Stratified random sampling is a probabilistic sampling option. The first step in stratified random sampling is to split the population into strata, i.e. sections or segments. The strata are chosen to divide a population into important categories relevant to the research interest.
For example, if interested in school achievement we may want to first split schools into rural, urban, and suburban as school achievement on average may be quite distinct between these regions. The second step is to take a simple random sample within each stratum. This way a randomised probabilistic sample is selected within each stratum. Each strata should be mutually exclusive (i.e. every element in the population can be assigned to only one stratum), and no population element can be excluded in the construction of strata.
Stratified random sampling is used instead of simple random sampling when the categories of the strata are thought to be too distinct and too important to the research interest, and/or when investigators wish to oversample a particularly small group of interest. (Investigators oversample in the smaller strata in order to increase their sample size, which is necessary to conduct proper statistical analyses.) In practice, stratified random sampling along with other more complex sampling techniques are employed in large-scale surveys, especially governmental censes, to reduce some of the logistical costs associated with collecting information from a sample.
In Grady et al (2008) the study examines the views of research participants in a hypothetical HIV vaccine study. In particular, it examines views regarding individual compensation for participation and post-trial benefits to the community in which the trial took place. Here, the total population is stratified according to region – peri-urban, rural, and deep rural.
Advice for CHOOSING this option (tips and traps)
- Ask about the stratification procedure. What are the segments? Why these segments and not others? What is the sample size we need from each segment to conduct our statistical analyses?
- The same rules apply as in simple random sampling. Ensure that those individuals/observations that are selected are surveyed. The response rate must be high in order to create statistically valid inferences.
Advice for USING this option (tips and traps)
- Consider whether your population needs to be split into distinct categories before taking a random sample.
- If you are afraid that you will not obtain enough people (observations) for one of your key categories in a simple random sample, then a stratified random sample is a viable alternative.
- Segment your population into the predetermined key categories of interest. Within each category take a random sample.
- Note that this requires knowing the population of interest. For example, if interested in distance to clean water for populated areas in a region, you must first have a list of all of these regions. You could segment on rural, urban, suburban, and then take a random sample within each.
- Module 11: Sampling: Taken from Volume 1 of a course book by the International Development Research Center (IDRC). Module 11 take your step by steps through the process of sampling for research.
Agresti, A. and Finlay, B. (2008) Statistical Options for the Social Sciences, 4th edition. (Upper Saddle River, NJ: Prentice Hall).
Grady, C. et al. (2008) “Research Benefits for Hypothetical HIV Vaccine Trials: The Views of Ugandans in the Rakai District.” IRB: Ethics and Human Research 30 (2): 1-7