Entrepreneur: Start & Grow Your Business

Measuring heterogeneous preferences for cattle traits among cattle-keeping households in East Africa.


by Ouma, Emily^Abdulai, Awudu^Drucker, Adam

In spite of the crucial role livestock plays in the economies of many sub-Saharan African countries, livestock productivity remains relatively low in the region. Several studies have shown that factors ranging from technical, institutional, and infrastructural constraints related to feeding, animal health, and genotype are the main causes of the low productivity (Tano et al. 2003). Breed improvement programs serve as natural entry points for productivity increases. However, as argued by Tano et al. (2003), the tendency for genetic improvement programs to concentrate on one aspect, such as meat or milk production, in isolation from broader livelihood system needs often results in the substitution of exotic cattle for indigenous breeds. Although indigenous cattle are often less productive than exotic breeds, when traits such as milk and beef production are considered in isolation, they may be better suited to the conditions of the local environment. This has often led to low adoption rates of exotic or crossbred cows (Abdulai and Huffman 2005). De Haan (1995) points out that livestock technologies have been the source of puzzling outcomes more frequently than crop technologies, partly because new technologies that would improve productivity are simply not adopted.

The low productivity in Africa's livestock has also been partly attributed to the multiple functions that cattle perform in the livelihood system (Moll 2005). It is estimated that approximately 80% of the value of livestock in low-input developing country systems can be attributed to nonincome, sociocultural functions, while only 20% is attributable to physical products such as meat, milk, and wool. In contrast, over 90% of the value of livestock in high-input developed country production systems is attributable to direct production outputs (Gibson and Pullin 2005). Some of the important nonincome and sociocultural functions of cattle in developing countries are embedded in traits that are not traded in the market, therefore lacking price or market values. Hence, the utilization of profit functions for derivation of economic values for cattle traits in such systems would result in exclusion of such traits from the breeding objective, which may result in genotypes not capable of fulfilling the multiple objectives of the cattle enterprise. This calls for the employment of a valuation method that can also permit calculation of economic values of traits without market values to be included in the breeding objective.

Few economic studies have attempted to investigate phenotypic trait preferences for genetic resources using stated preference approaches. Some authors have used conjoint analysis to assess relative importance of traits (Sy et al. 1997; Tano et al. 2003). These studies utilizing conjoint analysis employ ordered probit or multinomial logit models to model preference behavior. A limitation of these models is that they do not explicitly account for heterogeneity of preferences among producers, rendering them less useful for analysis aimed at providing policy recommendations for different environments and production systems.

Recent studies have increasingly focused on the monetary value of traits by employing choice experiments and including a monetary cost or benefit as one of the traits. For example, Scarpa et al. (2003) employ a mixed logit model, a recent advancement in discrete choice analysis, in their empirical estimation of choice data to value the phenotypic traits expressed in indigenous breeds of livestock in a pastoral system in Kenya. Zohrabian et al. (2003) use a maximum entropy search-theoretic framework to estimate the value of precommercial germplasm contained in the U.S. national plant germplasm system. Kontoleon (2003) utilizes a latent class model to account for preference heterogeneity in GM foods. Kline and Wichelns (1998) note the significance of accounting for preference heterogeneity, since preferences often vary among individual decision makers according to their environment, socioeconomic characteristics, and tastes.

The objective of this study is to derive the economic values for cattle traits in Kenya and Ethiopia using a choice modeling approach. Specifically, a mixed logit model is employed to investigate the existence of preference heterogeneity, while a latent class model is used to examine the sources of heterogeneity across segments of cattle keepers. To the best of our knowledge, no previous attempt has been made to investigate the existence of endogenous preference segmentation for cattle traits among cattle keepers in sub-Saharan Africa. Given that trypanosomosis disease is a major challenge to livestock production in the study areas, the study was partly designed to provide a better understanding of farmers' preferences for cattle and to enable a reasonable assessment of programs aiming at encouraging the adoption and use of trypanotolerant breeds.

