Estimating Population Proportions with Random Samples

Lumi sets down the bowl and turns to face you. "Here's the situation," she says. "You almost never get to measure every single person or object in a group you care about — that group is called the **population**. So instead, you measure a smaller **sample** and use what you find to make an educated estimate about the whole population." She taps the whiteboard. "The critical question is: *which* sample members did you pick? If your selection method systematically favors some members over others, your sample is **biased**, and any conclusion you draw is not trustworthy — even if the sample is large. Bias is about *how* you selected, not *how many* you selected." "A **simple random sample (SRS)** fixes this. When every member of the population has an equal, independent chance of being chosen — exactly like drawing numbered tiles from this bowl without looking — no systematic preference sneaks in. That fairness is what makes your conclusion defensible." "Once you have an unbiased sample, you compute a **sample proportion**: divide the count of the trait you're tracking by the total sample size. For example, if 34 out of 80 randomly chosen students prefer morning classes, the sample proportion is 34 ÷ 80 = 0.425, or about 42.5%. You then argue — carefully — that the population proportion is *likely near* 42.5%." "But how near? That is where the **margin of error** comes in — and it must always be paired with a **confidence level**. A 95% confidence interval is the standard: it means that if you repeated this sampling process many times, about 95 of every 100 resulting intervals would contain the true population proportion. To *approximate* the margin of error, use this rule of thumb: **margin of error ≈ 1/√n**, where *n* is your sample size. With n = 80: 1/√80 ≈ 1/8.9 ≈ 0.112, or about ±11%. A more precise calculation uses the sample proportion and a z-score, but 1/√n gives you the right order of magnitude quickly." "A larger random sample produces a smaller margin of error. With n = 400: 1/√400 = 1/20 = 0.05, so ±5%. With n = 1600: 1/√1600 = 1/40 = 0.025, so ±2.5%. You can also use **simulation**: draw thousands of virtual random samples from a model population, compute each sample proportion, and observe the spread of results — that spread directly shows you what a margin of error means in practice." "When you write a conclusion, always state (1) what population you are describing, (2) how the sample was drawn, (3) whether the method was genuinely random, and (4) the confidence level for your interval. Those four elements are what make statistical inference defensible rather than just a guess." **Non-stranding hint — if you get stuck:** Ask yourself three questions: Did every member of the population have an equal chance of being selected? If no, name the bias. If yes, compute the sample proportion and then approximate the margin of error with 1/√n. That sequence will get you through any inference problem in this lesson.