Surveys and interviews

A survey tries to give an overview of a large population by collecting information from a carefully chosen sample. The people who answer are respondents; the questions are asked through a questionnaire or an interviewer.

• A census is a survey that covers every single member of a population. India's population census, held every ten years, is the largest such exercise in the world, involving lakhs of investigators. Most surveys instead use a sample, saving time and money.

Choosing a representative sample rests on two principles:

• Stratification: large populations are made up of distinct sub-groups (rural and urban, regions, sizes of village, levels of income). All the relevant strata must be represented in the sample. (This statistical meaning of 'stratification' is different from social stratification.)
• Randomisation: within each stratum, units must be chosen purely by chance, so that every unit has an equal chance of being picked. A sample chosen for convenience, such as only households a researcher already knows, is purposive, not random, and may not represent the population.
• Bigger is not automatically better: a large sample that is badly chosen can still mislead.

Errors in surveys:

• Sampling error: because a small sample stands for a large population, results may differ slightly from the true value. This is unavoidable, but its likely size can be calculated, and researchers must report their sample size, design and margin of error.
• Non-sampling errors come from mistakes in design or delivery: badly worded questions, poorly trained interviewers, respondents who misunderstand or do not answer truthfully.

Strengths and limits of surveys: they can cover large populations and reveal aggregate patterns that no single village study could show, such as the fall in India's child sex ratio. But they lack depth: questions must be simple and standard, there is little time to build trust, and complex or sensitive matters are hard to capture.

The interview is a guided conversation between researcher and respondent. It sits between the structured questionnaire and open-ended participant observation.

• Interviews can be structured (fixed questions in a fixed order), unstructured (free flowing), or in between.
• The chief advantage is flexibility: questions can be reworded, reordered or followed up when something interesting comes up.
• But the same freedom makes interviews hard to compare or generalise from, and success depends on the researcher's skill and on rapport (mutual trust) with the respondent.
• Example (new): after a survey shows that many students in a district drop out after Class 8, a researcher interviews ten families in depth to learn why: the survey tells 'how many', the interviews tell 'why'.