Overview
Improve turns verified causes into a proven solution. Design of experiments (DOE) covers terminology (factors, levels, responses, treatments, replication, randomization, blocking, interaction, confounding), planning and power, one-factor experiments, two-level full factorial and fractional factorial designs, design resolution, and calculating and interpreting main and interaction effects. Lean methods covers waste elimination tools such as pull systems, kanban, 5S, standard work and poka-yoke, and cycle-time reduction tools such as continuous flow, single-minute exchange of die (SMED) and kaizen. Implementation covers selecting among solutions, piloting, simulation, and planning a controlled rollout with risk assessment.
Candidates must choose an appropriate design, compute effects, judge what aliasing hides, match a lean tool to a waste, and decide how to test a solution before full deployment. Improve depends on Domain 6's verified causes and hands a working solution to Domain 8's control plan.
How it's tested
Expect DOE calculations (number of runs, main and interaction effects, degrees of freedom), design-choice items (full versus fractional, resolution, blocking versus randomization), lean tool selection for a described waste, and implementation items asking what to do before full rollout. Effects questions reward careful use of the coded design table.
-
Compute the effect
A small two-level design with responses is given and you calculate a main effect or interaction, or the number of runs a design needs.
-
Choose the design
Constraints such as number of factors, budget and suspected interactions are described and you choose full factorial, a fractional design of a given resolution, or blocking.
-
Match the lean tool
A waste or flow problem is described, such as long changeovers or excess inventory, and you choose SMED, kanban, 5S, poka-yoke or continuous flow.
A team has seven candidate factors and budget for about 16 runs; it suspects two-factor interactions matter. A full factorial needs 2^7 = 128 runs. A 2^(7−3) resolution IV design uses 16 runs and keeps main effects clear of two-factor interactions, though those interactions are aliased with each other. That fits the goal of screening main effects reliably, so it is the best choice.
Subsections 3
Official subsections from the ASQ outline, in outline order.
7.1 Design of experiments (DOE)DOE terminology, planning, one-factor, full factorial and fractional factorial designs.
- What it means
- DOE deliberately changes input factors at set levels and measures the response to establish cause and effect, estimate main effects and interactions, and find optimal settings. Key principles are randomization (protects against lurking variables), replication (estimates pure error and raises power) and blocking (removes known nuisance variation).
- What to understand
- A two-level full factorial with k factors needs 2^k runs per replicate; a fractional factorial 2^(k−p) uses fewer runs by aliasing some effects. Resolution III aliases main effects with two-factor interactions, resolution IV keeps main effects clear of two-factor interactions but aliases those interactions with each other, and resolution V keeps main effects and two-factor interactions clear of each other. A main effect is the average response at the high level minus the average at the low level.
- How it shows up
- A design table or design constraint is given and you compute an effect, the number of runs, or choose the design and resolution that meets the goal.
- Common confusion
- Treating a significant interaction as unimportant and setting factors by main effects alone; when an interaction is significant, the best level of one factor depends on the other.
7.2 Lean methodsWaste elimination and cycle-time reduction tools, and kaizen.
- What it means
- This covers tools that remove waste (pull systems and kanban, 5S, standard work, poka-yoke or mistake-proofing, visual management) and tools that shorten cycle time (continuous or one-piece flow, cellular layout, SMED for changeovers, line balancing to takt), plus kaizen and kaizen events for rapid improvement.
- What to understand
- Each tool targets a specific problem: SMED converts internal setup (machine stopped) to external setup (done while running) to cut changeover time; kanban caps work in process and signals replenishment; 5S (sort, set in order, shine, standardize, sustain) organizes the workplace; poka-yoke prevents errors or makes them immediately detectable.
- How it shows up
- A process symptom is described and you choose the lean tool that addresses it, or you calculate a simple lean metric such as kanban quantity or takt-balanced staffing.
- Common confusion
- Choosing 5S for every problem, when changeover time needs SMED and overproduction needs a pull system.
7.3 ImplementationSelecting solutions, piloting, simulation and planning a controlled rollout.
- What it means
- This covers generating and selecting solutions with criteria-based tools (prioritization or Pugh matrices, cost-benefit), assessing solution risk (for example with FMEA), confirming results with a pilot or simulation, and planning full implementation with resources, communication and training.
- What to understand
- A pilot tests the solution on a limited scale under real conditions to confirm improvement and expose unintended effects before full rollout. Simulation, such as discrete-event modeling, tests options when a physical pilot is costly or risky. Results from the pilot should be verified statistically against the baseline.
- How it shows up
- A promising solution is described and you choose the next step, usually a pilot with measurement, or decide how to select among competing solutions.
- Common confusion
- Rolling out a solution organization-wide because the experiment was significant, skipping a pilot under real operating conditions.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Full factorial run count
A two-level full factorial with k factors needs 2^k runs per replicate; with r replicates, total runs are r × 2^k.
- Remember
- 2^3 = 8, 2^4 = 16, 2^5 = 32 runs; run count doubles with each added factor.
- Tested as
- Run-count calculations and design-choice items.
- Trap
- Using k² instead of 2^k.
-
Fractional factorial 2^(k−p)
A 2^(k−p) design studies k factors in 2^(k−p) runs by using p generators, such as D = ABC, which alias some effects with others.
- Remember
- A 2^(4−1) with D = ABC has defining relation I = ABCD and is resolution IV.
- Tested as
- Screening-design questions with many factors and limited budget.
- Trap
- Believing fractional designs estimate every interaction independently.
-
Design resolution
Resolution III: main effects aliased with two-factor interactions. Resolution IV: main effects clear of two-factor interactions, which are aliased with each other. Resolution V: main effects and two-factor interactions clear of each other.
- Remember
- Resolution equals the length of the shortest word in the defining relation.
- Tested as
- Asked whenever a fractional design is proposed.
- Trap
- Choosing resolution III when two-factor interactions are suspected to matter.
-
Main effect
Main effect = average response at the factor's high level − average response at its low level.
- Remember
- Compute from the coded column: sum of responses with + minus sum with −, divided by half the number of runs.
- Tested as
- Core DOE calculation item.
- Trap
- Forgetting to divide by the number of runs at each level.
-
Interaction effect
An interaction exists when the effect of one factor depends on the level of another; the AB effect uses the product of the A and B coded columns.
- Remember
- Non-parallel lines on an interaction plot indicate interaction.
- Tested as
- Interaction plots and calculations appear in DOE items.
- Trap
- Setting factors by main effects alone when a strong interaction reverses the best setting.
-
Randomization, replication, blocking
Randomization spreads unknown nuisance effects across runs; replication repeats whole runs to estimate pure error; blocking groups runs by a known nuisance source such as raw-material lot or day.
- Remember
- Block what you can control, randomize what you cannot.
- Tested as
- Planning questions on how to handle a known nuisance variable.
- Trap
- Treating repeated measurements of the same run as replication.
-
Confounding and aliasing
Effects are confounded or aliased when the design cannot separate them; their estimates are combined.
- Remember
- Blocking also confounds the block effect with a chosen high-order interaction.
- Tested as
- Interpretation of fractional design results.
- Trap
- Attributing a significant aliased estimate to the main effect without considering its alias partners.
-
SMED
Single-minute exchange of die reduces changeover time by separating internal setup (done while the machine is stopped) from external setup (done while running), converting internal to external, and streamlining both.
- Remember
- Shorter changeovers enable smaller batches and more frequent runs of each product.
- Tested as
- Chosen when long setups force large batches and inventory.
- Trap
- Choosing 5S or TPM when the stated problem is changeover duration.
-
Kanban and pull
A pull system produces only when downstream consumption signals a need, often with kanban cards; number of kanbans N = (demand × lead time × (1 + safety factor)) / container quantity.
- Remember
- Kanban limits work in process; reducing kanbans exposes problems.
