SAT® Margin of Error and Sampling: What a Survey Can Actually Tell You
SAT Clarity Editorial ·
Editorial calendar: 2026-06-13. Actual publication date is shown above.
A survey estimate is not an exact population count. A margin of error describes sampling uncertainty under a statistical model; it does not repair a biased survey. On SAT® Math questions, the most useful first step is to identify the population, the sample, and the method used to select participants. This guide uses original examples to explain what changes when a sample grows, how to interpret a stated interval, and why a large number of responses is not enough to make every conclusion valid.
Population and sample are different quantities
A population is the full group a study aims to describe. A sample is the smaller group actually observed. If a school surveys 160 randomly selected students to estimate the percentage of all 1,600 enrolled students who support a new lunch schedule, the population contains 1,600 students and the sample contains 160. A statistic calculated from those 160 responses is an estimate of a population characteristic.
The target population is not automatically everyone mentioned in a question. A random sample drawn from one school's enrollment list supports inferences about that school's enrolled students, subject to the study's assumptions. It does not automatically represent every student in the district, every teenager in the country, or the school's future students. The selection frame limits what the sample can represent.
Underline the group from which participants were actually selected. Then compare it with the group named in each answer choice. Many reasoning questions can be resolved before calculating anything because an answer attempts to generalize far beyond the sampled population. This is a scope error, not an arithmetic error.
Read the estimate and margin together
Suppose a report estimates that 58% of students support a proposal and gives a margin of error of 4 percentage points at a stated confidence level. Subtract and add the margin: 58% − 4% = 54%, and 58% + 4% = 62%. The reported interval runs from 54% to 62%. The estimate is the center; the margin is the distance from the center to either endpoint.
The margin is not the interval's full width. The interval above is 8 percentage points wide, because there are 4 points on each side. If a question gives endpoints instead, find the midpoint and half-width: an interval from 41% to 47% has midpoint 44% and margin 3 percentage points. Label both quantities so that you do not report the width when asked for the margin.
Do not interpret the interval as saying that exactly 58% of the population supports the proposal. It also does not say that 4% of respondents answered dishonestly. Sampling variability exists because different random samples can produce different estimates, even when everyone reports their preference accurately. Measurement problems and dishonest responses are separate sources of uncertainty.
For an interval stated as estimate ± margin, the lower endpoint is estimate − margin and the upper endpoint is estimate + margin.
Percentage points are not percent changes
Moving from 58% to 62% is an increase of 4 percentage points. It is not a 4% relative increase in the original 58%. The relative increase would be 4/58, approximately 6.9%. Survey margins stated in percentage points use subtraction and addition on the percentage scale; you do not multiply the estimate by the margin percentage.
For example, an estimate of 40% with a margin of 5 percentage points gives 35% to 45%. Computing 5% of 40 and reporting 38% to 42% would answer a different calculation. When the problem states its margin in percentage points, keep that unit attached to your scratch work.
The same interval can be written using decimals: 0.40 ± 0.05 gives 0.35 to 0.45. Do not mix the representations by writing 0.40 ± 5. Consistent units are especially important when a calculator displays decimal probabilities but answer choices use percentages.
What a larger random sample changes
Holding the confidence level, population variability, sampling design, and other relevant conditions comparable, a larger random sample generally produces a smaller margin of error. A smaller sample generally produces a larger margin. The direction of this relationship is often all a question requires. More data can reduce random sampling variability without eliminating it.
In common simple-random-sampling settings where the usual approximation applies and the population is large relative to the sample, margin of error scales roughly as one divided by the square root of sample size. Under those conditions, multiplying the sample size by four approximately halves the margin. Doubling the sample size does not approximately halve it; the reduction is by a factor of the square root of two.
Treat that square-root relationship as a conditional approximation, not a universal command for every study. A question might directly supply a relationship, specify a different sampling design, or compare studies at different confidence levels. Use the information stated in the problem. If only a qualitative comparison is requested, avoid adding an unnecessary formula.
When comparing sample sizes, explicitly hold the confidence level and sampling method constant before applying the usual larger-sample, smaller-margin relationship.
Why sample quality still matters
Imagine a town wants to estimate how residents commute. One survey randomly selects residents from a comprehensive list. Another asks volunteers at a bicycle shop to respond. Even if the bicycle-shop survey gathers many responses, its participants may differ systematically from all town residents. Increasing that convenience sample's size does not by itself remove the selection bias.
A voluntary-response poll can overrepresent people with strong opinions or unusually high engagement. A survey sent only by email may miss residents without reliable internet access. A leading question can influence responses even in a well-selected sample. These problems differ from the random variation summarized by an ordinary margin of error.
Random selection helps make a sample representative in a probabilistic sense; it does not guarantee a perfect miniature of the population in every characteristic. Nonresponse can also create problems if selected participants who do not respond differ meaningfully from those who do. On a test question, attend to the selection method and any stated limitations rather than assuming that the word ‘survey’ ensures reliability.
