Society for the Scientific Study of Reading conference: Day 20 Replies
Thanks for the nice feedback about my SSSR conference Day 1 blog post, it’s helped motivate this one. Here are summaries of what I attended on the second day (any errors/misunderstandings are my own).
Neural deficits in dyslexia
Dr Tracy Centanni of the university of Florida talked about two genes on Chromosome 6 – KIAA0319 and DCDC2 – which are widely studied in dyslexia. If you suppress KIAA039, the left temporal area of the brain starts working inconsistently, either too fast or too slow. About 50% of kids with dyslexia have low neural consistency. This makes it harder to link speech sounds to letters.
DCDC2 is linked with white matter integrity in the language network, and helps neurons go where they should. Suppression of this gene in rats really messed with the speed of their performance, so perhaps this gene is linked to Rapid Automatised Naming (RAN), vital for fluent reading.
The most common neural deficit in dyslexia is low activation in the visual word form area (VWFA) in the left side of the brain. In non-literate people this area processes information about things like tools, faces and structures. Once learning to read begins, VWFA cells start to specialise in processing written words, linking the visual and language areas of the brain. Face processing gets partly shunted to the right side of the brain, but most people don’t notice they’ve lost half their face processing real estate. Even before dyslexic kids start school, their VWFAs respond less strongly to letters, but not nonsense objects.
Goals targeting writing skills in ASD kids’ IEPs
The Autism Spectrum is very diverse, and kids with ASD vary greatly in their literacy and broader educational needs. 70% of ASD students in the US have an Individualised Educational Program (IEP).
A/Prof Matt Zajic of Columbia University surveyed families of 954 ASD kids with an average age of 10, 80% of whom were male, and 60% of whom spent more than half their time in mainstream school. 80% had good cognitive skills, 89% had language disorder, 43% had ADHD. The survey asked whether the students had at least one IEP goal targeting handwriting, keyboarding, spelling, grammar, punctuation, or sentence/paragraph construction, but didn’t evaluate the quality, quantity or priority of these goals.
Data were analysed using Latent Class Analysis, a statistical method used to identify hidden subgroups. About 15% of kids, mostly younger ones, fell into a transcription subgroup (handwriting and spelling). 17% of kids, mostly older ones, fell into a text generation subgroup (sentence and/or paragraph construction).
About 24% had ‘most needs’, and 44% had ‘minimal needs’, which I think means 44% of kids were good at writing and 24% were seriously not, but that could be wrong, he was going very fast. Keyboarding goals were low across the board. Cognitive, language and/or attention diagnoses didn’t make much difference to the data, but grade, time spent in mainstream school and adaptive behaviour did make a difference. More research is needed on different profiles, relationship to reading goals and level of professional support.
TOPsy: the Test of Prosody via Syllable Emphasis
There’s a clear link between prosody (speech intonation/stress patterns) and reading in research. While we know prosody is important for communication, it’s poorly understood. Dr Srishti Nayak of Vanderbilt University Medical Centre has been working on a test of adult prosody. It takes 10 minutes, and has 28 items. Test words are said by a female American English speaker, and test-takers identify word stress.
However, Dr Nayak pronounces many of the test words differently from the test, because she speaks Indian English, so she doesn’t do very well on her own test. The TOPsy is thus quite accent specific, and how it might work in other languages is not clear, since some languages have mobile lexical stress (e.g. English, as in ‘PHOto’ but ‘phoTOGrapher’), while in other languages (e.g. French) lexical stress is fixed, or operates in other ways.
The TOPsy built on an existing internet-based silent reading fluency test from Hong Kong called WordSword. This presents digital text without spaces between words. Test-takers have four minutes to mark word boundaries. The TOPsy also drew on a large genetic association study of word reading, to combine prosody scores with measures of genetic predisposition for reading. I didn’t really understand how, so write to Dr Nayak if you’re curious.
Anyway, the lower your prosody score on the TOPsy, the more likely you are to have dyslexia. There was also a genetic association between prosody and reading. People being studied provided genetic material by mailing back saliva kits during the pandemic. Researchers are creative!
Reading skills and the digital divide
47% of kids living in poverty in the US lack high-speed internet. Research in North Carolina compared broadband speed and state reading test scores, and found a small but significant increase in scores thanks to faster broadband.
