Recently, I saw results of a meta-analysis that showed phonics instruction to have a much smaller effect size (.19) than many other approaches to reading instruction. Doesn’t that mean that we are overdoing phonics? If we want to improve reading comprehension it looks like it would make more sense to emphasize motivation, fluency, and inferencing than teaching phonics.
In 1986, Gough and Tunmer presented a model indicating that reading comprehension was the product of decoding ability (the ability to translate written or printed text into oral language — that is, the skills that would allow someone to read a text aloud) and language comprehension ability (listening comprehension which would allow an understanding of that oral rendition of text).
According to this so-called “simple view,” reading comprehension could be completely explained by those two sets of abilities — decoding and language comprehension.
Over time, data have accumulated supporting the key roles of both decoding and language in reading (Hoover & Tunmer, 2021; Sleeman, Everatt, Arrow, & Denston, (2022), and indicating diagnostic and pedagogical benefits to the scheme.
Nevertheless, the theory tends to break down around the edges.
Oral language and written language operate somewhat differently (Daniels & Bright, 1996) — complicating the idea that reading comprehension is no more than a listening skills applied to text. There are vocabulary words that appear often in text, but rarely in oral language (e.g., occur, peruse, enumerate, venerate). Likewise, written sentence complexity often outstrips what we confront orally. Reading is like oral language, but mostly when we are being lectured — think about the sustained attention and memory demands of listening to an extended monologue. Oral language usually tends more to dialogue, reading to monologue. Also, oral language tends to allow for interaction between speaker and listener; not so much in reading (Olson, 1994). Treating oral language development as the sole basis of reading comprehension would fall woefully short.
The simple view is not especially specific about the skills included in either of those two constellations. Does phonemic awareness belong in decoding? What about the roles of reasoning and knowledge in listening comprehension? How do I know if I’m omitting a critical part of decoding or language?
Another complaint is that the model makes it look like decoding and language are equivalent — in what it takes to learn them, in their developmental horizons, and so on (Catts, 2018). Many children, perhaps most, can gain full benefits of decoding instruction during the first two or three years of school. Who would claim this to be true of language development?
Perhaps most damning is that statistical analyses of reading can’t account for all the variation in reading ability with those two sets of variables alone (Wagner, Beal, Zirps, & Spencer, 2021). In fact, according to that rigorous analysis, the simple view only accounts for a bit more than half the reading variation – suggesting the need for additional variables or different ways of measuring the variables already identified.
In response to these limitations, Duke and Cartwright (2021) have advanced a more elaborate model of reading. Their Active View Model is more specific about what goes in those word reading and language comprehension bubbles. With their model you don’t need to guess about that. Enumerating those items complicates things, a bit, and evidentiary support for individual items is pretty uneven. Some of the variables have a great deal of research support, others not so much (as of yet, anyway).
Duke and Cartwright also have included domains not part of the simple view. For instance, their model includes an Executive Function bubble that oversees word reading and comprehension. Another new category holds variables that don’t fit neatly into either word reading or language. For example, research has found that vocabulary plays important roles both in decoding and comprehension. Two-headed abilities like that populate a “bridge variables” constellation.
Just as one can marshal evidence that both decoding and language comprehension are important parts of reading — one can provide similar evidence for the active view variables.
The study that you noted (Burns, Duke, & Cartwright, 2023) was such an attempt.
These researchers examined relevant meta-analyses reported since 2006 — a rather arbitrary cut point (and one particularly unfortunate for the decoding variables). Doing it that way ensures that the largest body of research on elementary phonics instruction (the National Reading Panel Report) would be omitted from consideration.
If this study was aimed at understanding the impacts of phonics instruction, this approach would likely have been shot down by reviewers. A major concern with meta-analysis is sampling error. Ignoring a major corpus of data without persuasive theoretical and/or methodological reasons would be unacceptable.
However, their purpose was not to be comprehensive or even to suggest the relative importance of the variables in the model. They simply wanted to demonstrate that each of the constellations was supported by some empirical evidence. If all the phonics studies were included, the overall effect size might have been a bit bigger — it certainly wouldn’t have been lower. But the absence of those data wouldn’t alter the point that the major domains included in the active view are supported by evidence; that would be also be true if the phonics effect size had turned out to be much larger.
This study accomplished its goals — it showed that the active view provides an efficient and coherent compendium of reading abilities (at least in terms of those major domains.
Perhaps this model will generate useful research or curriculum development going forward. But remember it’s just a model, and a partial one at that. This model is more complete than the simple view and does a better job of accommodating some of the knowledge about reading that has been developed over the past several decades. But it doesn’t suggest anything about how these variables fit together, how their relative importance changes with development, or many other issues relevant to reading instruction.
Other reasons not to be overly concerned about the relatively low phonics effect size in this study?:
1. The study put forth two phonics effect sizes: the one you noted for average readers, and one for striving readers. That second effect size, the one for the strivers, was .48. That put phonics in the top tier of interventions for kids who struggle with reading. That effect size is based on 32 independent studies (the .19 was based on only 8), and remember, these effects were in terms of impact on reading comprehension or overall reading achievement — not on decoding.
2. The reporting of this study poses some important challenges to scholars, since it is difficult to identify which studies contributed to these main effects estimates. Usually in a meta-analysis, the studies are chosen because they provide data about the effect of a particular variable or approach. In this case, there are 12 variables for which main effects are reported based on data from 27 meta-analyses. But there is no linkage between the studies and the outcomes. That makes it almost impossible to evaluate the appropriateness of the analyses for any of the variables.
