A result can be correct and still be interpreted too strongly.
A pattern can be real and still have more than one explanation.
This is why science benefits from separating three ideas:
observation, interpretation and evidence.
Observation: what was detected
An observation is tied to a method.
A sequencing assay detects DNA sequences. A microscope records spatial structure. A metabolomics experiment measures chemical features. A culture experiment shows growth under defined conditions.
Each observation answers a particular question.
It also has limits.
Detecting DNA does not automatically prove that a cell is alive or active. Finding two organisms together does not prove they interact directly. Measuring a metabolite does not automatically identify which organism produced it.
The observation should be stated at the level the method supports.
Interpretation: what the observation might mean
Interpretation connects results to explanations.
Researchers use prior knowledge, statistics, controls and biological reasoning to decide what an observation may imply.
This is necessary.
Data do not explain themselves.
But interpretation introduces choices: which comparison matters, which mechanism is plausible, which alternative explanations remain and how much uncertainty is acceptable.
Two research teams can sometimes implement an apparently similar observational question differently and obtain different results because design and analytic decisions differ.
Evidence: how much support has accumulated
Evidence is broader than one result.
It includes the quality of measurement, study design, sample size, controls, reproducibility, consistency with other work and whether alternative explanations have been tested.
Different kinds of evidence can contribute different strengths.
A controlled experiment can support a causal mechanism under defined conditions. An observational cohort can reveal real-world associations at scale. A longitudinal study can show temporal order. Replication can test whether a result survives another attempt.
No single design answers every question.
Association is not automatically causation
This principle is famous because it is repeatedly important.
If two variables change together, one may cause the other.
But the association may also arise because a third factor influences both, because the direction of causation is reversed, because of selection bias or because of chance.
Bradford Hill's classic discussion of association and causation emphasized reasoning across multiple considerations rather than treating one statistical result as proof by itself.
Uncertainty is part of evidence
Scientific uncertainty does not mean that nothing is known.
It describes the boundary between what is supported and what remains unresolved.
A strong article should therefore be able to say both:
What we know.
and
What remains uncertain.
What we know
Research methodology supports that:
- measurements depend on the methods used to obtain them;
- observation and interpretation are distinct stages of reasoning;
- association does not by itself establish causation;
- transparent methods improve reproducibility;
- conclusions become stronger when multiple forms of evidence converge.
What remains uncertain
There is no universal scoring system that converts every study into one simple evidence number.
Different questions require different methods. Replication can disagree for methodological as well as biological reasons. New evidence can revise earlier interpretations.
Science is strongest when those limitations are visible rather than hidden.
MICROBA Perspective
MICROBA begins by protecting the observation.
What did we actually see?
What did we measure?
What are we inferring?
What would we need to test next?
This separation allows curiosity without overclaiming.
Observation opens the question. Interpretation gives direction. Evidence determines how far the conclusion can responsibly go.
References
- Hill AB. The Environment and Disease: Association or Causation? Proceedings of the Royal Society of Medicine. 1965. https://journals.sagepub.com/doi/pdf/10.1177/003591576505800503
- Hripcsak G, et al. Reproducible variability: assessing investigator discordance across 9 research teams attempting to reproduce the same observational study. Journal of the American Medical Informatics Association. 2023. https://pmc.ncbi.nlm.nih.gov/articles/PMC10114120/
- Human Microbiome Project Consortium. Structure, function and diversity of the healthy human microbiome. Nature. 2012. https://www.nature.com/articles/nature11234