Same Test, Three Graders, One Rubric

The spread is more likely a rater-and-rubric problem than a difference in student learning, but the three teachers teach different sections, so student differences could contribute. I need evidence before concluding. First, check whether the sections were comparable: prior unit quiz means, ELL and IEP composition, and whether the modeling task was taught with the same materials. Second, sample 12 papers across the score range, remove names, and have all three teachers independently rescore them. Compute exact agreement and adjacent agreement; look at which criterion drives disagreement, especially 'model clearly explains' and 'uses appropriate vocabulary.' Third, examine item-level means: if multiple-choice means are similar but modeling means diverge, the task and rubric are the problem. Fourth, review student work for evidence that raters weighted neatness or length. I would bring anchor papers and a draft analytic rubric, not accusations. The goal is calibration: same evidence, same score. If after calibration the section means still differ, then investigate opportunity to learn-time on modeling, access to particle diagrams, and feedback-rather than assuming rigor differs.

Criterion: Explains floating with a particle-level model of density. 4-Model shows water molecules in ice held in an open hydrogen-bonded lattice with greater average spacing than in liquid water; explicitly links the lower particle density of ice to lower mass per volume and to buoyancy; labels are consistent. 3-Shows greater spacing in ice and states ice is less dense, but the link to buoyancy or the particle representation is incomplete. 2-Shows particles but spacing is not clearly different, or says 'ice is less dense' without a model-based reason. 1-Uses only macroscopic language ('ice is lighter') or a misconception ('molecules expand'). 0-No relevant model.

I would separate this row from a vocabulary row so a student can score 4 on reasoning with imperfect spelling. Calibration: all three teachers score the same five anchor papers blind, discuss one point of disagreement at a time, revise descriptors, then score a fresh set of 10. We would set a target of exact agreement within one level and double-score any paper near a grade boundary. We would also agree that labels in Spanish count if the scientific relationship is clear; the construct is bonding, not English spelling.

First, determine whether the scores reflect the science construct or the task's language and response demands. The IEP student's strong multiple-choice score and weak modeling score could mean the graphic organizer was not actually available during the task, that the task required holding too many constraints in working memory, or that the rubric penalized conventions. I would check the accommodation log and the student's work. The emergent bilingual student's correct particle arrangement and Spanish labels show science understanding; a 1/4 on explanation may be a language score masquerading as a science score. I would allow a bilingual glossary, accept labeled diagrams and short oral explanations, and score scientific accuracy separately from spelling and grammar. That is not lowering the standard; it is removing construct-irrelevant barriers. At the same time, I would not accept 'ice is lighter' as a full explanation. I would provide sentence frames and a word bank for all students, then use a common rubric. If the student cannot yet express the density relationship in English, an oral defense with a bilingual peer or translator can provide valid evidence. The department should document these decisions so rigor means evidence of reasoning, not compliance with one mode of expression.