What a 68 Percent Average Is Hiding

What the data support.

• A specific, systematic gap. Items 4, 11, and 17 measure the same subskill - reading slope as a rate with units in a real context. Missing all three is not carelessness spread across the test; it is a single conceptual target that was not learned. A 68% average that pools those items with procedural items hides a hard edge.
• The distractor is diagnostic, not noise. Forty-seven percent choosing 'the y-intercept is the rate of change' tells you students have memorized that m and b are the two important numbers in y = mx + b but have attached the labels to the wrong symbols. That is a labeling error layered on an interpretation error, and it will not be fixed by more graphing practice. The reteach must contrast the two quantities directly: what does b mean in a taxi fare, what does m mean, which one changes when the ride gets longer.
• Procedural skill is intact. Item 19 at 91% says solving for a variable is not the problem. Re-teaching procedures would consume the two weeks and address nothing.

Errors to avoid.

1. Treating 68% as 'middling.' The mean conceals a bimodal distribution - students who can compute but not interpret, and students who cannot do either. She needs the distribution, not the average, and per-item data disaggregated by section.
2. Concluding students 'did not study.' The homework data show the opposite: work was being completed, at least at home.
3. Re-teaching the whole unit. Two of the three problem clusters are mastered.

What to look at next. Item-level results by section to see whether one class period is an outlier (a scheduling or instructional difference worth investigating), and the written work on items 4, 11, and 17 to see whether students can compute slope but not name it in context, or cannot do either.

The problem: invalid evidence inflating a grade. The two students' homework scores are perfectly accurate records of what a parent produced, and perfectly useless as evidence about the students' learning. With homework at 30% of the grade, a student can pass the course while failing every test - which means the reported grade stops functioning as a statement about what the student knows. This is a validity problem before it is a fairness problem.

What the weighting does and does not show. Homework at 30% is defensible if homework is being graded for practice and completion and is reported as such. The defect is not the percentage; it is that the same 30% is being treated as achievement evidence when it is not comparable across students. What she needs is a source of independent evidence of the same skills.

Concrete steps.

• Keep the practice grade but change its content. Grade three items per assignment for completion and one for accuracy, and say so in writing to families.
• Add a short, in-class, unassisted skills check (four to six items, ten minutes) once a week. It is quick to score, it is independent evidence, and it can carry real weight.
• Add an oral or written 'explain your step' component on a few homework items, completed in class, so the parent-corrected answer can be traced back to the student.
• Contact the two families. Frame it as partnership, not accusation: the parent's help is valuable, but the teacher needs to see what the student can do alone so she can target instruction. Ask the parent to let errors stand.

Constraints she must respect. Changing weights mid-quarter is usually governed by district or school policy and requires administrator approval plus notification to students and families; she cannot simply re-weight retroactively. She should propose the change for the next grading period, document it, and meanwhile add the in-class checks - which are simply additional evidence, not a change to the syllabus. Any change should be checked against whether it disproportionately penalizes students whose only support at home is a parent.

Design constraint. She cannot re-teach the unit and she cannot skip the gap - slope as a rate is the foundation of the next unit on linear models and systems. The solution is a small, targeted intervention embedded in the new unit rather than a detour before it.

Plan.

• Ten minutes, three days a week, for two weeks, at the start of class: one contextual slope problem per day, drawn from the new unit's content (cost per mile, rate of change in a graph of a draining tank). Keep the context; drop the volume of items.
• Contrast pairs. Each set includes one item asking for the rate and one asking for the starting value, in the same context. Students must name which is which before computing. This attacks the exact confusion the distractor revealed.
• Small-group pull-aside during independent practice for the students who missed all three items, using a worked example and a fading prompt sequence (full example, then completion, then independent).
• Peer explanation routine: two students explain to each other what the slope means in words with units, then trade. Cheap, high frequency, and it makes the interpretation verbal.
• Do not replace the new unit. Slope interpretation is spiraled into it, which is both efficient and better for transfer than isolated review.

Data to collect within one week.

• A five-item exit ticket with three parallel slope-in-context items and two items from the new content. Score by item, not total. Success criterion: the share of students missing all three slope items drops from roughly 60% to under 35%, and no student repeats the intercept/rate label swap.
• Error coding on the distractor: is the label swap decreasing, or have students shifted to a different error (e.g., dividing in the wrong order)? A shifting error is progress; a stable error means the contrast pair was not strong enough.
• Comparison with another section teaching the same new unit, if a colleague will share exit-ticket data, to separate her intervention effect from the difficulty of the new content.
• Homework accuracy on slope items, now that she is grading a few items for accuracy.

Decision rule. If by day five fewer than half the target students show improvement, the problem is not dosage - it is that the misconception is deeper than labeling, and she should request a brief diagnostic conversation with the math interventionist before the benchmark. Setting that rule in advance keeps the decision data-driven rather than driven by how the lessons felt.