Whole Population or Sample?
Examining everything feels safer, and one slide in the quantitative research week argues that it often is not. This page takes that argument apart into the three claims it actually makes, shows which one is worth keeping, and marks the point where the slide asserts more than it has argued.
Two answers to the same question, six slides apart
The week answers "why not examine everything?" twice, and the two answers have different tempers. The first arrives as sampling is introduced and treats it as something you settle for. The second arrives at the end of the sampling sequence and treats it as something you gain by.
Sampling as a concession
- "Populations are usually too big or items too inaccessible for the whole population to be examined."
- "One may have to be satisfied with examining only a part, or sample, of the total population."
- A third constraint follows separately: you have many objects you could collect data on "but you only have the resources to collect data on a few of them".
- Nothing on that slide states what is lost by sampling. No sampling error, margin of error, precision or representativeness threshold appears there or on any of the six slides after it.
Sampling as an improvement
- "Cost frequently rules out examining the whole population."
- A whole-population exercise accumulates error of its own, exceeding "the error inherent in using sample results".
- "Taking only a sample results in improved accuracy…"
- Nothing on that slide acknowledges the earlier framing, and nothing on the earlier slide anticipates this one.
Both framings are usable and they are not the same claim. If you are asked to justify sampling in a research proposal, the first supports a feasibility argument and the second supports a quality argument — and the second is considerably more interesting, provided you read what it actually says. The definitions the two slides rest on are covered at population, sample and variables.
The cost and error slide, in full
The slide is titled "Sampling selection limits" and its content sits under a single sub-heading, "Cost/size". It is quoted here complete, including the sentence that does not resolve grammatically.
The second sentence is broken at its load-bearing clause. "Coupled with the monotony of examining large numbers of items, leads to that the overall cumulative error that is greater than" has no subject for "leads to", and "leads to that … that is greater" does not resolve. This library quotes it as it stands and does not repair it, because any repair would decide what the sentence claims — and what it claims is the interesting part.
One other term enters here without introduction. "Survey" appears for the first time in the week in the phrase "the value of the survey results", and the week defines neither survey nor questionnaire; both belong to the later data collection material. The slide assumes an instrument it never names.
The argument in three claims
Read as a sequence, the slide makes three distinct claims, and they are of three different kinds: an economic one, a causal one and a general one. Separating them is what makes the slide usable.
THE COST AND ERROR ARGUMENT, SEPARATED INTO ITS CLAIMS
| Claim | What it says | Status |
|---|---|---|
| 1. Cost | Examining a whole population costs more than examining a sample, the cost "could easily exceed the value of the survey results", and cost "frequently rules out" a census. | Economic and conditional. Stated without a single figure — no cost, no ratio, no point at which the value is exceeded |
| 2. Cumulative error | A whole-population exercise is large enough to require unskilled investigators doing monotonous work, and the errors they accumulate exceed the error inherent in sampling. | Causal and conditional, and it is the claim worth keeping. It depends on scale, on who does the work and on the monotony of it |
| 3. Accuracy | "Taking only a sample results in improved accuracy because careful attention is given to measurements that are made with a high degree of accuracy." | General and unconditional as written — and the condition that made claim 2 work has quietly gone |
The slide states none of these as numbered claims; the separation is this library's, made to show where the argument holds and where it stops holding.
Why the middle claim is worth having
Claim 2 is a real argument and a good one. It says that the error in a result is not only the error that comes from having examined a part rather than the whole: there is also error introduced in the act of measuring, and that second kind grows with the size and tedium of the job. Below some scale, the measurement component dominates, and a smaller job done carefully beats a larger job done badly.
It is also the only place in the supplied material where any error is broken into parts. Everywhere else, error is a single undifferentiated thing that data collection introduces and analysis inherits — including the one bias the week does name, at non-response and sampling bias. That makes this slide more valuable than its four lines suggest — and it makes the next section more important, because the source does not carry the decomposition any further.
Where the argument overreaches
Two problems, both of which go on the page rather than being smoothed away.
Two ways the slide states more than it has established
The third claim drops the second claim's condition
Claim 2 makes the accuracy advantage depend on conditions: unskilled investigators, monotony, large numbers of items. Claim 3 states the advantage with no condition at all — taking only a sample "results in" improved accuracy. As written, the slide asserts that sampling always improves accuracy, which is not what its own preceding sentence argued.
The deck and the week's own notes point in opposite directions
The slide says accuracy improves as the sample gets smaller. The week's study notes — at that point a reproduced article from another discipline — say the number of subjects you need depends on your design and that more of them buys tighter estimates. One document has precision rising with size, the other has accuracy rising as size falls, and neither refers to the other.
