Functional Data Analysis and Statistical Inference Methods

Functional data analysis (FDA) is the branch of statistics that treats individual observations as continuous functions, curves or surfaces rather than as finite-dimensional vectors. Central to FDA are methods for smoothing raw measurements, representing functions via basis expansions, and extracting dominant modes of variation through functional principal component analysis. Statistical inference in this setting encompasses functional linear models—regressing scalar or functional responses on scalar or functional predictors—as well as hypothesis testing frameworks such as functional analysis of variance (FANOVA) for comparing mean curves across groups and tests for equality of covariance operators. Recent methodological advances emphasise nonparametric estimation of mean and covariance functions, intervalwise inference for identifying regions of significant effect, and robust procedures for non-Gaussian or sparsely observed data. FDA techniques have found broad applications in biomechanics (gait and movement analysis), neuroimaging (time-series of brain signals), environmental science (pollutant concentration trajectories) and epidemiology (disease progression curves), providing a powerful toolkit for capturing dynamic patterns and enabling precise scientific conclusions.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

A novel basis expansion approach has been developed for functional analysis of variance with repeated measures. By expressing each observed function in a chosen basis and recasting the two-way FANOVA problem as a multivariate analysis of basis coefficients, this method accounts for full curve information under different experimental conditions. Extensive simulations demonstrate superior power and error-rate control, and applications to gait data illustrate its practical benefits in biomechanical studies. Another contribution introduces a two-sample inference technique for functional data leveraging characteristic functions. Functional observations are projected onto a basis and the characteristic functions of resulting coefficient vectors are compared via a weighted distance, with null distributions approximated through permutation. This framework offers strong finite-sample performance and flexibility in the absence of Gaussian assumptions. Complementing these parametric strategies, classical nonparametric tests—sign, Wilcoxon and Mann-Whitney—have been extended to functional data via random projections. Functions are projected onto random directions to yield scalar summaries, to which standard rank-based tests are applied. Empirical studies confirm good size control and robustness in non-Gaussian settings, underscoring the versatility of projection-based inference in functional contexts.

Functional Data Analysis and Statistical Inference Methods publication trend

The graph below shows the total number of articles in functional data analysis and statistical inference methods across all publications each year (not limited to Nature Index journals).

Technical terms

Functional data: Observations recorded over a continuum (for example time or space), treated as single entities in the form of curves or functions.

Basis expansion: Representation of functions as sums of weighted basis elements (such as splines or Fourier functions), reducing infinite-dimensional data to finite coefficients.

Functional principal component analysis (FPCA): Method for dimension reduction that identifies orthogonal eigenfunctions capturing the main modes of variation in a set of curves.

Functional analysis of variance (FANOVA): Extension of classical ANOVA to test differences among mean functions across multiple groups of functional observations.

Covariance operator: Generalisation of the covariance matrix to function space, describing how functional observations co-vary across their domain.

References

1. Basis expansion approaches for functional analysis of variance with repeated measures. Advances in Data Analysis and Classification (2022).
2. Two-sample Tests for Functional Data Using Characteristic Functions. Austrian Journal of Statistics (2021).
3. Sign, Wilcoxon and Mann-Whitney Tests for Functional Data: An Approach Based on Random Projections. Mathematics (2020).