Sampling Techniques for Student Research

TL;DR: Sampling has two families. Probability sampling (simple random, systematic, stratified, cluster) lets you generalise statistically; non-probability sampling (purposive, convenience, quota, snowball) does not, and pretending otherwise is what panels catch. For sample size, Slovin's formula is the Philippine default but is much narrower than modules suggest: published work in The Philippine Statistician found it applies only when estimating a proportion at 95 percent confidence, and works best when that proportion is near 0.5. Cochran's formula with a finite population correction is the more defensible route, and for a population of 380 the two land within a few respondents of each other anyway.

Sampling is where a lot of otherwise decent student research quietly breaks. The technique gets picked because it is the one the last batch used, the sample size gets computed with a formula nobody has examined, and the conclusions then claim more than the method can support. None of that is hard to fix, but it has to be fixed before data collection, because afterwards there is nothing to do about it.

DepEd's Practical Research 2 curriculum guide makes this an assessed competency in its own right: "describes sampling procedure and the sample" (CS_RS12-IIa-c-2), sitting between constructing the instrument and planning data collection. This guide covers the eight techniques you will realistically choose from, a worked stratified allocation, and an honest account of the sample size formula everyone uses.

Which subject is this for? Practical Research 1 and 2 are applied subjects of the 2016 K to 12 Senior High School curriculum, and that structure is being replaced. DepEd Order No. 017, s. 2026 institutionalises the Strengthened Senior High School Curriculum for Grade 11 from SY 2026-2027 and Grade 12 from SY 2027-2028, and its Annex A lists Research 1 and Research 2 as 80-hour Academic Track electives in the Field Experience cluster rather than applied subjects everyone takes. Grade 12 learners not enrolled in pilot schools continue under the 2016 curriculum for SY 2026-2027. The sampling logic below is unaffected. See our guide to the Strengthened SHS Curriculum.

Probability vs Non-Probability, in One Table

The "can you generalise" column is the one that matters. A study using convenience sampling can still be good research; it simply cannot conclude that "Grade 11 students in Cebu City" do anything. It can conclude that its respondents did.

The Four Probability Techniques

Simple random sampling requires a complete list of the population, a sampling frame, and a genuine randomisation method. Drawing names from a bowl counts. Picking "randomly" as you walk around does not. Use it when the population is homogeneous and you have the list.

Systematic sampling divides the population size by the sample size to get an interval k, picks a random start between 1 and k, then takes every kth name. For a population of 500 and a sample of 100, k is 5: pick a random start between 1 and 5, then take every fifth name. It is faster than simple random sampling and just as defensible, unless the list has a hidden cycle, for example alternating male and female entries, that lines up with your interval. It also works poorly when the sampling fraction is very high, since k collapses towards 1 and you are effectively surveying everyone anyway.

Stratified random sampling splits the population into non-overlapping strata that matter to your question, then samples randomly within each. This is the workhorse for school-based research, because strand, grade level, section and campus are natural strata and you almost always want each represented.

Cluster sampling randomly selects intact groups, whole sections or whole schools, then studies units within them. Use it when a complete list of individuals does not exist but a list of groups does. The catch is that people within a cluster resemble each other, so a cluster sample of 200 carries less information than a simple random sample of 200.

The Four Non-Probability Techniques

Purposive sampling selects people deliberately because they meet criteria that make them informative. This is the correct and expected choice for most qualitative work, where the goal is depth rather than representativeness. Our guide to qualitative research designs covers how the design constrains who qualifies.

Convenience sampling takes whoever is reachable. It is extremely common in student research and perfectly publishable when labelled. Name it, and carry the limitation into Chapter 5.

Quota sampling fills predetermined counts per category, say 50 male and 50 female, without randomising within the category. It looks like stratified sampling and is not, because the selection inside each quota is not random.

Snowball sampling relies on referrals and exists for hard-to-reach populations where no list is possible. It is a legitimate technique with an obvious bias, since people refer people like themselves.

How Many Respondents? Start With Slovin, Then Check It

Slovin's formula is the Philippine default:

n = N / (1 + Ne²)

where N is the population size and e is the margin of error as a decimal. For a target population of 380 at a 5 percent margin of error:

n = 380 / (1 + 380 × 0.05²)
  = 380 / (1 + 380 × 0.0025)
  = 380 / 1.95
  = 194.87, rounded up to 195

Round up, always. A fraction of a respondent is not a thing, and rounding down shrinks your precision below what you claimed.

Now the part most modules skip. Tejada and Punzalan of the University of the Philippines Diliman published a careful examination of the formula in The Philippine Statistician in 2012. Their conclusion is direct: the formula "is applicable only when estimating a population proportion and when the confidence coefficient is 95%," and "it is optimal only when the population proportion is suspected to be close to 0.5." They add a detail worth knowing: from their literature review, "there does not seem to be a person named Slovin who put forward the formula," with Yamane (1967) the oldest reference in which it appears.

