CQPA common mistakes
CQPA · Understand the traps and how to avoid them.
Treating the seven basic quality tools as interchangeable with the seven management and planning tools, and picking a tool by name recognition rather than by data type (attribute vs. variable, discrete count vs. continuous measurement).
Both tool sets are taught together in the Body of Knowledge, and candidates memorize lists without attaching each tool to the question it answers (cause sorting, prioritization, trend, distribution shape).
What to do instead
Build a one-page decision table: for each tool write the data type it accepts, the question it answers, and one cue phrase from item stems (e.g., 'narrow the many causes to the vital few' points to Pareto; 'process behavior over time with limits' points to control chart; 'relationship between two variables' points to scatter diagram). Drill by classifying stems before choosing an answer.
Computing Cp/Cpk and Pp/Ppk with the wrong sigma, or using Cpk when the question asks about process performance over the long term, and then selecting an answer that is numerically plausible.
Candidates conflate within-subgroup variation (short-term capability) with overall variation (long-term performance) and forget that Cpk uses the smaller of the two one-sided distances to the specification limits.
What to do instead
Before calculating, write down which sigma the stem implies: subgroup-based (R-bar/d2 or s-bar/c4) for Cp/Cpk, total sample standard deviation for Pp/Ppk. Always compute both one-sided indices and take the minimum. Sanity-check with the centering rule: Cpk = Cp only when the process mean sits exactly at the specification midpoint.
Reading a control chart as a capability chart: reacting to a single point near a limit while ignoring run rules, or declaring the process out of control because a point exceeds a specification limit that is not a control limit.
Specification limits and control limits look alike on a chart, and candidates under-practice the Western Electric/Nelson pattern rules (runs, trends, stratification, alternating patterns) that define special-cause signals.
What to do instead
State the rule being tested before evaluating the chart. Memorize the standard signal set (one point beyond 3 sigma; 2 of 3 beyond 2 sigma on one side; 4 of 5 beyond 1 sigma; 8 in a row on one side; 6-point trend; 14 alternating points). Practice distinguishing voice of the process (control limits) from voice of the customer (specification limits) with short chart-reading items.
Applying the wrong phase of DMAIC or skipping the gate: proposing a solution in the Measure phase, or moving to Improve without a validated root cause and verified measurement system.
Exam items describe a situation and ask what the analyst should do next; candidates jump to the action that sounds most valuable rather than the one the current phase permits.
What to do instead
Map each phase to its deliverable and gate question: Define (charter, VOC, CTQ), Measure (baseline, MSA, capability), Analyze (verified root cause), Improve (piloted, validated countermeasure), Control (documented, monitored, response plan). For 'next step' items, identify the phase from the artifact mentioned in the stem, then choose the deliverable that closes that phase.
Selecting a sampling plan or sample size by convenience, or interpreting an AQL as the acceptable defect rate the process is expected to produce rather than as a quality level tied to the acceptance risk.
Acceptance sampling terminology (AQL, LTPD/RQL, producer's and consumer's risk, OC curve) is dense, and candidates memorize definitions without linking them to the probability of acceptance.
What to do instead
Learn acceptance sampling through the OC curve: for a given plan, plot probability of acceptance against incoming quality; mark AQL at the producer's-risk point and LTPD at the consumer's-risk point. Practice reading how tightening the plan (larger n, higher acceptance number) shifts the curve. Treat AQL as an indexing quality level, not a target defect rate.
Conducting a Gage R&R or attribute agreement study with the wrong design: too few appraisers or trials, parts that do not span the process range, or no randomized order, then accepting the result at face value.
Candidates learn the arithmetic of percent GRR but not the design requirements, so they cannot spot an invalid study described in an item stem.
What to do instead
Memorize the design baseline: typically 10 parts spanning the process spread, 2-3 appraisers, 2-3 trials, randomized and blind. Know the interpretation thresholds (under 10 percent acceptable, 10-30 conditional on cost/risk, over 30 unacceptable) and the distinction between repeatability (equipment) and reproducibility (appraiser). For attribute studies, use kappa and within-appraiser agreement.
Confusing precision with accuracy, or confusing a measurement system's resolution with its bias, when diagnosing a measurement problem described in a scenario.
The terms are used loosely in practice, and items deliberately describe a system that is consistent but offset, or inconsistent but centered, to test the distinction.
What to do instead
Anchor each term to its evidence: accuracy/bias is a shift from the reference (linearity is bias across the range); precision is spread (repeatability and reproducibility); resolution is the smallest readable increment. When reading a scenario, ask whether repeated readings cluster (precision) and whether the cluster is on target (accuracy) before choosing a remedy such as calibration versus equipment upgrade.
Treating a statistically significant result as a practically important one, or ignoring sample size, power, and the difference between statistical and practical significance when interpreting a hypothesis test.
Candidates focus on the p-value threshold alone and do not connect effect size, confidence intervals, and power to the business decision in the stem.
What to do instead
For every inference item, identify the null and alternative, the decision rule, and the effect size. Remember that large samples make trivial differences significant and small samples make real differences invisible (low power). Prefer answers that cite confidence intervals and practical impact over answers that rely only on 'p is less than 0.05'.
Mixing up the cost-of-quality categories, classifying prevention and appraisal as failure costs, or assigning an internal failure to the external failure category.
The four categories are simple in isolation but scenario items place costs in ambiguous settings (rework before shipment, warranty after shipment, training, audit, inspection).
What to do instead
Use the control point test: money spent before defects occur is prevention (training, design review, supplier development) or appraisal (inspection, test, audit); money spent after a defect exists is internal failure (scrap, rework, downtime before shipment) or external failure (warranty, returns, recalls, liability). Classify by when the defect is detected relative to the customer receiving the product.
Answering audit and corrective-action items with the corrective action when the item asks for containment, or closing a corrective action after the fix without verifying effectiveness and updating documentation.
Containment, correction, corrective action, and preventive action are closely related and candidates treat them as synonyms; the verification step is easy to skip under time pressure.
What to do instead
Memorize the sequence: contain the effect (protect the customer), correct the specific instance, find the root cause, implement corrective action, verify effectiveness with data, then update control plans, FMEA, and standard work. For audit items, distinguish a finding from an observation and know that a nonconformity requires a response with root cause and corrective action, not just a promise to fix.