Tensions in Practice: Cheaper repeated readings in tension with independent error¶
Two stipulated measurement designs for one fixed quantity
Both designs produce illustrative readings 0 and 6, so their equally weighted mean is 3. Each reading has error variance 1, a measure of its noise spread. Repeated readings from one instrument share some error: covariance 0.5. A separately acquired second source has zero covariance with the first under the toy assumptions. The mean’s variance is then 0.75 versus 0.5. The extra source buys more independent information, not a different averaging formula.
Reuse the existing instrument
Take two readings without obtaining a separate measurement source.
Reduce shared error
Acquire a second source with independently validated errors.
Why these aims pull against each other
A second reading can be cheap while retaining common error; reducing that coupling can require another instrument and fresh calibration.
Choose an arrangement to see what changes and what remains difficult.
Diagonal cells are individual error variances; off-diagonal cells are covariance shared by the two errors. Observed readings and diagonal values stay fixed; only the shared part changes.
What this choice protects
What it costs
When it fits
Compare the arrangements
Repeat on one instrument
Take two readings whose equal variances are 1 and covariance is 0.5.
| R1 error | R2 error | |
|---|---|---|
| R1: 0 | 1 | 0.5Shared part |
| R2: 6 | 0.5Shared part | 1 |
- What it protects
- The existing instrument can provide both observations.
- What it costs
- Their shared error limits the variance reduction from averaging.
- When it fits
- Fits when acquisition cost matters and a mean variance of 0.75 is acceptable under this validated model.
Illustration note: Var((R1+R2)/2) = (1 + 1 + 2×0.5)/4 = 0.75. The shared part is counted, not ignored.
Acquire an independent source
Use two equal-variance readings with covariance zero.
| R1 error | R2 error | |
|---|---|---|
| R1: 0 | 1 | 0Shared part |
| R2: 6 | 0Shared part | 1 |
- What it protects
- The mean variance falls to (1 + 1)/4 = 0.5 under the declared error model.
- What it costs
- A second instrument or channel must be acquired and its independence and calibration checked.
- When it fits
- Fits when reduced measurement uncertainty justifies that cost and independence is supported.
Illustration note: The same illustrative mean 3 does not reveal the difference; the error relationship does.
What this illustration does—and does not—establish
The source supplies the tension. The invented setting, alternatives and any numbers illustrate a limited comparison; each arrangement retains its stated costs and conditions.
- Readings, variances and covariance are invented; no actual instrument performance is claimed.
- Errors are stipulated unbiased with the same fixed target. Bias, drift or shared environment can invalidate the model.
- Zero covariance suffices for this variance calculation; full independence is an additional declared condition, not inferred from two numbers.
- Neither mean is guaranteed close to truth on an individual run; uncertainty concerns the stated repeated-measurement law.
Source entries
Precision Weighting
This source passage supplies the contextual tension. The concrete arrangements and schematic examples are editorial illustrations, not measured findings.
Optimal Weighting versus Correlated Signals (coupling)
When sources are correlated (witnesses who conferred, models trained on shared data, sensors with a common disturbance), inverse-variance weighting double-counts the shared component and overstates the combined precision.
The source operation
Precision weighting is the structural pattern by which evidence about the same latent quantity receives influence in proportion to estimated reliability — specifically precision, the inverse of variance — so lower-noise evidence contributes more to the resulting estimate or update.