Dr Callie Little of the Florida Center for Reading Research examined the impact of pre-pandemic (2015-2017) broadband access on early reading nationally. Data were from 650 children involved in the National Project on Achievement in Twins, aged 5 to 8 years, 89% of whom were white. Parents ranked internet speeds from 1-5, and average download speed in megabytes per second in each census tract was also collected. The Child Opportunity Index is also organised by census tract, and rates opportunity from 1-100 based on educational, health, environmental and socio-economic factors. DIBELS composite scores were used to measure reading. Number of devices in the home were not measured.
No significant association between internet access and early reading achievement was found in this study. Broadband speed information for 2020 is now available, so researchers will look at the impact of COVID-19 on reading achievement, including for older kids who had to do more homework online.
UFLI foundations: an affordable, evidence-based Tier 1 program
Dr Holly Lane talked about the University of Florida Literacy Intervention (UFLI) Foundations program, which was developed over two years in response to school requests to improve phonics outcomes. University staff had been giving schools program-agnostic professional learning advice, assuming this would lead to improved instruction, and thus to better student outcomes.
However, they discovered that professional learning improves teachers’ knowledge, but their practice doesn’t change. Teachers also need an educative curriculum. So the university staff asked, “What if we just develop the lesson plans for you?”. Schools said “yes, please!”.
UFLI scope and sequence and lesson plans were developed to be very explicit and systematic, with lesson routines, many opportunities to practice, interleaving, the development of automaticity, progress monitoring and differentiation. UFLI was piloted in one school with 16 teachers in the first year, then usability and feasibility revisions were made e.g. they discovered word chains need to be provided, as teachers are more likely to skip them than make up their own.
In the second year they did a district-wide pilot, with 1670 students in Years K-1, and measured progress with DIBELS (8th edition). The comparison groups were children at the same levels in the previous year. School autonomy presented a problem in implementation, as not every teacher implemented UFLI with fidelity, and this was related to outcomes. However, Effect Sizes for UFLI were still remarkable: 1.44 for Kindergarten and 2.04 for Year 1 (an Effect Size of 0.8 is considered high).
In the US summer of 2022, UFLI Foundations was released. Each lesson plan is two pages, and the only other things teachers need are letter sets, markers, whiteboards and free stuff from an online toolbox e.g. decodable text, a slide deck, web-based apps, and printables. They are still adding to these resources. UFLI is thus very cost-effective to implement, you just need the book.
UFLI has now gone viral, about 180,000 teachers around the world are using it, and Dr Lane has just toured Australia (darnit, I was busy with other things and missed her).
Teacher-led Tier 2 in Aotearoa/NZ’s Better Start Literacy approach
Associate Dean Brigid McNeill of Canterbury University talked about the Better Start Literacy Approach‘s Tier 2 (small group, helping beginners keep up) intervention. This needs to be:
- Explicit and systematic across all Big Five domains of phonemic awareness, phonics, vocabulary, fluency and comprehension.
- Complementary/supplementary to classroom instruction (Tier 1), but providing more intensity and exposure.
- Implemented early, and then building towards evaluation.
The Better Start program is strengths-based, and includes online teacher professional learning completed whenever suits them, and resulting in a microcredential. Facilitators are paired with teachers, to provide coaching and mentoring. There is family engagement and culturally appropriate resources. Online monitoring assessments identify kids with difficulties, monitor learning, help with planning, and are adapted for children with complex needs.
Better Start is now in 833 schools across NZ, and 3650 teachers have been trained. Over 43,000 students have been involved, 23.4% Maori, 9.4 Pasifika, 12.4% Asian and 46.8% European. 14% of children involved have had tier 2 support.
The Ready To Read Phonic Plus series readers, which were co-constructed with teachers and informed by parent and child feedback, are used. There are some scripted lessons, and some that teachers develop.
Children have baseline assessment at school entry, then 10 weeks later any strugglers start Tier 2 intervention for 20 weeks, focusing on phonemic awareness, phonics and word decoding/encoding. The gap between the Tier 2 children and other children disappeared at 30 weeks on all measures except their scores on reading connected text, where controls reached a higher level, perhaps because they could process text faster.
Key aspects of this research were that it took a co-creation/partnership approach, had high stakeholder engagement, and quality professional learning and development delivered at scale, with in-context, ongoing support. Evaluation of impact was via teacher administered measures and integrated into the study design, and teachers could use their evaluations to drive teaching decisions.
Stay tuned for more on Better Start’s Tier 1, discussed on Day 3 of the conference. There’s also a Tier 3.
