Healthcare Administration & Management · Independent exam preparation
Lean Six Sigma Healthcare Green Belt Certification Study Guide
Practise DMAIC decisions, healthcare workflow redesign, risk analysis and statistical tool selection through 60 focused concepts, worked cases and original drills.
By CertMedbry · Updated 3 November 2026
Choose appropriate improvement tools, distinguish process signals from noise, calculate performance measures and interpret statistical evidence in healthcare operations. These 60 concepts connect each principle to a practical decision, an original example and a common mistake. Use the accompanying questions and flashcards to practise reasoning across the curriculum.
Try 20 free practice questionsWhat is the exam format?
Check the question types, timing and outline before planning your revision.
Which certification and level does this guide address?
This guide addresses SSGI’s Lean Six Sigma Healthcare Green Belt Certification. SSGI describes an intermediate-level healthcare program that prepares Green Belts to lead smaller improvement projects and support larger projects under a Black Belt. Its general Green Belt program and healthcare career-track upgrade are separate offerings; their requirements should not be substituted for this examination.
How many questions are on the examination, and which item styles are used?
The issuer’s healthcare certification page, documented on November 3, 2026, states 100 multiple-choice and true/false questions. It does not identify a scored-versus-pretest breakdown. The 20 questions in this guide are independent four-option learning drills and do not reproduce the full mixture of official item styles.
What are the duration, delivery method and passing requirement?
SSGI states that the examination is taken online at the learner’s convenience, with a two-hour limit and a 70% passing score. An examination center is not required. The supplied issuer information does not establish adaptive delivery or detailed proctoring arrangements. These practice questions do not predict examination success.
Which curriculum version and domain weights apply?
The official linked healthcare Green Belt curriculum reviewed October 19, 2026, is undated and has no stated version, effective date or examination topic weights. The eight study domains used here group related curriculum topics; they are not an issuer-published weighted blueprint. Coverage emphasizes breadth without assigning invented percentages. The complete official curriculum remains the full scope, including Taguchi loss, AI in healthcare, named examples and project work beyond this selection.
SSGI: Lean Six Sigma Green Belt Healthcare — Full Course · SSGI: Lean Six Sigma Healthcare Green Belt Certification
Are prerequisites, paid retakes or renewal required?
The issuer page states no prerequisites, unlimited examination retakes without an additional fee and no certification renewal requirement. It also describes lifetime access to the training. These statements apply to this SSGI healthcare Green Belt offering, rather than establishing requirements for other providers or belt levels.
How should the guide and references be used?
Use the official curriculum to check complete topic coverage and the substantive references to understand tool mechanisms and statistical assumptions. Healthcare scenarios here illustrate operational improvement decisions, rather than establish clinical treatment recommendations. Explain each tool choice before calculating, then check the denominator, assumptions and interpretation.
SSGI: Lean Six Sigma Green Belt Healthcare — Full Course · SSGI: What Is DMAIC?
This independent guide prepares learners for Six Sigma Global Institute’s intermediate-level Lean Six Sigma Healthcare Green Belt Certification, displayed by SSGI as LSSGBH. The workbook label LSSGB-H identifies the same credential. The eight study domains organize the linked curriculum; SSGI does not publish examination domain weights. This selective guide does not replace the complete official curriculum, which defines the full scope.
Outline used: Undated official curriculum linked from SSGI’s current healthcare Green Belt certification page, reviewed 2026-10-19. Neither page states a syllabus version, effective date or examination domain weights. The workbook label “Certified Lean Six Sigma Green Belt - Healthcare (LSSGB-H)” maps to SSGI’s Lean Six Sigma Healthcare Green Belt Certification; SSGI’s page uses LSSGBH. The assigned slug is preserved.
60 key concepts: understand, apply and avoid the traps
Work through the questions by exam domain. Explain the answer in your own words, try the example, then check the common pitfall.
Healthcare improvement foundationsnot stated
1.What makes an improvement valuable in healthcare?
Healthcare value connects process performance to what patients and other legitimate customers require. An improvement should address a defined outcome, such as reliable information or timely access, while preserving necessary quality requirements. Faster work alone does not establish greater value if errors or rework increase. Choose measures that show the desired outcome and balancing effects, and distinguish required supporting activities from activities that directly create value for the customer.
Apply it
A scheduling team shortens calls by omitting confirmation of appointment details. Repeat calls rise afterward. Evaluate successful bookings and correction calls alongside handling time before deciding whether the redesigned process improved value for patients and staff.
Common pitfall: Treating every reduction in staff time as improvement can hide transferred work. Check whether another department or the patient must perform the omitted work.
2.How does a Lean culture support improvement?
A Lean culture makes problems visible and engages the people doing the work in systematic improvement. Standard work provides a shared baseline for learning, while observation reveals where actual conditions differ. Leaders support investigation of process conditions rather than relying on individual blame. The discriminating clue is repeated learning from observed work: suggestions should lead to tested changes and revised standards, rather than an isolated cleanup event.
Apply it
Registration staff report that two screens require duplicate entry. A supervisor observes the work with them, tests a revised handoff and reviews correction rates. Staff then help update the standard procedure using the pilot findings.
Common pitfall: A suggestion board alone does not create continuous improvement. Without investigation, feedback and follow-through, visible participation may produce little change in daily work.
3.How does Y = f(x) guide root-cause work?
The expression Y = f(x) represents an outcome as a function of process inputs and conditions. It directs attention from repeatedly inspecting the outcome toward identifying which inputs influence it. Candidate inputs remain hypotheses until supported by observation, analysis or testing. Distinguish a useful explanatory input from an associated variable that merely tracks workload or another underlying factor. Improvement targets verified drivers that the team can influence.
Apply it
A referral team defines turnaround time as Y and lists missing fields, daily arrivals and review staffing as candidate inputs. It collects matched process data before deciding whether form completeness or workload explains the observed delays.
Common pitfall: Naming a plausible input does not prove causation. Changing it without verification may leave the true driver untouched or create an additional process problem.
4.How is the voice of the customer translated into a CTQ?
Voice of the customer captures expressed needs; a critical-to-quality requirement translates an important need into a measurable characteristic. A usable CTQ specifies what is measured, the operational definition and the required performance. For a request for clearer instructions, the team might define completeness against agreed information elements. Select requirements grounded in customer needs and process purpose rather than convenient metrics that do not represent the underlying concern.
Apply it
Patients describe appointment messages as confusing. The team identifies essential message elements, defines a complete-message measure and establishes an agreed target. It checks comprehension feedback before treating message length as a suitable proxy for clarity.
Common pitfall: A vague goal such as better communication cannot support consistent measurement. A convenient count of messages sent may measure activity while missing whether information was usable.
5.How does the voice of the process differ from customer requirements?
Voice of the process describes what the process actually produces, including its typical performance and variation. Customer requirements describe what is needed. Comparing them reveals a performance gap, but neither substitutes for the other. A stable process can consistently miss customer requirements, while a favorable average can conceal unreliable individual outcomes. Use measured process behavior to assess current performance and customer-defined requirements to judge whether that performance is adequate.
Apply it
A service promises confirmation within an agreed interval. Its chart shows stable performance, but many confirmations exceed that interval. The team treats the result as a capability problem requiring process redesign despite the absence of special-cause signals.
