Three Data Sets in One Leadership Meeting
You are presenting to the leadership team at a high school of 1,400 students. Three data sets are on the table. First: last year there were 1,180 discipline referrals; Black students are 22% of enrollment but received 46% of referrals coded 'defiance/disrespect' and 19% of referrals coded 'fighting.' Second: a new SEL screener was given in September and January; the schoolwide mean improved by 3 points, but the share of ninth graders in the 'elevated risk' band rose from 18% to 24%. Third: a biology teacher emails you, 'Marcus scored at the 22nd percentile on the fall math benchmark. He can't handle Honors Biology next year - can you pull his file and confirm?'
1. What can and cannot be concluded from the discipline data, and what is your next analytical step?
Read the worked answer
What the numbers support: a disparity exists. Students who are 22% of enrollment receiving 46% of defiance referrals is a relative risk of roughly 2.1 times, and it is large enough to justify an equity review.
What they do not support: any claim about who behaves worse, or any causal statement. These are descriptive counts, not rates, and the share-of-referrals comparison is the least precise version of the statistic.
Next step: rates by subgroup. Referrals per 100 students enrolled, per subgroup, per year - that is the defensible figure and it is the one to put on the slide.
Then split by referral type. If Black students are overrepresented on subjective codes and near-proportional on objective codes like fighting, the likeliest driver is adult perception and decision-making, not student conduct. That is a system question, and it points at adult behavior change rather than student discipline.
Then look at concentration. Which referrers account for the subjective referrals, which class periods, which locations, and how many students account for repeat referrals. Reading 20 actual referral narratives will tell you more than a hundred more correlations.
Present it as a system question, never a student deficit. The framing determines whether staff hears 'our students are violent' or 'our referral practices need calibration.'
2. How do you answer the biology teacher about Marcus?
Read the worked answer
Decode the score first. The 22nd percentile on a fall benchmark is below average but inside the broadly normal range. It is a screener measuring performance on a defined set of skills on one day - it is not a measure of ability, potential, or aptitude, and group-administered benchmarks have limited reliability for high-stakes individual decisions.
Refuse the request as posed. 'Pull his file and confirm' is a request to use a single score to close a gate. The answer is no, and the reason is that placement decisions rest on multiple measures, the student's own goals, and teacher judgment - not on one percentile.
Redirect to what would actually inform the decision. Did Marcus complete the prerequisite? What supports did he have? Is there a teacher recommendation route and does he want to take the course? A student who has never been told honors is an option rarely asks for it, and counselor gatekeeping is one of the quiet mechanisms behind enrollment gaps.
Handle the record properly. I would not conduct this exchange over email or discuss his scores in a group thread. I would speak with the teacher directly and then with Marcus, and route it to the problem-solving team.
Then examine the pattern. If this teacher is filtering honors placement this way, the question is how many other Marcuses never get asked.
3. The schoolwide SEL mean improved while ninth-grade elevated-risk grew from 18% to 24%. How do you interpret this for the team?
Read the worked answer
These two findings are not contradictory, which is the first thing to say. A mean can rise because students at the low end moved slightly while the tail got heavier; averages hide distributions. The team's instinct will be to pick whichever number supports the program.
Ask whether the change is meaningful, not just present. Three points on a screener may be within measurement error. Report the effect size, not only the mean shift, and check the threshold - did the cut score or the norms change between administrations?
Name the rival explanations out loud. Regression to the mean, a fall-to-winter seasonal pattern, response shift after students learn the items, and different completion conditions. Ask who was absent at each administration; if the students at highest risk were missing in September and present in January, the 'rise' is an artifact of sampling.
Cross-validate. Check ninth-grade attendance, nurse visits, counselor referrals, and course failure for the same period. If they moved together, believe it.
Ask the implementation question. Who received Tier 2 support, for how many sessions, and with what fidelity? A program that never reached the elevated-risk students cannot be evaluated by a schoolwide mean.
Close with the limitation. A screener is not a diagnostic instrument and a band shift is not 'six percent more disordered.' Recommend a ninth-grade problem-solving team, targeted supports, and a winter-to-spring comparison with a defined comparison group.