
describe the sampling distributions of the sample mean and sample proportion, relate their means and standard deviations to the population’s, and state and apply the central limit theorem.
Instruction
A statistic such as x̄ or p̂ is a random variable: its value depends on which sample you happened to draw. Its probability distribution over all possible samples of size n is its sampling distribution (PS.15). Inference works by knowing how a statistic behaves across samples, so that one sample can be interpreted.
The population. Daily precipitation at St. Louis Lambert International Airport, every day from 2000 to 2024 (NOAA GHCN-Daily, N = 9,132). Its parameters: μ = 0.1177 inch per day, σ = 0.3457 inch. It is extremely skewed: 69% of days have zero rain, and the wettest day had 8.64 inches.
What theory predicts (PS.16). For random samples of size n:
- The mean of the sampling distribution of x̄ equals the population mean: μx̄ = μ. The sample mean is unbiased.
- Its standard deviation is σx̄ = σ/√n. This follows from RLP Statistics A’s rules: for independent observations, the variance of a sum of n of them is nσ2, and dividing the sum by n divides the variance by n2, giving σ2/n. It holds exactly with replacement, and very nearly without replacement when n is less than 10% of the population.
- For a proportion, μp̂ = p and σp̂ = √(p(1 − p)/n).
What simulation shows. One run of 10,000 random samples of each size:
| n | mean of x̄’s | SD of x̄’s | σ/√n | shape (skewness) |
|---|---|---|---|---|
| 5 | 0.1173 | 0.1564 | 0.1546 | strongly right-skewed (2.65) |
| 30 | 0.1178 | 0.0625 | 0.0631 | still right-skewed (1.02) |
| 100 | 0.1181 | 0.0345 | 0.0346 | mildly skewed (0.54) |
The centers match μ = 0.1177, and the spreads match σ/√n to within about 1%. For the proportion of rainy days (p = 0.3117) with n = 100, the simulated p̂’s had mean 0.3116 and SD 0.0459; theory gives √(0.3117 × 0.6883/100) ≈ 0.0463.
The central limit theorem (CLT). For random samples from any population with finite σ, the sampling distribution of x̄ approaches a normal distribution as n grows. The more skewed the population, the larger n must be. Here the skewness of x̄ fell from 2.7 to 1.0 to 0.54 as n went from 5 to 30 to 100, on its way to 0 (normal). For this extreme population, “n ≥ 30” is not enough: the sampling distribution at n = 30 is still visibly skewed. For a mildly skewed population, such as daily high temperatures, n = 30 is plenty. The rule of thumb is a starting point; the population’s shape decides. PS.9 applies the same theorem to counts: a binomial count is a sum of 0/1 trials, which is why it becomes normal for large n.
What the CLT does not say. It does not make the data normal. A sample of 100 rainfall days is still mostly zeros. It is the distribution of the sample mean that becomes normal. And it says nothing about biased samples. A convenience sample’s mean is not centered on μ, however large n is.
Why it matters. Because x̄ and p̂ are approximately normal with known SDs, you can say how far a statistic is likely to be from its parameter. Unit 2 turns that into confidence intervals.
📖 Read for this lesson
The reading opens over the lesson; close it to come back to exactly where you were. The checks below draw on it.
Formative check
Work through these before moving on. They are not graded — they tell you whether the standard below has landed yet.
Samples of 30 days from St. Louis daily rainfall (69% zeros, very skewed). What will the histogram of 10,000 sample means look like?
With σ = 0.3457 and n = 100, the standard deviation of x̄ is 0.3457/ ≈ .
Simulation & tools
Build sampling distributions from real rainfall
Data: NOAA NCEI GHCN-Daily, St. Louis Lambert International Airport (USW00013994), 2000-2024 (9,132 days, retrieved 2026-11-10). The population is in this lesson’s reading: Table 1 groups all 9,132 days into seven precipitation ranges (6,286, 1,050, 598, 502, 393, 251, 52 days) with the mean of each group, and states the parameters μ = 0.1177 in, σ = 0.3457 in and p = 2,846/9,132 = 0.3117. The individual days are too many to print, so build a stand-in population from Table 1 (each group’s count of days, each valued at its group mean; its mean is 0.1177 and its SD 0.3313, so the SD is slightly low because variation inside the groups is lost) and use the printed μ and σ for the theory. The reading’s Records A and B are the recorded run to check against.
- Graph the stand-in population of PRCP (a histogram of Table 1 will do). Record its mean, SD and shape, and compare with the printed μ and σ.
