Assessing the Effect of Project-Based Learning on Science Learning in Elementary Schools

46 Michigan schools (23 treatment, 23 control), 91 teachers, 111 classrooms, 2,371 grade-3 students

Grade 3 science across four regions of Michigan; 48% female, 12% English learners, 60% free/reduced lunch, 30% Black, 44% White.

Cluster randomised trial of Multiple Literacies in Project-Based Learning (ML-PBL): four NGSS-aligned science units, each framed by a driving question and an anchoring phenomenon and culminating in a student product, plus teacher professional learning and post-unit assessments. Control was business-as-usual grade-3 science. WWC rated it "Meets WWC standards without reservations" with low cluster-level attrition and low individual non-response, and it carries a Tier 1 (Strong) rating on this single study. Two things must travel with the number. FIRST, INDEPENDENCE: Krajcik and Schneider are the creators of ML-PBL and led the evaluation of their own curriculum, funded by Lucas Education Research which also funded the curriculum. Per METHODOLOGY's Replication rule an unreplicated developer-led positive caps a novel practice at verdict `mixed`. SECOND, THE MEASURE IS DISPUTED. The published abstract says students "had higher scores on a standardized science test"; the project's own technical report describes it as "an objective and validated summative test designed by the Michigan Department of Education to meet the NGSS"; but the WWC study review classifies the science achievement outcome as a RESEARCHER-DEVELOPED assessment, as it does all four outcomes. Assembled-by-the-research-team from state-authored items is not the same object as a state accountability test administered to everyone, and no independent state assessment (e.g. M-STEP) result is reported alongside it. This archive records the WWC classification and flags the discrepancy rather than accepting the paper's own label.

Grade-3 students in ML-PBL classrooms outscored controls on the summative science assessment by 0.277 SD (0.266 in the project technical report; WWC improvement index +11 percentile points), roughly an 8-percentage-point difference. Self-reported intrapersonal outcomes also moved: reflection +21, ownership +17, collaboration +16 percentile points — all researcher-developed self-report instruments and all non-cognitive, so they carry no weight for content learning. The headline is real and well-identified as a trial, but it is a developer-led, unreplicated result on a measure the WWC calls researcher-developed and the authors call standardized, with no independent state-test outcome reported.

Low. School-level randomisation with low attrition; the live threats are developer allegiance and outcome-measure alignment, not heredity.

No independent replication of ML-PBL located. The closest independent-outcome evidence for project-based learning in science is Saavedra 2022 (AP Environmental Science, College Board exam), which is externally scored and gives ES around 0.19-0.30.