Literacy Boost South Africa Baseline Report

Transcription

1 Literacy Boost South Africa Baseline Report November 2013 Ntonto Vezi, Lebohang Mokoena, Elson Mkhoma, Suzanne Wessels, the Thabo Mofutsanyana District ECD Team and Amy Jo Dowd With special thanks to Save the Children staff and our team of enumerators: Gatebe Nomvula, Hlakutsa Katiso, Majoro Nthabiseng, Makgapetla Ntsamayeng, Maseko Lisbeth, Mokhatla Ntswaki, Mokoena Gcwalisile, Moloi Lekgula, Mosia Putswane, Mphuthi Ntebaleng, Mtshawe Nthabiseng, Nhlapo Thobile, Paulosi Radebe, Radebe Kefiloe, Shabalala Sithembile, Soke Hadio, Suping Teleko, Tumisi Krefisi, and Vezi Ntonto 2013 Save the Children

2 Executive Summary In May 2013, 275 grade 2 students in 14 primary schools in the Nketoana and Maluti A Phofung municipalities in the Free State of South Africa participated in a reading skills and habits survey. Children in seven of these schools will receive the Literacy Boost pilot program, so this report serves as the baseline. Literacy Boost includes teacher training, community reading activities, and ageappropriate local language material creation to support emergent literacy skills among early-grade children. Skills assessed include basic concepts about print, and letter awareness through reading fluency, accuracy, and comprehension in Sesotho, Isizulu and English. As part of Literacy Boost, students are periodically assessed in each of these skills through an adaptable assessment tool to inform programming and estimate program impact. This report investigates Literacy Boost schools and comparison school comparability, and trends in students reading skills and how to student background and home literacy environment affect them in ways important for shaping Literacy Boost programming. The comparison and Literacy Boost samples are similar on all reading skills measures as well as all but very few background variables, suggesting that the group of comparison students represents a satisfactory counterfactual for testing the impact of Literacy Boost. The data further highlight some of the strengths Literacy Boost can build upon and some of the challenges the program team will face in this population of children: 1. Few reading materials written for children (comics, coloring books, and stories) are available, and between 25 and 40 percent of children are not routinely read to, told stories or helped with homework; 2. In both home languages, words with double consonants and longer words challenged children more, so reading camp and in-class activities should reinforce sounding out long words, blending letters and daily reading practice. Supporting learners to tackle blends of consonants and decode longer words could be a key element of a model lesson for the next teacher training session 3. Literacy Boost should also aim to help students expand their English vocabulary and support the weakest students to strengthen their English word and text reading as well. 4. Repetition most consistently predicts reading skills at baseline: if a child repeated first and/or second grade, scores are lower. 5. The analysis unearthed two key target groups: those not yet reading text (in each target language) and those who have already repeated a grade. Training teachers to use formative assessment to identify and support these children is a good place to begin dialogue about supporting these potentially struggling readers. Through reading camps and parent workshops, Literacy Boost can promote the creation of reading materials to build home resources. Further, community and school book banks can promote borrowing to enhance variety of books in these and all children s lives.

3 Table of Contents Introduction...1 Context...1 Methods...2 Sample...2 Measurement...2 Analysis...2 Children s Background...3 Gender Differences...3 Children s Reading Skills...3 Concepts about Print...3 Letter Identification...3 Word Recognition: Most Used Words...4 Fluency and Accuracy...4 Comprehension...5 Profiles of reading skill scores...5 Children s Home Literacy Environment...7 Relationships between Skills and Home Literacy Environment...8 Background Characteristics...8 Conclusion: Benchmarks and Implications for Programming...9 Appendix A: Tables of Means Appendix B: Item tables for CAP, Letters and Words Appendix C. Regression Models... 15

