Statistical note

Selection bias and representativeness

Recognise when inclusion, exclusion, survival, or visibility mechanisms distort the analysed sample.

11 August 20266 min read

Does the design create independent, comparable observations that support the intended scientific claim? This note turns that question into a practical analysis workflow.

Conceptual guide to Selection bias and representativeness

Core idea

Recognise when inclusion, exclusion, survival, or visibility mechanisms distort the analysed sample. The method is useful only when its target, data-generating assumptions, and unit of analysis match the scientific question.

$P(S=1\mid Y,X)\neq P(S=1)$

The equation is a compact statement of the target; it is not a substitute for checking design and data quality. State what each observation represents, how it entered the sample, and which sources of dependence remain.

Practical workflow

Write the scientific question and estimand in one sentence.
Identify biological units, technical replicates, nesting, repeated measurements, and exclusions.
Visualise raw observations and group structure before fitting the method.
Check assumptions using design knowledge, diagnostic plots, and sensitivity analyses.
Report the estimate, uncertainty, effect magnitude, sample sizes at every level, and limitations.

import numpy as np
from scipy import stats

# Summarise at the independent experimental-unit level.
cell_values = np.array([[2.1, 2.3, 2.2], [2.7, 2.8, 2.6]])
replicate_means = cell_values.mean(axis=1)
print(stats.bootstrap((replicate_means,), np.mean).confidence_interval)

Interpretation and failure modes

Do not read a software output as an automatic scientific conclusion. Ask whether dependence was modeled, whether preprocessing used information from validation data, whether missingness or selection is informative, and whether the reported uncertainty covers every level of sampling. Prefer estimates and intervals to threshold-only language.

Microscopy case study

Excluding dim cells after segmentation can exaggerate mean marker intensity because visibility depends on the outcome.

Connections

Use the tags above to continue to notes on design, assumptions, uncertainty, diagnostics, and complementary methods. A robust analysis normally combines several concepts rather than selecting one test in isolation.