Skip to main content

Sampling in Quality Improvement

Types of sampling include:

CENSUS. Sometimes a population will be sufficiently small enough that every member can be sampled.

When the population is too large, you want to choose a representative sample. There are probability and non-probability methods for sampling. The aim is to minimize the introduction of bias, so that the sample closely matches the population.

SIMPLE RANDOM SAMPLING allows every member of a population to have an equal chance of being included in the sample (a probability sampling method). They can be chosen by computerized random number generators, tables of random numbers, or you can simply draw names/numbers out of a hat.

STRATIFIED RANDOM SAMPLING divides the population into groups or strata, based on knowledge of the population—e.g., types of patients, gender, race, etc. The proportion of cases randomly drawn within each strata should be the same as in the larger population and drawn using a random method (probability sampling).

SYSTEMATIC SAMPLING, also called the Nth name selection technique, draws every Nth record from a population. As long as the list does not contain a hidden order, this technique is as good as random sampling.

JUDGMENT or RATIONAL SAMPLING is used more frequently in quality improvement studies, relying on the knowledge of those with process knowledge. In this mode, data is looked at over time in a control or Shewhart chart (called analytical studies). Data are selected in a non-random method (non-probability sampling), taking small repeated samples from a process over time. This could be daily or monthly, for example, drawing a small number of cases (4-7 recommended) each time, for a minimum of 25 data points (Raymond Carey, Improving Healthcare with Control Charts, 2003). The problem with this method is that the sampling error is unknown.

CONVENIENCE SAMPLING is just that—you look at easy-to-view cases to allow you to get a feel for your population, inexpensively. It is often used in the exploratory stage of a study, just to get some gross estimates.

For other methods of sampling, see

Sampling saves time and money. Collect only as much data as you need to answer your question, trying not to introduce bias into your observations.