A non-parametric regression method in which the predicted value of the dependent variable is calculated as the weighted average of the data points, where the weights are assigned according to a so-called kernel function. Typically, a kernel function, or a kernel, is symmetric and depends on the distance between the data points. The rate at which the weights decrease with the distance is controlled by a smoothness parameter called ‘bandwidth’: the larger the bandwidth, the slower is the fall in weights, and the smoother is the fit of the regression. In practice, the fit of a kernel regression is more sensitive to the choice of the bandwidth than to the choice of a particular functional form for the kernel.