Journal of Research in Reading

Medium of instruction

Much research suggests that using a child's mother tongue as the medium of instruction raises student achievement. Thomas and Collier (2002) showed that the longer immigrant children in the United States were taught in their mother tongue, the better their academic achievement. The Ife project in Nigeria (Fafunwa, Macauley, & Sokoya, 1989) demonstrated that using Nigerian languages of instruction improved academic performance. Assessments from Ethiopia (Heugh, Benson, Bogale, & Yohannes, 2007), Zambia (Williams, 1996) and Mali (Fomba et al., 2003) have argued that the use of the mother tongue as the medium of instruction leads to better learning.

However, as Benson (2004) observed, parents, communities and policy‐makers often demand ex‐colonial languages more than the mother tongue. The result is that many countries utilise languages that are not understood by many children (Trudell, 2007). The complexity of book production in many languages often turns cost‐benefit analyses against local languages. If there is insufficient attention to teaching in local languages, if books are unavailable and if teachers are undertrained, then instruction in local languages will be poorly executed. A conventional wisdom then emerges that low quality and poor learning outcomes are an inevitable result of linguistic complexity.

Reading fluency

Reading fluency has been described in various ways (National Institute of Child Health & Human Development [United States], 2000; Pikulski & Chard, 2005; Samuels, 2006), but we define it as the ability to read accurately, quickly and with appropriate expression or prosody (Rasinski, 2003). Although this assessment did not gauge prosody – which is difficult to measure reliably (National Institute of Child Health & Human Development [United States], 2000; Rasinski et al., 2010; Schwanenflugel, Hamilton, Wisenbaker, Kuhn, & Stahl, 2004) – we use the broader term fluency to respond to the literature that uses fluency and reading rate interchangeably.

In some U.S. reading assessment contexts, the most common measure of reading fluency is the number of correct words read per minute (cwpm) (Deno, 1985). Some U.S.‐based researchers set the benchmark for English oral reading fluency needed for comprehension at 60–90 cwpm in Grade 1, 85–120 in Grade 2 and 115–140 in Grade 3 (Harris & Sipay, 1990; Hasbrouck & Tindal, 2006).

Oral reading fluency measures must be language‐specific because word length and orthographic complexity vary. Ellis et al. (2004) argued that the relationship between word identification fluency and word length is more linear in transparent orthographies. Trudell and Schroeder (2007) observed that, because of phonologically derived word‐break rules, Bantu language words are often long and polymorphemic, so they take longer to decode (more than 500 languages in eastern and southern Africa are in the Bantu family, including Gikuyu and Kiswahili). Cwpm measures are problematic in comparing fluency across languages; therefore, they cannot account for word length differences and grapheme complexity, and they ignore the fact that relationships between decoding speed and word length can be non‐linear (Ellis et al., 2004).

Implications for Kenyan languages

These linguistic features have an impact on reading fluency for the languages in this article. Fluent reading in Bantu languages such as Gikuyu and Kiswahili is aided by their transparent orthographies; on the other hand, fluency requires the reader to master both polyphonemic syllables and polymorphemic words. In contrast, the relatively short words in English – and its orthographic depth – make sight‐word recognition a more viable initial reading strategy than for many Bantu languages.

For this discussion, the most important feature of Dholuo is its orthographic underrepresentation (Okombo, 1997), both in the number of vowels and the lack of tone markings. As a result, Dholuo readers try to read text via a writing system that often fails them; it hides the linguistic cues and the phonemic contrasts that their spoken language provides. Dholuo has nine contrastive vowel sounds, but they are represented with only five vowel letters. The letter  < o>, for example, represents two vowel phonemes, /o/ and /ɔ/. <chanjo> or ‘limping’ is pronounced /tsanjɔ/, while < chanjo> or ‘to vaccinate’ is pronounced /tsanjɔ/. The Dholuo writing system does not show the contrast. Aside from thousands of vowel contrasts like this one, which are unwritten, Dholuo also has unwritten tonal contrasts. For example, <kor> /ḳór/, which means ‘a prophecy’, and < kor> /kor/, which means ‘a narrow path’, are spelled the same.

