Check continuous measurement quality
Establish what the measurement represents, preserve its units and scoring rules, and investigate agreement before combining devices or methods. Use synthetic examples for calculations and retain the evidence behind acceptable differences.
Worked decision
Hypothetical decision: A laboratory replaces a device midway through collection. On ten synthetic paired samples, the new device reads two units higher on every pair. Correlation alone would not reveal that constant offset as a lack of association. The team compares the paired differences with a prespecified acceptable disagreement, checks the range and calibration evidence, and investigates whether any conversion is justified before pooling. Ten artificial constant differences do not validate a correction for real specimens.
Record your decision
Are measurements comparable enough for the intended use? Confirm units and scoring first, then inspect paired differences across the relevant range. Correlation does not establish agreement, and a change of device may need bridging evidence. Record acceptable disagreement before inspecting results and explain remaining uncertainty instead of choosing a convenient cutoff.
Work through the evidence
- Translate the instrument manual into scoring rules (TXT)
- Check that measurement units represent the same quantity
- Verify a linear conversion of means, SDs and interval limits
- Retain the source and arithmetic for unit conversions (TXT)
- Distinguish measurement agreement from correlation
- Explore descriptive limits of agreement from paired differences
- Assess a device or assay change before pooling measurements (TXT)
- Review NIST guidance on graphical and quantitative diagnostics