are, however, substantial differences between par-ticipants who live in communities with an adult literacy program and par-ticipants in communities without a program. What is particularly interestingis that the literacy and numeracy rates are higher in communities without aprogram. For participants with no completed education the literacy rates are17.7% and 14.3% in communities without and with a program, respectively.Especially striking are the corresponding rates for those who have completedprimary school, where the literacy rate for participants in communities withouta program is 81.8% while it is only 52.8% among participants in communitieswith a program.

The high participation rate and the higher literacy and numeracy ratesamong individuals who state they have participated in an adult literacy pro-gram, yet live in communities where a program is not available, are importantbecause they influence the strength of our instrumental variable estimationstrategy. As discussed above, we include community fixed effects to controlfor endogenous program placement. Then to account for unobservable indi-vidual characteristics that may influence both the decision to participate andthe outcomes, we rely on the length of exposure to the program interactedwith observable characteristics to instrument for participation. This means thatour instruments, once we control for community fixed effects, will have noexplanatory power for individuals in areas without an adult literacy programsince exposure is zero. Given that the effect of participation is only identifiedfor those affected by the instruments, our instrumental variable estimationsessentially ignore anybody living in a community without an adult literacyprogram, which includes the subgroup of participants who have higher literacyand numeracy rates than other participants. We will discuss this further inSection V.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 16: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 31

TABLE 4PER ADULT EQUIVALENT EXPENDITURES BY PARTICIPATION AND EDUCATION LEVEL OF HOUSEHOLD

Participant Meana Min Max SD N

Complete sample:No 1,387 118 26,486 1,198 2,588Yes 1,296 183 11,110 1,150 631All 1,369 118 26,486 1,189 3,219

Post-middle:b

No 1,731 208 26,486 1,937 281Yes 945 183 2,657 530 42All 1,629 183 26,486 1,835 323

Middle:b

No 1,438 170 7,615 1,014 1,006Yes 1,151 236 11,110 1,000 229All 1,385 170 11,110 1,017 1,235

Primary:b

No 1,377 247 6,198 1,025 315Yes 1,261 252 8,510 1,260 74All 1,355 247 8,510 1,073 389

None:b

No 1,240 118 12,240 1,124 986Yes 1,473 195 9,287 1,268 286All 1,292 118 12,240 1,161 1,272

Note. Participation status reflects whether at least one member has participated in an adult literacyprogram.a Expenditures measured in 1,000 cedis.b At least one adult with this level of education and none with more.

Table 4 shows levels of consumption and associated statistics by participationand education using the household sample. In addition to the complete sample,households are divided into four exclusive groups by the highest educationlevel among its adult members. Hence, say, the “Middle” households all haveat least one member who has completed middle school and no members whohave completed an education level higher than that. Within each group thestatistics are presented for both those households that have had no membersparticipate in an adult literacy program, those that have had members par-ticipate, and the two categories combined.

As expected, consumption increases with the level of highest educationwithin the household, although the differences might not seem large. Whilethe difference in mean consumption between households where no adults havecompleted any level of schooling and those where at least one adult has com-pleted a post-middle school education is 26%, the differences between the no-education group and the middle and primary education groups are only 7.3%and 5.0%, respectively. Furthermore, even if we, in general, measure con-sumption more precisely than income, there is still potential for measurementerror. In the case of subsistence households, who may barter for the part oftheir consumption that they do not produce themselves, we impute expen-

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 17: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

32 economic development and cultural change

ditures, which obviously introduce the possibility of errors. This may explainthe very low values for the minimum observation expenditures in table 4,which are substantially below the lower poverty line, even for the bettereducated households.22

What is mainly of interest here are the differences between households withadults who have participated in a literacy program and those without. Theonly group where average consumption is higher among participants thannonparticipants is the no-education group, where there is almost a 19% dif-ference. For everybody else, the participant households have substantially lowerconsumption than the nonparticipant households; participant middle-educa-tion households have an average consumption that is more than 20% lowerthan nonparticipant households.23 Note that it is difficult to say anythingabout the causal impact of participating in an adult literacy program onhousehold consumption from these numbers. It is possible that the participanthouseholds would have done even worse in the absence of participation or thatthere are substantial differences in area characteristics that influence con-sumption.

To investigate these possibilities further, table 5 presents the statistics fromtable 4 stratified by adult literacy program availability, participation status,and education group. Two interesting findings emerge. First, communitieswith adult literacy programs generally appear to be poorer than those without.This holds whether we compare nonparticipant households across communitieswith and without literacy programs or compare participants and nonpartici-pants combined across communities. One interpretation of this is that less-developed communities are more likely recipients of adult literacy programsthan more developed ones. This underscores the potential importance of unob-servable community characteristics that affect the placement of programs evenafter the level of development or the schooling level of the community aretaken into account.

A second important finding is that for all but the post-middle-educationgroup, participant households that live in communities without an adult lit-eracy program have substantially higher average consumption than participanthouseholds that live in areas with an adult literacy program. The differencesare 46%, 64%, and 32% for the no-education, primary, and middle-schoolgroups, respectively. Furthermore, these households constitute more than 16%

22 It is, of course, possible that the poverty lines themselves are not perfect measures of the actualcost of the minimum nutritional requirements.23 The biggest difference is for post-middle education households, but there are very few participantsin this group.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 18: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 33

TABLE 5PER ADULT EQUIVALENT EXPENDITURE BY PARTICIPATION, PROGRAM AVAILABILITY, AND

EDUCATION LEVEL WITHIN HOUSEHOLD

Program Available Participant Meana Min Max SD N

Complete sample:No No 1,547 170 26,486 1,683 598

Yes 1,681 270 11,110 1,562 104Yes No 1,339 118 12,240 1,004 1,990

Yes 1,220 183 9,287 1,035 527All 1,369 118 26,486 1,189 3,219

Post-middle:b

No No 2,298 256 26,486 3,326 72Yes 871 428 1,695 456 8

Yes No 1,536 208 6,860 1,060 209Yes 962 183 2,657 551 34All 1,629 183 26,486 1,835 323

Middle:b

No No 1,632 198 7,615 1,360 204Yes 1,435 270 11,110 1,655 41

Yes No 1,388 170 6,009 899 802Yes 1,089 236 7,246 783 188All 1,385 170 11,110 1,017 1,235

Primary:b

No No 1,474 260 6,072 1,118 69Yes 1,895 670 6,884 1,795 11

Yes No 1,349 247 6,198 999 246Yes 1,151 252 8,510 1,126 63All 1,355 247 8,510 1,073 389

None:b

No No 1,285 170 9,133 1,236 253Yes 2,003 346 6,710 1,488 44

Yes No 1,224 118 12,240 1,082 733Yes 1,377 195 9,287 1,203 242All 1,292 118 12,240 1,161 1,272

Note. Participation status reflects whether at least one member has participated in an adult literacyprogram.a Expenditures measured in 1,000 cedis.b At least one adult with this level of education and none with more.

of all the households that have adult members who have participated in anadult literacy program. As discussed above, our instrumental variable esti-mations essentially ignore this subgroup of participants who appears to dovery well compared to other participants.

With the substantial number of individuals who have participated in anadult literacy program but live in a community where no program is available,and noting these participants’ overall higher welfare, it is worthwhile exam-ining possible explanations for how and why they came to participate. Themost obvious explanation is that people travel to neighboring communitiesto participate when one is not available in their own community. Second,individuals who have participated somewhere else could migrate in search of

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 19: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

34 economic development and cultural change

better opportunities. Finally, the community questionnaire could have mis-coded the presence of an adult literacy program. Starting with the latter, itis difficult to imagine why there would be a correlation between how wealthya community is and coding errors in the community questionnaire; nonetheless,there are three communities in which eight households (out of 20) have par-ticipated, but where the community questionnaire does not show an adultliteracy program. This is, of course, also a possible outcome if individualstravel to nearby communities to participate. Unfortunately, there is no infor-mation in the GLSS 4 about neighboring communities or where or whensomebody participated in a program, which means that we cannot determinethe extent to which this has taken place. We are in a better position to examinemigration.24 While the nonparticipants are similar with respect to migration,fewer participants in areas without a program are born in their current placeof residence than participants in areas with a program. Furthermore, substan-tially fewer have always lived in the village or town they are currently livingin and are more likely to have moved away from their current residence for ayear or more. Hence, we cannot rule out that the better performance of theparticipants in communities without a program is due to a combination ofboth participation in neighboring communities and selective migration. Forboth possible explanations it is, however, the case that these participants mightbe more motivated, which could explain why these participants do better onaverage both in terms of the consumption of their households and their skilllevels as discussed above.

V. Welfare Effects of ParticipationThe descriptive analyses indicate that while there is only a small effect ofattending an adult literacy program on literacy and numeracy, the effect onwelfare is sizable. The problem with the cross-tabulations is that they do nottake account of the problems of potential endogenous program placement andself-selection into participation and may therefore be biased. Hence, this sectionpresents estimates of the effects of participation on welfare, which is the mainoutcome of interest. We first present the results with community fixed effectsand compare them to the ordinary least squares (OLS) estimates. This isfollowed by a reduced form analysis and finally by the instrumental variablescommunity fixed-effects results.