4 Introduction This report examines the results of a learner background survey and reading assessment conducted between August and October 2013 among grade 2 students in seven primary schools receiving Save the Children (SC) s Literacy Boost program as well as seven nearby comparison primary schools in the Nketoana and Maluti A Phofung municipalities in the Free State of South Africa. The Literacy Boost program includes teacher training, community reading activities, and age-appropriate local language material creation to support developing reading skills among early-grade children, such as: concepts about print, letter awareness, single word reading, fluency, accuracy, and comprehension. As part of Literacy Boost, learners are periodically assessed to inform programming and estimate program impact. The data gathered from these schools is analyzed to present a snapshot of the emergent literacy skills of grade 2 learners in these schools and to inform the adaptation of SC s Literacy Boost program to this context. The key research questions explored in this report include: 1. How comparable are learners in Literacy Boost schools versus comparison schools in terms of reading skills, background characteristics, and home literacy environment? 2. What can the baseline tell us about learners emergent reading skills? What does this mean for Literacy Boost programming? 3. How do learners reading skills vary by student background and home literacy environment? What does this mean for targeting Literacy Boost s two strands of intervention? To investigate these questions, this report will first describe the research methods used; including sampling, measurement, and analysis. Next, in order to see if groups are statistically similar, the comparability of Literacy Boost and comparison schools will be examined through clustered t-tests. The comparability of Literacy Boost and comparison learners scores for each of the emergent literacy skills, exploring learners strengths and weaknesses in each skill will also be examined. The report will then examine what are the literacy skills that are already present in the sample, and what areas should Literacy Boost focus on. Finally, the report will investigate relationships between skills and student background, and home literacy practices & environment variables using multilevel regression analysis. Context The sample for this baseline assessment encompasses 275 grade 2 students, divided between seven Literacy Boost schools (n = 140) and seven nearby government comparison schools (n = 135). The seven Literacy Boost schools chosen for baseline assessment were those selected by SC South Africa country office staff to receive the Literacy Boost program based on the finding from the 2010 Strengthening Our Schools needs assessment which indicated an urgent need to improve Foundation Phase Literacy outcomes. The seven comparison schools were selected based on their proximity to the target schools as well as the perceived similarity between the quality of instruction offered at schools, and learners home and socio-economic conditions. These schools generally share the same environment as Literacy Boost schools. Since 2011, target schools participated in Save the Children s Strengthening Our Schools project, which aims to improve quality of teaching, learning environments, school management and governance 1

5 and district authorities capacity to provide support to whole school development. In July of 2012, six of the schools began community action component activities, including read-a-thons, reading camps, and parent reading awareness workshops; the seventh school began these in September In addition, the first, second and third grade teachers from these schools attended two non-literacy Boost training events. Comparison schools do not benefit from any intervention that aims to improve reading outcomes, other than routine government teacher training, as is the case for target schools. During analysis we will reflect on whether these intervention differences relate to differences between the treatment and comparison groups by mid Methods Sample The data collection team set out to sample 20 children in grade 2 at each of the Literacy Boost and comparison schools. Students were randomly selected where there were more than 20 children in the classroom. This approach would stratify the sample by gender so that boys and girls would each comprise 50% of each school sample. However, this was the case in just six of fourteen schools, while in seven there are either more girls (five schools) or more boys (three schools) available for testing due to transportation conflicts, and in one school only 15 children were assessed as there were not 20 children in that small school. Overall, there are 130 boys (59 in comparison and 71 in Literacy Boost schools) and 145 girls (69 in Literacy Boost and 76 in comparison schools) in the sample. Measurement The students in the sample were asked about their background characteristics (age, school history, socio-economic status, etc.). To capture health/nutrition status, students were also asked whether or not they had eaten breakfast the morning of the assessment, and their height and weight were measured. Finally, students were asked about their family members and their reading habits in the week prior to the assessment (who they had seen reading, who had read to them, etc). After collecting this background data, all students were also given an emergent literacy test composed of six components administered through five sub-tests: concepts about print, letter awareness, vocabulary (reading of most used words), reading fluency & accuracy (words per minute read correctly and total percentage of passage read correctly, both within the same sub-test), and either reading comprehension questions for those who could read independently or oral comprehension questions for those who could not. The vocabulary, fluency & accuracy, and comprehension sub-tests were conducted in both the child s home language (Sesotho or Isizulu) and English. Thirty-five children across two schools took the assessment in Isizulu; while the remaining children took it in Sesotho. All assessments instructions were given in the child s home language. Analysis The critical purpose of this analysis is to establish the skill and home literacy profile of learners prior to the intervention and to test whether the Literacy Boost learners and the comparison learners are equal in terms of background and skills. That is, do these learners possess the same resources and capabilities? This question is important so that at end-line, we can know how much Literacy Boost has, or has not, contributed to learners' accelerated reading development. 2

