research article improving child literacy in africa ...listen/pdfs/2010 itid korsah...

1

G. Ayorkor [email protected] CandidateRobotics InstituteCarnegie Mellon UniversityNewell-Simon Hall 21045000 Forbes AvenuePittsburgh, PA 15213USA(412)268-7147

Jack [email protected] ProfessorSchool of Computer ScienceCarnegie Mellon UniversityNewell Simon Hall 42135000 Forbes AvenuePittsburgh, PA 15213USA(412)268-1330

M. Bernardine [email protected] Research ProfessorRobotics InstituteCarnegie Mellon UniversityNewell-Simon Hall 21045000 Forbes AvenuePittsburgh, PA 15213USA(412)268-7147

Improving Child Literacy in Africa KORSAH, MOSTOW, DIAS, SWEET, BELOUSOV, DIAS, GONG

Research Article

Improving Child Literacy in Africa:Experiments with an AutomatedReading Tutor*Abstract

This paper describes Project Kané, a research endeavor aimed at exploring therole that technology can play in improving child literacy in developing commu-nities. An initial pilot study and a subsequent four-month-long controlled ªeldstudy in Ghana investigated the viability and effectiveness of an automatedreading tutor in helping urban children enhance their reading skills in English.In addition to quantitative data suggesting that automated tutoring can beuseful for some children in this setting, these studies and an additional prelim-inary pilot study in Zambia yielded useful qualitative observations regardingthe feasibility of applying technology solutions to the challenge of enhancingchild literacy in developing communities. This paper presents the ªndings, ob-servations, and lessons learned from the ªeld studies.

1. IntroductionLiteracy is a key part of the global development agenda. The UnitedNations recognizes literacy as a human right, noting that basic education,of which literacy is the key learning tool, was recognized as a human rightmore than 50 years ago in the Universal Declaration of Human Rights.

As a contribution to the discourse on applying information and com-munication technologies (ICTs) to address development challenges, wedescribe a study exploring the potential role of computing technology,speciªcally an automated reading tutor, in improving English literacyamong Ghanaian and Zambian children who attend school in English buthave low reading achievement levels.

Low functional literacy among individuals who have completed primaryschool is not an uncommon problem in developing communities. Forexample, UNESCO (2005) reports that, in 2000, more than one in threeadults with a ªfth-grade education in Chad and Niger reported that theycould not read. In other cases, individuals may ªnish primary school read-ing below the expected level. In a representative sample of Ghanaian pub-lic schools, reading achievement levels measured by the government-administered Criterion Referenced Test in 2000 indicated that fewer than10% of the children in grade six were able to read with grade level mas-tery (Lipson & Wixson, 2004).

*The research reported here was supported in part by the discretionary gifts to the TechBridgeWorld research group atCarnegie Mellon University, by the Qatar Foundation for Education, Science, and Community Development, by the Na-tional Science Foundation under ITR/IERI Grant No. REC-0326153, by the Institute of Education Sciences, U.S. Depart-ment of Education through Grant R305A080628 to Carnegie Mellon University, and by the Heinz Endowments. Theopinions expressed are those of the authors and do not necessarily represent the views of any of our sponsors.

© 2010 USC Annenberg School for Communication & Journalism. Published under Creative Commons Attribution-Non Commercial-Share Alike 3.0 Unported

license. All rights not granted thereunder to the public are reserved to the publisher and may not be exercised without its express written permission.

Volume 6, Number 2, Summer 2010, 1–19

2 Information Technologies & International Development

IMPROVING CHILD LITERACY IN AFRICA

Tracy Morrison [email protected] StudentDepartment of StatisticsCarnegie Mellon UniversityBaker Hall 132M5000 Forbes AvenuePittsburgh, PA 15213USA(412)268-1889

Sarah M. [email protected] ManagerRobotics InstituteCarnegie Mellon UniversityNewell-Simon Hall 21045000 Forbes AvenuePittsburgh, PA 15213USA(412)268-7147

M. Frederick [email protected] EngineerRobotics InstituteCarnegie Mellon UniversityNewell-Simon Hall 21045000 Forbes AvenuePittsburgh, PA 15213USA(412)268-7147

Haijun [email protected] FellowComputer Science DepartmentCarnegie Mellon University5000 Forbes AvenuePittsburgh, PA 15213USA(412)268-4447

Several factors contribute to this problem. For the average child from arural or low-income urban background in Africa, reading is not part ofdaily family life, and sometimes, parents are not themselves literate. InGhana and Zambia speciªcally, most children speak one of a number oflocal languages at home, but they attend school classes taught in English,the ofªcial language for both countries. Typically, under-resourced schoolswith overcrowded classrooms offer few opportunities for individual atten-tion while students are developing reading skills, and many areas have theadditional challenge of inadequately trained teachers (Clegg, Ogange, &Rodseth, 2003).

Project Kané, described in this paper, is a proof-of-concept studydesigned to investigate whether an automated computer-based readingtutor that provides guided reading practice can signiªcantly improve thereading proªciency of children in a developing community, even if theyhave no prior familiarity with computers. We focus our study on childrenin Accra, Ghana, and Mongu, Zambia. As illustrated in Figure 1, this workis part of a larger research endeavor exploring the role of technology inimproving English literacy in Africa.

We began by employing the Reading Tutor in a preliminary three-week-long pilot study in Accra, Ghana. The pilot study was used toexplore technical and operational feasibility, and to motivate partnershipsand funding for a longer term study. We followed up with a four-month-

Figure 1. Project outline.

long controlled study in Accra, in collaboration withthe Ghana-India Koª Annan Centre for Excellence inICT (AITI-KACE) and with input from Associates forChange, an educational research ªrm in Accra. Thecontrolled study aimed at quantitatively measuringthe educational effectiveness of the tutor when itwas employed with an out-of-school usage modelby children in Accra. Observations and lessons fromthese experiences fed into another pilot study inMongu, Zambia, this time employing an in-schoolusage model, with the plan of following up with anadditional controlled study in the future.

The main contributions of this work are a quanti-tative analysis of the effectiveness of the ReadingTutor, as well as a discussion of the lessons learned,concerning both study implementation and the via-bility of automated tutoring for this context.

Section 2 describes the skills that are involved inreading and introduces the automated tutor weused in this project. Section 3 discusses prior workon the use of technology to support literacy indeveloping communities. Section 4 outlines theoverarching study objectives. Sections 5, 6, and 7describe the pilot ªeld study in Ghana, the four-month-long controlled study in Ghana, and the pilotstudy in Zambia, respectively. Section 8 distills practi-cal lessons learned in the implementation of theseªeld studies, and Section 9 concludes with a discus-sion of future work.

