Preserve Board Pairing in a Before-and-After Process Comparison

When the same boards are measured before and after a process, their identities are part of the experiment. Comparing two unrelated column averages discards useful information and can conceal a matching error. Preserve the pair for each board and examine its change directly.

The NIST paired-observation guidance defines pairing through a natural relationship between observations and analyzes differences within those pairs. Use this structure only when the relationship is real; matching unrelated boards by spreadsheet row does not create a paired experiment.

Define the pairing key

Use a stable board identifier, measurement location, and operating condition. If several nodes are measured on one board, the board serial alone is insufficient. Include the node and condition so that each after value is matched to the intended before value.

Check for duplicate keys and missing observations before calculating differences. Preserve incomplete pairs in a separate accounting record. Removing them silently can hide a process that prevented some boards from completing the second measurement.

Calculate and inspect changes

Choose one subtraction direction and label it consistently in tables, figures, and conclusions. Plot individual differences against the starting value and relevant board characteristics. A small average can hide large changes in opposite directions or a response limited to one subgroup.

Separate treatment from elapsed time

1. Keep the measurement method consistent.
2. Record the exact treatment applied to each board.
3. Include an appropriate untreated comparison where feasible.
4. Review instrument and environmental changes between measurements.
5. Interpret the observed differences within those limitations.

A before-and-after change is not automatically caused by the intended treatment. Storage, temperature, repeated handling, or a changed test setup can also contribute. Design the comparison to distinguish these possibilities rather than relying on the paired arithmetic alone.

For formal inference, choose an analysis appropriate to the distribution and independence of the board-level differences. Repeated measurements on one board do not automatically supply additional independent treated boards. Keep the number of boards and the number of readings visible separately.

Report practical magnitude as well as statistical evidence. State whether the change matters relative to the product requirement, and include the incomplete-pair count. The useful result explains how identified boards changed under a defined process, rather than presenting two averages whose membership cannot be reconstructed.