J Dairy Sci 1997 Oct;80(10):2572-9
Institute of Animal Science, Swiss Federal Institute of Technology, Zurich, Switzerland.
An analysis of productive life of Swiss Brown cattle was performed using a mixed survival model based on Cox regression. Data included 52,862 daughters of 297 sires. The length of productive life was observed from May 1, 1985 through August 15, 1995. Records on cows that were still alive in the end of study (32.4%) were treated as censored. The probability of being culled (hazard) was defined as a product of a baseline hazard function and a function of explanatory variables. In addition to sire effects, the model included effects of age at first calving and the time-dependent variables herd by year, lactation number, stage of lactation, and milk production within the herd to account for culling because of low production. Solutions for fixed effects indicated a higher probability of being culled for primiparous cows, for cows in the end stage of lactation, and for cows with low production. The impact of censoring on the accuracy of estimation was investigated by computing the rank correlations between the estimated transmitting abilities (ETA) of sires using a simplified model from uncensored data (reference) and the ETA from several different data files with an increased proportion of censored records. The rank correlations among sire ETA decreased as number of daughters per sire decreased and as the proportion of censored records increased. The maximum number of censored records that is acceptable to obtain accurate results is 30 to 40%. The acceptable proportion of censored records would be higher if the reference ETA were obtained on a larger data file using daughters of old sires.