revise our panels using feedback: refit lines when an outlier pulls them, fix models and percents, and extend a finding with column relative frequencies.
Phase 5: Prove It
Do Now. Read NIST’s short page about a plot with one outlier. Then predict.
If you include one far-off outlier when you fit a line, what happens to the line?
The reading opens over the lesson; close it to come back to exactly where you were.
Revision is where good work gets great. Today you use your gallery feedback. Most fixes fall into three groups: the model, the numbers and the words.
Worked example 1: Priya refits around an outlier. Priya’s team added sample student K (9, 62) to the sleep data and drew a line through (6, 70) and (9.5, 84) to “split the difference.” Slope: 14 ÷ 3.5 = 4, and 70 = 4(6) + b gives b = 46, so y = 4x + 46. She checks it: at 9 hours it gives 82, but student I scored 88; at 8 hours it gives 78, but F and G scored 82 and 85. All ten regular points are on or above the line, with gaps up to 7 points. Like NIST says, the fit is poor almost everywhere. Without K, y = 6x + 34 stays within 3 points of every regular point. Her revision keeps y = 6x + 34, keeps K circled on the plot, and adds: “Student K was left out of the line because the point is far from the pattern.”
Worked example 2: Andre extends a table finding. His panel showed row percents (30% vs 60%). A board member might ask the reverse question: of the students who got fewer than 8 hours, what share had phones? Column: 42 ÷ 58 ≈ 0.724, or about 72%. Of those who got 8 or more hours, 18 ÷ 42 ≈ 43% had phones. He adds both, labeled as column percents.
Fix the words. Replace cause words (makes, causes, leads to) with association words (tended to, was more likely, is associated with). Add the model’s range: “for 6 to 9.5 hours of sleep.”
Extension choices. (1) Add a second two-way table from the survey. (2) Compare your sample model with a classmate’s model shown a different way. (3) Write a “What we would study next” box that names a lurking variable.
Practice the three kinds of fixes, then revise your panel.
Revision work time.
Make at least two revisions to your panel: one to a number or model and one to the wording. Show each check you used.
Choose one extension and complete it. Explain how it makes the dashboard more useful to a school board member.
Every revision is evidence that you listened and thought again.
Exit ticket: what was your most important revision today, and how do you know the panel is better now?
Pacing: warm-up and reading 6 minutes, mini lecture 8, guided practice 9, revision work 15, closure 2. This is a flex revision and extension lesson; students who finish revisions early choose an extension.
Likely misconceptions: deleting an outlier from the plot; thinking a line that touches more points must fit better; mixing up row and column percents.
Supports: small groups from the 1.23 exit tickets; a revision log with columns for feedback, change and criterion. Stretch: compute residuals for both lines and add them up.
Answer guide: warm-up: pulled toward the outlier; slope 4; 82; true; about 72%; “tended to go with”; 6 free throws for 10 minutes. Priya’s pulled line y = 4x + 46 gives 70, 72, 74, 74, 76, 78, 78, 80, 82, 84 for students A-J, so the residuals are 0, 0, +1, +4, +4, +4, +7, +4, +6, +6 (all ten on or above, none below) and K is −20. With y = 6x + 34, the largest gap for A-J is 3. Andre: 42 ÷ 58 ≈ 72% and 18 ÷ 42 ≈ 43%.
National Institute of Standards and Technology, NIST/SEMATECH e-Handbook of Statistical Methods: 1.3.3.26.10 Scatter Plot: Outlier · U.S. government work, public domain
The NIST handbook’s example gallery includes a scatter plot with an outlier. Most of its points show a basic straight-line relationship between X and Y. One point, at X = 375, sits far away from the rest.
The handbook defines an outlier in a special way: a data point that seems to come from a different model than the rest of the data. The other points look like they follow one line with a certain slope and spread. The outlier looks like it was produced by something else.
The handbook says finding outliers is important for good modeling. If all the data, outlier included, are used to fit a line, the fitted model will be poor almost everywhere. If the outlier is left out of the fitting, the fit will be excellent for every point except the outlier itself. For that reason, NIST says outliers should be left out when the model is fitted.
Leaving a point out of the fit is not the same as hiding it. On our dashboard, the outlier stays on the plot, circled, with a note that explains why the line does not follow it. Honest analysts show the reader everything they saw.
Source: National Institute of Standards and Technology, NIST/SEMATECH e-Handbook of Statistical Methods: 1.3.3.26.10 Scatter Plot: Outlier (itl.nist.gov)