Tables 6 and 7 show the results for the community fixed-effects models forthe individual and household samples, respectively. When using the full sam-ples, the estimated effect of participation in an adult literacy program is

24 The tables are available upon request.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 20: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

35

TABLE 6INDIVIDUAL PARTICIPATION’S EFFECT ON HOUSEHOLD WELFARE: COMMUNITY FIXED EFFECTS

Level of Completed Education in Householda

VariablesComplete

Sample Post-Middle Middle Primary None

Attended literacy program .052** -.057 .019 .031 .108***(.020) (.055) (.026) (.043) (.028)

Female .020* -.011 -.020 .012 .044***(.010) (.020) (.015) (.024) (.015)

Age:30–44 .006 .079** -.013 -.063 .030

(.015) (.036) (.023) (.039) (.029)45–59 .021 .035 -.027 -.036 .066**

(.016) (.036) (.023) (.045) (.030)60–99 .054** .005 -.017 -.078 .119***

(.021) (.043) (.032) (.072) (.034)Primary education .024*** -.011 -.008 -.013

(.008) (.016) (.010) (.023)Middle education .033*** -.003 -.011

(.010) (.023) (.014)Post-middle education .039*** -.012

(.013) (.026)Owns land .044 .086 .059 .092 -.018

(.027) (.104) (.037) (.087) (.057)Nonfarming household .070 .143 .119 .062 -.077

(.046) (.112) (.092) (.167) (.072)Percent with primary educa-

tion.052 .155 -.001 .338**

(.039) (.264) (.082) (.145)Percent with middle education .179*** .161 .339***

(.030) (.151) (.077)Percent with post-middle edu-

cation.380*** .630***

(.067) (.191)Household size -.089*** -.077*** -.084*** -.089*** -.112***

(.008) (.016) (.010) (.025) (.012)Number of adults in house-

hold.022 .069** .022 .076 -.015

(.014) (.029) (.021) (.056) (.034)Constant 14.023*** 13.751*** 14.030*** 13.822*** 14.075***

(.038) (.193) (.083) (.157) (.064)Observations 6,713 883 2,849 779 2,202Communities 166 127 161 135 162R2 .19 .21 .21 .16 .27

Note. Robust standard errors in parentheses. Dependent variable is log of per adult equivalent expen-diture per year.a At least one adult with this level of education and none with more within household.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 21: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

36

TAB

LE7

HO

USE

HO

LDPA

RTI

CIP

ATI

ON

’SE

FFE

CT

ON

HO

USE

HO

LDW

ELF

AR

E:

CO

MM

UN

ITY

FIX

ED

EFF

EC

TS

Leve

lof

Co

mp

lete

dE

duc

atio

nin

Ho

useh

old

a

Vari

able

sC

om

ple

teSa

mp

leP

ost

-Mid

dle

Mid

dle

Pri

mar

yN

one

Att

end

edlit

erac

yp

rog

ram

.055

***

�.1

08.0

24.0

31.1

39**

*(.0

21)

(.104

)(.0

35)

(.069

)(.0

27)

Ow

nsla

nd.0

48*

.029

.052

.096

.008

(.026

)(.1

20)

(.038

)(.0

91)

(.049

)N

onf

arm

ing

hous

eho

ld�

.027

.016

.025

.001

�.1

04(.0

44)

(.125

)(.0

72)

(.158

)(.0

75)

Ave

rag

eag

e:30

–34

�.0

53**

�.0

21�

.063

�.0

86�

.082

(.026

)(.1

05)

(.040

)(.0

93)

(.056

)35

–39

�.0

14.2

72**

�.0

70*

�.0

38�

.008

(.031

)(.1

34)

(.039

)(.0

96)

(.061

)40

–49

�.0

40.1

13�

.084

*�

.192

**.0

27(.0

31)

(.135

)(.0

47)

(.091

)(.0

55)

50–5

9.0

34.1

04.0

14�

.184

.038

(.035

)(.1

44)

(.060

)(.1

24)

(.054

)60

–99

.075

*.1

97�

.031

�.0

78.0

72(.0

38)

(.220

)(.0

90)

(.184

)(.0

57)

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 22: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

37

Per

cent

wit

hp

rim

ary

educ

atio

n.1

21**

*.5

37**

.028

.272

(.037

)(.2

68)

(.075

)(.1

83)

Per

cent

wit

hm

idd

leed

ucat

ion

.251

***

.325

*.3

57**

*(.0

32)

(.196

)(.0

73)

Per

cent

wit

hp

ost

-mid

dle

educ

atio

n.4

04**

*.7

52**

*(.0

66)

(.255

)H

ous

eho

ldsi

ze�

.118

***

�.0

95**

*�

.103

***

�.1

18**

*�

.144

***

(.006

)(.0

19)

(.010

)(.0

20)

(.010

)N

umb

ero

fad

ults

inho

useh

old

.036

***

.064

*.0

28.0

89.0

21(.0

13)

(.036

)(.0

20)

(.055

)(.0

30)

Co

nsta

nt14

.191

***

13.7

56**

*14

.147

***

14.0

93**

*14

.238

***

(.036

)(.2

51)

(.083

)(.2

03)

(.061

)O

bse

rvat

ions

3,21

932

31,

235

389

1,27

2C

om

mun

itie

s16

612

716

113

516

2

No

te.

Ro

bus

tst

and

ard

erro

rsin

par

enth

eses

.A

hous

eho

ldis

cons

ider

edto

have

atte

nded

ifat

leas

to

neo

fit

sm

emb

ers

has

par

tici

pat

edin

anad

ult

liter

acy

pro

gra

m.

aA

tle

ast

one

adul

tw

ithth

isle

velo

fed

ucat

ion

and

none

wit

hm

ore

wit

hin

hous

eho

ld.

*St

atis

tica

llysi

gni

fican

tat

10%

.**

Stat

isti

cally

sig

nific

ant

at5%

.**

*St

atis

tica

llysi

gni

fican

tat

1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 23: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

38 economic development and cultural change

equivalent to a more than 5% increase in consumption for both the individuallevel and household level estimations. Furthermore, this effect is statisticallysignificant for both samples. It is, however, likely that this overall effect maskssubstantial differences in the effect of participation across different types ofhouseholds. Hence, as for the descriptive statistics in tables 4 and 5, we dividethe samples into groups depending on the highest education level completedby an adult in the household. The idea behind this is that the marginal effectof participation should be stronger if the human capital level of the householdis lower.

The most interesting results are the estimates for the households withoutcompleted education. They show an increase in consumption of more than10% for the individual level estimation and almost 14% for the householdlevel estimation, and both estimates are strongly statistically significant. Thiseffect is substantial, especially considering that most of the literature has foundmodest or no effect of literacy programs. It is worth keeping in mind, however,that the Ghanaian literacy programs take substantially longer to complete thanmost and teach topics other than literacy, as detailed in Section I. A moresuitable comparison might therefore be the return to education in Ghana,which Canagarajah and Thomas (1997) found to be between 4% and 6% peryear of schooling completed. Given the length of the program and the effortrequired by participants, these returns are comparable.

The effect of participation is substantially lower for the other groups anddeclines with increasing levels of completed education. For primary-educationhouseholds, the estimated effect of participation is just over 3%, while theeffect of participation for the middle-education households is around 2%,although none of these effects are statistically significant. The effect for the“post-middle” group is negative and large, although again not statisticallysignificant. In interpreting these results, it is worth keeping in mind that thesamples are relatively small. This might especially be a problem for the “pri-mary” group since there are only 389 households in that sample but they arespread over 135 communities. Hence, it is possible that the statistical insig-nificance for this group is mainly due to the small sample size combined witha large number of community fixed effects.25

Comparison of the community fixed-effects results with the standard OLSresults provides an indication of whether endogenous program placement is asignificant problem. Appendix tables A1 and A2 show the equivalent results

25 The “post-middle” group consists of only 323 households, of which 42 have had at least onemember participate in an adult literacy program, while in the “middle” group there are 1,235households with 229 participating.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 24: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 39

to tables 6 and 7 except that community fixed effects are not included. TheOLS results show that the estimated average effect of participation in an adultliteracy program is about a 10% increase in consumption per adult equivalent,while the OLS effect for the no-education group is over 20%. In other words,the OLS estimates are twice the size of the community fixed-effects estimates.The large differences between the OLS and fixed-effects estimates indicate thatendogenous program placement is a real concern in this case and lead tosubstantial biases in the estimated effects of participation. Furthermore, itappears that programs are allocated, at least partly, based on where they wouldgenerate the highest return.

A. Reduced Form ResultsWhile community fixed effects control for endogenous program placement,there is still the potential issue of self-selection into participation based onunobservable characteristics that could also influence the welfare of the house-hold. As we discuss above, finding suitable instruments is a challenge oncewe include community fixed effects. To provide an idea of how the instru-ments—exposure-interacted with a set of individual and household charac-teristics—influence welfare, tables 8 and 9 show reduced form estimationswith the instruments included instead of participation in adult literacy pro-gram. These results are shown only for the no-education sample for four modelswith an increasing number of variables included.26

For the sample of individuals, the main statistically significant effects arefrom the interactions between exposure and being in age group 45–59 andhousehold size. The effects for the other two age groups relative to the 18–29 age groups are substantially smaller. This seems to indicate that the effectof having an adult literacy program in the community is mainly concentratedin the middle-aged group.

The results for the sample of households show a strong effect of exposureinteracted with average age of adults being 40–49. The estimates for otheraverage age groups interacted with exposure indicate an effect that is almostas large in size, although none of the other age groups interacted with exposureare statistically significant. Finally, note that the interaction variables are jointlystatistically significant for the household level estimations but not for theindividual level estimations.

