Design of Experiments (DOE) is one of the most powerful tools in the Lean Six Sigma toolkit, and JMP Design of Experiments functionality makes it accessible without drowning you in manual calculations. JMP, developed by SAS, gives engineers, analysts, and Black Belts a visual, interactive platform to plan experiments, identify critical factors, and optimize processes with statistical rigor.
But knowing the software exists and actually using it well are two different things. If you’ve opened JMP’s DOE menu and felt unsure about which design to choose, how to set up your factors, or what to do once the data comes back, you’re not alone. These are exactly the kinds of gaps we address at Lean Six Sigma Experts through our training and consulting work, where we help teams move from theory to hands-on implementation of tools like DOE.
This guide walks you through building and analyzing a designed experiment in JMP, step by step. You’ll learn how to select the right experimental design, configure your factors and responses, collect data efficiently, and interpret results so you can make real process improvements, not just generate charts.
What you need before you start in JMP
Before you open JMP’s DOE interface, a few prerequisites will make the difference between an experiment that delivers clean answers and one that generates noise. Getting these in order upfront saves you from scrapped runs, wasted budget, and inconclusive results.
The right JMP version and license
JMP Pro and the standard JMP license both include full DOE functionality, including the Custom Designer you’ll use throughout this guide. If you’re working inside a company, confirm with your IT team that your version is JMP 16 or later, since newer releases include improvements to Model Screening and Augmented designs. You can verify your version by going to Help > About JMP.
If your organization runs an older version, some menu paths in this guide may differ slightly, so check the JMP documentation for your specific release before you begin.
Process knowledge you need before day one
JMP handles the statistics, but you supply the process knowledge. Before building any design, you need at minimum a preliminary list of factors (inputs) you can control, a measurable response variable, and a realistic operating range for each factor. Without these, the software has nothing meaningful to work with.
For example, if you’re optimizing a chemical mixing process, you need to know things like temperature range (say, 150°F to 200°F), mixing speed range, and what you’re measuring as output (viscosity, yield, or defect rate). Write these down before opening JMP.
A quick pre-experiment checklist
Run through this list before starting any JMP Design of Experiments session:
| Item | Why it matters |
|---|---|
| Defined response variable | Tells JMP what you’re optimizing |
| Factor list with ranges | Required to build any design |
| Measurement system confirmed | Garbage data ruins clean designs |
| Run budget approved | Determines design size and complexity |
| Team alignment on protocol | Keeps execution consistent |
Step 1. Define your goal, response, and factors
This step happens before you touch JMP at all. Skipping this stage is the single most common reason DOE projects fail, and no software can compensate for unclear objectives. Write down your goal in one sentence, then identify your response variable and your candidate factors before opening any menu.
Set one clear response variable
Your response variable is what you measure at the end of each experimental run. It must be numeric and continuous if possible, since JMP’s analysis tools work best with continuous data. For example, "reduce coating thickness variation from 12 microns to under 5 microns" is a clear, measurable response. Avoid vague targets like "improve quality" because JMP cannot optimize something you cannot measure.
Choose and categorize your factors
Factors fall into two categories that JMP Design of Experiments handles differently: continuous factors (like temperature or pressure) and categorical factors (like machine type or operator). For each factor, confirm you can actually control it during the experiment and define a realistic low and high setting.
| Factor type | Example | JMP setting |
|---|---|---|
| Continuous | Temperature: 150 to 200°F | Continuous |
| Categorical | Machine A or Machine B | Nominal |
Keep your initial factor list to five or fewer; too many factors inflate run counts and make your results harder to interpret.
Step 2. Build an efficient design in Custom Designer
With your factors and response defined, open JMP and navigate to DOE > Custom Design. The Custom Designer is the most flexible tool in the JMP Design of Experiments suite because it builds a statistically efficient design around your exact run budget rather than forcing you into a fixed template. Enter your response variable first, then add each factor with its type and range.
Configure factors and run count
In the Custom Designer panel, you’ll see separate sections for Responses and Factors. Add your response, set your optimization goal (minimize, maximize, or match target), then add each factor with the type and operating range you established in Step 1. For continuous factors, type in your low and high values. For categorical factors, enter each level name.
Once your factors are loaded, set your run count in the Number of Runs field. JMP suggests a minimum, but add at least 4 to 6 extra runs to give the model room for error estimation.
A run count that matches the bare minimum leaves no degrees of freedom to estimate error, which weakens your ability to confirm which effects are truly significant.
After clicking Make Design, JMP generates a randomized run table you can export directly to a spreadsheet or print for the team executing the experiment.
Step 3. Run the DOE with randomization and discipline
JMP generates a randomized run order for a reason: randomization protects your results from systematic bias caused by time-related factors like tool wear, temperature drift, or operator fatigue. Run every experimental combination in the exact order JMP specifies, even when it feels inefficient to jump between settings. Discipline during execution is what separates a clean, actionable dataset from one you can’t trust.
Follow the randomized run order
Resist the urge to reorder runs for convenience. If you group all low-temperature runs together to save setup time, you introduce a confounding pattern that makes it impossible to separate the effect of temperature from the effect of time. Trust the randomized sequence JMP Design of Experiments produces and document any deviations immediately if they occur.
If you must deviate from the run order, record the reason and the actual sequence in your data table so your analysis accounts for it.
Log actual conditions during each run
Record actual factor settings for every run, not just the planned target values. Machines drift, operators adjust, and real conditions rarely match planned conditions exactly. Log these values directly into the JMP data table alongside your response measurements so the analysis reflects what actually happened, not what you intended to happen.
Step 4. Analyze effects, interactions, and optimize settings
Once your data is collected, go to Analyze > Fit Model in JMP and load your response and effects. The jmp design of experiments workflow brings you here automatically if you built your design in Custom Designer, since the data table already stores the analysis script. Run the model, then focus on the Effect Summary report first.
Read the Effect Summary and Scaled Estimates
The Effect Summary table ranks each factor and interaction by p-value. Any effect with a p-value below 0.05 is statistically significant and deserves your attention. Look at the Scaled Estimates plot next to confirm direction and magnitude: a large positive bar means increasing that factor raises your response, and a large negative bar means the opposite.
If two factors show a significant interaction, you cannot interpret them independently. Optimize them together.
Set Optimal Factor Levels in the Prediction Profiler
Use the Prediction Profiler to turn your analysis into action. Click Desirability at the top of the profiler, then select Maximize Desirability to let JMP calculate the factor settings that hit your response goal. Verify that the predicted response aligns with your practical process constraints before you lock in those settings for a confirmation run. If the profiler suggests a setting that is physically impossible to maintain, adjust the factor range and re-run the desirability optimization before you commit.
Wrap it up and decide what to run next
At this point, you have a confirmed model, optimized factor settings, and at least one validation run that tests whether your predicted response holds in practice. A single DOE rarely answers every question, so treat each completed study as a stepping stone. Document what you learned, which factors mattered, and which you can now hold constant in future experiments.
Your next move depends on what the data showed. If the prediction profiler pointed to factor settings near the boundary of your original ranges, run a follow-up experiment that extends those ranges. If interactions dominated the model, design a response surface study to map the curvature more precisely. JMP design of experiments tools make both of those follow-up paths straightforward from the same Custom Designer menu.
Ready to put this into practice with expert guidance behind you? Contact the Lean Six Sigma Experts team to learn how we help organizations build repeatable DOE programs that deliver real process results.