Statistical Functions and Copilot as a Checked Assistant

Describe a distribution honestly, then let Copilot draft formulas you can still verify.

This week

Week 11 · Excel Ch. 8 · Excel Projects 12–13. Next: Python as automation (lecture 10).

Objectives

After this lecture you should be able to
  • Explain why mean, median, and spread can disagree
  • Compute AVERAGE, MEDIAN, STDEV / STDEV.S, and a simple percentile on a Table column
  • Use Copilot (or an equivalent assistant) to draft or explain a formula
  • Verify against a hand-checked example, a boundary, and a type trap
  • Distinguish allowed assistance from submitting unreviewed generated work

A distribution is more than a mean

The mean is one sentence about a list
  • AVERAGE of {10, 10, 10, 100} is pulled by 100 (lecture 4)
  • Median: the middle after sorting — often closer to “a typical row”
  • Spread: standard deviation, IQR, min/max — “how much do rows disagree?”
  • A PivotTable of averages by group can hide that one group is tiny (lecture 7)
Science, not vibes
  • GE Goal 3: a number is a measurement with a method
  • Report: “median attendance 42 (n = 18); mean 61 because of one festival”
  • A chart from lecture 5 (histogram or sorted column) is the picture of the same claim
  • Dropping the festival without saying so is not “cleaning”; it is a different dataset

Statistical functions in Excel

A starter set (Ch. 8)
Question Function Note
Typical value AVERAGE, MEDIAN Blanks vs zeros (lecture 4)
Spread STDEV.S (sample), STDEV.P (whole population) For a class of these students, sample vs population is a judgment
Position MIN, MAX, QUARTILE.INC / PERCENTILE.INC Know inclusive vs exclusive if the textbook distinguishes them
Count in a band COUNTIFS “How many between 0.7 and 0.8?”
Rank RANK.EQ Ties
STDEV.S vs STDEV.P
  • Sample (S): you have some of the rows, infer spread
  • Population (P): this list is the whole set you care about
  • For a single course roster, instructors often treat the roster as the population — say which you used
  • Copilot may pick .P or .S at random; you choose from the question
Check a tiny list by hand
  • Data: 2, 4, 4, 4, 5, 5, 7, 9
  • Median, mean, and a rough sense of spread on paper
  • Then Excel — if they disagree, your range includes a header or a total row

What Copilot is and is not

A practical definition
  • Copilot in Microsoft 365 is a language model wired into Excel, Word, PowerPoint, and Outlook
  • It predicts useful text and formulas from your prompt and (sometimes) the file context
  • It does not know whether STDEV.S matches your scientific claim
  • It does not replace SLO5: you determine correctness
Strengths
  • Remembering syntax (PERCENTILE.INC vs older PERCENTILE)
  • Explaining a formula you do not yet read fluently
  • Drafting a caption or data-dictionary stub in Word
  • Suggesting a PivotTable layout from a stated question
Failure modes you should expect
  • Invents a column that is not in the sheet
  • Mixes up relative and absolute references
  • Treats IDs as numbers, dates as text, or percents as already-multiplied values
  • Sounds confident while being wrong (fluency ≠ accuracy)
  • Picks STDEV.P when you meant a sample — or the reverse

Draft, explain, verify

The same loop as every Excel week
flowchart LR
  Q[Your question] --> P[Prompt]
  P --> D[Draft formula or text]
  D --> E[Explain it back in your words]
  E --> V[Verify on known rows]
  V --> K[Keep, fix, or discard]
  • If you cannot explain the draft, you do not understand the solution yet
  • If it fails the known list (2,4,4,4,5,5,7,9), it is not done
Verify like a lab
  • Keep a known example: tiny list or one student computed by hand
  • Test a boundary: exact 70%, blank, zero
  • Test a type trap: ID with a leading zero
  • Change one input; confirm the output moves in the direction you expect
Prompts that help
  1. Question: median lab score for on-time rows, or STDEV.S of Weighted
  2. Shape: Table name, column headers, types
  3. Success: “Match the paper median of the Checks list; ignore the Totals row”
  • Weak: “Do stats on this sheet”
  • Stronger: “Table Scores, column Weighted (numbers 0–1). MEDIAN and STDEV.S of that column only, not the Totals row. Explain sample vs population in one sentence.”
Copilot in Word and PowerPoint
  • Word: turn the stats into a methods + results paragraph; you still own the numbers
  • PowerPoint: three slides (question, distribution, caveat) — no invented statistics
  • Ask Copilot to outline; you paste values from Excel yourself

Academic integrity

Allowed in this course (unless an exam says otherwise)
  • Using Copilot to learn syntax and to draft formulas you then test
  • Pasting a generated formula after you can explain it and after it matches your checks
  • Citing assistance in a brief note when the assignment asks how you worked
Not allowed
  • Submitting a workbook you cannot explain in office hours
  • Generating an entire project and turning it in unreviewed
  • Using Copilot during a closed-resource Excel exam
  • Presenting Copilot’s prose as your reflection without revision (GE Goal 1a)

Try this in Microsoft 365

Lab 12 — Describe a distribution
  1. On Scores[Weighted], compute n (COUNTA or COUNT), AVERAGE, MEDIAN, MIN, MAX, STDEV.S
  2. Hand-check median on a copy of 7 values in Checks
  3. One-sentence result + caveat in Word (outlier? missing zeros?)
  4. Optional: quartile box or a simple column chart of sorted totals (lecture 5)
Lab 13 — Audit a generated formula
  1. Ask Copilot for nested IF letter grades from Weighted or for a COUNTIFS band
  2. Paste into a new column, not over working formulas
  3. Hand-compute two students and the tiny Checks list
  4. Find at least one bug or ambiguity (boundaries, rounding, Totals row, .P vs .S)
  5. Fix it yourself; comment: what Copilot got wrong
  6. Close the assistant; write the explanation of the fixed formula from memory
Discussion
  • Designate a note-taker
  • If Copilot and your hand calculation disagree, which do you trust first?
  • How is this different from asking a classmate to “just send the file”?

Takeaways

Carry these into Python
  • Mean, median, and spread are different claims; say which you computed
  • Copilot is a productivity tool inside the question → check loop
  • Fluency is not correctness; known rows and boundaries are
  • Next: when the same Excel steps must run every week, Python repeats a verified analysis
Placeholder notes for expansion
  • Histogram / FREQUENCY if the textbook covers it
  • Screenshots of Excel Copilot vs formula bar
  • Department AI-use statement, if updated