Compare observational groups without inventing causation

A filter changes the evidence you inspect

Select all species on Biscoe, then all species on Dream. A difference between the displayed means is an association with the recorded island label. The filters do not move birds between islands or control other conditions.

Inspect species composition before proposing an island mechanism. This file records Adelie on all three islands, Chinstrap on Dream and Gentoo on Biscoe. Thus the all-species island comparison changes both location and the mixture of species. It cannot isolate an island effect.

Make a narrower comparison

Keep the island at Biscoe and select Adelie. Capture the result. Change only island to Dream and capture again. This holds the recorded species label constant, which addresses one source of compositional difference. It does not hold sex, year, food supply, age or measurement circumstances constant, and it is not randomized assignment.

Write two sentences with different scopes: “Among the selected recorded Adelie birds, the mean…” and “This comparison cannot establish…”. A careful answer can describe the observed difference without pretending that a filter made the study experimental.

Investigate an empty combination

Predict what should happen for Chinstrap on Torgersen, then inspect the result. The file has no matching records. The correct output is an unavailable mean and an empty plot, not a mean of zero or proof that no Chinstrap penguin could occur there.

Compare three situations: no selected records, selected records with missing measurements, and selected records with complete measurements. Each implies a different evidence limitation. A program that converts all three to zero hides those distinctions.

Design a follow-up

Choose a question that could be addressed with observational evidence, such as whether a mass difference persists after comparing the same species, sex and season across locations. Specify which variables would be needed, how sites and individuals would be selected, and what residual confounding would remain. Do not propose moving or experimentally handling wild animals as an independent learner activity.

For an unfamiliar case, imagine a survey finds larger fish downstream than upstream but samples different species at each site. Explain how the penguin comparison helps diagnose the inference problem, then identify what additional observations would help. Similar reasoning applies across organisms, but an analogy does not prove that the two ecological systems have the same mechanisms.