In 2018, three cognitive scientists published a paper in Psychological Science in the Public Interest titled “Ending the Reading Wars.” The lead author, Anne Castles of Macquarie University, had spent decades studying how children learn to read. Her coauthors, Kathleen Rastle of Royal Holloway and Kate Nation of Oxford, had spent comparable decades. The 51-page review covered hundreds of studies. Its central conclusion was not subtle: the question of whether systematic phonics instruction works better than whole-language or “balanced literacy” approaches had been answered, repeatedly, by converging evidence from cognitive science, classroom trials, and large-scale randomized studies. Phonics works. Whole language does not work as well, particularly for the children most at risk of reading failure. There was, the authors argued, no scientific reason for the “reading wars” to continue.

The paper landed in a US education system where, eighteen years earlier, the federally commissioned National Reading Panel had reached substantially the same conclusion after reviewing more than 100,000 studies. The 2000 panel had been explicit: systematic phonics instruction significantly improves reading achievement, especially for at-risk students. That report was on the public record. It was cited in federal policy documents. It was known to every school of education in the country.

And yet, in the eighteen years between the National Reading Panel and the Castles review, the dominant approach in US elementary schools remained some version of whole language — increasingly relabeled as “balanced literacy” — in which children were encouraged to predict words from context, look at pictures for clues, and learn to read by being immersed in real texts rather than by being systematically taught letter-sound correspondences. Teacher-training programs taught these methods. Curriculum publishers sold them. Districts adopted them. And reading scores, particularly for children from lower-income backgrounds who could not compensate through home reading environments, lagged.

This is a story about how a research consensus and a practice consensus can sit in the same room for two decades and never speak to each other. It is also a story about what finally broke the standoff — investigative journalism by Emily Hanford at APM Reports, state legislative action starting in Mississippi, and the rise of what the field now calls the “Science of Reading” movement. For strategists and product builders who routinely encounter “evidence-based” claims in education and corporate training, the reading wars are the cleanest contemporary case study in what happens when practice systematically ignores the evidence, and what it takes to close the gap.

What The Reading Wars Actually Were

To understand why the resolution mattered, you have to understand what the two sides were claiming.

The phonics view is, roughly, that English is an alphabetic writing system: letters and letter combinations map onto sounds, those sounds combine into words, and the most efficient way to teach a child to read is to teach those mappings systematically and explicitly. A child taught phonics learns that the letter combination “sh” makes a particular sound, that “ai” makes a long-a sound, that the silent-e at the end of a word lengthens the preceding vowel. The child practices decoding new words by sounding them out. The cognitive bet is that once a child can decode reliably, they can read any word they can pronounce, including words they have never seen before. Vocabulary, comprehension, and fluency build on top of that decoding foundation.

The whole-language view, which rose to prominence in the US in the 1970s and 1980s associated with figures like Kenneth Goodman, Frank Smith, and the broader “psycholinguistic” school, argued that reading is fundamentally a meaning-making process, not a decoding process. On this view, fluent reading is more like solving a puzzle than like sounding out a code: skilled readers use context, syntactic cues, prior knowledge, and partial visual information about words to predict and confirm meaning. Children should be immersed in rich, meaningful texts from the beginning. They should learn words as wholes in context, the way they learned spoken language. Explicit phonics instruction was viewed as unnecessary at best, and at worst as boring drill that turned children off reading and obscured the real cognitive task.

The whole-language view had aesthetic and pedagogical appeal. Teachers preferred reading real children’s literature to drilling letter sounds. Parents preferred their children loving books to their children completing phonics worksheets. The whole-language framing aligned with broader currents in education that valued discovery learning, child-centered classrooms, and intrinsic motivation. By the late 1980s, the State of California had formally adopted a whole-language framework. Other states followed. Teacher-preparation programs increasingly trained teachers in whole-language and, later, “balanced literacy” methods.

The empirical problem was that the whole-language view rested on an inaccurate model of how skilled reading actually works. The “skilled readers predict from context” claim, advanced most prominently by Frank Smith in the 1970s, turned out to be backwards. Decades of cognitive science research — eye-tracking studies, lexical-decision tasks, neuroimaging of reading processes — converged on the opposite finding: skilled readers do not skip over words or predict them from context. They fixate on essentially every word, and they decode them through orthographic-to-phonological mappings that have become so automated they happen unconsciously. The “guessing from context” strategy is what beginning and struggling readers do because they have not yet built the underlying decoding system. It is a sign of weakness, not strength.

This is the empirical core of the reading wars. One side had built a pedagogy on a model of skilled reading that the cognitive science had falsified. The other side’s pedagogy was supported by both the cognitive model and the classroom-trial evidence. The wars persisted not because the evidence was genuinely ambiguous, but because the practice community and the research community had stopped talking to each other.