Common pitfall: Control limits are not customer acceptance limits. A process can remain statistically predictable while producing outcomes that customers or the organization consider unacceptable.
SSGI: Lean Six Sigma — The Definitive Guide · NIST: What Is Process Capability?
6.Which costs belong to poor quality?
Cost of quality distinguishes prevention, appraisal and failure costs. Prevention seeks to avoid defects; appraisal assesses whether requirements are met. Internal failure costs arise when defects are found before delivery to the customer, while external failure costs follow delivery. Cost of poor quality focuses on failures rather than treating every quality-related expense as waste. Identify when the defect was discovered and why the expenditure occurred before assigning its category.
Apply it
A billing office pays for form-design training, routine audits, correction of rejected internal drafts and handling complaints about issued bills. Classify these as prevention, appraisal, internal failure and external failure, respectively, to expose avoidable rework.
Common pitfall: Reducing inspection spending does not automatically reduce total quality cost. If defects escape and generate more failures, the apparent saving can conceal a larger loss.
7.When should each DMAIC phase drive the next decision?
DMAIC improves an existing process through Define, Measure, Analyze, Improve and Control. Define establishes the problem and requirements; Measure establishes a trustworthy baseline; Analyze verifies causes; Improve tests targeted changes; Control sustains results. The key distinction is the evidence needed before advancing. A suspected cause belongs in analysis until verified, and an attractive solution belongs in a pilot before broad implementation. Phase outputs should support the next decision.
Apply it
A team proposes automated routing for delayed requests. Before purchasing software, it defines the delay metric, checks timestamps and analyzes where requests wait. It pilots routing only if the evidence supports routing as a meaningful cause.
Common pitfall: Assigning a tool to a phase does not establish completion. A fishbone diagram lists possibilities; it does not, by itself, satisfy the need to verify causes.
8.How does measurement systems analysis separate process and measurement variation?
Observed differences can reflect the process, the measurement system or both. Repeatability concerns variation when the same operator measures the same item under comparable conditions; reproducibility concerns differences across operators or relevant conditions. Bias is systematic departure from a reference value. A system can be repeatable yet biased. Verify definitions, instruments and operator agreement before interpreting small process changes, especially when measurement error is large relative to the difference of interest.
Apply it
Two analysts time the same recorded registration sessions using a common start-and-stop definition. Repeated measurements assess within-analyst consistency; comparisons between analysts assess reproducibility. A verified recording clock helps investigate whether both analysts share a systematic timing bias.
Common pitfall: Agreement between observers does not prove accuracy against a reference. Both observers can consistently apply the same incorrect definition or use the same miscalibrated instrument.
NIST: Gauge R & R Studies · 2.4.4.1. Analysis of repeatability · 2.4.4.2. Analysis of reproducibility · 2.4.5. Analysis of bias · SSGI: What Is DMAIC?
Process flow, delays and wastenot stated
9.How do processing time and elapsed time reveal different delays?
Processing time measures time spent performing work, while elapsed lead time includes waiting and other intervals between defined start and finish points. Cycle-time terminology can vary, so state the operational definition before comparing results. When elapsed time greatly exceeds active processing time, queueing and handoffs deserve attention. A faster individual task may barely affect the total journey if most time is spent waiting elsewhere in the process.
Apply it
A request spends 12 minutes in active review and 108 minutes waiting between steps. Its elapsed time is 120 minutes and active processing occupies 10%. Investigate the waiting intervals before expecting major gains from faster typing.
Common pitfall: Adding staff task times does not necessarily equal the customer’s elapsed time. Parallel work and waiting make these quantities different, so define the measurement boundary.
10.When is SIPOC the right mapping tool?
SIPOC summarizes suppliers, inputs, a high-level process, outputs and customers. It is useful early in a project when boundaries, dependencies and output recipients need agreement. The process portion stays broad enough to show the overall transformation rather than every decision. Use SIPOC to clarify what enters and leaves the process, then use a detailed map when investigating specific handoffs, loops or waiting points inside those boundaries.
Apply it
A discharge-document project starts with documents approved for release and ends when the receiving office has them. A SIPOC identifies upstream information suppliers and downstream recipients before the team debates individual electronic routing steps and exceptions.
Common pitfall: Filling a SIPOC with dozens of task boxes defeats its scoping purpose. It also cannot establish which internal step causes the observed performance gap.
SSGI: Lean Six Sigma — The Definitive Guide · SSGI: What Is DMAIC?
11.How does value stream mapping expose system-wide waste?
Value stream mapping connects end-to-end work with the information that authorizes and coordinates it. A current-state map shows actual flow, processing and waiting; a future-state map describes a coherent improved system. This broader view helps prevent one department from accelerating output into another department’s queue. Choose value stream mapping when the problem spans several activities and information handoffs, and validate the map through observation of actual work.
Apply it
A referral pathway crosses intake, authorization and scheduling. Its current-state map shows repeated requests for missing information and long queues between departments. The future-state design aligns information completeness and downstream readiness before increasing the intake team’s output.
Common pitfall: A map of the intended procedure can miss everyday workarounds. Improving one box without considering downstream flow may simply relocate the queue or workload.
12.How should a process map represent decisions and rework?
A detailed process map shows the sequence of activities, decisions, handoffs and returns that produce an output. Decision branches need explicit criteria so different observers classify the same case consistently. Rework loops reveal repeated effort that a simple straight-line description can hide. Choose the detail level needed for the problem, and distinguish observed pathways from assumed ones. A useful map makes the next data-collection question clearer.
Apply it
An intake map includes a completeness decision with separate routes for complete and incomplete submissions. The incomplete route returns to the sender. Counting entries into this loop reveals how often the same request consumes intake effort again.
Common pitfall: A clean sequence of boxes can conceal exceptions and returns. Omitting these routes understates workload and can make the wrong activity appear to be the bottleneck.
13.How does an X-Y matrix prioritize candidate inputs?
An X-Y matrix links candidate process inputs to important outputs. Teams weight outputs by importance, rate each input-output relationship and sum weighted relationship scores to prioritize investigation. For input i, its total is the sum of output weight multiplied by relationship rating across outputs. The matrix organizes expert judgment; it does not estimate causal effects from data. Its best use is narrowing an extensive input list into a manageable study plan, with transparent scoring and sensitivity checks when rankings depend heavily on subjective ratings. Method reference: [Minitab’s X-Y matrix instructions](https://support.minitab.com/en-us/workspace/help-and-how-to/forms/types-of-forms/process-improvement/c-e-matrix-x-y-matrix/).
Apply it
A team weights timely completion more heavily than formatting consistency, then scores how missing fields, routing rules and layout affect both outputs. High-ranked inputs receive measurement attention first; the team does not automatically implement changes based on rank.
Common pitfall: A high total is a prioritization signal, not statistical proof. Subjective relationship ratings can reproduce the team’s assumptions unless subsequent analysis tests them.
SSGI: Lean Six Sigma Green Belt Healthcare — Full Course · SSGI: Lean Six Sigma — The Definitive Guide
14.How does Kanban control replenishment?
Kanban is a signal that authorizes production or movement within a pull system. Downstream consumption triggers replenishment instead of upstream activity being driven solely by a forecast. The signal must correspond to defined quantities, locations and replenishment rules. A visible card without an authorization rule is merely a label. Select Kanban when consumption and replenishment can be connected reliably, and monitor shortages and excess inventory after implementation.
