Sampling
If you are conducting a survey or doing interviews, you'll need to decide how many people to include.
If the population in question is large, you will likely choose a sample — the number of people you will survey or interview — from that population to participate. The first step is to consider who and how many will be selected. The sample size depends on the level of detail you are interested in, as well as the level at which you want to draw conclusions. To be able to detect small degrees of change and to be able to generalize information to a larger population (calculating statistical significance) requires a large sample size. Also, when the purpose of the survey is to detect changes produced by a program, you need a certain minimum amount of data. Consult a sampling expert if you intend to detect statistically significant change in a population.
Because interviewing can be labor intensive, it generally involves a more limited sample than a survey. The number of interviews you conduct will depend in part on how many different types or groups of informants you want to include or how diverse their opinions may be. Try to interview enough people from each subgroup to be representative of the group as a whole and to provide a range of perspectives. If you are targeting a specific group and it is a relatively small number of people, you could include them all.
The following are three common ways to select a sample:
-
Simple random sample: Each member of a population has an equal chance of being chosen. The best illustration of a simple random sample is drawing names from a hat.
-
Systematic sample: Similar to the simple random sample method, names are chosen from a list by starting from a randomly selected point and picking names at a standard interval. For example, if you select 15 as the standard interval and randomly start with the 10th name on the list, you would select #10, #25, #40, #55, etc. from the list for your sample.
-
Stratified sample: Divide the population of interest into subgroups based on characteristics (e.g., gender, age, geography, participant, eligible but not participating) and randomly select a sample from each stratum. This method is used when you have specific subgroups that you want to balance, or when you want to compare data across different subgroups.
Another common type of sampling is called convenience sampling, in which subjects are recruited because they are easy to reach. Such samples can be obtained in a variety of ways, such as by choosing the first ten names on a list, selecting students who volunteer for an experiment, or approaching customers who enter a store one morning. Because subjects are not selected randomly and certain groups are excluded (e.g., those who don't volunteer, those not present), convenience sampling is not considered to be representative of an entire population, so it may be difficult to make generalizations based on the findings. However, convenience sampling can be useful if you need to document the presence or occurrence of something specific (e.g., to demonstrate that new mothers do not all suffer from post-partum depression, you only need to find one new mother who does not). Convenience sampling can also be useful in pilot studies, where a random sample may not be needed.
Often, you already know which individuals' opinions would be important to gather and you can start there. You might also use snowball sampling—a method where you start interviewing a core group of people and ask them for suggestions of other knowledgeable people to whom you might talk. If you are working with a hard-to-reach population or have only a limited list to start with, using this method can expand your sample relatively easily. But, as with convenience sampling, this method is also subject to biases and may not be appropriate for generalizing your findings.
Northwest Center for Public Health Practice (2011). Data collection for program evaluation.
Additional Considerations in Data Collection