2. BackgroundAs a cognitive proªciency, reading involves severalcomponent skills, such as phonemic awareness,decoding, ºuency, vocabulary, and comprehension(National Institute of Child Health and HumanDevelopment [NICHD], 2000). Phonemic awarenessis the ability to perceive individual sounds or pho-nemes in words. Building on this awareness, a childlearns the alphabetic principle that spelling generallymaps systematically to pronunciation. The child alsolearns speciªc letter-sound correspondences and theability to correctly pronounce written words (decod-ing). Fluency is the ability to read text accurately,quickly, and expressively, and it is an importantfoundation for comprehension. A rich vocabulary isessential for comprehension and effective communi-cation. Children develop these skills through a vari-ety of experiences, including skilled instruction.Research has shown that regular guided oral reading

plays an important role in developing reading skills,particularly ºuency and comprehension (NICHD,2000).

Such guided oral reading may happen in smallgroups in a classroom setting, or with parents athome. Technology may also assist in this process:Scientiªc Learning’s Reading Assistant, evaluated inAdams (2006), and Project LISTEN’s Reading Tutor(Mostow & Aist, 2001) are examples of computer-based tools that use automated speech recognitionto provide a guided reading experience for the user.

The project described in this paper used ProjectLISTEN’s Reading Tutor, which has demonstratedsuccess in improving the reading abilities of children,both those whose ªrst language is English (Mostow& Aist, 2001; Mostow et al., 2008) and those learn-ing English as a second language (ESL) in the UnitedStates (Poulsen, Hastings, & Albritton, 2007) andCanada (Reeder, Shapiro, Early, Kendrick, &Wakeªeld, 2005).

The Reading Tutor displays stories on a screenand “listens” to a child read aloud. By using speechrecognition to analyze the child’s reading, the Read-ing Tutor is able to give graphical and spoken feed-back. It gives help when it detects a long pause, aseverely misread word, or a skipped word, and alsowhen the reader clicks for help. The tutor mayspeak the whole word out loud or decompose theword into syllables or phonemes and speak out eachpart while highlighting it. It may also give a “rhymeswith” hint, or read the sentence by playing a ºuenthuman narration to model expressive reading. TheReading Tutor includes a wide variety of stories atdifferent reading levels. It takes turns with the childin selecting a story to read. It monitors the child’sreading progress and selects stories at an appropri-ate level for the child. The Reading Tutor also pro-vides activities to let children author new contentthemselves. For readers at early stages of develop-ment, the Reading Tutor includes word-buildingexercises to develop knowledge of spelling-to-soundcorrespondences. Videos of the Reading Tutor in usemay be found at Project LISTEN’s website(www.cs.cmu.edu/�listen).

3. Relation to Prior WorkThere is a great deal of discussion, such as that byWagner, Day, and Sun (2004), about the potentialfor ICTs to make an impact on education in the

Volume 6, Number 2, Summer 2010 3

KORSAH, MOSTOW, DIAS, SWEET, BELOUSOV, DIAS, GONG

developing world. There is now a recognized needfor a greater focus on rigorous evaluations of theimpact of technology on educational outcomes indeveloping communities as a ªrst step in enablingpolicy makers to make effective decisions (Wagner,2005). There is some evidence, so far, of a positiveimpact on student learning from ICT tools targetedat speciªc subject areas (Kozma, 2005). For exam-ple, projects in India (Banerjee, Cole, Duºo, & Lin-den, 2007) and South Africa (Louw, Muller, &Tredoux, 2008) have shown improvements in mathe-matics performance as a result of using math-relatedsoftware. Evaluations such as these, still quite rare,show the potential impact of ICTs on education indeveloping communities.

In the domain of literacy and language learning,one of the most signiªcant projects related to ourwork is the MILLEE project (Kam et al., 2008), whichdevelops mobile phone games to assist with Englishas a Second Language (ESL) learning. User-centereddesign and evaluation of these games have beenfocused on rural communities in India. The mostrecent quantitative evaluation of this tool showed asigniªcant positive impact on ESL learning by chil-dren in an after-school program (Kam, Kumar, Jain,Mathur & Canny, 2009).

The work reported in the current paper contrib-utes to the discourse by studying a different part ofthe research space than has been previously studied:the use of technology to support English literacyamong urban children in English-speaking Africa. Itis important to evaluate the effectiveness of differ-ent kinds of tools, as the capabilities of differentplatforms vary greatly. Mobile phones are muchmore pervasive in the developing world than per-sonal computers. However, for reading and writinginstruction, the personal computer has some keyadvantages over the mobile phone: It has an easierinput device and a large screen that can display fullparagraphs of connected text. In the work reportedin this paper, we take advantage of the fact thatcomputers are more pervasive and viable in urbanareas than they are in rural areas. We evaluate theuse of a computer-based automated reading tutor inthis context, exploring both in-school and out-of-school usage scenarios.

4. Study ObjectivesThe primary objective of our studies was to evaluatethe feasibility and effectiveness of the automated

reading tutor in the setting of interest. Twosecondary objectives were to explore practical usagescenarios for the tutor, and to learn lessons relatedto conducting such studies in this environment. Assuch, we comment on all three aspects in this paper.

In Ghana, we focused on an out-of-school usagemodel for the practical reason that most of theschools with which we worked did not have com-puter labs. This is typical of most under-resourcedpublic schools in this setting, and therefore, it wasimportant to explore what would be possible with-out signiªcant additional infrastructure invest-ment in the schools. In Zambia, we focused on anin-school usage model, since the schools weworked with had computers donated by a non-governmental organization.

5. Ghana: Pilot Study

A. GoalThe 2005 Ghana pilot study aimed to evaluate thepracticality of a technological approach to guidedreading practice in Accra, and to investigate the fea-sibility of conducting a longer-term controlled study.

Speciªcally, we wished to answer the followingquestions:

1. Partners and logistics: How feasible is it toengage partners and arrange the logisticsfor such a study?

2. Learning to use the Reading Tutor: Howquickly do children with no prior computerexperience learn to operate the Reading Tu-tor, and what instruction do they need in or-der to do so?

3. Speech: Does the speech recognition soft-ware perform acceptably with Ghanaian ac-cents, and can the students understand thenarrated speech?

4. Tutor content: Do the children ªnd the read-ing material in the tutor engaging?

5. Usage sessions: What is an effective lengthfor a tutoring session?

6. Reading measures: Which test instrumentscan be used easily and effectively to assessreading proªciency in the Ghanaian setting?

B. Participants and MethodologyWe chose to focus on the needs of children fromlow-income families attending public school becausethey have a high risk of low achievement in reading.



In consultation with Associates for Change, we tar-geted children in grades two through four, since thisis a key period for developing reading skills after theinitial adjustment to the primary school environ-ment. We restricted our work to an urban environ-ment where computers are more readily available inthe community, but where signiªcant literacy chal-lenges still exist. As the school did not have its owncomputer lab, we employed an out-of-school usagemodel in which practice with the Reading Tutor wassupplemental to regular school activities.

The pilot study involved qualitative observationsof children as they used the Reading Tutor. Thestudy was conducted by the ªrst author, who is anative of Ghana, assisted by a local volunteer. Twogroups of children participated in this study. Onegroup comprised 12 children in grades two throughfour from an under-resourced public school. Theyused the Reading Tutor at an Internet café near theirschool for 20–30 minutes each day over a three-week period. The other group comprised six childrenfrom a mixed low- and middle-income neighbor-hood. They used the tutor on laptops in the homeof the researcher, for 20–30 minutes each day, threedays a week, over the same period.