Random sampling versus random assignment
Random sampling concerns who enters a study from a population. Random assignment concerns how study participants are allocated to treatment groups. They serve different purposes. A random sample can support generalization to the sampled population. Random assignment in a well-designed experiment can support a causal comparison between treatments, subject to the experiment's conditions.
Suppose researchers observe that students who attend an optional study club have higher average Math scores. Those students may also differ in prior preparation, motivation, available time, or access to support. The observed association alone does not establish that the club caused the higher scores. A large observational sample does not automatically solve that causal problem.
If eligible volunteers are randomly assigned to a new study program or a comparison condition, the design is stronger for evaluating the program's causal effect within the study. However, volunteers from one school are not automatically a representative sample of every student. A study can be strong for a causal comparison while limited in population generalizability. Keep the two questions separate.
Compare two survey reports carefully
Survey A reports 51% support with a margin of 5 percentage points. Survey B reports 54% support with a margin of 2 percentage points. Their intervals are 46% to 56% and 52% to 56%. Survey B's estimate is higher, and its interval is narrower. Those are direct descriptive comparisons that the supplied numbers support.
You should not conclude from the point estimates alone that population support definitely increased by 3 percentage points. The surveys may involve sampling variability, different respondents, different dates, or different methods. A formal claim about a difference requires an appropriate comparison and assumptions. Overlapping intervals should not be treated as a universal standalone significance test either.
If the question asks which report is more precise, the smaller margin indicates greater precision at the stated confidence level and under comparable assumptions. Precision is not the same as absence of bias. A narrow interval produced by a flawed selection process can still center on an estimate that systematically misses the intended population value.
Work through original practice examples
Example 1: a random survey estimates that 36% of a town's adult residents use the public library monthly, with a margin of 3 percentage points. Find the interval. The endpoints are 33% and 39%. If the town has 20,000 adult residents and the question asks for the count corresponding to the point estimate, calculate 0.36 × 20,000 = 7,200. That is an estimated count, not a direct census result.
Example 2: two otherwise comparable random surveys use samples of 400 and 1,600 people. Under the usual square-root approximation, the second sample has four times as many people, so its margin is approximately half as large. If the first margin is 6 percentage points, the second is approximately 3. The estimate itself need not remain unchanged; sample size affects uncertainty, not a guarantee of the same observed proportion.
Example 3: a principal asks the first 100 students entering the cafeteria whether the school day should start later, then claims that the result represents every district student. Two issues are visible: selection was based on arrival at a particular location, and the conclusion extends beyond the school sampled. The number 100 does not independently establish that the selection is random or that the district is represented.
Example 4: a report gives an interval of 68 to 76 minutes for an estimated mean commute time. The midpoint is (68 + 76)/2 = 72 minutes, and the margin is (76 − 68)/2 = 4 minutes. A margin of error is not restricted to percentages; its units follow the quantity being estimated.
Build a four-question checking routine
First ask, ‘Who is the population?’ Second ask, ‘How was the sample selected?’ Third ask, ‘What is being estimated, and in what units?’ Fourth ask, ‘Does the proposed conclusion stay within the design's limits?’ This sequence works for verbal survey questions as well as numerical interval questions.
When checking a calculation, confirm that the point estimate lies at the center of a symmetric estimate-plus-or-minus-margin interval. Confirm that subtracting the lower endpoint from the estimate gives the same margin as subtracting the estimate from the upper endpoint. These quick checks catch using the full width as the margin or mixing percentages and decimals.
For a missed reasoning question, write the unsupported step in one sentence. Examples include ‘I assumed volunteers were a random sample’ or ‘I confused association with causation.’ Repeating the calculation will not fix those mistakes. A targeted explanation will.
Frequently asked questions
Does a margin of error guarantee that the population value lies inside the reported interval? No. A confidence procedure has a stated long-run coverage interpretation under its assumptions; an individual interval is not an absolute guarantee. For introductory questions, describe the reported range as an estimate with uncertainty and avoid words such as ‘certainly’ unless the problem supports them.
Does a 95% confidence level mean 95% of individual observations fall inside the interval? No. An interval estimating a population proportion or mean concerns that population parameter, not a range containing a specified share of individual observations. Distinguish a confidence interval for a mean from the distribution of individuals.
Can the margin shrink while bias remains? Yes. More observations under the same flawed selection process may reduce random variability without removing a systematic selection problem. Good survey reasoning examines both the method and the numbers.
Sources and further practice
These examples are original teaching material. Check College Board's current Math-domain descriptions for the exam's scope, and use official practice for authentic question presentation. Our broad Math guide can help you connect sampling questions with ratios, percentages, and data interpretation.
For your next session, explain one survey design aloud before doing any arithmetic. Then calculate one interval from a center and margin, and reconstruct a center and margin from endpoints. Combining interpretation with calculation is more useful than memorizing an isolated rule about large samples.