Statistics Glossary → Ridge analysis (response surface)
Ridge analysis (response surface)concept
A method for a second-order response surface model that traces the factor settings giving the largest (or smallest) predicted response on spheres of increasing radius around the centre of the experimental region.
Response surface regression models an outcome with the linear, squared and interaction terms of several factors and locates the stationary point of the surface (a peak, a trough or a saddle). Often, though, the stationary point is a saddle or lies outside the region where the experiments were run, and the model's prediction there is unreliable. Ridge analysis instead finds, among all points at a given distance (radius R) from the centre in coded units, the one with the largest predicted response, and repeats this as R is increased step by step. The resulting path shows the most promising direction for improvement.
In coded units, a radius of 1 corresponds to the edge of the experimental region; predictions beyond it are extrapolations and should be used only as guidance for further experiments. The standard error of the predicted response should be tracked along the ridge path as well, because uncertainty grows with distance from the centre. The method assumes that the second-order model fits adequately within the region.
An everyday example
A bakery models how well a cake rises as a function of oven temperature and baking powder, using a response surface. The model's stationary point turns out to be a saddle and does not point straight to the best recipe. Ridge analysis follows the settings giving the highest rise on circles widening step by step around the current recipe and suggests 'raise the temperature a little, reduce the baking powder a little'; the bakery runs its next trials in that direction.
In MerQur
In Response Surface Regression, the 'Ridge analysis' option (Maximum response, Minimum response, Both, None) together with the 'Maximum radius' (default 1.0) and 'Radius step' (default 0.1) fields; the radius is in coded units (1 = the edge of the experimental region).
Analyses where this concept appears
Response Surface Regression (RSREG)
Related concepts
Stationary point and canonical analysis (response surface)Coded factors (response surface)