Infinite Parameter Estimates in Logistic Regression: Opportunities, Not Problems

2002 ◽  
Vol 27 (2) ◽  
pp. 147-161 ◽  
Author(s):  
David Rindskopf
Keyword(s):  
Logistic Regression ◽  
Maximum Likelihood ◽  
Confidence Intervals ◽  
Maximum Likelihood Estimates ◽  
Parameter Estimates ◽  
Strict Sense ◽  
Hypothesis Tests ◽  
Infinite Slope ◽  
Alternative Method

Infinite parameter estimates in logistic regression are commonly thought of as a problem. This article shows that in principle an analyst should be happy to have an infinite slope in logistic regression, because it indicates that a predictor is perfect. Using simple approaches, hypothesis tests may be performed and confidence intervals calculated even when a slope is infinite. Some functions of parameters that are infinite are still well defined, and reasonable estimates of these quantities (in particular, LD50) may be obtained even when the maximum likelihood estimates do not, in a strict sense, exist. Surprisingly, these techniques can provide more reasonable and useful results than the most popular alternative method, exact logistic regression.

Reliability of Pharmacodynamic Analysis by Logistic Regression

2000 ◽  
Vol 92 (4) ◽  
pp. 985-992 ◽  
Author(s):  
Wei Lu ◽  
James M. Bailey
Keyword(s):  
Logistic Regression ◽  
Confidence Intervals ◽  
Coefficient Of Variation ◽  
Parameter Estimates ◽  
True Parameter ◽  
True Value ◽  
Coefficients Of Variation ◽  
Effect Relation ◽  
Log Normal ◽  
Multiple Patients

Background Many pharmacologic studies record data as binary yes-or-no variables, and analysis is performed using logistic regression. This study investigates the accuracy of estimation of the drug concentration associated with a 50% probability of drug effect (C50) and the term describing the steepness of the concentration-effect relation (gamma). Methods The authors developed a technique for simulating pharmacodynamic studies with binary yes-or-no responses. Simulations were conducted assuming either that each data point was derived from the same patient or that data were pooled from multiple patients in a population with log-normal distributions of C50 and gamma. Coefficients of variation were calculated. The authors also determined the percentage of simulations in which the 95% confidence intervals contained the true parameter value. Results The coefficient of variation of parameter estimates decreased with increasing n and gamma. The 95% confidence intervals for C50 estimation contained the true parameter value in more than 90% of the simulations. However, the 95% confidence intervals of gamma did not contain the true value in a substantial number of simulations of data from multiple patients. Conclusion The coefficient of variation of parameter estimates may be as large as 40-50% for small studies (n < or = 20). The 95% confidence intervals of C50 almost always contain the true value, underscoring the need for always reporting confidence intervals. However, when data from multiple patients is naively pooled, the estimates of gamma may be biased, and the 95% confidence intervals may not contain the true value.

Likelihood-based confidence intervals of relative fitness for a common experimental design

2005 ◽  
Vol 62 (3) ◽  
pp. 693-699 ◽  
Author(s):  
Steven T Kalinowski ◽  
Mark L Taper
Keyword(s):  
Maximum Likelihood ◽  
Experimental Design ◽  
Confidence Intervals ◽  
Maximum Likelihood Estimates ◽  
Relative Fitness ◽  
Literature Analysis ◽  
Confidence Limits ◽  
Sample Sizes ◽  
Statistical Inferences ◽  
Different Types

Statistical inferences concerning the relative fitness of different types of individuals in a population have not been well developed. We present a method for calculating confidence intervals for maximum likelihood estimates of relative fitness obtained from an experimental design that is common in the fisheries literature. Analysis and simulation show that these confidence limits are reliable. We also show that the bias of the estimates is low for realistic sample sizes.

scholarly journals A New Maximum Likelihood Estimator Formulated in Pole-Residue Modal Model

