Climate change adaptation and livestock activity choices in Kenya: An economic analysis

Authors


Abstract

This paper examines the impact of climate change on the decision of farmers to engage or not to engage in livestock activities and also on the choice of different livestock species in Kenya. To this end, cross-sectional household level data supplemented by long-term averages of climate data are used. The probit model is employed to derive the response of the probability of engaging in livestock activities to climate change. Probit and multivariate probit methods are employed to model the choice of different livestock species. Atmosphere–ocean global circulation models are used to project the impact of different climate scenarios on the probability of engaging in livestock activities and also of adopting different livestock species according to variations in climate. The results suggest that farmers adapt livestock management decisions to climate change. At low levels of temperature increase, the probability of engaging in livestock activities falls, but at higher levels of climate change, the probability rises. The results further show that as it gets hotter, farmers change their livestock choices from dairy cattle and sheep to beef cattle and goats.

1. Introduction and motivation of the study

Kenya has climate and ecological extremes, with altitude varying from sea level to over 5,000 m in the highlands. The mean annual rainfall ranges from less than 250 mm in the arid and semi-arid areas to 2,000 mm in high potential areas. Out of a total area of 580,367 square kilometres, only 12% is considered high potential for farming or intensive livestock production. A further 6%, which is classified as medium potential, mainly supports livestock, especially sheep and goats. The other 82% of the total land is classified as arid and semi-arid lands (ASALs) and is largely used for extensive livestock production (ranching and pastoralism), and wildlife. It is estimated that the ASALs support about 25% of the nation's human population and slightly over 50% of its livestock. The country can be divided into seven agro-climate zones using a moisture index based on annual rainfall expressed as a percentage of potential evaporation (Sombroek et al., 1982). Areas with an index greater than 50% have high potential for cropping and are designated zones I, II and III (Table 1). These zones account for about 18% of Kenya's land area. The semi-humid to arid regions (zones IV, V, VI and VII) have indexes of less than 50% and a mean annual rainfall of less than 1,100 mm. These zones are generally referred to as the Kenyan range-lands (the ASALs). Ninety percent of the arid and semi-arid areas lie below 1,260 m and mean annual temperatures range from 22 °C to 40 °C.

Table 1. Characteristics of agro-climate zones and farming systems in Kenya
ZoneMoisture index (%)Climate classificationAverage annual rainfall (mm)Average annual potential evaporation (mm)VegetationFarming system
  1. Source: Sombroek et al. (1982); Jaetzold & Schmidt (1982).

I>80Humid1,100–2,7001,200–2,000Moist forestDairy, sheep, coffee, tea, maize, sugarcane
II65–80Sub-humid1,000–1,6001,300–2,100Moist and dry forestMaize, pyrethrum, wheat, coffee, sugarcane
III50–65Semi-humid800–1,4001,450–2,200Dry forest and moist woodlandWheat, maize, barley coffee, cotton, coconut, cassava
IV40–50Semi-humid to semi–arid600–1,1001,550–2,200Dry woodland and bush landRanching, cattle, sheep, barley, sunflower, maize, cotton, cashew nuts, cassava
V25–40Semi-arid450–9001,650–2,300Bush landRanching, livestock, sorghum, millet
VI15–25Arid300–5501,900–2,400Bush land and scrublandRanching
VII<15Very arid150–3502,100–2,500Desert scrubNomadism and shifting grazing

The vulnerability of agriculture in general and livestock production in particular to climate change is an important issue in many African countries as they are highly dependent on the sector. In Kenya, agriculture contributes a substantial share to national income, export earnings, rural employment and non-farm enterprises; both rural and urban populations depend on agriculture for food security. However, over the last three decades, the relative contribution of the agricultural sector to the livelihoods in many African countries including Kenya has continued to decline. Multiple factors are responsible for this decline, including poor initial resource endowments and endogenous factors such as population growth among others.

There is ample evidence that the temperature is increasing (IPCC, 2001). The most significant damages of climate change are predicted to occur in the agricultural sector in sub-Saharan Africa given the semi-arid nature of a large portion of the continent and the fact that the region already endures high temperatures and low precipitation, frequent droughts and water scarcity. In Kenya, global circulation models predict that, by the year 2100, climate change will increase temperatures by about 4 °C and cause variability of rainfall by up to 20%. From these predictions, the two extreme climate events that are expected to adversely affect the agricultural sector are drought and flooding in both the ASALs and the high potential areas. The ASALs will however bear the most substantial impact of global warming (Kabubo-Mariara and Karanja, 2007; see also Galvin et al., 2001).