6 To test the comparability of learners in Literacy Boost and comparison groups, this report will use comparison of means through t-tests, with clustered standard errors to account for the grouping of student-level data within schools. Summary statistics, accompanied by clustered t-tests, will be used to analyze learners performance in each of the reading sub-tests. Finally, this report will look to multilevel regression models to explore relationships between literacy skills and student background characteristics, and home literacy environment. Children s Background The students are, on average, nearly eight years old, live with four other family members, and most have electricity, televisions, toilets and refrigerators in their homes. More than three quarters attended ECD before primary school and more than a third have repeated at least one grade. Most ate breakfast and feel healthy, but 16 percent are stunted according to WHO standards. Literacy Boost and Comparison groups do not differ in any of these and additional background characteristics listed with group averages in Appendix A, Table A.1. Gender Differences On average, girls are significantly younger, significantly poorer and have repeated grades significantly less often than boys. These are differences to keep in mind as we consider the reading skills of children and, later, factors that influence these skills. Children s Reading Skills Concepts about Print The first sub-test of the reading assessment consisted of 11 concepts about print (CAP) questions. These questions concern familiarity with books where to start, which way to read, etc. On average, learners answered correctly 89 percent of the concepts about print assessed on average. This indicates learners have a high level of exposure to and understanding of books and reading materials. There were no differences in CAP scores between Literacy Boost and comparison students; and no differences between boys and girls scores. Letter Identification The next sub-test examined learners letter knowledge. Isizulu learners were shown a chart of 52 letters: 26 lowercase and 26 uppercase; while Sesotho learners were only shown 44. The students were asked to name the letter or pronounce the letter sound. On average, Literacy Boost and comparison learners who speak Sesotho at home correctly identified 93.6 percent (41.2) of letters. There were no significant differences by group or sex. IsiZulu-speakers correctly identified 87.2 percent (43.6) of letters, again with no differences between groups or between boys and girls. Table 1 shows the easiest and hardest letters, but note that even the hardest letters are known by nearly or more than three quarters of the children assessed. Overall most children know most of their letters. 3

