Decide what the evidence can support next

Try

A voluntary poll of10,000 visitors reports support to three decimal places. Before interpreting that precision, ask who the visitors are, who responded and whose opinion the report claims to represent. A small numerical standard error can coexist with large selection bias. Statistical inference needs a credible connection between the observed data and the target.

Predict

Predict whether a95% confidence interval describes95% of individual observations, a moving population parameter or a repeated procedure’s coverage. Also predict whether a paired study with20 people measured twice contains40 independent units. The unit of analysis and the source of randomness must be identified before choosing a method.

Experiment

Audit this original synthetic plan: select120 distinct adults randomly from a frame of2000 eligible residents, and record a binary response. Suppose72 respond yes and all selected people provide answers. The sample proportion is72/120=0.6. The sampling fraction is120/2000=0.06, below the common10% approximation threshold. The observed success and failure counts are72 and48.

Now consider a separate null model p0=0.2 with n=80. Its expected counts are16 successes and64 failures. Those use the proposed model, not observed counts. Write a checklist identifying which counts belong to which question. This lesson checks readiness; it does not turn these examples into an unintroduced confidence interval or test.

Understand

A parameter describes the target population or process, such as a success probability p or mean μ. A statistic is calculated from the sample, such as p̂ or x̄. The sampling distribution describes how that statistic varies under repeated sampling. A standard error quantifies its sampling spread under a model, often using estimated quantities in practice.

An inference plan needs a clear target, suitable collection design, appropriate units and justified distributional conditions. For proportions, a normal approximation commonly requires enough successes and failures; the relevant theoretical or estimated probabilities depend on the procedure. For means, small samples with severe skew or outliers can make routine methods unreliable. For paired data, inspect and analyze the within-pair differences, while checking independence between pairs.

The10% condition for sampling without replacement supports an approximate independence calculation when sampling randomly from a much larger finite population. It does not establish representative selection, eliminate nonresponse or make dependent repeated measurements independent. Separate design checks from numerical checks rather than treating any one threshold as a certificate of validity.

A confidence level describes the long-run coverage of an interval procedure under its assumptions. For a fixed population parameter, repeated samples yield different intervals; about95% cover the parameter for a valid95% procedure. After observing one interval, the parameter and endpoints are fixed in this frequentist interpretation. It is not an interval containing95% of individuals, nor a guarantee about the particular interval obtained.

Hypothesis testing asks how unusual data are under a specified model and alternative. A p-value is not the probability that the null hypothesis is true. A statistically detectable difference need not be practically important, and failure to reject does not prove equality. Exact procedures, confidence intervals, test statistics, error types and power need their own teaching and practice, which are outside this foundations unit.

Apply

For the paired trial in practice, state that20 differences are analyzed and define the intended mean difference. Examine recruitment, assignment, compliance, outliers and dependence before proposing a procedure. Write a qualified conclusion template that leaves numerical inference blank until a justified method is actually carried out. Honest incompleteness is better evidence than an invented p-value.

Continue

Continue to the forthcoming inference sequence for one- and two-proportion procedures, association tests and one-, paired- and two-mean inference. These foundations provide preparation, not complete Fall2026 AP Statistics coverage or an independently assessed statistics credential.