C. Results and ObservationsThe school and Internet café agreed without hesita-tion to participate in the project, and logistics werearranged very quickly. Teachers assisted by providingclass lists from which the researchers selected partic-ipants at random, but because the study took placeoff-site at an Internet café, the teachers were notdirectly involved in the pilot study. The parents werehappy to have their children participate in a projectthat could potentially be of beneªt to them, andthat did not have a fee for participation. The chil-dren were excited at the opportunity to use a com-puter and honored to be selected to participate. TheInternet café donated time on four desktop comput-ers at a time when the Reading Tutor could be used.We engaged the remaining children in other activi-ties while they waited their turn. This worked forthe pilot project, but it would not scale for a largerstudy.

Although one child had previously played a com-puter game, none of the other children had used acomputer before. The children were given an initial10–15 minute hands-on lesson introducing them tothe computer and showing them how to use themouse and keyboard. We then launched the Read-

ing Tutor and had them go through the built-inautomated tutorials on its operation. We found thatthe children had trouble understanding these auto-mated tutorials due to the unfamiliar narrationaccent, so we explained the tutorials one-on-one tothe children during their ªrst two sessions, instead.We later re-narrated these tutorials in a Ghanaianaccent. The children appeared to understand allother prompts given by the Reading Tutor. By theirsecond or third session, most of the children wereable to operate the Reading Tutor without help.Sometimes, the tutor would time-out if a child wasidle for a while, and the project staff would need toshow him or her how to log in again. On otheroccasions, a child would need to be encouraged tospend less time browsing the list of stories and moretime reading. The project staff did not give anyreading instruction or help to the children. Thespeech recognition capability appeared to work ade-quately with the children’s accents. Also, based onthe graphical feedback given by the tutor, the chil-dren quickly learned to recognize those occasionswhen it did not “hear” them correctly, and torepeat themselves when necessary. Although quanti-tatively analyzing the accuracy of the speech recog-nition was outside the scope of the project, we didnot see any qualitative evidence of speech recogni-tion problems serious enough to interfere with theperformance of the Reading Tutor.

Overall, the children were very enthusiastic aboutusing the Reading Tutor. Most of them appeared toenjoy the word-building exercises and the stories.However, we noticed that a couple of children, whowere older than the norm (11 or 12 years old) buthad a kindergarten or ªrst grade reading level, werenot as engaged by the content of the simple storiesavailable for their reading level. Furthermore, webelieve it would be good to incorporate more localcontent, such as Ghanaian folk tales, into the tutor.Unfortunately, time constraints limited our ability todo so. We noticed that the better readers could usethe tutor for more than half an hour at a time with-out getting bored, whereas the more challengedreaders would tire after about 20 minutes, but stilllooked forward to their next turn. We thus decidedto limit the length of usage sessions in the con-trolled study to 20–30 minutes.

Although the pilot study did not aim to quantita-tively measure reading progress, we chose an oralreading ºuency test (Deno, 1985) and the Test ofWritten Spelling (TWS) (Larsen, Hammill, & Moats,



1999) to assess reading ability, and we experi-mented with both of these test instruments duringthe pilot study, in preparation for the controlledstudy. These are both timed, hand-scored testswhich we chose to use because they have been fre-quently used in previous studies measuring theeffectiveness of the Reading Tutor, and they arepsychometrically reliable, fast, and easy to adminis-ter. The ºuency test is a reading exercise where thechild is given a grade-appropriate story to read. It isscored as the number of words read correctly in oneminute. The TWS is a dictation exercise where thechild writes words read by the tester in order ofincreasing difªculty. It is scored as the number ofwords spelled correctly. The students did not havetrouble with the format of either test. We modiªedthe ºuency test passages to use Ghanaian namesand referred to the tests as “exercises” to preventstudent anxiety about being tested.

6. Ghana: Controlled Study

A. GoalFollowing on the success of the pilot study, the goalof the controlled study in Ghana was to quantita-tively measure the efªcacy of the Reading Tutor inhelping children improve their reading skills. Spe-ciªcally, we wished to determine:

1. Does regular use of the Reading Tutor im-prove oral reading ºuency and spelling?

2. Do treatment effects depend on other fac-tors, such as school/socioeconomic back-ground, gender, or grade level?

Another goal of the controlled study was to learnabout the operational feasibility of an out-of-schoolusage model, taking into consideration installationand maintenance of the software, training of staffresponsible for the day-to-day running of the proj-ect, transportation of the children between theschool and the project site, and other logistics. Thecontrolled study was deployed by project staff atAITI-KACE, with remote training and support by ourteam in Pittsburgh.

B. Study Design/MethodologyThe controlled study involved 89 children from threeschools. The participating schools, recruited by Asso-ciates for Change, effectively catered to childrenfrom three different socio-economic backgrounds,respectively: S1, a private school in a middle-incomecommunity; S2, a public school in a low-incomecommunity; and S3, an informal educational pro-gram for highly disadvantaged children who havenever attended formal school. The study involvedchildren in grades two through four of S1 and S2,and in the “Intermediate” and “Advanced” levels(roughly corresponding to grades two and three,respectively) of S3.

The children were split randomly across school,grade, and gender boundaries into two groups,Tutor-1st and Control-1st, as shown in Table 1. Weused a two-treatment crossover study design, illus-trated in Table 2. For the ªrst half of the study, theTutor-1st group used the Reading Tutor while theControl-1st group had no additional reading inter-vention, and for the second half, the roles wereswitched: the Control-1st group used the ReadingTutor while the Tutor-1st group had no additionalreading intervention. Each half of the study lastednine weeks (two months), during which time therewere daily usage sessions of approximately half anhour per child, although attendance and usage var-ied considerably. The statistical reason for a cross-over design was to control for between-studentvariance by using each child as his/her own control,that is, by answering the question of whether a par-ticular child gained more while using the ReadingTutor than while not using the Reading Tutor.Another advantage of a crossover design is thatfewer participants are needed than would beneeded for a parallel group trial with the same sta-tistical power (Senn, 1993). The crossover designalso had the advantage of equity, in that all partici-pating children had the opportunity to use the



Figure 2. A child in Accra reading a story using theReading Tutor.

Reading Tutor at some point. The major potentialdisadvantage of crossover designs is that carryovertreatment effects from the ªrst half may affect thesecond half. The standard structure of the two-treatment crossover, in which each half of the par-ticipants experiences a different treatment order, canhelp identify such an effect during analysis.

A battery of three ºuency tests and one TWSwere administered to all the children at the begin-ning of the study (pre-testing), between the twohalves (mid-way testing), and at the end (post-testing). The testing was conducted at the children’sschools by AITI-KACE project staff. Three ºuencytests were used to better estimate reading ability.The passages used for these tests respectively corre-spond roughly to ªrst, second, and third grade read-ing levels, since a test close to the child’s currentreading level should give a more sensitive measureof progress. We found that the scores on the threepassages were highly correlated, so, for analysis pur-

poses, these scores were combined into a singlemean ºuency score.