The National Reading Panel, 2000

In 1997, the US Congress directed the National Institute of Child Health and Human Development, in consultation with the Secretary of Education, to convene a national panel to assess the effectiveness of various approaches to teaching children to read. The National Reading Panel was chaired by Donald Langenberg, then Chancellor of the University of Maryland System, and included reading researchers, educators, and a representative of parents. The panel’s mandate was to apply rigorous evidence-evaluation standards — limited to experimental and quasi-experimental studies with control groups and quantitative outcome measures — to the question of what works in reading instruction.

The panel screened more than 100,000 studies. It applied inclusion criteria that knocked most of them out, ending up with roughly 400 to 500 studies that met the methodological bar across the panel’s five focus areas: phonemic awareness, phonics, fluency, vocabulary, and comprehension. The phonics subgroup, working from a meta-analytic database of 38 high-quality experimental and quasi-experimental studies, conducted what remains the most influential meta-analysis of phonics instruction ever published.

The phonics meta-analysis was published independently in 2001 as Ehri, Nunes, Stahl, and Willows, “Systematic phonics instruction helps students learn to read: Evidence from the National Reading Panel’s meta-analysis,” in Review of Educational Research, volume 71, issue 3 (DOI: 10.3102/00346543071003393). The headline effect size for systematic phonics instruction on reading outcomes was approximately d = 0.41 — a moderate, robust, statistically significant effect. The effect was larger for younger children, larger for at-risk readers, and persisted across different program designs as long as the phonics instruction was systematic and explicit.

The panel’s final report, Teaching Children to Read: An Evidence-Based Assessment of the Scientific Research Literature on Reading and Its Implications for Reading Instruction (NIH, 2000), summarized the phonics finding in clear language: “Systematic phonics instruction produces significant benefits for students in kindergarten through 6th grade and for children having difficulty learning to read.” The report did not say that phonics was the only thing that mattered. It identified five pillars — phonemic awareness, phonics, fluency, vocabulary, comprehension — and argued that all five needed to be addressed. But on the central question of the reading wars, the report was unambiguous: systematic, explicit phonics instruction is better than non-systematic or no phonics instruction, and the evidence base supporting this is large and consistent.

The panel’s report became the formal evidentiary basis for the federal “Reading First” initiative under the No Child Left Behind Act of 2001, which provided roughly $1 billion a year through 2007 for state and local reading programs aligned with the panel’s findings. Reading First required grantees to adopt scientifically based reading programs, which in practice meant programs that included systematic phonics instruction. This was, on paper, the moment the policy environment caught up with the research.

It was not actually that moment. Reading First was bureaucratically contentious — there were Inspector General findings about conflicts of interest in the technical assistance contracts, and the program’s third-year impact evaluation in 2008 found mixed results on comprehension even as decoding outcomes improved. The political backlash against Reading First, combined with broader hostility to No Child Left Behind, created an opening for whole-language and balanced-literacy approaches to reassert themselves at the state and district level. By the early 2010s, with Reading First defunded and No Child Left Behind discredited, balanced literacy was again the dominant approach in US elementary schools. The research consensus had not changed. The practice consensus had quietly rotated back.

Castles, Rastle & Nation, 2018

The 2018 review by Anne Castles, Kathleen Rastle, and Kate Nation in Psychological Science in the Public Interest, volume 19, issue 1 (DOI: 10.1177/1529100618772271), is the most comprehensive synthesis of the reading-instruction evidence available. The journal in which it appeared is published by the Association for Psychological Science specifically to commission long-form, evidence-evaluation papers on questions of public consequence — it is the venue where the scientific community speaks to the public-policy community. The choice of venue was deliberate.

The review covers reading acquisition across the developmental arc, from the earliest stages of letter-sound learning through fluent adult reading. It synthesizes evidence from cognitive psychology, neuroscience, behavioral genetics, and intervention studies. On the phonics-versus-whole-language question, the conclusion is direct: “systematic phonics instruction is more effective than nonsystematic or no phonics instruction in helping children to develop the basic reading skills required to become a skilled reader.” The authors emphasize that this finding is not new — it had been the conclusion of the National Reading Panel eighteen years earlier — but they document that the evidence base has only grown stronger in the intervening years, with additional intervention studies, larger samples, and more sophisticated cognitive models all converging on the same finding.

