Overview
In the Improve phase the team moves from knowing what causes the problem to changing the process. Design of experiments (DOE) is the statistical core: choosing factors and levels, running a full or fractional factorial with randomization and replication, and reading main effects and interaction plots to find settings that move the response. Implementation planning covers selecting among solutions, piloting, cost-benefit and risk analysis, and managing the change with stakeholders, often through a plan-do-check-act (PDCA) cycle.
Lean tools complete the section: 5S, standard work, takt time and line balancing, setup reduction (SMED), mistake-proofing (poka-yoke), pull and kanban, and kaizen events. Candidates must calculate simple DOE effects and takt time, interpret interaction plots, and choose the improvement approach that fits the verified cause from Analyze rather than a favorite tool.
How it's tested
Improve questions include short DOE calculations and plot interpretation, lean calculations such as takt time, and judgment items about how to implement a solution. Wording such as BEST next step, MOST effective tool or FIRST action is common, and the right answer usually tests or pilots before full rollout and matches the tool to the stated waste or cause.
-
Compute or read a DOE effect
A two-level factorial table or a main effects or interaction plot is given and you calculate a main effect, count runs for 2^k designs, or decide whether an interaction exists.
-
Match the lean tool to the waste
A described symptom, such as long changeovers, wrong parts installed, or excess work-in-process, asks which lean tool addresses it most directly.
-
Plan the rollout
A solution has been chosen and you decide the next step: pilot, risk analysis, cost-benefit, stakeholder communication, or full implementation with controls.
A two-factor experiment shows that raising temperature increases yield at low pressure but decreases it at high pressure. A candidate reports that temperature has no effect because the averages cancel out. Reading the interaction plot, the lines cross, which signals a strong interaction. The right conclusion is that the best temperature depends on pressure, so the factors must be set together rather than one at a time.
Topics 3
Official topics from the ASQ outline, in outline order.
5.1 Design of experiments (DOE)DOE terms, factorial designs, randomization, replication, and main effects and interaction plots.
- What it means
- DOE changes several input factors at planned levels at the same time to see their effects on a response. Key terms include factor, level, treatment, response, run, replication, randomization, blocking, main effect, interaction, and confounding (aliasing). A full factorial with k factors at two levels needs 2^k runs; a fractional factorial uses fewer runs at the cost of confounding some effects.
- What to understand
- A main effect is the average response at the high level minus the average at the low level. An interaction exists when the effect of one factor depends on the level of another, shown by non-parallel lines on an interaction plot. Randomization protects against lurking time-related variables; replication estimates experimental error; blocking removes known nuisance variation.
- How it shows up
- Questions give a small 2^2 or 2^3 table and ask for a main effect or number of runs, or show plots and ask which factor matters most or whether an interaction exists.
- Common confusion
- Candidates believe one-factor-at-a-time testing is equivalent; it needs more runs and cannot detect interactions, which is why DOE is preferred.
5.2 Implementation planningSelecting solutions, piloting, cost-benefit and risk analysis, PDCA and managing the change.
- What it means
- Implementation planning takes a proven improvement into routine operation. It includes evaluating candidate solutions against criteria, piloting on a small scale, analyzing costs, benefits and risks, planning communication and training, and using the plan-do-check-act cycle to test, learn and adjust before full deployment.
- What to understand
- A pilot confirms the solution works under real conditions and exposes side effects before large investment. Cost-benefit analysis compares one-time and ongoing costs with expected savings, often as payback period or return on investment. A risk review, such as updating the FMEA for the new process, should precede rollout. Resistance is managed by involving the people who do the work.
- How it shows up
- A scenario describes a solution that worked in analysis and asks what to do next, where the expected answer is often a pilot with measurement, or asks how to handle resistance from operators.
- Common confusion
- Candidates jump from a significant DOE result to full-scale rollout; the exam expects confirmation runs or a pilot first.
