Collection of Data

Any statistical study begins by gathering facts and figures — this is the collection of data, the first stage of a statistical investigation. Data means a set of numerical facts or information collected for a definite purpose, such as the marks of students or the daily wages of workers.

Before collecting data, the investigator must be clear about the purpose, scope, and units of measurement of the enquiry. Data are of two main types depending on who collected them:

• Primary data — data collected for the first time, directly by the investigator, for a specific purpose. They are original and have not been used before (e.g. a student surveying classmates about their pocket money). Primary data are more reliable and suited to the purpose, but they take more time and money.
• Secondary data — data that have already been collected by someone else and are used again. They are second-hand (e.g. figures taken from a newspaper, a government report or a website). Secondary data save time and cost, but they may not exactly fit the new purpose and their reliability must be checked.

The chief sources of secondary data are published sources (government reports, the Census of India, RBI bulletins, newspapers, journals, international bodies like the UN) and unpublished sources (private records, registers, research files). The golden rule: before using secondary data, check who collected it, how, when and for what purpose.

When collecting primary data about a group, the investigator must decide how much of the group to study. The whole group being studied is called the population (or universe), and each member of it is a unit. There are two methods:

• Census method (complete enumeration) — every single unit of the population is studied. For example, the Census of India counts every person in the country. This method is very accurate and gives complete information, but it is costly, slow and needs a lot of effort — practical only when the population is small or when total accuracy is essential.
• Sample method — only a part (a sample) of the population is selected and studied, and the results are taken to represent the whole. For example, a TV-rating agency surveys a few thousand households, not every household. This method is cheaper, faster and needs less effort; it is the most common method in practice.

A good sample must be representative (it should mirror the whole population) and large enough. The main sampling techniques are: random sampling (every unit has an equal chance of being chosen, like a lottery — this avoids bias), stratified sampling (the population is divided into groups/strata and samples taken from each), and systematic sampling (every nth unit is chosen). The danger of sampling is sampling error — the sample may differ from the true population — which is reduced by choosing a proper, large, random sample.

When primary data must be collected, the investigator can choose from several methods, depending on the purpose, area, time and money available:

• Direct Personal Investigation — the investigator personally contacts and questions each informant face-to-face. It is accurate and reliable but suitable only for a small area; it is costly and time-consuming for large enquiries.
• Indirect Oral Investigation — instead of the informants themselves, the investigator questions other people (witnesses) who know about them. Used when informants are unwilling or hard to reach (e.g. enquiry into drinking habits). A police investigation works this way.
• Information through Correspondents — local agents or correspondents in different places collect and send information regularly. Newspapers gather news this way. It is cheap and wide-reaching but less accurate, as it depends on the correspondents.
• Mailed Questionnaire — a list of questions (a questionnaire) is mailed/emailed to informants, who fill it in and return it. It covers a wide area cheaply, but only literate people can respond and many may not return it (low response rate).
• Schedule Method — trained workers called enumerators carry the questionnaire (here called a schedule) to the informants and fill it in for them by asking the questions. This is how the Census of India is done; it suits large enquiries and even illiterate informants, but it is costly.

The key difference between a questionnaire and a schedule: a questionnaire is filled in by the informant himself, while a schedule is filled in by an enumerator on the informant's behalf. Whichever method is used, the questions must be clear, short, unambiguous and not personal or leading.