Population and Sampling

In a survey or analysis, the whole set of things whose characteristics you want to learn is called the population (母集団).

What counts as the population depends on the purpose of the survey; to learn how long high school students in a certain city spend commuting, for instance, the population is every high school student in that city. The part chosen from the population to actually examine is called a sample. Results from the sample are the material for estimating the characteristics of the population.

AnalogyTo gauge how worn a library's whole collection is, every book in it is the population, and the few dozen you pick up and check are the sample.

How to rememberThe Japanese term 母集団 starts with 母, "mother": every sample is born from the population, so the character itself traces the relationship.

Choosing the part of a population to be surveyed (the sample) is called sampling (標本抽出).

A survey that examines only the chosen sample is a sample survey; it saves the time and cost of examining everything while still letting you estimate the population from the results. Results shift somewhat depending on which elements happen to be chosen, and this variation is called sampling error. Samples picked by the surveyor's own judgment tend to be biased, so random sampling, which leaves the choice to chance, is the standard approach.

AnalogyIt is like checking a big batch of cookies by taking a few from the pile as they come out of the oven. Pick only the nicest-looking ones and you will misjudge the whole batch.

How to rememberThe Japanese word for a sample, 標本, is the same word used for an insect specimen: an example kept at hand to represent the whole.

The national census (国勢調査) is a government statistical survey that covers every person and household living in the country, and it is the classic example of a complete survey, which examines the entire target group.

In Japan it is held every 5 years in principle, producing figures such as population and household composition that underpin government planning and other statistics. Because the target is so large, it takes a great deal of staff and time from preparation to the publication of results.

AnalogyIt is like a school health checkup that measures the height of every student and records them all in a roster.

How to rememberIn Japanese, 国勢 means the state of the nation, its people and households. To capture that whole picture, everyone has to be counted.

Gathering opinions or facts by having people answer a form that lists questions is called a questionnaire survey (アンケート調査).

When there are many people to ask, it is usually run as a sample survey, sending the form to a part chosen from the population. How respondents are chosen, and who actually replies, can pull the results away from the true population. For example, if people who care strongly about the topic are more likely to reply, the collected opinions lean toward their side.

AnalogyIt is like finding out what students think of a new cafeteria dish by asking a few people who happen to be eating there at lunch.

How to rememberThe Japanese word アンケート comes from the French enquête, meaning "inquiry."

A survey that examines every element of the population without exception is called a complete survey (全数調査, also 悉皆調査).

The more targets there are, the greater the cost, time, and staff required, and the longer the results take. In some cases, such as tests that destroy the product being checked, examining everything would make the survey itself impossible. A complete survey is therefore chosen when the number of targets is limited or when an exact figure for the whole is essential.

AnalogyIt is like a stocktake where a shop counts every item one by one. Nothing is missed, but the more products there are, the longer the shop has to stay closed.

How to rememberIts other Japanese name, 悉皆調査, uses 悉, read "kotogotoku" and meaning "every last one."

Simple random sampling (単純無作為抽出) chooses a sample by chance, for example with random numbers, so that every element of the population has the same probability of being selected.

You prepare a numbered list of every element in the population and draw numbers at random to decide the sample. Because the surveyor's preferences play no part, it prevents bias from the way elements are chosen. On the other hand, it needs a list of the entire population, and when the targets are spread over a wide area, visiting them takes a lot of effort.

AnalogyIt is like putting a slip with everyone's name into a box, shaking it well, and drawing with your eyes closed.

How to rememberThe Japanese 無作為 reads as "without contrivance": the characters say outright that the chooser's intentions are kept out.

Stratified sampling (層別抽出) divides the population into groups of similar elements (strata) and chooses a sample at random from each stratum.

Strata are formed by traits likely to affect the results, such as age or region. If each stratum's sample size is set in proportion to its size, the sample mirrors the population's makeup. Simple random sampling can by chance pick almost no one from some stratum, but dividing into strata first keeps that kind of bias in check.

AnalogyIt is like checking the quality of a fruit gift set by taking a few pieces from each of the apple, orange, and grape boxes.

How to rememberGeologists call each layer of rock a "stratum." Picture taking soil from every layer in a cliff face to test it.

Multistage sampling (多段抽出) repeats random selection in stages, from large units down to smaller ones, and finally picks individuals or households as the sample.

Stage 1 picks large units such as municipalities, and each later stage chooses the next units only from within those picked in the stage before. There is no need to build a list of the whole population, and the areas to visit are clustered, which saves effort in nationwide surveys. The trade-off is that each added stage brings its own selection variation, so sampling error tends to grow.

AnalogyIt is like checking stock in a large warehouse: first pick a few shelves, then pick boxes from those shelves, and finally inspect the items inside the boxes.