Previous Page Table of Contents Next Page


Considerations in the design of on-farm livestock experiments and evaluation of results

G.I. Mlay and N.A. Urio
Departments of Rural Economy and Animal Science and Production, Sokoine University of Agriculture, P.O. Box 3004, Morogoro, Tanzania


Abstract
Introduction
Extraneous factors to be considered in on-farm livestock experiments
Statistical considerations
Experiences with dairy feeding systems project: Hai District-Tanzania
Materials and methods
Proposals for evaluating results when statistical methods are invalid or inadequate
References

Abstract

The paper identifies major extraneous factors which can have a significant effect on the variation of response variables in on-farm livestock research. The implications of such factors on the design of experiments and evaluation of results are discussed. Important statistical considerations are presented and their importance in guiding experimental designs demonstrated. The paper uses a case study based on a feeding trial of dairy cattle under small farms to illustrate the design, management and evaluation problems resulting from the extraneous factors. A case is put forward to support the use of farmer-evaluation and economic analysis as additional approaches to statistical methods in evaluating research results.

Introduction

On-farm research can be considered as an intermediate step between station level research and development. Interest in on-farm research has arisen out of difficulties encountered in the large-scale extension of technical innovations developed under controlled research environment. The rate of adoption of such technologies has been particularly disappointing in the case of small farmers (Jouve and Mercoiret, 1987).

The movement of a research process from a controlled laboratory environment to farm conditions entails additional considerations in experimental design and the evaluation of research results. Not only does the process introduce more extraneous factors whose control may be difficult, but an additional dimension, the farmer, has to be considered. Arbodela (1987) identifies the researcher, technology, the farmer and farmer's values as important components in on-farm research. While the researcher makes his recommendations based on an objective evaluation of the results arising from the methodology used in the research, the farmer makes a decision about the recommendation on the basis of an overall assessment of the research in the light of his values (Arbodela op. cit.).

Devendra (1987) identifies the following shortcomings based on experiences from Asian countries: "that methodologies presently used are haphazard, lack sophistication and control and are unimaginative; the nature of the work does not allow for statistical analyses of the results; and ad-hoc innovations are often imposed in the hope of demonstrating causes and effects, usually in quantitative terms". These shortcomings are a reflection of the special problems encountered at farm level which make a direct transfer of on-station research approaches difficult or very costly to implement.

This paper highlights the problems of experimental design of dairy cattle feeding trials and proposes alternative evaluation criteria where formal designs cannot be implemented.

Extraneous factors to be considered in on-farm livestock experiments

Inferences made about population parameter variables measured in experiments can be seriously affected depending on how extraneous factors are handled in a design. If a researcher completely ignores extraneous factors, then the variation that cannot be accounted for by treatments will be lumped together as error. Since the error mean square is involved in computing test statistics, this may lead to wrong inferences. In addition when significant extraneous factors are ignored, the usual assumption of homoscedastic error term is no longer valid and hence variance estimates and test statistics computed from such data will be biased. Therefore the usual practice is to attempt to stratify the experimental units on the basis of non-experimental factors which are thought to show significant variation across the experimental units. These are subsequently separated from the error sum of squares components in the analysis.

In on-farm livestock research, where weight gain or milk yield are the key response variables to be measured, the extraneous factors which the researcher will have to contend with may include: location (when the area being covered shows large climatic variation), breed, time, age, stage of lactation, lactation number and differences in the level of management across farmers.

The larger the number of extraneous factors to be handled in an experiment, the more complex the design becomes and the requirements in terms of resources also increase. In addition, interpretation of results become difficult particularly when interactions of higher orders are involved.

Statistical considerations

In order to apply statistical methods in data analysis and to make statistical inferences from the results, the design of experiments must satisfy the following conditions (Anderson and Mclean, 1974):

- The inference space must be defined. These are the limits within which the results will apply. The definition of an inference space will determine a relevant sample size to use.

- The experimental units must be randomly selected. Random selection is necessary to protect against bias in the experiment which could be the result of some unknown factor having had prior influence on the experimental units in some systematic fashion. When the experimental units turn out not to be homogeneous then stratification based on some inherent characteristic(s) will be necessary.

- Assignment of treatments to experimental units likewise must be random.

With the assistance of a statistician, a mathematical model evolving as a result of the problem to be studied, factor levels to be used in the experiment and the conditions listed above needs to be written down. The mathematical model will give rise to the ANOVA table which at this stage will consist of degrees of freedom, and expected mean squares for each of the specific factors selected. The expected mean squares will provide information on the various factors which will have tests available. The researcher can review the ANOVA table and if some of the assumptions and conclusions implied by the table are not realistic or practical, the design will have to be changed.

- Extreme care in data collection is essential. Other factors given, the success of a scientific investigation depends upon the validity of all data obtained.

In the light of the farm conditions under which livestock trials have to be conducted, it is clear that either some of the conditions listed above will be violated or high costs will have to be incurred in terms of both financial and human resources in an attempt to satisfy the stated conditions.

