Sampling depends on research questions and the circumstances of the research project—for instance, whether this is to be a cross-sectional (quantitative survey of as many homegardens as possible at one point in time) or a longitudinal study (study of processes and dynamism in a few locations for a longer period of time). These obviously require different sampling strategies. For our predominantly cross-sectional projects, we use a four-step approach.
1. Whenever possible, we prefer to use random sampling, for purposes of more rigorous statistical analysis and to allow the generalization of conclusions about larger populations. In a small village of indigenous farmers, as in the Chiapas and Kalimantan research, in which we could easily count and mark the units (homegardens), or in the case of Tyrol, where we have access to a complete list of all farmers in the study area, random sampling is relatively easy.
2. We use the randomly selected gardeners to find other informants in a manner similar to what Bernard (2002) calls snowball sampling. We also consider including people and homegardens encountered by chance in the course of a study.
3. We include an explicit historical perspective in our studies; therefore, we search for the eldest persons in the study area who might know something about the history of gardening, and we use the data for qualitative descriptive analysis.
4. Finally, we also contact all persons who have important functions related to gardening in the study area, such as members of gardening associations, extension agents, or owners of nurseries. Even if they are not included in the above samples, they often provide valuable information about gardens, gardening practices, and socioeconomic conditions in the study area. For studies using a diachronic approach, we suggest limiting the sample size but repeating the studies every few years, especially during years of unusual climatic variability or economic crisis, to study gardeners’ responses to stressful situations.