Livestock Research for Rural Development 19 (8) 2007 Guide for preparation of papers LRRD News

Citation of this paper

Characterization of smallholder dairy production systems for livestock improvement in Kenya highlands

L M Mburu, J W Wakhungu* and W G Kang'ethe**

Ministry of Livestock and Fisheries Development, P.O. Box 47010-00100, Nairobi-Kenya
leonardmburu@yahoo.com
*University of Nairobi, Department of Animal Production, P.O. Box 29053-00625, Nairobi- Kenya
**University of Nairobi, Department of Agricultural Economics, P.O. Box 29053-00625, Nairobi- Kenya
 

Abstract

Principal components analysis and cluster analysis were used to classify smallholder dairy farms in terms of risk management strategies, level of household resources, dairy intensification and access to services and markets in Kenya highlands. Four clusters of smallholder dairy systems were identified. Cluster 1, 2, 3 and 4 had 11.9%, 11.2%, 35.1% and 41.8% of households respectively. Cluster 1 had majority of farmers (56%) in lower highlands and no farmers in upper midlands. In cluster 2, majority of farmers (40%) were in lower midlands. Cluster 3 had majority of farmers (62%) in upper midlands. In cluster 4, majority of farmers (50%) were in lower highlands.

Characterization of smallholder dairy production systems in Kenya highlands is critical in understanding the constraints and opportunities that exist within the farming systems. It allows better targeting of dairy improvement research and development. Therefore, information obtained can be valuable for detailed analysis of constraints and opportunities found in smallholder dairy systems and to design policies and strategies to support smallholder dairy development programs in Kenya highlands under differing intensification one has to be aware of the challenges.

Key words: Access to services and markets, cluster analysis, dairy intensification, level of household resources, risk management strategies


Introduction

Developing appropriate interventions to assist smallholder dairy households, and identifying those which should be targeted requires a clear understanding of the dairy systems. Characterization is the grouping of farmers with similar practices and circumstances for whom a given recommendation would be broadly appropriate (Byerlee et al 1980). World livestock production systems were classified and characterized based on consideration of socio-economic and agro-ecological factors into two basic types namely solely livestock and mixed farming systems (Sere and Steinfeld 1995). Dairy production systems are considered a subset of the farming systems (Wilson 1994). In western Niger, cluster and discriminant analysis were used to classify crop-livestock producers in three villages into four recommendation domains using a combination of production and marketing variables (Williams 1994).

Several studies have been executed to characterize farming systems in Eastern Africa. In the Lake Crescent Region of Uganda, a variety of peri-urban smallholder dairy farm types were identified (Fonteh et al 2005). The most common (representing about 70%) was characterized by limited land availability (< 2acres), located at the outskirts of town (between approximately 5 and 10 Km away from town) and five or less cows.

Using principal components analysis and cluster analysis characterized dairy systems supplying the Nairobi milk market into four recommendation domains based on level of dairy intensification, farm/household resources and access to services and markets (Staal et al 1998; Staal et al 2001). In 1998 characterization, the four main domains of farmers distinguished were the informal resource poor, the cooperative resource poor, the elite and the specialists while in 2001 the four main domains of farmers distinguished were the informal resource poor, the intensive part time, the extensive landed and specialists farmers were distinguished.

Milk production systems in Kenya vary widely with breeds of animals used, intensity of land and labor use and feeding systems in Kenya (Wakhungu 2001; Muriuki et al 2003). This necessitate the need to characterize the smallholder dairy production systems in Kenya highlands for livestock improvements based the level of intensification of the farm dairy system, risk management strategies, level of access to output markets and input services, and farm / household resources available.
 

Methodology

Description of study site and data collection

The study used conceptual framework for dairy systems analysis of production-to-consumption approach developed by ILRI (Rey et al 1993). Data was collected from December 2004 to March 2005 through a survey questionnaire in Central Province located in Kenya highlands in three agro-ecological zones: Lower highlands, Upper midlands and Lower midlands (Jaetzold and Schmidt 1983). Primary data were collected through personal interviews by trained enumerators using a survey questionnaire covering measures from resources to parameters reflecting farm functioning from one hundred and thirty four households with at least one dairy cow at the time of survey. All information collected referred to the situation of the day before the survey.

Sampling procedure

Purposive multi stage design using Probability Proportion to Size (PPS) sampling design was used. Three agro-ecological zones: Lower highlands, Upper midlands and Lower midlands (Jaetzold and Schmidt 1983) were chosen purposively. Within the agro-ecological zones, eight research locations were selected based on household density: low, medium and high. Locations with a higher population size (CBS 2001) had a proportionately higher sample size in the survey. In order to capture as much local variations as possible, the sample in each zone was spread across the 27 sub-locations among farms selected as randomly as possible. In some, if the farmer could not be reached or did not wish to participate in the study, another one in the locality was substituted.

The sample size was obtained from estimating the number of observations potentially needed to distinguish between the three agro-ecological zones by a difference of 30% in some of the important farm/household variables. Assuming a desired confidence interval of 95%, and using a coefficient of variation of 68%, which was the observed co-efficient of variation of households in Kiambu dairy herd from previous studies (Kaguongo et al 1997); a minimum sample size of 40 in each agro-ecological zone was calculated (Poate and Daplyn 1993).

The calculation of sample size in each stratification class, to estimate a difference, was based on the equation:

Where:

n = minimum sample size,
z = 1.96 for 95% confidence interval,
c = Coefficient of Variation,
d = Level of difference [Poate and Daplyn 1993].

The chosen sample required then 14 observations in each location. However, in order to maintain proportionality, the number of observations in each location was adjusted to reflect the proportion of the number of households, resulting in sample sizes of 6 to 28 in each location. After maintaining a minimum of 10 observations in each location, the total sample size obtained was 134 households (or 0.07 % of the households in Kiambu district).

Identification of principal components

In order to distinguish characteristic patterns of dairy activity existing among households in Kiambu district, a clustering method was applied to some of primary variables. The method uses principal component analysis followed by cluster analysis. Principal components analysis is used in survey research in data reduction without omitting potentially important information (Mick 1990).

In principal components analysis, factors are extracted sequentially, such that the first accounts for the maximum common factor variance across all variables (i.e. more than any other factor that could be extracted). Thereafter, a second factor (reference factor) is then extracted which is at right angles to the first factor (i.e. orthogonal to it) such that the maximum amount of the common factor variance remains (Mick 1990). In the process the apparently most important variation from a larger set of variables are identified and then used to cluster the household observations. Typically factors are extracted as long as the latent roots (e.g. Eigen values) are greater than one. If less than one, they can be alternatively chosen by reference to significant gaps between them. Thereafter, based on these rules, the chosen principal components are rotated then orthogonally to improve interpretability. Only loadings above 0.30 or below -0.30 should be considered as significant (Mick 1990).

In the second step, farm/ households are then scored along the new vectors, and those created are used in standard cluster analysis. Since the variables were standardized in the analysis to have mean 0 and 1 variance, a correlation coefficient or weighting of 1, indicates strong correlation, 0 is neutral and -1 shows strong negative correlations. Therefore, negative means indicate levels lower than the overall sample mean.

Selection of variables used in principal component analysis

The groups of variables used in the principle component analysis that might distinguish between clusters were selected apriori. The themes chosen were the level of intensification of the farm dairy system, risk management strategies, level of access to output markets and input services, and farm / household resources available. These four themes thus formed the conceptual framework used in the principal component analysis and cluster analysis. For each theme a set of variables considered to reflect the primary measures of variability within that theme were chosen (Table 1).


Table 1.   Means of variables used in principal components analysis and cluster analysis classified according to clusters

Description of variables

Lower

highlands

Upper

midlands

Lower

midlands

Overall

Level of intensification of the dairy system

 

 

 

 

Price of milk, KES /kg

17.4

17.4

20.5

16.8

Distance to market, km

1.92

5.83

1.71

1.62

Maximum distance between farms, km

1.21

4.57

1.17

1.04

Milk marketing channel, 1=cooperative and other informal channels, 0=informal channels only

0.75

0.8

0.00

0.928

Member of dairy cooperative, 1= Yes, 0=No

0.88

0.87

0.13

0.98

Dairy cooperative source of information, 1=Yes, 0=No

0.69

0.07

0.13

0.55

Risk management strategies

 

 

 

 

Leases land, 1=Yes, 0=No

0.25

0.266

0.404

0.16

Number of farms cultivated

2

1.45

1.91

1.45

Cost of home grown fodder, KES / day

22.8

9.18

5.08

6.72

Access to credit, 1=Yes, 0=No

0.88

0.53

0.11

0.7

Number of animals left the farm per year

0.94

0.55

0.52

0.636

Level of access to output markets and input services

 

 

 

Household head age, years

58.3

58

50.3

52.4

Household head works off-farm, 1=Yes, 0=No

0.063

0.53

0.68

0.36

Total farm area under household care, acres

6.71

3.54

1.7

2.07

Total household income from off-farm, KES /month

1,438

3,967

6,347

4,019

Farm/ household resources

 

 

 

 

Weight of concentrates per TRLU, kg/day

4.11

2.12

1.98

2.61

Cost of purchased fodder per TRLU, KES /day

1.14

1.68

2.54

3.7

Total household farm size per TRLU, acres

2.74

1.19

0.95

0.81

Napier grass planted per TRLU, acres

1.13

0.37

0.34

0.297

Cost of milk output per TRLU, KES/ day

127

89.1

111

95

Source: Estimations from the survey data collected in 2004/2005 by the authors

The principal components analyses were carried out on the four variables using the data from the 134 household observations. Finally, each of the 134 households was given a score along the new variables generated that consisted of the sum of the products of the weightings and their scores along the original variables. Anderson-Rubin method of estimating factor score coefficients that ensured orthogonality of the estimated factors was used. The scores produced had a mean of 0, a standard deviation of 1, and are uncorrelated (Mick 1990). However, the recipients of cluster solutions should always be wary about the validity of the clusters, as cluster analysis is not based on stochastic foundations.
 

Results

Variables used in principal component analysis

These four themes that formed the conceptual framework used in the principal component analysis and cluster analysis the level of intensification of the farm dairy system, risk management strategies, level of access to output markets and input services, and farm / household resources available. There were considerable variations between the themes (Table 1).

Principal components analysis by level of access to services and markets

Six principal components selected to indicate level of access to services and markets (Table 1) in the principal components analysis yielded two factors with an Eigen value greater than 1, which explained 63.4% of the variation in selected variables. Provision of extension services by cooperatives is a proxy for availability of cooperative services. Cooperative membership is a proxy for access to both input and output markets.

Depending on the level of weighting, factors 1 and 2 defined new variables arbitrary called COOPPART (Cooperative participation) and MKTDIST (Distance to market) respectively (Table 2).


Table 2.  Rotated correlation co-efficient factor pattern level of access to services and markets

Description of variables

Components

Factor 1

COOPPART

Factor 2

MKTDIST

Price of milk, KES / kg

-0.843

-0.0815

Distance to market from farm, km

0.0545

0.849

Milk marketing channel, 1=cooperative and other informal  channels, 0=informal channels only

0.919

0.0757

Member of dairy cooperative, 1= Yes, 0=No

0.888

0.106

Dairy cooperative source of information, 1=Yes, 0=No

0.511

-0.443

Maximum distance between farms, km

0.0814

0.497

Source: Estimated from the survey data collected in 2004/2005 by the authors

Principal components analysis by level of risk management strategy used

Five principal components selected as important measures of risk management strategies (Table 1) yielded two factors with an Eigen value greater than 1, which explained 65.7% of the variation in selected variables. Risk management strategies are important as they affect the level of farm profit. Private management of risks can occur at two levels through income and consumption smoothing (Murdoch 1995). Depending on the level of weighting, factors 1 and 2 defined new variables arbitrary called CONSMOOT (Consumption smoothing) and INCSMOOT (Income smoothing) respectively (Table 3).


