Annex 6: Comparing Random and Purposive Sampling Methods

Names

Scientific sampling, random sampling, QUANT sampling

Purposeful sampling

Nonrandom sampling

QUAL sampling

Mixed methods

Overall purpose

To generate a sample that will address QUANT research questions.

To generate a sample that will address QUAL research questions.

To generate a sample that will address research questions.

Generalizability

To generalize findings to a defined target population (such as all low-income households).

More limited transferability based on judgmental considerations. Some QUAL evaluators question the appropriateness of generalization from research findings.

Focus on external validity issues for some components of the evaluation and transferability issues for others.

Techniques

Simple, stratified, cluster*, random-route*, quota sampling.*

There are a wide range of sampling techniques including: convenience; typical case and representative samples;  quota*; random- route*; cluster*; unique, deviant, extreme and negative cases; range; snowball;  critical case; and reputational samples.

All of the techniques used for probability and purposive sampling.

Rationale for selecting cases

Selection of cases that are together representative of the total population.  Random selection means that many cases will have a low information value.

To address specific purposes related to the research questions. Each case is selected to address a particular set of questions so that each case has a high information content/value. 

Representativeness for some components and depth and richness for others.

Sample size

For social science research samples will often include 500+ cases. For psychological and some kinds of medical and educational research the minimum size will often be set at 50+.

Usually small, often fewer than 30 cases.

There will often be multiple samples with size being determined by whether the purpose of a particular component is more QUANT or QUAL. More rigorous size estimation procedures may be used for selecting quota samples, case studies, participants in focus groups, etc. to ensure a minimum level of statistical representativeness.

Determinants of sample size

Sample size is estimated to achieve a given level of statistical precision defined in terms of the estimated effect size and required power of the test.

Size is determined judgmental and where time and resources permit researchers continue to sample more cases until no new information is being obtained.

Randomization size estimation procedures will be used for QUANT  components. The size of purposive samples will be determined by combining normal QUAL criteria with samples large enough to ensure general comparability with QUANT samples (when required). 

Depth/breadth of information per unit

Focuses on breadth of information and the ability to provide estimates for the total population of interest.

Focuses on depth of information.

Combining depth and breadth.

When sample is selected

Before data collection begins.

Before the study begins, during data collection or after the initial round of data collection has been completed. Sequential sampling can continue for as long as deemed necessary by the researcher.

The overall sampling strategy will normally be defined before data collection begins, but sub-samples may be identified as the analysis evolves. 

Selection method

Using statistical formula

Judgmental

An overall sampling strategy must be developed with different sub-samples being generated from this. It is essential to ensure that procedures ensure the different samples are comparable.

Sampling frame

Formal sampling frame covering the whole population of interest. For considerations of time and cost an existing sampling frame (such as a list of families registered with an agency) may be used.

Judgmental. Based on advice from experts combined with researchers' judgment and possibly checked through rapid exploratory studies (for example when the sampling unit is the community).

A master sampling frame from which different sub-samples are generated.

Note: * These techniques may be considered as random or purposive depending on how they are applied.

Source: Adapted from Teddlie and Tashakkori 2008 Table 8.1