Linear model vs. survival analysis for genetic evaluation of sires for longevity in Chianina beef cattle

The objective of this study was to compare sire EBVs for longevity in Chianina beef cattle estimated with linear models and survival analysis. Two datasets were created, one considered all data (SURVall), the other only uncensored records (SURVun). The linear models were used to analyze longevity measured as three correlated dichotomous (yes/no) measures of survival in the first three parities (LIN-S3) and as an overall measure of lifespan in months (LIN-LPL). Correlation between sire EBVs from the two survival analyses were 0.85. For LrN-S3 the correlations of EBVs across parities were between 0.69 to 0.93. Medium correlations (from 0.50 to 0.62) were found when only uncensored data (SURVun) were compared to the linear model (LIN-S3). Higher correlations (from 0.71 to 0.93) were found when EBV based on both censored and uncensored data (SURVall) were compared to LIN-S3. Heritability was estimated at 0.11, 0.09 and 0.08 for SURVall, SURVun and LIN-LPL, respectively; and 0.05, 0.02 and 0.02, respectively, for survival in the first three parities according to UN-S3. Linear and non-linear models differed in many aspects; the most precise EBV were obtained when all data was used in the evaluation. (c) 2005 Elsevier B.V. All rights reserved.