Exploratory Data Analysis and Pattern Discovery
Mira Dageraad needs to plan warehouse capacity, but an average hides the long waits. Learn to count the right orders, describe the centre and tail, and check unusual records before acting. You work on the records of Raket Parts, a fictional maker of rocket parts.
You will be able to
- describe a distribution honestly and decide which unusual records need checking
- compare groups fairly, show uncertainty and avoid turning a gap into a cause
- run a logged exploration and explain what was checked, what was found and what remains uncertain
- form testable questions
More than an average
Describe a distribution honestly and decide which unusual records need checking.
Fair comparisons
Compare groups fairly, show uncertainty and avoid turning a gap into a cause.
Explore with a log
Run a logged exploration and explain what was checked, what was found and what remains uncertain.
Does the pattern hold?
Count the groups you looked at, register candidates before seeing fresh data, and check each one on a later period as held, weakened or failed.
A full exploration
Run a whole exploration on a new brief, from the first log entry to a findings note that gives each finding its status.
Exploratory Data Analysis / Module 1
More than an average
Mira Dageraad needs to plan warehouse capacity, but an average hides the long waits. Learn to count the right orders, describe the centre and tail, and check unusual records before acting.
Sections
Exploratory Data Analysis / Module 1 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Exploratory Data Analysis / Module 2
Fair comparisons
Bram wants to rank countries by invoice value, but different group sizes and customer types can mislead. Learn to inspect the population, compare like with like and state what a gap cannot prove.
This module builds on module 1. If you are new to this, start with module 1.
Sections
Exploratory Data Analysis / Module 2 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Exploratory Data Analysis / Module 3
Explore with a log
Maaike has a broad question about webshop orders. Learn to form testable questions, keep a five-part record of each check, recognise source breaks, and present candidate findings without overstating them.
This module builds on module 2. If you are new to this, start with module 2.
Sections
Exploratory Data Analysis / Module 3 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Exploratory Data Analysis / Module 4
Does the pattern hold?
Mira wants to know which customers' orders get split into several shipments, since each extra shipment means more packing and freight. A first look at January to March throws up 14 striking groups. Jelmer asks her to count what she looked at, write the candidates down and check them on April to June before anyone acts.
This module builds on module 3. If you are new to this, start with module 3.
Sections
Exploratory Data Analysis / Module 4 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Exploratory Data Analysis / Module 5
A full exploration
Mira's brief is short: “More than half our order lines arrive late. Where should we look first?” This module runs the whole method on it, from the first log entry to a findings note Jelmer can review.
This module builds on module 4. If you are new to this, start with module 4.
Sections
Exploratory Data Analysis / Module 5 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Exploratory Data Analysis / Module 1 / 1.2
1.2 The typical value
Section 1.2 complete
Next: 1.3 Spread and the long tail
Exploratory Data Analysis / Module 1 / 1.3
1.3 Spread and the long tail
Section 1.3 complete
Next: 1.4 Odd values: three kinds
Exploratory Data Analysis / Module 1 / 1.4
1.4 Odd values: three kinds
Section 1.4 complete
Next: Module 1 check
Continue to the module check Recap this section Back to module 1
Exploratory Data Analysis / Module 1 / Check
Module 1 check
Module 1 complete
Next: Module 2, Fair comparisons
Continue to module 2 Try it on your own question Recap module 1 Back to module 1
Exploratory Data Analysis / Module 2 / 2.1
2.1 Small groups jump
Section 2.1 complete
Next: 2.2 Mix inside the groups
Exploratory Data Analysis / Module 2 / 2.3
2.3 Overlap
Section 2.3 complete
Next: 2.4 What a comparison cannot show
Exploratory Data Analysis / Module 2 / 2.4
2.4 What a comparison cannot show
Section 2.4 complete
Next: Module 2 check
Continue to the module check Recap this section Back to module 2
Exploratory Data Analysis / Module 2 / Check
Module 2 check
Module 2 complete
Next: Module 3, Explore with a log
Continue to module 3 Try it on your own question Recap module 2 Back to module 2
Exploratory Data Analysis / Module 3 / 3.1
3.1 From a brief to questions
Section 3.1 complete
Next: 3.2 A log entry has five parts
Exploratory Data Analysis / Module 3 / 3.3
3.3 Breaks in the data
Section 3.3 complete
Next: 3.4 Candidates, not conclusions
Exploratory Data Analysis / Module 3 / 3.4
3.4 Candidates, not conclusions
Section 3.4 complete
Next: Module 3 check
Continue to the module check Recap this section Back to module 3
Exploratory Data Analysis / Module 3 / Check
Module 3 check
Module 3 complete
Next: Module 4, Does the pattern hold?