What that means in practice:

• If your study estimates a proportion (what percentage of students do X) at 95 percent confidence, Slovin's formula is defensible.
• If your study estimates a mean (average screen time in hours, average score), the formula was never derived for that case and you should use a mean-estimation formula instead.
• If you want a confidence level other than 95 percent, the formula cannot express it. There is no place in n = N/(1 + Ne²) to put a z value.

The more defensible route is Cochran's formula, which Slovin's is a simplification of. Cochran's initial estimate is n₀ = z²pq / e². At 95 percent confidence (z = 1.96) with the most conservative proportion (p = q = 0.5) and a 5 percent margin of error:

n₀ = (1.96² × 0.5 × 0.5) / 0.05² = 0.9604 / 0.0025 = 384.16, so 385

Then apply the finite population correction for a population of 380:

n = n₀ / (1 + (n₀ - 1)/N) = 384.16 / (1 + 383.16/380) = 384.16 / 2.00832 = 191.3, so 192

Cochran gives 192 and Slovin gives 195. They nearly agree, and that is exactly the point: Slovin's formula is what Cochran's collapses to when you fix the confidence level at 95 percent and set p at 0.5. Inside those assumptions it is fine. Outside them it is a shortcut that no longer computes what you think it computes.

If your adviser requires Slovin's formula, use it and state the assumptions. Adding one sentence, "the formula assumes proportion estimation at 95 percent confidence with p = 0.5," is the difference between following a recipe and understanding a method.

Worked Example: Proportional Stratified Allocation

This example is illustrative, constructed for this guide. The school, the enrolment figures and the strand distribution are invented to show the arithmetic.

Illustrative situation: a study of Grade 11 students in one senior high school, target population 380, sample size 195 from the computation above, stratified by strand because the research question compares strands.

Proportional allocation gives each stratum a share of the sample equal to its share of the population: n per stratum = 195 × (stratum size / 380).

Two things to check every time. First, the stratum samples must add back to your total; rounding sometimes puts you one over or under, and you adjust the largest stratum to reconcile. Second, within each stratum you still draw at random. Allocating 62 slots to STEM and then handing questionnaires to the first 62 STEM students you meet turns a stratified design into a convenience sample with extra steps.

Matching Technique to Design

Our guides to quantitative research designs and qualitative research designs cover the design half of that table in full.

Common Mistakes That Cost Marks

• Calling convenience sampling random. The most common single error in student methodology sections.
• Using Slovin's formula on a qualitative study. Saturation is the standard there; a formula signals the design was not understood.
• Using Slovin's formula to estimate a mean. Outside its derivation, per Tejada and Punzalan (2012).
• Reporting a sample size with no margin of error. The number 195 is meaningless without the 5 percent that produced it.
• Stratifying, then not randomising within strata. Half a stratified design is not a stratified design.
• Sampling from a frame that does not match the population you claim. If your list is one section's group chat, your population is that group chat.
• Ignoring non-response. The sample you computed and the respondents you got are different numbers, and both belong in your write-up. See our guide to describing the respondents of the study.

Before You Collect Anything

1. Write down the population and where the list comes from.
2. Choose the technique from the design, not from the previous batch's thesis.
3. Compute the size, state the margin of error and confidence level, and note the formula's assumptions.
4. Decide the randomisation mechanism concretely: which app, which lottery, which interval and random start.
5. Confirm the number of instruments you need to print matches your allocation table, then check it against your research questionnaire plan.
6. Plan for non-response before it happens, and record distributed, returned and usable counts as you go.

Ready to find the right school? Browse schools or compare programs and reviews across hundreds of Philippine schools on SchoolFinderPH.

Disclaimer: departments differ on which sample size formula they accept and on how sampling must be documented. Confirm your own research manual's requirements before computing anything, and treat this guide as background rather than as your department's rule.

Sources

• Tejada and Punzalan (2012), "On the Misuse of Slovin's Formula," The Philippine Statistician, Vol. 61, No. 1, pp. 129-136 — Philippine Statistical Association; the applicability conditions, the Cochran derivation, and the note on the formula's origin
• Philippine Statistical Association — The Philippine Statistician, issue listing
• DepEd — K to 12 Senior High School Applied Track, Practical Research 2 Curriculum Guide — competency CS_RS12-IIa-c-2
• DepEd — K to 12 Senior High School Applied Track, Practical Research 1 Curriculum Guide
• Cochran, W. G. (1977), Sampling Techniques, 3rd edition, John Wiley and Sons — the finite population correction used above
• SchoolFinderPH — Respondents of the Study
• SchoolFinderPH — Quantitative Research Designs