Measure phase: data collection, measurement basics and process mapping
Measure establishes a factual baseline: what you will measure, how it is defined, whether the measurement system can be trusted, and how the process actually flows. Exam questions here are usually "what should the team do next" scenarios.
Types of data
- Continuous (variable) data: time, length, cost, temperature. More information per observation, so smaller samples suffice.
- Discrete (attribute) data: pass/fail, defect counts, categories. Needs much larger samples and is best expressed as counts per defined opportunity.
Operational definitions and data collection plans
An operational definition makes a characteristic measurable by anyone: "callback time = minutes from the customer's voicemail timestamp to the agent's logged outgoing call." The data collection plan records what, why, how, who, when, where, and the stratification factors (shift, site, product, operator) needed later.
Measurement system analysis
- Accuracy/bias: closeness of the average to the true value.
- Linearity: consistent accuracy across the range.
- Stability: consistency over time.
- Repeatability: same appraiser, same part, same instrument - equipment variation.
- Reproducibility: different appraisers, same part - appraiser variation. Gage R&R rule of thumb: under 10% acceptable, 10–30% conditional on cost and risk, above 30% unacceptable - fix the system before collecting the baseline.
Sampling
Random sampling supports inference; stratified sampling guarantees subgroup coverage; systematic sampling takes every k-th unit (watch for cycles); convenience sampling is fast but biased and rarely defensible as a baseline.
Process mapping
- Flowchart: detailed decision logic.
- Swimlane: handoffs and ownership across departments.
- Value stream map: material and information flow with cycle and lead times, exposing delay.
- Spaghetti diagram: physical travel of people or parts.
Worked example
Two operators each measure the same ten parts twice with one caliper. Within-operator spread is small, but their averages differ by 0.4 mm against a 1.0 mm tolerance. Diagnosis: repeatability is acceptable, reproducibility is poor - appraiser technique differs. Retrain and re-run the study before trusting any baseline.
Traps that cost marks
- Collecting data before agreeing operational definitions.
- Judging a process on attribute data from a handful of units.
- Reporting a baseline from a convenience sample.
- Choosing a value stream map when the question is about departmental handoffs (use a swimlane).