Table 3.  Rotated correlation coefficient factor pattern level of risk management strategy

Description of variables

Components

Factor 1

CONSMOOT

Factor 2

INCSMOOT

Leases land, 1=Yes, 0=No

0.898

-0.14

Number of farms cultivated

0.917

0.0686

Cost of home grown fodder, KES / kg

-0.0946

0.71

Access to credit, 1=Yes, 0=No

-0.17

0.686

Number of animals left the farm per year

0.251

0.732

Source: Estimations from the survey data collected in 2004/2005 by the authors

Farmers can smooth income the flow of income to the household through making conservative production choices combining production enterprises that generate returns during different times of the year, and diversifying economic activities. They can also smooth income by borrowing and saving; depleting and accumulating non financial assets, including livestock; undertaking migration; and relying on implicit of informal insurance arrangements. These mechanisms take force after shocks occur and help insulate consumption patterns from income fluctuations.

Principal components analysis by level of household resources

Four principal components selected as important measures of household resources (Table 1) yielded one factor with an Eigen value greater than 1, which explained 47% of the variation in selected variables. Income from off-farm employment is important to dairy intensification through their effects on increasing working capital. One factor is not rotated and depending on the level of weighting, factor 1 defined a new variable arbitrary called HHEADXS (Household head characteristics) (Table 4).


Table 4.   Un-rotated correlation co-efficient factor pattern of level of household resources

Description of variables

Component

Factor 1 HHEADXS

Household head age, years

0.817

Household head works off-farm, 1=Yes, 0=No

-0.586

Total farm area under household care, acres

-0.535

Total household income from off-farm employment, KES / day

0.764

Source: Estimations from the survey data collected in 2004/2005 by the authors

Principal components analysis by level of dairy intensification

Five principal components selected as important measures of dairy intensification (Table 1) yielded two factors with an Eigen value greater than 1, which explained 69.3% of the variation in selected variables. Depending on the level of weighting, factors 1 and 2 defined new variables arbitrary called ONFARMFO (On-farm fodder production) and OFFARMFO (Off-farm fodder production) respectively (Table 5).


Table 5.   Rotated correlation co-efficient factor pattern level of dairy intensification

Description of variables

Components

Factor 1

ONFARMFO

Factor 2

OFFARMFO

Weight of concentrates per TRLU*, kg/day

0.199

0.833

Cost of purchased fodder per TRLU*, KES / kg

-0.442

0.426

Total household farm size per TRLU*, acres

0.88

0.0794

Napier grass planted per TRLU*, acres

0.867

0.34

Cost of milk output /TRLU*, KES

0.101

0.851

Source: Estimations from the survey data collected in 2004/2005 by the authors
*
TRLU (Tropical Ruminant Livestock Unit): A standard unit by which livestock of

different species are compared. It is defined as an animal with the equivalent of 250 kg

live weight, e.g. camels- 1.0 TRLU, cattle- 0.7 TRLU (ILCA 1993).


Discussion

Four clusters were identified using principal components analysis and cluster analysis (Table 6).