Of the participating schools, S1 had a computerlab, while S2 and S3 did not. To ensure similar studyconditions for all participants, children from all threeschools used the Reading Tutor in a computer lab atAITI-KACE. The computers used for the project had2.4GHz Pentium IV processors, 256MB of RAM, and16GB hard drives (shared with other projects). Thesuggested (not necessarily minimum) speciªcationsfor running the Reading Tutor are a 700 Mhz pro-cessor, 256 MB of RAM, a 40 GB hard drive, fullduplex audio card, and Windows 2000 or WindowsXP. Because of the smaller hard-disk space on themachines used, speech logs from the tutor had tobe backed-up regularly.

C. Deviation from Study DesignIn a normal crossover study, experimental conditionsare held constant over the course of the study. How-ever, due to logistical challenges that delayed the



Table 1. Study participants by school, grade, and gender.

Number of Children

Tutor-1st Control-1st

School Grade Female Male Female Male

S1(29 children)

Grade 2 3 2 2 3

Grade 3 5 2 1 3

Grade 4 1 3 3 1

S2(30 children)

Grade 2 2 3 3 2

Grade 3 3 2 1 4

Grade 4 3 2 3 2

S3(30 children)

Intermediate 3 4 3 5

Advanced 3 5 5 2

Table 2. Cross-over study design.

Tutor-1st Group Control-1st Group

Pre-testing of all children

First half of cross-over(1st nine weeks)

Reading Tutor Control: no special intervention

Mid-way testing of all children

Second half of cross-over(2nd nine weeks)

Control: no special intervention Reading tutor

Post-testing of all children

start of the project, the ªrst half of this study tookplace while the students were attending school,whereas the second half overlapped with the schoolvacation. Because of the different conditions inplace during the two halves of the study, we ana-lyzed the results as two different experiments, ratherthan as a single crossover study. The ªrst experimentmeasured the effect of the Reading Tutor while the

children attended school (and thus attended Englishclass as usual). The second experiment measured theeffect of the tutor while the children did not attendschool (and so did not attend English class). Thismodiªed study design is illustrated in Table 3.

D. Results and AnalysisFigure 3 illustrates the pre-test scores of children atthe three schools, representing their reading

proªciency going into the study. Itis clear that the S1 children hadmuch higher levels of readingachievement than the S2 children,who, in turn, had higher levelsthan the S3 children. Pre-testingfor our study occurred about twomonths before the end of theschool year. To provide some con-text for these scores, Table 4compares the ºuency pre-testscores with end-of-year normsfrom schools in the United States(Hasbrouck & Tindal, 2006). Bythis standard, the average readingproªciency of the S1 studentsappears to be at or above theU.S. average, whereas those ofthe S2 and S3 children are lower.Norms from schools in Ghana arenot available for comparison.

In analyzing each experiment,we focus on gains in ºuency andspelling test scores. For experi-ment 1, the gain is the differencebetween pre- and mid-test scores;for experiment 2, the gain is thedifference between the mid- andpost-test scores. A positive gain



Table 3. Modiªed study design.

Tutor-1st Group Control-1st Group

Pre-testing of all children

Experiment 1(1st nine weeks)

School 1 Reading Tutor Control: school only

Mid-way testing of all children

Experiment 2(2nd nine weeks)

Control: No school No school 1 Reading tutor

Post-testing of all children

Figure 3. Pre-test scores by school and grade.

indicates improvement in the child’s reading pro-ªciency.

In each experiment, we use a standard statisticalt-test to compare the mean gains of the treatmentand the control group. This test yields a p-value indi-cating how signiªcant the difference is between themeans of the two groups. For this analysis, we con-sider p-values of less than 0.016 to indicate statisti-cal signiªcance. This value is smaller (and thus moreconservative) than the commonly used threshold of0.05, because multiple comparisons (due to the dif-ferent schools) require an adjustment of signiªcancelevels (Miller, 1981). We also compute a measure ofeffect size—that is, the magnitude of the treatmenteffect. The measure we use for effect size is Cohen’sd: the difference between group mean gains dividedby the within-group pooled standard deviation(Cohen, 1988). The effect size is computed for eachschool because the treatment effect depends on theschool. An effect size of 0.2 is generally consideredsmall, 0.5 is considered medium, and 0.8 is consid-ered large (Cohen, 1992).

Experiment 1 results by school: Table 5 showsthe results and t-test analysis for experiment 1,

when the children were attending school. Theseresults are illustrated graphically in Figure 4. Thereading proªciency of the S2 children who used theReading Tutor (the “treatment” group) improvedsigniªcantly more than those who did not (the“control” group), as evidenced by larger gains inboth ºuency and TWS. The S3 treatment groupsigniªcantly out-gained the control group in ºuency,but not in spelling, although there was a positivetrend. Finally, there was no signiªcant difference ingains between the S1 treatment and control groups,for either test. Omitted from this analysis are ªve S1children who were not present for the midway test-ing. Due to the small numbers, we did not breakdown the data to examine gains for each gradelevel within each school.

Experiment 2 results by school: Table 6 showsthe gains (from mid-test to post-test) for experiment2, when the children were not attending school. Theresults are illustrated graphically in Figure 5. Interest-ingly, there were negative ºuency gains for the S2children over the vacation, and these were sig-niªcantly more dramatic for the control group thanfor the treatment group. We discuss this observation



Table 4. Fluency pre-test scores compared with U.S. norms.

Mean pre-test ºuency (standard deviation)Year-end U.S. Norm*(50th percentile)Grade / Level S1 S2 S3

2 / “Intermediate” 82.4 (44.8) 32.3 (25.9) 11.1 (8.5) 89

3 / “Advanced” 123.3 (31.4) 45.8 (37.4) 20.4 (20.8) 107

4 132.7 (32.7) 59.8 (41.5) — 123

* Data from Hasbrouck & Tindal (2006)

Table 5. Comparison of treatment and control gains for experiment 1.

Treatment(Tutor-1st group,N 41)

Control(Control-1st group,N 42)

t-test comparison oftreatment & controlgains

Test School Mean Gain (SD) p-value Effect size

Fluency(# words readcorrectly per min)

S1 8.6 (20.5) 21.5 (11.7) �0.775

S2 61.4 (22.9) 23.4 (21.0) 0.0001 �1.731

S3 13.4 (9.3) 5.1 (4.4) 0.0054 �1.144

TWS(# words spelledcorrectly)

S1 �1.9 (4.9) �0.2 (2.6) �0.457

S2 3.9 (2.7) 0.3 (3.4) 0.0039 �1.153

S3 2.8 (4.0) 0.6 (1.7) �

in the Discussion subsection that follows. For the S2children, the difference in TWS test score gainsbetween the treatment and the control group wasnot statistically signiªcant. Finally, there was nosigniªcant difference between the treatment and

the control group in the othertwo schools, either for meanºuency or for TWS. Omitted fromthe analysis are two S1, two S2,and 10 S3 children who wereabsent for testing.

