Almost every methods chapter contains a sentence like “participants were recruited from a Canadian university.” Almost every committee asks the same three questions about it: which population do your conclusions apply to, how did people get into the sample, and who could never have been in it. This page answers those questions in advance. It separates the four decisions that sit behind one sampling paragraph, works six of those paragraphs out in full by discipline, and lists what a Canadian committee asks about each.
This is the “who and how” half of sampling. The “how many” half — power, saturation and the number itself — is a separate problem, handled in the sample size guide. Get the who and how right first; a sample size calculated for the wrong population is precise and useless.
The paragraphs below are shapes, not claims: no real study, no real statistic, [figure] where a number of yours belongs.
The four decisions, in order
| Decision | The question | What goes wrong if you skip it |
|---|---|---|
| Target population | Who do the conclusions apply to? | You generalise to a group you never had access to |
| Accessible population | Who could you actually reach? | The gap between target and accessible goes unreported, and an examiner finds it |
| Sampling frame | What concrete list or route enumerates them? | “Recruited online” — no one can assess coverage |
| Technique | By what rule did cases enter the sample? | Convenience is presented as if it were random |
The target/accessible distinction is the single most common omission in Canadian master’s theses, and it is cheap to fix: two sentences. Your target population is the group your research question is about. Your accessible population is the subset you could actually approach, given geography, permissions, language and time. If the two are identical, say so; usually they are not, and the difference is where your generalisability argument lives.
Choosing a technique, honestly
| Technique | Use it when | What you must report |
|---|---|---|
| Simple random | You have a complete frame and can draw from it | The frame, the draw method, the seed or procedure |
| Stratified | A known subgroup must be represented proportionally or equally | The strata, why those, and the allocation rule |
| Cluster | Individuals are only reachable through units such as schools or clinics | The clusters, how they were chosen, and that you will account for clustering in analysis |
| Systematic | You have an ordered list and no periodicity in it | The interval, the random start, and a check that the order is not patterned |
| Purposive / criterion | Qualitative work where specific characteristics are required | The criteria, in full, and who applied them |
| Maximum variation | You want the range of a phenomenon, not its average | The dimensions you varied on, and the range achieved |
| Snowball | The population is hidden or hard to reach | The seeds, the chain lengths, and the resulting homogeneity risk |
| Convenience | Nothing better is available | Say the word “convenience”. Do not dress it up. |
The last row matters more than the others. A convenience sample honestly labelled and carefully bounded is publishable. A convenience sample described as “a sample of Canadian nurses” is a generalisability claim you cannot support, and that is a much larger problem than the sampling itself.
Six worked sampling paragraphs
1. Nursing — a single hospital network, quantitative
The target population is registered nurses working in adult acute-care inpatient units in Canada. The accessible population comprises the
[figure]registered nurses employed on the seven adult medical and surgical units of the participating health network as of the study start date; nurses on units outside that scope, agency staff and those on leave exceeding three months are excluded because the exposure of interest is continuous unit-level practice. The sampling frame is the network’s own unit staffing roster, obtained under the data-sharing agreement described in section 3.7. All eligible nurses were invited, making this a census of the accessible population rather than a sample;[figure]responded, a response rate of[figure]per cent. Non-response is examined in section 4.2 by comparing respondents and the full roster on unit, shift pattern and years of service, the three variables available for both groups.
What the committee asks. “Is this a sample or a census?” — answered. “Can you say anything about the people who did not respond?” — answered, and the answer names exactly which variables make the comparison possible. Naming the three available variables is the move; it shows you understood the limits of your own frame.
2. Education — clustered by school
The target population is secondary teachers of examinable subjects in publicly funded schools in the province. The accessible population is teachers in the
[figure]secondary schools of one board whose research office granted approval; approval was sought from three boards and granted by one, and this selection is not random, a limitation revisited in section 3.9. Because teachers are reachable only through schools, a two-stage cluster design was used: all approved schools were retained at stage one, and within each school all teachers of the relevant departments were invited at stage two. Analyses account for the clustering of teachers within schools, since teachers in the same school share leadership, timetable and assessment culture and cannot be treated as independent observations.
What the committee asks. “Did you account for clustering?” This is the question that sinks education theses, and it belongs in the sampling paragraph rather than being left for the analysis chapter to discover. The sentence about three boards asked and one granting is also doing work: it reports the selection mechanism instead of concealing it.
3. Psychology — online panel, quantitative
The target population is community-dwelling adults aged 18 to 65 in Canada. The accessible population is adults registered with the research panel used for recruitment who meet the age and residency criteria and who completed the screening items. Quotas were set on age band, gender and region to approximate the distribution reported in the most recent census figures for these characteristics, and recruitment continued within each cell until the quota was met. This is a quota sample, not a probability sample: the panel is self-selected, panel members differ systematically from non-members in ways this design cannot measure, and quota matching on three variables does not make the sample representative on any fourth. Attention-check items were embedded, and the pre-registered exclusion rule is stated in section 3.6 rather than applied after seeing the results.
What the committee asks. “Is this representative?” The paragraph refuses the word and explains exactly why, which is a stronger position than claiming it. Naming the exclusion rule as pre-specified also pre-empts the question about researcher degrees of freedom.
4. Public health — administrative data
The study uses de-identified administrative records rather than recruited participants, so sampling operates as case selection. The target population is adults discharged from acute care with the index condition in the study province. The accessible population is defined by the extract available through the data custodian: all discharge records meeting the diagnostic and date criteria for fiscal years
[figure]to[figure]. No sampling was applied within that extract; the full cohort is analysed. Three exclusions are applied and reported with counts at each step in a flow diagram: records with invalid identifiers, records for non-residents of the province, and records with a length of stay above the pre-specified plausibility ceiling. Because inclusion depends on contact with acute care, individuals who were never hospitalised are absent by construction, which bounds the conclusions to the hospitalised population.
