Parameter in AP Statistics

One fixed population value vs. a sample number that keeps changing

A parameter summarizes a variable for the whole population. A statistic summarizes the same variable for a sample. Topic 1.2 (Variables) asks you to sort numbers in a study into these two bins, and the test is simple: ask whether the number came from everyone in the population or from a subset.

The symbols help. Parameters use Greek letters or plain p (μ, σ, p). Statistics use Roman letters with bars or hats (x̄, s, p̂). If a stem says "the mean of all 2,400 students at the school is 3.1 hours," that's μ. If it says "a random sample of 50 students averaged 3.4 hours," that's x̄.

The tricky case is a census, which collects data from every member of the population. Suppose a report says 45% of a city's residents support a law. If every resident was surveyed, 45% is the parameter p. If 600 residents were randomly sampled, it's the statistic p̂. The number alone can't tell you, so check how the data were collected.

A statistic from one sample often won't equal the parameter, and it would change if you drew a different sample. The parameter doesn't move. That gap between a fixed target and a varying estimate is what the rest of the course is built to handle.

μ_X, σ_X, n, and p as the parameters of a probability model

Topic 2.9 (Parameters of Random Variables) extends the idea from populations to probability distributions. The expected value E(X) = μ_X = Σ xᵢ·P(xᵢ) and the standard deviation σ_X are parameters because they're fixed properties of the model, not results of one run of it.

Interpret them as long-run behavior in context. If X is the number of defective bulbs in a box and μ_X = 0.8, a strong answer says that over many, many boxes, the average number of defective bulbs per box would be about 0.8. Saying "the box has 0.8 defective bulbs" misses the long-run meaning.

In Topic 2.10 (The Binomial Distribution), the number of trials n and the success probability p define the model, and they produce its mean np and standard deviation √[np(1 − p)]. A simulation can estimate binomial probabilities, but the simulated proportions are estimates. The p you put into the model is the parameter.

How parameters set the center and spread of a sampling distribution

Parameters do their heaviest work in Unit 3 (Inference for Categorical Data: Proportions), starting with Topic 3.2 (Sampling Distributions for Sample Proportions), where the question is how far a sample result tends to land from the population value. With independent sampled values, the sampling distribution of p̂ has mean μ_p̂ = p and standard deviation σ_p̂ = √[p(1 − p)/n]. The center is the parameter itself, which is why p̂ is an unbiased estimator of p.

The spread depends on the parameter too. For a fixed n, p(1 − p) is largest at p = 0.5, so a population proportion of 0.5 gives the most variable sample proportions. The large-counts check (np ≥ 10 and n(1 − p) ≥ 10) is also written in terms of the parameter.

Means work the same way in Topic 4.1. Say a population of adults has systolic blood pressure with μ = 120 mmHg and σ = 15 mmHg. For random samples of 100, the sample means center at 120 mmHg with standard deviation 15/√100 = 1.5 mmHg. So sample means typically fall about 1.5 mmHg from the true population mean. Topics 3.9 and 4.6 apply the same logic to differences, with centers at p₁ − p₂ and μ₁ − μ₂.

When the parameter is unknown, as in a real confidence interval, you plug in the statistic and the result is called a standard error. In Topic 5.5, the slope b and intercept a of a least-squares line are statistics for the same reason. They come from sampled values, and a new sample would give a different line.

Naming the parameter in hypotheses, intervals, and study questions

On Question 3 (Inference), a complete answer starts by defining the parameter in context. Take a store testing whether the proportion of customers using its loyalty app has changed from last year's 0.22. You'd write: p = the true proportion of all this store's customers who use the loyalty app, with H₀: p = 0.22 and Hₐ: p ≠ 0.22. The sample result of 0.25 is p̂, and it belongs in the calculation, not the hypotheses.

Putting a statistic in a hypothesis, like H₀: p̂ = 0.22, costs credit. You already know p̂ exactly, so there's nothing to test about it. A second mistake is a vague definition such as "p = the proportion of customers." To define the parameter in context, say "true" or "all," and name the population.

The same habit shows up in Question 1 (Multi-Focus, Practices 1 and 2). Topic 1.10 says an investigative question should identify the parameter, plus the direction of the alternative for a test or the goal of estimating a range of values for an interval. It should also name the population the conclusion applies to.

Multiple-choice items cover the other pieces. Some give μ, σ, or p in the stem and ask for a sampling distribution's mean or standard deviation. Others ask which statement correctly separates a population value from a sample result. In Topic 3.8, the true parameter decides which error is even possible. A Type II error requires the parameter to actually differ from the null value, and power goes up as that true value sits farther from the null, with the other factors held fixed.