PALM — Permutation Analysis of Linear Models — is a tool that allows inference using permutation methods, offering a number of features not available in other analysis software. These features currently include:

• Ability to work with volumetric and surface-based formats, including facewise data, as well as with non-imaging data;
• A range of various regression and permutation strategies;
• Statistics that are robust to heteroscedasticity;
• Shuffling of sets of observations, to allow, for instance, the analysis of certain designs with repeated measurements, with no missing data;
• Shuffling of observations with complex, tree-like covariance structure (such as for the Human Connectome Project);
• Permutation with sign-flipping (wild bootstrap);
• Modified Non-Parametric Combination (NPC) for joint inference over multiple modalities, or multiple contrasts, or both together, with various combining functions available;
• Classical multivariate statistics (MANOVA, MANCOVA, CCA) for joint inference over multiple modalities, assessed through robust permutation methods, and also parametrically when such approximations exist;
• Correction over multiple contrasts, multiple modalities, for images with or without the same size or geometry, including non-imaging data, controlling the FWER or FDR;
• Various acceleration methods based on the test statistics and their distributions.

PALM requires Matlab or Octave. It can be executed from inside either environment, or directly from the shell. It can also be called from scripts.

PALM is experimental software. As novel features are introduced, tested, verified, and validated, eventually they will be implemented and made available in randomise or in other tools. PALM is for users who are familiar with statistics and willing to use experimental analysis tools. Bugs (real or suspected) can be reported via GitHub or through the FSL mailing list.

To download PALM and for installation instructions, please visit the User Guide.

---

References

The main reference for PALM is the same as for randomise:

For Non-Parametric Combination (NPC; options -npc, -npcmod and -npccon), classical multivariate tests (MANOVA, MANCOVA, CCA; options -mv and -cca) assessed with permutations, and for correction over contrasts and/or modalities (options -corrcon and -corrmod), the reference is:

For the multi-level exchangeability blocks (options -eb, -vg, and for HCP data), the reference is:

For the accelerated permutation inference (options -accel or -approx), the reference is:

For additional theory of permutation tests in neuroimaging, please see and cite:

---

CategoryStatistics