Continue to module 4 Try it on your own question Recap module 3 Back to module 3
Exploratory Data Analysis / Module 4 / 4.1
4.1 Look hard enough and something appears
Section 4.1 complete
Next: 4.2 Count what you looked at, then register
Exploratory Data Analysis / Module 4 / 4.2
4.2 Count what you looked at, then register
Section 4.2 complete
Next: 4.3 Open the holdout and recompute
Exploratory Data Analysis / Module 4 / 4.3
4.3 Open the holdout and recompute
Section 4.3 complete
Next: 4.4 What the holdout cannot tell you
Exploratory Data Analysis / Module 4 / 4.4
4.4 What the holdout cannot tell you
Section 4.4 complete
Next: Module 4 check
Continue to the module check Recap this section Back to module 4
Exploratory Data Analysis / Module 4 / Check
Module 4 check
Module 4 complete
Next: Module 5, A full exploration
Continue to module 5 Try it on your own question Recap module 4 Back to module 4
Exploratory Data Analysis / Module 5 / 5.2
5.2 Describe how late
Section 5.2 complete
Next: 5.3 Compare fairly and register
Exploratory Data Analysis / Module 5 / 5.3
5.3 Compare fairly and register
Section 5.3 complete
Next: 5.4 Validate and report
Exploratory Data Analysis / Module 5 / 5.4
5.4 Validate and report
Section 5.4 complete
Next: Module 5 check
Continue to the module check Recap this section Back to module 5
Exploratory Data Analysis / Module 5 / Check
Module 5 check
Module 5 complete
Next: Module 6, coming later
Back to the Path Try it on your own question Recap module 5 Back to module 5
Exploratory Data Analysis / Module 1 / Recap
Recap card
1.1 One row, one order
One row, one order
- One order reference is one order, not a shipment.
- Count blank dispatch durations separately.
Common mistake
Calling blank zero. Missing shipment timing cannot be treated as immediate dispatch.
Exploratory Data Analysis / Module 1 / Recap
Recap card
1.2 The typical value
The typical value
- Use the median for the middle recorded order.
- Show the mean too when long waits affect planning.
Common mistake
Mean equals typical. Here the long waits make the mean much larger than the median.
Exploratory Data Analysis / Module 1 / Recap
Recap card
1.3 Spread and the long tail
Spread and the long tail
- Count how many orders exceed a meaningful waiting time.
- Compare the mean with the median to spot a long upper tail.
Common mistake
Hiding the extremes. Reporting only the centre conceals the orders that take much longer.
Exploratory Data Analysis / Module 1 / Recap
Recap card
1.4 Odd values: three kinds
Odd values: three kinds
- Keep genuine extremes and show them.
- Check possible errors and report different case types separately.
Common mistake
Delete every odd row. Removal can hide real operational pressure.
Exploratory Data Analysis / Module 1 / Recap
Recap card
Module 1: More than an average
One row, one order
- One order reference is one order, not a shipment.
- Count blank dispatch durations separately.
Common mistake
Calling blank zero. Missing shipment timing cannot be treated as immediate dispatch.
The typical value
- Use the median for the middle recorded order.
- Show the mean too when long waits affect planning.
Common mistake
Mean equals typical. Here the long waits make the mean much larger than the median.
Spread and the long tail
- Count how many orders exceed a meaningful waiting time.
- Compare the mean with the median to spot a long upper tail.
Common mistake
Hiding the extremes. Reporting only the centre conceals the orders that take much longer.
Odd values: three kinds
- Keep genuine extremes and show them.
- Check possible errors and report different case types separately.
Common mistake
Delete every odd row. Removal can hide real operational pressure.
Exploratory Data Analysis / Module 2 / Recap
Recap card
2.1 Small groups jump
Small groups jump
- Display how many customers support an average.
- Treat tiny groups cautiously and define a minimum before ranking.
Common mistake
Equal trust in unequal groups. A high average from few customers can be unstable.
Exploratory Data Analysis / Module 2 / Recap
Recap card
2.2 Mix inside the groups
Mix inside the groups
- Different kinds of customer can change headline averages.