Choice Modeling

The conceptual framework for choice experiments arises from the consumer theory developed by Lancaster (1966), which postulates that preferences for goods are a function of the traits or characteristics possessed by the good rather than the good per se. An important implication of this theory is that overall utility of a good can be decomposed into separate utilities for its constituent characteristics or traits. In terms of the utility function, this translates into using the characteristics of goods as the arguments of the function. Hence, a good can be described by the characteristics that generate utility or disutility to individuals. For cattle breeding, this permits the analysis of farmers' preferences in terms of the utility they perceive to result from various cattle traits.

Choice experiments are based on the assumption that an individual n receives utility, U, from choosing an alternative A equal to [U.sub.nA] = U([X.sub.nA]) from a finite set j of alternatives in choice set k, if and only if, this alternative generates at least as much utility as any other alternative, with [X.sub.nA] denoting a vector of the attributes of A. Utility is represented as two components, where one portion is deterministic and depends on the attributes of the alternative, and the remainder is stochastic. This can be specified as

(1) [U.sub.nA] = [V.sub.nA] + [[epsilon].sub.nA]

where [V.sub.nA] = g([X.sub.nA]) is the deterministic component and [[epsilon].sub.A] is a random component of the utility function. The probability of individual n choosing alternative A can then be specified as

(2) P(A) = Prob{[V.sub.nA] + [[epsilon].sub.nA] [greater than or equal to] [V.sub.nj] + [E.sub.nj]; A [not equal to] j, [for all] j [member of] k}.

Choice experiments and hedonic price analysis are alternative empirical applications to the Lancaster consumer theory. The strength of both hedonic pricing and choice experiment techniques is the ability to decompose revealed preference data, that is, prices of goods in the case of hedonics and choice of goods' profiles by individuals in the case of choice experiments, into marginal values (Dalton 2004). A number of recent studies, such as Barrett et al. (2003) and Green et al. (2006), have employed hedonic pricing to value livestock traits in eastern Africa. The use of hedonics is nevertheless limited to the valuation of existing traits and not prospective traits that may be of interest to breeders. Choice experiments overcome this limitation since preferences are measured directly, and then related to utility, making it possible to estimate economic values of prospective traits. Choice experiment technique is employed in the present study.

Estimation Techniques and Econometric Models

Discrete choice models are normally used to model the choices made by the decision makers from the choice experiments. Mixed logit and latent class models have recently been developed to relax the limiting assumptions associated with conventional logit and probit models. McFadden and Train (2000) and Train (2003) describe mixed logit as a highly flexible model that can approximate any random utility model. It relaxes the limitations of standard logit by allowing the taste parameters to vary randomly according to a parametric distribution. In addition, it allows for unrestricted substitution patterns and correlation in unobserved factors over time (Train 2003; Hensher and Greene 2002). (1) We refer the interested reader to the above literature for a more in-depth description.

The mixed logit model applies the usual framework of random utility models outlined in the previous section. As indicated earlier, a sampled individual n faces a choice of selecting the preferred alternative amongst a set of j cattle profiles, representing different traits and trait levels in each of the t choice situations. In our case, the number of choice situations is constant per respondent and we assume a linear utility function. An individual, n, is assumed to consider the full set of offered cattle profile alternatives in choice situation t and to choose the alternative with the highest utility. The utility associated with each set of j alternatives as evaluated by each individual n in choice situation t is represented in a discrete choice model by a utility expression of the general form,

(3) [U.sub.njt] = [[beta].sub.n][x.sub.njt] + [[epsilon].sub.njt],

where [x.sub.njt] is a vector of observed variables that includes the cattle traits and socioeconomic characteristics of cattle owners. Coefficient vector [beta] is unobserved for each n and varies in the population with density f([[beta].sub.n] | [theta]), where [theta] is a vector of parameters of a continuous population distribution, [[epsilon].sub.njt] is an unobserved random term that is assumed to be identically and independently distributed. The focus in mixed logit shifts from finding estimates of [[beta].sub.n] to finding the estimates of [theta], the population parameters (e.g., location and spread parameters), which determine the behavior of [[beta].sub.n]. Conditional on [[beta].sub.n] the probability that person n chooses alternative i, in choice situation t, is the conditional logit specification

(4) [ILLUSTRATION OMITTED]

Given that [[beta].sub.n] is unknown to the analyst, the unconditional probability is usually employed. The unconditional probability is the integral of the conditional probability over all possible values of [beta], which depends on the distribution of [beta]. This integral takes the form

(5) [P.sub.nit]([theta]) = [integral] [L.sub.nit]([[beta].sub.n])f([[beta].sub.n] | [theta])d[[beta].sub.n].

