AP Statistics: Sampling Distributions of Sample Proportions

Understanding how sample results vary is the cornerstone of statistical inference. The sampling distribution of a sample proportion provides the theoretical model that allows us to predict, quantify, and make sense of this variability. Mastering this concept is your bridge from simply calculating a statistic from one sample to making reliable claims about an entire population.

The Core Idea: From One Sample to All Possible Samples

When you take a single random sample and calculate the proportion of individuals with a certain characteristic, you get one sample proportion, denoted $\hat{p}$. For example, if you survey 100 students and 45 support a new policy, your sample proportion is $\hat{p} = 0.45$. If you took a different random sample of 100 students, you would almost certainly get a different value for $\hat{p}$. This natural fluctuation is called sampling variability.

The sampling distribution of the sample proportion is the distribution of all possible values of $\hat{p}$ from all possible random samples of the same size $n$ from the same population. Instead of imagining one sample, you imagine taking every conceivable sample of size $n$, calculating $\hat{p}$ for each, and plotting a histogram of all those values. This theoretical distribution has predictable, describable properties.

The Shape, Center, and Spread of the Sampling Distribution

For a categorical variable, the population has a true proportion of success, $p$. The sampling distribution of $\hat{p}$ has a clear relationship to this population parameter.

• Center: The mean of all possible $\hat{p}$'s is equal to the population proportion $p$. In symbols, the mean of the sampling distribution, $\mu_{\hat{p}} = p$. This means the sampling distribution is centered on the true population value; $\hat{p}$ is an unbiased estimator of $p$.
• Spread: The standard deviation of the sampling distribution, called the standard error of the sample proportion, measures how much $\hat{p}$ typically varies from $p$. Its formula is:

$$\sigma_{\hat{p}} = \sqrt{\frac{p(1-p)}{n}}$$ This formula reveals two crucial insights: variability decreases as sample size $n$ increases (larger samples give more precise estimates), and variability is greatest when $p = 0.5$.

• Shape: This is where a powerful theorem comes into play. The Central Limit Theorem for Proportions states that for a sufficiently large sample size, the sampling distribution of $\hat{p}$ will be approximately Normal.

The Conditions for Normality

You cannot safely use the Normal model for every sampling distribution. You must check these two conditions:

1. Random Condition: The data must come from a random sample from the population of interest. Without this, the mathematics of the sampling distribution does not apply.
2. Large Counts Condition: This ensures the sample size is large enough for the Normal approximation to be valid. You must check that both:

$$np \ge 10$$ $$n(1-p) \ge 10$$ Here, $n$ is the sample size and $p$ is the population proportion you are testing against. In practice, if $p$ is unknown, you use your sample proportion $\hat{p}$ to check the condition.

If these conditions are met, you can state: *The sampling distribution of $\hat{p}$ is approximately Normal with mean $p$ and standard deviation $\sqrt{p(1-p)/n}$.*

Calculating Probabilities: Putting the Model to Work

With an approximately Normal model, you can calculate probabilities about your sample statistic using z-scores. This answers questions like, "How likely is it to get a sample proportion this extreme if the population proportion is $p$?"

Example Scenario: Suppose 60% ($p=0.6$) of all voters support a candidate. You take a random sample of $n=150$ voters. What is the probability that your sample shows less than 55% support ($\hat{p} < 0.55$)?

Step 1: Verify Conditions.

• Random: We assume a random sample.
• Large Counts: $np = 150 * 0.6 = 90$ and $n(1-p) = 150 * 0.4 = 60$. Both are $\ge 10$. Conditions are met.

Step 2: Describe the Sampling Distribution. The distribution of $\hat{p}$ is approximately Normal with: Mean: $\mu_{\hat{p}} = p = 0.6$ Standard Error: $\sigma_{\hat{p}} = \sqrt{\frac{0.6(1-0.6)}{150}} = \sqrt{\frac{0.24}{150}} \approx 0.04$

Step 3: Calculate the z-score and probability. We want $P(\hat{p} < 0.55)$. $$z = \frac{\hat{p} - \mu_{\hat{p}}}{\sigma_{\hat{p}}} = \frac{0.55 - 0.6}{0.04} = -1.25$$ Using the standard Normal table or technology, $P(z < -1.25) \approx 0.1056$.

Interpretation: If the true population support is 60%, there is about a 10.6% chance of getting a sample proportion of 55% or lower in a random sample of 150 voters.

Connecting to Inference: The Foundation of Confidence and Evidence

This entire framework is the engine behind statistical inference.

• Confidence Intervals: When we build a confidence interval for a population proportion $p$, we are essentially "reversing the direction" of the sampling distribution. We use our single $\hat{p}$ as the center and use the standard error (estimated from $\hat{p}$ itself) to create an interval that we believe captures the true $p$.
• Significance Tests: When we conduct a test about $p$, we assume a null hypothesis value ($p_0$) is true. We then use that $p_0$ to construct the sampling distribution (just like in the example above). We calculate the probability of obtaining our observed $\hat{p}$ or something more extreme from that distribution. This probability is the P-value. A small P-value indicates our sample result would be very unlikely if the null hypothesis were true, providing evidence against it.

Common Pitfalls

1. Confusing the Distribution of the Sample with the Sampling Distribution: The distribution of the sample is the set of data you collected (e.g., a list of "support" and "don't support" responses). The sampling distribution is a theoretical distribution of a statistic ($\hat{p}$) across many samples. They are fundamentally different objects.
2. Misapplying the Standard Error Formula: The formula $\sqrt{p(1-p)/n}$ uses the *population proportion $p$*, not the sample proportion $\hat{p}$. When $p$ is unknown (which is almost always), you use $\hat{p}$ to estimate the standard error for a confidence interval, but for the theoretical model in a hypothesis test, you use the $p_0$ from the null hypothesis.
3. Neglecting the Conditions: Using the Normal model without verifying randomness and the Large Counts Condition is a critical error. If $np$ or $n(1-p)$ is less than 10, the sampling distribution may be skewed, and Normal-based calculations will be inaccurate.
4. Forgetting the Square Root in the Standard Error: A frequent computational mistake is writing the standard error as $\frac{p(1-p)}{n}$ instead of $\sqrt{\frac{p(1-p)}{n}}$. Always remember the square root.

Summary

• The sampling distribution of the sample proportion $\hat{p}$ describes how this statistic varies across all possible random samples of size $n$ from a population with proportion $p$.
• Its mean is $p$, and its standard deviation (standard error) is $\sigma_{\hat{p}} = \sqrt{p(1-p)/n}$.
• Provided the data are from a random sample and meet the Large Counts Condition ($np \ge 10$ and $n(1-p) \ge 10$), this sampling distribution is approximately Normal.
• This Normal model allows you to calculate probabilities about $\hat{p}$, which forms the direct basis for confidence intervals and significance tests for a population proportion.