Data Collection and Sources
What you must understand
Every statistical investigation begins with a question and a plan. Candidates should distinguish primary data (collected first-hand by the investigator, e.g. a survey you run) from secondary data (already published, e.g. census tables, school records, newspaper reports). Know the main collection methods: questionnaires (cheap, wide reach, but risk of vague wording and non-response), interviews (allow clarification, but costly and open to interviewer bias), observation (records behaviour directly, but people may change behaviour when watched), experiments (allow control of variables), and extraction from existing records.
You should understand populations and samples, censuses versus surveys, and the difference between quantitative (numerical) and qualitative (categorical) variables. Quantitative variables may be discrete (countable, e.g. number of siblings) or continuous (measured, e.g. height). Data may be raw, grouped, or ungrouped.
Traps that appear in questions
- Calling a published table "primary" because you found it yourself; source of origin, not who uses it, decides.
- Confusing discrete with continuous: money, height and time are continuous; counts are discrete.
- Ignoring bias: leading questions, samples drawn only from volunteers, or a survey of a school's sports teams used to describe the whole school.
- Forgetting that a census is expensive but gives no sampling error, while a sample is cheaper but carries sampling error.
Worked example
A student wants the average weekly pocket money of her 900 schoolmates. She asks the 30 members of her netball squad. Identify the population, the sample and two sources of bias.
Answer: Population = all 900 students; sample = the 30 netball players. Bias: the sample is not random (squad members may be wealthier or older), and the small size of 30 gives a large sampling error. A better plan is a random sample drawn from a complete list of students.
Rule of thumb: always ask who collected it, from whom, how, and why before trusting any data set.