A review of methods used to adjust for cluster effects in explanatory epidemiological studies of animal populations

Abstract

This paper reviews explanatory studies in the veterinary epidemiology literature, in which clusters (herds) were sampled and individual responses were measured. The studies were taken from the first ten volumes of Preventive Veterinary Medicine. The objectives of the review were: (1) to describe the statistical methods used to adjust for cluster effects; (2) to provide estimates of intracluster correlations for various health and production outcomes; and (3) for studies ignoring cluster effects, to estimate the proportion of significant inferences that would be rejected if cluster effects were included. Of the 67 papers reviewed, 36 (54%) used some form of adjustment for clustering. Using a fixed effect for herd was employed most frequently (27 papers). Four of the 67 papers used a matched design, two used a weighted analysis, and three papers included a random herd effect. An estimate of the intracluster correlation (^ϱ) was calculated when sufficient data were present. This estimate ranged from 0.0017 (lamb mortality data), to 0.46 (prevalence of brucelence serotiter). Most intracluster correlations ranged form 0.01 to 0.15. In the papers where ϱ was estimated from the data, the variance inflation factor ranged from 1.1 to 13. In 36 of the 67 (54%) papers reviewed, all inferences were accepted. In 30 of these 36, the authors correctly accounted for clustering, and in the other six the inference would not have changed after adjustment for clustering. In the remaining 31 (46%) papers, changes in inference were predicted. The proportion of incorrect inferences could be estimated in 20 of these 31 papers.