Apply it
An office uses two bins of standardized forms. Emptying the first bin releases a replenishment card while the second supports use during replenishment. Staff order the defined quantity only when the signal occurs, then review stockouts and excess.
Common pitfall: Increasing every bin quantity can hide replenishment problems while creating excess inventory. Kanban requires reliable rules and adequate replenishment performance, not just visible containers.
15.How is takt time calculated and interpreted?
Takt time equals available production time divided by customer demand over the same period. It expresses the pace needed to match demand, rather than the measured duration of an existing task. Compare task or process capacity with takt to identify a mismatch, using consistent units and realistic available time. In variable healthcare operations, this planning relationship does not require every individual encounter to take an identical amount of time.
Apply it
A document service has 420 available minutes and demand for 140 completed packets each day. Takt is three minutes per packet. A four-minute completion interval indicates insufficient capacity to match that demand under the stated operating conditions.
Common pitfall: Dividing demand by available time gives a rate rather than takt time. Using scheduled time without accounting for unavailable periods can overstate the capacity available.
16.Which eight wastes help identify unnecessary work?
The eight-waste framework draws attention to defects, overproduction, waiting, unused talent, transportation, inventory, motion and extra processing. Categories guide observation rather than replace an explanation of the problem. Transportation concerns movement of items or information; motion concerns movement by people. Overproduction creates output earlier or in greater quantity than needed. Identify the mechanism that consumes resources without advancing the required outcome, then consider how removing it affects the system.
Apply it
A coordinator walks repeatedly to a distant printer, produces unused packet copies and corrects duplicated data entry. Classify these observations as motion, overproduction and extra processing or defects as appropriate, then quantify their actual frequency and consequences.
Common pitfall: Arguing over a category can distract from improvement. Some observations fit multiple wastes; describe the actual activity and its impact before choosing an intervention.
17.How does 5S make workplace performance more reliable?
5S organizes work through sorting, setting in order, cleaning, standardizing and sustaining. Its purpose extends beyond appearance: needed items have defined locations, abnormal conditions become visible and routines maintain the arrangement. Standards should reflect actual use and be practical to follow. Choose 5S for searching, clutter or inconsistent placement problems, and verify whether search time and missing-item events improve after the initial organization activity is complete.
Apply it
A shared office removes obsolete forms, places current versions in labeled locations and assigns a daily check. After several weeks, the team measures searching and use of outdated forms to determine whether the arrangement remains effective during normal work.
Common pitfall: A one-day cleanup addresses only part of 5S. Without ownership and sustained standards, clutter and inconsistent placement can return despite an attractive initial result.
Errors and risk mitigationnot stated
18.How should an error be distinguished from its cause and effect?
An error or failure mode describes how a process fails to perform its intended function. A cause explains why that failure occurs, while an effect describes what follows. Keeping these levels separate supports targeted prevention and meaningful risk assessment. Describe the failure in observable process terms before proposing remedies. Broad labels such as human error usually provide insufficient detail to identify the conditions that make the failure possible.
Apply it
For an appointment message, an incorrect date is the failure mode, selection from an outdated calendar is a possible cause, and a missed appointment is a possible effect. The team investigates the calendar-selection mechanism before choosing a safeguard.
Common pitfall: Calling the effect the cause directs attention to the wrong part of the chain. A label such as confusion needs a specific process mechanism before action.
19.When is a checklist an effective safeguard?
A checklist supports reliable execution by making critical steps explicit at the point where they must occur. It is most useful for omission risks in a defined workflow, with clear items and an accountable user. Its effectiveness depends on actual use and the ability to act on findings. A completed box alone does not prove the underlying task occurred. Evaluate omissions and usability rather than counting completed forms alone.
Apply it
A records-transfer team adds a short check of recipient, attachment and required identifiers immediately before sending. During a pilot, observers compare checklist use with actual completeness and examine whether staff can resolve missing information before transmission.
Common pitfall: A long checklist can become a documentation exercise. Items that are vague, mistimed or impossible to verify encourage superficial completion without reducing the underlying failure.
SSGI: Lean Six Sigma Green Belt Healthcare — Full Course · Institute for Healthcare Improvement: FMEA Tool
20.How does Poka-Yoke prevent or immediately detect mistakes?
Poka-Yoke, or error-proofing, changes the process so a mistake is prevented or detected at its source. Examples include incompatible connections, constrained choices and checks that block an incomplete transaction. Prevention removes the opportunity; detection identifies an error before it travels downstream. Prefer a mechanism linked directly to the failure over a general reminder. Verify that the safeguard handles legitimate exceptions without encouraging routine bypass or creating a different error.
Apply it
A submission system refuses to send a request when a required routing field is blank and identifies the missing field immediately. The team tests ordinary requests and valid exceptions to confirm that the block prevents omissions without unnecessary workarounds.
Common pitfall: A warning that users routinely dismiss offers weaker control than preventing the invalid action. Automation alone is not error-proofing unless its mechanism addresses the failure.
21.How does FMEA prioritize prospective process risks?
Failure mode and effects analysis prospectively examines process steps, possible failures, their causes and their effects. In IHI’s method, severity, occurrence and detection ratings each range from 1 to 10, and RPN equals severity × occurrence × detection. Higher severity means more severe consequences; higher occurrence means more likely failure. Crucially, a higher detection rating means the failure is less likely to be detected: 10 represents very unlikely detection. Better detection therefore lowers this rating. Scores organize attention but are not precise probabilities or complete decisions. Review severe consequences explicitly and select actions that prevent occurrence or detect failure before downstream consequences.
Apply it
Before introducing electronic routing, a team maps each handoff and assigns owners for safeguards. In an explicitly hypothetical example, severity 8, occurrence 4 and detection 7 give RPN = 8 × 4 × 7 = 224. A tested alert improves detection, reducing its rating to 2 while severity remains 8 and occurrence remains 4; RPN becomes 64. The alert changes detectability, not the inherent severity of an escaped failure.
Common pitfall: Different rating combinations can produce the same product. Never prioritize solely by RPN or ignore a severe failure because its total is modest. Reassess ratings after testing the safeguard.
Institute for Healthcare Improvement: FMEA Tool · QIToolkit_FailureModesandEffectsAnalysis.pdf
22.How should a risk-reduction change be tested before expansion?
A risk-reduction proposal needs evidence that its mechanism works in actual process conditions. Define the targeted failure, the expected effect and relevant unintended consequences, then test on a manageable scale. Track both failures and opportunities for failure so changes in workload do not mislead. A successful pilot should support a decision about refinement or expansion, with updated procedures and ownership rather than assuming that implementation alone establishes effectiveness.
Apply it
A team pilots a required-field rule on one request type. It measures incomplete submissions per eligible request, completion delays and bypass attempts. The results guide whether to adjust the rule before applying it to other request categories.
Common pitfall: A fall in the number of errors can reflect lower volume. Without a denominator and balancing measures, the team may expand a safeguard that merely obstructs work.
SSGI: What Is DMAIC? · Institute for Healthcare Improvement: FMEA Tool
Graphical analysis and process variationnot stated
23.What can a histogram reveal about process data?