5.3 Lean tools5S, standard work, takt time, SMED, poka-yoke, kanban and pull, and kaizen.
- What it means
- Lean tools remove specific wastes. 5S (sort, set in order, shine, standardize, sustain) organizes the workplace; standard work fixes the best current method; takt time paces work to customer demand; SMED (single-minute exchange of die) cuts changeover time; poka-yoke prevents or detects human error; kanban signals pull production; kaizen events make rapid focused improvements.
- What to understand
- Takt time = available working time / customer demand in the same period; 450 minutes for 900 units gives 0.5 minutes, or 30 seconds, per unit. If a step's cycle time exceeds takt, it is a bottleneck. SMED converts internal setup (machine stopped) to external setup (machine running) and streamlines the rest.
- How it shows up
- A symptom is described and you pick the tool, or you calculate takt time and decide whether a workstation can keep up with demand.
- Common confusion
- Takt time is set by customer demand, while cycle time is how long the process actually takes; candidates use them interchangeably.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Main effect
The average response when a factor is at its high level minus the average response when it is at its low level, across all levels of the other factors.
- Remember
- In a two-level design the regression coefficient is half the effect, so software output and hand calculations can differ by a factor of two.
- Tested as
- Two-by-two factorial tables ask you to compute an effect and pick the most influential factor.
- Trap
- Reporting the coefficient (half the effect) or the difference between two single runs rather than the averages.
-
Interaction
An interaction occurs when the effect of one factor changes depending on the level of another, shown by non-parallel or crossing lines on an interaction plot.
- Remember
- Parallel lines mean no interaction; crossing lines mean a strong one, and main effects then can mislead.
- Tested as
- Plot-interpretation questions and the reason DOE beats one-factor-at-a-time testing.
- Trap
- Concluding a factor does not matter because its main effect is near zero when it has a strong interaction.
-
Full factorial runs (2^k)
A two-level full factorial with k factors has 2^k treatment combinations: 4 for two factors, 8 for three, 16 for four, multiplied by the number of replicates.
- Remember
- Fractional factorials such as 2^(k-1) halve the runs but alias some effects with others.
- Tested as
- Counting runs and recognizing when a fractional design is used.
- Trap
- Forgetting to multiply by replicates when asked for total runs.
-
Randomization, replication and blocking
Randomization runs treatments in random order to spread unknown time-related effects; replication repeats treatments to estimate experimental error; blocking groups runs to remove a known nuisance source such as different shifts or batches.
- Remember
- Randomize what you cannot control, block what you can identify but not remove.
- Tested as
- Experiment-planning questions ask which principle addresses a described risk.
- Trap
- Choosing replication when the risk described is a drift over time, which randomization addresses.
-
Takt time
Takt time = available production time / customer demand for the same period, the pace at which units must be completed to meet demand.
- Remember
- Available time excludes planned breaks; takt changes when demand changes, cycle time does not.
- Tested as
- Short calculations and line-balancing judgments.
- Trap
- Dividing demand by time, giving units per minute instead of minutes per unit.
-
SMED (single-minute exchange of die)
A setup reduction method that separates internal setup (done only with the machine stopped) from external setup (done while running), converts internal to external where possible, and streamlines both.
- Remember
- Shorter changeovers allow smaller batches, which reduces inventory and lead time.
- Tested as
- Scenarios with long changeovers and large batch sizes.
- Trap
- Choosing 5S or kanban for a problem that is clearly long setup time.
-
Poka-yoke (mistake-proofing)
Devices or methods that prevent an error from happening or make it immediately obvious, such as asymmetric connectors or sensors that stop the line when a part is missing.
- Remember
- Prevention-type poka-yoke is stronger than detection-type, and both beat inspection or retraining for inadvertent errors.
- Tested as
- Choosing the most effective fix for occasional human slips.
- Trap
- Choosing retraining or warning signs for errors that trained, careful people still make.
-
5S
Sort, set in order, shine, standardize and sustain: a workplace organization method that makes abnormalities visible and reduces motion and searching.