B. Instrumental Variables EstimationsAn important aspect of the instruments is their ability to explain the decisionto participate. Tables 10 and 11 show the estimates for individual and house-26 The corresponding results for the other groups are available upon request.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 25: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

40 economic development and cultural change

TABLE 8INDIVIDUAL LEVEL REDUCED FORM EFFECTS ON HOUSEHOLD WELFARE FOR

NO-EDUCATION SAMPLE: COMMUNITY FIXED EFFECTS

Variables Model I Model II Model III Model IV

Female .011 .009 .009 .003(.025) (.026) (.026) (.026)

Age:30–44 .026 .023 .023 .004

(.050) (.051) (.051) (.049)45–59 .008 .001 .001 �.003

(.051) (.052) (.052) (.051)60–99 .083 .075 .075 .080

(.059) (.059) (.059) (.057)Owns land �.021 .081 .081 .084

(.056) (.089) (.089) (.086)Nonfarming household �.091 �.070 �.070 �.064

(.073) (.105) (.105) (.106)Household size �.113*** �.113*** �.113*** �.091***

(.012) (.012) (.012) (.017)Number of adults in household �.016 �.017 �.017 �.046

(.034) (.034) (.034) (.041)Female exposure# .005 .005 .005 .007

(.004) (.004) (.004) (.004)Age # exposure:

30–44 .002 .003 .003 .008(.006) (.007) (.007) (.006)

45–59 .013* .015** .015** .017**(.007) (.008) (.008) (.007)

60–99 .007 .009 .009 .009(.008) (.008) (.008) (.008)

Owns land exposure# �.021 �.021 �.021(.014) (.014) (.014)

Nonfarming exposure# �.005 �.005 �.006(.018) (.018) (.018)

Household size exposure# �.005*(.003)

Number of adults exposure# .006(.006)

Observations 2,202 2,202 2,202 2,202Communities 162 162 162 162

Note. Robust standard errors in parentheses. Sample consists of households where no adult memberhas completed any level of schooling.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

hold participation, together with the test statistics for all identifying in-2x

struments being jointly equal to zero. In both cases, the instrumental variablesperform well for the complete sample and the no-education households. Thebest performance is for the individual sample with only no-education house-holds, where the identifying instruments are jointly statistically significant atless than the 1% level, while the corresponding level for the sample of house-holds is worse at the 8% significance level.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 26: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

41

TABLE 9HOUSEHOLD LEVEL REDUCED FORM EFFECTS ON HOUSEHOLD WELFARE FOR

NO-EDUCATION SAMPLE: COMMUNITY FIXED EFFECTS

Variables Model I Model II Model II

Owns land .001 .110 .113(.048) (.078) (.078)

Nonfarming household �.125 �.104 �.101(.077) (.114) (.114)

Average age:30–34 .000 �.015 �.034

(.088) (.091) (.089)35–39 �.062 �.059 �.076

(.096) (.095) (.099)40–49 �.060 �.070 �.080

(.088) (.090) (.090)50–59 �.004 �.018 �.023

(.079) (.082) (.082)60–99 .012 �.002 �.003

(.091) (.094) (.094)Household size �.146*** �.146*** �.133***

(.010) (.011) (.016)Number of adults in household .028 .028 .005

(.032) (.032) (.039)Average age # exposure

30–34 �.017 �.013 �.010(.014) (.015) (.014)

35–39 .012 .013 .017(.013) (.013) (.014)

40–49 .018 .021 .024*

(.013) (.014) (.014)50–59 .010 .015 .017

(.012) (.013) (.014)60–99 .012 .016 .017

(.014) (.014) (.014)Owns land exposure# �.023* �.023

(.013) (.014)Nonfarming exposure# �.004 �.005

(.020) (.020)Household size exposure# �.003

(.003)Number of adults exposure# .005

(.006)Observations 1,272 1,272 1,272Communities 162 162 162

Note. Robust standard errors in parentheses. Sample consists of households where no adultmember has completed any level of schooling.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 27: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

42

TABLE 10INDIVIDUAL PARTICIPATION IN LITERACY PROGRAM: COMMUNITY FIXED-EFFECTS LOGIT

Level of Completed Education in Householda

VariablesComplete

Sample Post-Middle Middle Primary None

Female �.438*** 1.715** �.120 �.174 �.863***(.157) (.873) (.284) (.546) (.247)

Age:30–44 .778*** 1.778 .953** .646 .562

(.238) (1.455) (.372) (.937) (.417)45–59 1.096*** .455 1.603*** 1.487 .786*

(.240) (1.145) (.394) (.915) (.418)60–99 .815*** 2.039 1.868*** 1.428 �.156

(.255) (1.405) (.446) (1.011) (.432)Primary education �.688*** .646 �.560* �.388

(.209) (.905) (.324) (.405)Middle education �2.254*** -1.795** �2.012***

(.199) (.775) (.255)Post-middle education �2.932*** �2.291***

(.503) (.835)Owns land �.007 1.202 .096 .326 �.112

(.206) (3.005) (.356) (.880) (.377)Nonfarming household �.194 1.426 .355 . .776

(.378) (3.249) (.750) . (.587)Percent with primary edu-

cation�.306 .682 .362 .210

(.410) (4.848) (.940) (1.992)Percent with middle edu-

cation.547* �.706 .042

(.327) (4.514) (.807)Percent with post-middle

education.302 �9.396*

(.778) (5.293)Household size �.035 1.081** .007 �.152 �.090

(.040) (.425) (.065) (.197) (.070)Number of adults in

household.029 �2.529** .075 .451 �.137

(.098) (.996) (.174) (.490) (.223)Female exposure# .023 �.168 .030 .029 .020

(.022) (.114) (.039) (.072) (.038)Age # exposure:

30–44 �.016 �.256 �.036 .138 �.013(.034) (.205) (.052) (.147) (.061)

45–59 �.031 .021 �.058 .147 �.059(.035) (.149) (.054) (.145) (.063)

60–99 �.009 �.271 �.092 .136 .039(.037) (.191) (.061) (.157) (.065)

Primary (%) exposure# .071 �.597 .047 .041(.048) (.674) (.113) (.227)

Middle (%) exposure# .025 .267 .085(.043) (.733) (.105)

Post-middle (%) expo-#

sure.058 1.596*

(.107) (.827)

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 28: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 43

TABLE 10 (Continued)

Level of Completed Education in Household

VariablesComplete

Sample Post-Middle Middle Primary None

Owns land exposure# .036 �.407 .011 .164 �.008(.031) (.375) (.054) (.155) (.056)

Nonfarming exposure# �.166** �1.167* �.113 1.395 �.568***(.072) (.633) (.125) (179.708) (.149)

Household size expo-#

sure.014** �.091 .009 .042 .018*

(.006) (.071) (.010) (.030) (.010)Number of adults expo-#

sure�.045*** .227 �.044* �.126* �.027

(.014) (.143) (.024) (.068) (.034)Observations 5,813 287 2,093 360 1,598Communities 145 37 104 49 106

test: Instruments p 02x 20.47** 10.60 10.13 8.13 20.41***Prob 21 x .039 .477 .429 .521 .009

Note. Robust standard errors in parentheses. Dependent variable is whether the individual has partic-ipated in an adult literacy program or not.a At least one adult with this level of education and none with more within household.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

The instruments perform substantially worse for the other three subgroupsand are far from being statistically significant.27 A major reason for this poorperformance is likely to be the low number of observations, which is exacerbatedby our use of community fixed effects, which both restricts the number ofavailable instruments and lowers the number of degrees of freedom. Thishighlights the trade-off between controlling for endogenous program place-ment and controlling for self-selection into program participation when eval-uating programs: the more finely we define a community, the harder it becomesto find instruments that can be used to deal with unobservable characteristicsthat influence both participation and the outcomes of interest. As we mentionedabove, the standard OLS results are twice as large as the community fixed-effects results, indicating that endogenous program placement may indeed bea serious concern in this case. There are, however, clear advantages to usingcommunity fixed effects beside accounting for endogenous program placement.In our data, as in most other household surveys, there is little or no informationon a number of factors, such as land quality, which may affect the decision to

27 No convergence was achieved for the sample of households where the highest education levelwas “post-middle.”

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 29: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

44

TABLE 11HOUSEHOLD PARTICIPATION IN LITERACY PROGRAM: COMMUNITY FIXED-EFFECTS LOGIT

Level of Completed Education in Householda

VariablesComplete

Sample Post-Middleb Middle Primary None

Owns land .101 .236 3.227** �.083(.235) (.432) (1.353) (.415)

Nonfarming household �.103 .319 �2.089 .569(.380) (.862) (1,509.256) (.586)

Average age:30–34 .100 .067 �2.782 .183

(.334) (.546) (1.806) (.676)35–39 .666** 1.064** �3.842* .594

(.315) (.491) (2.127) (.606)40–49 .559* .728 �3.489* .327

(.313) (.523) (1.816) (.596)50–59 .684** 1.436** �5.687*** .689

(.341) (.635) (2.197) (.578)60–99 .286 .495 �2.239 �.018

(.353) (.937) (2.232) (.567)Percent with primary ed-

ucation�.984** �1.187 .651

(.384) (.997) (2.793)Percent with middle edu-

cation�1.306*** �3.157***

(.327) (.889)Percent with post-middle

education�2.334***

(.766)Household size �.029 .046 �.237 �.096

(.047) (.080) (.295) (.081)Number of adults in

household.518*** .409** 1.979** .254

(.113) (.205) (.906) (.238)Avg age # exposure:

30–34 .008 .002 .588* .042(.049) (.078) (.333) (.101)

35–39 �.019 �.049 1.097*** �.062(.046) (.071) (.410) (.091)

40–49 �.073 �.079 .962*** �.075(.045) (.072) (.356) (.092)

50–59 �.022 �.062 1.063*** �.041(.050) (.087) (.396) (.088)

60–99 .028 .110 .872** �.017(.054) (.115) (.402) (.090)

Primary exposure# .084 .119 �.000(.054) (.139) (.311)

Middle exposure# .015 .089(.048) (.120)

Post-middle exposure# .130(.114)

Owns land exposure# .032 �.013 �.316 .017(.037) (.065) (.214) (.064)

Nonfarming exposure# �.174** �.135 �.595 �.474***

(.070) (.139) (215.033) (.150)

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 30: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 45

TABLE 11 (Continued)

Level of Completed Education in Household

VariablesComplete

Sample Post-Middle Middle Primary None

Household size expo-#

sure.016** .011 .055 .019

(.007) (.012) (.048) (.012)Number of adults ex-#

posure�.048*** �.050* �.250** �.029

(.016) (.029) (.126) (.038)Observations 2,820 891 164 938Communities 145 102 44 106

test: Instruments p 02x 22.78** 10.43 10.55 15.45*

Prob 21 x .030 .492 .394 .079

Note. Robust standard errors in parentheses. The dependent variable is whether at least one of thehousehold’s members has participated in an adult literacy program or not.a At least one adult with this level of education and none with more within household.b No convergence was obtained, so results have been excluded.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

participate and the outcomes. To the extent that most of the variation of thesevariables is between villages, the use of community fixed effects will helpalleviate the bias from these unobserved variables.

