Sample Data, Population Parameters, and Margin of Error

Pulling 1,000 people from a city of a million works because of how randomness behaves at scale. If the sample is chosen without bias, the people in it tend to split on an issue roughly the way the whole city would. The sample percentage will not match the city's true percentage exactly, but it lands close, and how close is predictable from the size of the sample alone — not the size of the city.

Margin of error turns that one sample number into a range. If a sample says 52 percent support something, with a margin of error of 3 points, the true population value is plausibly anywhere from 52 - 3 = 49 percent to 52 + 3 = 55 percent. The sample figure sits at the center, and the margin is added and subtracted to build the interval. A bigger sample shrinks that margin, because larger samples track the population more tightly — but the relationship isn't one-for-one. Margin of error shrinks with the square root of the sample size, so to cut it in half, the sample has to quadruple: going from 400 to 1,600 people (four times as many) cuts the margin of error in half, not to a quarter.

This connects directly to the probability and simulation ideas nearby: margin of error exists because sampling is a random process, and the interval it produces is a statement about plausibility, not certainty. That's why an occasional poll misses its own interval — it's built into how the method works, not a sign that the method failed.