Teaching Excellence · Domain 4 · Classroom Climate
What the room said
Read this lightly. The survey was a hook: a 30-second warm-up to open a 5-minute talk, written to make one point memorable. It was not a study, and it was never a test of anyone. The numbers below are honest, but they describe what one room of colleagues tapped on their phones during a PD session, not how anyone teaches. See the limitations before quoting any of it.
A word to everyone who got version A. If you read that student as making a choice, you did exactly what good teachers do all day: you responded thoughtfully to what was in front of you. Nothing in your version hinted at what was going on at home. That is the whole point of the exercise. The backstory is invisible in real life too, so no one “missed” anything. And even without it, most of you chose to quietly check in with him first. That says a lot about this faculty.
In short
- Three sentences changed how the room read the same student.With only what they could see, colleagues read the behavior as mostly a choice (5.1 of 7), a reasonable read of the information they had. Those who also learned his mother had just been diagnosed with cancer rated it 3.7. Very few took it personally in either version: “disrespect” was low from the start (2.6) and dropped a little more (1.9).
- The same people moved when they learned the news.After answering, the version-A group saw the backstory and answered again. 25 of 33 lowered their "choice" rating, by 1.6 points on average. Only one raised it.
- The room’s first instinct was already kind.Even without the backstory, 81% chose a quiet check-in first and only 2 of 37 chose a warning. The information changed how people read the student far more than what they would do, because what they would do was already caring.
What people saw
Each phone was randomly assigned one of two versions, balanced in pairs so the room split evenly. Nobody was told there were two versions.
Both versions: Third period. A junior walks in just after the bell and drops into his seat. When you ask him to take out his notebook, he says “I’m good” without looking up, and puts his head down.
Everyone answered three questions: how much the behavior is a choice (1–7), how much it is about disrespecting the teacher (1–7), and their first move (five options, shown in a random order). After submitting, version A saw the same backstory (“What you didn’t know”) and answered the three questions again. The scenario is adapted from the “Julie” case in Souers & Hall, Fostering Resilient Learners, ch. 2.
Responses came in between 8:06 and 8:22 a.m. Central: 37 in version A and 35 in version B. 33 of the 37 version-A phones also answered the follow-up.
Version A vs. version B
The two groups overlapped a lot, and both kinds of answer are defensible. Version A clustered around 5. Version B spread across the whole scale: some colleagues felt that a student who is grieving still makes choices, and that’s a fair view too.
The same people, before and after
On “choice,” 25 lowered their rating, 7 kept it, and 1 raised it. On “disrespect,” 27 lowered it, 5 kept it, and 1 raised it. Changing your answer once you learn something new is exactly what we’d hope for. This is the most striking result, but also the one to lean on least as evidence (see limitations): everyone was told the news and then asked again.
First moves
| First move | Version A n = 37 | Version B n = 35 | A, after learning n = 33 |
|---|---|---|---|
| Quiet check-in after the task starts | 81% (30) | 71% (25) | 67% (22) |
| Email his counselor | 8% (3) | 9% (3) | 24% (8) |
| Leave him alone | 5% (2) | 9% (3) | 9% (3) |
| Warn him or give a consequence | 5% (2) | 11% (4) | 0% (0) |
| Give him a demerit | 0% (0) | 0% (0) | 0% (0) |
Almost no one reached for a consequence in either version: 2 of 37 in A and 4 of 35 in B chose a warning or demerit (p = .355; with numbers this small, the slight difference means nothing). After learning the news, 2 of the version-A group who had chosen a warning switched away from it, and nobody switched to one (exact McNemar p = .500). The noticeable shift was toward emailing the counselor: 8 of 33 after the reveal, up from 2 before.
What surprised us
Setting the caveats aside for a moment, here is what stood out when we looked more closely at the answers.
1. “It’s a choice” didn’t mean “it needs a consequence.”
In version A, 28 colleagues rated the behavior a 5 or higher, mostly or completely a choice. 23 of those 28 still chose a quiet check-in as their first move. And the few warnings in version B came mostly from people who rated it low on choice (3, 3, 3, 5). In this room, how people judged the behavior and how they chose to respond were almost unrelated. You can think “he’s choosing this” and still meet him with care, and that’s what most of the room did.
2. Learning it late and knowing it from the start ended in the same place.
After the reveal, the version-A group’s “choice” ratings averaged 3.45. Version B, who knew from the start, averaged 3.66, and the two spreads of answers look almost identical. People who had already given an answer didn’t hold on to it; they moved right to where the informed group was. On “disrespect” they went a little further (1.48 vs. 1.94). That openness to new information is a good sign for any conversation about students.
3. Knowing more made the room disagree more.
Without the backstory, answers clustered tightly around 5 (standard deviation 1.31). With it, they spread across the whole scale (1.59 in version B, 1.68 after the reveal), and about a third of version B (12 of 35) still rated it 5 or higher. Context didn’t produce one “right” answer. It opened a real question (how much responsibility does a grieving 16-year-old carry?), which might make a great conversation for a department meeting.
