Matthew Effects in Reading

If you only read one paper on the theory of reading disability, find the one by Dr Keith Stanovich called “Matthew effects in reading: Some consequences of individual differences in the acquisition of literacy”.

Its title comes from a paraphrase of the Gospel of Matthew: “The rich get richer and the poor get poorer”. In this 1986 paper, Stanovitch described how early success in reading can lead to dramatic advantage over time. He started with the premise that readers build the skills of reading BY reading. Better skills make reading more pleasurable, encouraging the strong reader to read more. More reading helps the strong reader develop better skills. More reading makes reading more pleasurable, encouraging even more reading. It is a virtuous feedback loop.

By contrast, early struggle in reading often leads to catastrophic failure. Students who struggle with reading tend to avoid reading and get less practice. Lack of practice slows their acquisition of key reading skills such as vocabulary, fluency, comprehension strategies, and world knowledge. As they get older, the texts they encounter become harder and more opaque, and lack of component skills makes reading increasingly laborious and unpleasant. So poor readers increasingly avoid reading and fall farther behind. It’s the same feedback loop in reverse.

The usual interpretation of the Matthew Effect is that a small variation in initial success causes a vast chasm to open between strong readers and weak ones, and locks the weak reader into a permanent disadvantage. Once a weak reader has fallen substantially behind, catching up requires that he must read MORE than the strong reader. Yet it is physically impossible for a student who painfully labours with every sentence to read more than a fluent reader who seeks out books independently for pleasure.

Conceptual Framework

Of course this can’t be taken literally. Children follow every imaginable path to both strong reading and reading disability. But Stanovich was trying to build a conceptual framework of reading development, a model that could bring order to conflicting mountains of experimental data.

To be useful, a model must be able to make predictions. Newton’s three laws of motion are the most famous example of this kind of model. They describe a limited set of operations in a frictionless ‘billiard ball’ world. They don’t explain ‘how’, but they give an accurate prediction when we ask a suitable question. We know them to be incomplete since they do not account for thermodynamics, relativity or quantum effects. But Newton brought order to chaos, and his simple laws are the foundation on which engineers have built the modern world.

Stanovich was doing the same for reading, although he might blush to be compared to Newton. He was facing a mountain of conflicting and confusing experimental studies; researchers had measured correlations and differences in every conceivable cognitive skill between normal and dyslexic readers, young and older readers, wealthy and poor readers, and emerging readers developing skills over time. But it was like blind men examining an elephant and announcing their contradictory evidence. Worse, because readers might achieve comparable reading skills with different profiles of foundation skills, researchers hypothesized different causes for the same outcome. Stanovich set out to bring order to this chaos.

First, a word about correlation and causation might be in order. A correlation is just a relationship between two or more things in a population, such as age and vocabulary among elementary students. If they are correlated to some degree, then we say that one ‘predicts’ the other. But it is sometimes not clear which direction the causation works, or if the cause of the link is a third factor. Older students tend to have larger vocabularies, so age and vocabulary are correlated. It is obvious that vocabulary does not ‘cause’ age, but also unlikely that students acquire vocabulary just from getting older.

That particular correlation isn’t very interesting. But how to explain this one: readers with better comprehension skills tend to have larger vocabularies. So is it that having a bigger vocabulary helps you understand stories better, or that understanding stories better that helps you learn more words? Or is it some third factor that connects them?

Stanovich set out a framework of “reciprocal relationships” where the foundation skills of reading bootstrap each other over time. Unlike a causal relationship where one side is the cause and the other is the effect, reading seems to be a network of mutually reinforcing skills.

In Stanovich’s model, larger vocabulary helps you understand stories better AND better comprehension returns the favor by helping you acquire vocabulary. Both drive you to read faster, and faster reading in turn accelerates your vocabulary growth and comprehension. And so with every other component skill, driving cycles of interacting advantages.

Conversely, a poor vocabulary makes it harder to understand stories, and poor comprehension slows your vocabulary growth, and both cause you to be a slower reader, Across the multiple component skills of reading, this reciprocal feedback drives cycles of interacting deficits – voila, the Matthew Effect.

At first you might think that this reciprocal causation model doubles the number of causal relationships and complicates the understanding of reading. But it makes understanding reading very simple because most of the paired correlations that researchers rush into publication are no longer important. It is true that certain types of eye movements are correlated with reading fluency, but you can throw out all the research that tries to link them directly or diagnose inefficient eye movements as the cause of reading failure. Strong readers develop efficient eye movements, and the rest follows.

