Investigate composition, concentration and missingness
Learn the reasoning. Try the task. Compare your answer.
Prepared for DataCareerHub · Curriculum version 2026-10-15.6
By the end of this lesson
- Explain why the aggregate and segments tell different stories.
- Keep verified outliers and labelled sensitivity scenarios distinct.
- Use coverage and sparse-group evidence to limit interpretation.
Fictional guided portfolio · Source from Module 4 · Local spreadsheet or SQLite · Open the worked case study ↗
Understand the method
A portfolio project needs investigation beyond a correct total. Use the monthly baseline to identify a material observation, then ask what could qualify it. Northstar’s completed value rises by €520, or 17.9%, with the same completed-order count in each month. That implies a higher overall AOV, but it does not establish broad improvement within every group or a beneficial effect of an initiative.
Examine the composition
Web AOV falls from €100 to €90 and store AOV from €550 to €180. Store’s share of completed orders changes from 10% to 90%. Overall AOV uses the actual order shares: April is 0.9 × 100 + 0.1 × 550 = €145; May is 0.1 × 90 + 0.9 × 180 = €171. The aggregate rises while both components fall because their weights change. Show the counts and equations, not only a name for the pattern.
An equal-channel scenario would be €325 in April and €135 in May. Label those as hypothetical 50/50-mix averages. They are not observed AOVs and should not replace actual totals. Equal weights are a teaching choice, not an official target or a causal adjustment. Explain why you chose a scenario and what question it answers.
Check concentration
Across 40 completed orders, the mean value is €158 and the median €110. Using the inclusive quartile convention from Module 4 gives Q1 €100, Q3 €180 and an upper 1.5-IQR fence of €300. A020 at €1,000 is the only flagged high order. The reference verifies it, so it remains in the baseline. A statistical flag is a request to investigate, not authority to delete.
P7 produces frequency counts of 7, 14, 18, 0 and 1 for the specified EUR bins. P8 reports the all-completed baseline separately from a named A020-exclusion scenario. The latter has 39 orders, €5,320 and mean €136.41. This is a changed population, not a corrected baseline. April store’s average is especially fragile because its group contains only two orders, including A020.
Keep missingness in view
April delivery is measured for 14/20 completed orders; May for 19/20. The observed means, about 3.64 and 3.53 days, describe those measured subsets. Filling unknown durations with zero would create a different, misleading calculation. All seven missing completed durations belong to web orders in this fixture, a pattern worth investigating without claiming to know its cause.
Connect findings to the decision
The investigation supports a careful next step: check comparable periods, explain the channel-volume change and obtain costs before recommending expansion. The source contains no margin, advertising spend or experiment assignment. Do not transform a descriptive association into a causal story. In the portfolio, show at least one alternative interpretation and the evidence that would distinguish it. This demonstrates judgment rather than merely producing more charts.
Practical exercise
Complete P7/P8 and prepare a small investigation page. Include actual weighted AOV equations, an explicitly hypothetical equal-mix comparison, the distribution/verified high-value observation and a delivery-coverage table. Write one alternative explanation for the aggregate change and one source request that would help assess it.
Deliverable: a labelled baseline/sensitivity table and a concise investigation note. Keep every exclusion, scenario weight and population change visible. Do not change the source merely to improve the appearance of a trend.
Open the worked solution
| Analysis | Worked result |
|---|---|
| Actual AOV | April €145 → May €171 |
| Channel AOV | Web €100 → €90; store €550 → €180 |
| Store order share | 10% → 90%, +80 percentage points |
| Hypothetical equal-mix AOV | €325 → €135, not actual AOV |
| Completed baseline | 40 orders / €6,320 / mean €158 |
| Exclude verified A020 scenario | 39 / €5,320 / mean €136.41 |
P7’s five counts sum to 40. Retaining a zero-count €200–<€300 bin is useful: its absence would otherwise be ambiguous. The final bin includes €1,100 at its upper boundary, even though this fixture’s highest order is €1,000. Check the stated bin convention when reproducing the graphic.
A reasonable alternative explanation is a shift in the mix of recorded orders rather than uniform within-channel improvement. The sparse April store group and large verified order also influence the comparison. Request wider comparable channel history and source context for the mix change. These observations do not identify the initiative’s incremental impact or profitability.
One more question.
How these learning materials are prepared and maintained · Report a correction: support@datacareerhub.io
Read and practice without an account. Sign in with an existing account to save lesson completion.
Next lesson →Fictional training data. Portfolio practice is not a claim of client work or an accredited qualification. Explore practical lessons, worked examples and projects. Follow a published path, practise the method and build work you can explain.