- Repeat a comparison within one matching segment.
Common mistake
Call a mix difference a location effect. Holding segment constant is a safer starting point.
Exploratory Data Analysis / Module 2 / Recap
Recap card
2.3 Overlap
Overlap
- A difference between averages does not sort every individual record.
- State how much of one group crosses the other group’s middle value.
Common mistake
No overlap from different means. Distributions can overlap even when averages differ.
Exploratory Data Analysis / Module 2 / Recap
Recap card
2.4 What a comparison cannot show
What a comparison cannot show
- Report the group sizes and the matching comparison.
- A difference does not demonstrate its cause.
Common mistake
Turn an observed gap into a causal story. The data here support a comparison, not an explanation.
Exploratory Data Analysis / Module 2 / Recap
Recap card
Module 2: Fair comparisons
Small groups jump
- Display how many customers support an average.
- Treat tiny groups cautiously and define a minimum before ranking.
Common mistake
Equal trust in unequal groups. A high average from few customers can be unstable.
Mix inside the groups
- Different kinds of customer can change headline averages.
- Repeat a comparison within one matching segment.
Common mistake
Call a mix difference a location effect. Holding segment constant is a safer starting point.
Overlap
- A difference between averages does not sort every individual record.
- State how much of one group crosses the other group’s middle value.
Common mistake
No overlap from different means. Distributions can overlap even when averages differ.
What a comparison cannot show
- Report the group sizes and the matching comparison.
- A difference does not demonstrate its cause.
Common mistake
Turn an observed gap into a causal story. The data here support a comparison, not an explanation.
Exploratory Data Analysis / Module 3 / Recap
Recap card
3.1 From a brief to questions
From a brief to questions
- Say which records can answer each question.
- Define the group, measure and date or period first.
Common mistake
Start with an untestable why. Check observable patterns before asking what explains them.
Exploratory Data Analysis / Module 3 / Recap
Recap card
3.2 A log entry has five parts
A log entry has five parts
- Record question, check, result, decision and status.
- A check that goes nowhere still belongs in the log.
Common mistake
Keep only successful paths. Omitting discarded checks makes the exploration impossible to review.
Exploratory Data Analysis / Module 3 / Recap
Recap card
3.3 Breaks in the data
Breaks in the data
- Large changes require a source-quality check.
- Log reconstructed periods and report uncertainty without filling gaps.
Common mistake
Explain away the break. A source reconstruction does not permit an invented trend correction.
Exploratory Data Analysis / Module 3 / Recap
Recap card
3.4 Candidates, not conclusions
Candidates, not conclusions
- Log the number and kind of checks you performed, even dead ends.
- Describe patterns as candidates until independently checked.
Common mistake
Promote exploration to proof. An attractive pattern is not the same as a tested explanation.
Exploratory Data Analysis / Module 3 / Recap
Recap card
Module 3: Explore with a log
From a brief to questions
- Say which records can answer each question.
- Define the group, measure and date or period first.
Common mistake
Start with an untestable why. Check observable patterns before asking what explains them.
A log entry has five parts
- Record question, check, result, decision and status.
- A check that goes nowhere still belongs in the log.
Common mistake
Keep only successful paths. Omitting discarded checks makes the exploration impossible to review.
Breaks in the data
- Large changes require a source-quality check.
- Log reconstructed periods and report uncertainty without filling gaps.
Common mistake
Explain away the break. A source reconstruction does not permit an invented trend correction.
Candidates, not conclusions
- Log the number and kind of checks you performed, even dead ends.
- Describe patterns as candidates until independently checked.
Common mistake
Promote exploration to proof. An attractive pattern is not the same as a tested explanation.
Exploratory Data Analysis / Module 4 / Recap
Recap card
4.1 Look hard enough and something appears
Look hard enough and something appears
- Chance standouts: 14 of 30 groups stood out; 5 repeated
- Gap: a group's share minus its own quarter's overall share
Common mistake
Treating a standout as a finding. INDUSTRIAL's +7.0 in January to March was -1.7 in April to June.
Exploratory Data Analysis / Module 4 / Recap
Recap card
4.2 Count what you looked at, then register
Count what you looked at, then register
- Comparison count: 37 groups, including two cuts never written down
- Register first: candidates are fixed before the holdout opens
Common mistake
Counting only what you wrote. 30 written groups became 37 once n_lines and n_cancelled were counted.