3. An example, of the kind of further analysis that would be needed to evaluate a specific statistic like effect size for phonics instruction is posed by the Galuschka et al. (2014) meta-analysis. Galuschka combined the effects of studies that I would think of as learning trials, rather than efforts to raise general reading comprehension or overall reading achievement. Some of the phonics studies included in that meta considered instruction in which the “phonics” entailed no more than 4 half-hour lessons in which students memorized 25 two-letter syllables each day. I couldn’t figure out how any of the studies in that meta-analysis fit the purpose or selection standards of this Burns et al. study. My concerns about the inclusion of that odd meta-analysis doesn’t alter my overall estimate of the value of the Burns study, but it reveals why I wouldn’t be overly concerned about a specific effect size being higher or lower than you anticipated given that it isn’t clear which data contributed to it.
4. Another example of my concerns about the original meta-analyses that were the basis of this study is presented by the Suggate, 2016 study. My concern about that one is that it focused on long-term benefits of skills (most of the other meta-analyses were more immediate in,focus). Including long-term outcomes for some variables, but not for others presents an an unfortunate confound if the purpose was to compare variables — as it would suppress the relative impact of some variables. This is especially challenging given the odd classifications of original studies in the Suggate meta. For example, several studies conducted by Patricia Vadasy and her colleagues were classified as fluency interventions — not phonics, despite their focus on phonemic awareness, phonics, and code-oriented instruction (not fluency). This apparent misclassification may matter since these studies reported some of the biggest effect sizes in that analysis.
5. Another problem for the Burns et al. study is its failure to focus on interventions that addressed a single issue. Motivation, for example, was rarely if ever a variable on its own. A study included in the motivation set might have taught reading comprehension strategies along with some student choices for books, while the control groups received neither the strategy teaching, those books, or the chance to make choices. Attributing outcomes from such studies to motivation alone is misleading.
6. The developmental nature of reading raises additional concerns. Decoding has been identified as a skill set with a relatively low ceiling. The importance or value of phonics instruction depends upon how well students can decode. Young children are likely to benefit more from phonics than older ones. Struggling readers will usually benefit more from such teaching than average readers, especially with older students. Just comparing effect sizes across very different interventions with very different samples of students cannot provide meaningful relative estimates of importance. (This is also true for vocabulary and fluency development –— their value in supporting comprehension changes over time.)
7. Decoding is often described by scientists as a necessary but insufficient condition. That is, you can’t learn to read without learning to decode, but learning to decode will not be sufficient to make you a reader. This is like the food groups in nutrition. No nutritionist would ask, “Which food groups do we need to provide children?” They would recognize it as a trick question — to be healthy kids need all of these food groups, of course — it isn’t a competition between proteins and carbohydrates. In reading, making sure all kids reach threshold levels of decoding ability (Wang, Sabatini, O’Reilly, & Weeks, 2019) should be a non-negotiable — no matter the relative effect sizes in this kind of rough analysis.
8. The simple view is unable to account for all the variance in reading ability, which makes the identification of a more complete model a worthwhile pursuit. The active view model seems to provide greater completeness. However, this first attempt to quantify the additional power that this model provides for explaining variation in reading attainment is not convincing. The new model with its new domains and it additional variables was only able to pick up an additional 2% of variance. This 2% was statistically significant, but I am dubious as to its eduational importance. Given the problems with this analysis, I suspect the 2% added value is meaningless. That rather modest supposed added value wouldn’t convince me to treat decoding or language differently than in the past.
Basically, this study has nothing to say about the relative value of phonics instruction (or of instruction of any of the dozen variables it included).
Burns, M. K., Duke, N. K., & Cartwright, K. B. (2023). Evaluating components of the active view of reading as intervention targets: Implications for social justice. School Psychology, 38(1), 30-41. doi:https://doi.org/10.1037/spq0000519
Catts, H. W. (2018). The Simple View of Reading: Advancements and False Impressions. Remedial and Special Education, 39(5), 317–323. https://doi.org/10.1177/0741932518767563
Daniels, P. T., & Bright, W. (Eds.). 1996. The world’s writing systems. New York: Oxford University Press.
Duke, N. K., & Cartwright, K. B. (2021). The science of reading progresses: Communicating advances beyond the simple view of reading. Reading Research Quarterly, doi:https://doi.org/10.1002/rrq.411
Gough, P., & Tunmer, W. (1986). Decoding, reading, and reading disability. Remedial and Special Education, 7, 6–10.
Hoover, W. A., & Tunmer, W. E. (2021). The primacy of science in communicating advances in the science of reading. Reading Research Quarterly, doi:https://doi.org/10.1002/rrq.446
Olson, D. R. (1994). The world on paper: The conceptual and cognitive implications of writing and reading. Cambridge: Cambridge University Press.
Sleeman, M., Everatt, J., Arrow, A., & Denston, A. (2022). The identification and classification of struggling readers based on the simple view of reading. Dyslexia: An International Journal of Research and Practice, 28(3), 256-275. doi:https://doi.org/10.1002/dys.1719
Wagner, R. K., Beal, B., Zirps, F. A., & Spencer, M. (2021). A model-based meta-analytic examination of specific reading comprehension deficit: How prevalent is it and does the simple view of reading account for it? Annals of Dyslexia, 71(2), 260-281. doi:https://doi.org/10.1007/s11881-021-00232-2
Wang, Z., Sabatini, J., O’Reilly, T., & Weeks, J. (2019). Decoding and reading comprehension: A test of the decoding threshold hypothesis. Journal of Educational Psychology, 111(3), 387-401. doi:https://doi.org/10.1037/edu0000302