Overview
Measure establishes a trustworthy baseline. The section moves from documenting the process (flowcharts, procedures and work instructions) through the statistics a Green Belt needs to describe it: probability rules, the central limit theorem, and the normal, binomial, Poisson, t, chi-square and F distributions. It then covers how to collect data well (data types, measurement scales, sampling and data collection plans), how to summarize it numerically and graphically, how to prove the measurement system is adequate with measurement system analysis (MSA), and how to compare process output to specifications with capability and performance indices.
Candidates must calculate as well as interpret: a z-score, a probability, a %GRR, or a Cpk from given numbers. Measure feeds Analyze directly, since hypothesis tests and regression assume the data types, distributions and measurement quality established here.
How it's tested
Measure questions are frequently calculations with a short context: compute a z-value, a probability, a capability index or a gage R&R percentage, then interpret it. Others ask which distribution, data type, sampling method or graph fits a situation, or what to do FIRST before collecting process data. Bring a clear formula sheet and practice table lookups.
-
Calculate and interpret capability
Specification limits, a mean and a standard deviation are given; you compute Cp, Cpk, Pp or Ppk and decide whether the process is capable or off-center.
-
Choose the distribution or data type
A situation such as defects per page, pass/fail out of 50, or time to complete is described and you pick the matching distribution, data type or measurement scale.
-
Judge the measurement system
Gage R&R results, bias, linearity or stability findings are presented and you decide if the gage is acceptable and whether repeatability or reproducibility dominates.
A process has USL 50, LSL 30, mean 44 and short-term sigma 2. A candidate sees Cp = 20 / 12 = 1.67 and calls it capable. But the mean is off-center, so check Cpk: (50 - 44) / 6 = 1.0 and (44 - 30) / 6 = 2.33, so Cpk = 1.0. The process has the potential but is centered poorly; the best action is to shift the mean toward 40 before reducing variation.
Topics 6
Official topics from the ASQ outline, in outline order.
3.1 Process analysis and documentationProcess maps, flowcharts, written procedures, work instructions and identifying inputs and outputs.
- What it means
- Before measuring, the team documents how the work is actually done: detailed flowcharts or swim-lane maps, spaghetti diagrams for physical movement, and written procedures and work instructions. It also identifies process inputs (Xs) and outputs (Ys) so data collection targets the variables that matter.
- What to understand
- Map the as-is process by observation, not from the manual, and mark decision points, rework loops and handoffs. Know the difference between a procedure (what is done and by whom, at a higher level) and a work instruction (step-by-step how to do one task). Classify inputs as controllable or noise.
- How it shows up
- A scenario shows a team mapping from the official procedure and finding no problems, and asks what they should do instead, which is to walk and observe the real process.
- Common confusion
- Candidates treat the documented procedure as the process; the exam assumes the real process often differs and that the gap is itself a source of variation.
3.2 Probability and statisticsBasic probability rules and the central limit theorem.
- What it means
- Probability quantifies how likely events are. Key rules are the complement (P(not A) = 1 - P(A)), addition (P(A or B) = P(A) + P(B) - P(A and B)), and multiplication for independent events (P(A and B) = P(A) x P(B)). The central limit theorem says sample means are approximately normally distributed for reasonably large samples, with standard error equal to sigma divided by the square root of n.
- What to understand
- Know when events are mutually exclusive (the joint probability is zero) versus independent (one does not change the other's probability), and recognize that the CLT is why X-bar charts and tests on means work even when individual values are not normal.
- How it shows up
- Expect short probability calculations (two independent inspections both missing a defect) and questions on how the spread of sample means changes when the sample size quadruples, which halves the standard error.
- Common confusion
- Mutually exclusive events cannot be independent unless one has zero probability; candidates use the multiplication rule for events that cannot occur together.