Comprehension

Gough and Tunmer's (1986) model simplifies reading comprehension as the product of a child's decoding and listening comprehension skills. Reading comprehension research indicates that readers construct meaning by combining the information in the text with what they already know (Piper, Zuilkowski, & Mugenda, 2014; Samuels, 2006). Snow (2002) calls reading comprehension ‘the process of simultaneously extracting and constructing meaning through interaction and involvement with written language’ (p. xiii). Comprehension includes the ability to use prior knowledge to derive meaning from what is read (Hudson, 2007). Literature that reflects the linguistic and social contexts with which the readers are familiar will help them develop comprehension skills from the moment they enter school (Schroeder, 2007).

Regardless of first‐language reading skills, failure to grasp the oral syntactic and morphological features of the second language impedes text comprehension, although this can be overcome by high‐quality instruction (Lesaux, Rupp, & Siegel, 2007; Lesaux & Siegel, 2003).

Relating fluency skills to comprehension skills

Discussion of reading skills often refers to lower‐level (decoding or word‐attack) skills and higher‐level (comprehension) skills. This is perhaps a simplistic description of the relationship between the skills; Stecker, Roser, and Martinez (1998) noted that ‘fluency has been shown to have a “reciprocal relationship” with comprehension, with each fostering the other’ (p. 306). The view that comprehension is dependent on fluency also assumes that the reader has oral language skills in the language read. These readers often use word‐decoding skills, and, as they become more accomplished readers, rely more on visual recognition (Just & Carpenter, 1987).

However, when learners in a second language are already literate in their first language, they use their knowledge of the orthographic and syntactic features of both languages to make sense of second‐language text (Hudson, 2007). Where first‐language readers are concerned, accuracy and speed of word‐decoding skill are strong predictors of comprehension (Hudson, 2007). However, when the learners' reading fluency is assessed in a language in which they lack oral skills and vocabulary, as is likely in a multilingual setting such as Kenya, the relationship between fluency and comprehension documented with English‐speaking monolinguals may differ (Gough & Tunmer, 1986).

Reading and the multilingual environment

Understanding effective reading instruction in multilingual primary classrooms requires attention to context. Children are not usually fluent speakers of their first language by the time they begin primary school, given that mature speech patterns are not present until age 10 (Baker, 2006; Cummins, 1984; Thomas & Collier, 2002). Introducing a second language as a medium of instruction before children master the first language can cause delay (Baker, 2006). Multilingual education is most likely to succeed (a) in settings in which children have the chance to fully develop the first language (Cenoz & Genesee, 1998) and (b) when children are specifically taught how reading in one language is similar to or different from the second language (Lau & Liow, 2005; Nikolopoulos & Goulandris, 2000; Patel, Snowling, & de Jong, 2004; Scholfield & Chwo, 2005; Seymour, 2006).

Transferring reading skills

After a learner is a proficient reader in a given language, several reading skills readily transfer to a second language, including the knowledge that print carries meaning and is related to speech. Koda (2008) noted that the transfer of literacy ability from one language to another may occur after the reader grasps three universals as follows: (a) print relates to speech; (b) speech can be segmented into a sequence of sounds; and (c) these sounds relate systematically to the graphic symbols in the writing system.

Non‐transferrable reading skills are the mapping details (e.g. particular sound–symbol correspondences and grammatical features; see Koda, 2008) in the first language. The ability to transfer reading ability from one language to another is affected by the linguistic similarity of the two languages (Cenoz & Genesee, 1998). As noted earlier, two languages in our study – Kiswahili and Gikuyu – are related members of the Bantu language family; Dholuo is Nilo‐Saharan and English Indo‐European.

Successful reading acquisition in a second language depends on phonological awareness, decoding skills and oral reading fluency in the first language, along with oral fluency in the second language (Gove & Cvelich, 2011; Hudson, 2007). The goals of formal education in Kenya revolve around developing reading fluency and comprehension in English, which is either a second or a foreign language for most students, as described previously. If the goal is reading fluency and comprehension in both the first and subsequent languages, then sufficient time must be given to developing oral and reading skills across languages. Even if the goal is solely second‐language fluency (Kamwangamalu, 2010), attention to first‐language reading skills – and second‐language oral fluency – needs to be central to instruction. However, empirical research from countries like Kenya on learning outcomes across languages is lacking.

Language use in Kenyan schools

The unique language‐of‐instruction data in this article give a clear picture of language usage in Kenyan primary schools. Based on the documented observations, in early‐grade classrooms, the students' mother tongue was used only 14.1% of the time (Piper & Miksic, 2011). English was used more frequently than Kiswahili and the mother tongue in both urban and rural locations. Use of the mother tongue was primarily limited to the subject called ‘mother tongue’ and was not the medium of instruction in the content areas of social studies, mathematics, life skills or science.

