The aim of this research was to evaluate and compare four methods for the selection of
animals for future breeding, in flocks/herds where animals are not uniquely tagged. Tagging of
animals requires money and technology; both of which are difficult to come by in many
rangeland systems in developing countries, or even hill farms in the UK. These methods are
seen as an introduction to simplified recording, with the view to farmers then becoming more
integrated into breeding schemes that operate in their areas.
Each of the four methods calculates threshold value(s), and those animals that fall above
or on the threshold value can be selected for future breeding. The threshold values take into
account the top x% (% to Select) that the farmer wishes to select. The four methods were
evaluated using example data sets for three traits, namely Muscle Depth (MD) (with 4 data sets
for years 1991-1994), Milk Yield (MY) (with 4 data sets for farms A-D) and Ovulation Rate
(OR) (with 4 data sets for years 1988-1989 and 1991-1992) over% to Selects of 5, 15, 25, 35
and 45. To obtain threshold values, method 1 used a sample of the data set; method 2 used
historical data; method 3 used historical data (as in method 2) along with the first record of the
data set as its starting threshold value and then updated the threshold value each time a new
animal was recorded; method 4 used the sample threshold value from method 1 along with the
first record of the data set as its starting threshold value and then updated the threshold value as
each animal's record was reached.
An experiment using two procedures (A and B) was run to find an appropriate sample
size to be used for methods 1 and 4. Method A was derived from a published formula and
focussed on accuracy. Method Bused the example data sets to test whether the true mean of the
data set could be predicted consistently in random samples. On examination of sample size
range 5 - 50%, it was concluded that a sample size of 10% was appropriate for the traits to be
studied in this thesis.
The initial experiment of the methods showed that for MD and MY, method 4 had a
higher success rate (SR) than all other methods: 0.78 and 0.86 for MD and MY, respectively. A
% to Select> 5% resulted in better SR values. For OR, method 2 achieved the highest SR (0.73)
but the OR data was highly skewed and the results should therefore be treated with caution. The
most notable finding for historical effect was for MY farm D, where high historical mean
compared to the data set resulted in poor SR (0.19).
The second experiment concentrated on modifying methods 2 and 3 alone; renamed 2M
and 3M. To alleviate the problem of extreme values for historical data, the data was replaced by
using the minimum and maximum values of the actual data sets for the calculation of the
threshold values. MY SR results showed that methods 2M and 3M significantly improved on
methods 2 and 3 by 12 and 3%, respectively (Psignificantly differ from methods 2 and 3. For OR SR, methods 2M and 3M significantly
decreased SR compared to methods 2 and 3 by 49% and 4%, respectively (PSelect > 5% resulted in better SR values for MD and MY, while for OR, a % to Select of 5
resulted in a higher SR.
The third experiment concentrated on OR data only. The OR values were transformed
using log(IO) (L) and square root( ) as a result of the OR data not being normally distributed,
with methods renamed with a L or ✓ after the number. Transformation resulted in a wider range
of results. Methods 2L and ✓ significantly decreased SR compared to method 2 by 26% and
31%, respectively (PIL increased on method 1 by 3%. SR for methods 3L, ✓, 4L and ✓ all increased SR compared
to methods 3 and 4, by 9, I, 12 and 10%, respectively. % to Select for ✓ methods can be
flexible, but for L methods, % to Selects ~ 15 are less favourable.
The fourth experiment concentrated on MD and MY data only. The data were corrected
for environmental effects, with methods renamed with an E after the number. Correcting for
environment effects significantly affected method 2E, which decreased SR compared to method
2 by 17% (PConsistently good SR (>0.78 for MD and MY) can be obtained using method 4. Method
4 could be used as the standard with the possibility of adjusting for environmental factors if
information on environmental factors is available. For a low-tech method, methods 1 or 2 could
be used if the sampling/historical values are reliable and/or available. All methods should be
used with caution for OR data. Potential application of the methods is discussed.