NGSS-Designed Curriculum Materials for the Middle Grades: Results from a Study on Student Learning in Physical Science

29 middle schools randomized (15 treatment / 14 control); analytic sample 18 schools, 19 treatment and ~14 control science teachers, 1,953 grade-7 students (913 treatment / 1,040 control) across 3 districts.

US grade-7 physical science (Matter and Its Interactions), three districts including two large cities (>250,000 population); 11 of 18 schools Title I; 41% White, 23% Hispanic, 21% Asian, 4% Black; 10% IEP, 5% ELL.

School-level cluster RCT (matched pairs/triads, randomized within district) of Amplify Science 6-8 — one of the first widely available NGSS-designed middle school programmes — as materials PLUS teacher professional learning, against business-as-usual NGSS instruction (control districts used a redesigned NGSS curriculum, a district-adopted textbook, or an open-source PBL curriculum; both arms accountable to the same NGSS performance expectations). Amplify Science was developed by UC Berkeley's Lawrence Hall of Science; the evaluation team (WestEd, LFC Research, edCount) is independent of the developer and NSF-funded (DRL-1913317). TWO SERIOUS CAVEATS. First, the outcome is a RESEARCHER-DESIGNED seven-task NGSS assessment built by the study team (informed by the Next Generation Science Assessment project), not a standardized state test — though it was deliberately aligned to the state-standard performance expectations both arms were required to teach, and scoring was blinded and double-scored on 20%+ with third-scorer adjudication. Second, cluster attrition was 37.9% (18 of 29 schools) with 9.5% differential attrition (33.3% treatment vs 42.9% control), plus high student non-response driven by COVID closures; baseline differences of g = −0.11 (ELA) and −0.06 (maths) favouring control were statistically controlled. That attrition is why this is graded C rather than B. Read in full text.

This is the answer to 'is there ANY causal evaluation of an NGSS-aligned curriculum at scale?' — yes, one, and it is positive: g = 0.40 (p<.001), a 2.07-point gain on a 0-25 scale (13.77 vs 11.76), equivalent to about 16 percentile points. Effects were similar across gender, race/ethnicity and prior maths/ELA achievement. Read it with the archive's discipline, though: the treatment is an entire curriculum-plus-PD package, the outcome is a researcher-built test, and 38% of schools left the study. The instructional-practice exploratory analysis is arguably more informative about WHAT changed: treatment teachers reported far more reading of scientific texts (+1.38), writing in notebooks (+0.87) and communicating scientific thinking (+0.73), while CONTROL teachers reported more hands-on experiences (−0.61) and demonstrations (−0.57) and much more use of mathematics/computational thinking (−1.28), analysing data (−0.59) and asking questions (−0.46). The winning arm was the more text- and discourse-heavy one. Nothing here isolates sequencing: the coherent-storyline design principle travels with materials, PD, digital simulations and literacy scaffolds, so the 0.40 cannot be attributed to the order in which content was taught.

LOW-MEDIUM. School-level randomization defeats family selection, but 37.9% cluster attrition with 9.5% differential attrition reopens the door to non-random composition differences; the authors control baseline ELA/maths and ethnicity in response.

Unreplicated. The OpenSciEd efficacy study (AIR) uses propensity-score matching designed to meet WWC standards with reservations and has not published results. The closest adjacent causal result is Schneider et al. 2022 (Educational Researcher 51(2):109-121), a cluster RCT of a high school chemistry/physics project-based learning package in >4,000 students that found ~0.20 SD on an independently developed summative assessment and met WWC standards with reservations.