Experiences with dairy feeding systems project: Hai District-Tanzania

Smallholder farmers in Hai District keep an average of 4 heads of cattle per household. These are mainly cross breeds (Zebu crossed with exotic dairy cattle) and a few pure breeds, mainly Jersey cattle. In a diagnostic survey conducted in 1984, one of the factors identified as constraining smallholder dairy production was the availability of adequate feeds in terms of quantity and quality. After some assessment of feeds situation in the district, it was decided to test the impact of introducing bean haulms/chaff and molasses-urea mixture in basal rations on milk yield under smallholder farmer conditions.

The problems that arose with regard to the design of experiments were (a) several levels of a factor could not be implemented within a household; (b) the dairy cattle being kept varied in breeds/crosses and age, and were at different stages of lactation. Often the farmers were unable to provide an accurate history of their cattle; (c) the basic management, including housing and feeding varied across households; (d) the fact that the project site is 500 km from the home base meant that the day-to-day management of experiments and keeping of records was to be left under the control of farmers. That the success of the experiments depended on the willingness and the ability of the farmers to manage the trials meant that purposive selection of farmers and hence the cattle (experimental units) was unavoidable.

Ideally the experiments would need to satisfy the statistical considerations mentioned earlier and the blocking for the above extraneous factors would be necessary. If these were satisfied, a linear model of the following form would be specified:

Yijklmno = A + Bj + C (j) j + Dk + El + Fm + Gn + Ho + (Interactions) + Eijklmno

Where

Yijklmno is milk yield of the jth cow of the kth breed, in 1th lactation, managed by the mth farmer in nth location, fed ith ration on the oth day.

A is the overall mean

Bi is the effect of the ith ration on milk yield

C(i) j is the effect of the jth cow on the ith ration

Dk is the effect of the kth breed

El is the effect of the 1th lactation

Fm is the effect of management by mth farmer

Gn is the effect of nth location

Ho is the effect of the oth day

Eijklmno is the error term

The model as presented is still too basic. Considerations by an animal scientist on important interactions are essential, and thereafter, practicality of implementing the experiment will need to be considered. Where restrictions are necessary, these will have to be introduced in the model as restriction errors since these have an important implication on the resulting ANOVA table and the tests which will be subsequently available.

The actual feeding experiments conducted did not satisfy the criteria mentioned, and therefore subjecting the data to statistical analysis and drawing inferences from that basis is not warranted. The design adopted reflected practical considerations in the project site and resources at the disposal of the researchers. In addition, researchers were satisfied that the trade-off between design quality and developmental effects of the research was worthwhile in the initial stages of the work.

Materials and methods

Participating farmers were purposely chosen from three villages: 10 from Ng'uni, 5 from Mowo-Njamu and 5 from Wandri. Discussion was held with the farmers on the objectives of the experiments and the tasks they were expected to perform. Information was sought on the history of the cattle they kept. For practical reasons, the farmers were to continue with their usual feeding routines, only that supplementation with molasses-urea mixture sprinkled on bean haulms was introduced. Molasses-urea mixture and bean haulms were provided at cost. Daily feeding of the latter two feeds was at the rate of 2 kg and 8 kg per animal respectively. Farmers were asked to record, on a daily basis, milk yield, types of basal feeds being used and types and quantities of other concentrates fed. The experiment was continued for a period of 6 months.

Analysis and evaluation of results

The design used suffered from the following weaknesses:

(a) The effects on milk yield of other diets whose feeding varied within and across households were not controlled or accounted for the design.

(b) The design lacked control treatment.

(c) Only one level of the factor was considered.

As a result of the above weaknesses, it was not possible to assess the treatment effects on milk yield through statistical methods.

A subsample of the data collected is used here to illustrate the importance of controlling or accounting for extraneous factors when designing on-farm trials. The results were based on one-way analysis of variance by cow, farmer, breed, village (location) and ration. The results are presented in Tables 1 to 5. It is shown that with the exception of location (village), all other factors have significant effects on the variation of daily milk yield at 0.01 probability level.

Table 1. Average daily milk yield by cow.

Cow no.

Mean yield (litres)

Standard deviation

No. of records

1

5.6

0.8

35

2

2.2

0.5

35

3

4.0

0.6

30

4

10.3

1.3

30

5

11.4

1.4

28

6

7.2

0.8

21

7

8.9

1.5

21

8

3.1

0.5

23

9

7.2

1.0

35

10

10.2

1.5

33

11

5.2

1.0

18

12

7.8

0.8

20

13

7.5

1.3

27

Table 2. Average daily milk yield per cow by 9 farmers.

Farmer

Mean yield (litres)

Standard deviation

Samples size

1

3.9

1.8

70

2

8.5

3.5

88

3

8.1

1.5

42

4

3.1

0.5

23

5

5.2

1.0

18

6

7.8

0.8

20

7

7.5

1.3

27

8

7.2

1.0

35

9

10.2

1.5

33

Table 3. Daily average milk per cow by breed.

Breed

Mean yield (litres)

Standard deviation

Sample size

Jersey

8.6

3.2

153

Friesian

7.3

1.0

168

Crosses

5.4

2.2

35

Table 4. Average daily milk yield per cow in two villages.

