1. Describe the primary sampling units (PSU) and secondary sampling units (SSU) used in country C and in Kinshasa province in the study by Burnett et al.

Country C: two-stage stratified cluster sampling design

PSU: clusters, selected on the basis of PPS

SSU: households, selected using systematic random sampling from a list of households in each cluster

Kinshasa province: three-stage cluster survey design in 12 zones

PSU: neighborhoods, selected using systematic probability proportional to estimated size

SSU: households

2. The Kinshasa province survey described in Burnett et al. had 3 sampling stages, what was the third sampling stage (hint: look under “Survey Objectives and Sample Size”?)

The third stage sampling stage corresponds to infants 6-11 months. In HH with more than 1 infant, one is to be randomly selected.

3. Based on the definitions of probability sampling and sampling frame found in the 2015 Vaccination Coverage Survey Reference Manual (section 3.6, section 6.2 and annex A), what do you think of the sampling frames used in country C vs. the sampling frame used in the survey described by Burnett et al.? Describe potential limitations of the frames used and how they may relate to sampling bias

Country C: sampling frame corresponds to the list of Enumeration areas obtained from the National Statistics Office.

Kinshasa: sampling frame initially corresponded to the list of neighborhoods, based on polio microplan data. However, the boundaries of the neighborhoods were unknown and finally random points were selected using Arc GIS.

The sampling frame in Kinshasa has more limits than the one used in Country C. In Kinsahsa the sampling frame did not include a complete list of all HHs as GIS points were instead used. The selection of neighborhoods could not be proportional to population size as the population was unknown. It is not clear how HHs were selected around the GIS points and how the boundaries of the PSU were established. A HH could have been selected in two different neighborhoods. This sampling frame could have also left out some HH or nomadic populations depending on how HHs were selected.

In Country C, HHs were selected inside each cluster using random sampling. Consequently there is less bias from interviewers in the selection of HHs (for instance HHs were no-one is available).

4. In the Kinshasa province survey described in Burnett et al., the expected sample size was not reached. The authors describe two potential factors that may have contributed to this. How could have this been prevented? What are the main consequences of not reaching the expected sample size?

Completing a HH census before the survey would have avoided the problem of not reaching the sample size. A census is time consuming, but is a preliminary task that ensures the correct sampling of HHs, in number and also by indicating the specific HHs to be visited by the interviewers.

The main consequence of not reaching the sample size is a loss of power, in this case the power to detect a 20% difference in the proportion of infants 6-11 months of age with a vaccination status that was up-to-date for age between HH questionnaires and HF records.