A histogram groups numerical observations into intervals and displays their frequencies. It reveals distributional features such as spread, skewness, multiple peaks and possible extreme values. Its appearance depends partly on interval width and available data. Because observations are grouped by value, time order is lost. Use a histogram to explore distribution shape, then examine time-ordered or stratified data before concluding that the process is stable or identifying a cause.
Apply it
A histogram of turnaround times shows two peaks. The team separates requests by service route and checks collection times. Distinct workflows may explain the pattern, but the pooled histogram alone cannot identify whether a change occurred during the month.
Common pitfall: A bell-shaped histogram does not establish statistical control. A process can change over time while its pooled observations still form an apparently smooth distribution.
24.How does a Pareto chart guide prioritization?
A Pareto chart orders categories by their contribution to a chosen measure, often adding a cumulative percentage. It helps identify where concentrated effort may address the greatest measured burden. The measure matters: frequency, cost and time can produce different rankings. Use consistently defined categories and a common observation period. The chart identifies large contributors but does not establish their causes or require an exact eighty-twenty relationship in every dataset.
Apply it
An office classifies 200 correction requests by primary reason and orders the categories by count. It then compares a second Pareto chart based on correction hours because a less frequent category may consume substantially more total staff time.
Common pitfall: The most frequent defect is not automatically the most consequential. Choosing a frequency chart when severity or cost drives the decision can misdirect improvement effort.
25.How should a fishbone diagram be used in analysis?
A fishbone diagram organizes possible causes of a specified effect into useful categories. It broadens inquiry and makes assumptions visible, helping teams avoid fixing on the first explanation. Its branches represent hypotheses rather than verified causal relationships. Write a precise effect, involve people familiar with the process and select plausible branches for observation or measurement. Verification requires additional evidence that connects a proposed cause to the actual performance gap.
Apply it
For delayed request completion, a team lists possible causes involving methods, information, staffing and technology. It chooses missing authorization details for further study and compares delay patterns across complete and incomplete requests before proposing a form redesign.
Common pitfall: A crowded fishbone can look authoritative while containing only opinions. Counting how often a cause is mentioned does not establish its influence on the outcome.
Institute for Healthcare Improvement: Cause and Effect Diagram
26.Why does common-cause variation require a system response?
Common-cause variation arises from the ordinary, continuing conditions of a process. When a suitable control chart shows stable behavior without defined signals, isolated ups and downs should not automatically trigger different local corrections. Improving that stable pattern usually requires changing the system that generates it. Distinguish stability from adequacy: predictable performance can still be poor. Evaluate sustained process changes rather than blaming individuals for expected random differences between observations.
Apply it
Daily processing averages fluctuate within a stable chart pattern, yet service expectations remain unmet. The team redesigns the shared intake method and evaluates its effect instead of asking each day’s slowest employee to explain ordinary variation.
Common pitfall: Adjusting the process after every ordinary fluctuation can add variation. A low day and a high day do not necessarily represent separate causes requiring separate fixes.
27.When does special-cause variation call for investigation?
Special-cause variation reflects a departure from the established process pattern. A point beyond an appropriate control limit or another prespecified nonrandom pattern provides evidence for investigation. The chart does not identify the cause or prove that an error occurred. Confirm data quality and examine contemporaneous process conditions. An unusual improvement also deserves attention because understanding its cause may reveal an opportunity to reproduce better performance deliberately and safely.
Apply it
A completion-time chart shows a point beyond its upper control limit on the day a routing outage occurred. The team verifies timestamps, investigates the outage and documents its effect rather than automatically recalculating limits to include the disruption.
Common pitfall: Deleting an unusual point merely because it is inconvenient hides information. Exclusion or new limits require a documented process explanation and an appropriate stable baseline.
28.How do multi-vari displays guide further investigation?
Multi-vari displays compare variation across structured sources, such as positions within a unit, units within a period and periods over time. Their purpose is to show where the largest differences appear so subsequent investigation can focus efficiently. Sampling must preserve the structure being compared. A visible separation suggests a source worth studying; it does not eliminate confounding or prove causation. Choose categories that correspond to plausible process mechanisms.
Apply it
A team samples comparable requests from several desks during morning and afternoon periods. A multi-vari display shows much larger differences between periods than between desks. The next study examines time-related workload and routing conditions rather than immediately retraining one desk.
Common pitfall: Convenience samples can confound the displayed groups. If one desk handles only complex requests, its apparent effect may reflect request type rather than desk performance.
29.Why must the process mean and dispersion be assessed separately?
The mean describes location, while dispersion describes how widely observations vary around that location. Two processes can share the same mean but differ greatly in reliability. Moving the mean toward a target does not necessarily reduce variation, and reducing variation does not necessarily correct an off-target mean. Select measures for both characteristics when consistency matters, and inspect distribution shape because a single average can conceal skewness or distinct process populations.
Apply it
Two offices average 20 minutes per request. One usually takes 18–22 minutes; the other ranges from 5–35 minutes. A team evaluating reliability compares their spread and time patterns rather than declaring equivalent performance from their means.
Common pitfall: An improved average can coexist with more extreme delays. Reporting only the mean may conceal a widening distribution that matters to customers and process owners.
Probability and process performancenot stated
30.How is standard deviation interpreted in process data?
Standard deviation summarizes spread around the mean in the original measurement units. The sample calculation squares deviations, sums them, divides by n minus one and takes the square root. Squaring gives large deviations substantial influence, so extreme observations can strongly affect the result. Standard deviation describes variability rather than measurement accuracy or the precision of the sample mean. Compare it only across appropriately defined and comparable process populations.
Apply it
For observed times of 8, 10 and 12 minutes, the mean is 10. Squared deviations total eight; dividing by two gives sample variance four. The sample standard deviation is therefore two minutes, describing spread among the observations.
Common pitfall: Standard deviation and standard error answer different questions. Dividing by the square root of sample size estimates mean precision, rather than reducing actual process variation.
NIST: Measures of Scale · NIST: Confidence Limits for the Mean
31.When are normal-distribution coverage rules useful?
A normal distribution is symmetric and characterized by its mean and standard deviation. Approximately 68.27%, 95.45% and 99.73% of its population lie within one, two and three standard deviations of the mean. These are model-based coverage statements, not universal properties of process data. Use them when a normal model is reasonable, and investigate skewness, mixtures or instability before applying normal tail probabilities to operational performance.
Apply it
A stable, approximately normal processing measure has mean 30 minutes and standard deviation four minutes. About 95.45% of values are expected between 22 and 38 minutes under that model, rather than within four minutes of the mean.
Common pitfall: Applying normal coverage to strongly skewed waiting times can misstate tail risk. A mean and standard deviation alone do not establish that normal modeling is suitable.
32.How does a Z value locate an observation?
A standardized Z value equals the observation minus the population mean, divided by the population standard deviation. It expresses distance from the mean in standard-deviation units; its sign indicates direction. Standardization itself does not make a nonnormal distribution normal. To translate Z into a normal tail probability, justify the normal model. For a threshold, distinguish the probability below it, above it or between two limits before consulting a table.
Apply it
Under an appropriate normal model, mean completion time is 40 minutes and standard deviation five minutes. A 50-minute threshold has Z equal to two. The relevant exceedance probability is the upper tail beyond two, not the area below it.