Previous literature on the link between climate change and livestock production presents evidence that climate can affect livestock both directly and indirectly. Direct effects impinge on animal growth, animal products and reproduction. Indirect effects influence the quantity and quality of feedstuffs such as pasture, forage, grain and the severity and distribution of livestock diseases and parasites (Seo and Mendelsohn, 2006a). Extreme climate events have had devastating effects on the livestock sector in Kenya in the past, leaving marginal groups without a source of livelihood. Adaptation options and other strategies for farmers to reduce the adverse impact of climate change are therefore crucial, particularly in the ASALs. However, climatic uncertainty and drought as well as environmental changes driven by shifts in land use patterns have seriously diminished the ability of pastoral communities to cope using traditional strategies (Kabubo-Mariara, 2005a). According to Swinton (1988), drought is a regular phenomenon in semi-arid climates. Households located in such areas are chiefly concerned with ensuring their survival and may seek to minimize risks and also to limit losses following a production failure. Risk minimization strategies in the face of adverse climate conditions may include among others, crop cultivation and mixed crop livestock enterprises (Luseno et al., 2003; Sherlund et al., 2002; Little et al., 2001; Smith et al., 2001; Swinton, 1988). Loss management may involve sale of livestock and other assets, migration with livestock or even introduction of more drought-resistant livestock species (Swinton, 1988; Campbell, 1999; Kabubo-Mariara, 2003, 2005a, 2005b; Kazianga and Udry, 2004).

Though there are a number of studies on the impact of climate change on crop agriculture in developing countries, there is limited literature on its impact on livestock production (see Seo and Mendelsohn 2006a, 2006b, 2006c). A number of studies in Kenya have investigated the response to land pressure and drought focusing on ASALs (Campbell, 1999; McPeak, 2003; Kabubo-Mariara, 2005a) but there is a dearth of literature on the impact of climate change on the livestock sector. In addition, adaptive mechanisms used by livestock farmers to circumvent the impact of climate change have not been studied in Kenya. This paper addresses these research gaps and examines the impact of climate change on the choice decision of farmers to engage or not in livestock activities, and upon choosing to engage, the choice of different livestock species. The paper argues that choice of species is a form of adaptation of livestock management to climate change. The paper further simulates the expected effect of various long-term climate change scenarios on future livestock choice decisions. Understanding the impact of long-term climate change on livestock choice is crucial for future livestock policies and interventions in Kenya. Of particular interest are interventions that address the livestock sector in particular and sustainable livelihoods in general.

The rest of the paper is organized as follows. Section 2 presents the methods, section 3 discusses the study site and data; section 4 presents the results and section 5 concludes.

2. Theory and methods

Livestock production is a choice of household's resource allocation and a livelihood diversification strategy. In this paper, we first focus on the determinants of the decision of the household to enter or not to enter livestock production. Second, we focus on the determinants of choice of livestock species. Following Seo and Mendelsohn (2006a) we start by assuming that a livestock farmer chooses the outputs and inputs that maximize net revenue subject to the prices, climate and other external factors that he or she faces. The farmer must first determine whether or not it is profitable to engage in livestock production and also choose which species to adopt.

Suppose the profit from managing livestock is given by inline image where X is a vector of regressors composed of climate and other socio-economic factors. The disturbance, ɛ, is unknown to the econometrician but may be known to the farmer (the farmer is more likely to choose an animal that is most profitable (Seo and Mendelsohn, 2006b), but the cumulative distribution function (CDF) is a function f(ɛ) that is known up to a finite parameter vector.

The profit maximizing farmers will then choose to keep livestock if inline image or ɛ < . The probability that this occurs, given X, is P(ɛ < ) =F(). The likelihood function can be defined as:

image(1)

If F(ɛ) is a standard logistic CDF, then the probability can be defined as

image(2)

If instead we assume that ɛ is IN(0, δ2), the decision to hold livestock can be estimated using a probit model (Maddala, 1995). If we define yi as a binary response variable, the probit model can be defined as:

Pi(yi≠ 0 | X) =F() (3)

The farmer then compares the profits from different species in order to choose which one to adopt. Different approaches can be used to model choice of livestock species (Seo and Mendelsohn, 2006a). A primary animal model assumes that the only choice of importance to the farmer is the species that earns the greatest net revenue on the farm. The farmer must consequently choose a single primary animal from the list of available species. A portfolio model examines all possible combinations of species that a farmer can choose. This model treats specific combinations of species as distinct choices. These two approaches are not suitable for the Kenyan case because the list of choices for both must be mutually exclusive and the farmer can select only one choice. An alternative approach is to estimate a system of demand equations for each animal. The farmer determines whether a species is profitable. The more profitable the species, the more likely it is that the farmer will adopt it. The choices in the system of equations are not mutually exclusive and farmers can select more than one species. This system of equations can be estimated using multivariate probit.

We can extend equation (3) to the multivariate model by introducing a binary response of farmer i on animal j denoted as Yij. The collection of responses on all j animals is Yi= (Yi1, ... , Yij). The probability that Yi=yi conditioned on parameters β, Σ, and a set of covariates xij, is given by

image(4)

Where φj(t | 0, Σ) is j-variate normal distribution with mean vector 0 and correlation matrix Σ={σjk} and Aij is the interval

image(5)

Equation (5) presents a system of probit equations for each species and estimation accounts for possible correlation across errors in the regressions. The alternatives are not mutually exclusive and the sum of probabilities is greater than one. The demand system approach is however best suited to the case when the choice of each species is independent of the choice of others (Seo and Mendelsohn, 2006a). In Kenya, the assumption of independence of alternative choices may not hold due to resource constraints and also due to the need to minimize risks and diversify livelihoods, which make farmers hold different species of livestock (Kabubo-Mariara, 2005a, 2005b, 2007). Given these restrictions and data limitations, this paper employs both the univariate and multivariate probit approaches to model the choice of each livestock type.