The effects of the other explanatory variables on participation are generallyin line with what we would expect. Females are less likely to participate thanmales, although it appears that most of this effect comes from the groupwithout any education. For those with post-middle, middle, or primary schooleducation, women are, in fact, more likely to participate, although the effectis only statistically significant for the males. As expected, given that we measureage at the time of the survey rather than age when participating in the program,the age profile is relatively flat; it is, however, consistent with the programbeing in the areas for up to 15 years. The more education an individual hascompleted the less likely she is to participate. Neither of the variables relatingto land are statistically significant.

With respect to the instruments, none of the age groups have a statisticallysignificant effect on participation. The negative estimated effects for the twoyoungest groups indicate that the longer the program has been available, theless age matters for whether somebody has participated; however, part of thateffect could come from simple age progression where those who participatedearly move up to another age group. The nonfarming interaction with exposureis large and negative relative to the estimated direct effect on participation ofbeing a nonfarming household. The likely interpretation is not so much that

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 31: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

46 economic development and cultural change

nonfarming households become less likely to participate the longer the programhas been available, but rather that the other types of households pull furtheraway from the nonfarming ones and that the gap widens over time.

Tables 12 and 13 present the instrumental variable community fixed-effectsresults for the effect of participation on consumption. As is clear from above,the instruments perform satisfactorily only for the complete sample and forthe no-education group; nonetheless, the estimates for the other groups areincluded for completeness. Again, the most interesting results are for the no-education group. Using the individual level data, the estimated effect of par-ticipation increases slightly after instrumentation from 0.108 to 0.117, whilethe household level data show a decrease from 0.139 to 0.085. As expected,especially given the large number of community fixed effects, the standarderrors for both samples increase substantially, and none of these effects arestatistically significant. Hence, the point estimates still indicate a substantialeffect of participation in an adult literacy program on expenditure, and it doesnot appear that there is a large effect of unobservable individual and householdcharacteristics on the participation decision.

In summary, there is a strong effect of participation for households whereno adult has completed any level of schooling with an estimated effect ofbetween 8.5% and 14% increase in expenditures, depending on specification.The instruments are not strong enough to produce believable results for house-holds other than the no-education households, but the results from the standardcommunity fixed-effects estimations are interesting.28 The effect of partici-pation for households where the most educated adult has completed primaryschool is around 3%, while for households where the most educated adult hascompleted middle school the effect is between 1.9% and 2.4%. Clearly, theeffects are only statistically significant for no-education households, but theseresults show that there might be benefits to participation even for householdswith higher levels of human capital and in line with the standard theory ofdecreasing marginal return to human capital the effect of participation declineswith prior level of household human capital.

VI. What Explains the Effect on Welfare?This section examines some of the potential pathways through which partic-ipation in adult literacy programs affect consumption and whether there arealternative explanations for the effect on welfare. Given that the strongest andmost robust results are for the group of households where no adult memberhas finished even primary school, this section focuses on this group. In addition,

28 Those estimates obviously do not control for possible self-selection into participation.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 32: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

47

TABLE 12INDIVIDUAL PARTICIPATION’S EFFECT ON HOUSEHOLD WELFARE: IV COMMUNITY FIXED EFFECTS

Level of Completed Education in Householda

VariablesComplete

Sample Post-Middle Middle Primary None

Attended literacy pro-gram

.064 �.166 �.129 �.060 .117

(.184) (.153) (.389) (.374) (.224)Female .027* �.008 �.009 .007 .045

(.014) (.056) (.022) (.053) (.035)Age:

30–44 .007 .159*** �.021 �.031 .015(.018) (.061) (.030) (.082) (.040)

45–59 .033 .100 �.018 �.036 .070*(.023) (.063) (.050) (.132) (.039)

60–99 .061*** .069 .001 �.050 .108***(.023) (.079) (.060) (.136) (.037)

Primary education .030 .001 �.019 �.021(.031) (.109) (.050) (.066)

Middle education .042 �.013 �.034(.040) (.087) (.079)

Post-middle education .051 �.034(.055) (.084)

Owns land .037** �.004 .072*** .091 �.072*(.017) (.088) (.027) (.103) (.039)

Nonfarming household .055* .099 .084 .072 �.115*(.031) (.149) (.055) (.135) (.068)

Percent with primaryeducation

.031 �.327 �.019 .273*

(.039) (.233) (.072) (.150)Percent with middle

education.170*** �.067 .292***

(.031) (.173) (.056)Percent with post mid-

dle education.312*** .635***

(.059) (.221)Household size �.090*** �.059*** �.067*** �.093*** �.121***

(.003) (.019) (.006) (.017) (.007)Number of adults in

household.023*** .064 .005 .085** .012

(.008) (.040) (.014) (.041) (.020)Constant 14.031*** 13.638*** 14.044*** 13.833*** 14.103***

(.043) (.209) (.091) (.159) (.094)Observations 5,813 287 2,093 360 1,598Communities 145 37 104 49 106

Note. Robust standard errors in parentheses.a At least one adult with this level of education and none with more within household.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 33: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

48

TABLE 13HOUSEHOLD PARTICIPATION’S EFFECT ON HOUSEHOLD WELFARE : IV COMMUNITY FIXED EFFECTS

Level of Completed Education in Householda

Variables Complete Sample Post-Middleb Middle Primary None

Attended literacy pro-gram

.570 1.214** .519 .085

(.376) (.608) (.334) (.252)Owns land .018 .040 �.043 �.032

(.031) (.063) (.175) (.050)Nonfarming household �.003 �.015 .209 �.156*

(.050) (.113) (.246) (.080)Average age:

30–34 �.059* �.091 �.062 �.127*(.035) (.072) (.189) (.077)

35–39 �.042 �.250** �.208 �.041(.046) (.108) (.211) (.072)

40–49 �.036 �.159** �.254 .033(.034) (.078) (.174) (.067)

50–59 .020 �.165 �.270 .058(.050) (.149) (.188) (.071)

60–99 .052 �.254 �.303 .037(.045) (.184) (.283) (.063)

Percent with primaryeducation

.158*** .123 .210

(.053) (.158) (.297)Percent with middle

education.338*** .795***

(.070) (.269)Percent with post-mid-

dle education.479***

(.095)Household size �.123*** �.110*** �.127*** �.151***

(.007) (.018) (.032) (.010)Number of adults in

household.016 �.009 .075 .035

(.020) (.035) (.086) (.029)Constant 14.127*** 13.771*** 14.041*** 14.312***

(.064) (.239) (.351) (.095)Observations 2,820 891 164 938Communities 145 102 44 106

Note. Robust standard errors in parentheses. A household is considered to have attended if at leastone of its members has participated in an adult literacy program.a At least one adult with this level of education and none with more within household.b No convergence was obtained, so results have been excluded.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 34: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 49

TABLE 14EFFECT OF INDIVIDUAL PARTICIPATION ON INDIVIDUAL LITERACY FOR NO-EDUCATION

SAMPLE: LINEAR PROBABILITY MODEL

Variables OLS FE FE-IVa

Attended literacy program .114*** .111*** .364***(.019) (.019) (.112)

Female �.050*** �.052*** �.034*(.009) (.009) (.017)

Age:30–44 .014 .020** .018

(.009) (.010) (.020)45–59 .016 .010 .002

(.011) (.010) (.020)60–99 .011 .005 .018

(.011) (.010) (.019)Owns land .027** .011 .011

(.011) (.016) (.019)Nonfarming household .046** .005 .051

(.018) (.023) (.034)Household size .002 .001 .002

(.002) (.002) (.003)Number of adults in household �.012** �.009* .002

(.005) (.006) (.010)Constant .046*** .057*** �.042

(.015) (.017) (.047)Observations 2,202 2,202 1,598Communities 162 106R2 .07

Note. Robust standard errors in parentheses. Sample consists of all individuals living in house-holds where none of the adult members have completed primary schooling. All models arelinear probability models. Hence, the parameter for “Attended literacy program” shows thepredicted increases in the probability of being able to read from having participated in anadult literacy program.a First stage estimated using community fixed-effects logit shown in table 10. Predicted prob-ability of participation used as instrument for participation in linear second stage communityfixed-effects model.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

this group has the highest rate of participation and accounts for over one-thirdof the total sample of households. The section first presents an analysis of theeffects of participation on literacy and numeracy skills, followed by an inves-tigation of how participation affects business ownership, credit, and sale ofgoods. Finally, the section contains a discussion whether other governmentalinterventions might be responsible for the positive effect found.

