The Art of Statistics

Study Design Qualitydesign lever

The rigour with which an investigation is planned and executed, including use of randomization, control groups, blinding, representative sampling, adequate sample size and pre-specified protocols to support valid inference.

Data Qualitycontextual condition

The reliability and validity of collected data, encompassing accurate measurement, low systematic bias, completeness of follow-up, and the absence of errors or excessive missingness that distort what the data represent about reality.

Communication and Framing Choicesdesign lever

The deliberate choices in presenting statistical results, including positive versus negative framing, use of absolute versus relative risks, expected frequencies, graphical design and transparency, which shape how audiences perceive and understand findings.

Model and Algorithm Complexitydesign lever

The degree of flexibility built into a predictive or explanatory model or algorithm, ranging from simple rules to highly parameterized deep models, which trades off bias against variance and affects generalization.

Uncertainty Quantificationpsychological state

The extent to which variability and epistemic uncertainty in estimates are properly captured and expressed through confidence or uncertainty intervals, margins of error, P-values and acknowledgement of systematic error.

Systematic Biasbehavioral pattern

Non-random, directional error introduced through unrepresentative sampling, confounding, measurement bias, reverse causation, or selective processes that cause estimates to systematically depart from the true target quantity.

Over-fittingbehavioral pattern

The condition in which a model adapts too closely to idiosyncrasies of training data, fitting noise rather than signal, so that predictive performance declines on new independent data.

Questionable Research Practicesbehavioral pattern

Conscious or unconscious analytic flexibilities such as multiple testing, selective reporting, HARKing, optional stopping and post hoc subgroup analysis that inflate apparent statistical significance and produce exaggerated or false findings.

Audience Comprehensionpsychological state

The degree to which the intended audience accurately understands the meaning, magnitude and uncertainty of statistical findings, free from misinterpretation, exaggeration or emotional distortion.

Reliability and Trustworthiness of Conclusionsoutcome metric

The overall validity, accuracy and reproducibility of the conclusions drawn from an investigation, reflecting whether claims hold up to replication and accurately represent the target reality with appropriate humility about limitations.