Comparison of Alternative Calibration Procedures for Infra-red Milk Analyzers

Abstract

Alternative calibration procedures were used to produce instrumental estimates based on reference sample signals and chemical values collected over a period of one year in the Ontario Central Milk Testing Laboratory, Guelph, Ontario, Canada. Mean difference (estimate less chemical values) and standard deviation of differences from selected combinations of calibration conditions were evaluated using two ANOVA models. Model I considered four types of fat calibration [Fat A, Fat B, Fat ApB (73% B and 27% A), and Fat A+B (50% A and 50% B)], three levels of the number of samples between calibration, three levels of the number of samples used for calibration and three instruments. Model II compared the fixed intercorrection factor (FIF) approach to the multiple linear regression (MLR) approach using the same levels of samples for and between calibrations, as in Model I. The results showed that the MLR approach was statistically equivalent to the FIF approach, and Fat ApB (Fat A proportion B) calibration was equivalent to Fat A+B but slightly better than Fat B alone, which was slightly better than Fat A alone. The optimum samples for and between calibrations were 50, and 25 or 50, respectively. In most cases, instrument effects were significant and were similar in magnitude to treatment effects.