[Paper Review] Nonparametric testing for no-effect with functional responses and functional covariates

Valentin Patilea, César Sánchez-Sellero+1 more|arXiv (Cornell University)|Sep 10, 2012

Statistical Methods in Clinical Trials20 references3 citations

TL;DR

This paper proposes a nonparametric test for detecting the effect of a functional or multivariate covariate on a functional response without assuming a specific regression model. Using univariate nearest neighbor smoothing and a dimension reduction step to project high-dimensional covariates into a univariate index, the test statistic follows a standard normal asymptotic distribution, enabling detection of both linear and nonlinear effects under general error and design conditions.

ABSTRACT

This paper examines the problem of nonparametric testing for the no-effect of a random covariate (or predictor) on a functional response. This means testing whether the conditional expectation of the response given the covariate is almost surely zero or not, without imposing any model relating response and covariate. The covariate could be univariate, multivariate or functional. Our test statistic is a quadratic form involving univariate nearest neighbor smoothing and the asymptotic critical values are given by the standard normal law. When the covariate is multidimensional or functional, a preliminary dimension reduction device is used which allows the effect of the covariate to be summarized into a univariate random quantity. The test is able to detect not only linear but nonparametric alternatives. The responses could have conditional variance of unknown form and the law of the covariate does not need to be known. An empirical study with simulated and real data shows that the test performs well in applications.

Motivation & Objective

• To develop a nonparametric goodness-of-fit test for the no-effect of a covariate on a functional response without assuming a parametric model.
• To handle covariates that are univariate, multivariate, or functional, by reducing their dimension to a single index.
• To construct a test statistic that detects both linear and nonlinear effects, even when the conditional variance is unknown.
• To ensure asymptotic validity under minimal regularity conditions, including unknown covariate distribution and general error structure.
• To provide a practical method for model validation in functional regression by testing whether a predictor influences the response.

Proposed method

• Apply a dimension reduction technique to project a high-dimensional or functional covariate $X$ into a univariate index $\langle X, \gamma \rangle$ using a direction $\gamma$ estimated from data.
• Use univariate nearest neighbor smoothing to estimate the conditional expectation $\mathbb{E}(U \mid \langle X, \gamma \rangle)$, where $U$ is the functional response.
• Construct a quadratic form test statistic based on the smoothed residuals, which measures deviation from zero conditional expectation.
• Derive asymptotic critical values using the standard normal distribution under the null hypothesis $\mathbb{E}(U \mid X) = 0$ almost surely.
• Implement a sequential numerical algorithm to search for the optimal direction $\widehat{\gamma}_n$ that maximizes sensitivity to alternative effects.
• Justify the algorithm's consistency under the alternative by showing that the set of directions yielding zero conditional expectation has Lebesgue measure zero.

Experimental results

Research questions

• RQ1 Can a nonparametric test detect the absence of any effect of a functional or multivariate covariate on a functional response without assuming a specific model?
• RQ2 How can the effect of a high-dimensional or functional covariate be summarized into a univariate index for testing purposes?
• RQ3 Does the proposed test maintain correct size and power under general conditions, including unknown conditional variance and unknown covariate distribution?
• RQ4 Can the test detect nonlinear and nonparametric alternatives that linear models or FPC-based tests might miss?
• RQ5 Is the sequential algorithm for estimating the optimal index direction theoretically justified and consistent under the alternative hypothesis?

Key findings

• The proposed test statistic asymptotically follows a standard normal distribution under the null hypothesis of no effect.
• The test is valid under minimal assumptions: no parametric model is required, the conditional variance may be unknown, and the covariate distribution need not be known.
• The method detects both linear and nonlinear effects, overcoming limitations of existing FPC-based tests that fail to detect nonlinear alternatives.
• The dimension reduction step ensures that the test remains feasible and powerful even when the covariate is functional or high-dimensional.
• Theoretical justification shows that the set of directions yielding zero conditional expectation has Lebesgue measure zero, supporting the existence of informative directions.
• Empirical studies with simulated and real data (e.g., Canadian Weather and fruit fly egg-laying data) demonstrate good performance in finite samples.