Multi-response Statistical Process Control — Multivariate SPC
Also known as: Multivariate SPC, MSPC, Multi-response SPC, Multivariate statistical process control
Multi-response statistical process control (multivariate SPC) extends classical univariate control charting to processes where two or more correlated quality characteristics must be monitored simultaneously. By treating all responses as a joint distribution, it detects shifts that would be invisible when each response is charted independently, reducing false alarms and improving the sensitivity of process monitoring in manufacturing and service contexts.
Key highlights
- Accounts for correlation among responses: detects multivariate shifts that are invisible to a battery of independent univariate charts.
- Controls the overall false-alarm rate at the desired level regardless of the number of responses monitored.
- A single chart replaces multiple individual charts, simplifying operator monitoring in high-dimensional processes.
- The MYT decomposition provides interpretable diagnostics pinpointing which responses drive an out-of-control signal.
- Extendable to CUSUM and EWMA variants for improved detection of small or gradual process shifts.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use multi-response SPC when a process produces two or more correlated quality characteristics that must all remain stable, and when a change in the joint distribution is operationally meaningful even if no individual characteristic crosses its own univariate limit. Typical settings include machining, chemical processing, pharmaceutical manufacturing, and food production. The method requires continuous measurement data and an in-control Phase I reference dataset. Do not use it as a simple replacement for univariate charts when responses are known to be uncorrelated — in that case, independent Shewhart charts are more interpretable. Avoid it when the number of responses p approaches or exceeds the Phase I sample size, as the covariance matrix becomes ill-conditioned; consider dimension reduction (e.g., PCA-based MSPC) instead.
Strengths & limitations
- Accounts for correlation among responses: detects multivariate shifts that are invisible to a battery of independent univariate charts.
- Controls the overall false-alarm rate at the desired level regardless of the number of responses monitored.
- A single chart replaces multiple individual charts, simplifying operator monitoring in high-dimensional processes.
- The MYT decomposition provides interpretable diagnostics pinpointing which responses drive an out-of-control signal.
- Extendable to CUSUM and EWMA variants for improved detection of small or gradual process shifts.
- Requires a sufficiently large Phase I dataset to estimate the covariance matrix reliably; performance degrades when Phase I sample size is small relative to the number of responses.
- Assumes multivariate normality of the quality characteristics; non-normal processes require non-parametric or transformed variants.
- A single T² signal indicates that something has changed but does not immediately identify the responsible response; additional decomposition is required.
- Covariance estimation and UCL calculation are more complex than univariate charting, requiring statistical software and specialist knowledge.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
When should I use multi-response SPC instead of separate univariate control charts for each response?
When the responses are correlated. Independent univariate charts inflate the joint false-alarm rate (by the Bonferroni effect) and can miss shifts in the correlation structure itself. If responses are uncorrelated, independent charts are simpler and equally effective. Check the Phase I correlation matrix; if off-diagonal entries are large in absolute value, multivariate SPC is warranted.
How many Phase I subgroups do I need to estimate the covariance matrix?
As a practical rule, collect at least 20–30 subgroups, and in any case many more subgroups than responses (m >> p). With too few subgroups the estimated covariance matrix is singular or nearly so, leading to unstable control limits. Bootstrapped or shrinkage estimators can help when Phase I data are scarce.
What do I do when a T² signal fires?
Apply the MYT (Mason–Young–Tracy) decomposition to partition the T² value into unconditional and conditional contributions from each response. The conditional T² for a response quantifies its contribution after accounting for the other responses, pointing directly to which variable or variable combination is responsible for the out-of-control state.
Can I use multi-response SPC with non-normal data?
The standard T² chart assumes multivariate normality. For mildly non-normal data the chart is reasonably robust with moderate subgroup sizes (n ≥ 4–5). For severely non-normal or count-type responses, use non-parametric multivariate charts (e.g., data-depth-based charts) or apply variance-stabilising transformations before charting.
How does PCA-based MSPC differ from the T² chart?
When the number of responses is large (say, p > 20) relative to the sample size, direct T² charting is infeasible because the covariance matrix cannot be inverted reliably. PCA-based MSPC first reduces dimensionality to a small set of principal components, then applies T² to the scores and a separate Q (SPE) chart to the residuals. It trades some interpretability for the ability to handle very high-dimensional measurement spaces.
Sources
- 1.Lowry, C. A., & Montgomery, D. C. (1995). A review of multivariate control charts. IIE Transactions, 27(6), 800–810.
- 2.Mason, R. L., & Young, J. C. (2002). Multivariate Statistical Process Control with Industrial Applications. ASA-SIAM.ISBN 978-0898715033
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Multi-response statistical process control. ScholarGate. https://scholargate.app/experimental-design/multi-response-statistical-process-control