Case 4 — When evidence makes the preferred method claim smaller

What a reader needs to know

Principal component analysis (PCA) looks for a compact summary of how observations differ. Ball Mapper groups observations into nearby, sometimes overlapping neighbourhoods so researchers can inspect local similarities. The social-science decision here is whether the more elaborate representation adds enough insight once a simpler method describes most of the overall variation.

Research setting

The project had a substantive methodological interest in Ball Mapper.

A comparator analysis produced an inconvenient result: the first principal component explained approximately 97.31% of the variation in the three published PISA domain means.

Consequential challenge

If the global structure is overwhelmingly one-dimensional, why is a topological representation needed rather than a simpler scalar summary?

The project did not suppress that challenge.

Further comparator work showed that conventional summaries captured the dominant global achievement axis very well. Ball Mapper retained some local overlapping co-membership and profile information, but not a hidden alternative global taxonomy.

Outcome

The manuscript became more restrained:

• PCA remained the preferred compact global summary;
• hard clustering largely reproduced the dominant achievement axis;
• Ball Mapper was retained only as a complement when the substantive question concerned overlapping local neighbourhoods and relative profiles.

Reason-provenance lesson

Robust criticism should be allowed to reduce the claimed contribution of the method the project initially wanted to showcase.

Methodological commitment is not strengthened by ignoring an inconvenient comparator.

Evidence status

The 97.31% result was locally verified in the empirical workflow. The public case uses the frozen project result rather than an AI-generated claim.

Evidence available to public readers

The approximately 97.31% figure is reported from frozen, locally verified project outputs; the underlying computational inputs and outputs are not provided here for independent re-estimation. Readers can inspect the stated inferential boundary, not reproduce that number from this page.

Boundary

The comparator evidence does not establish that Ball Mapper is generally necessary, superior or unnecessary. It constrained the claim in this application.