Although mixed logit accounts for preference heterogeneity by allowing taste parameters to vary randomly over individuals, it is not well suited to explaining the sources of heterogeneity (Boxall and Adamowicz 2002). These sources often relate to the socioeconomic characteristics and taste of the decision maker. Though it is possible to account for the socioeconomic characteristics of the decision maker by interacting key individual characteristics with the traits, this requires a priori selection of key limited individual-specific variables.

Latent class models are more suited in explaining the sources of heterogeneity, since individuals are intrinsically sorted into a number of latent classes. Each is characterized by homogenous preferences though heterogeneous across classes (Boxall and Adamowicz 2002). However, its limitation is the assumption of local independence, where the choice variables are assumed to be independent of each other, conditional on latent class membership. Assignment of individuals into classes is probabilistic based on their sociodemographic characteristics. In this article we use a mixed logit model to identify the existence of preference heterogeneity and latent class modeling to estimate segment or class-specific parameters.

In the latent class logit, the mixing distribution f([[beta].sub.n] | [theta]), is discrete, with [[beta].sub.n] taking a finite set of distinct values (Train 2003). In this case, it is assumed that individuals are intrinsically sorted into a number of classes based on their tastes. Members of each class have similar tastes. However, the classes are latent, not observable by the analyst. Within the class, the individual choices from one choice situation to the next are assumed to be independent and choice probabilities are assumed to be generated by the logit model (Greene 2002). For instance, the probability that individual n chooses alternative i in a given number of choice situation t, given that he belongs to latent class c is

(6) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [x.sub.nit] is a vector of observable traits associated with alternative i, and [[beta].sub.c] is a class-specific parameter vector; t denotes the number of choice situations for person n, while [[beta].sub.c] is used to capture heterogeneity in preferences across classes. Since the classes are not observable, class probabilities are specified by the multinomial logit form

(7) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [z.sub.t] is a set of observable characteristics that enter the model for class membership. The cth parameter vector is normalized to zero to ensure identification of the model. This model does not impose the independence of irrelevant alternatives on the observed probabilities. For a given individual n, the model's estimate for the probability of a specific choice is the expected value, over classes, of the class-specific probabilities. In our case, each person makes repeated choices for either eleven bull profiles or twelve cow profiles. The repeated choices permit an examination of how the levels of various traits influence individual utility and a comparison with a priori expectations.

Survey Design and Data Description

The choice experiment household surveys were conducted in Narok and Suba districts in Kenya and the Ghibe Valley in Ethiopia, representing different livestock production systems in trypanosomosis prevalent areas. The three areas were selected after spatial mappings of tsetse fly distribution and cattle densities in Kenya and Ethiopia were done to target areas with trypanosomosis challenge. A purposive random sampling procedure to identify cattle-keeping households was employed to select 172 households in three divisions of Narok district, 132 households in two divisions of Suba district, and 204 households in the lower and upper Ghibe Valley. Sampling was done along line transects within the lowest administrative unit in the two countries. The sample size was calculated as a proportion of the total number of households in the administrative unit. The survey was administered between September and December 2004 in Kenya and April to May 2005 in Ethiopia by locally trained enumerators.

In order to identify the relevant cattle traits to be included in the choice experiment, informal focus group meetings were held with cattle keepers in the study sites. Cattle keepers were asked to identify the cattle traits that they considered important in meeting their objectives, while taking into consideration their prevailing local and environmental conditions. Pair-wise ranking technique was then employed to select the relatively highly preferred traits. A total of seven preferred traits for cows and six for bulls, each with two to three levels were identified for inclusion in the design of the choice experiment. An additional monetary trait, purchase price, was included to capture the marginal willingness to pay for the traits. The different purchase price levels considered were based on the prevailing market prices of bulls and cows. Table 1 presents the traits included in the choice experiment and their levels.