If we define z as a vector of climate variables, the marginal effect of climate change on the probability of holding livestock can be defined as:

image(6)

3. The data

3.1. Household data

The primary data for this study were based on a sample of 816 households in Kenya. The data were collected from 38 districts drawn from six out of eight provinces in Kenya between June and August 2004. The districts chosen captured variability in a wide range of agro-climatic conditions (rainfall, temperature and soil), market characteristics (market accessibility, infrastructure, etc.) and agricultural diversity, among other factors. Each district was divided into agro-ecological zones and samples of three different farm types/sizes: large (>8 hectares), medium (2–8 hectares) and small (0–2 hectares) chosen from each ecological zone. The sampling procedure was purposely designed to target at least four households from each agro-ecological zone, comprising at least one household from each farm type.

Though this paper is based on livestock production, most of the farmers earn revenue from both crops and livestock. Only 8% of all households in the sample specialized in livestock production, while 12% of the households specialized in crop production. Mixed crop livestock farmers constituted 80% of the sample (722 households). The key household variables of interest for this paper include diversified livestock species held by farmers, costs associated with livestock inputs (including labour) and household characteristics. The data show that households hold a diversified portfolio of animal species, with cattle, chicken, goats and sheep forming the main livestock types. The major livestock types and average endowments by agro-ecological zones are presented in Table 2. The table shows that dairy cattle and chickens were reared by the largest percentage (66) of households. Consequently, the main livestock products were milk and eggs. The data further show that though households located in low potential zones rely more on livestock than those in high-potential zones, there is significant livestock production in high potential zones as well, with dairy cattle, goats and chickens as the most common species. Dairy cattle are much more prevalent in households in high-potential zones than those in low-potential zones, who keep relatively more beef cattle. Though sheep are reared by a relatively small number of households in high-potential zones, these households keep many more sheep than their counterparts in low-potential zones. Some farmers owned additional animals such as camels, donkeys, rabbits, ducks, turkeys, bees and fish. The large standard deviations across all species and zones reflect high inequalities in livestock holding in Kenya.

Table 2. Average livestock holdings by agro-ecological zone
Livestock typeHigh potential zonesMedium & low potential zonesFull sample
% of HouseholdsMeanStd. Dev.% of HouseholdsMeanStd. Dev.% of HouseholdsMeanStd. Dev.
Beef Cattle 9337 965149264,833236953,822
Dairy Cattle293992,961371271,224662472,170
Bulls 6 23  93 9 40 21715 34 180
Goats234912,454194662,412424792,431
Sheep136263,995222271,137353752,597
Pigs 3 10  45 3  4  14 6  7  32
Oxen 6  4  12 9  4   715  4   9
Chicken30 66 26036 27  9066 45 188

3.2. Climate data

In addition to the household data, the study also makes use of satellite and Africa Rainfall and Temperature Evaluation System (ARTES) climate data. The temperature data came from satellites which measure temperatures twice daily via a Special Sensor Microwave Imager mounted on US Defense Department satellites (Basist et al., 1998). The ARTES dataset was interpolated from weather stations by the National Oceanic and Atmospheric Administration based on ground station measurements of precipitation and minimum and maximum temperature (World Bank, 2003). The data were constructed from a base with data for each month of the survey year and for morning and evening. The monthly mean temperatures were estimated from approximately 14 years of data (1988–2003) and the mean monthly precipitation estimated for 1960–1990 to reflect long-term climate change. In the final estimating equations, we use seasonal, wet and dry and annual climate variables. The summary statistics are presented in Table 3. The long and short rains refer to the extended wet and dry conditions respectively. In Kenya, long rains fall between March and May and short rains between October and December. The extended rain seasons are however longer to cover the cropping season. Long rain crops planted in early March are harvested in August. Farms are then prepared and planted in September and the crops harvested in February. In this paper, the long rains season is therefore defined as March to August and the short rains season as September to February.

Table 3. Sample statistics for temperatures and precipitation by season
SeasonTemperatures (°C)Precipitation (mm/mo)
MeanStd dev.MeanStd dev.
Fall (December–February)19.292.67 88.841.45
Summer (March–May)19.072.74103.7131.57
Winter (June–August)18.502.36 62.440.82
Spring (September–November)19.092.66 71.8926.95
Long rains (March–August)19.332.73 90.934.97
Short rains (September–February)18.652.46 81.2723.71
Annual average18.992.58 84.5318.6

4. Empirical results

4.1. Decision to engage in livestock activities

The first aspect of the empirical investigation is the choice of whether or not to engage in livestock production. As noted earlier, 88% of all households in the sample were found to hold livestock. We present two variants of the livestock ownership model: a climate only model, and a model with some household characteristics. First we model the impact of seasonal climate variables on the decision to hold livestock (Table 4). All probit results in this paper present the marginal effects (rather than the coefficients) of each variable on the probability of engaging in livestock production. This is because marginal effects are much easier to interpret than coefficients. All seasonal climate variables except winter temperatures are significant determinants of the decision to hold livestock. High summer and winter temperatures are negatively correlated with the decision to hold livestock. The largest marginal effects are from summer and fall temperatures. Increasing summer temperature by 1 °C would reduce the probability of holding livestock by 0.21%, while a similar increase in fall temperature would increase the probability of holding livestock by 0.13%. The impact of the other variables can be interpreted in a similar manner. When we include household characteristics, only summer temperature is significant. The significant impact of summer temperature implies that households may reduce their stocks when summer temperatures are extreme. Though these results are inconsistent with findings for Africa by Seo and Mendelsohn (2006a), they support literature which argues that farmers in Kenya would dispose of their livestock to reduce risks and also minimize their losses during periods of excessive drought (Campell, 1984; Kabubo-Mariara, 2005a). The location of Kenya at the equator and the subsequent composition of livestock species may also explain the differences in adaptation between Kenya and Africa as a whole.

