Sample Data set

Evaluation of growth rate in two indigenous sheep breeds of Ethiopia

The data set, background text and SAS programme runs were provided by Dr Markos Tibbo, ILRI Addis Ababa, Ethiopia. Mr James Audho and Drs Okeyo Mwai and Julie Ojango, ILRI, Nairobi, Kenya, prepared the GENSTAT runs and the rest of the texts.

Background information

Growth in animals can be measured by the increase in live weight over time. Different factors tend to have differential influence on growth depending on the period of growth the animal is undergoing. For example, early growth in lambs is influenced by the breed, sex of lamb, litter size born in, seasonal fluctuation in feed availability, hence the milk yield of the dam. The growth rate of sheep could be improved through selection, provided there are records available on a reasonable number of animals over several years.

Sheep from two indigenous sheep breeds of Ethiopia, the Menz and the Horro were reared at the ILRI Debre Berhan research station and their performance recorded from 1992 to 1997. Data obtained from this station will be used to illustrate the steps and procedures followed in statistical analysis of field (livestock) data.

Possible breeding goal:

To have indigenous sheep flocks that are well adapted to the local environment as well as productive (i.e. good reproductive performance and fast growth rates), hence have high off-take rates and be profitable to the producers.

Research questions

The following general research questions could be raised:

1. What environmental factors have the greatest influence on growth rate of sheep in Ethiopia?

2. Is growth rate at different ages the same for Horro and Menz sheep?

3. At what stage of growth does the dam have the greatest influence?

4. What proportion of the variation in growth can be attributed to the genes the individual animals posses (inherited from its parents)?

Source of Data

Two indigenous sheep breeds of Ethiopia, the Menz and the Horro, were reared at the ILRI Debre Berhan research station and records maintained on their performance from 1992 to 1997. The animals were retained as pure-breds. To avoid haphazard mating, different sexes of animals were grazed in separate paddocks. Lambing was planned to occur either in June at the beginning of the wet warm season (long rains from July to September) or in October at the onset of the dry cold season (November to January). During the breeding seasons, single sire mating groups of 20-25 ewes per ram were maintained. Within each mating season, ten rams from each breed were used. Details on the animal management are described by Tibbo (2006).

Data collection

Data used was collected over a period of five years (1992 to 1997). Records maintained on individual animals included weights at birth, 3 months and 12 months, and dates of birth, lambing and mating. The data collected was maintained in a [Microsoft Access] database.

The fields in the database are animal id, computer generated newid, breed, sex, birth date, birth weight, season of birth, sire id, computer generated new sire id, dam id, computer generated new dam id, litters id (lambs from the same dam within parity receives similar id), birth type (1, 2, 3), mating group, year of birth, weaning date, weaning weight, age at weaning date, pre-weaning average daily weight gain (ADG1), yearling date, yearling weight, age at yearling date, post-weaning average daily weight gain (ADG2), and mortality date [Sheep Data1], [Sheep Data-Raw], [Sheep Data Analysis1].

Data Exploration

Data collected must first be explored and errors in data entry checked. Data is then categorized into groups, and various distributions (scatter plots, histograms etc.) plotted to examine any patterns within the data. Observations noted outside the general pattern are checked to ensure they are not errors in the data [Biometrics example 1].

Raw statistics are then obtained on each variable of interest. Two statistical packages, SAS and Genstat were used to illustrate the steps and procedures. Links to the two sets of files are made for ease of reference (See Module 4, Section 2) [SAS] and [GENSTAT programmes].

The summary of the results from the runs of the two input programs are presented in Tables 1 to 4 to illustrate the point that both programs give similar results.

The 1^st step is to run the descriptive statistics, then use the results (see Table 1a) to clean the data (see SAS program). Results of subsequent analysis are presented in Table 1b. [SAS and GENSTAT Programmes]

Table 1a: SAS and GENSTAT outputs showing descriptive statistics for raw data on birth weight (BWT), weaning weight (WWT) and yearling weight (YWT) of Sheep from Debre Berhan

Table 1b: SAS and GENSTAT outputs showing descriptive statistics for edited data on birth weight (BWT), weaning weight (WWT) and yearling weight (YWT) of Sheep from Debre Berhan

Once one is sure of the quality of data at hand, then further analyses of various traits of interest can be performed. For example, one can analyse the factors affecting weights of animals at different stages of growth such as: birth weight, weaning weight and yearling weight. These are termed dependent variables (See Module 4, section 3.1). The next step is the model building step, where both dependent and independent variables are specified. The models used in this example include the fixed effects of breed (two levels); sex (two levels); birth type (three levels); dam parity (6 levels); season (two levels) and year of birth (five levels). Age of lamb at weaning was fitted as linear covariate to adjust for variation that might arise from imposition of a weaning date to every animal weaned older or younger than the assumed 90 days,(See Module 4, Section 3.2).Similarly age of lamb at yearling was fitted as linear covariate to adjust for such variations. Also see the relevant sections in the [SAS] and [GENSTAT] Program files.