- With the Sampler plugin, draw 500 samples of size 5 and collect the mean of PRCP from each. Graph the 500 means; record their mean and SD and compare with μ and σ/√5.
- Repeat for n = 30 and n = 100. At which n does the histogram of means look roughly symmetric?
- Repeat step 2 with a rainy indicator (1 for the 2,846 days with PRCP ≥ 0.01, 0 for the other 6,286) for n = 100, and compare the SD of the p̂’s with √(p(1 − p)/100).
Spreadsheet route: =INDEX(range, RANDBETWEEN(…)) samples as in RLP Statistics A Lesson 3.2. Python route (optional): JupyterLite with random.choices and statistics.mean.
Open in a new tab ↗ · CODAP, Concord Consortium · MIT open source
The central limit theorem guarantees that a large convenience sample’s mean will be close to the population mean.
How confident are you that you can describe the center, spread and shape of the sampling distributions of x̄ and p̂, and explain the central limit theorem?
Practice
Work these on paper or in your notebook, then open Check your answer. Aim for all of Fluency and Application; try at least one Challenge.
Printable version: this unit’s practice workbook (PDF)
Fluency
Build speed and accuracy with the core skill.
- Daily highs at St. Louis (2000-2024) have σ ≈ 20.05°F. Find the SD of x̄ for random samples of 25 days, 64 days and 100 days.
Check your answer
Answer: 4.011, 2.507 and 2.005°F.σ/√n. - The population mean daily high is 67.59°F. What is the mean of the sampling distribution of x̄ for n = 40?
Check your answer
Answer: 67.59°F.x̄ is unbiased. - The proportion of St. Louis days at or above 90°F is p ≈ 0.1380. Find the SD of p̂ for n = 50 and n = 200.
Check your answer
Answer: 0.0488 and 0.0244.√(p(1 − p)/n). - A population has σ = 12. What sample size makes the SD of x̄ equal to 2?
Check your answer
Answer: n = 36.12/√n = 2. - Is the 10% condition met for a sample of 500 days from the 9,132-day St. Louis record?
Check your answer
Answer: Yes: 500 < 913.2. - By what factor does the SD of x̄ change when n goes from 25 to 100?
Check your answer
Answer: It is halved. - True or false: the central limit theorem says that a large sample of data will be approximately normal.
Check your answer
Answer: False.It is the sampling distribution of x̄ that approaches normal, not the data.
Application
Use the skill in context. Show your reasoning.
- Use the CLT to estimate the probability that a random sample of 36 St. Louis days has a mean high above 75°F (μ = 67.59, σ = 20.05). A simulation of 5,000 such samples gave a proportion of 0.0092; compare.
Check your answer
Answer: SD of x̄ = 20.05/6 ≈ 3.342; z ≈ 2.22; P ≈ 0.0133. The simulation (0.0092) is close: both say about 1%. The small gap reflects simulation variability and the mild skew of daily highs. - A reporter takes the first 30 days of July at St. Louis as a ‘sample’ to estimate the station’s mean annual daily high. Does the CLT make x̄ a good estimate? Explain.
Check your answer
Answer: No. The CLT describes random samples; 30 consecutive July days are not random and are systematically different (summer), so x̄ is biased however normal its sampling distribution would be. - Daily rainfall at St. Louis (σ ≈ 0.346 in, extremely skewed) and daily highs (σ ≈ 20.1°F, mildly skewed): for which population is n = 30 enough for x̄ to be approximately normal? Explain.
Check your answer
Answer: Daily highs: the population is only mildly skewed. For rainfall (69% zeros, skewness about 6.0), x̄ at n = 30 is still visibly skewed; a larger n is needed.
Challenge
Stretch problems. Expect to think before you write.
- What is the smallest sample size for which the SD of x̄ for St. Louis daily highs is at most 1°F?
Check your answer
Answer: n ≥ (20.05)² ≈ 402.2, so n = 403.σ/√n ≤ 1.
Review
Keep earlier skills sharp.
- X is normal with mean 500 and standard deviation 100. Find P(289 < X < 614).
Check your answer
Answer: 0.8554Standardize both ends, then subtract the cumulative areas. - (Lesson 1.2) For N(56.64, 1.88), find P(X > 59).
Check your answer
Answer: ≈ 0.1047.Compare with the actual proportion of years, 8/75 ≈ 0.107: the normal model is a good approximation for these annual means. - (RLP Statistics A) A random variable has E(X) = 5 and SD 2. Find the mean and SD of 3X + 1.
