Effect of missing data on the estimate of average daily feed intake of growing pigs.

The effect of missing records for feed intake per day (FID) on the estimate of average daily feed intake (DFI) during the test period of individual pigs was studied. Data from 192 growing pigs tested with single-space computerized feeding stations during an average of 93 d were used. True DFI was computed by averaging FID records per pig, individually. A first- and third-degree polynomial and a nonlinear function were fit to FID records per pig to estimate DFI by averaging estimated FID records per pig, individually. The three functions showed small differences for goodness of fit. Missing FID records were simulated by random as well as period-wise deletions of FID records. The effect of missing FID records was judged on the Pearson correlation between true and estimated DFI. Deleting randomly up to 70% of FID records per pig before fitting each function reduced this correlation only from 1.00 to .96 for each function. Deleting 25 successive FID records (approximately 27% of records) before fitting reduced the correlation to values ranging from .92 to .96 and from .59 to .96 for the first- and third-order degree polynomial function, respectively, and from .80 to .97 for the nonlinear function. Using iteratively reweighted least squares regression methods to exclude undesirable effects of outlier values gave similar results for the effect of missing FID records on estimated DFI. Results imply that considering incorrect FID records as missing is a good alternative for adjusting incorrect data in combination with using functions to estimate DFI of growing pigs. Use of a first-degree polynomial function is recommended. Moreover, use of functions enables a more efficient use of feeding stations by recording feed intake data during only parts of the test period.