E. Discussion

Effectiveness of the ReadingTutor: The results provide evi-dence that, during the schoolterm, the S2 students (and to alesser extent, the S3 students)who used the tutor gained con-siderably more than those whodid not use the tutor. Thus, theReading Tutor was helpful for theS2 children. This is a positive out-come, since the S2 group mostclosely represents our target pop-ulation of children from low-income families attending publicschool. The proªciency of the S1children did not appear to beinºuenced by reading practicewith the tutor. A possible expla-nation for this is that the S1 chil-dren might not have had muchroom to beneªt from the tutor,since they were ºuent readersgoing into the study. Observa-

tions of the three groups of students using the tutorshowed that the S2 students appeared to be themost engaged overall, reading so enthusiasticallythat the room literally buzzed with noise. Some ofthe S1 students appeared to get bored easily—the



Figure 4. Comparison of treatment and control gains for Experiment 1,highlighting statistically signiªcant differences.

Table 6. Comparison of treatment and control gains for experiment 2.

Treatment(Control-1st group, N 33)

Control(Tutor-1st group, N 37)

t-test comparison oftreatment & controlgains

Test School Mean Gain (SD) p-value Effect size


S1 �3.8 (22.6) 0.222

S2 9.1 (17.1) 39.8 (16.8) 0.0001 1.816

S3 �0.665


S1 �0.197

S2 �0.068

S3 �0.3 (2.2) 0.474

researcher once noticed a child simultaneously read-ing a story book while using the tutor. As a group,the S3 students, who had a much lower proªciencyin speaking and understanding English, in additionto having the lowest reading pre-test scores, hadmore trouble using the tutor than the other twogroups: They required more frequent clariªcation oftutor instructions by the project staff (e.g., when achild appeared to be sitting idle and not respondingto tutor prompts).

Loss in proªciency over vacation period: Of thestatistically signiªcant results highlighted in Figure 5,the negative gains of the S2 children in experiment2 stand out. A possible explanation for this might befound in studies in the United States which havedocumented that reading achievement test scoresfor children from low-income families deterioratesigniªcantly over the summer vacation (a phenome-

non referred to as the “summerreading setback”), whereas thosefor children from middle-incomefamilies remain steady or increaseslightly (Allington & McGill-Franzen, 2003). This trend hasbeen attributed to the discrep-ancy in the reading opportunitiesand materials available to thesetwo groups over the vacation pe-riod. In light of these studies, it isinteresting to note that, for theS2 children, those who used theReading Tutor did not deterioratein reading ability as much asthose who did not use it. How-ever, the sessions with the Read-ing Tutor (totaling, on average,11.5 hours of reading per child),without English class at school,were not enough to prevent neg-ative gains.

Effect of pre-test scores: Acomplicating factor in the dataanalysis is the observed unequalaverage pre-test scores of the twogroups, Tutor-1st and Control-1st,despite supposed random assign-ment of children to the twogroups. Table 7 shows that thisdisparity is statistically signiªcant

for the S2 children. Higher pre-test scores can bethe cause of greater gains (Stanovich, 2000), and sothis difference in pre-test scores raises the questionof whether the observed reading improvements aredue to the use of the Reading Tutor, or to thehigher pre-test scores. To investigate this, we com-puted the correlation between pre-test scores andgains, and we found no signiªcant correlation. Thissuggests that the greater gains of the Tutor-1stgroup in experiment 1 are, indeed, attributable tothe Reading Tutor, rather than to their higher pre-test scores.

Tutor usage: Another signiªcant differencebetween the two experiments is that tutor usage inexperiment 2 was much lower than in experiment 1,due to a higher level of absenteeism from the studyduring the school vacation. As Table 8 shows, S3students’ usage dropped to under a third of what it



Figure 5. Comparison of treatment and control gains for Experiment 2,highlighting statistically signiªcant differences.

was in the ªrst half, although, on the days they didattend, the length of their Reading Tutor sessionswere similar to what they were in the ªrst half.Nineteen children who were absent for testing,either at the beginning or end of each experiment,are not included in this summary.

The reduced usage for S3 students may partlyexplain why there is no statistically signiªcant effectof the Reading Tutor for S3 children in experiment2. We found that there was a slight positive correla-tion between amount of usage and gains in the ªrstexperiment, but not in the second, and that usagewas not a strong (statistically signiªcant) predicatorof gains in either experiment. This does not neces-sarily indicate that no relationship exists betweenamount of usage and test score gains. It is likely thatwe lacked the statistical power to detect this rela-tionship due to the small sample size and the fact

that school is a confounding variable. This is anexample of where it would have been advantageousif we had been able to analyze the study as a cross-over study as originally planned, with each childserving as his or her own control.

7. Zambia: Pilot StudyThe work in Zambia complements the prior studiesin Ghana by investigating an in-school usage modeland testing the tutor in a different English-speakingAfrican country. ProjectEDUCATE, a nonproªt orga-nization, introduced us to two under-resourced pub-lic schools in Mongu, Zambia. The schools hadreceived donated computers and were enthusiasticto test the Reading Tutor. However, the Mongu Dis-trict Education Board Secretary’s ofªce requestedthat we select only one school for the pilot test.



Table 7. Comparison of pre-test scores for the Tutor-1st group and the Control-1st groups.

Pre-test score

Tutor-1st (N 41) Control-1st (N 42)

Test School Mean (SD) Mean (SD) p-value


S1 122.0 (45.9) 99.2 (33.9) 0.137

S2 65.8 (33.9) 26.2 (26.9) 0.0015

S3 18.6 (13.2) 12.1 (19.1) 0.308


S1 24.7 (9.7) 24.0 (9.6) 0.850

S2 11.1 (7.0) 4.3 (3.7) 0.003

S3 2.1 (2.2) 1.9 (3.1) 0.840

Table 8. Tutor usage in each experiment.

SchoolExperiment 1Mean (SD)

Experiment 2Mean (SD)

Number of days of tutoruse per student

S1 29.6 (12.0) 22.2 (9.4)

S2 36.9 (2.4) 24.0 (11.7)

S3 30.7 (2.8) 12.6 (3.9)

Total time spent readingwith tutor per student(hours)

S1 12.4 (5.4) 11.9 (5.2)

S2 18.6 (3.1) 11.5 (6.7)

S3 17.3 (2.7) 6.8 (2.5)

Average daily timespent reading with tutorper student (minutes)

S1 22.9 (7.1) 32.6 (5.3)

S2 30.1 (3.7) 26.6 (7.0)

S3 33.9 (4.6) 31.9 (5.3)

A. GoalThe goals of the Zambian pilot study were toanswer the following questions:

1. Computing infrastructure: What is the stateof the school’s computer lab, how is it cur-rently used, and can it feasibly support ses-sions with a computer-based reading tutor?

2. Training Teachers: How long does it take totrain teachers to guide students in the useof the tutor?

3. Children’s response: How do the children re-spond to the Reading Tutor?

4. Test instruments and Reading Tutor content:What test instruments and Reading Tutorcontent are appropriate for the Zambian set-ting?