Check your answer
Answer: Mean 16, SD 6.
Nine Thousand Days of Rain: A Skewed Population and a Simulation You Can Check
NOAA NCEI GHCN-Daily, station USW00013994 (St. Louis Lambert International Airport), element PRCP, 2000-2024 (N = 9,132 days); simulation records generated for this course with Python’s random module, seed 2026; retrieved 2026-11-10 · Public domain (works of the United States government), with SLS original simulation records
Why this reading. The lesson’s table of simulated sampling distributions comes from a run you did not see, on a population of 9,132 days you do not have. The workshop asks you to build the same distributions in CODAP or a spreadsheet, which needs the download and the tool. Here is the route that needs neither: the population summarized closely enough to recover its parameters, and a simulation recorded with its seed and code, with thirty samples of five printed in full so you can verify the arithmetic of a sample mean.
Table 1. The population: daily precipitation at St. Louis Lambert International Airport, 2000–2024
NOAA NCEI GHCN-Daily, station USW00013994, element PRCP, every day from January 1, 2000 to December 31, 2024. N = 9,132 days, none missing. Retrieved 2026-11-10. The individual days are too many to print; grouped, the population looks like this.
| Daily precipitation (in) | Days | Fraction | Mean within group |
|---|---|---|---|
| exactly 0 | 6,286 | 0.688 | 0.000 |
| 0.01–0.09 | 1,050 | 0.115 | 0.037 |
| 0.10–0.24 | 598 | 0.065 | 0.162 |
| 0.25–0.49 | 502 | 0.055 | 0.359 |
| 0.50–0.99 | 393 | 0.043 | 0.701 |
| 1.00–1.99 | 251 | 0.027 | 1.355 |
| 2.00 or more | 52 | 0.006 | 2.748 |
Parameters, computed from all 9,132 days: μ = 0.1177 inch, σ = 0.3457 inch (population SD, dividing by N), skewness 6.0. Days with measurable rain (at least 0.01 in): 2,846, so p = 2,846/9,132 = 0.3117. Wettest day 8.64 in (2022-07-26, the day of the St. Louis flash flood). Check μ from the grouped table: Σ(days × group mean)/N = 0.1177, within rounding of the true mean. The population is 68.8% zeros with a long thin tail; no histogram of it looks anything like a bell.
Record A. Thirty samples of five days, in full
Generated in Python with rng = random.Random(2026) and rng.choices(population, k=5) repeated, where population is the 9,132 PRCP values in date order. Sampling is with replacement, which for 5 of 9,132 is indistinguishable from without. Anyone who runs those lines on the same file gets this table.
| Sample | Day 1 | Day 2 | Day 3 | Day 4 | Day 5 | x̄ |
|---|---|---|---|---|---|---|
| 1 | 0.00 | 0.00 | 0.27 | 0.00 | 0.22 | 0.098 |
| 2 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.000 |
| 3 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.000 |
| 4 | 0.55 | 0.00 | 0.00 | 0.04 | 0.50 | 0.218 |
| 5 | 0.00 | 0.42 | 0.00 | 0.03 | 0.00 | 0.090 |
| 6 | 0.01 | 0.12 | 0.02 | 0.00 | 0.00 | 0.030 |
| 7 | 0.00 | 0.05 | 0.00 | 0.00 | 0.00 | 0.010 |
| 8 | 0.00 | 0.90 | 0.00 | 1.72 | 0.00 | 0.524 |
| 9 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.000 |
| 10 | 0.00 | 0.00 | 0.03 | 0.00 | 0.00 | 0.006 |
| 11 | 0.00 | 0.00 | 0.21 | 0.00 | 0.00 | 0.042 |
| 12 | 0.13 | 0.05 | 0.00 | 0.07 | 0.00 | 0.050 |
| 13 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.000 |
| 14 | 0.00 | 0.18 | 0.00 | 0.00 | 0.00 | 0.036 |
| 15 | 0.05 | 0.00 | 0.00 | 0.00 | 0.01 | 0.012 |
| 16 | 0.04 | 0.00 | 0.00 | 0.00 | 0.00 | 0.008 |
| 17 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.000 |
| 18 | 0.00 | 0.00 | 0.24 | 0.00 | 0.00 | 0.048 |
| 19 | 0.00 | 0.00 | 0.00 | 0.00 | 0.48 | 0.096 |
| 20 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.000 |
| 21 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.000 |
| 22 | 0.00 | 0.26 | 0.00 | 0.00 | 0.03 | 0.058 |
| 23 | 0.00 | 0.00 | 0.00 | 0.02 | 0.00 | 0.004 |
| 24 | 0.00 | 0.00 | 0.00 | 0.33 | 0.00 | 0.066 |
| 25 | 0.00 | 0.02 | 0.09 | 0.00 | 0.00 | 0.022 |
| 26 | 0.00 | 0.00 | 0.02 | 0.26 | 0.00 | 0.056 |
| 27 | 0.00 | 0.11 | 0.00 | 0.00 | 0.00 | 0.022 |
| 28 | 0.08 | 0.00 | 0.00 | 0.32 | 0.17 | 0.114 |
| 29 | 0.00 | 0.14 | 0.00 | 0.00 | 0.00 | 0.028 |
| 30 | 0.00 | 0.00 | 0.00 | 0.48 | 0.22 | 0.140 |
7 of the thirty samples are all zeros, and their means are 0. The thirty means have mean 0.0593 and SD 0.1014, against μ = 0.1177 and σ/√5 = 0.1546. Thirty repetitions is not many: the mean of 30 means has its own standard error, about 0.0282. That is why the lesson uses 10,000 repetitions, and why Record B uses 1,000.