What the committee asks. “Who is missing from your data by construction?” Secondary-data theses live or die on that sentence. The flow diagram with counts at each exclusion is now expected, not optional. Where the data come from and what access route applies is set out in the Statistics Canada microdata guide.
5. Social work — qualitative, purposive with maximum variation
The target population is front-line practitioners delivering the program under study. Participants were selected purposively against three criteria: at least twelve months of delivery experience, current caseload in the program, and employment in an agency that had implemented the program before the study period. Within those criteria, maximum variation was sought on agency size, urban or rural setting, and years of practice, with the achieved range reported in Table 3.2 rather than merely intended. Recruitment proceeded through agency directors, which introduces a gatekeeping effect discussed in section 3.9: practitioners critical of the program may be less likely to be forwarded an invitation. Recruitment continued until no new codes were generated across two consecutive interviews, with the decision documented in the audit trail.
What the committee asks. “Achieved variation, not intended variation?” — answered with a table reference. “What did gatekeeping do to your sample?” — named rather than waited for. The saturation claim is tied to a documented rule instead of asserted.
6. Engineering — non-human units
The units of analysis are pipe segments rather than people. The target population is buried potable-water distribution segments in Canadian municipalities of comparable size and climate. The accessible population is the
[figure]segments in the participating municipality’s asset register with complete material, diameter and installation-year fields for the observation window. Segments lacking any of these fields were excluded, and the count and characteristics of excluded segments are reported, because incompleteness is not random: older records are more often incomplete, which biases the accessible population toward newer infrastructure. No further sampling was applied. Generalisation beyond municipalities with a comparable register and climate regime is not claimed.
What the committee asks. “Is your missingness informative?” The paragraph answers before being asked, which is the single most valuable habit in a sampling section. Note also that sampling language applies unchanged when units are not people.
The template
- The target population is [the group the research question is about].
- The accessible population is [the subset reachable], which excludes [who, and why].
- The sampling frame is [the concrete list or route], obtained [how].
- The technique is [named technique], applied as follows: [the rule].
- This design cannot reach [who is absent by construction], which bounds the conclusions to [whom].
Five sentences. If you cannot write sentence three, you do not yet have a sampling plan — you have an intention.
Five things committees catch
- Target and accessible collapsed into one. The fix is two sentences and it changes what you are allowed to claim.
- Convenience sampling by another name. “Participants were recruited via social media using a purposive approach” is convenience sampling. Say so.
- Clustering ignored. If people were recruited through schools, clinics or teams, they are not independent, and both the sampling paragraph and the methodology chapter have to say what you did about it.
- Exclusions without counts. Every exclusion criterion needs the number it removed. A flow diagram is now standard in health and increasingly expected elsewhere.
- Inclusion criteria that do not match the construct. If you excluded people for a reason unrelated to your operational definitions, expect to justify it.
Where sampling meets ethics
Recruitment route and consent process are the same decision viewed from two angles, and a Canadian research ethics board will read your sampling paragraph as a recruitment plan. Gatekeeper approval, incentives, the wording of the invitation and whether anyone in the recruitment chain has authority over participants all belong in the ethics application under TCPS 2, and the answers you give there must match the paragraph in your thesis exactly. Where the sampling plan constrains the question rather than serving it, that is a signal to revisit the problem statement — a gap you cannot sample for is not yet a workable gap.
Write the paragraph once, properly
Tesify drafts the sampling and population section from your own design decisions, including the sentences about who is absent by construction, so the paragraph is complete before a committee has to ask for it. Start your thesis with Tesify — 100% written by you.
Frequently asked questions
What is the difference between target and accessible population?
The target population is the group your conclusions are about; the accessible population is the subset you could actually approach. Naming both is what allows you to state, honestly, how far your findings travel.
What is a sampling frame?
The concrete list or route that enumerates your accessible population — a staffing roster, a class list, an asset register, a panel. “Online” is not a sampling frame.
Is convenience sampling acceptable in a master’s thesis?
Yes, when nothing better is available and you label it accurately, bound your claims to the accessible population, and discuss the resulting bias. What is not acceptable is presenting it as something else.
Do qualitative studies need a sampling section?
Yes. The logic differs — purposive rather than probabilistic — but the criteria, who applied them, the achieved variation and the stopping rule all have to be stated.
What is purposive sampling?
Selecting cases because they have characteristics your question requires. The criteria must be written out in full, and the achieved sample reported against them rather than only the intended sample.
Do I have to report non-response?
Report the response rate, and compare respondents to the frame on whatever variables exist for both. If no such variables exist, say that, because it is itself a limitation.
What is snowball sampling and when is it justified?
Recruiting through participants’ own networks, used when a population is hidden or hard to reach. Report the number of seeds and the chain lengths, because long chains from few seeds produce a homogeneous sample.
My sample is one organisation. Is that a problem?
Not if you frame it as one. A single-site study answers a single-site question well; it becomes a problem only when the conclusions are written as though they covered the sector.
Does sampling apply when my units are not people?
Yes, unchanged. Segments, documents, sites, records and firms all have target populations, accessible populations, frames and selection rules.
Where does the sampling section go?
In the methodology chapter, after the design and before the instruments, because what you measure depends on whom you can reach.