Village

Mean yield (litres)

Standard deviation

Sample size

Ng'uni

7.0

3.2

291

Mowo-Njamu

6.9

1.5

65

Table 5. Average daily milk yield per cow by ration.

Ration

Mean yield (litres)

Standard deviation

Sample size

Only molasses-urea

5.7

2.1

33

Cottonseed cake plus molasses-urea

7.4

3.2

248

Wheat pollard plus molasses-urea

6.0

2.1

75

Proposals for evaluating results when statistical methods are invalid or inadequate

From experience in the Dairy Feeding Systems Research Project, it is proposed that farmer-evaluation of the research, and economic evaluation of benefits and potentials for development should always be either in conjunction with statistical methods, or on their own when statistical methods are invalid.

Farmer-evaluation is important particularly when it is considered that recommendations from on-farm research are expected to be a basis for development of technologies for adoption. As pointed out earlier, farmer-evaluation takes into account other dimensions which are not handled by statistical methods. A survey carried out at the end of the experiment which covered both participating and non-participating farmers, showed that the research was addressing an important problem and that there had been a positive impact on milk yield performance. The results of the survey are summarised in Tables 6 to 10.

An economic analysis of on-farm research can take several forms. The simplest analysis is that restricted on costs and returns. In order to examine input-output relationships and resource use efficiency of research trials a production function approach can be used. In the cases where a farmer is involved in several enterprises (crops and livestock) budgeting and linear programming techniques can be used to arrive at optimum combination of enterprises and employment of various resources. Mdoe (1986) using a multiperiod linear programming model was able to demonstrate the effects of alternative dairy production technologies on optimum enterprise combinations and farm incomes in Hai District.

Table 6. Impact of the research project on dairy management.

Management practices

Number of farmers

Ng'uni

Mowo-Njamu

Wandri

Total

No change in management

1

0

0

1

Record keeping

8

3

2

13

Use of molasses

7

5

4

16

Measuring milk production

3

2

2

7

Chopping of maize stover

1

0

1

2

Increase use of crop residues

1

0

0

1

Pasture management

0

1

0

1

Table 7. Dairy cattle performance.



Percentage of farmers

Ng'uni

Mowo-Njamu

Wandri

Performance remained the same

0

20

0

Performance improved

100

80

100

Performance declined

0

0

0

Table 8. Ng'uni vallage: milk yield before and during the project period (litres).

Farmer number

Yield before project

Yield during project

Change in yield

1

6.5

8.0

1.5

2

4.0

5.0

1.0

3

4.5

5.0

0.5

4

4.0

5.5

1.5

5

6.0

7.0

1.0

6

4.0

4.5

0.5

7

3.0

4.0

1.0

8

5.0

6.0

1.0

9

4.0

5.0

1.0

Total

41.0

50.0

9.0

Mean

4.5

5.5

1.0

S.D.

1.10

1.3

0.4

Table 9. Mowo-Njamu: Milk yield before and during project period (litres).

Farmer number

Yield before the project

Yield during the project

Change in yield

1

6.0

7.0

1.0

2

7.0

8.0

1.0

3

5.0

6.5

1.5

4

3.0

4.0

1.0

5

3.0

4.0

1.0

Total

24

29.5

5.5

Mean

4.8

5.9

1.1

S.D.

1.8

1.8

0.2

Table 10. Wandri village milk yield before and during the project period (in litres).

Farmer
number

Yield before
the project

Yield during
the project

Change
in yield

1

7.0

8.5

1.5

2

5.0

6.0

1.0

3

5.5

6.5

1.0

4

3.5

4.5

1.0

Total

21.0

25.5

4.5

Mean

5.3

6.4

1.1

S.D.

1.4

1.7

0.3

References

Anderson, V.L. and Mclean, R.A. 1974. Designs of experiments: A realistic approach. Marcel Dekker, Inc., New York. 418 pp .

Arboleda, C.R. 1987. Methodological and institutional considerations in applying statistical approaches to on-farm animal research. In: Proceedings of a workshop on on-farm research/extension and its economic analysis, 1923 January 1987. Southeast Asian Regional Centre for Graduate Study and Research in Agriculture, Los Banos, Laguna, Philippines. pp. 77-73.

Devendra, C. 1987. The relevance of on-farm animal production research in Asia. In: Proceedings of a workshop on on-farm research/extension and its economic analysis, 19-23 January 1987, Southeast Asian Regional Centre for Graduate Study and Research in Agriculture, Los Banos, Laguna, Philippines. pp. 13-18.

Jouve, M.P. and Mercoiret, M.R. 1987. Research and development: A method of putting farming systems at the service of rural development. Paper presented at the Agrarian Systems Seminar, 19 May 1987, Montpellier, France.

Mdoe, N.S. 1986. An economic analysis of alternative dairy feed management systems in the highlands of Kilimanjaro, Tanzania. M.Sc. thesis, University of Guelph, Ontario, Canada.


Previous Page Top of Page Next Page