Piper and Miksic (2011) found statistically significant relationships between English language use and English achievement and Gikuyu language use and Gikuyu achievement. These relationships were relatively weak in magnitude, explaining only 6.9% and 7.2% of the variation in English and Gikuyu oral reading fluency outcomes, respectively. On the other hand, no statistically significant relationship was found between the percentage of classroom Kiswahili language use and students' Kiswahili oral reading fluency. Similarly, there was no relationship between Dholuo language use and fluency. Of course, these relatively simple analyses cannot estimate the causal processes responsible for fluency and comprehension, and more analysis is necessary to understand how language choice overlaps with learning outcomes. Research on bilinguals' oral comprehension skills in their first and second languages is in its early stages. In addition, the transferability of these skills is affected by the unique characteristics of the individual languages involved (Gathercole, 2013).

Given the lack of research in Kenya that empirically investigated the relationships between oral reading fluency and comprehension in English, Kiswahili and mother tongue, we were interested in answering the following research questions.

1. Do children read more fluently in English, Kiswahili or mother tongue? Does this differ by urban or rural location?
2. Do children comprehend better in English, Kiswahili or mother tongue? Does this differ by urban or rural location?
3. If oral reading fluency scores are similar, will children comprehend better in English, Kiswahili or mother tongue?
4. Do oral reading fluency levels and reading comprehension scores in one language predict oral reading fluency and reading comprehension scores in another language?

Sample

The data analysed in this paper were initially collected as part of a language‐of‐instruction and student literacy achievement study undertaken by RTI International with funding from the William and Flora Hewlett Foundation and technically supported by the National Assessment Centre. The sampling framework utilised in this study was a three‐stage stratified random sample. Within the Central Province and Dholuo‐speaking portions of Nyanza Province, districts were randomly selected after stratifying by urban/rural status (Table 1). Schools were randomly selected within districts using proportional‐to‐population sampling methods. At the school level, Standard 3 children were sampled – stratified by gender – using systematic random sampling. The assessors identified the mother‐tongue status of children by asking the children what language they spoke at home to their mother (or other female guardian). The age range of these Standard 3 children was between 6 and 15, with 83.2% of the children between 8 and 10 years old. Table 2 describes the sample.

Specific language assessments

Language was a key variable in the research design. In schools in urban Central Province (Nyeri County) and urban Nyanza Province (Kisumu County), children were assessed in both Kiswahili and English. In schools in rural Central Province and rural Dholuo‐speaking Nyanza Province, children were assessed in English, Kiswahili and their mother tongue: Gikuyu for Central Province and Dholuo for Nyanza. These multiple‐language assessments allowed us to perform within‐child analyses investigating whether children who could decode words in one language could also decode in another and whether children who could comprehend in one language could comprehend in another.

Research instruments

The EGRA is an assessment tool used in more than 70 countries and 100 languages between 2007 and 2014 (A. M. Mulcahy‐Dunn, personal communication, 17 October 2014). It is a criterion‐referenced instrument that investigates key early reading skills. The EGRA for this study was developed in Kenya, utilising language experts from the universities, the MoEST and the Kenya Institute of Curriculum Development (KICD). Draft tools were evaluated by a technical panel including participants from the relevant MoEST directorates during an adaptation workshop in September 2009. The workshop participants ensured that the assessment was based on the KICD curriculum for Standards 1–3. The Kenyan EGRA included several subtasks, which were assessed in all languages (Piper & Miksic, 2011). The tasks analysed included the following:

1. Oral reading fluency – the ability to read a passage of about 60 words. Measured by cwpm. Figures 1 and 2 are sample excerpts from the instruments for two of the four languages: the assessment protocols for English and Dholuo connected text oral reading fluency.
2. Reading comprehension in connected text – the ability to orally answer several comprehension questions about a passage. Measured by (a) the percentage correct of five comprehension questions and (b) the percentage correct of the five comprehension questions attempted. Again as a sample, Figures 1 and 2 provide comprehension questions for English and Dholuo.

