Source: W1 PDF pp. 17-20, 52, 56-60; L1 marking table; L2 introduction and submission checklist; L4 pp. 1, 7.
1.1 The exam strategy follows the assessment
The Week 1 slides describe the exams as written, in person, and applied: you are shown a table, result, short code excerpt or chart and asked to judge it. They explicitly say the papers do not ask for definitions alone. Learn terms so that you can use them to explain a decision; memorising a glossary is insufficient.
A strong answer does four things: makes the requested decision, cites the relevant numerical or contextual evidence, explains the mechanism, and names the remaining limitation. For example: "Keep the verified 47-year-old student. Their age exceeds an IQR fence, but source verification places them in the target population. Deleting them changes who is counted and can lower the reported withdrawal rate without improving any student's outcome."
The assessment weights reported in W1 are: midterm 20% in Week 6; final exam 25%; weekly labs 15% (best 9 of 12); Data Quality Report 10%; EDA and Visualization Brief 10%; Modelling Assignment 10%; capstone and dashboard 10%. These are the supplied slide schedule, not independently checked current administrative instructions. Follow any later announcement from your lecturer.
Scope warning: the uploaded pack is a strong foundation for Weeks 1-5 and preparation for the scheduled Week 6 midterm. It is not complete preparation for an exam covering Weeks 6-12. Confirm the assessed scope with the lecturer.
1.2 A reusable answer method
- Question and claim: what is being asked? Description, comparison, prediction or causal effect?
- Population and row: who should the conclusion concern, and what does one record represent?
- Denominator and units: how many eligible records, how many valid measurements, and in what units?
- Method and evidence: show the calculation or source evidence that supports your choice.
- Interpretation: translate the result into the context rather than repeating a number.
- Limit or next check: identify the specific uncertainty that could change the decision.
For short questions, compress this into two or three sentences. Do not write all six labels mechanically if the question asks only for a calculation. Always answer the requested part first.
Weak: "The data are biased."
Stronger: "Evening students are 20/100 of invitees but only 8/80 of respondents. Complete-case analysis therefore halves their representation from 20% to 10%. The pooled respondent mean may not represent the invited cohort."
1.3 A 14-day preparation plan
This is a suggested plan, not an official course requirement. Allocate about two focused hours per day; extend the calculation sessions if they are new to you.
| Day | Focus | Evidence that you have learned it |
|---|---|---|
| 1 | Week 1: questions, claim ladder, populations and proxies | Repair three vague questions; identify what each claim needs |
| 2 | Week 1 + Lab 1: rows, keys, joins and denominators | Predict merge size and explain each TODO without notes |
| 3 | Week 2: mean, median, variance and SD | Complete the eight-number exercise by hand with units |
| 4 | Week 2: quartiles, boxplots, shape and z-scores | Explain fences versus whiskers and solve three z-score tasks |
| 5 | Lab 2: apply summaries to a decision | Write a 150-word recommendation using your own output |
| 6 | Week 3: quality, conversions and duplicate decisions | Build a decision table with evidence and alternatives |
| 7 | Week 3: missingness, sensitivity and audit | Explain MCAR/MAR/MNAR; calculate representation and bounds |
| 8 | Week 4: distances, scaling, log and encoding | Reproduce the A/B/NEW example and the future value 1.5 |
| 9 | Week 4: availability, leakage and pipeline | Audit a broken workflow; transform the future vector by hand |
| 10 | Lab 4: combined student exercise | Reproduce saved statistics, distances and test error |
| 11 | Week 5: EDA, rates, plots and unknown outcomes | Choose appropriate plots and compare rates with denominators |
| 12 | Week 5: confounding and Simpson's paradox | Reconstruct 66% versus 74% and the common-mix rates |
| 13 | Closed-notes practice exam in this guide | Mark every answer and identify why marks were lost |
| 14 | Repair weak areas, then a mixed retest | Answer changed examples without copying remembered numbers |
Daily pattern: 15 minutes recall yesterday from memory; 35 minutes study; 40 minutes solve or interpret; 20 minutes explain aloud or in writing; 10 minutes record mistakes. Revisit a difficult idea after one, three and seven days. These timings are recommendations, not a scientifically guaranteed schedule.
If you have only three days: day 1 covers Weeks 1-2 and their labs; day 2 covers Weeks 3-4; day 3 covers Week 5 and mixed timed practice. Prioritise row meaning, denominators, SD/quartiles, cleaning evidence, leakage and weighted rates. This compressed plan trades depth for coverage.
1.4 Study actively
Before running code: predict the number of rows, missing values, direction of change or nearest student. After running: cite what happened and explain any revision. Keep your original prediction visible in assessed notebooks.
Maintain an error log with four columns: question, my error, correct reasoning, new example. Categorise errors as arithmetic, denominator, unit, assumption, timing or overclaim. A repeated conceptual error deserves more attention than another reread of a familiar slide.
Use the labs as oral-exam practice: cover the output and explain what each line should do. Change the example: a repeated ID may be a copy in one table and a valid second week in another. Transfer is the skill the exam descriptions emphasise.