Common pitfall: Automatically applying a 1.5-standard-deviation shift changes the stated model. Use the distribution and assumptions supplied by the problem rather than importing an unstated convention.
33.How do defects per unit differ from defective-unit proportion?
Defects per unit equals total defects divided by units examined. A unit may contain several defects, so this measure can exceed one. Defective-unit proportion counts each unit with at least one defect once, regardless of how many defects it contains. Choose the measure according to whether the question concerns defect burden or the chance that a unit is unacceptable. Keep unit and defect definitions consistent across comparisons. Formula reference: [Minitab’s defect metrics](https://support.minitab.com/en-us/minitab/help-and-how-to/quality-and-process-improvement/capability-analysis/supporting-topics/capability-metrics/what-are-dpu-dpo-and-dpmo/).
Apply it
An audit finds 30 errors across 100 records, with errors concentrated in 20 records. Defects per unit is 0.30, while defective-unit proportion is 0.20. The two measures describe different aspects of the same audit results.
Common pitfall: Treating every error as a separate defective record inflates the defective proportion when records contain multiple errors. Count units and defects separately before calculating either measure.
SSGI: Lean Six Sigma — The Definitive Guide · SSGI: Lean Six Sigma Green Belt Healthcare — Full Course
34.How is DPMO calculated with a defensible denominator?
Defects per million opportunities equals defects divided by total eligible defect opportunities, multiplied by one million. If every unit has the same number of defined opportunities, the denominator is units times opportunities per unit. Otherwise, total the eligible opportunities directly. Opportunities must represent meaningful, consistently applied requirements. Comparisons become misleading when one process uses a broader or more generous opportunity definition merely to produce a smaller reported DPMO. Formula reference: [Minitab’s DPO and DPMO definitions](https://support.minitab.com/en-us/minitab/help-and-how-to/quality-and-process-improvement/capability-analysis/supporting-topics/capability-metrics/what-are-dpu-dpo-and-dpmo/).
Apply it
A review examines 200 forms with five eligible required fields each and finds eight field errors. There are 1,000 opportunities, giving 8,000 DPMO. The calculation assumes all five defined fields are genuinely applicable to every reviewed form.
Common pitfall: Using the number of defective forms as the defect count changes the metric. Adding irrelevant opportunities also dilutes DPMO without improving any actual process outcome.
SSGI: Lean Six Sigma — The Definitive Guide · SSGI: Lean Six Sigma Green Belt Healthcare — Full Course
35.What does first-time yield reveal about hidden rework?
First-time yield is the proportion of units that meet the defined requirements on their initial pass without rework or correction. Final acceptance after repairs answers a different question and can conceal repeated effort. Define the process boundary and what counts as rework before collecting data. A process with excellent final acceptance can still have poor first-time yield, signaling substantial hidden work even when every output is eventually acceptable. Definition and formula reference: [ASQ’s first-pass-yield definition](https://asq.org/quality-resources/quality-glossary/f).
Apply it
Of 120 packets entering a review step, 102 pass immediately, 15 pass after correction and three remain incomplete. First-time yield is 102 divided by 120, or 85%, regardless of how many corrected packets ultimately become acceptable.
Common pitfall: Counting repaired units as first-pass successes masks the hidden workload. Final yield and first-time yield should remain distinct when evaluating process quality and improvement.
SSGI: Lean Six Sigma — The Definitive Guide · SSGI: Lean Six Sigma Green Belt Healthcare — Full Course
36.How does rolled throughput yield describe an entire pathway?
Rolled throughput yield describes the probability of completing a multistep pathway without a defect or rework at any step. Multiply the relevant first-pass yields across the pathway, using conditional yields for units reaching each step without earlier failure. The product exposes cumulative losses that strong-looking individual percentages can conceal. Ensure that the step definitions and populations align; multiplying unrelated marginal percentages does not necessarily estimate the pathway’s actual defect-free performance. Formula reference: [Minitab’s throughput-yield definitions](https://support.minitab.com/en-us/minitab/help-and-how-to/quality-and-process-improvement/six-sigma/supporting-topics/what-are-throughput-yield-ytp-and-rolled-throughput-yield-yrt/).
Apply it
Three sequential steps have conditional first-pass yields of 0.90, 0.95 and 0.98 for the relevant pathway. Rolled throughput yield is 0.8379, or 83.79%. The result is lower than every individual step yield because losses accumulate.
Common pitfall: Averaging step yields does not measure successful passage through every step. Likewise, final acceptance after rework cannot substitute for the defect-free pathway measure.
SSGI: Lean Six Sigma — The Definitive Guide · SSGI: Lean Six Sigma Green Belt Healthcare — Full Course
Statistical process control and response plansnot stated
37.How should sampling support a process decision?
A sampling plan defines the population, selection method, timing, sample size and measured characteristics. Representative selection reduces systematic bias, while adequate sampling across relevant conditions supports useful conclusions. Larger samples improve precision only for the population actually sampled; they do not repair exclusion of evenings, complex requests or another important segment. For monitoring, consistent sampling also matters because changing the collection method can create an apparent process change.
Apply it
A service operates across three shifts, but its audit covers only weekday mornings. The team samples across shifts and days using a defined selection procedure before estimating overall completeness or building a baseline intended to represent the whole service.
Common pitfall: A very large convenience sample can produce a precise estimate of the wrong population. Increasing sample size does not remove selection bias or inconsistent definitions.
SSGI: Lean Six Sigma Green Belt Healthcare — Full Course · NIST: What Are Control Charts?
38.Which control chart matches the data structure?
Control-chart selection begins with the measurement type and sampling structure. Numerical measurements collected in rational subgroups can use mean charts paired with range or standard-deviation charts. Binary conforming/nonconforming outcomes can support a p chart of proportions. Defect counts are different from defective-unit counts. Select a chart whose probability model and denominator match the observations, and preserve time order so the chart can detect changes in process behavior.
Apply it
An audit records whether each request is complete and reviews different numbers each day. A p chart fits the daily incomplete proportion. A mean-and-range chart would instead require numerical measurements grouped under an appropriate sampling plan.
Common pitfall: Using a p chart for total errors treats multiple defects within one unit as separate binary outcomes. Chart choice depends on what each observation represents.
NIST: Proportions Control Charts · NIST: Shewhart X-bar and R and S Control Charts
39.Why does rational subgrouping matter?
Rational subgrouping collects observations so within-subgroup variation reflects short-term common conditions, while differences between subgroups can reveal changes over time. Subgroups should follow a defensible process rationale rather than administrative convenience. Combining unlike streams within each subgroup may inflate estimated background variation and conceal meaningful differences. Define subgroup membership, timing and size before analysis, and verify that the observations are sufficiently comparable for the selected chart model.
Apply it
A team collects five consecutive completion times every hour under comparable routing conditions. It avoids creating each subgroup from one observation per shift, because that mixture would blend potential shift differences into the estimate of within-subgroup variation.
Common pitfall: A subgroup is not automatically rational because its size is convenient. Mixing process streams can widen limits and make real changes harder to detect.
40.How do control limits differ from specification limits?
