Both country teams have used the sampling frame suggested in the initial research protocol, although in Indonesia more households than originally intended have been visited. This purposive sampling frame was designed to cover a variety of health care settings. In each country, two districts or provinces were chosen, sampling within each district three settings: one urban, one suburban, and one rural. In each of the three settings two communities were sampled in such a way that they varied in terms of socio-economic profile, relative distance to health facilities, household structure and level of urbanization. In each community a random sample of 60 families with pre-school children (under five years of age) was selected by cluster sampling, bringing the total of households visited in each district to 360. The study's sampling frame is presented in Figure 1.

Figure 1: Original sampling frame, household survey - Uganda & Indonesia

Within the households the person most responsible for health care of the household members (usually the mother) should be interviewed. Pre-school children were selected as the target-group because they are ill most often and they are a target audience of most PHC-programmes. Since only these households have been sampled in both countries, the prevalence rates calculated from the data collected in household questionnaires are not representative of the population as a whole, but of families with pre-school children only.

In Uganda, the study was conducted in Ankole, Western Uganda and in Busoga, Eastern Uganda. In each of these provinces, three settings were selected (urban, semi-rural and remote). A total of 360 households in each province was visited twice. Two weeks after the initial interview, the household was visited again and the prevalence of injection use in the past two week recall period was recorded. This reduces memory bias in respondents as the interviewer can refer to a specific event: the last time (s)he visited the home. In Uganda the data collected during the follow-up visit produced slightly lower, but probably more reliable injection rates. No households were lost in the follow-up.

In Indonesia the study was conducted in two districts, Lebak (West Java) and the island of Lombok (east of Bali). In each district three sub-districts (urban, suburban, rural) were selected. The sampling frame of Figure 1 was not exactly followed by the Indonesian team, resulting in the total number of households amounting to 407 in Lebak and 409 in Lombok. All households were visited once.

There were no serious difficulties in sampling procedures or in conducting the surveys. In Uganda, some problems were reported as to who should be interviewed. In the provisional research protocol mothers had been envisaged as the prospective respondents, while in the actual research phase the head of the household, who was subsequently the key respondent, was often a man. The research team in Uganda solved this problem by asking which member of the household was most responsible for matters of health, and this person was asked to serve as the main respondent during the interview. In the actual interviews, it also quite regularly occurred that other household members 'joined in' the conversation. The Ugandan team decided to view additional information as vital, and included it into the survey results.