Table 4. Probit model results of livestock adoption: Seasonal climate variables
VariableClimate Variable ModelAll Variable Model
  1. Notes: Robust z statistics in square brackets. * significant at 10%; ** significant at 5%; *** significant at 1%.

Summer temperature−0.2059−0.1597
[3.23]***[2.62]***
Fall temperature0.12660.0851
[1.67]*[1.18]
Winter temperature−0.00850.0205
[0.14][0.36]
Spring temperature0.08560.0492
[2.19]**[1.28]
Summer precipitation−0.0026−0.0025
[2.54]**[2.50]**
Fall precipitation0.01170.012
[2.48]**[2.56]**
Winter precipitation0.0020.0024
[1.73]*[2.15]**
Spring precipitation−0.0189−0.0204
[2.29]**[2.50]**
Log household size 0.1163
 [3.91]***
Age of household 0.0016
 [2.06]**
Average years of education of household members −0.0023
 [0.74]
Number of observations816816
Wald chi2(*)48.32***63.38***
Pseudo R20.08170.1158

The results for precipitation are more interesting. Summer and spring precipitation are inversely correlated with the decision to hold livestock, but winter and fall precipitation have a positive impact. It would seem as though heavier precipitation during the short rains period (summer and spring spanning September to February) make farmers reduce their livestock and probably turn to crop farming. Heavy rains during the long rains (fall and winter spanning March to August) encourage farmers to keep more livestock due to availability of water, pasture and forage. The marginal effects of precipitation are however quite low. In spite of the insignificance of the temperature variables, the marginal effects suggest that the decision to hold livestock is more sensitive to temperature than to rainfall variations. Introduction of the quadratic terms of the seasonal variables render all coefficients insignificant due to collinearity. No combination of seasonal variables is robust, though precipitation remains significant but with low marginal effects.

The results of the wet/dry climate conditions model are quite surprising. Coefficients for wet conditions (long rains) variables have significant effects when evaluated without the dry conditions (short rains) and vice versa. A combination of the two sets however yields all insignificant impacts. On the other hand, combining the linear and quadratic terms for the two sets yields significant results for some variables. The results are presented in Table 5 (see also Table 6 for overall marginal impacts of climate variables). Though short rain temperatures and long rain precipitation have insignificant impacts, the results are close to the linear seasonal climate results discussed above. The livestock response to short rain temperatures is hill-shaped but the response to long rains temperatures is U-shaped. These results are robust with findings by Seo and Mendelsohn (2006a) using seasonal climate variables, except for significance of variables. Introduction of household characteristics however render most climate variables insignificant.

Table 5. Probit model results of livestock adoption: Wet/dry and annual climate conditions
VariableWet/dry climate conditionsAnnual climate conditions
  1. Notes: Robust z statistics in square brackets. *** significant at 1%.

Short rains temperature0.1767 [0.55] 
Short rains temperature squared−0.0009 [0.51] 
Long rains temperature−0.4893 [1.53] 
Long rains temperature squared0.0028 [1.55] 
Short rains precipitation−0.0497 [4.35]*** 
Short rains precipitation squared0.0003 [4.23]*** 
Long rains precipitation−0.007 [1.12] 
Long rains precipitation squared 0 [1.39] 
Annual temperature −0.3384 [4.17]***
Annual temperature squared 0.0081 [4.11]***
Annual rainfall 0.061 [5.60]***
Annual rainfall squared −0.0004 [5.67]***
Log household size0.1189 [3.84]***0.1206 [3.97]***
Age of household0.0021 [2.58]***0.0021 [2.66]***
Average years of education of household members−0.0027 [0.90]−0.0027 [0.89]
Number of observations722722
Wald chi2(*)64.11***59.22***
Pseudo R20.10870.1056
Table 6. Marginal effects of livestock adoption
 Marginal impactStd. error
  1. Notes: ** significant at 5% level, *** significant at 1% level.