A. Participation and Literacy and NumeracyTables 14 and 15 present the determinants of literacy and numeracy for in-dividuals living in no-education households.29 All models are linear probability29 The corresponding results for the other groups are available upon request.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 35: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

50 economic development and cultural change

TABLE 15EFFECT OF INDIVIDUAL PARTICIPATION ON INDIVIDUAL NUMERACY FOR NO-EDUCATION SAMPLE:

LINEAR PROBABILITY MODEL

Variables OLS FE FE-IVa

Attended literacy program .097*** .074*** .074(.020) (.018) (.115)

Female �.063*** �.075*** �.083***(.011) (.011) (.018)

Age:30–44 �.020 �.014 �.008

(.019) (.015) (.020)45–59 �.055*** �.043*** �.051**

(.019) (.016) (.020)60–99 �.087*** �.088*** �.093***

(.019) (.017) (.019)Owns land .041*** �.035* �.069***

(.014) (.019) (.020)Nonfarming household .068*** .032 .013

(.025) (.028) (.035)Household size �.004 �.003 �.004

(.003) (.003) (.003)Number of adults in household �.015** �.000 .008

(.006) (.007) (.010)Constant .166*** .161*** .165***

(.025) (.023) (.048)Observations 2,202 2,202 1,598Communities 162 106R2 .06

Note. Robust standard errors in parentheses. Sample consists of all individuals living in house-holds where none of the adult members have completed primary schooling. All models arelinear probability models. Hence, the parameter for “Attended literacy program” shows thepredicted increases in the probability of being able to do written calculations from havingparticipated in an adult literacy program.a First stage estimated using community fixed-effects logit shown in table 10. Predicted prob-ability of participation used as instrument for participation in linear second stage communityfixed effects model.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

models, where the first column shows the results using standard OLS, thesecond column includes community fixed effects, and the final column thecommunity fixed-effect instrumental variable results using the method de-scribed above. There is relatively little difference between the OLS and thefixed-effects results for both literacy and numeracy, with the effect slightlysmaller for the fixed-effects estimates. According to the fixed-effects estimates,participation in an adult literacy program increases the probability of beingable to read by just over 10 percentage points and the probability of beingable to do written calculations by around 7 percentage points. All of the OLSand fixed-effects estimates of the effect of participation are strongly statisticallysignificant.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 36: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 51

While there is little difference between the OLS and the community fixed-effects results, using instrumental variables leads to substantial changes in theestimated effects of participation on literacy. The effect of participation onliteracy becomes three times as large, at 36 percentage points, and remainsstatistically significant. For numeracy the point estimate remains the same asthe fixed-effects one, but the standard error increases dramatically, making theeffect far from statistically significant. One interpretation of the increase inthe estimated effect on literacy is that unobserved ability is negatively correlatedwith participation. This could happen if the value of time for a higher abilityindividual was substantially higher than for a lower ability individual or ifthe return to participation was expected to be relatively low. While thisexplanation might be seen as supported by the slight increase in welfare forthe individual level estimation when instrumenting, it does not explain whythere is no change in the estimated effect of participation on numeracy.30

Furthermore, given the descriptive statistics for literacy and participation inadult literacy programs, it is hard to imagine that there is sufficient bias fromselection on unobservables and measurement error to explain the estimatedeffect.

A substantial part of the problem is likely to be the instruments. Differentsets of instruments make the first-stage instruments perform substantiallyworse but do not change the estimated effect of program participation onliteracy much.31 This reinforces the point made above that using communityfixed effects to control for endogenous program placement makes it moredifficult to also take account of self-selection into the programs and thisproblem is exacerbated when analyzing subsamples, which limits the numberof available instruments and observations even further.

Despite these issues, it seems clear that participating in an adult literacyprogram does have positive, albeit small, effects on literacy and numeracy.Given the large effects of participation on welfare, it seems unlikely that theseeffects are explained solely by the increase in literacy and/or numeracy.32 There-fore, we now turn to the effects of participating in an adult literacy programson other aspects that might increase household welfare.

30 Note also that the household level estimations show a decrease in the estimated effect ofparticipation on consumption.31 This is only the case for literacy. For welfare and numeracy there are large changes, and theestimates become too large to be believable.32 At least that would seem to require that the returns to literacy and numeracy would be a factorof 10 larger than returns to schooling.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 37: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

52

TAB

LE1

6B

USI

NE

SSO

WN

ER

SHIP

,C

RE

DIT

,A

ND

LITE

RA

CY

PR

OG

RA

MPA

RTI

CIP

ATI

ON

:C

OM

MU

NIT

YFI

XE

D-E

FFE

CTS

LIN

EA

RP

RO

BA

BIL

ITY

MO

DE

L

Loan

Sour

ceLo

anP

urp

ose

Vari

able

sO

wns

aB

usin

ess

Cur

rent

Loan

Form

ala

No

tFa

mily

bIn

vest

men

tB

usin

ess

Ag

ricu

ltur

e

Att

end

edlit

erac

yp

rog

ram

.030

.023

.020

.015

.045

.006

.045

(.031

)(.0

29)

(.032

)(.0

63)

(.065

)(.0

57)

(.050

)O

wns

land

�.0

79**

�.0

19�

.040

�.1

14.1

32.0

08.1

24**

(.036

)(.0

33)

(.040

)(.0

78)

(.080

)(.0

71)

(.061

)N

onf

arm

ing

hous

eho

ld.1

02**

�.0

95**

�.0

51�

.024

.001

.163

�.1

62*

(.050

)(.0

46)

(.056

)(.1

10)

(.114

)(.1

00)

(.087

)A

vera

ge

age:

30–3

4.0

17�

.054

.010

�.1

01�

.133

�.0

15�

.115

(.051

)(.0

47)

(.058

)(.1

13)

(.116

)(.1

03)

(.089

)35

–39

�.0

24�

.046

.028

�.1

69�

.046

.015

�.0

57(.0

50)

(.046

)(.0

55)

(.107

)(.1

11)

(.097

)(.0

84)

40–4

9�

.034

�.0

11.0

40�

.115

�.0

96�

.029

�.0

62(.0

48)

(.044

)(.0

52)

(.101

)(.1

04)

(.092

)(.0

79)

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 38: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

53

50–5

9�

.068

�.0

30.0

07�

.074

�.2

50**

�.1

57*

�.0

99(.0

48)

(.044

)(.0

51)

(.100

)(.1

03)

(.091

)(.0

79)

60–9

9�

.111

**�

.203

***

�.0

03�

.245

**�

.068

�.1

03.0

41(.0

46)

(.042

)(.0

54)

(.106

)(.1

09)

(.096

)(.0

83)

Ho

useh

old

size

.015

**.0

06�

.028

***

�.0

01.0

18.0

16.0

00(.0

07)

(.007

)(.0

08)

(.015

)(.0

16)

(.014

)(.0

12)

Num

ber

of

adul

ts�

.003

�.0

16.0

78**

*.0

23.0

68.0

23.0

43(.0

21)

(.019

)(.0

26)

(.050

)(.0

51)

(.045

)(.0

39)

Ob

serv

atio

ns1,

272

1,27

234

234

234

234

234

2C

om

mun

itie

s16

216

211

511

511

511

511

5

No

te.

Ro

bus

tst

and

ard

erro

rsin

par

enth

eses

.Sa

mp

leco

nsis

tso

fho

useh

old

sw

here

noad

ult

mem

ber

has

com

ple

ted

pri

mar

ysc

hoo

l.A

hous

eho

ldis

cons

ider

edto

have

atte

nded

ifat

leas

to

neo

fit

sm

emb

ers

has

par

tici

pat

edin

anad

ult

liter

acy

pro

gra

m.

All

mo

del

sar

elin

ear

pro

bab

ility

mo

del

sw

ith

com

mun

ity

fixed

effe

cts.

Hen

ce,

the

par

amet

erfo

r“A

tten

ded

liter

acy

pro

gra

m”

sho

ws

the

pre

dic

ted

incr

ease

sin

the

pro

bab

ility

fro

mha

ving

par

tici

pat

edin

anad

ult

liter

acy

pro

gra

m.

aA

tle

ast

one

loan

ob

tain

edfr

om

afo

rmal

inst

itut

ion,

whi

chin

clud

est

ate

ban

k,p

riva

teb

ank,

coo

per

ativ

e,g

ove

rnm

ent

agen

cy,

NG

O,

bus

ines

sfir

m,

and

oth

erfo

rmal

.b

At

leas

to

nelo

ano

bta

ined

fro

ma

sour

ceo

ther

than

rela

tive

,fr

iend

,o

rne

ighb

or.

*St

atis

tica

llysi

gni

fican

tat

10%

.**

Stat

isti

cally

sig

nific

ant

at5%

.**

*St

atis

tica

llysi

gni

fican

tat

1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 39: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

54 economic development and cultural change

B. Business Ownership, Credit, and Sale of Agricultural Produce

Since the Ghanaian adult literacy programs have a substantial focus on areasother than literacy and numeracy, it is worthwhile attempting to furtherunderstand how participation in a program affects the households. Incomedata are available in the GLSS 4 but are of substantially worse quality thanthe expenditure data which precludes using that to understand how literacyprograms work. It is, however, possible to examine whether participation affectsfactors that can impact income such as whether the household owns a business,its access to and use of credit, and whether it sells various goods.

Table 16 presents community fixed-effects linear probability model resultsfor the effects of participation on business ownership and various aspects ofcredit market participation using the household level data.33 The outcomesexamined are whether the household owns a business (not including agricul-ture), whether it currently has a loan, if it has had a loan any time over the12 months before the survey, what the source of the loan was, and for whatpurpose the loan was taken.