5. Feasibility of long-term study: Would a re-mote partnership between the school andour research group be a feasible model for alonger-term controlled study?

B. Methodology and ImplementationThe selection of the school for the pilot test wasdone after meeting with the headmasters andteachers at both schools and assessing the state ofthe computer labs and the potential for a successfulstudy. Subsequently, the pilot study consisted ofconducting interviews with teachers at the selectedschool, providing training for the teachers, and mak-ing qualitative observations of teachers and studentsas they used the Reading Tutor. We also exploredwith the teachers possible details of a longer con-trolled study.

C. Results and ObservationsBoth schools had labs with roughly 20 266 MHzPentium II computers with 64–128MB of RAM and10GB hard drives. Some teachers from each schoolhad taken a computer skills training course whenthe computers were donated, but had not had theopportunity for additional training or guided prac-tice. The schools did not have Internet access, andthe computers were used primarily by teachers fortyping exam questions. Given the large class sizes(ranging from 50 to more than 100 students) andthe strict timetable, the teachers could not feasiblyuse the lab to teach all students about computers.Some computers at both schools were not function-ing due to broken keyboards, mice, and power

strips; some computers were not being protectedfrom dirt during the dusty winter months; poweroutages were a daily occurrence; and for oneschool, even maintaining electricity for the computerlab was a challenge due to limited ªnancialresources. For the pilot study, we selected the schoolwith better maintained equipment and a higher like-lihood of maintaining communication by telephoneand e-mail.

We trained three teachers to use the ReadingTutor. The hour-long session covered the educationalfeatures of the tutor, as well as administrative tasks,such as managing users. After basic instruction andsome time to practice on their own, the teacherswere able to guide students in using the tutor. Theywould often give feedback to the children as a com-plement to the tutor when the students haddifªculty reading stories. For example, they wouldinstruct the children to click for help when theyneeded it, or they would sometimes correct a mis-pronounced word that the tutor did not detect. Wethink this involvement of teachers in the early stagesas the child gets used to the tutor could be animportant part of longer-term use of the ReadingTutor in a school setting. The teachers especiallyappreciated the ability to track their students’ per-formance using the tutor.

We observed a group of 11 children in gradestwo through four as they read one or two storiesfrom the tutor. The students were selected by theteachers and had varying levels of reading ability,English comprehension, and speaking ºuency. All ofthe students except one were completely new tocomputers. We introduced them, as a group, to thebasic components of the computer, and then wegave them the opportunity to demonstrate use ofthe mouse and keyboard to each other. We pro-vided verbal instruction to students individually asthey began reading stories from the Reading Tutor.Some grade-two students seemed to have difªcultyunderstanding English, since English is introduced asa language of instruction only in grade two, as apart of the Zambian Ministry of Education’s initiativeto encourage basic literacy by teaching in a familiarlanguage in the ªrst year of school. The studentsenjoyed using the Reading Tutor; when we returnedto the school on another day to meet with theteachers, the group of students was using the word-building exercises in the Reading Tutor on their owntime.



Building on the studies in Ghana, which focusedon assessing reading with the same standard mea-sures that have been used with previous studieswith the Reading Tutor, we were able, in Zambia, todiscuss with the teachers additional reading assess-ment options for a longer-term study. We learnedthat, through the Zambia Primary ReadingProgramme (PRP), the students’ literacy levels aremeasured by reading standardized storybooks aloudto their teachers. Each color-coded book is associ-ated with a given reading level. Once the teacherdetermines that the child can successfully read at agiven level, he or she moves the child up to a higherlevel reading group with a different set of books.The teachers were interested in engaging the stu-dents with the computers in a way that would sup-

port curricular requirements fromthe Ministry of Education, andthis was also emphasized to us bythe Ministry of Education ofªcials.We also saw that several of theexisting Reading Tutor stories hadlittle relevance to children’s livesin Mongu.

Together with the teachers, weexplored the feasibility of a year-long, in-school controlled studyusing the Reading Tutor. Theteachers suggested that a ran-dom subset of the children ingrades two through four could beselected to participate in a con-trolled study. They explained thatstudents in these grades have 60minutes of reading class per dayand suggested that the interven-tion group of students couldspend 30 minutes of that timeworking with the Reading Tutor,with teacher supervision, whilethe control group of studentswould remain in class. Thus, itwould be logistically feasible toincorporate a study into theschool schedule. However, it wasclear that there would be severalchallenges in conducting thisproject as a remote ªeld study.Communication is difªcult, giventhe school’s limited Internet

access and frequent disruptions in telephone com-munications. Success would depend on the teacherstaking ownership of the project as a result of theirown enthusiasm; however, from our experience inGhana, we realized the importance of close commu-nication and collaboration to ensure that the individ-uals running the study have the needed support. Wehoped that maintaining communication with localcontacts would help address some of the expectedchallenges. However, it has not yet been feasible toconduct a longer-term structured study, eventhough we are told that some children at the schoolhave been using the Reading Tutor. The experiencein Zambia so far has helped to identify some of thechallenges with in-school use of a tool such as theReading Tutor.



Figure 6. Teachers and a child test the Reading Tutor in Mongu.

Figure 7. Students in Mongu using word-building exercises as a group.

8. Lessons LearnedThe lessons gleaned from our experiences in Ghanaand Zambia can be broadly categorized into thoseabout study implementation, and those aboutlonger-term viability of automated tutoring for thissetting. The lessons captured below are particularlyrelevant, due to the current dearth of literature onthe evaluation of educational technology in Africancontexts. We believe that this is a fertile and impor-tant area for further research, and we recognize thevalue of discussing not only technical results, butalso lessons about process, in order to encourageand assist further work in this area.

A. Study Implementation

1) Building relationshipsAs is well-appreciated in this ªeld of research, animportant requirement for running these studieswas the process of developing partnerships at eachstage in the project. Signiªcant time must bedevoted to building relationships and developing ashared vision. Researchers have articulated thisrequirement for a variety of contexts—e.g.,Schwartzman and Parikh (2007). An interesting fea-ture of the project described in the current paper isthat it involved, at different stages, both ªeldresearch by the authors and remote collaborationwith implementation partners. In developing thevarious partnerships at each stage, we found thatphone conferences can be a useful tool, but someface-to-face meetings are essential, particularly inthe early stages of project planning. A staged imple-mentation, as in a pilot study followed by a con-trolled study, helps to distill key questions to ask,reªne design and implementation decisions, andidentify potential problems.

2) LogisticsWe learned that it was important for someoneinvolved in the local, day-to-day running of the proj-ect both to have signiªcant decision-making powerand to feel ownership of the project. This individualmust have the ability, for example, to replace equip-ment or interface with representatives of the schooladministration. Otherwise, problems can easily stallprogress. It is also essential to have access to somelocal technical expertise to troubleshoot and repairproblems—building this local support base duringthe pilot studies is crucial.

For the controlled study in Ghana, we found that

the remote collaboration required regular, some-times daily, communication among the project part-ners; e-mail, instant-messaging, and VoIP were costeffective and feasible means of achieving this.