Prior to tool finalisation, a pilot survey was undertaken with 100 children in each of the four languages. A Rasch model was fit to examine item behaviour, and the tools were revised prior to data collection. In the full assessment, the tools performed quite well. Piper (2010) showed that the Cronbach's alpha reliability statistics for the entire English (0.89), Kiswahili (0.90), Gikuyu (0.93) and Dholuo (0.88) assessments were above acceptable levels. The assessments were administered by trained assessors (with interrater reliability scores above 0.94; Piper, 2010). The assessor sat in a quiet location with the child and administered one language assessment, before sending the pupil to another trained assessor to administer the EGRA in another language.

Data analysis

In order to answer the research questions in this paper, we undertook general linear hypothesis (GLH) tests and fit ordinary least squares (OLS) regression models in the Stata statistical software package. In order to ensure external validity to Central Province and Dholuo‐speaking portions of Nyanza Province, we used the svyset weighting procedure in Stata to account for the sampling of children, schools and districts. Analyses in this study were undertaken on weighted rather than raw data to account for the sampling design. That is, the svyset procedure in Stata allows for the data to be weighted and for standard errors to account for the nested nature of education in Kenya. In Kenya, pupils are nested with schools, and schools are nested within districts. The standard errors account for the similarities of students within those schools and schools within those districts. Our regression modelling was carried out in the svy set of commands in Stata, with fluency or comprehension scores as the dependent variable. The methods used were simple. To answer Research Question 1, we undertook GLH tests to compare whether mean scores were statistically significantly different between languages. To answer Research Question 2, we compared R², the variance explained by particular relationships across models. The R² was calculated after OLS regressions were fit on the weighted data.

For each language, the pupils were asked five comprehension questions that matched with approximately 12 words of the story they had just read. The questions were designed so that each language had two explicit, two textually implicit and one implicit question (refer to Figures 1 and 2). All questions were answered orally.³

Summary of results

Given the dominance of English as the language of instruction in Central and Dholuo‐speaking Nyanza schools, it is not surprising that these children read English words somewhat more readily than words of the other languages (Piper & Miksic, 2011). Children's oral reading fluency scores in their first language were more predictive of reading comprehension than were those in their second language. However, these children's reading comprehension scores were significantly higher in the mother tongue than in English. In urban Central, children's comprehension was 61.9% higher in Kiswahili than in English; in urban Nyanza, comprehension was 141.5% better in Kiswahili than in English. Children understood what they read in Kiswahili 292.7% better than in English in rural Central and 193.3% better in rural Dholuo‐speaking Nyanza. The mother tongue had an even larger difference, with children understanding their mother tongue 465.6% better than in English in rural Central and 748.5% better in rural Nyanza, even though they read significantly less fluently in their mother tongues. In addition, children's oral reading fluency scores in their first language were found to be more predictive of reading comprehension than their oral reading fluency scores in a second language.

Discussion

The results show that although these children could recognise English words readily, their understanding of English remained quite limited. The comprehension questions on the English assessment were very simple, yet children struggled to decipher meaning from what they had read in order to answer them. On the other hand, children could understand much more of what they read in their mother tongue, although their skills in decoding words in the mother tongue were weaker than they were in English.

These findings call into question earlier perspectives that have ignored the multilingual nature of reading instruction in Kenya, and they show that the relationship between oral reading fluency and reading comprehension depends on children having the requisite vocabulary in the language being used. It appears that after three years of predominantly English instruction, children are able to recognise common words and have very modest decoding skills, but lack the English vocabulary necessary to understand the meaning of those words.

Our findings confirmed that the English fluency of these primary‐school children was significantly weaker than their fluency in their mother tongue. The patterns were consistent: If oral reading fluency rates were held constant, Kiswahili comprehension rates would be 48.1% higher than English in urban Central and 47.9% higher in urban Nyanza. For rural Central, the comprehension rates at similar fluency rates would be 288.6% higher for mother tongue than for English and 185.5% higher in rural Nyanza. These children simply understood better in their mother tongue, even if their oral fluency rates in English were higher; the basic reading comprehension levels expected of those tested by the EGRA were quite easily attained in the mother tongue, even with minimal instructional time. The implication is that using English as the medium of instruction is extremely inefficient. The study indicates that children's decoding abilities in their mother tongue correlated with their decoding ability in English. However, reading comprehension scores in mother tongue and Kiswahili were not as predictive of English scores. Urban children's Kiswahili comprehension predicted only 15.5% of the variation in Central children's English scores, and 28.6% of the variation for Nyanza children. For mother tongue in rural schools, the relationship was even weaker, with Gikuyu predicting only 5.5% of the variation in English comprehension and Dholuo predicting only 11.2% of the variation in English comprehension scores in rural Nyanza.