3.3 Statistical distributionsNormal, binomial, Poisson, chi-square, t and F distributions and when each applies.
- What it means
- Distributions model how data behave. The normal describes many continuous measures; the binomial counts defectives in a fixed number of trials with constant probability; the Poisson counts defects or events in an area of opportunity; the t, chi-square and F distributions underpin tests on means with unknown sigma, on variances and on ratios of variances.
- What to understand
- Use z = (x - mean) / sigma and a normal table to find the proportion beyond a limit. Binomial mean is np; Poisson mean and variance are both lambda. Pick binomial for pass/fail counts out of n, Poisson for counts with no fixed upper limit such as scratches per panel.
- How it shows up
- A stem describes the data and asks which distribution fits, or gives mean, sigma and a specification limit and asks what percentage falls outside.
- Common confusion
- Defectives (units that fail) follow the binomial; defects (flaws that can occur several times per unit) follow the Poisson, and candidates swap them.
3.4 Collecting and summarizing dataData types, measurement scales, sampling, data collection plans, descriptive statistics and graphical methods.
- What it means
- Data are continuous (variables) or discrete (attributes), measured on nominal, ordinal, interval or ratio scales. Sampling methods include simple random, stratified, systematic and rational subgroup sampling. Descriptive statistics summarize center (mean, median, mode) and spread (range, standard deviation, variance), and graphs such as histograms, box plots, scatter diagrams, run charts and normal probability plots show shape and patterns.
- What to understand
- Continuous data carry more information per observation, so they need smaller samples. A data collection plan states what to measure, operational definitions, how, who, when and sample size. Use the median when data are skewed or have outliers. A box plot shows median, quartiles and outliers.
- How it shows up
- Questions ask which scale a variable uses (temperature in Celsius is interval, weight is ratio, satisfaction ranking is ordinal) or which sampling method avoids bias when shifts differ.
- Common confusion
- Interval data have no true zero, so ratios are meaningless (20 degrees Celsius is not twice as hot as 10); candidates treat interval and ratio scales as the same.
3.5 Measurement system analysis (MSA)Gage repeatability and reproducibility, accuracy, bias, linearity, stability and attribute agreement.
- What it means
- MSA checks whether the measurement system is good enough before the data are trusted. Accuracy concerns bias (average offset from the true value), linearity (bias change across the range) and stability (bias change over time). Precision concerns repeatability (equipment variation, same appraiser and part) and reproducibility (appraiser variation, different appraisers on the same parts).
- What to understand
- Common AIAG guidelines: %GRR under 10% of total or tolerance variation is acceptable, 10% to 30% may be acceptable depending on the application and cost, over 30% is unacceptable; number of distinct categories should be at least 5. High repeatability points to the gage; high reproducibility points to appraiser method or training.
- How it shows up
- A gage study result is given and you decide whether to proceed with data collection or fix the measurement system first, and which fix fits the dominant error.
- Common confusion
- Calibration addresses bias (accuracy) but does nothing for repeatability or reproducibility, yet candidates choose recalibration to fix a precision problem.
3.6 Process and performance capabilitySpecification limits versus process spread, capability studies, Cp, Cpk, Pp, Ppk and short-term versus long-term performance.
- What it means
- Capability compares the voice of the process (its natural spread) with the voice of the customer (specification limits). Cp and Cpk use short-term (within-subgroup) sigma; Pp and Ppk use overall (long-term) sigma. A capability study requires a stable process, valid measurement and a check of the data's distribution.
- What to understand
- Cp = (USL - LSL) / 6 sigma-within; Cpk = minimum of (USL - mean) / 3 sigma and (mean - LSL) / 3 sigma. Pp and Ppk use the same formulas with overall sigma. Cpk equals Cp only when the process is centered. A Cpk of 1.33 is a common minimum target, and a large gap between Cpk and Ppk signals instability or shifts over time.
- How it shows up
- Calculations with off-center processes, or questions asking why capability indices are invalid when the control chart shows special causes.