- Remember
- Sustain is the hardest step and depends on audits and leadership attention.
- Tested as
- Symptoms of searching for tools, clutter or unclear storage.
- Trap
- Treating 5S as a one-time cleanup rather than a standard that must be kept.
-
Kanban and pull
Kanban is a signal, often a card or empty bin, that authorizes production or replenishment only when downstream consumption occurs.
- Remember
- The number of kanbans caps work-in-process; reducing kanbans exposes problems.
- Tested as
- Overproduction and excess work-in-process scenarios.
- Trap
- Using kanban to fix quality defects, which it does not address.
-
PDCA (plan-do-check-act)
An iterative cycle for testing a change: plan it, do it on a small scale, check results against the prediction, then act by standardizing or adjusting.
- Remember
- The do step is a small-scale trial, which is why PDCA naturally pairs with piloting.
- Tested as
- Implementation sequencing and continuous-improvement questions.
- Trap
- Treating do as full-scale implementation.
-
Cost-benefit and payback
Compares implementation and ongoing costs with expected savings; payback period = initial investment / annual net savings.
- Remember
- Include hidden costs such as training, downtime and validation, and confirm savings with finance where possible.
- Tested as
- Solution-selection questions and simple payback calculations.
- Trap
- Choosing the solution with the largest gross savings without netting costs.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
DOE vs. one-factor-at-a-time (OFAT) testing
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Design of experiments | Changes several factors together according to a planned design | Estimates main effects and interactions efficiently | Skipping randomization because the run order is convenient |
| One-factor-at-a-time | Changes one factor while holding others fixed | Cannot detect interactions and needs more runs for the same information | Believing it is safer because it is simpler |
Takt time vs. cycle time vs. lead time
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Takt time | Available time divided by customer demand | Set by demand; the required pace | Trying to reduce takt by working faster |
| Cycle time | Time the process or step actually takes to complete one unit | Set by the process; must be at or below takt | Calling it takt time |
| Lead time | Total elapsed time from order or start to delivery, including waiting | Includes queues and inventory, usually much longer than cycle time | Improving cycle time and expecting lead time to drop by the same amount |
SMED vs. poka-yoke vs. kanban vs. 5S
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| SMED | Reduces changeover time | Targets long setups and large batches | Applying it to error problems |
| Poka-yoke | Prevents or flags human errors | Targets inadvertent mistakes | Relying on training instead |
| Kanban | Signals pull replenishment | Targets overproduction and work-in-process | Expecting it to reduce defects |
| 5S | Organizes and standardizes the workplace | Targets searching, motion and hidden abnormalities | Doing it once without a sustain plan |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
No main effect, so no influence
- What candidates pick
- Drop a factor whose main effect is near zero.
- Why it looks right
- A small main effect looks unimportant.
- Why it's wrong
- If the factor interacts strongly with another, its effect reverses across levels and averages to near zero; dropping it loses control of the response.
Clue to watch for Crossing lines on the interaction plot or a large interaction term in the table.
-
One-factor-at-a-time is safer
- What candidates pick
- Test each setting separately while holding the others constant.
- Why it looks right
- It seems simpler and easier to explain.
- Why it's wrong
- OFAT needs more runs and cannot detect interactions, so it can miss the true best settings; a factorial design is more efficient.
Clue to watch for Two or more factors that may influence each other, and limited runs available.
-
Full rollout after a good result
- What candidates pick
- Implement the new settings on every line immediately after the experiment.
- Why it looks right
- The data were significant and the sponsor wants results.
- Why it's wrong
- Experiments run under controlled conditions; a pilot or confirmation run checks the solution in normal operation and reveals side effects before large investment.
Clue to watch for The question asks for the NEXT step after identifying optimal settings.
-
Retrain for slips
- What candidates pick
- Retrain operators who occasionally install the part backwards.
- Why it looks right
- Training feels like the direct fix for a human error.
