standard sampling equation to determine a sample size
Use the standard sampling equation to determine a sample size
sufficient for generalizing results to represent the target
In determining an appropriate sample size (i.e., the number of units to be surveyed),
consider the sampling methodology (i.e., how sampling units, such as households,
will be selected) and the analysis plan for the data collected. Discuss these
considerations with a technical M&E staff person and your head of office or head of
programming, as these will ultimately shape the framework for your survey and
Calculate the sample size based on the confidence level and level of standard error
(also known as the confidence interval) appropriate for your survey and whether
you will cluster or stratify your sample. Each of these terms is defined and explained
below. Given the possible variations,
What confidence level is acceptable given your information needs? It is
customary to use a 95 percent confidence level.
What level of standard error is acceptable for the survey? Common levels of
standard error are +/-6 percent and +/-7 percent. Aim for the minimal level of
error that is feasible given time or logistical constraints and project survey
Monitoring data generally use a higher level of error and a smaller sample size. The
higher level of error is appropriate for monitoring surveys because they are conducted
repeatedly and must produce quick results to feed into ongoing project management
Interpret survey results based on the confidence level and the level of standard error
selected. If the survey results stated that 55 percent of households were displaced by
the flood, with a level of standard error of +/-6 percent and a confidence level of 95
percent, this means that you can be 95 percent confident that the actual proportion of
displaced households was between 49 percent and 61 percent (within +/-6 percentage
points of 55 percent).
Will you cluster your sample? Clustering a sample refers to first selecting
clusters (such as communities or schools) and then selecting the actual units
(households or schoolchildren) from within these clusters. Clustering a sample
usually reduces the time required for fieldwork and travel time, but requires an
increased sample size to account for the error it introduces. It is advisable to
cluster your sample if:
o You do not have a complete list of all sampling units in your sample
population (e.g., a complete list of all households in your targeted
o Conducting fieldwork within a few smaller geographic areas would save
considerable time and resources.
Will you stratify your sample? Stratified samples allow for statistical
comparisons between key subgroups. Stratification requires an increased sample
size so that each subgroup can be adequately represented. Common comparisons
are between socioeconomic groups, districts or states, flood-affected and
drought-affected areas, project participants and nonparticipants, and men and
women. Are any comparisons between subgroups required by your analysis
Remember—there is no magic 10-percent sampling rule. It is important to note that
the sample size is not related to the size of the population being sampled. A frequent
mistake is to conduct surveys among 10 percent of a given population; in fact, it is
likely that 10 percent of the population is either too many or too few households. With
too many households, the survey is using excessive resources and time; with too few
households, the sample will not adequately represent the population.
Account for nonresponse. Due to challenges in data collection, it is common practice
to increase the sample size by 10 percent to account for nonresponse. Nonresponse
may be due to difficulty in locating all the selected units (e.g., individuals or
households), to unwillingness of a unit to respond, or to data collection errors.
If the number of sampling units (e.g., households) is less than the calculated sample
size, include all units.
Document the confidence level and the level of standard error used in the methodology
section of your report so you can interpret the results within these boundaries.