The slow development of fast word recognition
Most children get better at recognising both written and spoken words in their first few school years. To understand what we hear and read, we must identify words efficiently. Spoken words are presented temporally in fractions of a second, but when reading, we must rapidly tell the difference between words with shared letters, by building evidence for the correct word, and suppressing competitor words.
Professor Bob McMurray of the University of Iowa talked about research in which 280 children in Grades 1-3 saw briefly presented words and then had to click on matching pictures. Some of the competing pictures represented words with shared letters e.g. for the word ‘ship’, pictures were ‘shin’, ‘chip’, ‘shop’, ‘snip’ and ‘coin’. They also had to identify spoken words, but I didn’t write down how that was done, sorry. Write to him if you need to know.
There were concurrent relationships between spoken and written word identification, which suggest a common factor. More robust and efficient spoken word recognition leads more efficient written word recognition. Efficient spoken word recognition might reflect better phonological processing (discerning the structure of spoken words). To understand how children become efficient readers, we might need to better understand how they become efficient listeners.
The words children see and hear
Dr Luan Li of East China Normal University spoke about research into lexical variability (diversity in words used) in language seen and heard by school-aged children in China. Most research in this area has focussed on preschoolers, but in the early school years there is a vocabulary spurt, and it is a critical period for the development of abstract ideas.
Vocabulary diversity was studied in three sources of language input: child-directed speech (1.8 million words), animated cartoons/movies (1.8 million words), and picture books (1.5 million words). Picture books had the most diverse vocabulary, but cartoons/movies had greater contextual and semantic diversity, i.e. they used words in a wider variety of ways/to express more varied meanings. Child-directed speech had the least variability. The words kids see and hear affect language and reading development. If you can read Chinese, more information is here.
DIBELS 8th edition as a dyslexia screener
Most US states now require students to be screened for dyslexia, but there aren’t any screening tests specifically validated for this purpose (I hope EarlyBird will help fill this niche). Dr Patrick Kennedy from the University of Oregon spoke about outcomes of the first two of four years of research into the use of a general reading screening test – DIBELS 8 – as a dyslexia screener.
Preliminary results suggest that DIBELS can be used as a dyslexia screener, though it’s difficult to decide exactly where to put cutoff scores (e.g. 5th, 15th or 25th percentile).
Computer-adaptive reading screening
Dr Emily Farris from Middle Tennessee State University talked about research comparing three computer-based reading screening tests: Istation Indicators of Progress Early Reading, MAP-Reading and Star Reading. Third graders’ results on these screening tests were compared with the state achievement test, which 55% of children currently fail.
Sadly, none of the computer-based screeners were very sensitive or specific, they all identified far too many kids. These results suggest it’s probably not a good idea to rely on a single measure to identify children at risk. Screening in third grade is also too late, it needs to be done much earlier.
Dr Kristina Breaux from educational publisher Pearson, developer of the widely-used fourth edition of the Weschler Individual Achievement Test, spoke about students who seem to read within normal limits, but still find reading difficult, and are reluctant to do it. These kids have above-average listening comprehension (standard scores of 105+), below-average phonological processing and pseudoword reading (standard scores of 85 or lower), but real word reading in the average range (standard scores of 85-90+ across four word reading subtests).
The WIAT-4 standardisation sample of 1800 school-aged students suggests that 1-3% of students might be considered compensating readers.
The text-complexity leap between third and fourth grade
In the US, nearly two-thirds of 4th and 8th grade kids are not reading at grade level, but ‘grade level text’ is poorly defined. Text complexity in grade level text increases substantially between third and fourth grade (from Lexile 420 to 740, a 76% jump).
Dr David Paige of Northern Illinois University discussed work developing systems which would allow more fine-grained tracking of progress, and earlier foundational work, so there is a smoother transition to reading at a fourth-grade level. This is needed if most fourth graders are to succeed on their state tests. The systems he was discussing are US ones, so I didn’t understand all of it, sorry. There is an article here.
Awards, including Distinguished Scientific Contributions Award
The awards ceremony was the last event on Friday, and I didn’t write down all the award and recipient names because I assumed they’d be on the SSSR website Awards page after the conference. They’re not there yet, and a new SSSR website is being developed, so perhaps the update is going straight onto the new site, which should be available soon.
Charles Hulme (whose name I have been pronouncing ‘Hulm’ in my head for decades, but it’s ‘Hume’, gah) won the Distinguished Scientific Contributions award and gave a talk titled: “What we talk about when we talk about reading“. I can’t do this talk justice here, but wrote down a few things:
- Causes of problems like dyslexia and poor reading comprehension are theoretical statements, not real Things. They only exist in the context of a well-specified theory/model, developing out of correlations and operating forwards in time.
