Interpretation guidance

FPC signs are arbitrary

An eigenfunction can be multiplied by -1 with its scores multiplied by -1 without changing the model. Interpret the two ends of the component contrast, not the sign itself.

Variance explained is descriptive

High variance explained does not make a component causal, theoretically important, or predictive.

Component shapes are estimated

A smooth-looking FPC is not fixed truth. Inspect matched bootstrap stability and, when shape interpretation matters, descriptive component envelopes.

When eigenvalues are close, component labels can swap across resamples. Matching similarity should be interpreted before pointwise envelope width.

Scores inherit preprocessing

A score derived after normalization, registration, smoothing, or coordinate centering represents that transformed process.

Prefer trajectory language

Good: “Higher FPC1 scores represented earlier rightward evidence-panel excursions followed by return toward the decision region.”

Weak: “FPC1 represented attention.”