The identified traits and their levels were then combined according to an experimental design to create choice sets. A full factorial design, which includes all possible combinations of the traits, would yield 864 ([2.sup.5] x [3.sup.3]) possible choice sets for cows and 432 ([2.sup.4] x [3.sup.3]) for bulls. Since it is not practically feasible to work with such a large number of choice sets, a randomized, partially orthogonal main effects only combinations was created using Choice-Based Conjoint (CBC) software, which resulted in twelve generic choice sets for cows and eleven for bulls. (2) These were then used to construct cards with pictorial profiles describing the differences in traits and the levels to demonstrate each choice set to survey respondents. The choice experiment was administered as part of a household-level questionnaire survey using in-person interviews in respondents' homes. The rest of the questionnaire covered socioeconomic aspects such as household and farm characteristics as well as market and resource access.

The administration of the choice experiment was conducted in the following manner: respondents were first introduced to the type of choice task and asked to indicate their preferences between cows and bulls for purchases. The respondents were then presented with choice tasks for either bulls or cows and shown two pictorial profiles based on each choice set. They were then asked to choose the most preferable animal profile to purchase for rearing. Each of them was presented with either twelve choice sets in the case of cows or eleven choice sets in the case of bulls. In each case, a "don't buy" option was also included for respondents who preferred neither of the two profiles presented. A total of 253 complete choice experiment interviews were carried out for bulls and another 253 for cows, yielding 2,783 completed choice sets for bulls and 3,036 for cows. Twenty-three choices (two interviews) were discarded due to incomplete information.

Table 2 presents a description of the choice experiment variables used in estimating the mixed logit and latent class models. To avoid confoundment with the grand mean, we employed effects coding for the choice experiment variables to measure nonlinear effects in the trait levels (Hensher, Rose, and Greene 2005). (3) Effects coding uses values for codes, which when summed over any given trait column, equal zero. From an a priori perspective, we expect that the trait levels of trypanotolerance, high fertility in bulls, high reproduction potential in cows, and high milk yield should increase an individual's utility, as should an increase in live weight. A high reproductive performance and high fertility are expected to have a positive impact on herd productivity and herd size, while trypanotolerance would increase the competitiveness of the cattle enterprise by reducing the costs associated with the treatment of trypanosomosis disease.

Similarly, low watering requirement for the animals should increase utility, since the watering points are located far from the homesteads at an average distance of 3 km. The coefficient values for coat color and traction potential may plausibly take either negative or positive sign depending on an individual's preference. Coat color is an important stimulus in attracting tsetse flies to a target and also has a role in the landing response of the flies. The strongest landing responses have been found to be on black surfaces relative to lighter ones. For this reason, cattle keepers in the Ghibe Valley prefer light-coat-colored cattle in order to minimize the likelihood of trypanosomosis infection through tsetse fly bites. In the Narok district, however, there is preference for blackcoat-colored bulls with white spotted dewlap, features that are important in the selection of bulls for use in ceremonial functions. Finally, the trait coefficients associated with monetary expenditure, that is, purchase price of the animal and the need for purchased feed supplements is expected to have a negative sign due to the positive marginal utility for income generally exhibited by most individuals.

Information on the socioeconomic characteristics of the sample households used in the econometric modeling is presented in table 3. Fifty percent of the households use tsetse fly control methods, which include spraying, applying pour-on insecticides, and clearing bush as a way to control the disease. Generally, illiteracy levels of household heads are high in the sample population. The average number of years of schooling for a household head is 4.9 years. In terms of access to market infrastructure, the households are located relatively far from market centers, at an average distance of 4.8 km. Average cattle herd size per household is thirty animals, but is significantly different between the pastoral, agro-pastoral, and crop-livestock production systems at the 1% level. In the pastoral system, a household owns an average of seventy-three cattle, while in the crop-livestock system in Kenya, households own twelve cattle, mainly comprising the local Zebu breeds. In Ethiopia, the average cattle herd size is six animals, mainly comprising male stock, which is used for traction. In pastoral systems, livestock keeping is a central component of the livelihood system and is closely linked to the cultural and social lives of the communities, where livestock numbers are an important means of demonstrating wealth and a mode of cushioning high livestock losses experienced during major droughts or disease outbreaks.