The review also addresses why whole-language and balanced-literacy approaches have proven so durable in practice despite the evidence. The authors are diplomatic but clear. They note that “three-cueing” — the practice common in balanced-literacy classrooms of teaching children to identify unknown words by checking whether the guess makes sense semantically, fits syntactically, and matches some visual features — actively trains children to use the inefficient strategies that struggling readers default to, rather than building the decoding fluency that skilled readers actually rely on. Three-cueing is not, in the cognitive-science view, a less effective shortcut to skilled reading. It is a strategy that, taught systematically, can interfere with the development of skilled reading.

The Castles, Rastle & Nation paper is the document that, more than any other, made the reading-wars resolution intellectually unavoidable for anyone reading the primary literature. The 2000 National Reading Panel report had been dismissable by critics as a politically influenced exercise — there had been formal minority dissents within the panel itself, and No Child Left Behind politics colored the reception. The 2018 review came from a different scientific community, published in a different venue, written by authors with no entanglement in US policy battles, and reached the same conclusion. The empirical case was, at that point, as settled as such cases get in social science.

Emily Hanford And The Science Of Reading Movement

The Castles review was a scientific paper. It did not, on its own, change practice. What changed practice was a series of investigative reports by Emily Hanford at American Public Media (APM) Reports, beginning in 2018 and continuing through 2024.

Hanford’s seminal report, “At a Loss for Words: How a Flawed Idea Is Teaching Millions of Kids to Be Poor Readers,” was published by APM Reports in August 2019. It was followed by additional documentaries and articles, most notably “Sold a Story: How Teaching Kids to Read Went So Wrong,” a six-part podcast series released in 2022. The Hanford reporting did several things that academic publications had not been able to do.

First, it named names. Hanford’s reporting identified the specific authors, publishers, and curriculum developers whose products dominated US elementary reading instruction — most prominently Lucy Calkins of Teachers College Columbia and the Teachers College Reading and Writing Project, and Fountas and Pinnell, whose “Leveled Literacy Intervention” and “Benchmark Assessment System” were used in tens of thousands of US classrooms. The reporting traced the lineage of three-cueing back to its originators, documented its persistence in widely sold curricula, and showed that the curricula were still being marketed as evidence-based even as the cognitive-science evidence against them mounted.

Second, it told the story through individual children and families. Hanford’s reporting featured children who had been put through balanced-literacy programs, had failed to learn to read, had been diagnosed as having learning disabilities or attention problems, and had then been pulled out and taught using systematic phonics — at which point many of them learned to read within weeks or months. The implication, supported throughout the reporting by interview material from cognitive scientists and reading specialists, was that a substantial fraction of US children diagnosed with reading disabilities did not have reading disabilities. They had been taught using methods that the research evidence said would not work for them.

Third, it reached a non-specialist audience. The “Sold a Story” podcast became one of the most-listened-to education podcasts ever produced. State legislators, school-board members, district administrators, and parents who would never have read a paper in Psychological Science in the Public Interest listened to the podcast and understood, often for the first time, that the curriculum their children were being taught with was not consistent with the research evidence.

The political and policy response was the most significant US K-12 education reform of the 2020s. By 2024, more than thirty states had passed legislation requiring evidence-based reading instruction aligned with what was now broadly called “the Science of Reading” — which, in legislative practice, meant requirements for systematic, explicit phonics instruction, restrictions on the use of three-cueing, mandatory teacher training in the cognitive science of reading, and procurement requirements that prevented districts from using curricula judged inconsistent with the evidence base. Mississippi, which had passed an early version of this legislation in 2013 and seen substantial gains on the National Assessment of Educational Progress fourth-grade reading scores, became the policy template. Tennessee, Colorado, Ohio, North Carolina, California, and many others followed with varying versions.

The cumulative effect was that, twenty-four years after the National Reading Panel report and six years after the Castles review, US K-12 practice had finally moved into substantial alignment with the research evidence. The vehicle was not the research itself. It was journalism, supported by state-level political organizing.

Why Practice Diverged From Evidence For Three Decades

The reading wars are a long-form case study in the conditions under which practice diverges from evidence. Several mechanisms operated simultaneously.

The pedagogy was aesthetically more appealing than the research model. Whole-language and balanced-literacy approaches were built around real children’s literature, child choice, and the romantic image of immersion in meaningful text. Systematic phonics instruction, by contrast, requires teaching letter-sound correspondences explicitly and in sequence — work that is more structured, less child-led, and harder to make visually appealing on a classroom-tour brochure. Teachers, who generally chose the profession because they love books and children, found the whole-language frame more emotionally resonant than the phonics frame. This was not stupidity. It was a real conflict between the pedagogy that felt right and the pedagogy the research supported.

