Multi-response Response Surface Methodology
Also known as: Multi-response RSM, MRSM, Multi-objective RSM, Multiple response optimization
Multi-response Response Surface Methodology (MRSM) extends classical RSM to situations where an experiment generates two or more response variables that must be optimized simultaneously. Rather than tuning factor settings for a single output, MRSM fits a separate second-order polynomial model for each response, then combines them — most commonly via Derringer and Suich's desirability function — to find factor settings that satisfy all objectives at once.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
+3 more
When to use it
Use MRSM when a designed experiment yields two or more response variables that must all be brought to acceptable levels simultaneously and when the responses can be reasonably modeled by second-order polynomials over a bounded factor space. It is well-suited to process optimization in manufacturing, chemical engineering, food science, and pharmaceutical development where throughput, quality, and cost objectives conflict. Do not use MRSM if responses are non-continuous (binary, count, categorical) without transformation, if the factor space is highly non-convex or high-dimensional (more than ~8 factors), if response surfaces exhibit severe non-polynomial curvature, or if only a single response matters — standard RSM is sufficient in that case. Also avoid when physically meaningful objective weights cannot be assigned, as the desirability aggregation is sensitive to weight choices.
Strengths & limitations
- Simultaneously optimizes multiple competing responses from a single designed experiment with no additional runs.
- The desirability function provides an interpretable 0–1 scale that accommodates different response types (maximize, minimize, target) and allows explicit trade-off weighting.
- Built on the well-validated RSM polynomial framework; extensive software support (JMP, Minitab, R rsm/desirability packages, SAS).
- Graphical overlaid contour plots make the feasible region and trade-offs visible to engineers and decision-makers.
- Confirmation experiments provide an empirical check on model validity before committing to new process settings.
- Individual response models must each be adequate; a poor fit for any single response compromises the joint optimum.
- Desirability function weights and bounds are analyst-specified and subjective; different weight assignments can yield different optimal settings.
- Second-order polynomial models may be insufficient for responses with sharp optima, strong higher-order effects, or discontinuities.
- The geometric mean aggregation can mask unacceptable individual responses when other desirabilities are very high.
- Assumes the experimental region captures the true optimum; extrapolation beyond the design space is unreliable.
Frequently asked
How is MRSM different from running separate RSM optimizations for each response?
Separate single-response RSM optimizations will almost certainly yield different, conflicting optimal factor settings. MRSM finds a single compromise set of factor levels that is collectively acceptable across all responses. The desirability function explicitly quantifies and balances the trade-offs, whereas running separate optimizations forces you to choose one response's optimum and accept suboptimal performance on all others.
How do I choose the weights in the desirability function?
Weights reflect the relative business or engineering importance of each response. Start by engaging stakeholders to rank responses by criticality. Equal weights are a reasonable starting point when priorities are genuinely equivalent. Conduct a sensitivity analysis by varying weights across a plausible range and checking whether the optimal factor settings change substantially; robust solutions are preferable when they exist.
Which experimental designs work best with MRSM?
Central Composite Designs (CCD) and Box-Behnken Designs (BBD) are the most common choices because they support second-order polynomial fitting with a reasonable number of runs. CCDs are preferred when the experimenter can run star points outside the factor range; BBDs avoid extreme corner settings and work well when those extremes are impractical or costly.
What if the desirability score is very low at the optimum?
A low overall desirability (e.g., D < 0.4) signals that the response requirements are difficult to satisfy simultaneously within the explored factor space. Consider relaxing one or more response bounds in consultation with stakeholders, expanding the experimental region if safe to do so, or investigating whether an uncontrolled noise factor is responsible for poor response performance.
Can MRSM handle non-normal or binary responses?
Standard MRSM assumes continuous responses that can be reasonably modeled with ordinary least squares. Binary or count responses require generalized linear model extensions of RSM (GLM-RSM). Moderately skewed continuous responses can often be handled via variance-stabilizing transformations (log, square root) applied before model fitting.
Sources
- Derringer, G., & Suich, R. (1980). Simultaneous optimization of several response variables. Journal of Quality Technology, 12(4), 214–219. DOI: 10.1080/00224065.1980.11980968 ↗
- Myers, R. H., Montgomery, D. C., & Anderson-Cook, C. M. (2016). Response Surface Methodology: Process and Product Optimization Using Designed Experiments (4th ed.). Wiley. ISBN: 978-1118916025
How to cite this page
ScholarGate. (2026, June 3). Multi-response Response Surface Methodology. ScholarGate. https://scholargate.app/en/experimental-design/multi-response-response-surface-methodology
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Box-Behnken DesignExperimental design↔ compare
- Central Composite DesignExperimental design↔ compare
- Design of experimentsExperimental design↔ compare
- Optimization-assisted response surface methodologyExperimental design↔ compare
- Quality Function DeploymentExperimental design↔ compare
- Response Surface MethodologyExperimental design↔ compare