Short rains temperature0.1660.154
Long rains temperature−0.327**0.155
Annual temperature−0.017***0.005
Short rains rainfall−0.002***0.001
Long rains rainfall0.0010.0001
Annual rainfall0.001***0.001

The last model for the decision of whether or not to keep livestock is based on annual climate variables. The results indicate that climate change significantly affects livestock holding decisions. The Chow tests (Wald chi(2)) show that though the overall fit of the model is quite poor, the model fits the data better than the intercept only model. The individual results can be interpreted as follows: a 1 °C change in the linear annual temperature will increase the probability of holding livestock by 0.38, but a similar change in the quadratic value will lead to a 0.009 decline in the probability of holding livestock. Further computations would be needed to derive the total marginal impact of temperature on the decision to hold livestock. All other probit results can be interpreted in the same manner. The results show that all climate variables have significant impacts on the decision to hold livestock. Further, the results also show that the livestock response to annual temperature is U-shaped but the response to precipitation is hill-shaped. Though the coefficients of the climate variables are significant in both models, the marginal effects are almost zero for the quadratic term. The results suggest that the probability of engaging in livestock production decreases up to some threshold with an increase in annual temperatures, then increases, but the reverse effect is observed for precipitation. We discuss this turning point later in the paper. The non-linear relationship between global warming and the decision to engage in livestock production suggests that farmers make adjustments to climate change as global warming rises. The marginal impacts of temperature are much higher than for precipitation.

Introduction of household characteristics increases the overall fit of the model by 4%. The results suggest that larger households are more likely to engage in livestock production than smaller households. Though livestock production may not be labour intensive in pastoral regions, it is quite labour intensive in smallholder farming due to scarcity of pasture. Larger households will therefore be less labour-constrained than their smaller counterparts. Age of the household head is positively correlated with the probability of engaging in livestock production. This could be explained by the fact that most rural dwellers who own land are the more elderly in society and therefore have more resources to keep livestock than their offspring (younger adults). These results are not uncommon in the literature (Kabubo-Mariara, 2005a, 2007; Dercon, 1998; Imai, 2003).

4.2. Choice of livestock species

The second issue is to investigate the choice of livestock species to adopt, upon making the decision to engage in livestock production. Unlike the latter, the analysis of choice of species is based on an annual climate model because no significant impact of seasonal and wet/dry conditions variables was found. The results focus on the decision to hold dairy and beef cattle, goats and sheep, and chickens. First, the results of the probit equation, which are easier to interpret as we report the marginal effects, are discussed. The results (Table 7a) suggest that the models fit the data quite well, and the Wald chi(2) tests show significant results for all individual probits. The overall fit of the models is however poor, which is not uncommon in cross-sectional data. The coefficients for the climate variables are all significant for the choice of cattle and chickens, but not consistent for the other livestock species.

Table 7a. Probit model results for choice of livestock species
 Dairy cattleBeef cattleGoatsSheepChicken
  1. Notes: Robust z statistics in brackets. * Significant at 10%; ** significant at 5%; *** significant at 1%.

Annual temperature−0.79560.28450.04560.157−0.4905
[6.01]***[2.63]***[0.31][1.21][3.74]***
Annual temperature squared0.0181−0.00630.0005−0.00430.0116
[5.73]***[2.45]**[0.13][1.37][3.69]***
Annual rainfall0.11−0.0570.0389−0.02510.1172
[6.15]***[3.59]***[2.02]**[1.36][6.37]***
Annual rainfall squared−0.00070.0003−0.00020.0001−0.0007
[6.43]***[3.68]***[1.87]*[1.11][6.31]***
Log household size0.09710.14940.1610.06780.0862
[1.78]*[3.05]***[2.53]**[1.24][1.55]
Age of household head0.00060.0030.00450.004−0.001
[0.41][2.45]**[2.89]***[2.76]***[0.72]
Household's average years of education0.0074−0.0031−0.0125−0.01030.0117
[1.50][0.65][2.20]**[2.02]**[2.31]**
Number of observations722722722722722
Wald chi2(7)58.1546.2596.7737.5237.40
Pseudo R–squared0.070.060.110.040.05

The results suggest that global warming affects the choice of livestock species kept by households. The results are however complex and differ from one species to another. The results further show that choice of livestock species is more responsive to temperature than to precipitation (see also Table 7b). The probability of holding dairy cattle exhibits a U-shaped relationship with annual temperature, but a hill-shaped relationship with precipitation. The marginal effects of individual variables suggest that temperature changes are the key drivers of the decision to hold dairy cattle. The results are consistent with Seo and Mendelsohn (2006a) who find an inverted U-shaped response of dairy cattle to summer temperature. Household size has a positive impact on the probability of holding dairy cattle but we uncover no significant impact of age and education variables. The impact of climate variables on the probability of holding beef cattle is the reverse of the impact on dairy cattle. The probability of keeping beef cattle exhibits a hill-shaped relationship with annual temperatures, while annual precipitation exhibits a U-shaped relationship. A unit increase in annual temperature would increase the probability of selecting beef cattle by 0.015, but the impact of rainfall is almost negligible. The marginal effect on choice of dairy cattle is twice as large as the effect on beef cattle. The impact of the linear precipitation variable for beef cattle supports previous studies on drought and livestock in Africa (Seo and Mendelsohn, 2006a; Swinton, 1988).

Table 7b. Marginal effects of choice of livestock species
Climate change scenarioDairy CattleBeef CattleGoatsSheepChicken
  1. Notes: *** Significant at 1% level. Std. errors in parentheses.

