CQE common mistakes
CQE · Understand the traps and how to avoid them.
Computes Cpk by dividing by the overall sample standard deviation s (or the population sigma) instead of the within-subgroup estimate sigma_hat = R-bar/d2 or s-bar/c4, then reports a capability index that is too low.
Most candidates memorize 'Cpk = min(USL - xbar, xbar - LSL) / 3sigma' as a single formula and never internalize that the sigma in the denominator distinguishes Cp/Cpk (short-term, within-subgroup) from Pp/Ppk (long-term, overall).
What to do instead
Before any capability calculation, write one line at the top of your scratch work: 'within-subgroup sigma = R-bar/d2 (Cp/Cpk) vs overall s (Pp/Ppk)'. If the problem gives subgroup data, compute both and state which index each feeds. If it gives only a single sample with no subgroups, the answer is a P-index.
Uses z = 1.96 for a one-sided 95% confidence bound, or z = 1.645 for a two-sided 95% interval, and also uses z instead of t when n is small and sigma is unknown.
Candidates memorize the single number 1.96 as 'the 95% number' and never link the multiplier to the number of tails or to whether the population standard deviation is known.
What to do instead
Draw a 10-second sketch of the curve and shade the area the problem asks about before selecting a multiplier. Two rules to apply every time: two-sided 95% gives 1.96, one-sided 95% gives 1.645; if sigma is unknown and n < 30, use t with n-1 degrees of freedom.
Rejects the null hypothesis and then writes the conclusion as 'the null hypothesis is proven false' or, conversely, fails to reject and writes 'the null hypothesis is true', rather than stating the decision in terms of the evidence.
Everyday language treats 'not proven' and 'proven true' as opposites, but in hypothesis testing failing to reject only means the sample did not supply sufficient evidence against H0, not that H0 is correct.
What to do instead
Use a fixed sentence template: 'At alpha = ___, there is / is not sufficient evidence to conclude that ___.' Never write 'accept H0' or 'prove'. Also pre-write the alpha/beta trade-off (lowering alpha raises beta, power = 1 - beta) so those items take seconds.
On an attribute control chart item, chooses a c or np chart when the subgroup size varies from period to period, or chooses a p chart when the sample size is constant and the count of defectives is what is being tracked.
Candidates map the mnemonic letters to 'defectives' and 'defects' without checking the two gating conditions - constant vs varying sample size, and count vs proportion.
What to do instead
Run a two-question decision gate before choosing: (1) Is the sample size constant or varying? (2) Am I plotting a count or a proportion/rate? Constant + count of defectives = np; constant + count of defects = c; varying + proportion defective = p; varying + defects per unit = u. Write the chart name and both gating answers on your scratch sheet.
Reacts to a point outside the control limits and a run of seven points on the same side of the centerline in exactly the same way - treating both as a special cause and prescribing a process adjustment - or the reverse, dismissing a run rule as 'just common cause'.
Candidates learn 'points outside the limits = out of control' as the whole rule set and do not apply the run/sensitizing rules, or they apply the rules without asking whether the variation is sustained (common cause) or a discrete upset (special cause).
What to do instead
Memorize the split: a single point beyond 3 sigma is a special-cause signal, and so are runs, trends, and 2-of-3 beyond 2 sigma - all are detectable special-cause patterns. Sustained, stable variation within the limits with no rule violations is common cause, and adjusting the process then is tampering. For each item, first label the pattern, then label the action.
Reads a Gauge R&R table and swaps repeatability and reproducibility, or calls a %GRR of 12% 'unacceptable' because it is not zero, or blames the equipment when the appraiser-to-appraiser spread is the larger component.
Both terms start with 're' and both are expressed as a percentage of total variation, so candidates assign the labels by position in the table rather than by definition.
What to do instead
Anchor the definitions to a single phrase each: repeatability = same appraiser, same part, same instrument (equipment variation, EV); reproducibility = different appraisers (appraiser variation, AV). Then apply the standard acceptance bands (under 10% acceptable, 10-30% conditional on importance and cost, over 30% unacceptable). Identify the larger component before recommending a fix.
Classifies incoming inspection and supplier audits as prevention costs, and rework or warranty claims as appraisal costs, when asked to place a cost in the prevention / appraisal / internal failure / external failure model.
Candidates sort costs by 'when it happens in the flow' rather than by 'is it avoiding defects, finding them before shipment, or finding them after', and inspection feels preventive because it happens early.
What to do instead
Use one test per category: prevention = the money is spent so a defect never exists; appraisal = the money is spent to find defects before the customer sees them; internal failure = found before shipment and the cost is rework/scrap; external failure = found after shipment (warranty, returns, recalls, liability). Incoming inspection is appraisal, not prevention.
Labels a test that confirms a design meets its stated specification as 'design validation', and labels a field trial with actual users as 'design verification'.
Both terms sound like 'checking the design', and candidates recall them from a list rather than from the one-word distinction - requirements vs needs.
What to do instead
Memorize the one-word test: Verification answers 'did we build it right?' against the design inputs/specifications; Validation answers 'did we build the right thing?' against user needs and intended use. If the item mentions user needs, customer use conditions, or clinical/field performance, it is validation.
Places a tool in the wrong DMAIC phase - for example putting measurement system analysis or baseline capability in the Improve phase, or putting the FMEA and root-cause analysis in the Control phase.
Candidates learn the tools as a flat list and never build the phase-to-tool map, so on an exam item they reason from the tool's name instead of the phase's question.
What to do instead
Build a five-column map once and reproduce it on scratch paper at the start of the exam: Define - project charter, SIPOC, VOC/CTQ; Measure - data collection plan, MSA/GR&R, baseline capability; Analyze - root cause, FMEA, hypothesis tests; Improve - DOE, pilot, solution selection; Control - control charts, control plan, mistake-proofing, response plan. Assign the tool by asking which phase question it answers.
On an FMEA item, assigns a low detection number to a failure mode with no current control because 'it would be easy to notice', and then claims a design change that adds an alarm lowers the severity rating.
Candidates carry the intuition that a high score is bad for severity and occurrence, and then apply the same direction to detection, forgetting that the detection scale runs the other way (10 = cannot be detected).
What to do instead
Write 'detection 10 = undetectable, 1 = certain detection' at the top of your scratch sheet before any FMEA item. If no control exists, detection is 10. Remember that severity reflects the effect of the failure on the customer and cannot be reduced by detection or prevention controls - only a design or process change to the effect can lower it. And do not treat an RPN threshold as the only prioritization rule; severity alone can drive action.