A household is considered to own a business if any member of the householdhas worked for himself, other than on a farm or raising animals. While thereis a 3% increase in the probability of someone in the household being self-employed from having participated in an adult literacy program, this effectis not statistically significant. The effect of participation on whether the house-hold owes money to someone is also positive but again statistically insignificantand small.

For the loan source and loan purpose we use only those households thatcurrently have a loan outstanding, since there unfortunately is no informationabout loans that have been paid off (unless the household has a current loan).This substantially reduces the sample size; there are now only 342 householdsfrom 115 communities. For loan sources two possibilities are considered. Thefirst is formal sources, which include state bank, private bank, cooperative,government agency, NGO, business firm, or other formal. The increased like-lihood of lending from a formal source after participating in a program issmall at 2 percentage points and not statistically significant. The same is thecase for having at least one loan that is not from relatives, friends, or neighbors.

The last three columns show whether participating in an adult literacyprogram has an effect on the likelihood of obtaining a loan for investmentpurposes, which consists of two possible uses, business expansion and agri-

33 Community fixed-effects logit results are available on request but show marginal effects thatare qualitatively similarly to the results in table 16.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 40: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 55

culture.34 Neither of these effects is statistically significant. The effect onbusiness expansion is essentially zero, while there is a much larger effect onthe likelihood of using a loan for agricultural investment. The latter increasesby 4.5 percentage points, and while it is not statistically significant, this ishardly surprising given the low number of degrees of freedom after controllingfor other characteristics and for community fixed effects.

To examine whether participation leads households to be more engaged inmarket activities, table 17 shows community fixed-effects linear probabilitymodel results for the effects of participation on whether the household sellsvarious agricultural goods. The five groups of goods are livestock; grains andcash crops; roots, fruits, and vegetables; other produce; and processed produce.35

For the first four, the household is asked if it has sold any of the goods in theparticular category during the 12 months prior to the survey, while for thelast category the question is only asked for the 2 weeks prior to the survey.

The largest and most statistically significant impact from participation isthe increased likelihood that a household engaged in selling of goods in thelivestock category (which also includes output from fishing). Households thathave participated in an adult literacy program are 7 percentage points morelikely to have sold some type of livestock over the 12 months prior to thesurvey. There are also statistically significant effects of participation on theprobability that a household has sold roots, fruits, and vegetables, and otherproduce over the last year, but for those two categories the effect is smallerat around a 4 percentage point increase. Interestingly, there is no statisticallysignificant effect of participation on whether the household sells grain andcash crops, although the effect is still positive.

The effect on sale of processed produce is small and far from statisticallysignificant. This is surprising given that one would expect these goods to behigher value goods and to be goods that require more investment and humancapital, which can be gained through participation in an adult literacy program.A possible explanation for the small, insignificant effect is that only 13% ofall households report selling these goods, which may be a result of the shortperiod covered for these goods as mentioned above. It is worth noting thatthe share of households who sell these goods is almost twice as high amongparticipants as among nonparticipants (18% vs. 10%, respectively).

The conclusion is that there is not one factor that explains the large, positiveeffect of participation on household consumption. It does appear that house-holds that have participated are more likely to be engaged in market activities,

34 Agriculture investments include land, equipment, and inputs.35 A more detailed description of each category is given in the notes to the table.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 41: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

56

TAB

LE1

7SA

LEO

FA

GR

ICU

LTU

RA

LG

OO

DS:

CO

MM

UN

ITY

FIX

ED

-EFF

EC

TSLI

NE

AR

PR

OB

AB

ILIT

YM

OD

EL

Vari

able

sLi

vest

ock

aG

rain

san

dC

ash

Cro

psb

Ro

ots

,Fr

uits

,an

dVe

get

able

scO

ther

Pro

duc

edP

roce

ssed

Pro

duc

ee

Att

end

edlit

erac

yp

rog

ram

.073

**.0

30.0

39*

.043

*.0

13(.0

29)

(.029

)(.0

23)

(.025

)(.0

23)

Ow

nsla

nd.0

34.1

35**

*.1

16**

*.0

15�

.018

(.033

)(.0

33)

(.027

)(.0

29)

(.026

)N

onf

arm

ing

hous

eho

ld�

.083

*�

.488

***

�.1

14**

*�

.112

***

�.0

76**

(.046

)(.0

45)

(.037

)(.0

40)

(.036

)A

vera

ge

age:

30–3

4�

.042

�.0

23.0

18.0

38�

.002

(.047

)(.0

47)

(.038

)(.0

41)

(.037

)35

–39

�.0

07.0

16.0

47.0

21.0

07(.0

46)

(.046

)(.0

37)

(.040

)(.0

37)

40–4

9�

.007

.006

.005

.013

.008

(.044

)(.0

44)

(.035

)(.0

38)

(.035

)50

–59

.014

�.0

60�

.010

.010

.043

(.044

)(.0

44)

(.035

)(.0

38)

(.035

)60

–99

�.0

22�

.059

.005

�.0

57�

.007

(.042

)(.0

42)

(.034

)(.0

37)

(.033

)

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 42: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

57

Ho

useh

old

size

.010

�.0

06�

.005

�.0

04.0

13**

(.007

)(.0

07)

(.005

)(.0

06)

(.005

)N

umb

ero

fad

ults

.049

**.0

71**

*.0

31**

.019

.005

(.019

)(.0

19)

(.016

)(.0

17)

(.015

)O

bse

rvat

ions

1,27

21,

272

1,27

21,

272

1,27

2C

om

mun

itie

s16

216

216

216

216

2

No

te.

Ro

bus

tst

and

ard

erro

rsin

par

enth

eses

.Sa

mp

leco

nsis

tso

fho

useh

old

sw

here

noad

ult

mem

ber

has

com

ple

ted

pri

mar

ysc

hoo

l.A

hous

eho

ldis

cons

ider

edto

have

atte

nded

ifat

leas

to

neo

fit

sm

emb

ers

has

par

tici

pat

edin

anad

ult

liter

acy

pro

gra

m.

All

mo

del

sar

elin

ear

pro

bab

ility

mo

del

sw

ith

com

mun

ity

fixed

effe

cts.

Hen

ce,

the

par

amet

erfo

r“A

tten

ded

liter

acy

pro

gra

m”

sho

ws

the

pre

dic

ted

incr

ease

sin

the

pro

bab

ility

fro

mha

ving

par

tici

pat

edin

anad

ult

liter

acy

pro

gra

m.F

or

allg

roup

so

fg

oo

ds,

exce

pt

for

“Pro

cess

edP

rod

uce,

”th

ere

leva

ntp

erio

dco

vere

dis

12m

ont

hs.

For

“Pro

cess

edP

rod

uce”

onl

yth

ela

st2

wee

ksb

efo

reth

esu

rvey

are

cove

red

.a

Dra

ught

anim

als

(do

nkey

,ho

rse,

bul

lock

),ca

ttle

(incl

udin

gco

ws)

,sh

eep

,g

oat

s,p

igs,

rab

bit

s,ch

icke

n,o

ther

po

ultr

y,o

ther

lives

tock

,fis

h,cr

ab,

and

oth

er.

bC

oco

a,co

conu

t,co

ffee

,co

tto

n,ke

nef,

rub

ber

,w

oo

d,

gro

und

nut/

pea

nut,

mai

ze,

rice

,so

rghu

m/m

illet

/gui

nea

corn

,b

eans

/pea

s,to

bac

co,

sug

arca

ne,

and

oth

ercr

op

s.c

Avo

cad

op

ear,

ban

anas

,co

lanu

t,m

ang

o,

oil

pal

m,

ora

nges

,p

awp

aw,

pla

ntai

ns,

oth

erfr

uit

tree

s,ca

ssav

a,co

coya

m,

oni

on,

swee

tp

ota

toes

/po

tato

es,y

am,g

ard

eneg

gs/

egg

pla

nt,

leaf

yve

get

able

s,o

kro

,p

epp

er,

tom

ato

es,

oth

erve

get

able

s,an

dp

inea

pp

le.

dH

unti

ng(in

clud

ing

snai

lco

llect

ion)

,ho

ney,

dri

nks

(pal

mw

ine/

akp

etes

hie,

pit

o,

mm

edaa

,et

c.),

frui

t/b

erri

es,

milk

and

oth

erd

airy

pro

duc

ts,

egg

s,hi

des

,w

oo

l,sk

in,

and

mus

hro

om

s.e

Mai

zeflo

ur,

flour

fro

mo

ther

gra

ins,

husk

ed/p

olis

hed

rice

,ho

me-

bre

wed

dri

nk,

cass

ava

flour

,sh

elle

dg

roun

dnu

ts,

pro

cess

edfis

h,g

ari,

shea

but

ter,

oth

ernu

ts.

*St

atis

tica

llysi

gni

fican

tat

10%

.**

Stat

isti

cally

sig

nific

ant

at5%

.**

*St

atis

tica

llysi

gni

fican

tat

1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 43: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

58 economic development and cultural change

owning a business and especially getting a loan to invest in agriculture, al-though none of these effects are statistically significant. There are, however,significant effects on the likelihood of a household selling at least some typesof agricultural good, but even here the increases are not large. Hence, as wediscuss below, understanding how the adult literacy programs affect householdwelfare is an important area for future research.