Annual temperature−0.0293***0.0146***0.0167***−0.0017 −0.0129***
(0.0039)(0.0039)(0.0039)(0.0037)(0.0037)
Annual rainfall0.0001−0.0005 0.0012***−0.002***0.0019
(0.0004)(0.0004)(0.0004)(0.0004)(0.0004)

The different effects of global warming on the probability of engaging in dairy and beef cattle management supports findings on droughts and consumption smoothing, which have shown that households may or may not adjust herd size to droughts depending on prevailing factors (Fafchamps, 1998; Kazianga and Udry, 2004). Fafchamps (1998) has also shown that two distinct forces are capable of inducing producers to hold onto livestock even when they anticipate losing many of their animals to global warming: (i), the desire to smooth consumption when livestock make an essential contribution to household income and other assets are not available; and (ii), the desire to maximize profits when demand for livestock products is inelastic. Kazianga and Udry (2004) however find that households in Burkina Faso rely almost exclusively on self-insurance in the form of livestock sales to smooth out consumption.

The decision to adopt goats and sheep is less responsive to climate change than that to hold cattle. Though the probability of holding goats and sheep shows a hill-shaped relationship with temperature, no significant impact was uncovered and the marginal effects were quite modest. Annual precipitation exhibits a hill-shaped relationship with the probability of engaging in goat rearing. Both the linear and quadratic terms are significant but the marginal probabilities are very low. None of the climate variables are significant for the sheep model. Though this result does not support findings on the impact of climate change and livestock adaptation in Africa (Seo and Mendelsohn, 2006a), it seems to support literature that argues that small ruminants are more adaptable to harsh agro-climatic conditions than cattle (Kabubo-Mariara, 2007). Only age of household head and the household's average level of education significantly affect the probability of holding sheep. Age is particularly quite significant, though the marginal effect is quite low. The elderly without labour support may turn to sheep (and goat) rearing because these activities are less labour demanding. Education is negatively correlated with both sheep and goat rearing probabilities. This supports studies which suggest that more educated farmers are likely to keep less livestock than their less educated counterparts because education broadens alternative income earning opportunities (Kabubo-Mariara, 2007). The probability of rearing chickens is significantly affected by all climate variables and the marginal effects are quite high compared to the impact on other livestock species. Annual temperatures exhibit a U-shaped relationship with this decision choice, but rainfall exhibits a hill-shaped relationship. Education is positively correlated with the decision to hold chickens, implying that though education may give the household alternative income earning opportunities outside livestock, some basic skills are required.

The multivariate probit results presented in Table 8 show the coefficients rather than the marginal effects presented in the probit model. The coefficients of the probit models (not presented) are however quite close to those of the multivariate probits. The results are consistent with the probit model results, but the climate variables in the multivariate probit model are all significantly correlated with the probability of choosing different species except for sheep. The Chow test indicates that the overall model is quite significant and also suggests that the system of equations is quite stable. The results for choice of dairy cattle, goats and chicken show a U-shaped response to temperature, but a hill-shaped response to precipitation. Contrary to the probit results, the choice of goats is significantly affected by temperature. Choice of beef cattle and sheep however show a hill-shaped response to temperature but a U-shaped response to changes in rainfall.

Table 8. Multivariate probit results for choice of livestock species
 Dairy cattleBeef cattleGoatsSheepChicken
  1. Notes: Robust z statistics in brackets.

Annual temperature−1.13550.5174−0.75920.2073−1.3095
[5.36]***[2.27]**[3.68]***[1.00][6.38]***
Annual temperature squared0.0247−0.01080.022−0.00640.0309
[4.86]***[1.98]**[4.41]***[1.29][6.30]***
Annual rainfall0.2968−0.1960.1209−0.04970.3196
[5.80]***[3.49]***[2.47]**[0.99][6.41]***
Annual rainfall squared−0.00180.0011−0.00070.0002−0.0018
[5.97]***[3.53]***[2.41]**[0.72][6.35]***
Log household size0.36230.48740.29980.15610.2134
[2.49]**[2.99]***[1.92]*[1.09][1.43]
Age of household head0.00190.00930.0110.0105−0.0029
[0.47][2.18]**[2.80]***[2.71]***[0.75]
Household's average years of education0.0201−0.0083−0.0309−0.02630.0323
[1.51][0.52][2.13]**[1.89]*[2.37]**
Number of observations 816     
Wald chi2(42) 947.79***     

4.3. Predicting impact of global warming on livestock choice

The probit regression results were used to project the impact of climate change on the decision of whether or not to engage in livestock production and also the choice of species to adopt. Evidence shows that there have been very large geographical disparities in the trend patterns for both temperature and rainfall in Kenya (Kabubo-Mariara and Karanja, 2006). Estimates show that there has been a tendency for annual rainfall to decrease in the arid and semi-arid areas but to increase over Lake Victoria and the coastal and neighbouring regions. This implies that climate change may cause precipitation increases in some regions while others may experience reductions, which is consistent with findings in other countries (see for instance, Gbetibouo & Hassan, 2005; Deressa et al., 2005; Seo et al., 2005). This implies that it is more appropriate to simulate the impact of different climate scenarios for different regions in Kenya separately. This would facilitate design of region specific adaptation policies rather than a uniform Kenya-wide approach. However, due to poor fit of regional models (high vs. low potential zones), this paper presents a simulation of the impact of climate change for the whole country. Since the only alternative simulations available are Africa-wide (Seo and Mendelsohn, 2006a), the simulations here still provide useful insights for general livestock adaptation and development in Kenya.

