Set for Variability - a fancy term for 'map it backwards'

Typical Saturday morning. I’d decided not to think about work (word mapping!) and then a Facebook post from The Reading League pops up with the statement:

“Did you know that set for variability can be a stronger predictor of word-reading skill for students with dyslexia than phonological awareness (Steacy et al., 2022)?”

And the usual thought kicks in: here we go, teachers will now assume that’s accurate without looking at the actual research. I would hate to see the Steacy et al. finding overinterpreted on social media and taken as a green light to emphasise SfV over phonemic foundations. If you are interested in me picking it apart, jump to the end of this blog.

The Reading League quote uses the term phonological awareness. Here I’ll use phonemic awareness, since it’s phoneme-level skill that dyslexic learners struggle with most and that underpins decoding and spelling. So, if a child knows the grapheme markers, the plausible pronunciations for most of those graphemes (enough to make partial decoding useful), and can blend (has good phonemic awareness), then yes, word knowledge facilitates set for variability. But poor phonemic awareness is a core marker of dyslexia, so that has to be addressed before SfV can be used effectively.

Why? Because it is the bonding of speech sounds, spelling, and meaning that facilitates storage in the orthographic lexicon and enables self-teaching (Speech Sound Mapping Theory).

They also mention ‘solid grapheme–phoneme knowledge’. But how are they quantifying that? How many of the 350+ correspondences, as seen in the Spelling Clouds®, are they assuming children know? Commercial phonics programmes only explicitly cover around 100 – in England, this is across two years (KS1). If children must rely only on plausible pronunciations of graphemes, they are often stuck. So they end up relying on guesswork from partial decoding, and that can be exhausting, especially for dyslexic learners.

In England, synthetic phonics is mandated, yet 1 in 4 children still cannot read or spell at a minimum level by age 11 (DfE 2024). These children are taught around 100 GPCs. My findings in schools show that strict adherence to “no guessing” means many miss out on the benefits of SfV. When I talk about SfV in Word Mapping Mastery teacher training, teachers often worry they will be going against DfE guidelines.

At the Early Dyslexia Screening Centre we actually find that what may be referred to as three-cueing is more beneficial than relying only on the limited GPCs taught. For example, when a child sees “Sam has so **** apples. Poor Fred does not have *** apples” they can deduce the missing words many and any. They then map the phonemes and graphemes (m–a–n–y to /m ɛ n i:/, and a–n–y to /ɛ n i:/). If they try another word, like a number, it will not fit the graphemes or the initial phoneme. Dyslexic students who work this way move towards self-teaching earlier. Three-cueing is fine — it can be brilliant — if it leads to phoneme–grapheme mapping.

We also have word mapping tech to show them the grapho-phonemic structure of any word, i.e. both the phoneme value and grapheme markers, with Phonemies® and our Code Mapping® algorithm (now being patented). It maps all words of English.

Skilled readers do use three-cueing to reach self-teaching. The key difference between those who map words easily and those who do not comes down more to phonemic awareness than to word knowledge. That is why this post worried me a little, because of how teachers may interpret it. They are time-poor and often do not dig deep enough into the research.

And here is the elephant in the room: what counts as the “actual pronunciation” of the word? The way the teacher says it? The child’s own version? Or the universal spelling code represented by IPA (Stanislas Dehaene, Reading in the Brain)? I talk about 'plausible pronunciations here)

Ok, now I’m off to walk the dogs and try to forget about word mapping for the day. My brain, of course, is chuckling: “Yeah, right, good luck with that.”

Miss Emma MEd SEN Neurodivergent Reading Whisperer®

As promised for fellow research nit-pickers Key limitations / caveats in interpreting the Steacy et al. result

1. Correlational design, not causalThe study is correlational / predictive. Even though SfV “dominates” in their regression models, that does not prove causal primacy over PA. It’s possible that SfV as measured is partly a downstream manifestation of underlying phonological or decoding ability, vocabulary, orthographic knowledge, or other latent traits. In other words, SfV might encapsulate or partly mediate the effect of PA + decoding + lexical knowledge. The regression partitioning could overstate SfV’s “independence.”

2. Measurement overlap / construct interdependenceSfV tasks often involve phonological‐orthographic “clean-up” or “realignment” of mispronounced forms (i.e. noticing that wæsp should be wɒsp). That inherently draws on phonological skill, decoding skill, vocabulary (lexical knowledge), morphological and orthographic knowledge. In other words, SfV is not a “pure” separate construct but a hybrid one. Some portion of its predictive power may derive from the overlap with phonological awareness or decoding skill. (The authors acknowledge that SfV variance is partially accounted for by phonological and orthographic predictors in prior work)

3. Population / sample limitations and generalisability

   • Their sample is of typically developing and struggling readers in mainstream classes, not necessarily a “diagnosed dyslexia-only” group. Thus, the result is for a mixed sample, not confirmed dyslexic children only.

   • The quantile analysis shows that SfV’s dominance is more apparent in the lower quantiles of reading skill, but that does not guarantee it holds for all children with dyslexia, particularly those with severe PA deficits.

   • Cultural, linguistic, and instructional context matters. Their data is U.S.-based (or at least non-UK), so transfer to England’s phonics mandate, different teaching practices, dialect variation, and orthographic exposures may differ.

4. Residual variance and practical effect sizesEven with SfV’s “dominance,” the total variance explained is still limited. SfV accounts for unique variance beyond other predictors, but a lot of variance remains unexplained by any predictor. Also, the difference between 15 % (SfV) and 1 % (PA) depends heavily on which predictors are included, how they are measured, and the model specification. If a stronger, more precise PA measure or decoding measure were used, PA might contribute more. The low ~1 % figure for PA might partly reflect the particular PA tasks or sample used.

5. Task specificity and item-level effectsThe SfV task is operationalised via mispronunciation correction / “cleanup” tasks (e.g. giving a non-word or mispronounced form and asking the child to recover the “actual” word). That is itself a specialized task. Performance can be influenced by familiarity with vocabulary, metalinguistic awareness, attentional control, context cues, etc. It is not strictly “set for variability in real reading contexts.” The translation from performance on this task to actual everyday decoding in texts is nontrivial.

6. Ignoring the necessity of foundational phonological skillsEven if SfV predicts more variance in their models, it does not mean that phonemic awareness or phonological decoding is unimportant or superseded. If a child doesn’t have sufficient phonemic awareness first, they can’t reliably engage in SfV. The Steacy et al. model doesn’t address intervention sequencing or the minimal thresholds of PA needed to even apply SfV. In practice, many children with dyslexia may need focus on PA first before variability strategies can be effective. The study doesn’t (and can’t) tell us which comes first in instruction; it only tells us which correlates better under certain conditions.