Relvora Prep

SAT Math lesson · Sampling & inference

SAT sampling and studies: estimates, margin of error, and what a study can conclude

How to answer SAT questions on inference: estimating a population count from a random sample, reading a margin of error, how far a sample's result generalizes, and when a study supports cause and effect.

The core idea

A random sample's percent estimates the percent in the population it was drawn from, and no larger group. A margin of error turns the estimate into a range of plausible values for that population; a larger random sample gives a smaller margin of error. Only random assignment to treatments supports a cause-and-effect conclusion.

Core ideas

Words used here

  • Population: the whole group a question is about, such as all 3,600 students at a school.
  • Sample: the people (or things) actually measured.
  • Random sample: a sample in which every member of the population had the same chance of being chosen.
  • Estimate: a sample's result used as a guess for the population.
  • Margin of error: how far the population value plausibly is from the estimate, in either direction.
  • Observational study: researchers record what people already do.
  • Experiment: researchers assign people to treatments. With random assignment, chance decides who gets which treatment.
  • Estimating a count: population count ≈ (sample proportion) × (population size).
  • Margin of error: estimate ± margin gives the plausible range for the population value.
  • Sample size: a larger random sample gives a smaller margin of error.
  • Scope: results generalize to the population that was randomly sampled, and no larger.
  • Cause and effect: needs random assignment to treatments. Without it, only an association.

On the SAT.

  • Text questions: most of these questions describe a survey or a study in a few sentences and ask which conclusion is supported. Ask two things: was the sample random, and from which group? Were treatments randomly assigned?
  • Numbers: estimates use one proportion and one multiplication; the calculator helps with large populations.
  • Typed answers: an estimated count goes in the box without commas (44100, not 44,100).

Try it. A random sample of 200 of the 4,500 students at a school found that 48 walk to school. Estimate how many of the 4,500 students walk to school.

Show the answer

48/200 × 4,500 = 0.24 × 4,500 = 1,080 students.

Estimating from a sample

A random sample's proportion estimates the population's proportion. Multiply it by the population size to estimate a count. With a margin of error, the plausible counts run from (estimate − margin) to (estimate + margin), each applied to the population size. If each group was sampled separately at a different rate, estimate each group from its own sample and add.

Why it works. A random sample tends to mirror its population, so the share with a trait in the sample is the best single guess for the share in the population.

How to spot it: "based on the survey, what is the best estimate of the number of …", "the greatest plausible number", a table of groups with "number surveyed".

Example. A random sample of 300 of the 90,000 adult residents of a city found that 147 use public transit weekly.

  1. Sample proportion: 147/300 = 0.49.
  2. Estimate for the city: 0.49 × 90,000 = 44,100 residents.

Trap: answering 147 (the sample count) or dividing the wrong way (300/147 × 90,000).

Edge cases.

  • With a margin of error: 62% of a random sample of a school's 2,400 students ride a bus, with a margin of error of 4 percentage points. The plausible percents are 58% to 66%, so the least plausible number is 0.58 × 2,400 = 1,392 and the greatest is 0.66 × 2,400 = 1,584.
  • Groups sampled at different rates: in a middle school, 30 of the 300 grade 6 students were surveyed (12 said yes) and 20 of the 400 grade 7 students (5 said yes). Estimate 12/30 × 300 = 120 and 5/20 × 400 = 100, a total of 220. Pooling (17/50 × 700 = 238) over-weights grade 6, which was sampled at a higher rate.
  • A non-random sample gives no trustworthy estimate at all.

Check yourself. A random sample of 125 of a town's 5,500 voters found 97 who support a new library. Estimate the number of supporters in the town.

Check your answer

97/125 × 5,500 = 4,268.

Reading a margin of error

An estimate p with margin of error e says it is plausible that the population value is between p − e and p + e. The range is about the population, not the sample (the sample's value is known exactly), and it is plausible, not guaranteed. With everything else the same, a larger random sample gives a smaller margin of error.

Why it works. Different random samples give slightly different results. The margin of error measures how much they typically vary; with more people in each sample, the results vary less, so the margin shrinks.

How to spot it: "the margin of error is 3 percentage points", "which is the most plausible range", "which conclusion is most appropriate", two surveys of different sizes.

Example. A random sample of 500 of a city's residents found that 34% favor a new park, with a margin of error of 3 percentage points.

  1. Plausible range for all residents: 34 − 3 = 31% to 34 + 3 = 37%.
  2. Supported: "It is plausible that between 31% and 37% of all residents favor the park."
  3. Not supported: "exactly 34% of residents" (an estimate isn't exact), "between 31% and 37% of the 500 in the sample" (the sample's percent is exactly 34%), "must be between 31% and 37%" (plausible, not certain).

Trap: using the margin of error on one side only (34% to 37%), or doubling it.

Edge cases.

  • A mean with a margin of error works the same way: a mean of 6.8 hours with a margin of 0.3 hours gives 6.5 to 7.1 hours.
  • Two surveys: researcher A surveys 150 people and researcher B surveys 450 from the same population. B's margin of error is most likely smaller. That doesn't mean B's estimate must be closer to the truth.
  • Overlapping ranges (40% ± 4 and 44% ± 3) don't show a real difference between the two groups.