Using the probit results and variable means, we examine a set of climate change scenarios predicted by Atmosphere-Ocean General Circulation Models (AOGCMs). These climate scenarios reflect the A1 scenarios in the IPCC's Special Report on Emissions Scenarios (SRES) (IPCC, 2001) from the following models: Canadian Climate Center (CCC), Center for Climate System Research (CCSR), and Parallel Climate Model (PCM) (Kurukulasuriya et al., 2006; Seo and Mendelsohn, 2006a). In 2100, PCM predicts a 2 °C increase, CCSR a 4 °C increase and CCC a 6 °C increase in temperature in Africa. For precipitation, PCM predicts an average 10% increase, CCC a 10% decrease and CCSR a 30% average decrease in rainfall in Africa. Though the scenarios may not have a uniform impact across all Africa, they cover the range of two General Circulation Models (GCMs): namely the Canadian Climate Model (CCC) and the Geophysical Fluid Dynamics Laboratory model (GFDL), which have been found to give reasonable climate forecasts for Kenya (Kabubo-Mariara and Karanja, 2006).

First we examine the climate change scenarios for 2020, 2060, and 2100 from each of the AOGCMs. The results (Table 9) show that all three models predict that temperatures in Kenya will increase steadily up to the year 2100. Rainfall will however be noisier and will fluctuate across the three models. The overall impact of the CCC and CCSR predictions will be a fall in rainfall by 2100, but PCM predicts an increase in average rainfall.

Table 9. AOGCM climate scenarios
 Current202020602100
Temperature (°C)
CCC1920.3521.9024.44
CCSR1920.6221.3522.34
PCM1919.5420.3521.04
Rainfall (mm/month)
CCC85817669
CCSR85788166
PCM85958688

To get the new climate for each district, we added the predicted change in temperature from each AOGCM to the benchmark values, and then evaluated the impact on the probability of engaging in livestock management. We also adjusted benchmark precipitation by the predicted percentage to get the new precipitation levels. We repeated this exercise for the probability of choosing different livestock species. Due to space constraints, only predictions for the year 2100 are presented.

The predictions for probability of engaging in livestock management (Table 10) show that a warming of between 2 °C and 4 °C reduces the probability of holding livestock by about 7%. However higher levels of warming (a 6 °C rise in temperature) increase the probability of holding livestock by about 4%. The predictions suggest that there is a turning point in the response of the probability of holding livestock to increased temperatures. As temperature rises, farmers reduce their demand for livestock holding, until at a threshold of about 5.5 °C rise in temperature where the farmers are indecisive of whether to reduce their demand or not (change in probability is zero). Increases in temperatures of more than 5.5 °C will make farmers adapt their farming systems towards livestock production and therefore probability rises.

Table 10. Predicted change in probability of adopting livestock from different climate scenarios
Climate change scenarioPredicted probability*Change in probability% Change
  1. Note: * Base probability is 0.88.

PCM: +2°C temperature 0.83−0.059−6.69
CCSR: +4 °C temperature0.82−0.060−6.78
CCC: +6 °C temperature0.920.0343.84
PCM: +10% rainfall0.70−0.10−10.98
CCC: −10% rainfall0.84−0.046−5.16
CCSR: −30% rainfall0.48−0.409−46.26
PCM: +2 °C tempt & +10% rainfall0.73−0.16−17.99
CCSR: +4 °C tempt & −30% rainfall0.59−0.297−33.56
CCC: +6 °C tempt & −10% rainfall0.940.066.28

All the simulated effects of a change in precipitation predict a fall in the probability of holding livestock. The decline is highest for a 30% reduction in rainfall as predicted by the CCSR, followed by a 10% increase as predicted by the PCM. It is not clear whether there exists a turning point in the response of livestock demand to changes in precipitation. Simulated impacts of both increases and deceases in precipitation suggest options of reducing the demand for livestock management. A 10% increase in precipitation is predicted to reduce the probability of holding livestock by 11%, but a 10% decrease is predicted to reduce the probability by 5%. Though the results may seem to be conflicting, most likely they reflect the adaptation options available to the farmer. On one hand, a very high rise in temperature will discourage stocking, and farmers may sell their livestock. Livestock could also die due to extreme temperatures combined with lack of water, as happens many times in Kenya during severe droughts. On the other hand, an increase in precipitation may lead to substitution between crops and livestock, more so in the less arid zones.