Exploratory Data Analysis / Module 4 / Recap
Recap card
4.3 Open the holdout and recompute
Open the holdout and recompute
- Same definitions: split share of shipped orders, gap to the quarter's own share
- Status: ES, GB, NO, PL held; INDUSTRIAL, IT, NL failed
Common mistake
Comparing raw shares. INDUSTRIAL's 68.3% to 52.8% mixes it with a 7-point fall in every group.
Exploratory Data Analysis / Module 4 / Recap
Recap card
4.4 What the holdout cannot tell you
What the holdout cannot tell you
- Same customers: 389 of 402 Q1 customers are back in Q2
- No full year: seasonal candidates stay candidates
Common mistake
A country pattern that is one customer. Without X-CUS-0001, GB sits at -2.8 and +0.7.
Exploratory Data Analysis / Module 4 / Recap
Recap card
Module 4: Does the pattern hold?
Look hard enough and something appears
- Chance standouts: 14 of 30 groups stood out; 5 repeated
- Gap: a group's share minus its own quarter's overall share
Common mistake
Treating a standout as a finding. INDUSTRIAL's +7.0 in January to March was -1.7 in April to June.
Count what you looked at, then register
- Comparison count: 37 groups, including two cuts never written down
- Register first: candidates are fixed before the holdout opens
Common mistake
Counting only what you wrote. 30 written groups became 37 once n_lines and n_cancelled were counted.
Open the holdout and recompute
- Same definitions: split share of shipped orders, gap to the quarter's own share
- Status: ES, GB, NO, PL held; INDUSTRIAL, IT, NL failed
Common mistake
Comparing raw shares. INDUSTRIAL's 68.3% to 52.8% mixes it with a 7-point fall in every group.
What the holdout cannot tell you
- Same customers: 389 of 402 Q1 customers are back in Q2
- No full year: seasonal candidates stay candidates
Common mistake
A country pattern that is one customer. Without X-CUS-0001, GB sits at -2.8 and +0.7.
Exploratory Data Analysis / Module 5 / Recap
Recap card
5.1 Set up the project
Set up the project
- One row, one order line: the late share counts lines
- Holdout first: April to June is fixed before any pattern is seen
Common mistake
Dropping the open lines. 321 Q1 lines are still NOT_SHIPPED; they are logged as a limit and a question.
Exploratory Data Analysis / Module 5 / Recap
Recap card
5.2 Describe how late
Describe how late
- Middle and tail: median 2 days, average 6.89
- Second group: 485 lines 19 to 41 days late, all CAT
Common mistake
Turning how into why. 437 long-delay lines came in two shipments; that does not say why.
Exploratory Data Analysis / Module 5 / Recap
Recap card
5.3 Compare fairly and register
Compare fairly and register
- 26 groups compared, 15 candidates registered
- Look inside: NO is 80.6% late even without X-CUS-0006
Common mistake
Calling a candidate a cause. The price review note's +5.6 was a gap on 120 lines, nothing more.
Exploratory Data Analysis / Module 5 / Recap
Recap card
5.4 Validate and report
Validate and report
- Status: 8 held, 5 weakened, 2 failed against 53.2%
- Odd values: 5 lines 111 to 166 days late are kept and checked
Common mistake
Reporting candidates as findings. Only what held is a finding, with its limits.
Exploratory Data Analysis / Module 5 / Recap
Recap card
Module 5: A full exploration
Set up the project
- One row, one order line: the late share counts lines
- Holdout first: April to June is fixed before any pattern is seen
Common mistake
Dropping the open lines. 321 Q1 lines are still NOT_SHIPPED; they are logged as a limit and a question.
Describe how late
- Middle and tail: median 2 days, average 6.89
- Second group: 485 lines 19 to 41 days late, all CAT
Common mistake
Turning how into why. 437 long-delay lines came in two shipments; that does not say why.
Compare fairly and register
- 26 groups compared, 15 candidates registered
- Look inside: NO is 80.6% late even without X-CUS-0006
Common mistake
Calling a candidate a cause. The price review note's +5.6 was a gap on 120 lines, nothing more.
Validate and report
- Status: 8 held, 5 weakened, 2 failed against 53.2%
- Odd values: 5 lines 111 to 166 days late are kept and checked
Common mistake
Reporting candidates as findings. Only what held is a finding, with its limits.