Non-Sampling Error: The Mistake That Has Nothing to Do With the Sample
Imagine you're the class monitor asked to find the average height of students in your school. You decide to measure every single student — a complete census, not a sample. You get your measuring tape, write down heights, and calculate the average. But then you realise: the measuring tape was stretched, you misread some numbers, and a few students lied about their height because they were embarrassed. Your final average is wrong — even though you measured everyone.
That wrongness is a non-sampling error.
The Everyday Intuition
When you hear "error" in statistics, you probably think of sampling error — the natural difference between a sample and the population it came from. That's like tasting one spoonful of soup and guessing the whole pot's flavour; you might be off because you only tasted a small part.
Non-sampling error is different. It's the mistake that happens even if you taste the entire pot. It's the burnt tongue, the misread recipe, the faulty thermometer. It has nothing to do with how much you tasted — it's about how badly you measured, recorded, or processed what you had.
The Precise Meaning
A non-sampling error is any error in a survey, census, or data collection that is not caused by the fact that you took a sample. It arises from:
- Measurement errors: The instrument is faulty (a weighing scale that reads 2 kg too high), the question is confusing ("How much did you spend on food last year?" — who remembers that?), or the respondent lies or forgets.
- Coverage errors: Your list of people to survey (the sampling frame) is incomplete or has duplicates. For example, a telephone survey misses households without phones.
- Non-response errors: People you want to survey refuse to answer, or are not home. If the rich people in a neighbourhood systematically refuse to report their income, your average income will be too low — even if you contacted every house.
- Processing errors: You type "105" instead of "150", you misplace a decimal point, or your computer code has a bug.
- Interviewer bias: The surveyor's tone of voice, appearance, or leading questions influence the answer. ("You don't smoke, do you?")
Non-sampling errors are more dangerous than sampling errors. Sampling error shrinks when you increase sample size — take a bigger spoonful, get a better guess. Non-sampling error does not shrink with more data. In fact, a bigger survey can magnify a non-sampling error. A census (measuring everyone) eliminates sampling error entirely, but non-sampling errors can still ruin the result.
Why It Matters for Economics
In economics, you rarely have perfect data. The government's National Sample Survey Office (NSSO) conducts massive surveys on consumption, employment, and poverty. These surveys use careful sampling to keep sampling error small. But non-sampling errors are the real enemy.
Consider the Consumer Price Index (CPI) — a measure of inflation. To calculate it, field investigators visit thousands of shops and record prices of hundreds of items. If an investigator records the price of rice from a premium store instead of an ordinary one, that's a measurement error. If a shopkeeper gives yesterday's price because he's busy, that's a response error. If the list of items being tracked is outdated (say, it still includes cassette tapes), that's a coverage error. All of these are non-sampling errors, and they make the CPI less accurate — even though the sample size might be perfectly adequate. …