Skip to content
AP Statistics · Inference for Categorical Data: Proportions

Sample proportion

Sample proportion appears in 12 reviewed AP Statistics questions and sits in Inference for Categorical Data: Proportions. Below are worked examples showing exactly how the exam tests it, with the answer marked and every distractor explained.

How Sample proportion is tested

Example 1Easier
A town's public works department wants to estimate the proportion p of its 5,000 streetlights that are burned out. An inspector selects a simple random sample of 200 streetlights and finds that 18 of them are burned out, so p-hat = 18/200 = 0.09. Which of the following statements about p-hat as an estimator of p is correct?
  1. Ap-hat is a biased estimator of p because only 4% of the town's streetlights were sampled.
  2. BDoubling the sample size to 400 streetlights would make p-hat a less biased estimator of the true proportion p.
  3. Cp-hat is unbiased for p, so its values center at p over many samples, though 0.09 likely differs from p.Correct
  4. DBecause the sample was randomly selected, the observed value p-hat = 0.09 must be exactly equal to the true p.

Why C is right

With random sampling, the sample proportion p-hat is an unbiased estimator of the population proportion p: the mean of its sampling distribution equals p. Unbiased describes how the estimator behaves over many repeated samples, not any single sample. So the one observed value, p-hat = 18/200 = 0.09, is subject to sampling variability and will probably not equal p exactly. The sample is 200/5,000 = 4% of the population, which does not bias p-hat.
Why the other 3 choices miss
A
Wrongly thinks a small sampling fraction causes bias. Bias comes from how a sample is selected, not from what share of the population is sampled. A 4% random sample still gives an unbiased p-hat, and it even meets the 10% condition.
B
Mixes up bias with variability. p-hat is already unbiased for a random sample of any size. A larger sample reduces the spread of p-hat (its standard deviation, sqrt(p(1-p)/n), gets smaller), not its bias.
D
Confuses unbiased with exact. Random sampling removes systematic bias, but any single p-hat still differs from p because of sampling variability.
Example 2Easier
A large school district's records show that 30% of its students ride a bus to school. A researcher plans to take a random sample of students and compute p-hat, the proportion who ride a bus. If the sample size is increased from n = 100 to n = 400, how does the standard deviation of the sampling distribution of p-hat change?
  1. AIt decreases to one-fourth, about 0.011.
  2. BIt stays the same, at about 0.046.
  3. CIt decreases by half, from about 0.046 to about 0.023.Correct
  4. DIt doubles, to about 0.092.

Why C is right

The standard deviation of the sampling distribution of p-hat is sqrt(p(1-p)/n). With p = 0.30 and n = 100, it is sqrt(0.21/100) = sqrt(0.0021), about 0.0458. With n = 400, it is sqrt(0.21/400) = sqrt(0.000525), about 0.0229. Multiplying n by 4 divides the standard deviation by sqrt(4) = 2, so the spread is cut in half.
Why the other 3 choices miss
A
Divides the standard deviation by 4 instead of by sqrt(4) = 2. This forgets that n is under a square root (0.0458/4 is about 0.011).
B
Assumes the spread of p-hat depends only on p and not on sample size, so it ignores the n in the denominator of sqrt(p(1-p)/n).
D
Gets the direction wrong by thinking a larger sample gives more variability, perhaps mixing up the spread of a count with the spread of a proportion. A larger n makes p-hat less variable, not more.

Studied alongside

Other concepts from Inference for Categorical Data: Proportions.