National and subnational estimates, principles and purpose of stratification

Surveys designed to produce single representative national estimates of, for example, micronutrient status, anaemia, or intervention coverage provide useful information for reporting the national status and possibly for comparing that status to the results of a previous survey. However, national estimates are generally not useful in determining where a deficiency is most prevalent or where an intervention is performing poorly. A more common option is to design a survey that will produce representative estimates for two or more subnational areas, referred to as “strata.” A primary purpose of stratification is to improve the precision (reliability) of the survey estimates for areas of programmatic interest.

Stratification partitions the national listing of PSUs into mutually exclusive and collectively exhaustive subgroups. This means that there is no overlap between the populations of different strata, and the population of all strata combined represents the national population. Separate samples are then selected from each stratum to produce representative data for each selected subpopulation. This would show for example, differences in the prevalence of a specific micronutrient deficiency by province or by urban/rural location.

National estimates can be obtained by combining strata results after adjustment as appropriate. This is described in Module 15, Data processing and analysis).

Two important considerations in defining strata are:

• Familiarity with where and how policy and programme decisions are made. In countries with a decentralized administration, stratification should make it easier to provide data for decision-making at the appropriate subnational level. This may be less important for large scale food fortification which tends to be mandated at the national level.
• The expected status of the primary variable of interest for the survey. This should be as different as possible (heterogeneous) across the various strata. For example, if it is known that the status of the primary variable and any intervention is similar (relatively homogeneous) across two subregions, then these two subregions could be combined as one stratum. Note that this also depends on available information about the homogeneity of other variables of interest.

When a survey is being designed without recent data on the status of micronutrients or related interventions of interest, decisions about stratification could be based on geography, on other factors known to be associated with the primary variable, or on administrative areas where programme decisions are made.

Box 4.2 summarizes these points and additional factors to consider in the selection of strata.

Box 4.2 Main factors to consider when using stratified sampling design

1. What is the basis of the stratification, and what characteristics should be used to subdivide the population into strata?
In the case of micronutrient surveys, stratification could be based on:
• Known or suspected differences in the etiology and magnitude of micronutrient deficiencies according to:
  • Residence type, for example urban and rural
  • Geographical features, such as inland and coastal, mountains and lowlands
  • Administrative factors, such as districts with smoothly functioning health care systems compared to districts subject to conflict or other disruptive factors.
• Intervention-related issues, for example, if implementation of an intervention to improve micronutrient status is known to face specific challenges in geographically defined areas, then these areas should be considered as separate strata so that representative, programmatically useful data are obtained for each stratum.
• Programme/policy decision-making administrative units
• Stratification used in previous surveys. If the planned survey is a follow-up, using the same stratification will allow a comparison of stratified data between surveys.
2. The number of strata to be constructed and the stratum boundaries to be used relate to the basis for stratification.
3. The number of observations to be collected in each stratum relates to the desired precision within each stratum and nationally.
   National estimates (made up of all the strata) will have the largest sample size and tightest precision compared to stratum-specific estimates.
4. The available budget:
   A separate survey sample needs to be drawn for each stratum. Therefore, total resource requirements are dictated by the number of strata and the selected precision for the main indicator estimates in the survey. Some strata may require a larger budget if there are challenges related to logistics, terrain access issues, or insecurity.

Specifically defining the strata usually takes place during the survey design stage and is known as “explicit stratification.” Another type of stratification is “implicit stratification”. This relates to the order in which PSUs are listed prior to selection and is referred to briefly in Module 6, Selecting clusters.

The present module is based on considerations for the design of a national survey. However, the same principles apply where a survey aims to obtain information from a subnational area only. If micronutrient status or other indicators of interest vary considerably within that area, then stratification along lines that divide the population of the area into groups with more similar status should be considered.

During the survey planning stage, a range of sample sizes with associated precision for the indicator estimates and expected budget are calculated. When combined with information on survey objectives and available resources, these can be used to guide decisions about design. You can find an abbreviated example of this process in Box 4.2, and a detailed description of the process in Module 5, Sample size.