7 Table 1. Hardest and easiest letter by % correctly identifying (red were on IsiZulu form only) HARDEST EASIEST UPPERCASE Mean LOWERCASE Mean UPPERCASE Mean LOWERCASE Mean I 72% r 78% R 98% w 98% X 78% k 79% S 98% x 98% G 78% u 81% O 98% d 98% C 81% j 82% U 99% p 98% W 83% e 85% M 99% f 99% Word Recognition: Most Used Words The most used words (MUW) sub-test consists of a chart of 20 words that the student is asked to read. These 20 words were identified as most used by tabulating the number of times a word appeared in learners language arts textbooks. For this sample, the students were asked to read in two languages: Sesotho or IsiZulu, and English. The learners who read the Sesotho words in Literacy Boost and comparison schools, read 84 percent correctly; while the learners who read IsiZulu words read 92% of them correctly. All children read English and read 59 percent of English words correctly. There were no differences between Literacy Boost and comparison students; and no differences between girls and boys average scores. There were also no differences in the English reading of Sesotho and IsiZulu speakers. In Sesotho, kreiti, mmele, shebile and mofuthu were the most difficult words read correctly by percent of students, and bata, bala and rata were the easiest read correctly by nearly percent of students. Note that the hard words have double consonants and vowels, while the latter do not. In Isizulu, ehlathini, enikeza, and epulazini were the hardest read by 81 percent of students, while funda, hamba, ijuba, ilanga, ingozi, and ukudla were all read correctly by 97 percent of Isizulu speakers. In IsiZulu, the longest words represent a greater challenge. In English, learners least read correctly saw, eat, and are; while the, and, swim, and school were read correctly by over 70 percent of students. A full list of all words and the percent of children reading each correctly is presented in Appendix B, Table B.1. Fluency and Accuracy Fluency (words per minute read correctly) and accuracy (percent of the passage read correctly) are presented together here because they are measured together in a single subtest in which learners read a passage aloud. The number of words learners read correctly in a minute is tracked for fluency. As the student continues to read after the first minute, the total number of words read correctly from the passage as a whole, no matter how long it takes the student, is computed for accuracy. On this assessment, 17.3 percent of Sesotho-speaking children could read no words; and only one Isizulu-speaking child could not read in her own language. On average, Sesotho-speakers read at a rate of 40.2 words per minute correct, and 75.2 percent accuracy. IsiZulu-speakers, on average read 23.5 words per minute correctly with 90.5 percent accuracy. In English, there were 47.6 percent of learners who could not read any words of the text, and the remaining students read 25.4 words per minute correctly with 44.1 percent accuracy on average. There were no significant differences between the 4

8 fluency and accuracy skills of Literacy Boost and comparison students or between boys and girls in any language. Comprehension The final sub-test asked students a series of five comprehension questions for each passage. For those students who were unable to read a single word from the passage, the assessor read the passage to the student before asking the comprehension questions. Those who could read on their own were not read to by the assessors, and tested on the same comprehension questions. In Sesotho, on average, those who were read to answered 42.6 percent of comprehension questions correctly, while those who read themselves answered 77 percent. In English on average, those who were read to answered 19.8 percent of comprehension questions correctly, while those who read themselves answered 54 percent correctly. In Sesotho, there were no significant differences between the listening or reading comprehension scores of Literacy Boost and comparison students or between boys and girls. In English, Literacy Boost learners who listened to the passages had significantly higher comprehension (24.9 percent correct) than comparison learners (13.8 percent correct); but both had significantly lower comprehension than their peers who read to themselves (54 percent correct). Profiles of reading skill scores Figures 1-3 show the average reading skills scores for Literacy Boost students in Sesotho and IsiZulu, and then for all learners in English. 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 92.0% Sesotho Reading Skills (n=240) 88.5% % 51.0% letters words fluency accuracy listening comprehension 78.8% readling comprehension On average, Sesotho-speaking learners in grade 2 are headed to reading with comprehension. They know most letters, and read many common words seamlessly. While 17.3 percent are not yet reading text, the rest are honing their fluency and accuracy to ensure fuller comprehension in their first language as the school year goes forward. 5

9 IsiZulu Reading Skills (n=35) 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 84.5% 91.5% 90.3% 64.5% 23.9 letters words fluency accuracy comprehension Similarly, the smaller group of IsiZulu-speaking children is also showing near mastery of foundational skills and has high accuracy. Interestingly, rate of reading is slower but the words appear to be longer as well! This underscores the importance of considering the metric within language and not across them. We ll look for the endline IsiZulu fluency to be higher than 23.9 words per minute correct, and not judge it against the 42 words per minute correct of Sesotho-speakers in Figure 1. English Reading Skills (n=275) 70% 60% 61.4% 58.2% 50% 40% 42.3% 30% 20% 10% % 0% words fluency accuracy listening comprehension readling comprehension Finally, Figure 3 features the same measures for English and show that these skills, as expected, lag behind skills in the mother tongue on averages. It is worth noting that 46.7 percent of students read no words correctly of this English passage, while some children can read English, addressing the many zero scores in this profile will show program impact. There is further distance to cover in learning to read English with comprehension. 6

