Sampling Methods in Research: Probability vs Non-Probability

Sampling is the bridge between your population of interest and the participants you can actually study. The choice of sampling method shapes what you can claim from your data, what statistical tests are valid, and how examiners will judge your methodology chapter. This guide walks through both families, names every common method, and gives you a decision tree for matching method to research question.

Population, Sampling Frame, Sample

Three terms you need to use precisely:

• Population — everyone or everything your research question is about (“first-year undergraduates at Canadian universities”).
• Sampling frame — the operational list from which you actually draw participants (“the student registries at five GTA universities”).
• Sample — the people who consent and participate (“120 students”).

The gap between population and sampling frame is your coverage error. The gap between sampling frame and sample is your nonresponse error. Both matter; both need to be reported.

Probability Sampling Methods

Simple random sampling

Every member of the sampling frame has equal probability of selection. Implementation: list everyone, assign a random number, take the first N. Gold standard for unbiased estimation but rarely practical because you need a complete frame. Best for: small populations with an accessible list.

Stratified random sampling

Divide the population into mutually-exclusive strata (e.g., year of study, gender, institution). Randomly sample within each stratum. Strata can be sampled proportionally or oversampled for small groups. Gives more precise estimates per sub-group than simple random. Best for: studies where you need to compare across known sub-populations.

Systematic sampling

Pick every k-th member from an ordered list. If your list has 10,000 and you want 100, take every 100th (k = 100). Random starting point. Cheaper than simple random but vulnerable if there’s a periodicity in the list.

Cluster sampling

Randomly select clusters (whole groups), then study everyone or a sub-sample within selected clusters. Standard when individual sampling is impossible (e.g., randomly select 10 classrooms, then survey every student in those classrooms). Sacrifices some precision for cost. Best for: nationally-representative studies, school-based research, geographically dispersed populations.

Multistage sampling

Combine clustering with stratification across multiple stages (e.g., stage 1 random sample of provinces, stage 2 random sample of schools, stage 3 random sample of students). Standard in large national surveys (Statistics Canada PISA, Canadian Community Health Survey).

Non-Probability Sampling Methods

Purposive (judgemental) sampling

Researcher selects participants based on specific characteristics relevant to the research question. Variants: maximum-variation (deliberate diversity), homogeneous (deliberate similarity), critical-case (extreme or revelatory cases), typical-case. Standard in qualitative research.

Snowball sampling

Start with a few participants meeting your criteria; ask them to refer others. Useful for hidden populations (sex workers, undocumented migrants, members of stigmatised communities). Biased toward connected individuals; mitigated by respondent-driven sampling (RDS) which adjusts for network structure.

Quota sampling

Set quotas for sub-groups (e.g., 50 women, 50 men, 25 each from four provinces) then recruit until quotas are filled. Faster than stratified random but doesn’t allow probability-based inference because selection within quotas is convenience-based.

Convenience sampling

Recruit whoever is easily accessible (your university’s subject pool, social-media solicitations, MTurk, Prolific). Cheapest method. Generalisation is severely limited — results apply only to your sample, not to a broader population. Acceptable for pilot studies, exploratory research, and methodological development. Examiners flag overuse.

Hire A Professional Academic Writer

• PhD-qualified Canadian writer
• AI-free, plagiarism-free guaranteed
• On-time delivery, every time
• Unlimited free revisions 14 days
• 24/7 chat support

Order Now Learn More

Use code FIRST10 for 10% off your first order.

Probability vs Non-Probability: Decision Matrix

Sample Size: How Many Is Enough?

For probability samples answering quantitative questions, calculate sample size via power analysis (see our hypothesis-testing guide). Typical Master’s thesis: N = 100–300. PhD: N = 300–1,000+ depending on effect size and design.

For non-probability qualitative samples, the criterion is saturation — you stop recruiting when new interviews produce no new themes. Empirical work suggests saturation typically occurs at:

• 12–15 interviews for homogeneous populations exploring a focused question.
• 20–30 interviews for diverse populations or broad questions.
• 6–8 focus groups (4–8 participants each).

Smaller numbers are acceptable for phenomenological studies (5–8 deep interviews) and case studies (1–6 cases).

Common Sampling Errors

• Mismatched method — using convenience sampling and claiming national generalisation. Examiners will catch this.
• Undefined frame — saying “a random sample of Canadian students” without naming the registry, list, or recruitment channel.
• Self-selection bias unacknowledged — web-survey respondents differ from non-respondents in systematic ways. Report response rates and discuss potential bias.
• Survivor bias — sampling only those who completed a programme misses dropouts who are often the more informative group.
• WEIRD samples — Western, Educated, Industrialised, Rich, Democratic. Acknowledge if your sample is WEIRD; do not claim universal results.

Reporting Sampling in Your Dissertation

In the methodology chapter, dedicate one section to sampling. State:

1. Target population.
2. Sampling frame (specific list, registry, or platform).
3. Sampling method with full name and any sub-variant.
4. Inclusion and exclusion criteria.
5. Recruitment procedure (channels, incentives, timing).
6. Final sample size with justification (power analysis or saturation).
7. Response rate (probability) or saturation criterion (qualitative).
8. Demographic characteristics of the sample.
9. Discussion of bias and generalisation limits.

Frequently Asked Questions