4. The most certain moved the most, and a thoughtful few held steady.
In the follow-up, colleagues who first answered 7 (“completely a choice”) lowered their rating by 2.0 points on average, those who answered 5 or 6 by 1.8, and those at 4 or below by 1.0. In all, 7 people kept the same rating, 6 of them at 5 or above. That’s a coherent view, not stubbornness: you can feel deeply for a student and still believe he is choosing how to respond to a hard day.
5. Seeing a choice and feeling disrespected are separate things.
Only 6 of 37 colleagues in version A rated “disrespect” 4 or higher, even though most rated “choice” high; the two answers were only loosely related (correlation 0.43). The faculty’s default seemed to be “this may be a choice, but it isn’t about me.” Not taking a student’s bad moment personally is one of the hardest and most important parts of classroom climate, and this room already does it.
6. The one action that clearly changed: bringing in the counselor.
After the reveal, emailing the counselor went from 2 to 8 of 33 in the version-A group. Curiously, version B, who knew from the start, chose it only 3 of 35 times. Learning the news after forming a first impression seemed to make people want to be sure someone else knew and could support him. It’s a small number, but a lovely instinct.
What we might take from it
- Our instincts are a strength to build on. In this room, the first move was already a check-in, even when people read the behavior as a choice. So the book’s message may fit us best as a nudge about how we interpret students (which shapes our tone, our follow-up and how we talk about a student with colleagues) rather than as a correction to how we react.
- “As if” is the practical takeaway. Version B is the information we almost never get. Since the room read the student differently once it knew, a gentle habit worth sharing is to leave room for a story we can’t see, especially on the days a student is hardest to read.
- A future scenario could be harder. A case where the caring move is less obvious (a repeated pattern, a public challenge in front of the class) would tell us more about where colleagues genuinely differ, and would make for a richer conversation.
Limitations
- It was a hook, built to make a point. The scenario, the backstory and the questions were chosen to make a contrast vivid, and were tuned in the days before using AI role-play simulations. Nothing was pre-registered, and the p-values are descriptive, not tests of a hypothesis we committed to in advance.
- Who answered, and how. 72 phones from one faculty on one morning. We don’t know how many people were present but didn’t scan. Answers were anonymous and tied to a browser cookie, so someone could have answered twice from a second browser. There were no attention checks.
- The setting hints at the topic. This happened during a talk titled “Classroom Climate” at a PD session, and version B’s backstory sat in a highlighted box. It’s natural to answer with the talk’s theme in mind, and that alone could account for part of the difference.
- A calm room is not third period. A phone survey has no 25 watching students, no time pressure, no history with the student and no hard morning of our own. The answers are what we’d hope to do on a good day, which is valuable, but it isn’t the same as observing practice.
- The follow-up is a demonstration, not evidence. Version A was told the news and then asked the same questions again, so ratings would naturally drop. It shows how much the context mattered to each person, which is personal and hard to generalize.
- Phones could be compared. Neighbors may have seen each other’s versions. That would tend to shrink the difference, not create it.
- Single, made-up questions. Each idea was measured with one item written for this talk. “Choice” in particular can be read many ways, and there was no wrong answer to it.
- Tiny counts for consequences. Six warnings in total across 72 responses. Nothing about first moves should be read as a finding.
- One school, one case. One scenario about one kind of hardship, answered by our own colleagues. It says nothing about other schools, other students or other situations.
Numbers, for reference
| Measure | Without backstory | With backstory | Test |
|---|---|---|---|
| Choice, A vs. B | 5.11 [4.67, 5.54] | 3.66 [3.11, 4.20] | Welch t(65.9) = 4.22, p < .001, d = 1.00 |
| Disrespect, A vs. B | 2.59 [2.26, 2.93] | 1.94 [1.58, 2.31] | Welch t(69.3) = 2.67, p = .009, d = 0.63 |
| Warn or demerit, A vs. B | 5% (2/37) | 11% (4/35) | χ²(1) = 0.85, p = .355 |
| Choice, A before → after | 5.06 [4.62, 5.50] | 3.45 [2.86, 4.05] | paired t(32) = 6.53, p < .001, dz = 1.14 |
| Disrespect, A before → after | 2.61 [2.25, 2.96] | 1.48 [1.22, 1.75] | paired t(32) = 7.21, p < .001, dz = 1.26 |
| Warn or demerit, A before → after | 6% (2/33) | 0% (0/33) | exact McNemar, p = .500 |
Ratings are means with 95% confidence intervals in brackets. Shares are of each group.
How the AI preview compared
Before the talk, we simulated the survey with 20 AI-written teacher personas. It predicted the “choice” gap well (it expected about 4.5 vs. 3.2; the room gave 5.1 vs. 3.7) and matched version A’s “disrespect” rating almost exactly (2.6 vs. 2.6). It badly overestimated warnings: it expected about 30% in version A, and the room gave 5%. Our real colleagues were kinder in their first moves than the simulated ones.