10 Children s Home Literacy Environment An important aspect of reading development concerns the home literacy environment (HLE). How are children exposed to the printed word in the home? How much access do they have to books and print to practice their nascent reading skills? Many Literacy Boost activities are centered on helping parents and communities to enhance the HLE. As such, it is important to measure where learners' HLE begin, and how they change over the course of time. Figure 4 displays the different types of printed materials that learners may or may not have at home. Figure 4. Home Literacy Materials by Group comics coloringbks stories newspaper magazines religious textbooks 14% 13% 41% 37% 42% 45% 61% 68% 62% 55% 62% 76%* 99% 98% 0% 20% 40% 60% 80% 100% 120% Literacy Boost Comparison School It is clear that school and religious materials are the most common items that children report having at home, but fewer children have children s reading materials storybooks, comics, picture books coloring books at home. In fact, 41 percent of children in the sample have none of these at home. Only the percent of children who have newspapers at home is statistically significantly different between Literacy Boost and comparison groups or boys and girls, so access to children s materials is the same and access to any materials slightly better for newspapers only among Literacy Boost families. Given that books were purchased in 2012, follow up on their circulation and use might improve access moving forward. The HLE is not only about materials in the home, but how those materials are used to engage children in reading and learning. Hess and Holloway (1984) identified four additional dimensions of the home literacy environment that are theoretically related to reading achievement in children: the value placed on literacy, reading with children, press for achievement and opportunities for verbal interaction. We operationalize these by asking the learners whether last week they saw anyone reading, whether anyone read to them, helped with homework or told them stories. 7

11 80% 70% 60% 50% Figure 5. Home Literacy Interactions by Group 72%* 66% 68% 62% 58%* 49% 45% 44% 40% 30% Comparison School Literacy Boost 20% 10% 0% % seen reading % read to child % help child study % told stories Figure 5 shows that, overall, the vast majority of learners saw reading and were read to by about half of their family members. Statistically significant differences exist between Literacy Boost and comparison groups in reading and seeing reading, a possible effect of SOS/Literacy Boost already being underway in these communities. No differences exist for supporting studies or oral language development. Relationships between Skills and Home Literacy Environment This final section explores the results of a series of multilevel regressions, accounting for clustering of students within schools. To arrive at the most relevant regression model, the significance of student background and home literacy environment variables were tested in both univariate and combined multivariate models, using reading skill sub-tests as the dependent variables. Appendix B presents the results of the final multivariate models for each literacy subtest. It should be noted that these relationships convey correlation rather than causation. To establish causation requires further research and/or endline analysis. Background Characteristics Two consistent background characteristics negatively predict skill scores: grade repetition and being in a multi-grade class. As can be seen in Figure 6, children who repeated first or second grade have lower predicted fluency (-15 wpm in Sesotho and -17 wpm in English), accuracy (-16% in Sesotho and -29% in English), and reading comprehension (-14% in English) than those who did not repeat. 8