Check yourself. A survey estimates that 58% of a club's members want later hours, with a margin of error of 5 percentage points. What is the plausible range for all members?

Check your answer

53% to 63% of all the club's members.

Who the results apply to

A random sample represents the population it was drawn from, so its results can be generalized to that population, and to no larger group. A sample that isn't random (volunteers, people who are easy to reach, a subgroup) can be biased: it may over-represent people with one answer.

Why it works. Random selection gives every member of the population the same chance of being in the sample, so no kind of person is favored. People who volunteer or happen to be nearby are often different from the rest in exactly the way the survey asks about.

How to spot it: "to which group can the results be generalized?", "which method of choosing people is most appropriate?", "the largest population to which the results apply".

Example. A researcher selects 300 members at random from the 5,400 members of a fitness club and asks whether they would attend early-morning classes.

  1. The sample was drawn at random from the club's members, so the results apply to all members of the club.
  2. Not just "the 300 surveyed": they were chosen to represent the rest.
  3. Not "all adults in the region": no one outside the club could be chosen.

Trap: generalizing to a bigger, more interesting group than the one that was sampled.

Edge cases.

  • Good method: select at random from a list of the whole population.
  • Biased methods: surveying people at the club at 6:00 a.m. (they already like early hours), asking for volunteers online (strong opinions answer), or sampling at random from a subgroup such as the evening-class members (represents only them).
  • A bigger biased sample is still biased: size reduces random variation, not bias.

Check yourself. To estimate the share of a town's voters who support a new library, is it better to survey 100 people leaving the current library or 100 voters chosen at random from the voter list?

Check your answer

The random sample from the voter list: people leaving the library are more likely to support a new one.

Studies, experiments and cause and effect

Two questions decide what a study supports. Were the participants randomly assigned to the treatments? If yes, a cause-and-effect conclusion is supported; if no, only an association. Were the participants randomly selected from a population? If yes, the conclusion extends to that population; if no (volunteers), only to people like those in the study.

Why it works. Random assignment makes the groups alike in every way except the treatment (apart from chance), so a real difference in results must come from the treatment. When people choose their own habits, the people who choose differently may differ in other ways too.

How to spot it: "randomly assigned", "volunteers", "surveyed", "which conclusion is best supported by the study?"

Example. 100 volunteers are randomly assigned to eat breakfast each morning or to skip it for two weeks. The breakfast group scores significantly higher on an attention test.

  1. Random assignment: yes, so eating breakfast causes higher scores, for these participants.
  2. Random selection: no, they volunteered, so the conclusion applies to people like those in the study, not to all adults.

Trap: concluding "eating breakfast causes higher scores" from a survey of people who already eat breakfast. That study shows only an association.

Randomly assigned Not randomly assigned
Randomly selected Cause and effect, for the population Association, for the population
Volunteers Cause and effect, for people like those studied Association, for the people studied

Edge cases.

  • "Statistically significant" says the difference is unlikely to be chance; it doesn't make an observational study causal.
  • No relationship is never the supported conclusion when the study found a significant difference.
  • Reverse direction: an association between reading before bed and sleeping well could run either way, or come from a third factor.

Check yourself. A researcher randomly selects 200 employees of a company and records whether each stretches in the morning. Those who stretch report fewer backaches. What can be concluded?

Check your answer

An association between stretching and fewer backaches among the company's employees, but not that stretching causes it: no one was assigned to stretch.

How hard questions disguise it

For harder questions (L4–L5)

At L4, both design questions appear at once (random selection and random assignment), or two surveys of different sizes ask what the sample size says about the margins of error. At L5, estimates combine with the margin of error ("the greatest plausible number of the 2,400 students") or with groups sampled separately, where each group's estimate is made from its own sample. These connect to percents (a percent of a population) and to probability (expected counts from a table). Name the population, check how the sample was chosen, and only then compute.

Practice

Practice these patterns in timed sets on the sampling and studies practice page.

Common traps

  • Answering for the sample instead of the population

    147 of 300 surveyed is a count in the sample. For the 90,000 residents, apply the proportion: 147/300 × 90,000 = 44,100. When groups were sampled separately, estimate each group and add.

  • Misreading the margin of error

    An estimate of 34% with a margin of error of 3 percentage points makes 31% to 37% plausible for the population. It is about the population (the sample's 34% is exact), it goes both ways, and it is plausible, not certain.

  • Getting the effect of sample size backward

    With everything else the same, a larger random sample gives a smaller margin of error. That makes the method more precise; it doesn't guarantee that one particular estimate is closer to the truth.

  • Generalizing to the wrong group

    Results extend to the population the sample was randomly drawn from: not just to the people surveyed, and not to a larger group. Volunteers and convenience samples don't represent anyone beyond themselves.

  • Reading an association as cause and effect

    Only random assignment to treatments supports 'X causes Y'. When people choose for themselves (an observational study), the study shows an association: other differences between the groups could explain it.