Ten students take a test.
The class average is 70.
That single number is useful.
It is also radically incomplete.
Quick Read
Aggregation combines many observations into fewer values: totals, averages, rates, medians, proportions, categories or grouped summaries. It allows humans and systems to reason about large populations without carrying every individual record at once.
One-sentence answer: aggregation is lossy because many different underlying distributions can produce the same summary, causing internal variation and individual differences to disappear.
The Same Mean Can Hide Different Worlds
Imagine one class where every student scores 70.
Now imagine another where half score 40 and half score 100.
The mean is 70 in both.
The educational problem is completely different.
Totals Hide Distribution
A city may report rising total income while many households become poorer.
A hospital may report stable average waiting time while a small group waits far longer than before.
A school may improve its overall pass rate while one subject deteriorates.
Aggregation preserves one property and discards many others.
Group Statistics Can Mislead About Individuals
A relationship observed across groups does not automatically hold for every individual inside those groups.
This is one reason ecological fallacies occur: a group-level pattern is mistaken for an individual-level one.
The aggregate is a different object from the person.
Aggregation Can Hide Minorities
If 95% of users experience a service successfully, an overall success rate looks excellent.
But the remaining 5% may represent a specific group for whom the system fails consistently.
Minority failure can vanish inside majority success.
Bins Are Aggregation Too
Raw ages become 0–9, 10–19, 20–29.
Scores become fail, pass, distinction.
Income becomes low, middle, high.
Binning makes patterns visible and destroys within-bin detail.
Education: A Mark Is an Aggregate
A final mark may combine vocabulary, grammar, comprehension, writing, speed and examination execution.
Two students can receive the same score for entirely different reasons.
Diagnosis requires disaggregation.
Dashboards Need Drill-Down
Good dashboards start coarse and permit zooming into the data beneath the summary.
This preserves usability without pretending the top-line number is the whole system.
A summary becomes safer when it has an upstream route.
A Practical Aggregation Audit
- Summary: what property is the aggregate preserving?
- Distribution: what variation disappears inside it?
- Subgroups: do important groups behave differently?
- Outliers: can rare but consequential cases vanish?
- Level: am I making an individual claim from group data?
- Drill-down: can I inspect the underlying records?
Continue Through eduKateSG
Continue with How Lossy Works | Resolution and How Lossy Works | Smoothing. Resolution collapses fine distinctions; aggregation collapses many records into fewer summaries.
Final Thought
One number can be useful.
One number can also become a hiding place.
Aggregation works best when we remember that the people, events and values underneath it are still there.