Ordinary versus random-effects logistic regression for analyzing herd-level calf morbidity and mortality data

Charles R. Curtis, Robert H. Mauritsen, Philip H Kass, Mowafak D. Salman, Hollis N. Erb

Research output: Contribution to journalArticle

41 Scopus citations

Abstract

The analysis of data from populations organized into groups is frequently complicated by cluster (herd) effects. When present, clustering effects influence the probability of a health-related event in a way that is not readily accounted for with classical fixed-effects models (including unconditional logistic regression). Clustering introduces an additional source of (extra-binomial) variation into a logistic regression model, violating the independence and identical-distribution assumptions, and leading to biased variance estimators and spurious statistical significance. One remedy is to treat herd effects as random effects, so that variation even from unmeasured and unmeasurable (but clustered) sources can be accounted for. The logistic-normal regression model is introduced as one such random-effects model; its application is demonstrated using data from a previously described study of calfhood morbidity and mortality in 25 New York dairy herds.

Original languageEnglish (US)
Pages (from-to)207-222
Number of pages16
JournalPreventive Veterinary Medicine
Volume16
Issue number3
DOIs
StatePublished - 1993

ASJC Scopus subject areas

  • Animal Science and Zoology
  • veterinary(all)

Fingerprint Dive into the research topics of 'Ordinary versus random-effects logistic regression for analyzing herd-level calf morbidity and mortality data'. Together they form a unique fingerprint.

  • Cite this