C. Other Government InterventionsClearly, given the available data it is difficult to disentangle how participatingin an adult literacy program increases welfare. Part of the effect is likely throughskills acquisition, although the effect appears to be too small to explain theentire increase in consumption. Furthermore, while there are positive effectsof participation on factors that might be related to business and agriculture,these effects are small and only statistically significant for the sale of sometypes of goods. This makes it more important to examine whether othergovernment programs, interventions, or other community characteristics mightbe responsible for the observed effect. This could, for example, be the case ifother programs or interventions were rolled out at the same time and to thesame areas as the adult literacy programs. Table 18 therefore shows correlationcoefficients for a variety of government programs, interventions, and com-munity characteristics for the 166 communities in the base data set.

In addition to adult literacy program, s the community characteristics in-clude whether there is a drivable road close to the community and whetherthere is a bank, daily market, hospital, health post, family planning clinic, oragricultural extension center in the community. While there are many com-munity characteristics that are statistically significantly correlated, the onlycharacteristic that adult literacy program presence is statistically significantlycorrelated with is whether there is a drivable road near the community andeven that correlation is small at 0.14. The correlation between the availabilityof an adult literacy program and and whether there is a bank in the communityis negative and around 0.1, while the correlation with the presence of anagricultural extension program is around the same size but positive. Neithercorrelation is statistically significant. Hence, it is unlikely that other govern-ment programs, interventions, or community characteristics are responsible forthe positive effect of participation in adult literacy programs.

VII. ConclusionThis paper examines the effect of adult literacy program participation onhousehold consumption in Ghana. We use community fixed effects to dealwith possible endogenous program placement combined with instrumental

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 44: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

59

TAB

LE1

8C

OR

RE

LATI

ON

BE

TWE

EN

CO

MM

UN

ITY

CH

AR

AC

TER

ISTI

CS

Vari

able

sD

riva

ble

Ro

adB

ank

Dai

lyM

arke

tH

osp

ital

Hea

lth

Po

stFa

mily

Pla

nnin

gA

gri

cult

ural

Ext

ensi

on

Dri

vab

lero

ad1.

000

Ban

k.1

49*

1.00

0(.0

56)

Dai

lym

arke

t.1

76**

.458

***

1.00

0(.0

23)

(.000

)H

osp

ital

.053

.356

***

.301

***

1.00

0(.4

97)

(.000

)(.0

00)

Hea

lthp

ost

.230

***

.439

***

.347

***

.101

1.00

0(.0

03)

(.000

)(.0

00)

(.195

)Fa

mily

pla

nnin

g.1

39*

.433

***

.289

***

.171

**.6

61**

*1.

000

(.076

)(.0

00)

(.000

)(.0

28)

(.000

)A

gri

cult

ural

exte

nsio

n.1

47*

.381

***

.363

***

.163

**.5

58**

*.5

11**

*1.

000

(.059

)(.0

00)

(.000

)(.0

37)

(.000

)(.0

00)

Lite

racy

pro

gra

m.1

39*

�.1

08�

.075

�.0

38�

.016

.021

.105

(.074

)(.1

68)

(.335

)(.6

24)

(.839

)(.7

91)

(.181

)

No

te.

Leve

lof

sig

nific

ance

inp

aren

thes

es.

Co

rrel

atio

nco

effic

ient

san

dle

velo

fsi

gni

fican

cear

eb

ased

on

the

166

com

mun

itie

sin

the

bas

ed

ata

set.

*St

atis

tica

llysi

gni

fican

tat

10%

.**

Stat

isti

cally

sig

nific

ant

at5%

.**

*St

atis

tica

llysi

gni

fican

tat

1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 45: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

60 economic development and cultural change

variables to account for self-selection into program participation. We findsubstantial, positive, and statistically significant effects on consumption forhouseholds where no adult members have completed any formal education.For other households the effects are smaller and not statistically significantand become smaller the more educated the household is. Based on the results,our preferred estimated effect of participation in an adult literacy program isan approximately 10% increase in per adult equivalent consumption for no-education households. For comparision, Canagarajah and Thomas (1997) findthat the return to education in Ghana is between 4% and 6% per year. Giventhe length of the program and the effort required by participants, combinedwith the focus on skills directly applicable to improving participants’ welfare,these returns are approximately equal.

Another way of looking at the effectiveness of the program is to performa simple cost-benefit analysis.36 The average cost per participant to the gov-ernment of Ghana is U.S. dollars (USD) 33.40 (Oxenham 2002). The oppor-tunity cost of participation is based on 21 months of participation at 10 hoursper week with time valued at the average wage by sex of individuals who havenot completed any formal schooling and who reported wage labor.37 Benefitsare the net present value of the increase in consumption from when the par-ticipants “graduate” from the program until and including age 55 using ourpreferred estimate of 10% increase per year. Hence for, say, a 30-year-oldparticipant, we use the net present value of increased consumption from age32 up to and including age 55. To simplify and keep the results conservative,we assume only one adult equivalent in the household; obviously the impactswill be greater for larger families since the results above are per adult equivalent,not only for the participant. We use two levels of preparticipation expenditure:the lower poverty line of 700,000 cedis per year and the upper poverty lineof 900,000 cedis per year.38 The lowest internal rate of return based on theseassumptions is 9.2% for a 40-year-old male participant who began at the lowerpoverty line, while the highest is close to 20%. These social rates of returnare very respectable, especially considering that they do not include potential

36 The primary aim here is to illustrate the scale of benefits relative to the total cost of an adultparticipating in a literacy program, not to conduct a complete cost-benefit analysis, which wouldrequire data on many other outcomes besides earnings. For example, we do not include possibleeffects on health, fertility, or the education of children. See, e.g., Blunch (2006, 2011) for a discussionof literacy’s effect on these outcomes.37 As described above, most programs run three times a week with each session lasting up to 3hours; adding 1 hour for transport leads to 10 hours a week.38 Table A3 shows both the net benefits with a discount rate of 5% and the internal rates of returns(the discount rate at which the net present value of the benefits equal the combined costs).

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 46: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 61

benefits from participation other than the impact on future productivity andthe conservative nature of the assumptions.

The adult literacy programs in Ghana consist of several modules, and hencedetermining what makes the program effective is an important topic for futureresearch. While participants are significantly more likely to be able to readand do written calculations than nonparticipants, the literacy and numeracyrates for individuals without formal education after participation are not evenclose to those of individuals who have completed primary school (who haverelatively low rates because of the poor quality of schooling). Furthermore,there are no statistically significant effects on variables such as whether thehousehold owns a business or obtains a loan, although participants are sig-nificantly more likely to sell some types of agricultural goods. It would thereforebe tempting to conclude that the modules on income-generating activities aremore important for the positive effect of participation than the actual literacyand numeracy skills. While this is certainly a possibility, it is also possiblethat some or all of the positive effect comes from some other attribute of theprogram, such as an improvement in the participants’ network, or even fromthe combination of attributes. Furthermore, the fact that the effects for house-holds with more educated members are smaller than for the no-educationhouseholds, but still positive, seems to point toward a more standard increasein human capital rather than the income-generating modules, especially sinceone might expect these to be complementary to household human capital. Itis, however, clear that it is not other government programs and interventionsthat drive the effect; there is no significant correlation between having anadult literacy program in the community and all of the other communitycharacteristics except for whether there is a drivable road, and that correlationis small.

Part of the explanation for the difficulty in determining the pathwaysthrough which program participation affects welfare is that, in addition toliteracy and numeracy, there are up to 28 themes and these can vary by area,but we only have information on participation, with no information on theprovider or the timing of the program. A promising direction for future researchwould be to run randomized experiments with varying combinations of mod-ules to determine which ones are most important. It is already clear, however,that, despite the disappointing results from previous literacy programs, ap-propriately designed adult literacy programs can be an important componentin the effort to improve the livelihood of those who have not participated inthe formal education system.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 47: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

62

AppendixOLS Estimates and Cost-Benefit Analysis

TABLE A1INDIVIDUAL PARTICIPATION’S EFFECT ON HOUSEHOLD WELFARE: ORDINARY LEAST SQUARES

Level of Completed Education in Householda

VariablesComplete

Sample Post-Middle Middle Primary None

Attended literacy program .112*** �.105 .046 .002 .202***(.029) (.096) (.045) (.059) (.043)

Female .056*** .022 �.014 .044* .123***(.010) (.040) (.012) (.026) (.025)

Age:30–44 �.033 .121* �.024 �.043 �.067*

(.024) (.064) (.029) (.052) (.037)45–59 �.017 .036 �.074*** �.060 .013

(.016) (.053) (.021) (.057) (.031)60–99 .070*** .005 �.104** �.092 .167***

(.019) (.062) (.045) (.063) (.036)Primary education .038*** �.015 �.029** �.005

(.007) (.019) (.011) (.019)Middle education .062*** .005 �.025

(.007) (.020) (.016)Post-middle education .080*** .008

(.009) (.033)Owns land .116*** .104 .148*** .239*** .031

(.020) (.086) (.036) (.057) (.044)Nonfarming household .299*** .427*** .307*** .172** .183***

(.035) (.122) (.057) (.064) (.050)Percent with primary education .271*** .474** .121 .410***

(.043) (.210) (.097) (.119)Percent with middle education .436*** .611*** .530***

(.030) (.137) (.063)Percent with post-middle education .589*** 1.114***

(.061) (.184)Household size �.085*** �.033* �.076*** �.097*** �.108***

(.010) (.018) (.011) (.020) (.012)Number of adults in household �.006 .017 .022 .022 �.058*

(.014) (.042) (.021) (.052) (.034)Constant 13.895*** 13.210*** 13.844*** 13.884*** 14.072***

(.050) (.169) (.073) (.153) (.087)Observations 6,713 883 2,849 779 2,202R2 .24 .27 .22 .27 .28

