Totals do not identify the hidden links¶
Cross-Domain EchoesShared pattern · Aggregation
Imagine one precinct with 100 voters: 50 in group A, 50 in group B, and 50 votes for option X. Those totals fit a world in which all A voters chose X and no B voters did, or the reverse. The aggregate does not identify the link between group and vote. An R&D portfolio chart can hide a different kind of link: several apparently distinct projects may depend on the same unproven component. Their plotted spread need not reveal whether their risks move together. Both examples ask what relationship was lost when the whole was made legible. The voting example is a precise many-to-one inverse problem. A project view is a design choice whose missing dependencies can be recorded explicitly; it is not governed by the same statistical theorem.
Choose a role to see its counterpart in both examples. The diagrams show relationships, not measured quantities.
Population and voting analysis
Different individual patterns fit the same totals
Read Ecological Inference ProblemDomain-specific abstraction
Two hypothetical voting patterns produce identical group totals and vote totals while implying opposite subgroup behavior.
In this example: The toy arithmetic demonstrates non-identification from one set of marginals; it is not a result about any real electorate.
Research and development management
Apparent spread can hide a shared dependency
Read Portfolio ViewMechanism
Projects can occupy distinct chart positions while depending on the same component and sharing its risk.
In this example: The chart is incomplete only if the relevant dependencies are omitted; a good portfolio view keeps them attached.
The top boxes are alternative possible underlying structures, not contributions to add together.
Written comparison
Different hidden relationships
Population and voting analysis
Opposite allocations of X votes across A and B
Research and development management
Separate risks or a shared component dependency
The top boxes are alternative possible underlying structures, not contributions to add together.
The same visible representation
Population and voting analysis
Identical group and vote totals
Research and development management
A chart spread without its dependency links
The visible summary can leave the decision-relevant relationship unresolved.
What must be added
Population and voting analysis
Justified identifying structure or different data
Research and development management
Explicit dependency and correlation information
A narrower answer needs an additional source of information, not merely stronger language about the same summary.
What carries across
If a decision depends on relationships between items, retain or recover those relationships explicitly. More confidence in the totals cannot supply links the representation omitted.
Where the comparison stops
The voting illustration has exact arithmetic and logical non-identification. A portfolio chart does not inherit the same bounds, likelihood model or theorem.
- The top nodes depict alternative possibilities, not two sets to be pooled. In the toy precinct, the two scenarios are deliberately extreme to make the missing association visible.
- Additional assumptions can make ecological estimates possible, but the resulting precision is not supplied by the marginals alone. A portfolio view can avoid the depicted omission by retaining its dependency layer.
Conditions for this comparison
- Only group totals and outcome totals are available in the ecological illustration, without individual-level links.
- The project chart omits the dependency structure at issue; merely having different plotted positions is not evidence of independent risk.
Source entries
Shared pattern
Aggregation
Prime
Core Idea
Aggregation collapses many items into a unified form that retains chosen features while suppressing granular detail, formalized in classical statistics as the reduction of a sample to a summary statistic (Fisher, 1925).
Population and voting analysis
Ecological Inference Problem
Domain-specific abstraction
Core Idea
The problem is mathematically underdetermined: many different individual-level joint distributions are consistent with the same observed marginals, so the group-level data alone do not have a unique individual-level solution.
What It Is Not
Collecting more precincts of the identical aggregate form does not identify the answer — only additional structure (an assumption or a prior) or different data (the secret ballots themselves) can.
Research and development management
Portfolio View
Mechanism
Example
three "separate" projects all rely on the same unproven sensor, so their risks are correlated, not independent. The artifact flags that linkage rather than letting the portfolio look more diversified than it is.
When it helps, and when it misleads
a portfolio can look diversified while its holdings move together, so an aggregate exposure understates the real risk when a shared dependency or correlation links the items.
How it works
- Retain the dependencies. Keep item-level linkages and outliers attached to the view so the aggregate does not disguise correlated or concentrated risk.