Table 6.  The means for each variable within each final cluster, the frequency of households in each cluster and the significance levels (F)

Factors/ Clusters

1

2

3

4

F

Sig

Consumption smoothening

0.124

-0.149

0.33

-0.272

3.47

0.018

Income smoothening

1.45

0.00425

-0.619

0.104

28

0.000

Total farm income

-1.15

-0.201

0.485

-0.0244

14.3

0.000

Cooperative participation

0.539

0.303

-1.199

0.771

178

0.000

Distance to market centre

-0.332

2.17

-0.187

-0.328

65

0.000

On-farm produced fodder

1.95

0.0869

-0.192

-0.42

52.5

0.000

Off-farm produced fodder

0.529

-0.363

-0.116

0.0428

2.48

0.064

Frequency of households

16

15

47

56

 

 

Percentage cases

11.9

11.2

35.1

41.8

 

 

Lower highlands, %

56.3

26.7

14.9

50.0

 

 

Upper midlands, %

0.000

33.3

61.7

19.6

 

 

Lower midlands, %

43.7

40.0

23.4

30.4

 

 

Source: Estimations from the survey data collected in 2004/2005 by the authors

The F tests should be used only for descriptive purposes because the clusters have been chosen to maximize the differences among cases in different clusters. The observed significance levels are not corrected for this and thus cannot be interpreted as tests of the hypothesis that the cluster means are equal. Each cluster had unique constraints and opportunities, which helped define research priorities based on opportunities and constraints. Appropriate interventions should consider variations in all factors of production, and the relationships and patterns among the clusters.

Cluster 1 had 11.9% of the households (Table 6) with majority of farmers (56%) in lower highlands. Upper midlands had no farmers in this cluster. The cluster recorded highest levels of risk management strategy through income smoothing characterized by dairy farmers who relied on own produced fodder as they had largest parcels of land; use of on-farm and off-farm produced fodder and lowest total farm income. The cluster had high reliance on dairy cooperative as a source of information by households and composed of old farmers, majority of whom worked on-farm (Table 1).

Cluster 2 consisted of 11.2% of households (Table 6) with majority of farmers (40%) in lower midlands. It was characterized by furthest distance to the nearest market centre and between various parcels of land (Table 1). These farmers did not rely on off-farm produced fodder.

Cluster 3 contained 35.1% of households with majority of farmers (62%) in upper midlands (Table 6). Households in this cluster managed risks either through consumption or income smoothing. They marketed their milk through the informal market channels only and had lowest cooperative participation. Also, majority of household heads worked off-farm and had lowest mean age and total farm area (Table 1).

Cluster 4 which contained 41.8% of households with majority of farmers (40%) in lower highlands (Table 6) characterized by consumption smoothing as a risk management strategy through high cooperative participation and lowest reliance on on-farm produced fodder and. This cluster had least distance to nearest market centre and distance between farms and highest cooperative participation, which was characterized by lowest milk prices. Due to small farm sizes (Table 1) they had highest cost of purchased fodder per TRLU.
 

Conclusions

Acknowledgements

This study was supported by research funds from Deutscher Akademischer Austausch Dienst (DAAD). The assistance of Liston Njoroge (International Livestock Research Institute) in statistical analysis is also greatly acknowledged.
 

References

Byerlee D, Collinson M, Perrin R, Winklemann D, Biggs S, Moscardi E, Martinez J, Harrington L and Benjamin A 1980 Planning Technologies Appropriate to Farmers: Concepts and Procedures. CIMMYT El Batan Mexico 71 pp.

CBS 2001 The 1999 Population and Housing Census: Counting Our People for Development Volume 1. Ministry of Planning and National Development, Central Bureau of Statistics Nairobi, Kenya: 1-5.

Fonteh F A, Mubiru S, Tibayungwa F and Lammers W 2005 System analysis of peri-urban smallholder dairy farming in the Lake Crescent Region of Uganda. Livestock Research for Rural Development. Volume 17, Article #79. Retrieved March 22, 2007, from http://www.cipav.org.co/lrrd/lrrd17/7/font17079.htm

Go K 2005 Project Document: National Agriculture and Livestock Extension Program, Phase II July 2005- June 2010. NALEP II Report No. 1. Ministry of Agriculture and Ministry Livestock and Fisheries Development, Government of Kenya, Nairobi, pp 45.

