Local regression and likelihood

From MaRDI portal





Suppose a dataset consists of \(n\) pairs of observations \((x_1,Y_1),\;(x_2,Y_2),\ldots,(x_n,Y_n)\), where \(x\) is a (nonrandom) predictor variable (real or vector valued) and \(Y\) is a (random) response variable related to the predictor variable. The problem is to predict \(Y\) for a fixed value \(x\) which may not belong to the set \(\{x_1,x_2,\ldots,x_n\}\). It is obvious that a kind of interpolation (or perhaps extrapolation) between the points \(x_1,x_2,\ldots,x_n\) is needed, and that those of the points which are nearer to \(x\) should play a greater role. If a model of the form \(Y=\mu(x)+\varepsilon\) with a deterministic function \(\mu\) and random error \(\varepsilon\) is assumed, the problem is known as that of local regression. Ch. 2 of the book gives details. For a model of the form ``\(Y\) has a probability density function \(f(y,\mu(x,\theta))\) with given form of \(f\) and \(\mu(x,\theta)\), a local maximum likelihood estimate for the parameter \(\theta\) may be conctructed. The Local Likelihood Model is discussed in Ch. 4. Similarly one can formulate the problem of local density estimation (Ch. 5), local Survival and Failure Time Analysis (Ch. 6), and so on. The book gives us a complete and uptodate review of the state of the art. Practical applications and computer software (LOCFIT) are presented in detail.




Cited in
(only showing first 100 items - show all)


Describes a project that uses

Uses Software






This page was built for publication: Local regression and likelihood

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4267797)