The results from these analyses are presented in Tables 2, 3 and 4 as extracted from the outputs of the SAS and GENSTAT programmes.

Statistics	SAS				GENSTAT
	BWT	WWT	YWT		BWT	WWT	YWT
Overall Mean	2.45	10.57	17.85		2.45	10.58	17.85
Model DF	18	19	18		18	19	18
Residual DF	4343	3102	1748		4343	3102	1748
Residual SS	857	13107	17026		856.6	13107	17026
F ratio	144.74	83.15	81.57		144.74	83.15	81.57
Significance level	<.0001	<.0001	<.0001		<.001	<.001	<.001
R-Square	0.38	0.34	0.46		0.37	0.33	0.45

DF = Degree of Freedom
SS = Sum of Squares

	BWT			WWT			YWT
Effect	LSM	% FIT	Signif- level	LSM	% FIT	Signif- level	LSM	% FIT	Signif- level
Breed		8.43	* * *		2.53	* * *		3.71	* * *
Horro	2.39			9.78			18.49
Menz	2.05			8.95			16.66
Sex		0.97	* * *		0.43	* * *		6.17	* * *
Female	2.17			9.20			16.52
Male	2.28			9.53			18.63
Birth Type		22.02	* * *		13.5	* * *		5.03	* * *
Single	2.73			11.22			19.0
Twins	2.16			9.04			16.58
Triplets	1.77			7.84			17.13
Age		-			0.08	Ns		3.2	* * *
Parity of Dam		10.53	* * *		2.71	* * *		1.58	* * *
Year season of birth	See Fig.1	4.74	* * *	See Fig.1	20.6	* * *	See Fig 1	24.6	* * *

NS = not significant
*** = Significant at 0.001

	BWT			WWT			YWT
Effect	LSM	% FIT	Signif -level	LSM	% FIT	Signif -level	LSM	% FIT	Signif -level
Breed		5.76	* * *		1.87	* * *		4.19	* * *
Horro	2.64			11.04			18.35
Menz	2.30			10.22			16.65
Sex		0.89	* * *		0.37	* * *		7.16	* * *
Female	2.40			10.40			16.43
Male	2.51			10.72			18.57
Birth Type		20.11	* * *		9.56	* * *		3.15	* * *
Single	2.63			11.13			19.02
Twins	2.06			8.94			16.58
Triplets	1.67			7.74			16.90
Age		-			0.08	NS		3.2	* * *
Dam Parity		6.0	* * *	See Fig.1	1.29	* * *		1.47	* * *
Year season of birth	See Fig.1	4.74	* * *	See Fig.2	20.6	* * *	See Fig.1	26.5	* * *

NS = not significant (P>0.05)
*** = Significant at 0.001

Figure 1 Least Squares means for various traits over different Year-Seasons of birth

In the subsequent analyses breed was dropped from main effects since the analyses were done separately by breed. In this case, sire was fitted as random effect and lamb weaning age as linear covariate. [SAS and GENSTAT programmes]

The results obtained at this point indicate that there exists significant between and within breed variation for growth in the two indigenous breeds. However, to determine what proportion of the variation is due to genetic composition of the breeds, subsequent genetic analyses which enable, estimation of the variance components, covariances and genetic parameters (heritabilities and genetic correlations) for growth using programs such as WOMBAT, AIREML, ASREML, VCE, PEST, R and DMU are performed.

Exercises

Based on the above information:

1. Find out and indicate key reasons as to why cleaning of data was necessary?
2. Prepare 2 Tables (as Table 4) for SAS and GENSTAT outputs by breed.

References

Tibbo M. 2006. Productivity and health of indigenous sheep breeds and crossbreds in the central Ethiopian highlands. PhD dissertation. Department of Animal Breeding and Genetics, Faculty for Veterinary Medicine and Animal Sciences, Swedish University of Agricultural Sciences (SLU), Uppsala, Sweden.Available at: http://diss-epsilon.slu.se/archive/00001142/01/Markos_Tibbo_corrected.pdf