- Common confusion
- Capability indices are only meaningful for a stable process; candidates compute Cpk on an out-of-control process and treat it as a prediction.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Cp (process capability)
Cp = (USL - LSL) / (6 x within-subgroup sigma), the ratio of the specification width to the natural process spread, ignoring where the mean sits.
- Remember
- Cp shows potential only; it can look excellent while the process makes defects because it is off-center.
- Tested as
- Asked alongside Cpk to diagnose whether the problem is spread or centering.
- Trap
- Concluding a process is capable from Cp alone when the stem gives a mean near one limit.
-
Cpk
Cpk = min[(USL - mean) / 3 sigma, (mean - LSL) / 3 sigma], capability measured to the nearer specification limit using within-subgroup sigma.
- Remember
- Cpk is never larger than Cp, and the two are equal only when the mean is exactly centered between the limits.
- Tested as
- The most frequently calculated Measure index; one-sided specifications use only the relevant half of the formula.
- Trap
- Taking the larger of the two sides, or using 6 sigma instead of 3 sigma in the denominator.
-
Pp and Ppk
Process performance indices use the same formulas as Cp and Cpk but with the overall (long-term) standard deviation, which includes shifts and drifts between subgroups.
- Remember
- Ppk lower than Cpk by a wide margin means between-subgroup variation, such as shift-to-shift or lot-to-lot changes, is inflating long-term spread.
- Tested as
- Interpretation questions comparing short-term and long-term capability.
- Trap
- Treating Pp/Ppk and Cp/Cpk as interchangeable names.
-
Z-score
z = (x - mean) / sigma converts a value to standard deviations from the mean so a normal table gives the area beyond it.
- Remember
- About 68.27% of a normal distribution lies within 1 sigma, 95.45% within 2 and 99.73% within 3.
- Tested as
- Used to estimate the percentage out of specification and to relate capability to defect rates.
- Trap
- Reading the table's cumulative area as the tail area, or forgetting to add both tails for two-sided specifications.
-
Central limit theorem
The distribution of sample means approaches normal as sample size grows, with mean equal to the population mean and standard error sigma / square root of n.
- Remember
- Quadrupling the sample size halves the standard error.
- Tested as
- Explains why X-bar charts and z or t tests on means work for non-normal individual data.
- Trap
- Believing the CLT makes the individual data normal; it applies to the distribution of means.
-
Binomial vs. Poisson
The binomial models the number of defectives in n trials with probability p (mean np, variance np(1 - p)); the Poisson models count of defects in a fixed area of opportunity (mean and variance both lambda).
- Remember
- Defectives with a fixed sample size point to binomial; defects with no fixed maximum point to Poisson.
- Tested as
- Distribution-choice questions and probability calculations such as P(0 defects) = e^(-lambda).
- Trap
- Using Poisson for pass/fail results because the numbers are counts.
-
Gage R&R acceptance guidelines
Under widely used AIAG guidance, %GRR below 10% is acceptable, 10% to 30% may be acceptable depending on importance and cost, and above 30% is unacceptable; number of distinct categories should be 5 or more.
- Remember
- Repeatability is equipment variation; reproducibility is appraiser variation.
- Tested as
- Gage study output must be judged and the right corrective action chosen.
- Trap
- Retraining appraisers when repeatability, which comes from the gage itself, is the dominant component.
-
Accuracy vs. precision
Accuracy is closeness of the average measurement to the true value (bias, linearity, stability); precision is closeness of repeated measurements to each other (repeatability, reproducibility).
- Remember
- A gage can be precise but biased, which calibration can fix, or accurate on average but imprecise, which calibration cannot fix.
- Tested as
- Target-diagram and gage-study questions.
- Trap
- Choosing calibration to fix scatter in repeated readings.
-
Measurement scales
Nominal (categories, no order), ordinal (ordered ranks with unequal gaps), interval (equal gaps, no true zero) and ratio (equal gaps and a true zero).
