Accuracy concerns closeness to the true value; reliability concerns consistency; validity concerns whether the evidence answers the intended question.
A thermometer that always reads two degrees too high may give consistent results but inaccurate temperatures. Compare instruments with a trusted reference before fieldwork. Record the resolution: showing many decimal places does not remove systematic error.
Reliability improves when repeated observations under comparable conditions agree. Train observers, pilot the questionnaire, repeat measurements and use common definitions. Real change between noon and evening is not necessarily unreliable measurement; compare like conditions before judging consistency.
Validity asks whether the method measures the concept you intend and supports the conclusion. Counting benches may accurately describe furniture but poorly measure accessibility if the benches are unreachable or unsafe. Combine indicators and ask users about their experience.
Tie every improvement to the problem. Calibration targets a biased instrument. Repeated readings help identify random variation. Sampling missing groups improves the coverage of a social survey. No single improvement guarantees all three forms of quality.
Step by step
Identify the problem
Name the instrument, sampling or interpretation error.
Explain its effect
Show how it could change the result or conclusion.
Target the fix
Choose a practical change that addresses that mechanism.
Worked example: A stronger evaluation sentence
"Observers used different definitions of shade, so scores may reflect the observer rather than site conditions. Agree a shading rule, practise together and compare ratings before collecting the main dataset." This connects problem, consequence and improvement.
Watch out for this
Reliable data must produce a valid conclusion.
A method can consistently measure the wrong thing. Reliability helps, but validity also depends on relevance and interpretation.
Check your understanding
Every observer counts the same number of ramps, but steep gradients are ignored. What is the main issue when judging wheelchair access?
- The count must be unreliable.
- The indicator has limited validity for access.
- The sample is necessarily too small.