Hypothesis Tests for Differences in Population Proportions

• Two-Sample Z-Test: This inferential procedure evaluates whether the difference between two observed sample proportions ($\hat{p}_1 - \hat{p}_2$) provides enough evidence to reject the claim that the underlying population proportions ($p_1$ and $p_2$) are equal.

• Null Hypothesis ($H_0$): The baseline assumption is that there is no difference between the populations, expressed as $H_0: p_1 = p_2$ or $H_0: p_1 - p_2 = 0$.

• Alternative Hypothesis ($H_a$): This represents the research claim, which can be one-tailed ($p_1 > p_2$ or $p_1 < p_2$) or two-tailed ($p_1 \neq p_2$).

• Sampling Distribution: When samples are independent and large enough, the distribution of $\hat{p}_1 - \hat{p}_2$ is approximately normal with a mean of $p_1 - p_2$ and a standard deviation derived from the population parameters.

• The Pooling Logic: Because the null hypothesis assumes $p_1 = p_2$, we combine the data from both samples to create a single, more precise estimate of the common population proportion, known as the pooled proportion ($\hat{p}_c$).

• Standard Error (Pooled): The standard error for the test uses this pooled estimate to reflect the variability of the difference under the assumption that both samples come from populations with the same proportion.

Step 1: Calculate the Pooled Proportion

• Combine the successes ($X$) and total trials ($n$) from both groups:

$\hat{p}_c = \frac{X_1 + X_2}{n_1 + n_2}$

Step 2: Verify Conditions

• Independence: Samples must be independent of each other, and individual observations within samples must be independent (usually checked via random sampling and the 10% rule).
• Large Counts: Both samples must have at least 10 expected successes and 10 expected failures using the pooled proportion: $n_1\hat{p}_c \ge 10$, $n_1(1-\hat{p}_c) \ge 10$, $n_2\hat{p}_c \ge 10$, and $n_2(1-\hat{p}_c) \ge 10$.

Step 3: Compute the Z-Statistic

• The test statistic measures how many standard errors the observed difference is from the hypothesized difference (zero):

$z = \frac{(\hat{p}_1 - \hat{p}_2) - 0}{\sqrt{\hat{p}_c(1-\hat{p}_c)(\frac{1}{n_1} + \frac{1}{n_2})}}$

• Pooling vs. Non-Pooling: Pooling is mandatory for hypothesis tests of proportions because the null hypothesis explicitly states the proportions are identical. In contrast, confidence intervals do not assume equality, so they must use the separate sample estimates.

• Define Parameters: Always start by explicitly defining $p_1$ and $p_2$ in the context of the problem (e.g., 'where $p_1$ is the true proportion of adults who...').

• Check the Tail: For a two-tailed test ($H_a: p_1 \neq p_2$), remember to double the area found in one tail of the normal distribution to get the final p-value.

• Pooled Proportion Check: Ensure you use the pooled proportion $\hat{p}_c$ for both the standard error calculation and the Large Counts condition check.

• Conclusion Phrasing: If $p < \alpha$, 'reject $H_0$' and state there is 'sufficient evidence' for $H_a$. If $p > \alpha$, 'fail to reject $H_0$' and state there is 'insufficient evidence' for $H_a$.

• Mixing up X and n: A common error is confusing the number of successes ($X$) with the sample size ($n$) when calculating $\hat{p}_c$.

• Incorrect Standard Error: Students often use the standard error formula for confidence intervals (using $\hat{p}_1$ and $\hat{p}_2$ separately) instead of the pooled formula required for the test.

• Independence Violation: Forgetting to check the 10% rule when sampling without replacement can lead to an overestimation of precision.