elrm: Software Implementing Exact-like Inference for Logistic Regression Models

Abstract

Exact inference is based on the conditional distribution of the sufficient statistics for the parameters of interest given the observed values for the remaining sufficient statistics. Exact inference for logistic regression can be problematic when data sets are large and the support of the conditional distribution cannot be represented in memory. Additionally, these methods are not widely implemented except in commercial software packages such as LogXact and SAS. Therefore, we have developed elrm, software for R implementing (approximate) exact inference for binomial regression models from large data sets. We provide a description of the underlying statistical methods and illustrate the use of elrm with examples. We also evaluate elrm by comparing results with those obtained using other methods.

Post-intravitreal anti-VEGF endophthalmitis in the United Kingdom: incidence, features, risk factors, and outcomes

“…In cases where dichotomous categorical-independent variable cells had less than five cases contributing to the count, the Fisher's exact test was used instead. Confidence intervals (CIs) for the odds ratio (OR) were calculated using the exact distribution, and exact logistic regression was carried out using the elrm package in R. 26,27 These methods were required because some of the sample sizes in the exposure groups was small.…”

Monte Carlo Analysis in Academic Research

“…For larger problems, the MC hypothesis test approximates that distribution by sampling. Algorithms have been developed to extend the exact test to some logistic regressions, for example, (Hirji et al, 1987;Mehta & Patel, 1995), and yet for larger problems, an approximation by simulation is necessary (Zamar et al, 2007 Besag & Diggle, 1977;Ripley, 1977;Marriott, 1979). Random processes are hypothesized to cause things (animals, plants, etc.)…”

“…For larger problems, the MC hypothesis test approximates that distribution by sampling. Algorithms have been developed to extend the exact test to some logistic regressions, for example, (Hirji et al, 1987;Mehta & Patel, 1995), and yet for larger problems, an approximation by simulation is necessary (Zamar et al, 2007). The MC hypothesis test is not discussed in most statistics texts, perhaps a signal that it is not considered necessary for most common statistical problems.…”

Exploring The Development of Social Responses in Children with Callous and Unemotional Traits: An Examination of The Impact of Hypothesized Reinforcing and Aversive Stimuli

“…As it is not advisable to use logistic regression in small samples because of the possibility of biases in the maximum likelihood estimator (Mehta & Patel, 1995), exact logistic regression (Zamar, et al, 2009) was used to determine the predictive validity of CU traits for frequency, duration and physiological reactivity in the behavioral task. Exact logistic regression utilizes the sufficient statics for the covariates in the model and is able to withstand small sample sizes (Mehta & Patel, 1995).…”

“…Electrodermal data were initially viewed and evaluated using Affectiva's Q (2.01.56) and Q Live EDA output. The R package, elrm (1.2.2), was used to conduct exact logistic regressions (Zamar, McNeney & Graham, 2009). All session were video recorded using a Canon Vixia HF R300 digital video camera for purposes of timeline corroboration with the EDA recordings.…”