The dyslexia interventions evidence, sorted.

Two shelves in the dyslexia shop

Walk the exhibit hall of any education conference and you can buy two kinds of help for a child who cannot read. On one shelf: structured teaching programs — explicit work on speech sounds, letters and words, drilled daily. On the other: tinted overlays, coloured lenses, eye-tracking exercises, balance boards. Both shelves come with testimonials. Both take a school’s money and, more preciously, a struggling child’s time.

The difference between the shelves is not packaging. It is that one of them has survived randomized controlled trials and one of them has not. Reading science is often accused of moving slowly. On this question it has delivered something rare. A pooled analysis of 22 randomized trials points one way (Galuschka et al., 2014). A masked trial of the most popular alternative therapy found nothing (Ritchie, Della Sala & McIntosh, 2011). And the pediatric professions have stated jointly, in writing, that vision-based treatments lack scientific support (American Academy of Pediatrics et al., 2009).

This article lays out that evidence for teachers, school leaders and parents: what dyslexia is, why code-focused teaching works, how far it can go, and why the eye-based industry keeps failing its exams.

What dyslexia is — and isn’t

Dyslexia is a specific and persistent difficulty in learning to read and spell words, out of line with a child’s other abilities and their opportunities to learn. Decades of work converge on its core: a deficit in phonological processing — the brain’s handling of speech sounds — which makes it hard to connect sounds to letters and to blend them into words (Snowling, 2013). The problem sits in language, not vision. Children with dyslexia see print as clearly as anyone. What resists them is cracking the code that links what words look like to what they sound like.

Three facts about the condition matter for every decision that follows. It is dimensional: reading difficulty shades from mild to severe with no natural cutoff, so diagnosis is a line drawn on a continuum. It runs in families: a child with an affected parent carries something like a fourfold elevated risk, which makes family history a cheap early-warning signal (Snowling, 2013). And it is persistent but treatable: the deficit does not vanish with time or motivation, yet the right teaching reliably improves the skill it disrupts.

That definition already sorts the two shelves. If the deficit is phonological, help should target sounds and their mapping to print. Interventions aimed at the eyes are aimed at machinery that was never broken (Snowling, 2013).

The causal chain, tested

The phonological account is not a fashion; it was built on prediction and experiment. In 1983, Lynette Bradley and Peter Bryant reported in Nature that preschool children’s ability to categorize sounds — to hear that hat, cat and rat share an ending — predicted their reading years later, ahead of intelligence. Crucially, they embedded an experiment: children trained on sound categorization improved in reading, and the group whose sound training was connected to plastic letters improved most (Bradley & Bryant, 1983). Sound awareness was not just correlated with reading. Teaching it changed reading.

A decade later, Peter Hatcher, Charles Hulme and Andrew Ellis sharpened the recipe with struggling seven-year-olds. Four groups: reading practice alone, phonological training alone, the two explicitly linked, and controls. The linked group made the most reading progress — more than either ingredient alone (Hatcher, Hulme & Ellis, 1994). They called it the phonological linkage hypothesis: sound work pays off when it is welded to letters and real text, not delivered as ear training in isolation.

Those two studies fixed the template that every effective program since has followed: explicit, systematic work on phoneme awareness and letter-sound mapping, tied immediately to reading and writing words. The pooled trials would later confirm the template at scale (Galuschka et al., 2014).

The verdict of the pooled trials

By 2014, enough randomized controlled trials existed for a proper reckoning. Katharina Galuschka and colleagues pooled 22 of them — 49 treatment-control comparisons — covering children and adolescents with reading disabilities. The pool spanned every major treatment family: phonics instruction, phonemic awareness training, reading fluency training, auditory training, medical treatment and coloured overlays (Galuschka et al., 2014).

One approach passed. Phonics instruction — explicit teaching of letter-sound relationships and their use in decoding — was the most frequently tested approach. It was also the only one whose pooled effect on reading and spelling reached statistical significance, at g′ ≈ 0.32 (Galuschka et al., 2014). The other approaches produced small, unstable estimates from too few trials to trust. Some may yet prove useful in some dose; on the evidence assembled, none had earned a recommendation, and overlays and medical treatments sat among the weakest.

Two readings of that 0.32 are worth holding at once. It is moderate, not miraculous — a third of a standard deviation, won through months of structured work. And it is the only number on the board that clears the bar. In a field thick with claims, one treatment family has replicated, pooled evidence behind it. Buying anything else first is not a judgment call. It is a bet against the trials.

How far intensive help can go

Meta-analytic averages blur what an all-in intervention looks like. Joseph Torgesen and colleagues supplied the landmark case. Sixty children aged eight to ten, with severe reading disabilities, received 67.5 hours of one-to-one teaching in one of two contrasting phonemic programs. The dose ran at two 50-minute sessions a day for roughly eight weeks. The children were then followed for two years (Torgesen et al., 2001).

The results reset expectations for what “treatment-resistant” children could do. Word-attack skills — decoding unfamiliar words — jumped from the bottom of the distribution into the average range, and the gains held at one- and two-year follow-ups. Roughly 40% of the children no longer needed special education services during the follow-up period (Torgesen et al., 2001). These were children for whom ordinary remediation had failed.

