Discrete choice experiments (DCEs) are attribute-driven experimental techniques used to elicit consumers’ preferences, with potential to identify underlying preferences more clearly than survey data. DCEs describe behaviours, e.g. food choices, using particular characteristics or ‘attributes’, and must critically be informed by the particular study context. Greater rigour has been called for in implementing and reporting the process of designing DCEs, including attributes and attribute levels (i.a. Coast, 2007). This paper describes the process of and key issues encountered with developing a DCE for investigating food choice responses to a change in price of a staple (maize) in rural Malawi.
To design the DCE, attributes and attribute-levels were derived from household and market surveys conducted in May 2017 in Phalombe and Lilongwe Districts of Malawi. The household survey asked respondents to list foods, and their respective quantities, purchased from local markets by households in the previous seven-day period. Food quantities were elicited using standardised measures such as kilograms, or where non-standardised measures were used (e.g. a ‘bucket’ of maize), the study used two standardised cups. The ‘large cup’ held one litre of water; the ‘small cup’ 500 millilitres of water. A market survey collected market price data of the food products and quantities identified previously. The study estimated that on average, a household consumed food products valued at MK1000.00 (US$1.40) each two to three days. Based on this, a DCE was designed consisting of ten scenarios. Each scenario contained three baskets. Each basket contained maize (the staple), and a maximum of rice (maize substitute), cabbage (rare vegetable), dried fish (a healthy protein food), and the soda ‘Frozy’ (similar to Fanta; an unhealthy product), and costing between MK900 and MK1100. The DCE was piloted three times, with data collectors, in a rural area near Zomba city, and in a poorer region of Zomba.
Findings and Interpretations
Using a modified Federov algorithm in NGENE software, a d-error-minimising efficient design generated two sets of five DCE scenarios each (10 scenarios in total). One set of five scenarios had maize at a high price (MK400/kg); the other had maize at a low price (MK150)/kg). The scenarios were presented to 200 respondents in each district; and in different orders to avoid ordering bias. We will conduct analysis in STATA15 software, using multinomial, mixed and latent class logit models (analysis of pilot data already complete).
Four broad issues were faced: establishing an appropriate model of food choice when it was possible to have only five goods as attributes; constructing baskets that provided sufficient attribute variability; constructing a viable instrument, with appropriate, easily interpreted photographs; and establishing a choice scenario around prices and basket values was clear to respondents. Our piloting exercises provided an opportunity to understand the importance of assumption setting and implementation context.
We identified important issues regarding attribute inclusion and context setting that may be of interest to others seeking to use a DCE approach for understanding food choice. Our results show that the DCE approach can be a useful addition to the tools used by researchers.
Based on information derived from a household and a market survey in two districts of rural Malawi, the research team has designed a DCE to be used in assessing the impact of maize price changes on food choice and dietary diversity. The overall study will make some key contributions to the literature on food choice. First, we found that using a cup to standardise food quantities improved the internal validity of the study findings. Secondly, this is among a small number of studies to have used a DCE approach to study food choices, and the only study of which we are aware to have done this in a low- or middle-income country setting, and to document the process and issues faced with the DCE development. Data for the experiment were collected in January and February 2018, and we are currently analysing the study results. To date, very few studies have used DCE study approaches to elicit food choice preferences, and the growing number of studies expected in this area would likely benefit from awareness of the process, issues, and novel solutions that we have encountered and developed.