12 Figure 6. Predicted Reading Skills by Repetition Status fluency (wpmc) accuracy (%) single words (%) fluency (wpmc) accuracy (%) comprehension (%) 43.2 repeater non-repeater Sesotho English These children represent more than a third of the population and could benefit from greater support to make progress in these skills. Further, the children in multi-grade classrooms have lower predicted Sesotho word reading (-4 words), fluency (-17 wpm in both Sesotho and English) and Sesotho accuracy (-28%). Recall that these findings are correlational and do not show that repetition or being in a multigrade classroom causes lower scores but their consistency across so many skills identifies these learners as key targets for additional support. Conclusion: Benchmarks and Implications for Programming While Save the Children has used this approach to reading assessment and intervention in Bangladesh, Ethiopia, Guatemala, Indonesia, Malawi, Mali, Mozambique, Nepal, Pakistan, Peru, the Philippines, Sri Lanka, Uganda, and Zimbabwe, comparison across countries and languages is less helpful than more detailed contextual information for setting expectations of impact. For each measure used in these assessments, the upper end of the range of scores can be used to consider what is currently possible among these children. The scores at the 75 th percentile of each measure function as locally-derived benchmarks for Literacy Boost students at the end-line assessment. Below, we present these figures a alongside recommendations for programming to meet them based on the evidence in this report. After taking up in the intervention phase the recommendations for moving these key skills forward, the endline assessment will check our progress towards them. Table 2 outlines the skill benchmarks this evidence proposes for students as well as suggested in-school and out-of-school activities associated with improving these skills. 9

13 Table 2. Benchmarks and recommendations by skill Skill 75 th percentile Program implications CAP 100% Knowing the beginning, where to start reading, and what direction to go will help those lagging behind in print awareness to master this foundational skill. The program team, teachers and reading camp leaders should focus on activities that support children reading in groups in reading camps as well as reading buddy pairs to have the children who are listening or reading following the words with their finger as a more advanced reader reads to them. Letter identificat ion Reading Single Words Sesotho: 100% IsiZulu: 88% Sesotho: 100% IsiZulu: 100% English: 95% Fluency Sesotho: 63 IsiZulu: 40 English: 55 Accuracy Sesotho: 97% Comprehension IsiZulu:100% English: 91% Sesotho: 100% IsiZulu: 80% English: 60% In both languages, uncommon letters were most challenging, so the program team should encourage teachers to focus on activities/games/strategies that teach and review these letters and their sounds, as well as emphasizing the daily decoding of multi-letter words both in class and in reading camps. In both home languages, words with double consonants and longer words challenged children more, so reading camp and in-class activities should reinforce sounding out long words, blending letters and daily reading practice. Supporting learners to tackle blends of consonants and decode longer words could be a key element of a model lesson for the next teacher training session. Given the large number of children currently reading independently at baseline, the reading camps and buddies should emphasize reading aloud to strengthen fluency and feature lots of comprehension debates, discussions and games. At the same time, there is a subset of children not currently reading that program staff and teachers can identify and support to master foundational skills and get on track for progress in fluency, accuracy and comprehension. Tracking (following words while reading aloud by pointing with your finger) while reading to each other, and daily practice enhance fluency. Among current readers, the program can use their skills to support others and through this increased reading practice grow towards the reading goals in each target language. Program staff, teachers and reading camp leaders should encourage daily reading with reading buddies, family members and groups in reading camps and during these practice times to listen to each other carefully, follow along in the text and offer input to correct reading and intonation. Program staff should communicate with teachers, reading camp leaders and parents about the importance of not only reading with children, but also asking children questions and opinions about the text afterwards. This will enhance both understanding and confidence in expression. Given that children s reading comprehension scores were higher than their nonreading peers listening comprehension scores, program staff will also need to focus on promotion of oral language development activities included in the flipbook, parent workshops and camps to ensure that children have the vocabulary and verbal skills necessary to understand the text and demonstrate that comprehension. Finally, repeaters and the children in multi-grade classrooms are in need to additional supports to bring their average scores up and support programmatic progress overall. 10

14 Appendix A: Tables of Means Table A1: Average Background Characteristics by Sample Group GENERAL Comparison Schools (N=135) Literacy Boost (N=140) speaks home 16.30% 9.30% speaks home 83.70% 90.70% class is multi-grade 11.10% (1 school) 14.30% (1 school) age female 56.30% 49.30% attended ECD 86.70% 77.10% repeated grade % 29.30% repeated grade % 20.70% ever repeated 29.60% 39.30% SES Iron/tin roof 90.37% 90.71% brick wall 66.43% 70.37% tv 85.90% 84.30% electricity 89.60% 80.70% fridge 72.60% 72.10% bike 22.20% 15.00% toilet 96.30% 97.90% number in household SHN height (cm) weight (kg) breakfast 77.00% 82.10% Breakfast size lunch 99.30% 99.30% Feels healthy 97.00% 94.20% stunted 15.70% 16.40% wasting 5.90% 1.40% malnourished 9.60% 5.70% 11