Empirical Results

The simulated maximum likelihood estimates for the mixed logit model that allows for correlations between taste parameters are reported in table 4. (4) Following Hensher, Rose, and Greene (2005), we used a zero-based, asymptotic t-test for individual parameter standard deviations to determine the set of random parameters. From this, the traits trypanosomosis, traction, fertility, live-weight, purchase price, feeding requirement, and reproduction ability were entered as random parameters in the mixed logit estimations, while watering frequency and coat color were assumed fixed. Further, we assumed a normal distribution for all random parameters, except purchase price, which was assumed to have been drawn from a triangular distribution.

The model was estimated using NLOGIT software version 3.0 (Econometric Software, Inc. 2002) utilizing 100 Halton draws for the simulations. Given that the estimation of mixed logit first involves the estimation of a conditional logit model to derive initial start values for each of the parameters in the mixed logit model, the relative performance of the conditional and mixed logit models can be compared. A likelihood ratio test is performed to test the null hypothesis that the conditional logit fits the data better than the mixed logit. The sample values of the likelihood ratios are 941.8 and 793.6 for bulls and cows traits, respectively, with a critical value of [[chi square].sub.20,0.01] = 37.6, thus rejecting the null hypotheses. This finding indicates that the mixed logit model, which allows random taste variation, fits the data better than the conditional logit model, which assumes fixed taste parameters.

With the notable exception of the purchase price, the mean coefficients of the random parameters for bulls in table 4 are all significant at the 1% level. The model reveals preference for bulls that are trypanotolerant, cheap, highly fertile, have good traction potential, and high live-weight. Associated with each of the mean coefficient estimates of the random parameters are derived standard deviations calculated over the R draws, indicating the amount of spread that exists around the sample population. The standard deviation of each random parameter coefficient is highly significant, indicating that these coefficients are indeed heterogeneous in the population. The nonrandom parameter, watering frequency, is positive and highly significant, implying that there is a preference for bulls that are drought-tolerant (need to water only once in two days). (5)

Mean coefficient estimates for all random parameters for cow traits in table 4 are significantly different from zero at the 5% level, and the standard deviations are also highly significant, implying existence of preference heterogeneity in the population. The results indicate preference for cows that are cheap, trypanotolerant, have high milk yield, and high reproduction ability. Cows that need supplementary purchased feeds are not preferred, as indicated by the negative coefficient on feeding requirement. The constant variable in the model results in table 4 represents the "don't buy" choice option and is the base for the choice model, as it is associated with "zero" utility. It takes a value of one if the option is "don't buy" and zero otherwise. The results indicate a strong negative preference for this option, implying that the respondents preferred to select the other two choice options associated with various trait levels.

The estimated means and standard deviations of the coefficients provide information on the shares of the population that place a positive or negative value on the bull and cow traits. (6) Preference for the trait tolerance to trypanosomosis is skewed to the positive orthant, with 75% and 73% of the cattle keepers preferring trypanotolerant bulls and cows, respectively. A bull with good traction ability is preferred by 88% of the cattle keepers. Similarly, a cow with high reproduction potential is also preferred by 88% of the population, while 72% of the population has a negative preference for paying for feed supplements for cows. The estimated mean and standard deviation coefficients for the purchase price of cows reveal that 39% of the distribution is below zero and 61% is above zero. This indicates that two-thirds of the population display a negative marginal utility for income, implying that for this group, purchasing a cow may result in an increase in utility, which outweighs the decrease in utility from spending monetary resources on the purchase of the animal. This is even more pronounced for bulls where only 23% of the population displays a positive marginal utility for income. A plausible interpretation for these estimates is that cattle keepers may view the cost of purchasing cattle as a worthwhile expense, due to the multiple benefits associated with owning a cow or a bull. (7)

Of significant interest is the derivation of the marginal rate of substitution between the traits and the monetary coefficient (purchase price in our analysis) as it provides an estimation of implicit prices for the traits. (8) Hensher and Greene (2002) suggest that to derive behaviorally meaningful values from mixed logit, the distributions from which the random parameters are drawn need to be constrained. Although little is reported in the literature about the best constraint to implement, Hensher, Rose, and Greene (2005) argue that constraining the standard deviation parameter estimate to that of the mean of the random parameter for a triangular distribution guarantees nonnegative implicit price values. Implicit prices are therefore computed using conditional constrained parameters. (9) The results are reported in table 5. (10)