These findings show that many children were able to transfer the ability to identify words across languages. These findings held when other multiple regression models (not included in this article because of space constraints) were fit that controlled for the percentage of time each language was used. Therefore, even with limited instructional time in the variety of languages in multilingual Kenya, children had learned to decode basic words in other languages. On the other hand, children were unable to transfer comprehension skills across languages. Kenya is ‘buying’ English fluency by spending more than 60% of classroom time in English. This increased instructional time helps the children unlock the orthography challenges of English and gain some basic fluency. However, they are neither gaining proficiency in spoken English nor mastering English academic vocabulary. Thus, even if reading fluency can be bought, comprehension cannot.

We think that two key elements are likely to be heavily correlated with children's reading comprehension: oral vocabulary (Piper & Mugenda, 2014) and sufficient instructional time in mother tongue (Piper & Mugenda, 2012, 2013). Without oral vocabulary, children are unlikely to comprehend what they read, even if they are taught primarily in English. Unfortunately, the data used in this analysis did not include oral vocabulary measures. Without sufficient instructional time in reading, pupils do not have the time to practise these skills to develop them fluently.

The predictive power of the relationships between language outcomes was higher when the languages were more similar. In other words, Kiswahili reading outcomes were a better predictor for children in Central than they were for children in Nyanza. Conversely, English reading outcomes were a better predictor for children in Nyanza than for children in Central. This suggests that even in linguistically complex environments, reading skills transfer more easily across languages that are linguistically similar.

Conclusion

This analysis of language, oral reading fluency and comprehension in two regions of Kenya confirms observations that have been made by educators over a number of years about the shortcomings of the current practice of using English as the primary medium of instruction throughout the formal education system. Despite the national policy that supports the use of local languages as languages of instruction in Grades 1–3 (see, for example, Bunyi, 2005, 2008; Kioko, 2013; Trudell & Piper, 2013; Trudell & Schroeder, 2007), the data indicate that local choices about language of instruction are directly and consistently contravening national language and education policy. The result is lower student achievement.

The inconsistent policy implementation found in the schools studied can be readily seen in primary classrooms elsewhere in Kenya as well. In part, it is related to the tension between cultural and linguistic realities and perceptions of what education is for (Trudell & Piper, 2013). Powerful ideas regarding the centrality of formal, international language‐medium education in economic development (Chaudenson, 2008; Djité, 2008; Piper & Mugenda, 2014; Piper, Zuilkowski, & Mugenda, 2014; Trudell & Piper, 2013; Trudell, Young, & Nyaga, in press), as well as a keen national‐level desire for inclusion in the international (largely Anglophone) community of nations (Rizvi & Lingard, 2010), continue to influence local and national decisions concerning language of instruction in the classroom.

The evidence is that a single‐minded focus on English in Kenyan primary school education is not yielding strong learning outcomes. While prioritising English as the language of instruction has indeed resulted in children who are better at pronouncing English words than words in any other language, their mastery of the English language is inadequate for them to understand what they are reading. The data suggest that 3 years of using English as the predominant language of instruction can impart basic skills in decoding and recognising words, but not the level of English language mastery necessary to understand the meaning of those words. In contrast, it was found that basic reading comprehension levels are quickly and easily attained in the mother tongue, even with minimal classroom attention. Greater attention to the use of the mother tongue as a language of instruction in the classrooms studied would likely have yielded even stronger reading comprehension levels for these same children.

A weakness of the current study is that there was no measure of vocabulary, either expressive or receptive, and therefore, a full examination of whether the simple view of reading is appropriate in this multilingual context is not possible. Further research should more deeply examine the relationship between vocabulary skills and listening comprehension, as well as the interaction between vocabulary skills and fluency in producing reading comprehension in multilingual contexts such as Kenya. Nevertheless, an important recommendation from this study – even without the inclusion of vocabulary measures in the current analysis – is that Kenya should ensure that children have rich oral language experiences in English to combine with their skills in decoding English words.

This study suggests that the Kenyan education system is trying to ‘buy’ strong English‐language reading outcomes through an emphasis on English‐medium instruction; but what is actually being produced are children with limited fluency in either oral or written English and minimal reading skills in their own languages as well. Hence, even if a certain level of English oral reading fluency can be ‘bought’ in this way, English language comprehension cannot.