- Hulme has been trying to specify cognitive processes in learning to read which can be tested statistically. Famous statistician George Box said, “All models are wrong, but some models are useful”. William of Ockham (he of Ockham’s Razor) said “Plurality must never be posited without necessity”. Einstein said, “Everything should be made as simple as possible, but not simpler.” We look for the simplest way to explain available data.
- Path diagrams can be used to represent causal theories. Causes and consequences are linked with one-headed arrows. Understanding causes is essential for developing and evaluating interventions. Intervention studies allow us to test causal theories.
- There are at least three causal influences of individual differences in children’s ability to learn to decode: letter-sound knowledge, phoneme awareness (PA), and rapid automatised naming (RAN). The first two are the basis of the alphabetic principle. RAN seems to tap a separable mechanism concerned with the efficiency of retrieving names from visual inputs. Phonemic skills and letter knowledge should fairly clearly predict early reading, and affect each other.
- Hulme set out to prove that RAN wasn’t important, but the data did not agree. Alphanumeric RAN after three months of reading instruction predicted the later rate of growth in reading over two years (Lervag and Hulme 2009). RAN repurposes neural mechanisms meant for other tasks in left-hemisphere brain region networks. Changes in the volume of the arcuate fasciculus (a bundle of nerves connecting language and frontal areas of the brain) predicted reading. RAN taps the left-hemisphere naming system, so having effective connections between posterior and frontal areas seems essential for fluent reading.
- The three components of the Triple Foundation Model (PA, phonics, RAN) all centre on aspects of phonology. RAN seems to tap some neurally constrained processes for retrieving the names of printed items. RAN seems to operate more strongly in later stages of development. We can’t train it, so can’t be sure it’s causal, but it probably is (he said, reluctantly).
- The Simple View of Reading (SVoR) is a statistical model of the concurrent predictors of reading comprehension. There are four potential profiles: typical readers, hyperlexia, dyslexia and language disorder/poor comprehension. Word reading and language comprehension should both predict reading comprehension, and they do. They have double-headed arrows between them in path diagrams, indicating a reciprocal relationship.
- The Wellcome Language and Reading study was a seven year study. Hulme thought speech skills at age three-and-a-half would be critical for the development of reading ability. He was wrong. Language development strongly predicts RAN, phonics knowledge and phonological skills. Word-level literacy at five-and-a-half tells you a lot about reading comprehension at age eight. All this is in line with the SVoR. But language skills are critical for the development of phonological skills. Phonology grows out of much broader, language-based skills.
- However, persistent speech difficulties are related to reading problems. One study of 569 kids starting school followed them for four years, and 7% had speech difficulties. These were a powerful predictor of later reading difficulty. The development of PA and Letter Sound Knowledge (LSK) are affected by persistent speech problems, a clear risk factor for later decoding problems. It’s easy to identify kids with speech problems.
- The Wellcome study provided strong support for the SVoR, but makes it more complicated as language and phonology are highly correlated, not independent. But the simplicity of the SVoR makes it helpful.
- Code-related skills drive decoding which influences comprehension. We also have a direct effect from language to reading comprehension. Early language skills provide the foundation for both decoding and comprehension.
- The Nuffield Early Language Intervention (NELI) improves narrative, vocabulary and listening skills. There have now been several randomised controlled trials of the effects of this intervention. Teaching assistants deliver this pullout program targeting the bottom 20-25% of kids. It produces reliable improvements in kids’ oral language skills. Children who got intervention strongly changed their language post-test scores compared with kids on the waiting list (the difference was 0.8 of a Standard Deviation, which is A Lot). In their second year of school, the intervention kids had improved language, and this completely accounted for improvements in reading comprehension.
- A mobile app called Language Screen has been developed, which operates on an Apple or Android tablet or phone. It has four subtests and is an easy, automated way to assess kids’ language ability. Thanks to UK Education department funding the NELI program has now been provided to 100,000+ children and data on well over half a million kids is being assessed with Language Screen. They also have a preschool language enrichment program which runs for 20 weeks, which kids and teachers like, and which improves language skills by a quarter of a standard deviation at population level.
- However, there are still lots of people selling snake oil interventions that don’t have good evidence.
- The title of this talk was inspired by a rather pessimistic short story by Raymond Carver called “What we talk about when we talk about love”, which is a story about what love means. There’s much more reason for optimism in the field of reading difficulties!