- Why it's wrong
- Inadvertent errors happen to trained people; a poka-yoke such as an asymmetric fixture makes the error impossible or obvious.
Clue to watch for Experienced, trained staff making occasional, unintentional mistakes.
-
Faster work to meet takt
- What candidates pick
- Ask operators to work faster when a step's cycle time exceeds takt time.
- Why it looks right
- It appears to close the gap immediately.
- Why it's wrong
- Takt is fixed by demand; the process must be redesigned through line balancing, waste removal or added capacity so cycle time falls below takt sustainably.
Clue to watch for A workstation's cycle time is longer than the calculated takt time.
-
Kanban to fix quality
- What candidates pick
- Introduce kanban cards to reduce defects at a workstation.
- Why it looks right
- Kanban is a well-known lean tool.
- Why it's wrong
- Kanban controls production quantity and work-in-process; defect problems call for root cause work, poka-yoke or standard work.
Clue to watch for The symptom is defects or errors, not overproduction or excess inventory.
Answer framework
For Improve phase questions about experiments, lean tools or rollout, use these steps.
- Confirm that the change targets a root cause verified in Analyze, and reject options that address an unverified or different cause.
- For experiments, identify factors and levels, count 2^k runs, then calculate main effects as high-level average minus low-level average.
- Check the interaction plot before trusting main effects: crossing or non-parallel lines mean factors must be set together.
- For lean questions, name the waste or symptom first, then match the tool, such as SMED, poka-yoke, kanban, 5S or standard work.
- For implementation, choose the path that pilots, measures, reviews risks and costs, involves the people affected, and prepares controls before full rollout.
Practice question
Want more like this? The CSSGB practice test has questions for every domain.
Checklist
0 of 12
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Begin with DOE vocabulary and the 2^k structure, then practice computing main effects and reading interaction plots from small tables. Ten worked two-factor examples, including some with crossing lines, will cover most DOE questions at Green Belt depth. Keep the distinction between an effect and a coefficient clear, because it is the most common calculation slip.
Next study the lean tools as a symptom-to-tool map: long changeovers to SMED, inadvertent errors to poka-yoke, overproduction to kanban and pull, searching and clutter to 5S, inconsistent methods to standard work. Add takt time calculations and line balancing against takt. Then cover implementation planning, where the key habit is choosing pilots, risk reviews and stakeholder involvement before full rollout.
Candidates commonly underestimate implementation questions because they read like management common sense; the exam wants the disciplined, data-confirmed sequence. After this section, practice mixed Improve questions under time, then move to Control, where the settings and lean changes you made here must be locked in with SPC, control plans and lean control tools.
FAQ
How many Improve questions are on the CSSGB?
ASQ lists 15 of the 100 scored questions for this section.
How deep does DOE go at Green Belt level?
Expect terms, two-level full and fractional factorial structure, run counts, main effect calculations and interpretation of main effects and interaction plots, rather than detailed ANOVA analysis of complex designs.
What is the difference between a main effect and an interaction?
A main effect is the average change in the response when one factor moves from low to high. An interaction means that change depends on the level of another factor, shown by non-parallel lines on an interaction plot.
How do I calculate takt time?
Divide available working time by customer demand for the same period. For example, 450 available minutes and demand of 900 units give a takt time of 0.5 minutes, or 30 seconds, per unit.
Why pilot a solution that already worked in an experiment?
A pilot tests the solution under normal operating conditions, with real staff and variation, revealing side effects and implementation issues before large investment and full rollout.
How long should I spend on Improve?
Plan roughly 15–20 hours within a 120-hour plan, divided between DOE practice and the lean tool and implementation topics.
How does Improve connect to Control?
Every setting, procedure or lean change made in Improve becomes something the Control phase must document, monitor with SPC and protect with a control plan so the gain is sustained.
Other Quality Management certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 5 knowledge
You have covered DOE effects and interactions, piloting with PDCA, takt time and matching lean tools to wastes. Check whether you can apply them on exam-style scenarios.