Discrete choice experiments (DCE) is based on the characteristics theory of value (Lancaster, 1966; Rosen, 1974) and random utility theory (McFadden, 1974). Lancasterian theory of demand postulates that consumers not only derive utility from goods per se, but from the complex of different characteristics embodied in the product or service. The random utility assumes that farmers’ choose vaccine programs which they perceive will provide the maximum utility and this perceived utility has deterministic (measurable) and random components. The DCE method has, therefore, the advantage that the utility of hypothetically marketed good/service is divided into its components or attributes. Consumers in real markets make decisions among competing alternatives. Participants in DCE survey are asked to make a choice between alternatives with different attribute level combinations. Survey methods that ask consumers to make choices from experimental
choice sets enable researchers to learn about consumer preferences for products and attributes that do not yet exist in real markets (Carson et al., 1994).
The discrete choice experiment survey reported here involved several steps of designing processes. This process began with the collection of expert opinion on the attributes and attribute levels of potential NCD vaccination programmes. Then, further discussion was made with experts in the area of poultry health, and the literature was also used to validate attributes and attribute levels together with their definitions so that they could be used in the final processing of choice profiles and the survey. Two focus group discussions were also conducted in the study area, involving farmers and livestock agency workers of the district to gauge the practicality of communicating identified attributes and attribute levels to farmers. The discussions helped to identify five attributes and their levels for the choice experiment. The attributes were the route of vaccine administration, delivery mechanism, efficacy level, reduction in severity, and price. In village poultry, a possible route of NCD vaccine administration could include using an aerosol spray, giving the vaccine with feed and giving the vaccine with water. Hence, this attribute had three levels in the experimental design. NCD vaccine can be delivered in a number of ways in village poultry. Two ways were considered for this study based on past experience and literature. These were delivery by a veterinary technician (Development Agent in the village working on animal health) and delivery by trained farmers. Possible efficacy levels of the vaccine scenarios were given three attribute levels, based on findings from past studies. Efficacy in this study refers to effectiveness of the vaccine measured by proportion of chicken that develop immunity to NCD. Reduction in severity, which refers to a reduction in mortality of birds due to NCD once it has occurred, was also given three attribute levels in the experimental design. The final attribute of the vaccine profiles was the cost of the vaccine, which is for three vials of vaccine per annum, as the whole vaccine programme was designed to be administered three times a year for optimal control of the disease, considering the village poultry production system.
For analysis of data, the monetary attribute; the cost of NCD vaccine for three times in a year, was included in the models as a continuous variable with their actual levels. All other attributes of the designed NCD vaccine programme were treated as discrete variables. Therefore, for each attribute in the DCE with levels 𝐿, we created 𝐿 − 1 discrete variables to measure nonlinear effects in the trait levels confounding effect in the grand mean, as
suggested by Hensher et al. (2005). A summary of attributes and attribute levels used in the final designing process is presented in Table 5.1.
Table 5.1 Attributes and attribute levels in the choice experiment
No. Vaccine Attribute Attribute level Reference level
1 Efficacy 1. 30 %
2. 50 % 3. 70 %
50 percent
2 Delivery mechanism 1. By vet. technician 2. By trained farmers
By trained farmer
3 Route of administration 1. With water 2. With feed 3. Aerosol Spray
Aerosol spray
4 Reduction in severity after outbreak
1. 20 % 2. 40 % 3. 60 %
40 percent
5 Price of vaccine 1. ETB 60.00 2. ETB 80.00 3. ETB 100.00
Used as continuous
Experimental designs commonly used in valuation studies are orthogonal fractional factorial designs that aim to ensure statistical independence among the attributes. Preserving orthogonality at any cost can lead to decreased efficiency (Kuhfeld, 2010). However, the aim of experimental design is to create an efficient design that maximises the information in the experiment. Therefore, the use of information-efficient designs have been recommended (Kuhfeld, 2010) as these capture the maximum amount of information by minimizing the asymptotic joint confidence sphere surrounding the parameter estimates, although it is not necessarily orthogonal (Kanninen, 2002). Following Kuhfeld (2010) the more comprehensive experimental design approach, information-efficient and D-efficient, which generated a statistically efficient design with an SAS algorithm was employed in this study. The unlabeled DCE was designed to produce NCD vaccine programme profiles using the identified characteristics/attributes of vaccine. There are 162 (34*2) possible ways to combine the five selected vaccine attributes and their levels (see Table 5.1) to produce vaccine programme profiles. Using all of the possible profiles is
cognitively too challenging for respondents to produce a meaningful choice exercise for most practical situations. Therefore, 36 profiles were created out of the 162 possible combinations by applying the SAS algorithm D-efficiency criterion. The final design had a D-efficiency of 99.8, suggesting that the variance matrix should generate reliable estimates. These profiles were randomly classified into 18 vaccine profiles, each choice set having two profiles, and blocked into three. Hence each respondent was presented with six choice sets. An opt-out option was included to each choice set to avoid forced choice, so that the DCE is consistent with utility maximization and demand theory (Bateman et al., 2002). Accordingly, respondents were presented with six vaccine choice sets, each containing three alternatives: two vaccine profiles and opt-out option. Choice sets were supplemented by visual aids (pictures) to help communicate information about attribute levels.
The study was approved by the University of Liverpool Committee on Research Ethics (reference-VREC76) and the survey was conducted accordingly. The DCE survey information statement read to farmers described that the survey would involve choice of hypothetical NCD vaccine programmes that would be administered three times yearly (i.e. the cost of vaccine service is for three doses or vials of NCD vaccine) among other things. Prior to the formal survey, the questionnaire was extensively piloted and pre-tested among individuals and in focus group discussions during early January 2013. The pilot survey for the DCE indicated that communicating attribute and attribute levels was workable and respondents could complete the choice exercise. Following the feedback from pilot survey, logical ordering of the questionnaire presentation was re-arranged to maximise respondent attention for the choice task. The formal survey was conducted in February and March 2013.This DCE survey was administered to 450 farmers drawn from a list of farm households in the four ‘Gandas’(the lowest administrative unit in government structure consisting of several villages), provided by local development agents, employing sampling with probability proportional to size. The four ‘Gandas’ were selected by the project from two different market channels in the district. The survey was conducted by well-trained and experienced enumerators in close supervision with the researchers.