The estimates are validated using the household-level survey data and past studies. Estimates of willingness to pay (WTP) for traits parameters indicate that a trypanotolerant animal is valued at U.S. $25 more than a trypano-susceptible one. According to the survey data, the average cost of treatment or control of trypanosomosis per year per herd varies from U.S. $6 to U.S. $37 in Ethiopia, while in Kenya it is an average of U.S. $36 in crop-livestock systems but can be as high as U.S. $219 in pastoral systems, depending on the herd size and number of treatments per year. The choice experiment estimate of U.S. $25, therefore, appears plausible, although more information can be gained by assessing the distribution of the estimate in the population. An adaptive bounded kernel density approach has been utilized to examine the distribution of the WTP for the various traits that have values that are inherently bounded from below by zero. We employ Gaussian kernels with variable-bandwidth density estimator that supports boundary correction for variables with bounded domain (Jann 2005). (11) However, WTP trait values for traction potential may plausibly take positive and negative values, since it is a highly valued trait in cropping systems, yet most pastoral communities view use of cattle for draft power as a taboo and therefore exhibit negative preference for the trait. Similarly, cattle keepers have a negative preference for cows that require supplementary purchased feeds to boost milk production. This is probably because most of the systems in the study area are subsistence and not market oriented and additional costs associated with production result in a decrease in utility. These distributions are presented in figure 1. The distribution of WTP values for trypanotolerance presented in figure 1(a), shows a bimodal distribution having two peaks at U.S. $19 and 36. This indicates presence of heterogeneity in trypanotolerance trait valuations. Although little is reported in the literature about valuation of trypanotolerance using stated preference methods, Tano et al. (2003) show that disease resistance can be ranked very highly under production systems of similar types to those covered in this study.

The kernel density plot in figure l(b) for traction potential also provides evidence of heterogeneous preferences for traction potential in bulls, indicated by the substantial negative values and the bimodal distribution density having two peaks above zero. The mean of the WTP distribution is U.S. $58 for a bull with good traction potential. This finding is consistent with the finding by Tano et al. (2003) who found that fitness to traction was the highest ranked trait for bulls in West Africa.

[FIGURE 1 OMITTED]

Live-weight increase, which is associated with meat production, is valued at U.S. $1.05 per kg. This is comparable to the average slaughter weight value of U.S. $1.02 per kg found in Scarpa et al. (2003) for a pastoral system in Kenya. A bull that needs to be watered only once every two days, used as a proxy for drought tolerance, is valued at U.S. $7 more than one that needs to be watered twice per day. This is not surprising, since water is a constraint in the study areas especially during the dry seasons. The trait is, however, not significant for cows. The kernel density distribution of WTP values for low watering frequency in figure l(c), is tight with values concentrated around the mean, indicating homogeneity in preference for this trait.

An important attribute in cows is the ability to calve every year instead of once in two years. The Zebu cattle breeds, kept by most of the respondents in the study sites, tend to have relatively longer calving intervals of about 1.8 years. A calving interval trait of one year is valued at U.S. $9.4, which is even higher than the value of U.S. $8.1 for a cow with high milk production. The kernel density distribution of WTP for reproduction ability in figure 1 (f), indicates large variance for positive values. The value of purchased supplementary feeds for cows that cattle keepers would be willing to accept as compensation for utility reduction is U.S. $10.7. Cattle keepers are reluctant to raise cows requiring purchased supplementary feeds to boost milk production. Due to scarce financial resources, the cattle keepers prefer cattle that do not require externally acquired inputs. The kernel distribution associated with the implicit price for supplementary feeds in figure 1(d), has a mean substantially in the negative range indicating significant aversion to this trait, though there is a small fraction in the positive orthant.