The institutional infrastructure favored the established approach. By the time the National Reading Panel report was published, schools of education had been training teachers in whole-language methods for two decades. Curriculum publishers had built product lines around balanced literacy. District reading coaches had been certified in those methods. Reversing course would have required retraining hundreds of thousands of teachers, replacing tens of thousands of curricula, and admitting that decades of work had been built on a flawed foundation. The institutional cost of reversing course was very high, and the institutional incentives to keep going were significant.

The evidence was framed as one position among many rather than as a converging consensus. Through the 2000s and 2010s, the question of how to teach reading was often presented in education-policy discourse as a matter of philosophy or pedagogy or values, in which different schools of thought had legitimate disagreements. This framing made the evidence base seem like one input among many, rather than what it actually was: a strongly converging finding from multiple independent research literatures. The frame of “everyone has their opinion” obscured the frame of “the evidence on this question is unusually clear.”

There were no immediate consequences for practitioners who ignored the evidence. Individual children who failed to learn to read were typically diagnosed with reading disabilities — an attribution that placed the failure inside the child rather than inside the instruction. The aggregate effect on national reading scores was visible only at multi-year time horizons, by which point causal attribution to specific curricula was complicated by many confounders. The accountability loop between bad instructional choices and visible consequences was long enough that the consequences rarely flowed back to the people making the choices.

The journalistic exposure was the missing piece. The reason the policy response accelerated dramatically after 2019 is not that new research appeared — the Castles review summarized literature that was substantially in place by the early 2010s. It is that Hanford’s reporting closed the gap between the research community and the parents, legislators, and school-board members who had standing to demand change. Without that exposure, the research consensus would likely still be sitting on the shelf today.

This pattern — research consensus, practice indifference, journalistic exposure, political response — is not unique to reading. The same dynamic plays out in pharmaceutical practice when off-label prescribing diverges from indication evidence, in financial advice when commission incentives diverge from fiduciary research, and in corporate training when leadership-development products diverge from the deliberate-practice and feedback literatures. The reading wars are unusually well-documented but they are not unusual in structure.

Strategist Takeaway

For strategists and operators evaluating “evidence-based” claims in education, corporate training, or any field where research and practice can diverge, the reading wars yield several actionable heuristics.

When practice and research diverge, ask which one moves first. In the reading wars, practice changed slowly and only after sustained external pressure. Research had converged decades earlier. If you encounter a domain where the practice has been stable for a long time and the research community has been quietly clear for a long time, the burden of proof is on the practice, not the research.

Look for the journalism, not just the journals. Academic publications establish what is true. Journalism translates what is true into something the relevant decision-makers can act on. If you are trying to understand whether a research finding has actually changed practice, the most informative signal is whether a high-quality investigative reporter has translated the finding into a story that non-specialists can engage with. If yes, practice is likely moving. If no, practice is likely static regardless of how strong the research is.

Distrust “balanced” framings on questions where the evidence is asymmetric. “Balanced literacy” succeeded as a brand in part because the word “balanced” is hard to argue against. In practice, however, the framing functioned to preserve the status of an approach the research did not support, by suggesting it was being combined with phonics in some appropriate ratio. When you encounter a “balanced” or “best of both worlds” framing of a question where the underlying evidence base is genuinely lopsided, the framing is likely doing political work, not scientific work.

Be suspicious of pedagogies and methodologies that flatter the teacher or trainer. Whole-language pedagogy gave teachers a more romantic, child-centered, intellectually congenial role than systematic phonics did. Many corporate training methodologies similarly offer trainers a more flattering self-image than the underlying evidence supports. When a methodology’s appeal is meaningfully driven by how it makes the practitioner feel about themselves, the evidence for its effectiveness deserves additional scrutiny.

Track outcomes for the children — or the customers — least able to compensate. Whole-language approaches worked acceptably for children with literate, engaged parents who supplemented at home. They worked badly for children without those advantages. The aggregate national failure of whole language was concentrated in populations that had no compensating support. Across domains, this pattern recurs: marginal pedagogies and marginal products tend to produce acceptable results for advantaged populations and bad results for less-advantaged ones, and the bad results are easy to attribute to the population rather than the product. If you are evaluating a learning product or service, the outcomes for the population least able to compensate for product weakness are the most informative outcome data you can look at.

The reading wars resolved. They resolved late, at substantial cost to a generation of US children. The resolution was the product of cognitive science, federal evidence review, comprehensive academic synthesis, investigative journalism, and state-level political action working together over twenty-five years. That is roughly how long it takes to close a research-practice gap in a field where the institutional incentives favor the wrong answer. Knowing this should change how you read the next “evidence-based” claim you encounter — particularly in education, but really anywhere.

