Sample Statistics and Margin of Error

Why We Sample at All

Studying an entire population — every voter, every product, every student in a district — is usually impossible, so researchers examine a sample and use it to estimate the whole. The SAT tests whether you understand both the logic of that leap and its limits. Almost every question in this topic is a reasoning question, not a calculation: you interpret a reported statistic rather than derive one.

Random Sampling and Generalization

The single most important requirement is that the sample be random: every member of the population must have an equal chance of being selected. Only a random sample lets you generalize honestly to the population it came from.

Just as important, the reach of a conclusion is limited by who was actually eligible to be sampled. If a study randomly surveys shoppers at one grocery store, its results generalize to that store's shoppers — not to an entire city, and not to people who shop elsewhere. A valid SAT conclusion must name the same population the sample was drawn from at random. Widening the claim beyond that population is a trap answer.

Bias Can't Be Fixed by Math

When a sample is not random, it may be biased, and no calculation repairs a biased sample. Classic biased designs include self-selected samples (only people who volunteer to respond, e.g. an online poll) and convenience samples (whoever is easiest to reach, e.g. everyone leaving a library). Each systematically over- or under-represents part of the population — a reading survey taken at a library oversamples heavy readers. The SAT loves to offer a large sample size as a distractor: a big biased sample is still biased. Size never cures bias; randomness does.

Margin of Error

A sample statistic — a sample mean or a sample percentage — is only an estimate of the true population value. The margin of error expresses how far the true value might reasonably sit from that estimate. It is reported right alongside the statistic: "$46\%$ support the measure, with a margin of error of $3\%$."

That margin defines a plausible interval for the true population value:

$\text{interval} = \text{estimate} \pm \text{margin of error}$

So $46\% \pm 3\%$ means it is plausible that the true population percentage lies anywhere from $43\%$ to $49\%$. You do not compute the margin of error from scratch on the SAT; you interpret it and reason about what makes it larger or smaller. Two levers matter most:

A bigger random sample gives a tighter, more precise estimate — the intuitive core of nearly every SAT margin-of-error question. If a question asks how to make an estimate more reliable, "survey more people" (at random) is the answer.

What the Margin Describes

The margin describes uncertainty about the population parameter — the true mean or percentage for the whole group — not about any single individual. If a sample estimates the mean household water use at $4{,}200$ gallons $\pm 150$, the interval $[4{,}050,\ 4{,}350]$ is a plausible range for the population mean. One particular household could easily use far more or far less; the interval says nothing about individuals. Mistaking a population interval for an individual range is one of the two most common wrong answers on the SAT.

What "Confidence Level" Means

A "$95\%$ confidence level" is a statement about the method, not about one specific interval. It means that if the entire process — draw a random sample, build an interval — were repeated many times, about $95\%$ of the intervals produced would contain the true population value. It is not correct to say "there is a $95\%$ chance this one interval contains the true value," and it is not a probability attached to any individual. For SAT purposes, the practical takeaways are: (1) a higher confidence level widens the interval, all else equal, and (2) the language of confidence describes long-run reliability of the procedure.

Comparing Two Estimates

A favorite hard question gives two overlapping intervals and asks what you can conclude. If Candidate A polls $48\% \pm 4\%$ (interval $44\%$–$52\%$) and Candidate B polls $45\% \pm 4\%$ (interval $41\%$–$49\%$), the intervals overlap, so the poll cannot establish who truly leads — the difference is within the noise. Two estimates differ meaningfully only when their plausible intervals separate.

Common Mistakes to Avoid

• Generalizing beyond the population that was actually randomly sampled.
• Reading the margin of error as a range for one individual instead of the population parameter.
• Thinking a bigger margin of error is "better" — it means less precision.
• Believing a large sample size or a margin-of-error calculation can rescue a biased sample.