More isn't always better: The curvilinear relationship between inquiry-based teaching and student achievement in science

4,382 grade-8 students in 211 intact science classrooms (Norway, TIMSS 2015)

Norwegian grade-8 science, TIMSS 2015 nationally representative two-stage cluster sample; teachers averaged 13.1 years of experience.

Cross-sectional international-assessment data analysed with multilevel structural equation modelling — no causal identification, so this is grade D under METHODOLOGY regardless of the modelling sophistication. Inquiry is measured by TEACHER SELF-REPORT: five items selected from the TIMSS teacher questionnaire ("how often do you ask them to do the following"), on a 4-point never-to-every-lesson scale, covering observing/describing phenomena, designing investigations, conducting experiments, interpreting data and using evidence to support conclusions. That is frequency of activity, with no information about guidance or quality, which is exactly the conflation this archive warns about. Achievement is the TIMSS science scale (five plausible values, all incorporated). The paper's contribution is to add a QUADRATIC term rather than assume linearity.

The linear association between inquiry frequency and achievement was NOT significant (B=0.072, SE=0.096). Adding a quadratic term produced a significant negative coefficient (B=-0.264, SE=0.130, p<0.05) — an inverted U: a little inquiry is associated with higher achievement, a lot with lower. THE IMPORTANT CAVEAT, WHICH THE ABSTRACT DOES NOT CARRY: once classroom socioeconomic status is added to the model, the quadratic term drops to -0.174 (SE 0.116) and -0.179 (SE 0.131) and is NO LONGER STATISTICALLY SIGNIFICANT, while SES itself is a strong predictor (B=0.635-0.700, p<0.01). The likelihood-ratio tests still favour the curvilinear model over the linear one for every plausible value, so the shape claim has support; the effect-size claim does not survive SES adjustment. Classroom SES did not moderate the inquiry-achievement relationship. The defensible reading is: teacher-reported inquiry FREQUENCY is not linearly related to science achievement, and more is not monotonically better — which is a statement about dose without guidance, not a refutation of guided inquiry.

High for any causal reading. Cross-sectional survey data with no random assignment; classroom composition, family background and reverse causation (teachers assigning more hands-on work to weaker classes) are all live, and the curvilinear term itself weakens sharply once a single composition control is added — a direct demonstration of the confounding.

Consistent in direction with the PISA-based correlational literature already in the archive (OECD 2015; Jerrim 2019) and with Aditomo and Klieme 2020, which resolves the same pattern into guided versus independent inquiry. The Norwegian TIMSS 2015 result has not to our knowledge been directly replicated in other TIMSS countries with the same quadratic specification.