Do any GraphPad programs peform multivariate statistical tests?
Yes. Prism offers a wide range of analyses involving multiple variables, and recent versions have expanded these capabilities substantially. The short answer depends on what you mean by "multivariate," a term that gets used in two distinct ways in practice.
The strict definition: multiple outcome variables analyzed simultaneously
In formal statistical usage, a multivariate analysis examines several dependent (outcome) variables at once, treating them as a joint distribution rather than analyzing each one separately. Principal Component Analysis (PCA) and clustering analyses fit this definition. Both are available in Prism and are described below.
The common usage: multiple predictor variables
In everyday scientific writing, "multivariate" frequently means that a single outcome variable is modeled as a function of two or more predictors. Statisticians prefer the term "multivariable" for this, but the broader usage is widespread. Prism has extensive support here, covering continuous, binary, count, and survival outcomes.
Truly multivariate analyses in Prism
These analyses work across multiple variables simultaneously, without designating any single column as the outcome.
Principal Component Analysis (PCA)
PCA is a dimensionality-reduction method that uncovers the underlying structure in datasets with many variables. It is particularly useful for exploratory work when variables are correlated and you suspect a smaller number of patterns drive most of the variation. Prism computes principal components, loadings, PC scores, and eigenvalues, and generates score plots, loadings plots, biplots, scree plots, and proportion-of-variance plots. PC scores can be carried forward for principal component regression. PCA is available in all Prism editions.
Hierarchical clustering
Hierarchical clustering groups observations based on their similarity across all selected variables, producing a dendrogram that shows the nested cluster structure at every level. Prism offers multiple distance methods (Euclidean, Manhattan, and others) and linkage algorithms (single, complete, average, Ward's, and others). Available with Prism Pro and Enterprise.
K-means clustering
K-means clustering partitions observations into K groups by minimizing within-cluster distances across all selected variables. Prism implements the K-means++ initialization algorithm for more reliable results and provides 17 cluster validation indices to help select the optimal number of clusters, along with elbow plots, silhouette plots, and gap plots. Available with Prism Pro and Enterprise.
Multivariable analyses in Prism
These analyses model a single outcome as a function of multiple predictor variables.
Multiple linear regression
Models a continuous outcome as a linear function of two or more predictors, which can be continuous, categorical, or both. Prism reports parameter estimates, standard errors, confidence intervals, P values, R², an ANOVA table, residual diagnostics, and multicollinearity statistics. Poisson regression, for count-data outcomes, is available as an option within the same analysis.
Multiple logistic regression
Models the probability of a binary outcome (yes/no, present/absent, survived/died) as a function of multiple predictors. Prism reports odds ratios with confidence intervals, P values, goodness-of-fit statistics, pseudo-R² values, a classification table, and tools for comparing competing models. Simple logistic regression with a single predictor is also available.
Cox proportional hazards regression
The standard method for survival analysis when multiple predictors need to be accounted for simultaneously. Cox regression models the hazard rate as a function of both categorical and continuous predictors without assuming a specific distribution for survival times. Prism reports hazard ratios, confidence intervals, P values, model comparison results, residual diagnostics, and estimated survival curves stratified by covariate patterns. Available with Prism Pro and Enterprise.
Two-way ANOVA
Tests the effects of two categorical factors and their interaction on a continuous outcome. Prism supports ordinary and repeated-measures designs (or a mixed effects model when data are unbalanced or missing), along with a full suite of multiple comparisons tests. This is one of the most widely used multivariable designs in biomedical research.
Three-way ANOVA
Tests three main effects, three two-way interactions, and one three-way interaction for a continuous outcome across three categorical factors. The three-way ANOVA analysis in Prism supports 2×2×K designs. For more flexible three-factor designs with more levels, Multifactor ANOVA is the better choice.
Multifactor ANOVA (N-way ANOVA)
Handles any number of categorical factors and any number of levels per factor, testing all main effects and interactions up to three-way. This removes the design restrictions of the specialized three-way ANOVA and supports four-factor, five-factor, and more complex designs as well. Uses Prism's Multiple Variables data table format. Available with Prism Pro and Enterprise.
Mixed effects models
For repeated-measures experiments with missing data or unbalanced designs, Prism fits a mixed effects model that treats subject-level variability as a random effect and tests the fixed effects of the experimental factors. This is available alongside repeated-measures ANOVA as an alternative approach for two-way and three-way designs.
Working with the Multiple Variables data table format
Prism's Multiple Variables data table format (standard database or "tidy" format, with one row per observation and one column per variable) is the natural home for complex multivariable datasets. From a Multiple Variables table you can run:
- Multiple linear regression and Poisson regression
- Multiple logistic regression
- Multifactor ANOVA
- Principal Component Analysis (PCA)
- Hierarchical clustering
- K-means clustering
- Cox proportional hazards regression
- Correlation matrix analysis
- t tests and related analyses
- Kaplan-Meier survival analysis
- Nonlinear regression
This makes it straightforward to import data from other software or databases and apply multiple different analyses to the same dataset without restructuring. Prism also offers numerous useful data wrangling and prep tools such as in-table formulas to create new variables within the multiple variables data table.
Analyses not available in Prism
Prism is designed for the analyses most commonly needed in biomedical and life sciences research. A few techniques that are sometimes grouped under the "multivariate" umbrella are largely outside that scope: MANOVA, discriminant analysis, exploratory and confirmatory factor analysis, structural equation modeling, canonical correlation analysis, and partial least squares regression. Please reach out and let us know if your workflows require these analyses.
Note that stepwise regression is also not available in Prism. This is intentional: stepwise methods are widely recognized as producing unreliable results, inflated R² values, and misleading P values. The guide page "Why no stepwise regression?" covers the reasoning in detail.
Summary
| Analysis | Type | Available in |
|---|---|---|
| Principal Component Analysis | Truly multivariate | All editions |
| Hierarchical clustering | Truly multivariate | Pro, Enterprise |
| K-means clustering | Truly multivariate | Pro, Enterprise |
| Multiple linear regression | Multivariable | All editions |
| Multiple logistic regression | Multivariable | All editions |
| Poisson regression | Multivariable | All editions |
| Cox proportional hazards regression | Multivariable | Pro, Enterprise |
| Two-way ANOVA | Multivariable (2 factors) | All editions |
| Three-way ANOVA | Multivariable (3 factors, 2×2×K) | All editions |
| Multifactor ANOVA | Multivariable (N factors) | Pro, Enterprise |
| Mixed effects models | Multivariable | All editions |