The Chasm Opens

There are sweeping ideas in this paper. For example, if reading bootstraps other cognitive processes as we believe, then the chasm that opens up between strong and weak readers is much wider and more complex than simply whether a student will develop a love of reading. Strong readers will pick different friends and develop different interests. They will develop different attitudes about learning, and develop different levels of self-esteem and motivation, and these environmental differences and attitudes will further drive reciprocal improvements in reading.

Reading is critical to success. The SAT scores used by college and university admissions are frequently criticized as ‘just’ reading tests in disguise; a typical question presents a short text and then asks the student to make inferences about it. But universities require them because SAT scores correlate strongly with student’s success. The Matthew Effect explains why a strong reader is more likely to succeed at university.

But a disturbing consequence of the Matthew Effect is that poor reading skills can quickly snowball into something much worse. Stanovich charts how a specific reading deficit will predictably cause behavioral, cognitive, and motivational issues, and may spiral into a generalized learning disability. That seems to happen in practice. The Matthew Effect is invoked in a famous lawsuit in North Carolina over a school’s failure to provide suitable reading interventions (here), the sad details will sound familiar to every parent who has struggled to get help for their child. But note how the child’s IQ steadily declined over the years, from 127 to 102 from lack of reading.

This is the opposite of the “Gift of Dyslexia” premise, where poor reading is just a single weakness in a sea of compensating strengths. Instead, poor reading is an open drain that relentlessly weakens the development of other cognitive skills.

You can watch Dr Stanovitch discuss the Matthew Effect and its impact on cognitive skills in this six-minute video (here).

Attacking and Fixing Dyslexia

But the really interesting part of this paper is frequently overlooked. Towards the end of his paper, Stanovich looked at the problem of breaking the negative feedback cycles in poor readers. The Matthew Effect makes it clear that this is an extraordinarily difficult task, but gives insight into how it might be done.

If Dyslexia can be fixed, it is because Stanovich’s model doesn’t require any magical explanation of reading failure such as developmental delays, brain wiring or genetic misfortune. The Matthew Effect predicts the difficulties that a dyslexic reader will face, without requiring any cognitive difference or disorder. Stanovich takes a few paragraphs to dismiss the existence of a specific reading disability, but the reader will have already noticed that the entire paper dismisses traditional ‘dyslexia’ as an unnecessary and superfluous explanation for reading failure.

A severe reading deficit is just a compounding failure to develop component skills. Stanovich proposed to break the cascade of negative feedbacks by delivering an educational ‘surgical strike’ on one of the weak component skills, and then use it to start a positive bootstrap to other skills. He suggested the most promising target was the ability to blend and segment phonemes (‘phonological processing’).

This is exactly what we do at the Community Reading Project for students that come to us with weak phonological skills. It takes about two to three weeks of daily practice using our free Blending program to clear that roadblock. Then we jump in with fingerpoint reading of easy chapter books on top of those new phonological skills to unlock word recognition skills. Then we systematically attack comprehension, fluency, spelling, writing, and other component skills.

It sounds complicated, time-consuming, and hard. It is time-consuming, there is no short-cut. But it’s not complicated, not hard, and you can do it with your struggling reader. We’ll show you how.

Poor Research Techniques in Dyslexia

In another part of this sweeping paper, Stanovich looked at ‘Reading-Level Match’ research. These studies compare cognitive skills between, say, 6-year-old normal readers and 10-year-old dyslexics matched at the same reading levels. The idea is that if the older poor reader underperforms his age-peers on a skill such as vocabulary then that weakness could be attributed to lack of reading practice. But if he underperforms a younger child who obviously hasn’t had much practice, then poor vocabulary is evidently a cause of his poor reading. The Matthew Effect invalidates this kind of research, and any similar kind of research involving lags on a predetermined developmental path.

Yet these flawed studies continued to grow in popularity, and there are thousands of them. Dyslexics are claimed to have impaired oral comprehension in the presence of noise (here), specific language impairments (here), impaired phonological and visuospatial memory (here), and a host of similar correlations. Or not, depending on how the study was performed.  The studies are all likely ‘true’, but worthless in the sunlight of Stanovich’s framework.

Keith Stanovich was a beacon of originality and clarity in the field of reading research, and this paper should have started a revolution. But it did not. Reading research remained a persistent muddle of murky thinking, flawed statistics, and fanciful theories, and the Matthew Effect framework is ignored.

By the late 90’s Stanovich was obviously frustrated, and writing cranky editorials about how research should inform reading instruction, and sadly by 2000 he had moved on to work in the field of rational thinking. You can read a bit about his new interests in this accessible New York Times article (here).