In countries where sampling frames of fishing communities at the district or province-level were available or could be created at low cost, FAO asked the national project directors to develop probability sample designs. The objective was to sample five percent - at least five coastal fishing communities - from a sampling frame consisting of fishing communities in a particular district or province. Adjacent to each sampled fishing community, a nearby agricultural community would be identified and selected. Depending on the size of the selected communities, about 5–10 percent - at least 100 households were to be sampled.
Consequently, information obtained from a probability sample survey of households in fishing communities was representative of all fishing community households in the particular district or province. However, the above-mentioned selection method implies that results obtained from households in agricultural communities would not be representative for households in all the agricultural communities in that same district or province. Nevertheless, this was not considered significant as the reason for including agricultural communities was solely to obtain survey data from a comparison group of households living in the same environment, but with livelihoods unrelated to fisheries.
Guidelines for the sampling of fishing communities and households within fishing communities entailed sampling procedures for simple two-stage self-weighting sample designs. Self-weighting designs were proposed to avoid the computation and use of sample-design weights in the survey analysis stage. In a self-weighting design, the selection procedures were such that households that were in the sample had the same overall (i.e. ex-post) selection probability, irrespective of the number of households in the communities from which households were sampled. The guidelines proposed two model sample designs:
If a district- or province-level sampling frame could inexpensively include current population estimates for fishing communities, five or more communities could be sampled with probabilities proportional to the population (PPS) of communities in the sampling frame. This would be accomplished by applying the systematic selection method to a geographically sorted list of communities with reported cumulative population estimates for each village. Often the population of a community was unknown, because village census registers did not exist or were out of date. In that case, the last existing population census was chosen and the selection of communities was in proportion to their estimated population size (PPES). In the second stage, a fixed number of households (say 20) was selected in sampled communities, after the communities were screened, for the total number and location of households. If the proportionate distribution of population of sampled communities at the time of the last census was found to differ significantly from the proportionate population distribution derived from the screening of the sampled communities at the time of the survey, this difference was taken into account to correct the initial selection probabilities. The PPES sampling strategy results in a self-weighting design whereby the overall selection probabilities of households are the same, irrespective of the size of the community from which they come.
If a district- or province-level sampling frame existed with only the names of fishing communities but without their population size estimates, a somewhat different approach, though self-weighting, was followed: In the first stage, a fixed number of five fishing communities was sampled from the sampling frame of households with the help of a random number table. In the second stage, contact was made with the administration in the selected communities to obtain the total number of households in the community and their location. If such information was not available, a screening census was carried out in the communities to create a sampling frame of households. A fixed percentage of households was selected in the selected fishing communities. Although this method is easier to apply than the first, the disadvantage is that there is no a priori control over total sample size, because the number of households in the communities differ.
In country-specific studies, small sample size has analytical implications because statistically meaningful generalizations at the level of the district or province cannot really be made. In addition, limitations are posed on the kind of research questions that can be answered, particularly those that require creation and interpretation of multi-dimensional tabulations. Such tabulations will produce a number of empty cells and/or cells containing a fewnumber of cases only. Moreover, because the magnitude of variance estimates of averages and proportions is inversely related to sample size, the smaller the sample size, the higher the probability to make so-called “errors of omission” (i.e. Type-2 errors). That is, the error or failure to uncover significant variables, relations and true differences between groups of interest. As far as issues covered by the focus group discussions in the areas studied overlap with those of the household survey, the negative consequences of such errors can to some extent be minimized.
The above-mentioned analytical restrictions owing to small sample size become less significant when survey data feature in research questions involving comparative analyses across countries, in search of common features, differences and patterns of change between fisherfolk and other rural dwellers.
