Modal Regression Techniques in Statistical Inference

Modal regression focuses on estimating the conditional mode—the most probable outcome—given a set of predictors, thereby offering insights that complement classical mean and quantile regression. By targeting peaks of the conditional distribution, modal regression delivers robustness against skewness and outliers and can produce tighter prediction intervals in asymmetric scenarios. Techniques encompass nonparametric kernel‐based density derivative estimators, local polynomial modal regression, quantile regression–derived mode estimators and mean‐shift algorithms. A central challenge is the selection of smoothing parameters (bandwidths) to balance bias and variance in density estimation. Recent extensions address measurement error in covariates, multivariate and functional predictors, and integration with scalable learning frameworks. These advances have broadened applicability across economics, ecology, medicine and energy forecasting, enhancing inference when traditional regression assumptions are violated and enabling more precise characterisation of multimodal phenomena.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent work has introduced a local polynomial approach to modal regression that emphasises the conditional mode rather than mean or quantiles. This nonparametric estimator exhibits resilience to outliers and skewed distributions and has been extended via varying‐coefficient models to handle multivariate, functional and longitudinal data, with performance demonstrated through simulation studies and analyses of health‐care expenditure.

A semiparametric framework utilises linear quantile regression to approximate the conditional mode by inverting estimated quantile curves. Accompanied by asymptotic theory for its limiting distribution and methods for analytical and subsampling confidence intervals, this approach enables scalable conditional mode estimation and has been applied to predicting power‐plant energy output.

Another strand of research tackles measurement error in predictors by proposing two nonparametric modal estimators: one based on deconvolution of the joint density and the other on error‐prone conditional density estimation. Both approaches employ mean‐shift algorithms for mode seeking and integrate data‐driven bandwidth selection, showing improved accuracy compared to naïve methods in simulation experiments.

Modal Regression Techniques in Statistical Inference publication trend

The graph below shows the total number of articles in modal regression techniques in statistical inference across all publications each year (not limited to Nature Index journals).

Technical terms

Conditional mode: The value of the response variable that maximises its probability density conditional on given covariate values.

Kernel density estimation: A nonparametric technique that approximates an unknown probability density by averaging smooth kernel functions centred at each data point.

Bandwidth: A smoothing parameter governing the width of the kernel in density estimation, which controls the trade‐off between bias and variance.

Local polynomial regression: A regression method fitting low‐order polynomials within a neighbourhood around each target point using weighted least squares to capture local structure.

Mean‐shift algorithm: An iterative procedure that locates density modes by shifting data points towards regions of higher estimated density until convergence.

References

1. Data-driven density derivative estimation, with applications to nonparametric clustering and bump hunting. Electronic Journal of Statistics (2013).
2. Quantile regression approach to conditional mode estimation. Electronic Journal of Statistics (2019).
3. Nonparametric modal regression in the presence of measurement error. Electronic Journal of Statistics (2016).
4. Nonparametric statistical learning based on modal regression. Journal of Computational and Applied Mathematics (2022).