Two details travel beyond the study. First, the two programs — philosophically quite different in method — produced equivalent gains. What they shared was the active core: explicit phonemic work, linked to print, at high intensity. Brand mattered less than dose and content. Second, not everything normalized. Reading rate lagged behind accuracy; children decoded far better but many still read slowly, because years of avoided reading leave a practice debt that eight weeks cannot repay (Torgesen et al., 2001). Accurate first, fluent later — and only with mileage.

The overlay industry meets its trial

Now the other shelf. Since the 1980s, a persistent industry has sold the idea that many reading difficulties stem from “visual stress” or “scotopic sensitivity” — and that coloured filters fix them. Tinted overlays laid on the page, or precision-tinted lenses worn as glasses, are prescribed after in-house assessments, often at a cost of hundreds per child. Testimonials abound. Schools stock overlay packs the way they stock plasters.

The claim is testable, and in 2011 it was tested properly. Stuart Ritchie, Sergio Della Sala and Robert McIntosh ran a masked randomized study with children who had reading difficulties, including the subgroup formally meeting “Irlen syndrome” criteria after assessment by certified screeners. Each child read with their prescribed overlay, with a differently coloured control overlay, and with none — under conditions where neither child nor tester could steer the result (Ritchie, Della Sala & McIntosh, 2011).

The prescribed colour did nothing reliable. Reading rate with the “correct” overlay was statistically indistinguishable from the control colour and from no overlay at all — including for the children the method itself had certified as needing it (Ritchie, Della Sala & McIntosh, 2011). Remove the masking and the effect returns; that is how testimonial evidence is made. A theory whose benefits appear only when everyone knows the answer is describing expectation, not vision.

The pediatricians draw a line

Medicine noticed the pattern before most school systems did. In 2009, the American Academy of Pediatrics — jointly with the pediatric ophthalmology and orthoptic professional bodies — published a statement on learning disabilities, dyslexia and vision. Its conclusions were direct. Vision problems are not the cause of dyslexia. Eye exercises, behavioural vision therapy and special tinted filters or lenses are not supported by scientific evidence. Children with suspected reading disability should get educational evaluation and evidence-based teaching, with eye care reserved for actual eye disease (American Academy of Pediatrics et al., 2009).

Two years later the same coalition published an expanded technical report reviewing the claims study by study, and reaffirmed the line (Handler & Fierson, 2011). The reasoning bears repeating for every school office that fields a therapy brochure. Reading is a language act; the visual system of children with dyslexia performs normally on examination; and controlled studies of vision-based therapies fail to show reading gains. Meanwhile the therapies cost real money and — the report’s sharpest point — real time, during exactly the years when structured teaching pays most.

What the evidence doesn’t show

The trials sort the shelves decisively, but they do not license every strong claim made on the winning side. The honest limits:

• A cure. The pooled phonics effect is a moderate g′ ≈ 0.32; most children improve, few close the entire gap, and spelling and comprehension gains typically trail decoding (Galuschka et al., 2014).
• One winning brand. Torgesen’s two philosophically opposed programs produced equivalent gains; the shared core — phonemic content, print linkage, intensity — did the work, not the trademark (Torgesen et al., 2001).
• Fluency for free. Even after successful intervention, reading rate lags accuracy; the practice debt from years of avoided reading repays only through volume of reading (Torgesen et al., 2001).
• Overlay benefit for a subgroup. The masked trial found no reliable gain even among children certified as having “Irlen syndrome”; subgroup rescue claims rest on unmasked testing (Ritchie, Della Sala & McIntosh, 2011).
• A settled screening recipe. Family risk and early sound-categorization measures predict real risk, but which screens, at what age, at what false-positive cost, remains under study; response to good teaching is the most defensible filter (Snowling, 2013).
• A window that slams at eight. The pooled trials include adolescents, and effects did not vanish with age — later help works; it simply costs more per unit of gain than early help (Galuschka et al., 2014).

Where the evidence stops

1. 1A cure
2. 2One winning brand
3. 3Fluency for free
4. 4Overlay benefit for a subgroup
5. 5A settled screening recipe
6. 6A window that slams at eight

© 2026 FUTURE PROOF™

Choosing help by the evidence

For a school leader, a special-needs coordinator or a parent holding three brochures, the literature reduces to six rules.

Buy teaching that targets the code. The only approach with significant pooled evidence teaches letter-sound relationships explicitly and uses them in reading and spelling (Galuschka et al., 2014). If a program’s materials never show a child mapping sounds to print, it is not that approach.

Weld sound work to letters and text. Phonological training pays when explicitly linked to print — the linkage finding is thirty years old and still decisive (Hatcher, Hulme & Ellis, 1994). Ear-training apps and listening games, alone, underperform the same minutes spent on linked work.

Buy dose, not brand. Contrasting programs with the same active core produce the same gains (Torgesen et al., 2001). The questions that matter: how many minutes a week, in how small a group, with what phonemic content — not which trademark is on the folder.

Screen early, by response. Family history and preschool sound-categorization flag risk years before failure (Bradley & Bryant, 1983); the cleanest identification is watching how a child responds to good first teaching, so struggling readers surface in year one (Snowling, 2013).

Route eye claims to the statement. When the overlay brochure arrives, the reply exists in writing from the pediatric bodies: no scientific evidence, eyes are not the problem, spend the time on teaching (American Academy of Pediatrics et al., 2009).

Protect reading volume afterwards. Accuracy is the intervention’s gift; fluency is repaid mileage (Torgesen et al., 2001). A child leaving intervention needs a planned on-ramp of readable text, or the rate gap quietly reopens.