A defence of decodable books: Response to a recent systematic review

Well before the sparkling new Australian Curriculum 9.0 endorsed replacing predictable texts with decodable texts in early years classrooms, demand had begun to grow for decodable books. They are called ‘decodable books’ because almost all of the words are decodable for beginning readers who have just started to learn the alphabetic code. They are written specifically to provide a scaffolded way for students to practice reading words using the phonics knowledge and skills they are learning. They are closely aligned with a phonics scope and sequence so, as a child’s knowledge of phonics grows, the number and complexity of words they can decode also grows. Used properly, high quality decodable books also draw students’ attention to the meaning of what they are reading. Decodable books are best viewed as an instructional tool that is used for a limited time. Eventually, all books are decodable but not at the start of instruction.

Decodable books are never recommended as a replacement for shared reading of storybooks and picture books in all stages of learning. Shared reading means hearing the text and looking at the text as well as the pictures and other features – not just listening to it. Storybooks and picture books (and non-fiction) are important for oral language, vocabulary, comprehension, and the sheer delight of books. However, in an evidence-based approach to reading instruction, where students are always encouraged to decode unfamiliar words using phonics as a first strategy, decodable texts are recommended to replace predictable texts for reading practice. Predictable texts are designed to encourage students to use the inefficient and inaccurate ‘three-cueing method’ for reading unfamiliar words, which in practice means guessing what the word is using context and picture clues. They contain many words that beginning readers would only be able to work out by guessing.

There is a detailed comparison of decodable and predictable books, and a summary of the evidence supporting the use of decodable books, is in an earlier post that can be read here.

Since that post was published, a new research report has been released — A systematic review of decodable and levelled reading books for reading instruction in primary schools: An evaluation of quality research evidence (Birch et al., 2022). It has been shared by teachers and principals to a mixed response. It is sometimes being used as a justification not to use decodable texts in initial reading instruction.

One point to note is that the review does not distinguish between levelled reading books and predictable (sometimes called repetitive text) books. Predictable books are a specific sub-type of levelled reading book (see above). Birch et al. (2022) concur with the description of predictable books above when describing levelled books, albeit with a re-ordering of the cues from the typical configuration of ‘MSV’ (meaning, structural, visual) used in Running Records.

Levelled reading books support the theory that students learn language and subsequently learn to read through making meaning using the three-cueing system of visual (graphophonic), meaning making (semantic context), and structure (syntactic context; see for example, Clay, 1991). Students need multiple strategies to consider whether a word looks right, sounds right and makes sense (Fountas & Pinnell, 2006) (p.  20).

The key findings listed in the introduction of the report are fairly non-controversial in that they are ambiguous and can theoretically apply irrespective of your view of what constitutes effective reading instruction. However, the first clue that the authors of this report may have a view of effective reading instruction that is not based on a strong adherence to scientific reading research is on p.10, where Reading Recovery is included in the examples of “evidence-based approaches to the teaching of reading”. Even before the most recent study of Reading Recovery, which shows that it is not an effective intervention and may in fact be detrimental, there have been reams of evidence it doesn’t work for most students. On p.12, the report approvingly cites the unproven Four Resources Model of reading.

The part of the report that has drawn most attention is the authors’ conclusion that “the balanced approach and those using levelled reading books often fared better than those with an overt focus on phonics/decodables” (p. 45).

There are a few reasons to be wary of accepting this conclusion.

1.  The study results reported do not support the conclusion that the ‘balanced’ approach(ie. both levelled and decodable texts) and levelled texts had greater effects than decodables.

The authors’ own statement of their findings contradicts this conclusion.
 
“When investigated by the focus of the text used in the intervention, for the majority of outcomes, the difference in the average effect of intervention using a specific text type was marginal between the two text groups (~d ± 0.10). Of the outcomes with greater than d = 0.10 difference in the average effect and more than a single study contributing to the effect calculation, Fluency, Word Recognition and Decoding Skills outcomes were stronger among the decodable interventions” (p. 35).

In Table 2 where the effect sizes are reported, it can be clearly seen that the highest effect sizes are those associated with the use of decodable texts.

      2.    While the report is described as a systematic review, it goes a step further and tries to be a meta-analysis by combining the effect sizes from individual studies to calculate summary effect sizes. There are two problems with the statistical analysis.

a)    The summary effect sizes (Table 3) appear to have been calculated using simple averaging. This is not best practice. The simple averaging method gives the same weight to every study, irrespective of its size. Furthermore, they appear to have combined effects from studies making different comparisons, often comparing either levelled or decodable readers to some other condition, such as “business as usual”. Since the comparison conditions are different, and there appear to be few studies making direct comparisons, the effect sizes for levelled and decodable readers are highly unlikely to be comparable.

b)    At least one set of summary effect sizes in Table 3 is questionable — the effect sizes for levelled books for the outcome of ‘decoding skills’. The average summary effect sizes for these variables are reported as ES 0.50 (all) and ES 0.71 (low competence readers). The simple averages appear to have been calculated using these individual study effect sizes:

 All: Cheatham (ES 0.09), Denton (ES 0.16) and Jenkins (ES 1.25)

Low competence readers only: Denton (ES 0.16) and Jenkins (ES 1.25)

However, it is not explained why the effect size for the study by Kim (ES -0.04) is not included in these calculations. In Table 3, the Kim (2010) study is not coded for text focus and therefore excluded from the simple effect size averages, but in Table 1 the study by Kim (2010) is described as having an intervention text focus of ‘levelled’ and a sample population of low competence readers, therefore making it eligible for inclusion in both of the summary effect sizes.

