What is the CRA Framework in Math Intervention?
The CRA framework — Concrete, Representational, Abstract — is a three-stage explicit-instruction sequence in which students first manipulate physical objects (concrete), then work with two-dimensional drawings or diagrams (representational), and finally operate on symbolic notation alone (abstract). The most recent meta-analytic review of 30 single-case design studies, Ebner, MacDonald, Grekov & Aspiranti (2025, Learning Disabilities Research & Practice 40(1)), yielded a statistically significant Tau-BC effect size of 0.9965, and the What Works Clearinghouse synthesis by Bouck, Satsangi & Park (2018, Remedial and Special Education 39(4):211-228) confirms CRA meets the criteria as an evidence-based practice for students with learning disabilities across math domains.
Why CRA works — the theoretical basis
CRA traces its roots to Jerome Bruner's (1966) enactive-iconic-symbolic progression and is reinforced by Sweller's (1988) Cognitive Load Theory: novices acquire schemata most efficiently when new information is introduced in concrete form that reduces extraneous cognitive load, then progressively abstracted as the learner's schemata consolidate. Each stage scaffolds schema formation for the next; the symbolic notation of mathematics becomes meaningful only after the learner has anchored the referent in physical and pictorial experience.
The Wilson et al. (2019, Nature Communications 10:4646) "eighty-five-percent rule" establishes that training tasks are optimally learned at approximately 85% success rate — a finding that directly informs how CRA stages are paced. IncluMath applies the 85% target at baseline and 88-90% for ADHD profiles to reduce delay-aversion frustration (Sonuga-Barke 2005, Behavioral & Brain Sciences 28(3):397-419).
The three CRA stages
| Stage | What the student does | Example — 27 + 18 |
|---|---|---|
| Concrete | Manipulates physical objects | Base-10 blocks: 2 rods + 7 units; combines 1 rod + 8 units; trades 10 units for 1 rod. |
| Representational | Draws pictorial/diagram of same action | Draws sticks and dots; circles groups of 10; writes count. |
| Abstract | Performs symbolic algorithm | Writes the standard algorithm: 27 + 18 = 45, regrouping tens. |
The evidence base — what CRA meta-analyses show
- Ebner et al. (2025), Learning Disabilities Research & Practice 40(1). Meta-analysis of 30 single-case design studies. Tau-BC = 0.9965, statistically significant.
- Bouck, Satsangi & Park (2018), Remedial and Special Education 39(4):211-228. Evidence-based practice synthesis — CRA meets WWC criteria across multiple math domains (whole-number operations, fractions, algebra).
- Witzel, Riccomini & Schneider (2008), Intervention in School and Clinic 43(5):270-276. Secondary-students-with-LD study: students learning algebra through CRA significantly outperformed students using abstract instruction alone on post- and follow-up tests.
- Gersten, Chard, Jayanthi et al. (2009), Review of Educational Research 79(3):1202-1242. Foundational federal meta-analysis (k=42) across SPED math instruction. Explicit instruction g=1.22; visuals g=0.47; CRA components g=0.41.
- Dennis et al. (2016), Learning Disabilities Research & Practice 31(3):156-168. MLD instruction meta-analysis (k=25, N>4,800): CRA g=0.48 as a standalone component; technology-delivered g=0.55 — empirical equivalence with in-person delivery.
Expert perspective
"CRA instruction is an evidence-based practice for teaching mathematics to students with learning disabilities. When delivered with fidelity, it produces gains that transfer to the abstract notation students encounter on grade-level assessments."
Implementation principles
- Mastery drives advancement, not time. OLI/Smart Learning Environments (2018) indicates a minimum of 7 attempts; Doroudi (2020, AIED) proposes N=4 consecutive correct (N-CCR heuristic) adjusted for guessing.
- Explicit instruction at each stage. Gradual Release of Responsibility: I Do → We Do → You Do (Hughes, Morris, Therrien & Benson 2017, Learning Disabilities Research & Practice).
