Interpreting Margin of Error

A single survey only measures a sample, never the whole population, so the number it reports almost certainly isn't the exact truth. The margin of error admits this honestly. Instead of pretending the sample percentage or sample mean is exact, it builds a window around that estimate: estimate minus the margin, up to estimate plus the margin. Everything inside that window counts as a plausible value for the population.

Take a poll showing 52 percent support with a margin of error of 3 points, at 95 percent confidence. That claims the true population support plausibly sits between 49 percent and 55 percent, and that the sampling method used to build intervals like this one succeeds in capturing the real value about 95 percent of the time it's used. It is not a guarantee for this particular poll, and it is not the worst-case error pollsters could make. Bigger mistakes are always possible, from bad sampling to dishonest answers. The margin of error only accounts for the ordinary wobble you get from sampling a limited group at random.

Push the confidence level up from 95 percent to 99 percent and the interval has to widen, because demanding a higher success rate across repeated samples means leaving less room for error. This is the same tradeoff that runs through the rest of data analysis in this unit: tighter claims about a population always cost something, whether that's a wider interval, a bigger sample, or less confidence. Reading margin of error correctly means reading it as a statement about a method's reliability, not a promise about one result.