What the slide's title promises and does not deliver
The slide is titled "Sampling selection limits", which promises a set, and its sub-heading is "Cost/size", which promises two things. It delivers one limit and one half of the sub-heading.
- Cost is the only limit developed. Time, access, ethics, expertise, the availability of a sample and the willingness of the population to take part are named nowhere as constraints on sampling.
- "Size" appears in the sub-heading and nowhere in the body. No size, minimum, proportion or rule of thumb is stated — the size question is examined at how large should a sample be, where the finding is that the material never answers it.
- No number appears on the slide at all. "Higher costs", "a small sample", "frequently", "large numbers of items", "greater than" and "a high degree of accuracy" are the whole of its quantitative language. The week's only trade-off argument is stated entirely without quantities.
- No break-even point is offered. The argument turns on a scale at which measurement error overtakes sampling error, and the source gives no way of locating that scale for any study.
Deciding between a census and a sample in practice
The decision rules below are this library's synthesis, not the source's instruction — the source supplies no criterion for the choice. Each one is anchored to a claim the slide does make, so that you can trace what you are relying on.
Which way to go, and what to say about it
One boundary statement applies across every sampling page here. The week supplies definitions of population and sample, six named sampling types, one bias with a sound causal chain, and this cost and error argument. It supplies no procedure for executing any sampling type, no mention of a sampling frame, no sample size rule in its own voice, and no criterion for choosing between the types. Sampling is named in the later data collection week's only stated learning objective and was taught there in one sentence that recommended random selection and immediately withdrew it as usually impossible. Across the six weeks of teaching material now on record, fourteen objectives are stated: none fully delivered, four partially delivered, ten not delivered at all — the reconciled audit is at the six-week objective audit.
What to carry forward
- The source frames sampling twice and incompatibly: as a concession you settle for, and as a choice that improves accuracy.
- The cost argument is unconditional in its wording and unquantified in its content — no cost, no size, no error rate, no break-even point.
- The cumulative-error claim is genuinely valuable: below some scale, error introduced by measuring beats error caused by examining only a part.
- The slide then states that advantage without the condition that produced it, and the week's own notes point the other way on size and precision.
- The source never distinguishes sampling error from measurement error, so quote both positions, name the missing distinction, and take the distinction itself from a text you cite.
Frequently asked questions
Does the supplied material say a census is worse than a sample?
It says a census usually costs more than it is worth and that, at a scale requiring unskilled investigators doing monotonous work, the errors accumulated exceed the error inherent in sampling. It then states more broadly that taking only a sample results in improved accuracy, dropping the conditions. The conditional version is the defensible one.
How can examining everything be less accurate than examining a part?
Because error has more than one source. The source's argument is that a very large measurement job brings in less skilled effort and monotony, and the errors introduced in the measuring can exceed the error caused by having examined only a part. The material makes that argument once and never names the two kinds of error.
At what point does a census stop being worth doing?
The material gives no break-even point, no cost figure, no size and no error rate. Its entire trade-off argument is stated qualitatively. If you need a threshold, you will have to derive one from your own cost and effort estimates and say that you did.
The slide says smaller samples are more accurate but the study notes say bigger samples are better. Which is right?
This library does not adjudicate. The two statements are about different error sources and would be reconcilable if the material distinguished them, but it never does — and the notes are a reproduced article from another discipline rather than the subject's own voice. Report both, and name the distinction the source is missing.
How do I justify sampling in a research proposal using this material?
You have two arguments and they are stronger together. The feasibility argument is that the population is too large or too inaccessible and your resources are limited; the quality argument is that a smaller job can be measured more carefully. State the conditions that make the second one true in your case rather than asserting it generally.
References and source attribution
- The supplied teaching source: the quantitative research week's slide titled "Sampling selection limits", the slide introducing sampling, and the week's key terms slides; together with the same week's study notes.
- Hopkins 2000, 'Quantitative Research Design', a sport and exercise science web journal, vol. 4, no. 1 — the article set as the week's required reading and reproduced verbatim in the study notes, cited here to identify the register of the notes' position on sample size and precision.
- Bryman, A. 2016, Social Research Methods, 5th ed., Oxford University Press, Oxford. Standing reference for the sampling error and measurement error distinction the supplied material does not make.
- Veal, A. J. 2005, Business Research Methods: A Managerial Approach, Longman.
- O'Leary, Z. 2017, The Essential Guide to Doing Your Research Project, 3rd ed., Sage Publications, London.
Suggested questions for Ask KEVOS
- Set out the source's cost and error argument for sampling, claim by claim.
- Where does the supplied material contradict itself about sample size and accuracy?
- Help me justify sampling rather than a census in a research proposal.
- What limits on sampling does the material name, and which ones does it leave out?
- What is the difference between sampling error and measurement error, and does this source cover it?