YEAR 03 • CLINICAL TRANSITION

Research & Evidence-Based Medicine

20 learning units • 100 chapter sections • 200 explained questions. Build a question, judge the evidence and connect the answer with the person receiving care.

Original RAHBAR study material for the Year 3 learning pathway. Educational examples support supervised learning; institutional research and clinical requirements still apply. References & further reading ↓

MODULE 05 / 20

Variables, Measurement & Data Quality

Define what is measured, how it is measured and which errors could change the answer.

Quick revision ↓ Medical Dictionary

01Types of variable

Nominal variables have categories without a natural ranking, such as blood group. Ordinal variables have ordered categories, such as mild, moderate and severe symptoms, but the gaps need not be equal. Counts are discrete numerical variables; height is usually treated as continuous. Coding categories with numbers does not convert them into quantities for meaningful arithmetic. Variable type influences summaries, graphs and statistical models.

02Operational definitions

An operational definition states exactly how a concept becomes a measurement. For adherence, specify whether the measure is self-report, dispensing records or observed doses, and define the period and threshold. Predefine the outcome assessment time and procedure. Different definitions can identify different groups, so a paper's label is insufficient. Clear definitions support reproducibility and reduce opportunities to choose a favourable outcome after seeing results.

03Reliability and validity

Reliability concerns consistency when measurement is repeated under comparable conditions. Validity concerns whether the measurement reflects the intended construct or reference. A scale that always adds five kilograms can be highly repeatable but inaccurate. Inter-rater agreement addresses consistency between observers; correlation alone is not proof of agreement because two raters can rank patients similarly while systematically assigning different values.

04Sources of measurement error

Calibration, standardised procedures, training and blinded outcome assessment can improve measurement quality. Differential error varies between comparison groups and can distort associations in either direction. Non-differential misclassification is not universally guaranteed to bias every estimate toward the null; that familiar simplification depends on the setting and assumptions. Ask how the error arose and which variables it affects before predicting its consequence.

05Build a defensible dataset

Create a data dictionary with variable definitions, permitted values, units and missing-value codes. Preserve raw data, document cleaning decisions and distinguish a true zero from missing information. Check impossible dates, duplicate records and values outside plausible ranges against the source. Never silently delete inconvenient observations. Corrections need an audit trail so that an independent reviewer can reconstruct the analysis.

High-Yield Revision

Numerical codes do not make categories continuous.
Repeatability and accuracy are different properties.
Keep raw data and document every cleaning decision.

Common Pitfalls

Correlation is not sufficient evidence of agreement.
Do not code an unknown value as zero unless zero truly represents the observation.

References & Further Reading

Use the full methods and relevant chapters when appraising a paper. These sources support further study; the examples and questions above are original teaching exercises.

Reference review: 25 September 2026. Consult current institutional guidance for project approval and clinical decisions.