15 WORK washing 14.10% 23.60% cleaning 9.60% 16.40% tending 2.20% 2.90% Study time (minutes) Table A2. Home Literacy Environment by Sample Group Home Literacy Environment Comparison (N=135) Literacy Boost (N=140) Reading Materials in Home 100% 100% N of different types of reading materials at home Religious books 98% 99% Newspaper 20% 15% Textbooks 95% 97% Magazine 19% 14% Story books 29% 28% Significant Difference Coloring books 26% 25% Comics 20% 23% Picture books 25% 26% Total N of family seen reading at home % of Family Seen Reading 34% 35% Total N of family read to you at home % of Family Read to child 28% 30% 12

16 Appendix B: Item tables for CAP, Letters and Words Table B1 Percent answering correctly by item CAP % identified correctly CAP % identified correctly cap11 82% cap7 89% cap4 83% cap9 90% cap5 84% cap8 95% cap6 84% cap1 97% cap10 87% cap2 100% cap3 87% Table B2. Percent who read the word correctly by letter UPPERCASE % identified correctly LOWERCASE % identified correctly I 72% r 78% X 78% k 79% G 78% u 81% C 81% j 82% W 83% e 85% Y 85% z 86% Q 86% g 88% Z 91% c 91% J 91% q 92% H 92% a 92% V 94% l 95% D 94% s 95% T 95% y 95% P 96% h 96% F 96% n 96% L 96% o 96% K 97% i 96% N 97% t 96% E 97% v 97% A 97% m 97% B 97% b 98% R 98% w 98% S 98% x 98% O 98% d 98% U 99% p 98% M 99% f 99% 13

17 Table B3. Percent who read the word correctly by word and language Sesotho word % read correctly IsiZulu word % read correctly English word % read correctly kreiti 78% ehlathini 81% saw 19% mmele 79% enikeza 81% eat 33% mofuthu 81% epulazini 81% are 46% shebile 81% benza 84% all 48% metswalle 82% elikhulu 88% don't 50% thuto 83% enzicane 88% running 54% titjhere 83% amagama 91% when 56% sekolong 84% bukani 91% went 57% nako 84% umalusi 91% then 59% motho 84% vakasha 91% like 62% thabile 84% memeza 94% playing 64% phoofolo 85% sonke 94% tree 65% sefate 85% umngani 94% has 65% lapeng 86% funda 97% we 66% bapala 88% hamba 97% sun 68% monate 88% ijuba 97% go 70% buka 89% ilanga 97% the 71% bala 90% ingozi 97% and 72% bana 90% ukudla 97% swim 73% rata 92% school 82% 14

18 Appendix C. Regression Models VARIABLES (1) (2) (3) (4) (5) (6) (7) English Sesotho Sesotho word English English fluency accuracy reading fluency accuracy Sesotho word reading English comprehension Has Repeated a Grade ** * *** ** ** ** (1.231) (4.214) (0.0695) (1.119) (4.866) (0.0739) (0.221) Multi-grade classroom ** ** ** ** (1.268) (4.388) (0.0717) (1.961) (4.417) (0.105) (0.271) Constant 18.15*** 47.54*** 0.839*** 13.88*** 33.40*** 0.570*** 2.161*** (0.722) (3.487) (0.0422) (1.120) (5.218) (0.0663) (0.263) Observations R-squared Robust standard errors in parentheses *** p<0.001, ** p<0.01, * p<