| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
LETTERS TO THE EDITOR |
Lilly Hall of Life Sciences, 915 W. State Street, Purdue University, West Lafayette, IN 47907-2504, e-mail: aschinck{at}purdue.edu
Several issues in the article by Johnson et al. (2004)
, "Evaluation of procedures to predict fat-free lean in swine carcasses," J. Anim. Sci. 82:2428–2441, deserve attention. The first issue is the use of the regression of the residual values on the actual values as an indicator of bias. Regardless of the true magnitude of bias of the equation (St-Pierre, 2003), the expected value of the regression of residual values on actual values is (1 – r2). Equations with four levels of measurement errors were simulated, produced different levels of biases, and accounted for 100, 55, 32, and 20% of the total variation in fat-free lean mass between genetic populations (Schinckel et al., 2000). The values of regressions of the residuals on the actual values were all equal to 1 – r2 of each equation.
The magnitude of the biases must be compared with the actual genetic population, gender, or treatment differences. If the equations are to be used in the future, it is important to know what percentage of the total variation among the genetic populations is expected to be predicted (Gu et al., 1992; Schinckel et al., 2001). Johnson et al. (2004) did not present either the mean fat-free lean values or other statistics to evaluate the relative magnitude of the genetic population biases. For example, the 3.43-kg overprediction of Berkshire pigs with equations using carcass weight and backfat depth must be compared with the amount by which Berkshire pigs differ from other pigs.
The second issue is that the extent to which the carcass soft tissue was separated into fat-free lean and carcass fat by chemical (lipid extraction) or by physical (dissection) means is not clearly stated. Based on past results (Wagner et al., 1999), 30% of the total carcass fat-free lean is contained within the dissected ham and loin muscles, and 70% is contained within the other soft tissue mixtures. The level of dissection completed in this trial is far less than most previous trials used to determine fat-free lean mass (Schinckel et al., 2001). A related issue is the discussion of lipid-free lean on page 2437 of the article. Schinckel et al. (2001) evaluated the lipid-free soft-tissue mass of pigs, not the lipid-free lean mass, which was indicated by Johnson et al. (2004) as the lipid-free mass of the dissected lean. All but one of the other papers cited by Johnson et al. (2004) included the lipid-free mass of both the lean (muscle) and fat (adipose) tissue. Lipid-free soft-tissue mass and fat-free lean mass are different measures of carcass composition, with different marginal growth rates (Schinckel et al., 2001).
The third issue is the use of data over a broad range of BW. The relationships between carcass component mass and dependent variables change and become more nonlinear as the range of BW increases (Wagner et al., 1999; Schinckel et al., 2001). In past trials (Gu et al., 1992; Schinckel et al., 2001), BW range biases were evaluated as the mean residual values of pigs of specific BW ranges. Prediction equations developed over larger BW ranges can have a greater magnitude of BW range, BW x sex range, and genetic population x BW range biases (Schinckel et al., 2001).
The trial reported by Johnson et al. (2004) evaluated different sets of genetic populations of pigs at different BW ranges. If pigs of different genetic populations are evaluated at different BW, the relationships between the dependent and independent variables and among the independent variables can become strangely nonlinear. Much of the emphasis on the quadratic and cross-product terms is likely because different genetic populations were evaluated at different BW ranges.
The fourth issue is that the equation for the Ultrasonic Fat-o-Meater (UFOM) seems to be unrealistic. Measurement errors tend to be a constant percentage of the backfat and loin depth measurements (Boland et al., 1995). At the average HCW reported by Johnson et al. (2004), the prediction equation for the UFOM predicts a 0.76-kg increase in fat-free lean mass as UFOM backfat depth decreases from 10 to 9 mm, and only a 0.64-kg increase in fat-free lean mass as backfat depth decreases from 36 to 28 mm. At a constant HCW, the change in fat-free mass per change in backfat (kg/ mm) is b = – 1.04 + 0.030 UFOM backfat, mm, which is 0 at 34.66 mm; at >34.66 mm, the quantity of lean increases as backfat increases.
A final issue is the cause of the variance component attributable to slaughter day of measurement. If the measurement of fat-free lean mass varied from day to day, the magnitude of the variance caused by day would be nearly identical for each equation; however, in Johnson et al. (2004) the magnitude of the variance because of day varied among equations when the same fat-free lean observations were analyzed. Johnson et al. (2004) stated that this result could be partially caused by the existence of day-to-day measurement errors on the independent variables. Measurement errors, whether on a daily basis or on an individual observation basis, result in biased prediction equations (Neter et al., 1996; Schinckel et al., 2000). The regression equations assume that the random effect of day is independent of the values of the independent variables. Nonetheless, the data for these equations came from two serial slaughter experiments (target BW of 113 or 131.5 kg in one trial and 113, 131.5, and 150 kg of BW in another trial) and another two trials with pigs at close to 114 kg of BW. If the random effect of day is partially confounded with another factor (BW range, genetic population, or sex), the random effect of day will absorb some of the variation because of the factor.
 |
Footnotes |
|---|
|
Literature Cited |
|---|
 Top Literature Cited |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