- Virtual manipulatives are effective. Ebner et al. (2025) meta-analysis includes virtual manipulatives in the Tau-BC estimate; Bouck & Park (2024, Canadian J Science, Mathematics and Technology Education) document equivalent outcomes with blended concrete-plus-virtual designs.
- Students must manipulate, not merely view. IES guidance on virtual manipulatives — interactive manipulation produces equal-or-stronger gains vs physical; passive viewing produces weaker effects.
- Pair CRA with schema-based instruction for word problems. Jitendra, Harwell et al. (2017, J Learning Disabilities 50(3):322-336) cluster-RCT N=1,999: SBI g=0.37-0.45 immediate, g=0.34 at 9-week delay.
- Distribute practice. Wiseheart, D'Souza & Chae (2025, Behavioural Sciences) meta-analysis: distributed vs massed d=0.54.
Do NOT do — common CRA mistakes
- Don't skip the concrete stage for older students. Witzel et al. (2008) established CRA effectiveness for secondary students with LD. Age is not a reason to begin at abstract.
- Don't force all students through all three stages at uniform pace. Rapid advancement is appropriate for high-mastery students; expertise reversal means unsolicited scaffolds can harm learners with PFA > 0.85 (Kalyuga 2007, Educational Psychology Review 19:509-539).
- Don't combine worked examples with self-explanation prompts for students with LD. Barbieri et al. (2023, Ed Psych Review) — the combination can harm learning; choose one.
- Don't use productive failure with elementary students. Sinha & Kapur (2021, Review of Educational Research, N>12,000): productive failure g=0.36-0.58 only for ages 11+; harmful for K-5.
- Don't rely on growth-mindset prompts. Macnamara & Burgoyne (2023, Psychological Bulletin, N=97,672): d=0.02 ns in highest-quality studies.
How IncluShift implements CRA
IncluMath operationalizes CRA as a technology-delivered progression: BlockManipulative (concrete), sticks-and-dots drawing (representational), and symbolic algorithm (abstract). Advancement between stages is driven by the PFA (Performance Factor Analysis) adaptive engine (Pavlik, Cen & Koedinger 2009) targeting 85% success (Wilson et al. 2019, Nature Communications 10:4646). Scaffold tiers follow the on-demand-only rule — high-mastery students (PFA > 0.7) never receive unsolicited scaffolds to prevent expertise reversal (Kalyuga 2007). Cooldown protocol on three consecutive errors implements Balban et al. (2023, Cell Reports Medicine 4(1):100895) cyclic-sighing breathing to support math anxiety regulation. See IncluMath.
Key research citations
Ebner, MacDonald, Grekov & Aspiranti (2025). A meta-analytic review of the CRA math approach. Learning Disabilities Research & Practice, 40(1). [Tau-BC=0.9965]
Bouck, Satsangi & Park (2018). The CRA approach for students with LD: An EBP synthesis. Remedial and Special Education, 39(4), 211-228.
Gersten, Chard, Jayanthi et al. (2009). Mathematics instruction for students with LD: A meta-analysis. Review of Educational Research, 79(3), 1202-1242.
Witzel, Riccomini & Schneider (2008). Implementing CRA with secondary students with LD. Intervention in School and Clinic, 43(5), 270-276.
Wilson, Shenhav, Straccia & Cohen (2019). The eighty five percent rule for optimal learning. Nature Communications, 10, 4646.
Jitendra et al. (2017). SBI for students with math difficulty. J Learning Disabilities, 50(3), 322-336. [g=0.37-0.45]
Kalyuga (2007). Expertise reversal effect. Educational Psychology Review, 19, 509-539.
This page provides educational information about the CRA framework. IncluMath is a research-informed adaptive practice tool; it has not been evaluated in its own controlled study. Consult qualified special-education and math-intervention professionals before making placement decisions.