Considerations for household-based surveys
For individual-level indicators in a household-based survey where households are randomly selected from a complete listing of households in the cluster, the number of households to visit depends on:
- the number of individuals needed to obtain estimates with sufficient precision for the indicator within that population group;
- the average size of a household;
- the number of individuals from the population group of interest expected within each household; and
- the expected response rate.
More detail on how these factors are accounted for is described in Box 5.2.
The decision about selecting the survey sample from all households or from households that meet a specified criterion requires expert consultation to clearly understand the advantages and disadvantages of each approach and their effect on interpreting the resulting data.
If the number of survey households to include in the sample is based on the numbers required for preschool-age children, then it may not be necessary to collect data from all eligible WRA in the households. In such a case, it may be best to do a random selection of WRA. Possible methods include the random selection of one WRA per household, or selecting all WRA from every third household. The approach needs to be decided at the survey design stage and cannot be changed during the fieldwork. In all cases, it is important to document the total number of eligible individuals in each household, because this information will be needed to determine the sampling weight at the data management stage.
It is also important to keep in mind that not all information needs to be collected on every survey subject or household. For example, it may be reasonable to perform more expensive tests on a subsample of biological samples, such as every second survey participant within one population group, so long as minimum sample size requirements for that indicator are satisfied.
After making the initial calculations of sample sizes for the desired precision at the stratum level, decisions need to be made about feasibility. Where one population group (in this case children under 5 years of age) requires visiting significantly more households, then the following can be considered:
1) identifying in advance households with this population group and randomly selecting as many of these as are required to find 370 children (this would bias other indicators to be representative of households with children under 5).
2) accepting a reduced stratum level precision for estimates of 10% (stunting among children under 5), 11% (household coverage of adequately iodized salt), and 13% (anaemia and iron deficiency among children 6-59 months). (This would reduce the required number of households to approximately 250 per stratum).
3) it may be determined that reliable estimates for indicators among this group are only possible at the national level rather than at sub-national (e.g., stratum) levels.
Box 5.2. Household- versus individual-level indicators
For household-level indicators, the number of households to visit will be determined by the number of completed household interviews (and, where included, food tests or samples) required to obtain estimates with the desired precision, accounting for the expected number of occupied households and the response rate for interview and sample collection. For example, 95% of selected households may be occupied and have an adult household member willing to answer questions about the household, while food sample collection may only be feasible in a lower proportion due to non-availability of the food item and/or non-response for collection.
For individual-level indicators in a household-based survey with a random selection of households from a complete household listing, the number of households to visit depends on four factors:
- the required sample size for the number of individuals within a specific population group
- the average household size
- the proportion of the population comprising the population group of interest
- the expected household and individual response rates for population-group specific interviews and for sample collection.
Multiplying the average household size by the proportion of the population group of interest in the national population provides the average number of eligible individuals expected per household. The number of households that need to be visited to achieve the required sample size can then be calculated from this, taking into consideration the expected response rate.
The final number of households to be visited to obtain data for the required number of subjects from a specific population group may be calculated by dividing the sample size (in this case 766) by the product of: [the average household size (3.9) multiplied by the proportion of the specific population group in the population (0.31) multiplied by the expected response rate for households (90%) and individuals (85%)]. This must also take into account the design effect, DEFF, which is a measure of the homogeneity within a cluster and the variability between clusters.
As an example, where a survey sample size for non-pregnant women of reproductive age (WRA) in a geographic area has been determined to be 766 (already accounting for DEFF), the average household size is 3.9, the proportion of non-pregnant WRA in the population is 0.31, the household response rate is expected to be 0.9 (90%), and the individual response rate for consenting to biological sample collection is expected to be 0.85.
The number of this population group (WRA) per household would be expected to be 3.9 x 0.31 = 1.2. Therefore, the team can expect to obtain data from more than one eligible woman per household on average. However, the response rate also needs to be considered. The final number of households to visit to obtain information or samples from 766 WRA, based on the information above, could be calculated as:
