A sentence that turns up in methods chapters everywhere: “A random sampling method was used; the questionnaire was distributed via Google Forms to student groups on social media.” The sentence contradicts itself. Whoever saw the post, had time and cared about the topic responded. That is a self-selected sample, and nothing about it is random.
The interesting part is that convenience samples are nothing to be ashamed of. A great deal of good research uses them. The trouble lies only in mislabelling them and drawing conclusions the sample cannot carry. Know what kind of sample you have and what it buys you, and your methods chapter will stand up to scrutiny.
Two families of sampling and the real difference
Probability sampling means every member of the population has a known, non-zero chance of selection. The prerequisite is a sampling frame: a list of students, households or registered firms. Without a frame there is no probability sample, however carefully you randomise among the people you happen to meet.
Non-probability sampling has no such known chance. It is cheaper and faster, and it reaches groups for which no list exists: gig workers, users of a particular app, people who have lived through a specific event. The cost is that you cannot use sampling theory to attach a margin of error to population estimates.
Sampling methods compared
| Method | How it works | Works well when | Main risk |
|---|---|---|---|
| Simple random | Random draw from the frame | Small population, complete list | Laborious if population is dispersed |
| Stratified | Split into strata, draw randomly within each | Small subgroups need enough cases | Weights needed when combining |
| Cluster | Randomly select classes, villages or schools, then survey within | No list of individuals | Design effect inflates required n |
| Convenience | Whoever is reachable | Piloting, exploratory work | Self-selection, no generalisation |
| Quota | Convenience, but filled to target proportions | Matching population composition | Matches only the quota variables |
| Purposive | Select information-rich cases | Qualitative and case studies | Researcher selection bias |
| Snowball | Participants refer others | Hard-to-reach groups | Sample trapped in one network |
When a convenience sample is enough
There are three situations in which a convenience sample is perfectly defensible:
- Testing relationships between variables, not estimating prevalence. If your question is whether stress is associated with sleep quality, the relationship may be reasonably stable across samples even when the average level of each variable is not.
- Piloting and scale development. Thirty to fifty people to test wording, completion time and preliminary reliability.
- Randomised experiments. Participants may come from a convenience pool, but assignment to treatment or control is random, and internal validity comes from that assignment step.
By contrast, if your title contains “prevalence”, “proportion” or “level” in a specific population, a convenience sample will rarely do. An online survey on “the proportion of students with part-time jobs” is biased from the start: working students may be too busy to respond, or more motivated to. Nobody knows which way the bias runs.
Strengthening a non-probability sample cheaply
- Set quotas on two or three variables whose population distribution you know, such as gender, year of study and faculty. Close each cell once it is full.
- Diversify recruitment channels: not one online group but departmental email, in-person classes and student societies. Record each respondent’s channel and check whether results differ by channel.
- Compare sample and population on variables with official figures. A simple two-column table persuades reviewers more than any paragraph of justification.
- Track the response funnel where possible: how many were invited, opened the survey and completed it.
Sampling in qualitative research
Qualitative sampling aims at information richness, not statistical representativeness. Purposive sampling comes in several forms: maximum variation (an urban school and a remote rural one), typical cases, deviant cases (the unusually successful), or criterion sampling. State the criterion explicitly — “primary teachers with at least three years of multigrade teaching” — rather than “suitable participants”.
With snowball sampling, start from several independent seeds, three to five people who do not know each other, so the sample does not circle within one friendship group.
Frequent errors in methods chapters
- Calling a voluntary online sample “random”.
- Claiming the sample is “highly representative” without a single comparison table.
- Calculating sample size with a probability-sampling formula, then collecting a convenience sample as if a big number cured selection bias.
- Not reporting how many responses were removed for missing data, straight-lining or implausibly fast completion.
The third point deserves emphasis. A convenience sample of 2,000 is not more representative than one of 200. It simply produces biased estimates with greater precision.
A model limitations paragraph
“Data came from a convenience sample of 312 students recruited through departmental email and three online study groups. Compared with university enrolment figures, the sample contains more women (68% versus 55%) and over-represents business programmes. Descriptive proportions should therefore not be generalised to the whole student body; associations between variables were re-estimated separately by gender and showed a similar pattern.”
That paragraph hides nothing, yet it shows the author checked. Your next step: write down the sampling frame you have, or admit you have none, name your sampling method correctly, and prepare the sample-versus-population comparison table on at least two variables before you begin collecting data.
Câu hỏi thường gặp
Is an online survey a random sample?
Usually not. If anyone who sees a public link can respond, it is a self-selected convenience sample. It approaches a probability sample only if you randomly draw people from a list and send each a personal invitation.
Can I use convenience sampling in a dissertation?
Yes, when your question focuses on relationships between variables or is exploratory, and you name the method correctly and discuss its limits. Avoid using it to estimate prevalence in a defined population.
What is the difference between stratified and quota sampling?
Both divide the population into groups. Stratified sampling draws randomly within each group from a list, while quota sampling fills each group by convenience, so it offers no basis for calculating sampling error.
When should I use snowball sampling?
Use it for hard-to-reach or hidden populations where no list exists. Begin with several unconnected seed participants to reduce the risk of recruiting only one social network.
How do I show my sample is representative?
Compare your sample with known population figures on variables such as age, gender or field of study. Where they match you have partial support; where they differ, say so and consider weighting.