ILCA 1993 Handbook of livestock statistics. International Livestock Centre for Africa, Addis Ababa, Ethiopia. 66 pp http://www.ilri.org/InfoServ/Webpub/Fulldocs/X5482e/x5482e00.htm

Jaetzold R and Schmidt H 1983 Natural conditions and farm management information: Farm management Handbook of Kenya. Volume II. Part B, Central Kenya, Ministry of Agriculture, Nairobi: Pp 510.

Kaguongo W N W, Staal S and Ackello- Ogutu C 1997 Smallholder adoption of dairy Technologies: Effect of risk with management strategies in a liberalizing economy. In: Fungoh P O and Mbadi G C O (Editors) Focus on Agricultural Research for Sustainable Development in a Changing Environment: Proceedings of the 5th Kari Scientific Conference, Nairobi, Kenya. Pp 325-337.

Mick A 1990 Exploring hyperspace: A non-mathematical explanation of multivariate analysis. McGraw-Hill Book Company, London.139 pp.

Murdoch J 1995 Income Smoothing and Consumption Smoothing. Journal of Economic Perspectives 9(3): 103 -114.

Muriuki H, Omore A, Hooton N, Waithaka M, Ouma R, Staal S and Odhiambo P 2003 The policy environment in the Kenya dairy sub-sector: A review. SDP Research and Development Report No. 2. Smallholder Dairy (R and D) Project: Nairobi, Kenya, Pp 21.

Poate C D and Daplyn P F 1993 Data for agrarian development. Wye Studies in Agricultural and Rural Development, Cambridge, Cambridge University Press.

Rey B, Thorpe W, Smith J, Osuji P, Mullins G and Agyemang K 1993 Improvement of dairy production to satisfy consumer demand in sub-Saharan Africa: A conceptual Framework for Research. Market- oriented Smallholder Dairy Research Working Document No. 1. International Livestock Centre for Africa (ILCA), Addis Ababa, Ethiopia: Pp 13. http://www.fao.org/wairdocs/ilri/x5465e/x5465e00.htm

Sere C and Steinfeld H in collaboration with Groenwold J 1995 World Livestock Production systems: Current Status, Issues and Trends: In: Gardiner P and Devendra C (Editors) Global agenda for livestock research, Proceedings of a Consortium. ILRI, Nairobi, Kenya, 18-20 January 1995, ILRI (International Livestock Research Institute), Nairobi, Kenya pp11-39 http://www.fao.org/WAIRDOCS/LEAD/X6101E/X6101E00.HTM

Staal S J, Chege L, Kenyanjui M, Kimari A, Lukuyu B, Njubi D, Owango M, Tanner J, Thorpe W and Wambugu M 1998 Characterization of dairy systems supplying the Nairobi milk market: A pilot survey in Kiambu district for the identification of target groups of producers. KARI / MoA / ILRI Collaborative Research Project Report. Nairobi, Kenya: Pp 32.

Staal S J, Owango M, Muriuki H, Kenyanjui M, Lukuyu B, Njoroge L, Njubi D, Baltenweck F, Musembi F, Bwana O, Muriuki K, Gichungu G, Omore A and Thorpe W 2001 Dairy systems characterization of the Nairobi greater milk shed. SDP Research Report, KARI/ MoARD /ILRI/DFID Collaborative Research Project Report, Nairobi, Kenya: Pp 40.

Wakhungu J W 2001 Dairy cattle breeding policy for Kenya smallholders: An evaluation based on a demographic stationary state productivity model. PhD thesis, University of Nairobi: Nairobi, Kenya, Pp 189.

Williams T O 1994  Identifying target groups for livestock improvement research: The classification of sedentary livestock producers in western Niger. Agricultural Systems 46: 227-237.

Wilson T 1994 Integrated Livestock and Crops for Sustainable Use and Development of Tropical Agricultural Systems. AGSP-FAO, pp 84-92.



Received 27 March 2007; Accepted 10 June 2007; Published 6 August 2007

Go to top