Record B. One thousand samples at each of three sizes
Same seed and code, k set to 5, 30 and 100, each run restarted from random.Random(2026). (The first 30 samples of the n = 5 run are Record A.)
| n | Mean of the 1,000 x̄’s | SD of the x̄’s | σ/√n | Skewness | x̄ ≥ 0.20 | x̄ = 0 exactly |
|---|---|---|---|---|---|---|
| 5 | 0.1169 | 0.1552 | 0.1546 | 2.18 | 214 of 1,000 | 172 of 1,000 |
| 30 | 0.1167 | 0.0631 | 0.0631 | 0.81 | 108 of 1,000 | 0 of 1,000 |
| 100 | 0.1177 | 0.0355 | 0.0346 | 0.48 | 16 of 1,000 | 0 of 1,000 |
Distribution of the 1,000 sample means, counted in bins (inches):
| x̄ from…to | 0–0.02 | 0.02–0.04 | 0.04–0.06 | 0.06–0.08 | 0.08–0.1 | 0.1–0.12 | 0.12–0.14 | 0.14–0.16 | 0.16–0.2 | 0.2–0.3 | 0.3–0.5 | 0.5–1 | 1 or more |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| n = 5 | 349 | 104 | 57 | 50 | 57 | 42 | 52 | 21 | 54 | 100 | 84 | 27 | 3 |
| n = 30 | 16 | 72 | 101 | 140 | 138 | 108 | 97 | 100 | 120 | 101 | 7 | 0 | 0 |
| n = 100 | 0 | 3 | 30 | 111 | 185 | 222 | 188 | 144 | 101 | 16 | 0 | 0 | 0 |
Compare with the lesson’s run of 10,000: centers near 0.118 in both, SDs matching σ/√n within a few percent (n = 100: 0.0355 here, 0.0345 in the 10,000-sample run, 0.0346 from theory), skewness falling from about 3 to about 1 to about 0.5. Two honest simulations of the same population disagree in their third decimal, and neither is wrong. The n = 5 histogram has its largest bin at the very left; the n = 100 histogram is a slightly lopsided but recognizable hump between 0.04 and 0.20.
The proportion. For n = 100 and the rainy-day indicator, the 1,000 simulated p̂’s (same seed) have mean 0.3107 and SD 0.0467; theory says p = 0.3117 and √(p(1 − p)/100) = 0.0463.
A question the sampling distribution answers
Lesson 1.5 will ask you to use a simulated distribution to judge a sample. Rehearse it here: a random sample of 30 days averaged 0.20 inch. Of the 1,000 simulated means at n = 30, 108 were 0.20 or more, so a mean that large comes up about 10.8% of the time from this population (the 10,000-sample run gives 10.0%). Unusual, not astonishing. At n = 100 it came up 16 times in 1,000, and a 100-day sample averaging 0.20 would be strong evidence that the days were not drawn at random from this record, or that they come from a wetter period.
Sources: NOAA NCEI, Daily Summaries (GHCN-Daily) data search (ncei.noaa.gov); NOAA NCEI Access Data Service (the query used here) (ncei.noaa.gov); Python documentation, the random module (docs.python.org)