Random-effects linear model application to herd-level assessment of bovine hepatic trace mineral concentrations

Abstract

To evaluate the utility of random-effects linear modeling for herd-level evaluation of trace mineral status, we performed a retrospective analysis of the results for trace mineral testing of bovine liver samples submitted to the Michigan State University Veterinary Diagnostic Laboratory between 2011 and 2017. Our aim was to examine random-effects models for their potential utility in improving interpretation with minimal sample numbers. The database consisted of 1,658 animals distributed among 121 herds. Minerals were assayed by inductively coupled plasma–mass spectroscopy, and included cobalt, copper, iron, molybdenum, manganese, selenium, and zinc. Intraclass correlation coefficients for each mineral were significantly different ( p < 0.001) from zero and ranged from 0.38 for manganese to 0.82 for selenium, indicating that the strength of herd effects, which are presumably related to diet, vary greatly by mineral. Analysis of the distribution and standard errors of best linear unbiased predictor (BLUP) values suggested that testing 5–10 animals per herd could place herds within 10 percentile units across the population of herds with 70–95% confidence, the confidence level varying among minerals. Herd means were generally similar to BLUPs, suggesting that means could be reasonably compared to BLUPs with respect to the distributions reported here. However, caution in interpreting means relative to BLUPs should be exercised when animal numbers are small, the standard errors of the means are large, and/or the values are near the extremes of the distribution.