Control limits are calculated from process behavior to identify statistically unusual observations under a chosen chart model. Specification limits express requirements or acceptance boundaries established for the output. They answer different questions: whether the process has changed and whether its output meets requirements. Never substitute specifications for calculated control limits. A process can be stable but incapable, and an unstable process can temporarily produce only outputs within specification.
Apply it
A stable chart shows completion times centered above an agreed maximum. The team uses the chart to understand predictability and the requirement to identify unacceptable performance. It plans a system change rather than declaring success because points remain inside control limits.
Common pitfall: Moving control limits to match a target destroys their statistical purpose. Tightening a requirement also does not physically reduce process variation or improve process performance.
NIST: What Are Control Charts? · NIST: What Is Process Capability?
41.How do false alarms and missed shifts trade off in monitoring?
A monitoring false alarm signals a change when the process remains in its baseline condition; a missed shift fails to signal a real change. Narrower limits or additional signal rules generally increase sensitivity but can also increase false alarms. Wider limits reduce false alarms while making smaller shifts harder to detect. Choose the chart and prespecified rules for the monitoring purpose, recognizing that repeated observations and multiple rules affect overall alarm behavior.
Apply it
A team adds several run rules after missing a small sustained change. Before adoption, it considers the increased investigation burden and defines how staff will respond. It avoids changing rules whenever the latest result feels inconvenient or surprising.
Common pitfall: The false-alarm probability for one point is not the probability of ever seeing an alarm over many points. More opportunities to signal increase cumulative alarm risk.
NIST: What Are Control Charts? · NIST: What Are Statistical Tests?
42.How are paired X-bar and R charts constructed?
For equal-sized rational subgroups, calculate each mean and range. Let grand mean be the average subgroup mean and R-bar the average range. X-bar centerline = grand mean; limits = grand mean ± A2 × R-bar. R centerline = R-bar; lower and upper limits = D3 × R-bar and D4 × R-bar. Constants depend on subgroup size n. Examine dispersion first because mean-chart limits depend on its estimate.
Apply it
An office samples four comparable times hourly. Hypothetical baseline summaries are grand mean 20 minutes and R-bar 4 minutes. For n = 4, A2 = 0.729, D3 = 0 and D4 = 2.282. X-bar limits are 20 ± 2.916, or 17.084 and 22.916; R limits are 0 and 9.128 minutes. Plot subsequent means and ranges against these respective limits.
Common pitfall: SSGI labels its X-bar/R modules N<12. NIST generally recommends ranges through about n=10; this is a statistical recommendation, distinct from the issuer’s curriculum bands. Monitoring means alone misses dispersion changes.
NIST: Shewhart X-bar and R and S Control Charts · SSGI: Lean Six Sigma Green Belt Healthcare — Full Course
43.How are paired X-bar and S charts constructed?
Average the subgroup standard deviations to obtain S-bar. X-bar centerline = grand mean; limits = grand mean ± A3 × S-bar, where A3 = 3/(c4 × √n). S centerline = S-bar; limits = B3 × S-bar and B4 × S-bar. Here B3 = max(0, 1 − 3√(1 − c4²)/c4) and B4 = 1 + 3√(1 − c4²)/c4. The constants depend on n. Check the S chart first.
Apply it
A service collects 20 comparable observations per interval and uses the corresponding constants. The team checks subgroup standard deviations for unstable dispersion before interpreting subgroup means. Standard deviations use all observations and are generally preferred to ranges for larger subgroups.
Common pitfall: SSGI publishes X-bar/S bands 12<N<25 and an additional X-bar module N>25. Those titles leave N=12 and N=25 unspecified; they establish no boundary-case rule. Never substitute S values into R-chart formulas.
NIST: Shewhart X-bar and R and S Control Charts · SSGI: Lean Six Sigma Green Belt Healthcare — Full Course
44.How are p-chart limits calculated and interpreted?
A p chart monitors the proportion of nonconforming units under a binomial model. Estimate the center using total nonconforming units divided by total units, then calculate approximate three-sigma limits as p-bar plus or minus three times the square root of p-bar times one minus p-bar divided by subgroup size. Limits vary when sample sizes vary and are bounded by zero and one. Sparse outcomes may require more suitable binomial treatment.
Apply it
With baseline p-bar of 0.10 and a subgroup of 100 requests, the standard deviation is 0.03. Approximate limits are 0.01 and 0.19. A subgroup of 400 has narrower limits because its estimated proportion is more precise.
Common pitfall: A simple average of daily percentages misweights unequal samples. Also, a negative calculated lower limit represents no feasible proportion and is conventionally bounded at zero.
45.How do Cp and Cpk distinguish spread from centering?
For an appropriate stable normal model, Cp equals specification width divided by six process standard deviations. Cpk is the smaller distance from the mean to either specification limit divided by three standard deviations. Cp describes potential fit based on spread; Cpk also reflects centering. Cpk cannot exceed Cp under these definitions. Check stability, distributional assumptions and the variation estimate before interpreting either index, and use the relevant one-sided expression when appropriate.
Apply it
A stable process has limits 10 and 22, mean 18 and standard deviation two. Cp is one. Cpk is the smaller of 4/6 and 8/6, or about 0.67, showing that off-center performance limits capability.
Common pitfall: A favorable Cp can hide poor centering. Neither index is a substitute for checking the actual distribution, stable behavior and suitability of the specification limits.
46.Why must stability be established before capability is projected?
Capability projects how a process distribution fits requirements. If the process changes unpredictably, a single historical mean and variation estimate may not describe future behavior. Establish a defensible stable process state before making a continuing capability claim. Historical nonconformance can still be reported descriptively, but it is different from demonstrating predictable capability. Investigate distinct process states rather than pooling them into one apparently comprehensive index that obscures their differences.
Apply it
A monthly dataset combines performance before and after a routing failure. Instead of reporting one Cpk as future capability, the team investigates the change, separates justified process states and evaluates the stable operating condition with appropriate assumptions.
Common pitfall: More pooled data do not solve instability. A precise calculation from a mixture of changing conditions can give unjustified confidence about future process performance.
47.What belongs in an actionable control response plan?
A response plan specifies the monitored measure, sampling method, signal criteria, responsible person, investigation steps and escalation or containment actions. It connects detection to a consistent operational response. Include how data quality will be checked, how findings will be documented and when normal operation or revised limits are justified. The plan should distinguish a statistical signal from an acceptance failure because either can require action for a different reason.
Apply it
After a request-routing pilot, the owner receives a defined chart, collection schedule and escalation procedure. An upper-limit signal triggers timestamp verification and routing investigation; a missed service requirement follows the separately specified service-recovery procedure, even without a chart signal.
Common pitfall: A dashboard without assigned action can document deterioration without addressing it. Automatic limit recalculation after every signal can also normalize an unresolved process change.
Hypothesis testingnot stated
48.How should null and alternative hypotheses be specified?
A null hypothesis states the reference claim being tested, commonly equality or no difference. The alternative states the departure of interest. Choose a one-sided or two-sided alternative from the substantive question before examining results. The p value evaluates how incompatible the observed evidence is with the null model; it is not the probability that the null is true. Link hypotheses to a population parameter and a clearly defined measurement.
Apply it
A team asks whether a redesign changes mean completion time in either direction. It specifies equal means as the null and unequal means as the two-sided alternative before analysis, even though the team hopes the redesigned process will be faster.