- Remember
- The scale limits the statistics you can use: a mean of nominal codes is meaningless and ratios need a true zero.
- Tested as
- Classification questions and choosing valid summary statistics.
- Trap
- Calling Celsius temperature ratio data.
-
Data types and sample size
Continuous (variables) data are measured on a scale; discrete (attribute) data are counts or categories.
- Remember
- Continuous data need much smaller samples than attribute data to detect the same change, so convert to continuous measures where practical.
- Tested as
- Data collection planning questions and later chart and test selection.
- Trap
- Recording pass/fail when the actual dimension could be measured.
-
Descriptive statistics
Mean, median and mode describe center; range, variance and standard deviation describe spread; sample standard deviation divides by n - 1.
- Remember
- The median resists outliers and skew, so it is the better center for skewed data such as cycle times.
- Tested as
- Small calculations and choices of summary for skewed data or box plots.
- Trap
- Using the population formula (dividing by n) for a sample.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Cp vs. Cpk vs. Pp vs. Ppk
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Cp | Specification width over 6 within-subgroup sigma | Short-term potential, ignores centering | Calling the process capable when it is off-center |
| Cpk | Distance from mean to nearer limit over 3 within-subgroup sigma | Short-term, accounts for centering | Taking the larger side instead of the minimum |
| Pp | Specification width over 6 overall sigma | Long-term spread including shifts between subgroups | Using within-subgroup sigma |
| Ppk | Distance from mean to nearer limit over 3 overall sigma | Long-term actual performance with centering | Expecting Ppk to exceed Cpk; overall sigma includes between-subgroup shifts, so Ppk is normally equal to or lower than Cpk |
Repeatability vs. reproducibility vs. bias
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Repeatability | Variation when one appraiser measures the same part several times with the same gage | Equipment variation; fix the gage or fixture | Retraining appraisers for a gage problem |
| Reproducibility | Variation in averages between different appraisers measuring the same parts | Appraiser variation; fix method, training or operational definition | Replacing the gage when appraisers measure differently |
| Bias | Difference between the observed average and the reference value | An accuracy issue; fix by calibration | Expecting calibration to reduce scatter |
Normal vs. binomial vs. Poisson
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Normal | Continuous, symmetric bell-shaped distribution | Used for measured variables such as length or time | Assuming every dataset is normal without checking a probability plot |
| Binomial | Number of defectives in n independent trials with constant p | Fixed number of trials, two outcomes each | Using it for defects that can occur many times per unit |
| Poisson | Number of events or defects in a fixed area of opportunity | No fixed upper limit; mean equals variance | Using it for pass/fail results out of a known sample |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Cp says capable
- What candidates pick
- The process is capable because Cp is 1.67.
- Why it looks right
- Cp is the first index taught and 1.67 sounds strong.
- Why it's wrong
- Cp ignores where the mean sits; if the mean is near one specification limit, Cpk can be below 1 and defects are being produced.
Clue to watch for The mean given is not halfway between the limits.
-
Capability before stability
- What candidates pick
- Calculate Cpk immediately from the data collected.
- Why it looks right
- The stem provides everything needed for the formula.
- Why it's wrong
- Capability predicts future performance only for a stable process; if a control chart shows special causes, they must be removed first.
Clue to watch for A mention of points beyond control limits, trends or shifts in the baseline data.
-
Skipping MSA
- What candidates pick
- Start collecting baseline data right away to save time.
- Why it looks right
- The team is eager and the sponsor wants numbers.
- Why it's wrong
- If the measurement system contributes large variation, every baseline, capability and later hypothesis test is unreliable; MSA comes first.
Clue to watch for The question asks what to do FIRST in Measure, or mentions inspectors who disagree.
-
Calibrate to fix precision
- What candidates pick
- Recalibrate the gage when repeated readings scatter widely.
- Why it looks right
- Calibration is the familiar fix for measurement problems.