Note. Robust standard errors in parentheses.a At least one adult with this level of education and none with more within household.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 48: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

63

TABLE A2HOUSEHOLD PARTICIPATION’S EFFECT ON HOUSEHOLD WELFARE: ORDINARY LEAST SQUARES

Level of Completed Education in Householda

VariablesComplete

Sample Post-Middle Middle Primary None

Attended literacy program .094*** �.195* .031 �.017 .202***(.028) (.101) (.047) (.074) (.045)

Owns land .140*** .139* .157*** .232*** .093**(.020) (.074) (.035) (.056) (.039)

Nonfarming household .203*** .330*** .217*** .130** .138**(.034) (.095) (.051) (.058) (.053)

Average age:30–34 �.079* �.097 �.099** �.058 �.093

(.044) (.116) (.039) (.115) (.067)35–39 �.077 .036 �.119** �.123 �.026

(.051) (.121) (.056) (.104) (.073)40–49 �.123*** �.014 �.149*** �.279*** �.025

(.035) (.126) (.041) (.094) (.076)50–59 �.049 .123 �.078 �.210* �.028

(.034) (.126) (.074) (.102) (.060)60–99 .096** �.137 �.174* �.034 .116**

(.038) (.215) (.098) (.119) (.054)Percent with primary education .289*** .374* .135 .325*

(.041) (.205) (.092) (.162)Percent with middle education .456*** .600*** .509***

(.030) (.143) (.066)Percent with post-middle education .619*** 1.053***

(.053) (.188)Household size �.115*** �.046** �.099*** �.123*** �.143***

(.007) (.019) (.010) (.018) (.011)Number of adults in household �.006 .008 .018 .028 �.065**

(.011) (.037) (.018) (.057) (.025)Constant 14.153*** 13.467*** 14.030*** 14.182*** 14.327***

(.049) (.188) (.068) (.197) (.079)Observations 3,219 323 1,235 389 1,272R2 .30 .29 .29 .33 .32

Note. Robust standard errors in parentheses. A household is considered to have attended if at leastone of its members has participated in an adult literacy program.a At least one adult with this level of education and none with more within household.* Statistically significant at 10%.** Statistically significant at 5%.*** Statistically significant at 1%.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 49: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

64 economic development and cultural change

TABLE A3COST-BENEFIT ANALYSIS FOR ADULT LITERACY PROGRAMS

Females Males

Cost:Direct cost 33.40 33.40Opportunity cost 42.97 56.48Total cost 76.37 89.88

Benefits:Lower poverty line:

30-year-old 160.98 160.9840-year-old 115.48 115.48

Upper poverty line:30-year-old 206.98 206.9840-year-old 148.48 148.48

Net benefits:Lower poverty line:

30-year-old 84.61 71.1040-year-old 39.11 25.60

Upper poverty line:30-year-old 130.61 117.1040-year-old 72.11 58.60

Internal rate of returnLower poverty line:

30-year-old 14.7 12.240-year-old 12.3 9.2

Upper poverty line:30-year-old 19.4 16.340-year-old 17.6 14.1

Note. All figures are in USD with an exchange rate of 6,000 cedis to 1 USD.

ReferencesAbadzi, Helen. 2003. “Adult Literacy: A Review of Implementation Experience.”

OED Working Paper 29387, World Bank Operations Evaluation Department,Washington, DC.

Angeles, Gustavo, David K. Guilkey, and Thomas A. Mroz. 1998. “Purposive Pro-gram Placement and the Estimation of Family Planning Program Effects in Tan-zania.” Journal of the American Statistical Association 93, no. 443:884–99.

Appleton, Simon. 2002. “The Rich Are Like Us, Only Richer: Poverty Functionsor Consumption Functions.” Journal of African Economies 10, no. 4:433–69.

Blunch, Niels-Hugo. 2006. “Skills, Schooling, and Household Well-Being inGhana.” PhD diss., George Washington University.

———. 2011. “Human Capital, Religion, and Contraceptive Use in Ghana.” Pho-tocopy, Department of Economics, Washington and Lee University.

Blunch, Niels-Hugo, and Claus C. Portner. 2005. “Literacy, Skills, and Welfare:Effects of Participation in Adult Literacy Programs.” Department of EconomicsWorking Paper UWEC-2005–23, University of Washington, Seattle.

Boissiere, M., J. B. Knight, and R. H. Sabot. 1985. “Earnings, Schooling, Ability,and Cognitive Skills.” American Economic Review 75, no. 510:16–30.

Canagarajah, Sudharshan, and Claus C. Portner. 2003. “Evolution of Poverty and

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 50: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

Blunch and Portner 65

Welfare in Ghana in the 1990s: Achievements and Challenges.” Africa RegionWorking Paper Series 61, World Bank, Washington, DC.

Canagarajah, Sudharshan, and Saji Thomas. 1997. “Ghana’s Labor Market (1987–92).” Policy Research Working Paper Series 1752, World Bank, Washington, DC.

Carr-Hill, Roy A., Aikael N. Kweka, Mary Rusimbi, and Rustica Chengelele. 1991.“The Functioning and Effects of the Tanzanian Literacy Programme.” IIEP Re-search Report 93, International Institute of Educational Planning, UNESCO,Paris.

Carron, Gabriel, Kilemi Mwiria, and Gabriel Righa. 1989. “The Functioning andEffects of the Kenya Literacy Programme.” IIEP Research Report 76, InternationalInstitute of Educational Planning, UNESCO, Paris.

Central Bureau of Statistics (Ghana). 1984. “1984 Population Census of Ghana:Preliminary Report.” Technical Report, Central Bureau of Statistics, Accra.

Chiswick, Barry R., Yew Liang Lee, and Paul W. Miller. 2003. “Schooling, Literacy,Numeracy, and Labour Market Success.” Economic Record 79, no. 245:165–81.

Dustmann, Christian, and Francesca Fabbri. 2003. “Language Proficiency and LabourMarket Performance of Immigrants in the UK.” Economic Journal 113, no. 489:695–717.

Glewwe, Paul. 1999. “The Impact of Cognitive Skills on Wages.” In The Economicsof School Quality Investments in Developing Countries: An Empirical Study of Ghana,ed. Paul Glewwe. London: Macmillan.

Green, David A., and W. Craig Riddell. 2003. “Literacy and Earnings: An Inves-tigation of the Interaction of Cognitive and Unobserved Skills in Earnings Gen-eration.” Labour Economics 10, no. 2:165–84.

Ishikawa, Mamoru, and Daniel Ryan. 2002. “Schooling, Basic Skills, and EconomicOutcomes.” Economics of Education Review 21, no. 3:231–43.

Jolliffe, Dean. 1999. “The Impact of Cognitive Skills on Income from Farming.” InThe Economics of School Quality Investments in Developing Countries: An Empirical Studyof Ghana, ed. Paul Glewwe. London: Macmillan.

Mitra, Aparna. 2002. “Mathematics Skill and Male-Female Wages.” Journal of Socio-Economics 31, no. 5:443–56.

Moll, Peter G. 1998. “Primary Schooling, Cognitive Skills, and Wages in SouthAfrica.” Economica 65, no. 258:263–84.

Murnane, Richard J., John B. Willett, and Frank Levy. 1995. “The Growing Im-portance of Cognitive Skills in Wage Determination.” Review of Economics andStatistics 77, no. 2:251–66.

Ortega, Daniel, and Francisco Rodriguez. 2008. “Freed from Illiteracy? A CloserLook at Venezuela’s Mision Robinson Literacy Campaign.” Economic Developmentand Cultural Change 57, no. 1:1–30.

Oxenham, John. 2002. “Review of World Bank Supported Projects in Adult BasicEducation and Literacy, 1977–2002: Comparison of Costs.” Photocopy, WorldBank, Washington, DC.

Pitt, Mark M., Mark R. Rosenzweig, and Donna M. Gibbons. 1995. “The Deter-minants and Consequences of the Placement of Government Programs in Indo-nesia.” In Public Spending and the Poor: Theory and Evidence, ed. Dominique van de

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions

Page 51: Literacy, Skills, and Welfare: Effects of Participation in Adult Literacy Programs

66 economic development and cultural change

Walle and Kimberly Nead. Baltimore and London: Johns Hopkins UniversityPress for the World Bank.

Ribar, David C. 2003. “What Do Social Scientists Know about the Benefits ofMarriage? A Review of Quantitative Methodologies.” Photocopy, George Wash-ington University.

Rosenbaum, Paul R., and Donald B. Rubin. 1983. “The Central Role of the Pro-pensity Score in Observational Studies for Causal Effects.” Biometrika 70, no. 1:41–55.

Rosenzweig, Mark R., and Kenneth I. Wolpin. 1986. “Evaluating the Effects ofOptimally Distributed Public Programs: Child Health and Family Planning In-terventions.” American Economic Review 76, no. 3:470–82.

Tyler, John H., Richard J. Murname, and John B. Willett. 2000. “Do the CognitiveSkills of School Dropouts Matter in the Labor Market?” Journal of Human Resources35, no. 4:748–54.

Vijverberg, Wim P. M. 1999. “The Impact of Schooling and Cognitive Skills onIncome from Non-Farm Self-Employment.” In The Economics of School Quality In-vestments in Developing Countries: An Empirical Study of Ghana, ed. Paul Glewwe.London: Macmillan.

Wooldridge, Jeffrey M. 2002. Econometric Analysis of Cross Section and Panel Data.Cambridge, MA: MIT Press.

This content downloaded from 91.229.248.35 on Wed, 21 May 2014 14:32:09 PMAll use subject to JSTOR Terms and Conditions