J .A. Archer, L. Bergh Livestock Production Science 65 2000 47 –55
49
2.2. Analyses to the residual errors from the model. Relatively
small numbers of animals in each test meant that it Test durations ranging from 1 to 16 weeks were
was not possible to fit separate regression slopes for considered, with all tests beginning at the test start
each test, but results from Australian data [J.A. date, following a 28-day pre-test adjustment period.
Archer, P.F. Arthur, R.M. Herd and E.C. Richardson, The length of test was progressively increased in
unpublished data] suggest that where conditions are weekly increments by including an extra week of
relatively uniform across different tests, common feed intake and weight data in the data set analysed.
regression coefficients across tests with a fixed effect Growth of each individual was modelled by linear
to adjust for test effects are adequate. regression of weight against time on test, using the
Variance components for each trait and test dura- regression procedure of SAS 1989, and the regres-
tion were calculated from univariate analyses using sion coefficient estimates were used to calculate
ASREML
Gilmour et al., 1998. The model fitted average daily gain ADG during the test, and the
included age and weight at start of test as covariates, mid-weight mean of start and end weights. The
test unique test centre, year and test number combi- linear model fitted to weight data from each separate
nation as a fixed effect, and sire fitted as a random animal included an intercept, a linear term for
effect. A number of records had missing sire in- number of days on test and the residual error term.
formation see Table 1 for numbers missing in each Information on daily feed intake DFI and growth
breed. These records were retained in the data, as was used to calculate two indices of efficiency.
they provided information about the test group Firstly feed conversion ratio FCR was calculated as
effect, but did not contribute to the sire variance total feed intake divided by total gain. The second
component. Dam information was not used in the index of efficiency used was residual feed intake
analyses as there was a limited number of dams with RFI. Residual feed intake was calculated by model-
more than one progeny in the data, and the objective ling daily feed intake within each breed using the
was not to estimate genetic parameters. general linear model procedure of SAS 1989, with
terms fitted in the model including a class variable for each test defined as a unique combination of test
3. Results
centre, year and test number, and mid-weight raised to the power of 0.73 and linear average daily gain
A summary of the performance of the five breeds fitted as covariates. Residual feed intake was equated
over the 112 day test is given in Table 2. In general
Table 2 Summary statistics of bulls at start of test and performance over the 112 day test
Trait Afrikaner
Angus Bonsmara
Hereford Simmental
Start age Mean6S.D.
264623 239626
259624 254628
251626 days
Range 176–325
172–307 174–338
156–330 167–317
Start weight Mean6S.D.
231622 263625
275626 264627
318628 kg
Range 169–296
179–349 172–400
158–340 210–415
ADG Mean6S.D.
1.2360.17 1.7060.26
1.6160.24 1.5760.25
1.8360.23
21
kg day Range
0.65–1.89 1.08–2.63
0.67–2.68 0.75–2.53
0.98–2.68 DFI
Mean6S.D. 8.8461.08
11.1161.25 10.5261.31
10.1061.24 11.9161.20
21
kg day Range
5.38–12.23 7.87–15.43
5.88–15.06 6.38–13.91
8.16–15.84 FCR
Mean6S.D. 7.2860.90
6.6060.73 6.6260.78
6.5260.76 6.5760.72
21
kg kg Range
5.43–12.48 4.65–9.56
4.15–10.91 4.44–11.34
4.26–9.66 RFI
Mean6S.D. 0.0060.50
0.0060.50 0.0060.67
0.0060.59 0.0060.64
21
kg day Range
2 1.49–2.76 2 2.47–1.83
2 3.76–3.72 2 2.35–2.28
2 3.27–2.56
50 J
.A. Archer, L. Bergh Livestock Production Science 65 2000 47 –55
the performance and amount of variation in Angus, al. 1988 showed that pre-test environment and
Bonsmara, Hereford and Simmental were similar for selection effects can be important when using post-
most traits, while the performance of Afrikander weaning performance test data for genetic evalua-
tended to be lower than the other breeds. The breed tion. However, the sire variance component esti-
means for residual feed intake cannot be compared, mated here represents real biological variation in the
as the within breed regression meant that all means traits caused by genetic and possibly some pre-test
were zero, and rankings of animals are only relative environmental effects which has been separated
to other animals in the same population. from the residual variation which in turn includes
The magnitude of variance components for the measurement error among other things. Therefore the
five breeds considered is shown for ADG, DFI, FCR response in the variance components to increasing
and RFI in Figs. 1–4, respectively. The variance test duration is appropriate for assessing the duration
components are not intended to be used as genetic required to minimise measurement error.
parameter estimates, as any pre-test environmental The residual variance for ADG decreased up to 42
effects which may not have been removed by the test days for Afrikaner, Angus, Bonsmara and Hereford,
entry criteria are partially confounded with sire with no marked reduction occurring after this point.
effects. In addition, the bulls tested represent a The residual variance appeared to take slightly
selected sub-set of the South African cattle popula- longer to stabilise for Simmental, where approxi-
tion for each breed, and this selection has not been mately 56 days were required. Sire variance com-
accounted for in the variance estimates. De Rose et ponents had already stabilised by the time the
Fig. 1. Response in variance components of ADG to increasing test duration.
J .A. Archer, L. Bergh Livestock Production Science 65 2000 47 –55
51
Fig. 2. Response in variance components of DFI to increasing duration of test.
residual variance stabilised, and were of a similar the non-sensical estimates obtained was considered
magnitude to the estimate obtained for a 112-day to be a result of the poor distributional properties of
test. FCR when the denominator total gain is close to
For DFI, the response in residual variance com- zero or negative for some animals. This situation is
ponents to increasing test length was not as marked likely to occur when short tests are used, as the daily
as for growth rate. The residual variance appeared to variation in weight is large compared to the weight
stabilise after approximately 56 days on test for gain over a short period. Similar anomalies for FCR
Afrikaner, Angus, Bonsmara and Hereford, and after measured using short test durations were observed by
70 days for Simmental. However, as the response Archer et al. 1997. The results for FCR suggest
curves were fairly flat, it could be argued that that residual variances stabilised after approximately
shortening the test further from these durations e.g. 70 days in Afrikaner, Angus and Simmental al-
to 42 days would result in only marginal losses in though there was a perturbation at 77 days for
accuracy, and the savings by using a shorter test Afrikaner, and 84 days in Bonsmara and Hereford.
duration might be justified. The sire variance com- Residual variances for RFI stabilised after approx-
ponents for DFI were stable after around 21 days on imately 70 days for Afrikaner, Angus and Bonsmara,
test. and after 84 days for Simmental and Hereford.
Variance components for FCR measured using However, as with DFI the response curve for RFI
shorter tests behaved poorly, although they appeared was relatively flat, and so shortening the test duration
to stabilise for longer tests Fig. 3.. The reason for to as little as 49 days would result in only minor
52 J
.A. Archer, L. Bergh Livestock Production Science 65 2000 47 –55
Fig. 3. Response in variance components of FCR to increasing duration of test.
losses in accuracy for all breeds. Sire variance variation and other variation, including measurement
components stabilised after 49 days for most breeds, error. Heritability of the measured trait from differ-
with the exception of Afrikaner which took 56 days. ent test lengths has been used as a criterion in other
studies e.g. Liu and Makarechian, 1993b. This approach recognises that the variance of the test
4. Discussion outcome reflects both real biological differences