Sources

  • Castles, A., Rastle, K., & Nation, K. (2018). Ending the reading wars: Reading acquisition from novice to expert. Psychological Science in the Public Interest, 19(1), 5-51. DOI: 10.1177/1529100618772271
  • Ehri, L. C., Nunes, S. R., Stahl, S. A., & Willows, D. M. (2001). Systematic phonics instruction helps students learn to read: Evidence from the National Reading Panel’s meta-analysis. Review of Educational Research, 71(3), 393-447. DOI: 10.3102/00346543071003393
  • Hanford, E. (2019). At a loss for words: How a flawed idea is teaching millions of kids to be poor readers. APM Reports. (August 22, 2019.)
  • Hanford, E. (2022). Sold a story: How teaching kids to read went so wrong. APM Reports podcast series.
  • National Reading Panel. (2000). Teaching children to read: An evidence-based assessment of the scientific research literature on reading and its implications for reading instruction. National Institute of Child Health and Human Development, NIH Pub. No. 00-4769.
  • Seidenberg, M. (2017). Language at the Speed of Sight: How We Read, Why So Many Can’t, and What Can Be Done About It. Basic Books.

Frequently Asked Questions

Did the National Reading Panel really say phonics was the answer? It said that systematic, explicit phonics instruction significantly improves reading outcomes and that the effect is especially strong for younger children and for children at risk of reading failure. It did not say that phonics was the only thing that mattered — the panel identified five pillars of reading instruction. But on the specific question of whether systematic phonics is better than non-systematic or no phonics, the answer was yes, with a moderate, well-replicated effect size.

Is “balanced literacy” the same as whole language? Functionally, in most US implementations through the 2010s, yes. Balanced literacy was marketed as combining whole-language and phonics approaches in appropriate balance, but the most widely used balanced-literacy programs retained the core whole-language commitments — including three-cueing and an emphasis on predicting unknown words from context — that the cognitive-science evidence specifically did not support. The label change did not, in practice, change the underlying pedagogy meaningfully.

Are children in states with phonics mandates actually reading better? The evidence is partial but suggestive. Mississippi, which passed its phonics-mandate legislation in 2013, posted some of the largest fourth-grade reading score gains on the National Assessment of Educational Progress through the late 2010s, moving from one of the lowest-performing states to near the national average. The state-level evidence from other recent adopters is too recent for definitive evaluation, but the early signals are consistent with the research expectation that systematic phonics works better than the alternatives.

Does this mean whole language never works? Children who are read to extensively at home, who grow up in print-rich environments, and who have engaged literate parents often learn to read across a wide range of instructional approaches, including whole language. The reading wars were never about whether some children learn to read in whole-language classrooms. They were about whether systematic phonics produces better outcomes for the population as a whole, and especially for children who do not have the compensating home advantages. On that question, the evidence is clear.

Why did it take so long for the policy environment to catch up to the research? A combination of institutional inertia in teacher-preparation programs and curriculum publishing, the aesthetic appeal of the whole-language pedagogy to teachers and parents, the framing of the question as a matter of philosophical disagreement rather than empirical evidence, and the absence of accountability loops connecting instructional choices to visible consequences. Sustained investigative journalism, starting with Emily Hanford’s APM Reports work in 2018, was the proximate cause of the political response that finally moved practice.

Is the reading-wars resolution actually settled, or could the pendulum swing back? The empirical case is as settled as such cases get in social science, but the institutional and pedagogical pressures that produced whole-language dominance have not vanished. The Science of Reading legislative wave has run for roughly five years, which is short relative to the multi-decade dominance of balanced literacy. Vigilance is warranted. The pattern of practice drifting back toward the more emotionally appealing pedagogy, particularly after political attention shifts elsewhere, is well-established in education-policy history.

What is the right way for strategists to read claims of “evidence-based” education or training? Look at the specific evidence cited, not the marketing language. Check whether the cited studies are intervention studies with control groups or merely descriptive correlational work. Ask whether the population the product was tested on resembles the population you would deploy it to. Ask what outcomes were measured, and whether those outcomes are the outcomes you actually care about. And, especially, ask whether the practice community in the relevant domain has converged with the research community, or whether there is a research-practice gap of the kind that defined reading instruction for thirty years.

replication-crisis reading-wars phonics education-research evidence-evaluation
Share this article
LinkedIn (opens in new tab) X / Twitter (opens in new tab)
Atticus Li

Experimentation and growth leader. CXL-certified CRO practitioner, Mindworx-certified behavioral economist (1 of ~1,000 worldwide). 200+ A/B tests across energy, SaaS, fintech, e-commerce, and marketplace verticals.

About LinkedIn Newsletter