Not surprisingly, we learned that a process thatrequires a signiªcant change in behavior and extrawork on the parts of parents, such as having thechildren come to school or participate in a studyduring a vacation, is hard to sustain and should beavoided if possible. However, it was also clear thatunexpected situations are bound to arise for anystudy that involves cross-continent collaboration, soºexibility and the ability to adjust the study design ifnecessary is essential. For example, in the controlledstudy in Ghana, the second half of the study had tobe held during the school vacation, despite the factthat this was not the original plan.

B. Viability of the Automated ReadingTutor for this Context

1) Impact of prior knowledge and experienceA key lesson was that, even without prior computerexperience, the participating children were quicklyable to acquire the skills needed to use the tutor. Ingeneral, they were excited about using the com-puter and about the interactive features of theReading Tutor. Assistance from the project staff (inGhana) or from the teachers (in Zambia) helpedthose children who were initially nervous aboutcomputer use or about reading in English. In bothlocations, the children’s natural curiosity overcameany initial apprehension with the unfamiliar technol-ogy.

We noticed that children with a basic foundationin English and not much prior experience with com-puters, such as the S2 children in Ghana, were easilyengaged with the tutor. It was clear that insufªcientfamiliarity with the English language was a chal-lenge for some, particularly the S3 children in Ghanaand some of the grade two students in Zambia. Forchildren who did not yet understand English wellenough to beneªt from an automated tutor thatuses only English, a tool that bridges between thelocal language and English, perhaps by giving oralprompts or explaining words in the local language,might have been better. A challenge to take intoconsideration, however, is that, in this context, chil-dren are often not literate in their native languageeither, so the issues of literacy and language learn-ing are intertwined. The question of whether chil-



dren learn to read ªrst in their native language or inthe country’s ofªcial language is an important issuein the discourse on primary education in Africa. Ped-agogical ªndings support the former, as discussedby Clegg et al. (2003), but many practical issuesoften result in the implementation of the latter, par-ticularly in urban areas where several languages arespoken.

Finally, we noticed that the S1 children who werealready ºuent readers and experienced with com-puters seemed to get bored and distracted easilywhen using the Reading Tutor. This might have beenbecause the Reading Tutor displays the story beingread one sentence at a time, and there would some-times be a short delay in loading the next sentence.For a ºuent reader, even this short delay was notice-able and sometimes frustrating.

2) Usage modelsAlthough the Reading Tutor is designed for use by asingle user at a time, we noticed that, in both pilotstudies, children would gather around a single com-puter and try to help each other. This also happenedto a lesser extent during the controlled study inGhana, and suggests that we should investigatemulti-user scenarios.

In running the Ghana studies, transportation ofthe children was the greatest expense, so althoughscheduled out-of-school use in an ICT center wasthe model chosen for the controlled study, its viabil-ity as a long-term usage model will depend on fac-tors such as the proximity of appropriate ICT centersto the schools. As we learned in Zambia, there arealso many challenges to be addressed for an in-school usage model for under-resourced publicschools, even when the school already has a com-puter lab. It is clear that a careful analysis of theparticular context would need to inform decisionsabout what usage models to adopt, and that clearsteps would need to be taken to address the partic-ular challenges of that context.

3) UsabilityThere are several usability improvements that can bemade to the Reading Tutor to reduce the requiredtechnical support, particularly in a developing com-munity setting. These features would apply to anyPC-based intervention.

• Installation must be easy and straightforward.

• Customization must be easy (both for addinglocal content and to control for the installation

footprint for machines without much hard diskcapacity). Although it is straightforward to addnew stories to the Reading Tutor, the processof narrating these new stories after adding thetext can be rather tedious.

• An easy administrative interface is needed forcontrolling options such as level of logging,again to deal with limited-capacity machines.

4) SummaryThe studies showed that automated tutoring withthe Reading Tutor holds signiªcant promise for help-ing children improve their reading. For this to besustainably exploited in a developing community,several challenges need to be addressed. The mostimportant challenge is designing creative usage andsupport scenarios to enable children in under-resourced schools to beneªt from automated tutor-ing despite the resource constraints and lack oftechnical expertise and infrastructure faced by theschools. A further consideration is that the ReadingTutor in its current form is a research prototype andis not readily available for non-research use. In addi-tion, as discussed above, there are features whichcould be included that would make such a tool evenbetter suited to a developing community setting.Thus, a useful development might be an open-source tool that is freely available to schools andsupported via a public-private partnership within thecountry in question.

To address the needs of various contexts within acommunity with differing levels of availableresources and support, it might be useful to consid-er a continuum of literacy-supporting tools, begin-ning with simple exercises that run on platformsthat are truly ubiquitous, such as the mobile phonesexploited in prior related work, and leading up tothe more sophisticated automated tutors for per-sonal computers.

9. Conclusions and Future WorkThis paper presents our initial investigations into theviability and effectiveness of a computer-based read-ing tutor in improving English literacy among chil-dren in Ghana and Zambia, particularly thoseattending under-resourced public schools. This workis a useful contribution to the discourse on ICT indevelopment, both because it demonstrates thatthere is promise for the effectiveness of theapproach and because of the practical lessons



regarding the implementation of the study. We havepresented an initial proof-of-concept investigation.Many additional questions would need to beexplored to understand the potential for large-scaleapplication of these technologies.

Follow-up work for our project includes pursuingthe goal of conducting a year-long study to test thefeasibility of an in-school usage model in a develop-ing community setting. This study would both meas-ure the children’s reading gains over an entire schoolyear of using the Reading Tutor and compare thesegains to those obtained with a non-technologicalapproach to guided reading practice. The studywould also investigate the feasibility of incorporat-ing teachers closely into this work.

Additionally, we will continue to develop anunderstanding of what features would be requiredfor a continuum of literacy-enhancing tools devel-oped speciªcally for use by children in developingcommunities in Africa and elsewhere. What needscan be met with lighter-weight tools (e.g., mobilephones) versus computer-based tools? Others in ourresearch group have recently begun this processthrough an initial project in Tanzania that resulted inthe design and implementation of a prototypemobile phone-based literacy game for children andan accompanying Web-based content authoringtool for teachers (Abimbola et al., 2009). There issigniªcant scope for further research on the role ofcomputing technology in improving child literacy indeveloping communities. ■

AcknowledgmentWe express our appreciation to the Ghana-India KoªAnnan Centre for Excellence in ICT (AITI-KACE);Accra, Ghana, for their role in the controlled study,particularly to Dorothy Gordon for her support;Patricia Nyahe for her tireless management of thestudy; and the UNESCO Ghana Ofªce for sponsor-ing their work. We also thank Leslie Casely-Hayfordof Associates for Change, Ghana, for her advice andcontributions to the pilot and controlled studies inAccra. We acknowledge the valuable contributionsof Cybercity Internet Café and ProjectEDUCATE tothe pilot studies in Ghana and Zambia respectfully.We are grateful to Steve Fienberg for his advice onthe statistical analysis as part of the Statistical Prac-tice course at Carnegie Mellon University. Finally, weare indebted to all the participating children and

schools, and to all others who contributed to theproject in various ways.

