Sampling
Sampling is the process of selecting participants from a population for inclusion in a research study. How participants are selected determines the External Validity of findings — whether results can be generalised beyond the study's specific sample.
Sampling Methods
Probability Sampling
Every member of the population has a known, non-zero chance of selection. Enables statistical generalisation.
| Method | Procedure | Strength |
|---|---|---|
| Simple random | Every individual has an equal chance of selection | Unbiased; basis for inferential statistics |
| Stratified random | Population divided into subgroups (strata), then random selection within each | Ensures representation of key subgroups (e.g., L1 background, proficiency level) |
| Cluster | Random selection of groups (e.g., classes, schools), then all members included | Practical when individual random selection is impossible |
| Systematic | Every nth individual selected from a list | Simple to implement; approximates random sampling |
Non-probability Sampling
Selection is not random. Common in applied linguistics because researchers rarely have access to complete population lists.
| Method | Procedure | When used |
|---|---|---|
| Convenience | Whoever is available and willing | The default in most classroom SLA research — researchers use their own classes or accessible institutions |
| Purposive | Participants selected for specific characteristics | Qualitative Research: selecting information-rich cases (e.g., a teacher known for innovative practice) |
| Snowball | Participants recruit other participants | Hard-to-reach populations (e.g., undocumented immigrants learning English) |
| Quota | Non-random selection ensuring specified proportions | Ensuring balanced representation without randomisation |
The Convenience Sampling Problem in SLA
The vast majority of SLA research uses convenience sampling — intact university classes, accessible schools, or volunteer participants. This has significant consequences:
- Population bias — university students (typically young, educated, motivated) are dramatically overrepresented; children, older adults, migrant workers, and low-literacy learners are underrepresented
- Geographic bias — most published SLA research comes from North American, European, and East Asian universities
- Limited generalisability — findings from one institutional context may not transfer to others
- WEIRD problem — samples are disproportionately from Western, Educated, Industrialised, Rich, Democratic societies
Sample Size
Sample size affects statistical power — the ability to detect a real effect. In SLA classroom research:
- Individual classes typically contain 15-30 students
- Many studies compare just two classes (total N = 30-60)
- Plonsky (2013) showed that the median sample size in SLA research is too small to detect medium effect sizes reliably
- Power analysis (determining the required sample size before data collection) is still uncommon but increasingly recommended
Sampling in Qualitative Research
Qualitative Research uses purposive rather than random sampling. The goal is not statistical representativeness but information richness — selecting cases, participants, or sites that can provide the deepest insight into the research question. Sample size is determined by data saturation (the point at which new data no longer generate new themes), not by statistical power calculations.
Key References
- Dörnyei (2007) — sampling in applied linguistics research
- Plonsky (2013) — sample size and power in SLA research
- Patton (2015) — purposive sampling strategies in qualitative inquiry
- Mackey & Gass (2005) — participant selection in SLA research design