- Tested as
- Asked when overproduction or excess inventory is the symptom, sometimes with a calculation.
- Trap
- Using kanban in a process with highly unstable demand and no level scheduling.
-
Poka-yoke
Mistake-proofing devices or methods that prevent an error from occurring or make it immediately obvious, such as asymmetric fixtures or sensors that stop the line.
- Remember
- Prevention (control) poka-yoke is stronger than detection (warning) poka-yoke.
- Tested as
- Solution-selection questions for human error.
- Trap
- Choosing retraining or added inspection when a mistake-proof design is possible.
-
5S
Sort, set in order, shine, standardize, sustain: a workplace organization method that makes abnormal conditions visible.
- Remember
- Sustain is the hardest step and needs audits and leadership attention.
- Tested as
- Asked for disorganized workplaces, searching for tools, or as a foundation for other lean tools.
- Trap
- Treating 5S as a one-time cleanup.
-
Overall equipment effectiveness (OEE)
OEE = availability × performance × quality; for example 0.90 × 0.95 × 0.98 ≈ 0.84.
- Remember
- It shows which loss, downtime, speed or defects, dominates equipment performance.
- Tested as
- Lean equipment questions and calculations.
- Trap
- Adding the three factors instead of multiplying them.
-
Pilot
A limited-scale trial of the solution under real conditions with before-and-after measurement.
- Remember
- A pilot confirms the improvement and reveals side effects before full rollout.
- Tested as
- The usual NEXT step after a solution is selected or a DOE finds optimal settings.
- Trap
- Skipping the pilot because the experiment already showed significance.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Resolution III vs. IV vs. V
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Resolution III | Main effects aliased with two-factor interactions | Fewest runs; screening only when interactions are negligible | Used when interactions are likely |
| Resolution IV | Main effects clear of two-factor interactions; those interactions aliased with each other | Reliable main effects, ambiguous interactions | Assuming two-factor interactions are individually estimable |
| Resolution V | Main effects and two-factor interactions clear of each other | More runs; supports modeling interactions | Paying for it when only screening is needed |
Randomization vs. replication vs. blocking
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Randomization | Random run order | Protects against unknown, time-related nuisance | Skipped for convenience when factors are hard to change |
| Replication | Repeating complete runs | Estimates pure error and increases power | Confused with repeated measurements of one run |
| Blocking | Grouping runs by a known nuisance factor | Removes known nuisance variation from error | Treating the block as a factor of interest |
SMED vs. kanban vs. 5S vs. poka-yoke
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| SMED | Reduce changeover time | Targets setup, enabling small batches | Applied to general disorganization |
| Kanban | Signal-based pull replenishment | Targets overproduction and inventory | Used where demand is erratic and unleveled |
| 5S | Workplace organization | Targets searching, motion and hidden abnormalities | Expected to fix long changeovers on its own |
| Poka-yoke | Mistake-proofing | Targets human error at the source | Replaced by inspection or retraining |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Unaveraged effect
- What candidates pick
- The difference between the sums of high and low responses as the main effect.
- Why it looks right
- Subtracting sums is quick and uses the coded column directly.
- Why it's wrong
- The effect is the difference between averages; the sum difference must be divided by the number of runs at each level.
Clue to watch for A coded design table with responses for every run.
-
Cheapest fraction
- What candidates pick
- A resolution III design to save runs when the team believes temperature and pressure interact.
- Why it looks right
- Fewer runs save time and money.
- Why it's wrong
- Resolution III aliases main effects with two-factor interactions, so an important interaction would distort main-effect estimates.
Clue to watch for The scenario says interactions are suspected or important.
-
Ignoring the interaction
- What candidates pick
- Set each factor at the level with the better main effect.
- Why it looks right
- Main effects are simple to read.
- Why it's wrong
- A significant interaction means the best level of one factor depends on the other; use the interaction plot or cell means.
Clue to watch for An interaction term is significant or the interaction plot lines cross.