The maximum likelihood estimates for the latent class model for cows and bulls are reported in tables 6 and 7. To identify the number of latent classes to be used in the analysis, we employed the Bayesian Information Criterion (BIC) proposed by Boxall and Adamowicz (2002). This criterion is low and more intuitive for the model with three classes for bulls and cows. (12) Therefore, we estimated three-class latent class model (LCM) for cows and bulls. The LCM results indicate significant heterogeneity in preferences across latent classes as revealed by the differences in magnitude and significance of the utility function parameter estimates. For instance, table 6 indicates strongly positive trypanotolerance trait coefficient for class 3 in comparison to the other two classes while the traction coefficient is strongly positive for class 1. The class membership coefficients for the third class were normalized to zero in order to be able to identify the remaining coefficients of the model. The class membership coefficients for bulls and cows estimations indicate that the probability of being in a class is significantly related to the cattle keepers' production system. Table 6 shows that 54% of the respondents who participated in the choice experiment for bulls have a fitted probability to belong to class 1, which is strongly significant for agro-pastoralists. On the other hand, 19% and 27% of the respondents have a fitted probability to belong to class 2 and 3, respectively. Class 2 membership coefficients indicate that members of this class are likely to be pastoralists, younger household heads with low levels of education and high income relative to class 3. Class 3 could subjectively be associated with crop--livestock farmers. Class 1 for cow traits in table 7 is strongly significant for agro-pastoralists while class 2 is significant for crop--livestock farmers, who have a lower income, relative to class 3, and practice some tsetse fly control measures to control trypanosomosis. Class 3 could subjectively be associated with being a pastoralist.

Calculations of implicit prices of the traits across the latent classes show marked differences in preference structure as presented in table 8. Cattle keepers in class 1 attach a high value to good traction potential in bulls, which is even higher than the value they attach to trypanotolerance and high fertility in bulls. This is not totally unexpected since agro-pastoralists, the most likely members of this class, consider draft power capacity in bulls for ploughing and threshing grains an important reason for keeping cattle. This is further supported by the fact that their cattle herd contains more males (average of 2.2) than females (average of 1.7). Reproductive potential in cows is also an important trait for this class, while milk production is not statistically significant, suggesting that the trait is not important for this class.

Class 2, mainly associated with pastoralists (for bull traits), exhibits a different preference structure from the other two classes. In this class, high fertility in bulls is a highly valued trait, while trypanotolerance and live-weight have the same weight. Traction fitness is not highly valued in this class, a finding that is in line with expectations, since pastoralists do not use cattle for traction. They are more concerned with obtaining larger herd sizes, which is linked to high fertility trait in bulls. This class is also associated with crop-livestock farmers for cow traits who exhibit negative preferences for cows that need supplementary purchased feeds. High milk production is an important trait for this class, though reproduction potential was not found to be significant. Class 3, comprising mainly the crop-livestock farmers (for bull traits), displays a high-preference for trypanotolerant animals and bulls with good traction potential. High fertility in bulls is not a significant trait in this class. In this system, bulls are used for ploughing crop fields. For cows, high milk yield and high reproduction potential are valued by the pastoralists.

Conclusions

This article has employed mixed logit and latent class models to examine farmers' preferences for cattle traits in Kenya and Ethiopia, using choice experiment data for bulls and cows. The empirical results provide several insights to understanding cattle keepers' choice behavior. The results from the mixed logit model revealed significant preference heterogeneity among cattle owners, based on the environment and production system. Good traction potential, fertility, and trypanotolerance were found to be the most preferred traits in the model of bull preferences. The most valued traits in the cow preference models were trypanotolerance and reproductive performance.

Quite significant and interesting was the finding that traits related to beef and milk yield were ranked below the other factors such as traction, fertility, and resistance to trypanosomosis. This is probably because the cattle keepers in these systems are subsistence oriented and consequently place less emphasis on meat and milk production for the market. These findings are particularly interesting because traditional economic analyses on livestock and cattle breeding programs often focus on raising milk and meat productivity, with little emphasis on the nonincome traits such as traction and disease resistance. For example, the National Sahiwal Stud of Kenya, which was established with the objective of improving the breed for milk and meat production in marginal areas, focuses its selection criteria for the breeding stock on traits associated with meat and milk production such as lactation milk yield, age at first calving, calving interval, and growth rate without consideration of adaptation traits, which may be useful for marginal areas (Mpofu and Rege 2002). However, the results clearly suggest that breeding programs, as currently practiced in East Africa may be focusing on