THE SIZE AND COST OF THE ESTIMATION ERROR WHEN USING ANALYSIS OF A SAMPLE OF A DAIRY-HERD TO ASSESS THE WHOLE HERD PERFORMANCE

Whilst computerised herd health schemes have been available for almost 20 years in the UK and they have been shown to increase profitability by pound 45-65 per cow per year their commercial uptake has been disappointing. Previous surveys have shown that uptake is limited due to poor motivation amongst farmers who fail to see either the need or the benefits. In order to demonstrate the potential benefits to a farmer a certain amount of data analysis is required but manual data analysis is limited by the size of commercial herds and the amount of data involved, The analysis of a sample of the herd would be possible but the accuracy of the results is not known. Five hundred repeat samples of 10, 20 and 40 cows were taken from six large commercial herds (average N = 218, range 175-245), and the 95% confidence interval for fertility and disease parameters calculated. The estimation error (with 95% confidence limits) about the mean was expressed in financial terms using standard costs for performance shortfalls, For samples of 20 cows the estimation error averaged pound 39.20 for ten parameters over the six herds. For a sample size of 40 cows, the average was pound 26.39. The results indicate that sampling 40 cows from large commercial herds (about 200 cows) and 20 cows from smaller herds (approx. 100 cows) will allow the cost of poor performance to be estimated sufficiently accurately to determine whether further computer-assisted investigation is warranted. The errors were greatest for the annual culling rate and it is suggested that this be calculated for the whole herd.