4.4 Sampling and Inference Basics

Why This Section Matters

You will not design a full AP Statistics inference course on Praxis 5165, but you will judge whether a sampling plan is fair, whether a conclusion overreaches the data, and whether a simulation correctly models a random process. These ideas appear in the Statistics & Probability domain and in task-of-teaching scenarios about student projects.

Population, Sample, and Parameter

• Population: entire group you want to learn about (all students at a school, all voters in a district).
• Sample: subset actually measured.
• Parameter: numerical fact about the population (true mean homework time for all students).
• Statistic: numerical fact about the sample (mean homework time for 40 surveyed students).

We use a statistic to estimate a parameter. The quality of that estimate depends on how the sample was chosen and how large it is — not just on whether someone calculated a mean correctly.

Random Sampling and Bias

A simple random sample (SRS) from a roster gives each student an equal chance to be selected. That does not guarantee the sample perfectly mirrors the population, but it makes large bias less likely than convenience methods.

Praxis wording is direct: choose the plan that randomly selects from the entire target population unless stratification is explicitly justified.

Worked Example: Picking the Best Sample

A teacher wants to estimate mean nightly homework minutes for all 900 students in a high school.

• Plan A: Survey every student in one honors algebra class.
• Plan B: Ask volunteers at lunch to text their answers.
• Plan C: Number the roster 1–900, use a random number generator to pick 60 students, and survey those students.

Plan C is best. It targets the full roster and uses random selection. Plan A restricts to one advanced class (likely underestimates or overestimates depending on school culture, but clearly not representative). Plan B is voluntary and likely excludes busy or disengaged students.

From Sample to Inference (Informal)

Formal confidence intervals are not the focus, but you should understand the logic:

1. Collect data from a random sample.
2. Compute a statistic (sample mean, sample proportion).
3. Acknowledge sampling variability — another random sample would give a slightly different result.
4. Draw a limited conclusion about the population, not certainty.

If only 18 of 20 students in an after-school club prefer online homework, you cannot conclude that all 900 students prefer online homework. The sample is neither random nor representative of the whole school.

Simulation on Praxis 5165

A simulation imitates a chance process with a random model. Valid simulations match the sample space and independence assumptions of the real situation.

Worked Example: Simulating Two Dice

To model rolling two fair six-sided dice and recording the sum:

• Valid: Generate two independent random integers from 1 to 6 and add them. Each die outcome is equally likely and the dice do not affect each other.
• Invalid: Generate one digit and double it — that does not produce the same sum distribution as two dice (you cannot roll a sum of 1, and sums like 7 are wrong frequency).

Teaching items test whether students understand that each random device in the model must represent one part of the physical process.

Margin of Error Intuition

When a poll reports "52% support with margin of error ±3%," the plausible range for the population proportion is roughly 49% to 55% if methods are sound. Larger random samples usually shrink the margin of error. However, if the sample is biased, a precise-looking percentage can still be wrong — accuracy of the center matters as much as width.

Task-of-Teaching Moves

When a student generalizes from a biased sample, the strongest feedback:

1. Names the bias (who was left out).
2. Proposes a random selection from the defined population.
3. States the limit of what the current data can support.

Quick Exam Checklist

• Who is the population?
• Was every population member eligible to be selected?
• Is the conclusion about the population or only the sample?
• Does a simulation represent each stage of the random process independently?

Other Sampling Designs You May See

A stratified random sample randomly selects within subgroups (strata) such as grade level when those groups matter. A systematic random sample picks every k-th name from a shuffled roster. Both are acceptable when every population member has a fair chance to be included; convenience subsets are not.

Worked Example: Overgeneralizing

A student project finds that 18 of 20 club members prefer digital notes. The club meets after school and volunteers for technology tasks. A valid critique: the sample is neither random nor representative of all students, so the statistic 18/20 = 90% estimates club preference only — it cannot support a claim about the whole school without a new sampling design.

Polling Language on the Exam

Random-sample results support tentative population conclusions; biased-sample percentages stay wrong no matter how precisely they are reported.