ReferencesAbimbola, R., Alismail, H., Belousov, S. M., Dias, B.,

Dias, M. F., Dias, M. B., et al. (2009, August).iSTEP Tanzania 2009: Inaugural experience. (Car-negie Mellon University, Robotics Institute Techni-cal CMU-RI-TR-09-33). Retrieved April 25, 2010,from http://www.ri.cmu.edu/publication_view.html?pub_id�6422&menu_code�0307

Adams, M. J. (2006). The promise of automaticspeech recognition for fostering literacy growthin children and adults. In M. C. McKenna, L. D.Labbo, R. D. Kieffer, & D. Reinking (Eds.), Interna-tional handbook of literacy and technology, Vol-ume 2. Mahwah, NJ: Lawrence ErlbaumAssociates.

Allington, R. L., & McGill-Franzen, A. (2003, Sep-tember). The impact of summer setback on thereading achievement gap. Phi Delta Kappan,85(1), 68–75.

Banerjee, A. V., Cole, S., Duºo, E., & Linden, L.(2007, August). Remedying education: Evidencefrom two randomized experiments in India. Quar-terly Journal of Economics, 122(3), 1235–1264.

Clegg, J., Ogange, B., & Rodseth, V. (2003, April).Evaluating digital learning material for Englishlanguage development in African primary class-rooms. IMFUNDO KnowledgeBank Paper. Re-trieved April 25, 2010, from http://imfundo.digitalbrain.com/imfundo/web/papers/refpapers/?verb�view

Cohen, J. (1988). Statistical power analysis for thebehavioral sciences (2nd ed.). Hillsdale, NJ: Law-rence Earlbaum Associates.

Cohen, J. (1992). A power primer. Psychological Bul-letin, 112, 155–159.

Deno, S. L. (1985). Curriculum-based measurement:The emerging alternative. Exceptional Children,52(3), 219–232.

Hasbrouck, J. & Tindal, G. A. (2006). Oral readingºuency norms: A valuable assessment tool forreading teachers. The Reading Teacher, 59(7),636–644.



Kam, M., Agarwal, A., Kumar, A., Lal, S., Mathur,A., Tewari, A., et al. (2008, February). Designinge-learning games for rural children in India: Aformat for balancing learning with fun. Proceed-ings of the 7th ACM Conference on DesigningInteractive Systems (DIS ’08), 58–67.

Kam, M., Kumar, A., Jain, S., Mathur, A., & Canny,J. (2009, April). Improving literacy in rural India:Cellphone games in an after-school program.Proceedings of the 3rd International Conferenceon Information and Communication Technologiesand Development (ICTD 2009), 139–149.

Kozma, R. B. (2005, November). Monitoring andevaluation of ICT for education impact: A review.In D. A. Wagner, B. Day, T. James, R. B. Kozma,J. Miller, & T. Unwin (Eds.), Monitoring and evalu-ation of ICT in education projects: A handbookfor developing countries (pp. 11–18). Washing-ton, DC: infoDev/World Bank. Retrieved April 25,2010, from http://www.infodev.org/en/Publication.9.html

Larsen, S. C., Hammill, D. D., & Moats, L. C. (1999).Test of written spelling. Austin, TX: Pro-Ed.

Lipson, M., & Wixson, K. (2004, August). Evaluationof the BTL and ASTEP programs in the Northern,Eastern, and Volta regions of Ghana. Report pre-pared by the International Reading Associationfor The Education Ofªce, USAID/Ghana. Re-trieved April 25, 2010, from http://www.reading.org/General/CurrentResearch/Reports/GhanaReport.aspx

Louw, J., Muller, J., & Tredoux, C. (2008, February).Time-on-task, technology and mathematicsachievement. Evaluation and Program Planning,31(1), 41–50.

Miller, R. G. (1981). Simultaneous statistical infer-ence (2nd ed.). New York: Springer Verlag.

Mostow, J. & Aist, G. (2001). Evaluating tutors thatlisten: An overview of Project LISTEN. In K. D.Forbus & P. J. Feltovich (Eds.), Smart machines ineducation (pp. 169–234). Cambridge, MA: AAAIPress/The MIT Press.

Mostow, J., Aist, G., Huang, C., Junker, B., Kennedy,R., Lan, H., et al. (2008). 4-Month evaluation of alearner-controlled Reading Tutor that listens. InV. M. Holland & F. P. Fisher (Eds.), The path of

speech technologies in computer assisted lan-guage learning: From research toward practice(pp. 201–219). New York: Routledge.

National Institute of Child Health and Human Devel-opment (NICHD). (2000). Report of the NationalReading Panel. Teaching children to read: An evi-dence-based assessment of the scientiªc researchliterature on reading and its implications forreading instruction (NIH Publication No. 00-4769). Washington, DC: U.S. Government Print-ing Ofªce. Retrieved April 25, 2010, from http://www.nichd.nih.gov/publications/nrp/smallbook.cfm

Poulsen, R., Hastings, P., & Allbritton, D. (2007). Tu-toring bilingual students with an automatedReading Tutor that listens. Journal of EducationalComputing Research, 36(2), 191–221.

Reeder, K., Shapiro, J., Early, M., Kendrick, M., &Wakeªeld, J. (2005). The role of L1 in youngmultilingual readers’ success with a computer-based reading tutor. Proceedings of the Fifth In-ternational Symposium on Bilingualism.

Schwartzman, Y. & Parikh, T. S. (2007). Establishingrelationships for designing rural information sys-tems. CHI ’07 Extended Abstracts on Human Fac-tors in Computing Systems. New York, NY: ACM.Available at doi� http://doi.acm.org/10.1145/1240866.1240909

Senn, S. (1993). Cross-over trials in clinical research.Chichester, UK: John Wiley & Sons.

Stanovich, K. E. (2000). Progress in understandingreading: Scientiªc foundations and new frontiers.New York: Guilford Press.

UNESCO. (2005). Education for All global monitor-ing report 2005—Education for All: The qualityimperative. United Nations Educational, Culturaland Scientiªc Organization (UNESCO) Publishing.Retrieved April 25, 2010, from http://www.unesco.org/en/efareport/reports/2005-quality/

Wagner, D. A. (2005, November). Monitoring andevaluation of ICT4E: An introduction. In D. A.Wagner, B. Day, T. James, R. B. Kozma, J. Miller,& T. Unwin (Eds.), Monitoring and evaluation ofICT in education projects: A handbook for devel-oping countries (pp. 5–9). Washington, DC:



infoDev/World Bank. Retrieved April 25, 2010,from http://www.infodev.org/en/Publication.9.html

Wagner, D. A., Day, B., & Sun, J. (2004, March). In-formation technologies and education for the

poor in Africa (ITEPA). IMFUNDO KnowledgeBankPaper. Retrieved April 25, 2010, from http://imfundo.digitalbrain.com/imfundo/web/papers/refpapers/?verb�view



research article improving child literacy in africa ...listen/pdfs/2010 itid korsah...

Documents