Assignment 4 — Data Honesty (Block IV)

Due: week 28 · Weight: see assessment plan · Covers LO7, LO9, LO10 · Submit code + report.

This assignment is about the two data-science skills the course has been building since Block I: making a summary that survives scrutiny, and simulating what you cannot measure — without pretending the simulation is measurement.

The scenario

The university's sports centre claims: "Average gym visit lasts 52 minutes." You are given its log for one week — as a described dataset, not a file (construct it yourself per the spec below, with your generation code submitted).

The log (per visit): a start minute (0–959), a duration in minutes, and a member flag. The week's reality, which you must reproduce exactly:

• 400 visits on weekdays (Mon–Fri) in total, 150 at the weekend;
• durations: typical visits 30–70 minutes, but 5% of visits are equipment-tour sessions of exactly 15 minutes and 2% are half-day passes logged as 240 minutes;
• 30 of the weekday logs have duration missing (recorded as −1);
• 8 visits are exact duplicates of other visits (the door reader re-uploads them).

Task 1 — The claim audit (6 marks)

(a) Compute the claim's value as stated from your constructed log (the mean including everything, after your cleaning decisions) and say whether 52 is defensible on your numbers. Show the arithmetic. (3) (b) Re-compute the mean under two defensible alternative cleanings (e.g. with and without the 240-minute passes; with −1 rows dropped vs repaired). Present all three numbers in a table. (2) (c) In three sentences: which number would you put on the notice, and what must the notice say next to it? (1)

Task 2 — Cleaning ledger (5 marks)

Write the cleaning ledger for your log: every rule applied, the count of rows it touched, and the reason. Duplicates, missing durations, and the extreme values must each appear — with drop / fix / keep-and-label decisions argued, not defaulted. End with the total rows before and after, and reconcile the difference (numbers must add up).

Task 3 — Simulate the uncovered case (5 marks)

The centre plans a third gym floor and needs "how many visitors are inside at the peak minute?" (a) Define one simulation trial precisely: what is random, what comes from the log, when does the trial end? (2) (b) Implement it (standard library only), run 10,000 trials, and report the peak headcount distribution in five buckets. State your seed. (2) (c) Name the assumption in your simulation that the log cannot justify, and say what data would justify it. (1)

Task 4 — The honesty paragraph (4 marks)

One paragraph, ≤ 120 words, addressed to the sports centre: what the data supports, what it does not, and the one number you refuse to print without more data — with the reason. This is marked on calibration (claiming neither more nor less than the evidence), not on length.

Independent-reasoning component

Task 3(c) and Task 4 must be written from your own cleaning and simulation choices — they cannot be copied, because the numbers differ by the decisions earlier tasks forced. State any collaboration in one line at the top (see integrity).

Submission checklist

• Generation code + cleaning code + simulation code (standard library only; seed recorded).
• The cleaning ledger with reconciled row counts.
• The three-mean table and the notice paragraph.
• Length guide: code + 3–5 report pages.