Common pitfall: Choosing a one-sided test after seeing the observed direction makes the stated error control unreliable. The desired conclusion should not determine the hypothesis retrospectively.
49.How do Type I error, Type II error and power differ?
A Type I error rejects a true null hypothesis, with probability controlled by alpha under the test assumptions. A Type II error fails to reject a false null, with probability beta for a specified alternative. Power is one minus beta and depends on effect size, variability, sample size and the test. Failure to reject is not proof of equality. Plan sample size around a meaningful difference rather than interpreting a weak study as reassurance.
Apply it
A small pilot finds no statistically significant time reduction. Its confidence interval includes both useful improvement and no change. The team recognizes limited precision and plans adequate sampling for the meaningful effect instead of declaring the methods equivalent.
Common pitfall: Beta is not one universal value for a study; it varies with the true effect considered. A nonsignificant result can reflect insufficient power rather than equality.
50.How is a one-sample t-test carried through to a decision?
A one-sample t-test compares a population mean with a specified reference when the population standard deviation is unknown. Its statistic is t = (sample mean − reference)/(s/√n), with n − 1 degrees of freedom. Independent observations and a suitable sampling distribution are essential; small samples require particular attention to normality and extreme observations. Prespecify the alternative and alpha. Reject the null when the statistic enters the corresponding rejection region, or equivalently when the p value is below alpha; otherwise, do not reject.
Apply it
Prespecify H0: mean = 20 minutes, H1: mean ≠ 20 minutes and alpha = 0.05. For 16 independent times with mean 22 and s = 4, standard error = 4/√16 = 1 and t = 2 with 15 degrees of freedom. The two-sided critical magnitude is approximately 2.131. Because |2| < 2.131, do not reject H0. Equivalently, the two-sided p value is approximately 0.064 > 0.05. The evidence does not establish a mean difference at this level, and it does not establish equality.
Common pitfall: Using s alone as the denominator ignores sample size. A test of a mean does not establish that every request meets the reference, and non-rejection does not prove the reference mean is correct.
NIST: Confidence Limits for the Mean · 1.3.6.7.2. Critical Values of the Student's-t Distribution
51.When should independent groups be compared with a two-sample t-test?
A two-sample t-test compares population means using observations from two independent groups. The sampling relationship determines independence; matching or repeated measurement requires a different treatment. The pooled version assumes equal population variances, while Welch’s approach allows unequal variances. Choose the method from design and assumptions rather than automatically pooling. Examine distributions and unusual observations, and distinguish evidence of a mean difference from evidence that the group assignment caused that difference.
Apply it
Two offices contribute samples of different, unrelated requests. Their sample sizes and variances differ substantially. An independent comparison using Welch’s t procedure is more suitable than pairing requests by their arbitrary row order or assuming equal variances without justification.
Common pitfall: Before-and-after labels do not automatically imply pairing. Conversely, ignoring genuine repeated measurements discards the within-unit relationship and uses the wrong standard error.
52.Why does a paired t-test analyze differences?
A paired t-test applies a one-sample t procedure to within-pair differences. Pairing arises from repeated measurements on the same units or a defensible matching design. The relevant spread is the standard deviation of differences, rather than the separate spreads of the two measurements. Pairs should be independent of other pairs, and the distribution of differences must support the test. Consistent pairing can remove substantial between-unit variability from the comparison.
Apply it
The same 12 standardized requests are processed under two procedures in a controlled comparison. The analyst computes each request’s time difference and tests the mean difference, rather than treating the 24 recorded times as unrelated observations.
Common pitfall: Equal sample sizes do not create valid pairs. Pairing unrelated observations by spreadsheet position invents a relationship and can distort the estimated uncertainty of the comparison.
53.What does a one-factor ANOVA establish?
One-factor ANOVA compares population means across groups by assessing between-group variation relative to within-group variation. Its null states that all group means are equal. A significant overall result indicates that at least one mean differs; it does not identify the differing groups or show that every pair differs. The classical procedure assumes independent errors, normal error distributions and equal variances. Planned contrasts or adjusted follow-up comparisons address specific differences.
Apply it
A team compares average completion times across four independent routing methods. A significant ANOVA motivates appropriate follow-up comparisons. The team does not conclude that the method with the smallest sample mean differs significantly from every other method.
Common pitfall: Running every possible unadjusted t-test increases the chance of false discoveries. A significant overall ANOVA also does not by itself establish practical importance or causation.
54.How should a confidence interval for a mean be interpreted?
A confidence interval combines an estimate with uncertainty from a specified sampling method. For a mean with unknown population standard deviation, a conventional t interval uses the sample mean plus or minus a t multiplier times s divided by the square root of n. A 95% procedure covers the fixed population mean in 95% of repeated samples under its assumptions. It is not an interval containing 95% of individual observations.
Apply it
An estimated mean reduction is four minutes with a 95% confidence interval from one to seven minutes. The interval supports a positive mean reduction under the model, while showing that its magnitude remains uncertain and may need comparison with a meaningful operational threshold.
Common pitfall: A narrow interval can still be biased if sampling or measurement is flawed. Confidence describes the procedure’s coverage, rather than guaranteeing the accuracy of one dataset.
55.How are Wilcoxon rank-sum and signed-rank procedures selected?
Wilcoxon rank-sum, also called Mann–Whitney, compares two independent groups with independent, rankable observations. Its equal-distribution null supports exchangeability; interpreting a result as a median difference additionally requires a common-shape, common-spread location-shift model. Wilcoxon signed-rank instead analyzes paired numerical differences, requiring independent pairs and differences symmetric about the tested location. Under symmetry, the pseudomedian equals the median of differences, which need not equal the difference of marginal medians. Neither procedure generally tests means. Procedure reference: [R’s Wilcoxon documentation](https://stat.ethz.ch/R-manual/R-devel/library/stats/html/wilcox.test.html).
Apply it
Two independent service groups supply ordinal ratings with a ceiling effect and many ties. For rank-sum, assign tied values average ranks and use tie-aware permutation inference or a tie-adjusted approximation. Report rank evidence; the ceiling effect weakens a simple median-shift interpretation.
Common pitfall: Nonparametric does not mean assumption-free. Pairing changes the procedure, ordinal category codes do not automatically justify numerical differences, and ties require appropriate inference.
SSGI: Lean Six Sigma Green Belt Healthcare — Full Course · NIST: What Are Statistical Tests?
Regression analysisnot stated
56.How should a regression slope and prediction be interpreted?
In a simple linear model, the slope gives the predicted change in the response for a one-unit increase in the predictor. The intercept is the predicted response when the predictor is zero, which may lack practical meaning outside the observed range. A fitted relationship supports conditional prediction under the model, not automatic causal inference. Check predictor units and observed range before using the equation, especially when extrapolation would extend beyond available evidence.
Apply it
A model predicts completion time as 12 plus 0.4 times queue size. At a queue size of 20, the prediction is 20 minutes. The slope represents 0.4 additional predicted minutes per extra queued request within the modeled conditions.
Common pitfall: A positive slope does not prove that changing the predictor alone will change the outcome. Shared workload factors or other omitted variables may explain the association.
57.What does least-squares fitting minimize?
