Diagnostics › Method comparison

Precision evaluation

Estimates how repeatable a measurement procedure is by splitting its scatter into named variance components (repeatability, between-run, between-day, within-laboratory).

What is Precision evaluation?

"Precision" is the closeness of repeated measurements of the same sample — the opposite of scatter. But not all scatter is the same: two measurements minutes apart in one run agree more tightly than two measurements on different days. A precision study deliberately spreads replicates across runs and days so it can attribute the total variability to each source.

The engine fits a random-effects ANOVA. In the simplest form (one grouping factor) it reports repeatability (within-group) and a between-group component. Add the nested run-within-day factor and it performs the full nested decomposition: repeatability (within-run), between-run, between-day, and their pooled totals — within-day (repeatability + between-run) and within-laboratory (the grand total). This mirrors the standard CLSI EP05 precision experiment.

Each component is reported as a variance, an SD, a CV%, its share of the total variance, and a confidence interval — exact chi-square for repeatability, Satterthwaite (effective degrees of freedom) for the pooled terms. A measurements-by-day plot shows the day-to-day and within-day scatter at a glance.

When should I use Precision evaluation?

  • Validating or verifying the imprecision of an assay or measurement procedure.
  • Designing to a nested replicate × run × day layout (the classic EP05 design).
  • Reporting within-laboratory (total) CV against an analytical performance goal.

What data does it need?

A measurement column + a day/outer grouping column. Optionally add a run-within-day column for the full nested Day/Run/Error decomposition.

What does it report?

A component table (repeatability, between-run, within-day, between-day, within-laboratory) with SD, CV%, % of total variance and 95% CIs, the nested-ANOVA table (SS/DF/MS/expected-MS), and a measurements-by-day plot.

What does it assume?

  • Balanced design (equal replicates per run and runs per day) for the expected-mean-square coefficients.
  • Random, independent effects at each level; approximately normal errors.
  • The process is stable over the study window (no drift).

Formula

σ²_repeatability = MS_error
σ²_between-run = (MS_run − MS_error) / (reps per run)
σ²_within-lab = σ²_repeatability + σ²_between-run + σ²_between-day
CV% = SD / mean × 100

How do I interpret the result?

Read the within-laboratory CV against your analytical goal — it is the total imprecision a user of the assay experiences. If between-day or between-run dominates, the process is drifting and calibration/QC frequency may need attention; if repeatability dominates, the measurement itself is noisy.

The % of total column shows where to focus improvement: shrinking the largest component gives the biggest reduction in overall imprecision.

See also

References

  • CLSI EP05-A3. Evaluation of Precision of Quantitative Measurement Procedures.
  • Searle, Casella & McCulloch (1992). Variance Components.