2019 ◽  
Vol 9 (15) ◽  
pp. 3120
Author(s):  
Sandro Amador ◽  
Mahmoud El-Kafafy ◽  
Álvaro Cunha ◽  
Rune Brincker
Keyword(s):  
Maximum Likelihood ◽  
Least Squares ◽  
Maximum Likelihood Estimator ◽  
Confidence Intervals ◽  
Modal Parameters ◽  
Parameter Estimates ◽  
Modal Parameter ◽  
Likelihood Estimator ◽  
Pole Residue ◽  
Modal Model

Recently, a lot of efforts have been devoted to developing more precise Modal Parameter Estimation (MPE) techniques. This is explained by the necessity in civil, mechanical and aerospace engineering of obtaining accurate estimates for the modal parameters of the tested structures, as well as of determining reliable confidence intervals for these estimates. The Non-linear Least Squares (NLS) identification techniques based on Maximum Likelihood (ML) have been increasingly used in modal analysis to improve precision of estimates provided by the Least Squares (LS) based estimators when they are not accurate enough. Apart from providing more accurate estimates, the main advantage of the ML estimators, with regard to their LS counterparts, is that they allow for taking into account not only the measured Frequency Response Functions (FRFs) but also the noise information during the parametric identification process and, therefore, provide the modal parameters estimates together with their uncertainties bounds. In this paper, a new derivation of a Maximum Likelihood Estimator formulated in Pole-residue Modal Model (MLE-PMM) is presented. The proposed formulation is meant to be used in combination with the Least Squares Frequency Domain (LSCF) to improve the precision of the modal parameter estimates and compute their confidence intervals. Aiming at demonstrating the efficiency of the proposed approach, it is applied to two simulated examples in the final part of the paper.

scholarly journals Analyzing data from memory tasks - comparison of ANOVA, logistic regression and mixed logit model

Psihologija ◽  
2018 ◽  
Vol 51 (4) ◽  
pp. 469-488
Author(s):  
Milica Popovic-Stijacic ◽  
Ljiljana Mihic ◽  
Dusica Filipovic-Djurdjevic
Keyword(s):  
Logistic Regression ◽  
Confidence Intervals ◽  
Logit Model ◽  
Linear Models ◽  
Mixed Logit ◽  
Binary Logistic Regression ◽  
Standard Errors ◽  
Parameter Estimates ◽  
Mixed Logit Model ◽  
Mixed Logit Models

We compared three statistical analyses over binary outcomes. As applying ANOVA over proportions violates at least two classical assumptions of linear models, two alternatives are described: the binary logistic regression and the mixed logit model. Firstly, we compared the effects obtained by the three methods over the same data from a previous memory research. All three methods gave similar results: the effects of the tasks and the number of sensory modalities were observed, but not their interaction. Secondly, by using the bootstrap estimates of the parameters, the efficacy of each method was explored. As predicted, the bootstrap parameter estimates of the ANOVA had large bias and standard errors, and consequently wide confidence intervals. On the other hand, the bootstrap parameter estimates of the binary logistic regression and the mixed logit models were similar ? both had low bias and standard errors and narrow confidence intervals.

Infinite Likelihood

2017 ◽  
Author(s):  
Russell Cheng
Keyword(s):  
Maximum Likelihood ◽  
Asymptotic Normality ◽  
Confidence Intervals ◽  
Parameter Estimates ◽  
Threshold Parameter ◽  
Bootstrap Confidence Intervals ◽  
Stable Law ◽  
Log Likelihood ◽  
Consistency And Asymptotic Normality

This chapter examines methods that overcome a difficulty with infinite likelihoods. In shifted threshold distributions where the PDF has the form f(y) ∼ k(b,c)(y−a)c−1, if y tends to the threshold parameter a, then the log-likelihood tends to infinity if c < 1 and a also tends to y(1) the smallest observation. The maximum likelihood (ML) method fails in this case, yielding parameter estimates that are not consistent. A method is described overcoming this problem, called the maximum product of spacings method. This yields parameter estimates with the same consistency and asymptotic normality properties as ML estimators when these exist, and which yield, when c < 1 where ML fails, consistent estimates with that for a hyper-efficient. Confidence intervals for a are difficult to obtain theoretically when c < 2. A method is given using percentiles of the stable law distribution and this is numerically compared with bootstrap confidence intervals.

Sign in / Sign up

Export Citation Format

Share Document