 If the Kim study was included, the simple average effect sizes would be ES 0.36 (all) and ES 0.46 (low competence). Putting these revised figures into Table 3, these summary effect sizes are substantially and likely to be significantly lower than the equivalent for decodables. This would also affect the ‘grand average’ that led to the conclusions drawn by the authors regarding the overall effectiveness of levelled texts versus decodable texts. 

      3.    Birch et al. (2022) recognise that the stage of reading development in which particular texts are used is a very relevant factor.

 “At the earliest levels, it could be argued that reading books with highly, or a significant level of decodable features outperformed reading books with high frequency words” (p. 39)

 However, they do not consider the year group as a mediator in the statistical analysis.

      4.    Birch et al. (2022) recognise that the alignment between instructional approach and the texts used to support reading development is a highly relevant factor.

 “Importantly, the type of reading books used in reading instruction is impacted significantly by their integration with pedagogical approaches used by the teacher (Price-Mohr & Price, 2020)” (p. 40)

However, they do not include it among the mediators in the statistical analysis.

5.    Table 4. ‘Advantages and disadvantages of decodable and levelled reading books as identified in the systematic literature review’ purports to summarise the result of the studies included in the systematic review.

However, the relatively large section describing the ‘disadvantages of decodable books’ draws most heavily on a paper by Altweger et al. (2014) that wasn’t selected for inclusion in the systematic review. Statements from other studies seem to contradict each other, eg. “Usefulness of decodables on word recognition strategy (Mesmer, 2010)” is given as an advantage, while “Inconclusive evidence if decodable reading books improve accuracy (Mesmer, 2010)” is given as a disadvantage.

In the real world of schools, teachers often have to make decisions based on imperfect information. They need to use the findings of the best quality research available, but they also have to use deductive reasoning to fill in any gaps.

At the moment, this is what is known about decodable books and the reasons they should replace predictable (or repetitive) levelled books for initial reading instruction.

• There is a relatively small amount of research comparing the outcomes of using decodable and levelled texts. However, the existing evidence favours decodable texts for beginning and struggling readers who are still consolidating their phonic knowledge and skills.
• The evidence base that provides the rationale for using decodable texts instead of predictable texts is very strong. It is explained in detail in the earlier post but, in brief, reading unfamiliar words by phonic decoding is more efficient and accurate than using the three-cueing method. Therefore, using texts that facilitate the more accurate and efficient reading method is logical.
• The use of decodable books for reading practice does not imply that no other books are required or desirable — shared reading of a wide range of children’s literature and non-fiction is always important.

In support of their recommendation to use levelled texts, Birch et al. (2022) cite a number of articles that caution against a narrow interpretation or misrepresentation of the science of reading. One of the authors they cite is Mark Seidenberg. Interestingly, Mark Seidenberg has also explained the problems with the three-cueing method, and therefore the levelled reading systems that encourage it, in no uncertain terms.

The best “cue” to a word is the word itself. That is the great thing about alphabetic writing: the spelling of a word tells you what the word is.  B-o-o-k is the word BOOK, pronounced /bʊk/–rhymes with /tʊk/ and /lʊk/, similar in meaning to TEXT and MAGAZINE. The spelling is far more informative than strategies such as look at the picture, take a running start, skip the word and go back at the end, and other ways to “solve words”.

Word solving is the opposite of word recognition as I’ve described it. It is slow and effortful. It requires conscious awareness. Whereas the spelling decisively identifies the word, the [cueing] strategies are fallible and vary in how efficiently they guide the child to the word. Readers are meant to use “all types of information simultaneously,” but the process of remembering and combining relevant information is effortful and inefficient. (Seidenberg, 2021, para. 11-12)

Some people may be tempted to dismiss this report on the basis that it was financially supported by a publisher of levelled book series and reading programs aligned with their use. While that is worth bearing in mind, it should not be the only reason to be critical of its findings. It is possible to make an equivalent accusation of people who advocate for the use of decodable books and work for publishers of those book series. In my case, I was arguing in favour of decodable books long before I joined MultiLit. Rather, the veracity of reports should be judged on their merits. This one by Birch et al. (2022) does not stack up.

The author acknowledges valuable contributions by Nicola Bell and Mark Carter.

Image from ‘The Brave Dragons’, written by Barbara Healy and illustrated by Janet Tiitinen.