The compactification of generalized linear models
The main purpose of this paper is to unify and extend the existing theory of `estimated zeroes' in log-linear and logit models. To this end it is shown that every generalized linear model (GLM) can be embedded in a larger model with a compact parameter space and a continuous likelihood (a `CGLM'). Clearly in a CGLM the maximum likelihood estimate (MLE) always exists, easing a major data analysis problem. In the mean-value parametrization, the construction of the CGLM is remarkably simple; except in a rather pathological and rare case, estimated expected values are always finite. In the \(\beta\)-parametrization however, the compactification is more complex; the MLE need not correspond to a finite \(\hat\beta\), as is well known for estimated zeroes in log-linear models. The boundary distributions of CGLMs are classified in four categories: `inadmissible', `degenerate', `Chentsov', and `constrained'. For a large class of GLMs, including all GLMs with canonical link functions and probit models, the MLE in the corresponding CGLM exists and is unique. Even stronger, the likelihood has no other local maxima. Equivalent algebraic and geometric conditions [in the vein of \textit{S. J. Haberman}, see ``The analysis of frequency data. (1974; Zbl 0325.62017), and \textit{A. Albert} and \textit{J. A. Anderson}, see Biometrika 71, 1-10 (1984; Zbl 0543.62020)], necessary for the existence of the MLE in the GLM corresponding to a finite \(\hat\beta\), are given. For a large class of GLMs these conditions are also sufficient. Even for log-linear models this seems to be a new result.
- scientific article; zbMATH DE number 4215208
- scientific article; zbMATH DE number 3878188
- Using penalized likelihood to select parameters in a random coefficients multinomial logit model
- scientific article; zbMATH DE number 4043042
- scientific article; zbMATH DE number 3976123
- Jeffreys-prior penalty, finiteness and shrinkage in binomial-response generalized linear models
- A Systematic Theory of Exponential Families of Probability Distributions
- Convex Analysis
- scientific article; zbMATH DE number 3145638 (Why is no real title available?)
- scientific article; zbMATH DE number 44577 (Why is no real title available?)
- scientific article; zbMATH DE number 3509596 (Why is no real title available?)
- scientific article; zbMATH DE number 3602484 (Why is no real title available?)
- Maximum likelihood estimates in exponential response models
- On the existence and uniqueness of the maximum likelihood estimates for certain generalized linear models
- Wald's Test as Applied to Hypotheses in Logit Analysis
- On TLS formulation and core reduction for data fitting with generalized models
- Computationally efficient likelihood inference in exponential families when the maximum likelihood estimator does not exist
- A Compressive Sensing Based Analysis of Anomalies in Generalized Linear Models
- scientific article; zbMATH DE number 4215208 (Why is no real title available?)
- Maximum likelihood estimation in log-linear models
- Automated subset selection via information criteria optimization in generalized linear models
- Bayesian finite mixtures of Ising models
- Finite-sample bounds to the normal limit under group sequential sampling
- Verifying the existence of maximum likelihood estimates for generalized linear models
This page was built for publication: The compactification of generalized linear models
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4696593)