Ordinary least squares chooses model coefficients to minimize the sum of squared response residuals, where each residual is observed response minus fitted response. Squaring gives large residuals substantial influence and prevents positive and negative discrepancies from canceling. The criterion concerns vertical response differences, rather than the shortest geometric distance to a line. A best-fitting line under this criterion is not necessarily a scientifically suitable model or a reliable extrapolation tool.
Apply it
A request takes 28 minutes when the fitted model predicts 23. Its residual is positive five minutes and its squared contribution to the least-squares objective is 25. A second residual of negative five contributes the same squared amount.
Common pitfall: Minimizing raw residual sums would allow cancellation and would not define the usual regression fit. A small fitted error also cannot establish causal validity.
58.How do residual plots reveal model problems?
Residual plots assess whether unexplained variation behaves consistently with the fitted model. Curvature can indicate a missing nonlinear relationship; a widening spread can indicate nonconstant variance; time patterns can indicate dependence or an omitted process change. Inspect unusual observations rather than deleting them automatically. Ordinary least-squares inference commonly relies on independent, constant-variance errors, with normal-error assumptions supporting conventional small-sample tests. A high goodness-of-fit measure does not replace these checks.
Apply it
A linear workload model has residuals below zero at low and high workload but above zero in the middle. The team considers a curved relationship and checks process explanations before using the existing line for staffing-related predictions.
Common pitfall: A residual average near zero is expected in ordinary least squares with an intercept. It does not establish that the residuals lack curvature, dependence or changing spread.
NIST: Linear Least Squares Regression · 4.4.4. How can I tell if a model fits my data?
59.How does statistical significance differ from useful prediction?
A coefficient test evaluates evidence against a specified coefficient value, commonly zero, under the fitted model. Statistical significance does not establish a large effect, strong prediction or operational importance. Large datasets can identify very small relationships, while uncertain estimates can leave important effects unresolved. Examine coefficient magnitude, units, confidence intervals and residual variation together. A decision about usefulness should compare the estimated effect and prediction uncertainty with a meaningful process requirement.
Apply it
A model finds a statistically significant increase of 0.02 minutes per additional request, but individual prediction errors remain large. The team evaluates the effect over realistic workload changes and the uncertainty before deciding whether the model supports operational planning.
Common pitfall: A small p value is not a measure of prediction accuracy. A significant predictor may explain too little practical variation to support individual operational decisions.
NIST: Linear Least Squares Regression · NIST: Quantitative Techniques
60.How are coefficients interpreted in multiple regression?
Multiple regression models the response using several predictors. Each coefficient represents the predicted response change for a one-unit predictor increase while other included predictors are held constant. This adjusted association differs from a simple unadjusted comparison. Strongly related predictors can make individual coefficients unstable or imprecise, and omitted variables can still bias interpretation. Check model form, supported predictor combinations and residual behavior before using adjusted coefficients as a basis for decisions.
Apply it
A model includes queue size and request complexity. The queue coefficient describes the predicted time difference at the same modeled complexity, rather than the raw difference between busy and quiet periods. The team checks whether both predictors vary sufficiently independently.
Common pitfall: Holding included predictors constant does not remove all confounding. An adjusted coefficient remains model-dependent and cannot, by itself, establish the effect of an intervention.
What to focus on during revision
The following are CertMedbry study suggestions. Use the official outline to check the full scope of your examination.
- Explain the evidence needed to move from a DMAIC hypothesis to a verified cause, a tested change and sustained control.
- Select SIPOC, detailed mapping, value stream mapping, an X-Y matrix or a graphical tool from the decision being made.
- Calculate takt, DPU, DPMO, first-time yield and rolled throughput yield using explicitly defined units and denominators.
- Distinguish failure modes, causes and effects; explain how a proposed safeguard prevents or detects the specified failure.
- Separate mean from spread, common causes from special causes, and control limits from specification limits.
- Choose charts and statistical tests from measurement type, subgroup structure, independence and pairing.
- Interpret confidence intervals, capability indices and regression coefficients with their assumptions and practical limitations.
- Check the full official curriculum for additional scope, including Taguchi loss, AI in healthcare, issuer examples and project work.
A practical preparation plan
1. Build a decision map
Organize the official curriculum around DMAIC. For each tool, write the question it answers, required data and output needed for the next decision. Keep examination logistics separate from your substantive study notes.
2. Follow one healthcare workflow
Use an operational example such as referral handling. Define customer requirements, build a SIPOC and detailed map, locate waiting and rework, then propose measurements that could verify the suspected causes.
3. Practise denominators and units
Work takt, standard deviation, Z, DPU, DPMO, yield, p-chart and capability calculations by hand. State what each result means before checking arithmetic, and identify assumptions that would invalidate the interpretation.
4. Contrast neighboring tools
Compare histogram versus control chart, Pareto versus fishbone, checklist versus error-proofing, paired versus independent t-tests, and Cp versus Cpk. Write the single discriminating cue that changes your choice.
5. Explain every distractor
Complete the 20 learning questions before reading explanations. For each option, state why it fits or fails the scenario. Revisit the corresponding concept whenever the error concerns design, denominator, assumption or interpretation.
6. Close scope gaps and practise active recall
Use the flashcards without notes, then compare your coverage with the complete linked curriculum. Finish by explaining a control response plan and a regression result aloud, including what the evidence supports and what remains uncertain.
Use your sample score as a starting point
The free set contains 20 original practice questions. It is a small learning sample, not an official examination or a validated pass prediction. Review each explanation and its source before moving to another topic.
Start the free sampleOfficial sources and updates
- SSGI: Lean Six Sigma Healthcare Green Belt Certification
- SSGI: Lean Six Sigma Green Belt Healthcare — Full Course
- SSGI: What Is DMAIC?
- SSGI: Lean Six Sigma — The Definitive Guide
- NIST: Gauge R & R Studies
- Lean Enterprise Institute: Takt Time
- Lean Enterprise Institute: Kanban
- Institute for Healthcare Improvement: FMEA Tool
- Institute for Healthcare Improvement: Cause and Effect Diagram
- Institute for Healthcare Improvement: Pareto Chart
- NIST: Measures of Scale
- NIST: Normal Distribution
- NIST: Shewhart X-bar and R and S Control Charts
- NIST: Proportions Control Charts
- NIST: What Is Process Capability?
- NIST: What Are Control Charts?
- NIST: What Are Statistical Tests?
- NIST: Confidence Limits for the Mean
- NIST: Two-Sample t-Test for Equal Means
- NIST: One-Factor ANOVA
- NIST: Linear Least Squares Regression
- Lean Enterprise Institute: 5S
- Lean Enterprise Institute: Value Stream Mapping
- Lean Enterprise Institute: Error-Proofing
- NIST: Histogram
- ASQ: Cost of Quality
- NIST: Quantitative Techniques
- 2.4.4.1. Analysis of repeatability
- 2.4.4.2. Analysis of reproducibility
- 2.4.5. Analysis of bias
- QIToolkit_FailureModesandEffectsAnalysis.pdf
- 1.3.6.7.2. Critical Values of the Student's-t Distribution
- 4.4.4. How can I tell if a model fits my data?
Confirm current eligibility, fees, scheduling and policies directly with the examination provider. CertMedbry is independent and is not endorsed by these organisations. How we handle content and corrections.