The predicted changes in probability of selecting different livestock species are presented in Table 11. The predictions are based on the probit model results. The results suggest that global warming will reduce the probability of keeping dairy cattle. The highest fall will be from warming predicted by the CCSR model (43% and 23% from rainfall and temperature changes respectively). The CCSR also predicts the highest combined decline in probability of holding dairy cattle. At low levels of temperature, choice of livestock species seems to be more sensitive to temperature than to precipitation changes. The predicted changes in probabilities of holding beef cattle are exactly opposite those of dairy cattle. The probability of holding beef cattle rises as global warming increases. This implies that with warming, households substitute from dairy to beef cattle. Consistent with the results for dairy cattle, the highest predicted changes are from the CCSR model. Unlike the predictions for cattle, goats, sheep and chickens show mixed results. Increased temperatures and precipitation increase the probability of holding goats, but decrease the probability of holding sheep and chickens. However, higher temperatures increase the probability of keeping chicken. The predictions imply interesting scenarios in the response of different animal species choice to climate change. Dairy cattle and beef portray a reversed non-linear response to changes in temperature and precipitation. Goats and chickens portray a rising response to changes in temperature but sheep have a falling response. The response to changes in precipitation is falling for goats and chickens, but rising for sheep. The combined impacts suggest a difference in the response of farmers’ choices of different species to changes in temperature and precipitation. For instance, it seems like dairy cattle are much more responsive to changes in temperature than to precipitation, but the reverse is the case for beef cattle.

Table 11. Change in probabilities of selecting livestock species from different climate scenarios
Climate change scenarioDairy CattleBeef CattleGoatsSheepChicken
PCM: +2 °C temperature −20.5126.5030.44−5.92−8.88
CCSR: +4 °C temperature−23.0533.8862.31−17.68−6.54
CCC: +6 °C temperature−12.3421.9191.80−33.744.17
PCM: +10% rainfall−12.7821.992.52−8.68−6.48
CCC: −10% rainfall−3.0012.10−7.3214.33−10.57
CCSR: −30% rainfall−43.26110.27−35.5759.15−53.25
PCM: +2 °C tempt & +10% rainfall−24.8151.7931.93−14.06−12.81
CCSR: +4 °C tempt & −30% rainfall−51.05132.9587.4738.61−54.52
CCC: +6 °C tempt & −10% rainfall−10.2721.9226.79−22.392.08
Base probabilities0.660.220.410.350.65

5. Conclusion

This paper examines the impact of climate change on farmers’ decisions to engage or not to engage in livestock activities and also on the choice of different livestock species. The analysis is based on primary data are collected from a sample of 816 households from 38 districts in 2004. The primary data is enriched with secondary climate data, which reflect long-term climate changes in Kenya. The probability of engaging in livestock activities is analyzed using the probit model and is based on the entire sample. The choice of different livestock species is analyzed using univariate and multivariate probit approaches. Furthermore, the impact of different climate change scenarios predicted by AOGCMs on the probability of engaging in livestock management and also on choice of livestock species is examined.

The probit model results indicate that there is a non-linear relationship between global warming and the decision to engage in livestock production, suggesting that farmers adapt their livestock holding decisions to climate change. The results further suggest that climate change affects livestock preference in Kenya. The marginal impacts of temperature are much higher than the impacts of precipitation. The results strongly suggest that with increased warming, farmers will reduce dairy in favour of beef cattle. Goats and sheep are less responsive to climate change than cattle, implying that they can withstand harsher climate conditions than cattle.

The simulations of climate change scenarios suggest that at low levels of global warming, the probability of choosing livestock management falls, but as it gets much warmer, the probability rises. The simulated impacts of changes in precipitation suggest less robust results. The simulations also suggest that increased temperatures lead to adjustments in the decision to keep dairy cattle. When it becomes much warmer, farmers opt to move towards beef cattle but to reduce the demand for dairy cattle. The probability of choosing goats also increases with global warming while the demand for sheep rearing declines. The corresponding impact of warming through reduced rainfall also leads to substitution between dairy and beef cattle, and also goats instead of sheep.

These results imply that adaptation to climate change in Kenya is important if livestock holding households are to counter the expected impacts of long-term climate change on their livelihoods. Monitoring of climate change and disseminating information to farmers are critical policy interventions. There is a need to educate farmers on the vulnerability of specific species and the appropriate species mix, including drought resistant breeds so as to empower them to make appropriate adaptations to counter the adverse impact of climate change. These interventions will need to be complemented by livelihood supportive policies including livestock market development, risk-sharing and insurance options as well as policies for improving the socio-economic conditions of many livestock dependent households in Kenya.

This paper makes an important contribution to the literature on the impact of climate change on animal husbandry in Africa. We are not aware of any specific study that analyses livestock choice and adaptation to climate change in Kenya. Further research is however necessary to document the adaptation options by agro-ecological zones as this may have important implications on choice of livestock species given large geographical disparities and variability in trend patterns for both temperature and rainfall. It is also important to analyze the impact of climate change on choice of main livestock species using multinomial choice procedures because it would be expected that in resource-constrained situations, farmers compare profits from the available animals when they make choices of which species to manage. The scope of the study in this direction was limited by the available data, collected primarily to analyze the impact of climate change on crop agriculture.

Footnotes

  1. Dr. Jane Kabubo-Mariara is Senior Lecturer, School of Economics, University of Nairobi, Nairobi, Kenya. E-mail: Jmariara@mail.uonbi.ac.ke

Acknowledgements

The author is grateful to the Global Environment Facility/World Bank for financing the project under which the data used in this paper were obtained. The author is also thankful to the Centre for Environmental Economics and Policy in Africa (CEEPA), University of Pretoria, for housing the project and providing technical support. Comments and suggestions from an anonymous reviewer are also acknowledged. The author is however responsible for all errors and omissions in this paper.

Ancillary