- Why it's wrong
- Calibration corrects bias (accuracy); scatter is a repeatability or reproducibility problem needing gage maintenance, fixturing, a better gage or appraiser training.
Clue to watch for The same part measured repeatedly gives widely different values.
-
Poisson for defectives
- What candidates pick
- The Poisson distribution for the number of failed units in samples of 50.
- Why it looks right
- Both are counts of bad outcomes.
- Why it's wrong
- Units are either good or bad in a fixed number of trials, which is the binomial; Poisson is for defects with no fixed maximum.
Clue to watch for A fixed sample size and a pass/fail outcome per unit.
-
Averaging when data are skewed
- What candidates pick
- Report the mean cycle time as the typical value.
- Why it looks right
- The mean is the default statistic.
- Why it's wrong
- A few long delays pull the mean upward; the median better represents the typical case for skewed data.
Clue to watch for A box plot or histogram with a long right tail, or a few extreme values in the dataset.
Answer framework
For Measure phase questions, move from trustworthy data to the right statistic with these steps.
- Classify the data first: continuous or discrete, and for counts decide whether they are defects or defectives.
- Check the measurement system before trusting any number; if a gage study is given, judge %GRR and which component dominates.
- Choose the matching distribution or graph, such as normal for measures, binomial for defectives, Poisson for defects, box plot for skew.
- For capability, confirm the process is stable, then compute both Cp and Cpk (or Pp and Ppk) with the correct sigma.
- Interpret the result against the customer requirement, deciding whether the gap is spread, centering or measurement, and pick the matching action.
Practice question
Want more like this? The CSSGB practice test has questions for every domain.
Checklist
0 of 13
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Study Measure in the order the work happens. Start with process documentation and data types, because classifying data correctly drives every later choice of distribution, chart and test. Then cover probability rules and the central limit theorem, followed by distributions, practicing z-score lookups until reading the normal table is routine.
Spend the most time on measurement system analysis and capability. These are the highest-yield topics: be able to interpret a gage R&R table, decide whether to fix the gage or the appraisers, and calculate Cp, Cpk, Pp and Ppk from raw numbers, including one-sided specifications. Build a formula sheet with each formula, when it applies and one worked example, because the exam is open book but time is limited.
Commonly underestimated areas are measurement scales, sampling methods and the difference between defects and defectives, which look simple but decide later answers. After this section, take a timed set of mixed Measure calculations, then move to Analyze, where the distributions and data types you have learned determine which hypothesis test is valid.
FAQ
How many Measure phase questions are on the CSSGB?
ASQ lists 23 of the 100 scored questions for this section, tying it with Define as the largest section.
What is the difference between Cpk and Ppk?
Both measure capability to the nearer specification limit, but Cpk uses within-subgroup (short-term) standard deviation and Ppk uses overall (long-term) standard deviation, so Ppk includes shifts and drifts between subgroups.
Do I need to memorize statistical tables?
No. The CSSGB is open book, so bring tables and a formula sheet, but practice lookups in advance; the time cost of searching for a formula adds up across a 110-question exam.
What gage R&R result is acceptable?
Under common AIAG guidance, %GRR below 10% is acceptable, 10% to 30% may be acceptable depending on the application and cost, and above 30% is unacceptable; also look for at least 5 distinct categories.
Why must the process be stable before calculating capability?
Capability indices predict future output, and that prediction only holds if the process is in statistical control. With special causes present, the estimate of sigma and mean is not reliable.
How long should I study Measure?
Plan roughly 25–30 hours within a 120-hour plan, with most of it spent on calculations: z-scores, gage studies and capability indices.
How does Measure connect to the rest of the exam?
The data types and distributions chosen here decide which hypothesis tests apply in Analyze, and the capability baseline set here is what Improve must beat and what Control must sustain with SPC.
Other Quality Management certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 3 knowledge
You have worked through distributions, gage R&R and the Cp, Cpk, Pp and Ppk formulas. Test whether you can calculate and interpret them under time pressure.