-
Repeats as replicates
- What candidates pick
- Measure each run's output three times and treat it as three replicates.
- Why it looks right
- It triples data cheaply.
- Why it's wrong
- Repeated measurements capture measurement variation only; replication requires re-running the full treatment setup.
Clue to watch for The extra data come from re-measuring the same experimental unit.
-
Inspection over mistake-proofing
- What candidates pick
- Add an inspection step to catch parts assembled backward.
- Why it looks right
- Inspection is quick to add and catches the error.
- Why it's wrong
- Inspection detects after the fact and is itself error-prone; poka-yoke such as an asymmetric fixture prevents the error.
Clue to watch for The defect comes from a repeatable human error in orientation or omission.
-
Straight to full rollout
- What candidates pick
- Implement optimized DOE settings across all plants immediately.
- Why it looks right
- The experiment produced statistically significant results.
- Why it's wrong
- Experimental conditions differ from routine operation; a pilot confirms results and reveals side effects before scaling.
Clue to watch for The question asks what to do NEXT after finding optimal settings.
Answer framework
Use these steps for Improve-phase design, calculation and implementation questions.
- Confirm which verified causes the solution must address, and whether interactions among them are likely.
- Choose the design by factor count, budget and needed resolution, and plan randomization, replication and blocking before running anything.
- Compute effects from averages in the coded table, check interactions before main effects, and confirm significance with ANOVA or normal probability plots.
- Match lean tools to the specific waste, SMED for changeovers, kanban for overproduction, poka-yoke for human error and 5S for disorganization.
- Select the solution with weighted criteria, assess its risk, pilot it under real conditions, and verify results against the baseline before full rollout.
Practice question
Want more like this? The CSSBB practice test has questions for every domain.
Checklist
0 of 14
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Start with DOE vocabulary and principles, because many questions test randomization, replication, blocking and confounding without any calculation. Then work 2^2 and 2^3 designs by hand: build the coded table, compute main effects and interactions from averages, and sketch interaction plots. Repeat until the averaging step is automatic.
Next, learn fractional factorials. Practice writing the defining relation from a generator, listing aliases, and stating the resolution. Pair each resolution with the question it answers: screening, reliable main effects, or interaction modeling.
Then cover lean tools with a symptom-to-tool table: long changeovers to SMED, overproduction to kanban and pull, searching and clutter to 5S, repeated human errors to poka-yoke, imbalance to line balancing against takt. Add OEE and kanban quantity calculations.
The commonly underestimated area is implementation. Questions about pilots, solution selection and risk look like common sense but reward the cautious, data-confirmed answer.
Tab a DOE page in your open-book reference with run counts, resolution definitions and the effect formula. Afterwards, practice mixed Improve sets, then move to Control, which locks in the settings and lean standards established here.
FAQ
How many questions come from Improve?
ASQ assigns 21 of the 150 scored questions to Improve, about 14% of the exam.
How hard is the DOE math?
Moderate. Most items ask for run counts, main or interaction effects from a small two-level design, or the meaning of resolution. Practice the coded-table method until it is fast.
When should I use a fractional factorial?
When there are many factors and limited runs, usually for screening. Choose the resolution based on whether two-factor interactions are likely to matter.
What is the difference between blocking and randomization?
Blocking removes variation from a known nuisance factor by grouping runs; randomization protects against unknown nuisance factors by randomizing run order.
Which lean tool reduces changeover time?
SMED, which separates internal from external setup, converts internal steps to external, and streamlines what remains.
Why pilot a solution that already worked in an experiment?
Experiments run under controlled conditions. A pilot tests the solution in normal operation, confirms the gain against the baseline, and exposes side effects before full rollout.
How does Improve connect to Control?
Improve establishes new settings, standard work and mistake-proofing; Control documents them in a control plan, monitors them with SPC, and hands them to the process owner.
Other Quality Management certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 7 knowledge
You have covered design of experiments, lean waste and cycle-time tools, and controlled implementation. Test whether you can compute effects and choose designs at exam pace.