Sampling Errors and Non-Sampling Errors

No survey — census or sample — produces results that are perfectly exact; the gap between an estimate and the true value is called error. Statistics distinguishes two broad types.

Sampling Error arises solely because conclusions about the whole population are drawn from a part (a sample) of it, rather than from every unit. It is the difference between a sample statistic and the true population parameter that would have been obtained from a full census, purely because different samples give somewhat different results.

Illustration: Suppose the true (but otherwise unknown) average monthly sales of all 400 retail outlets in a district is $\mu = ₹50{,}000$. A sample of 40 outlets gives a sample mean of $\bar{x} = ₹47{,}000$. The sampling error here is $|47{,}000 - 50{,}000| = ₹3{,}000$. Sampling error exists only in the sample method — a census, by covering every unit, has no sampling error — and it tends to decrease as the sample size increases, because a larger sample represents the population more closely.

Non-Sampling Error arises from mistakes and shortcomings in planning, collecting, recording, or processing data, and can occur in both a census and a sample survey — a census, in fact, is often more exposed to it, simply because there is so much more data to handle. Common sources include:

• Poorly worded or ambiguous questionnaire items.
• Non-response or refusal by respondents, or deliberately incorrect answers.
• Bias or carelessness on the part of interviewers/investigators.
• Errors in recording, coding, tabulating, or processing the collected data.
• Incomplete coverage of the intended population (some units never get a chance to be included at all).

Controlling the two types of error: