Hypothesis Testing for Population Proportions: Comprehensive Guide

Fundamentals of Hypothesis Testing and Essential Ingredients

Hypothesis testing is a statistical method used to make decisions between two competing claims based on data.
General Four Steps of Hypothesis Testing: - Step 1: State Your Hypotheses: Define the Null Hypothesis ( $H_0$ ) and the Alternative Hypothesis ( $H_a$ ). - Step 2: Prepare and Collect Data: Determine what data to collect and the required sample size ( $n$ ). Larger sample sizes are preferred as they reduce uncertainty. - Step 3: Compute and Compare: Collect the actual data and compare the resulting sample proportion ( $\hat{p}$ ) with the expected proportion under the assumption that the null hypothesis is true. - Step 4: Interpret the Results: Decide if the observed results are too unusual to have occurred by random chance under the null hypothesis.
Parameters vs. Statistics: - Hypothesis testing is always conducted for the population parameter ( $p$ ), never the sample statistic ( $\hat{p}$ ). - We already know the sample statistic from our data; the goal is to infer if that statistic provides enough evidence to make a claim about the entire population.

Defining Null and Alternative Hypotheses

Null Hypothesis ( $H_0$ ): - Represents the "status quo," no change, no surprise, or no effect. - It is the claim that is initially assumed to be true. - It is often referred to as "H naught." - Rule: The mathematical expression for $H_0$ must always contain an equal sign ( $H_0: p = p_0$ ). - Example: In testing a coin for fairness, $H_0: p = 0.5$ .
Alternative Hypothesis ( $H_a$ ): - Represents the claim the researcher intends to demonstrate as true (the suspect change or effect). - It is often referred to as "H-A." - Rule: The mathematical expression for $H_a$ must always contain inequality signs: less than ( < ), greater than ( > ), or not equal to ( $\neq$ ). - Example: If a coin is suspected to be unfair after trials, $H_a: p \neq 0.5$ .

Directionality: One-Sided vs. Two-Sided Hypotheses

The comparison symbol used in the alternative hypothesis determines the "tailedness" of the test: - Left-Tailed Test: Used when looking for a decrease. H_a: p < p_0. - Right-Tailed Test: Used when looking for an increase. H_a: p > p_0. - Two-Tailed Test: Used when looking for any change or difference from the status quo. $H_a: p \neq p_0$ .
Retailer Example (Internet Sales): - Scenario: Historical online sales are $30\%$ . A retailer buys ads to increase this. - Increase Question: $H_0: p = 0.30$ ; H_a: p > 0.30 (One-sided, right test). - Decrease Question: $H_0: p = 0.30$ ; H_a: p < 0.30 (One-sided, left test). - Change Question: $H_0: p = 0.30$ ; $H_a: p \neq 0.30$ (Two-sided test).

The One-Proportion z-Test Statistic

The one-proportion z-Test Statistic is effectively the z-score for the sample proportion.
It measures how far (in standard deviations) the sample evidence ( $\hat{p}$ ) is from the null claim ( $p_0$ ).
Formula: - $z = \frac{\hat{p} - p_0}{SE}$ - Where $SE = \sqrt{\frac{p_0(1-p_0)}{n}}$ - $p_0$ represents the value claimed by the null hypothesis. - $\hat{p}$ represents the observed sample proportion.
Standard Error (SE) Calculation Note: When calculating the SE for hypothesis testing, use the population proportion from the null hypothesis ( $p_0$ ), not the sample proportion ( $\hat{p}$ ).

Characterizing p-values and Significance Levels

p-Value Big Picture: The p-value answers the question: "If the null hypothesis were true, how surprising is this result in terms of probability?" - Small p-value indicates results are very surprising under the null; provides strong evidence against $H_0$ . - Large p-value indicates results are not surprising; provides weak evidence against $H_0$ .
Calculation Definitions by Tail Type: - Left-Tailed p-value: Probability that the new sample proportion will be less than or equal to the observed sample proportion ( $P(Z \leq z_{observed})$ ). - Right-Tailed p-value: Probability that the new sample proportion will be greater than or equal to the observed sample proportion ( $P(Z \geq z_{observed})$ ). - Two-Tailed p-value: Probability that the new sample proportion will be farther away from $0$ than the observed statistic (2 \times P(Z > |z_{observed}|)).
Level of Significance ( $\alpha$ ): - This is the "cut-off" point for evidence chosen before running the test. - Typical values for $\alpha$ : $0.05$ ( $5\%$ ), $0.01$ ( $1\%$ ), or $0.10$ ( $10\%$ ). - The smaller the $\alpha$ , the stronger the evidence required to reject $H_0$ .
Decision Rules: - If \text{p-value} < \alpha: Reject $H_0$ . Evidence is strong enough to accept $H_a$ . - If $\text{p-value} \geq \alpha$ : Fail to Reject $H_0$ . Evidence is not strong enough to conclude $H_a$ is true.
Critical Terminology Caution: - Never say "Accept $H_0$ " or "Prove $H_0$ ". We only "Fail to reject $H_0$ ". - Analogy: In a court of law, a verdict of "Not Guilty" does not mean "Innocent"; it means there was not enough evidence to prove guilt.

Conditions for Using the z-Test (Central Limit Theorem)

To ensure the sampling distribution follows a Normal distribution ( $\hat{p} \sim N(p_0, SE)$ ), three conditions must be met: - 1. Random and Independent: The sample data must be collected randomly. - 2. Large Population: The total population should be at least $10$ times larger than the sample size ( $10n$ ). - 3. Large Sample Size: There must be at least $10$ expected successes and $10$ expected failures under the null hypothesis: - $np_0 \geq 10$ - $n(1 - p_0) \geq 10$

Detailed Case Studies and Examples

Example 1: Engineering Cracks

Scenario: Historically, $20\%$ of metal ingots had surface cracks. Engineers use a new method and test $400$ ingots, finding a $17\%$ crack rate.
Hypotheses: - $H_0: p = 0.20$ (No improvement) - H_a: p < 0.20 (Improvement)
Calculations: - $p_0 = 0.20$ , $n = 400$ , $\hat{p} = 0.17$ - $SE = \sqrt{\frac{0.20 \times 0.80}{400}} = 0.02$ - $z = \frac{0.17 - 0.20}{0.02} = -1.50$
p-value: P(Z < -1.50) = 0.067.
Conclusion: At $\alpha = 0.05$ , since 0.067 > 0.05, we Fail to Reject $H_0$ . There is not enough evidence to claim the new method reduced the crack rate.

Example 2: Ohio High School Graduation Rates

Scenario: Ohio's state rate is $80\%$ . A specific school has a rate of $85.3\%$ in a sample of $130$ students. Is it statistically higher at $\alpha = 0.05$ ?
Conditions Check: - np_0 = 130 \times 0.80 = 104 > 10 - n(1-p_0) = 130 \times 0.20 = 26 > 10
Calculations: - $p_0 = 0.80$ , $\hat{p} = 0.853$ - $SE = \sqrt{\frac{0.80 \times 0.20}{130}} = 0.035$ - P(\hat{p} > 0.853) = 1 - P(Z < 1.51) \approx 0.0655
Conclusion: Since 0.0655 > 0.05, Fail to Reject $H_0$ . The evidence is not significant.

Example 3: Major League Baseball (MLB) Pitchers

Scenario: $10\%$ of the general population is left-handed. A sample of $100$ MLB pitchers shows $29$ are left-handed. Is the proportion for pitchers higher?
Hypotheses: $H_0: p = 0.10$ vs. H_a: p > 0.10
Calculations: - $\hat{p} = 0.29$ , $p_0 = 0.10$ , $n = 100$ - $SE = \sqrt{\frac{0.10 \times 0.90}{100}} = 0.03$ - $z = \frac{0.29 - 0.10}{0.03} = 6.33$
p-value: Approximately $0$ .
Conclusion: Reject $H_0$ . There is strong statistical evidence that the percent of left-handed pitchers exceeds $10\%$ .

Example 4: Hispanic/Latino Origin Census

Scenario: National average is $16\%$ . A county sample of $437$ residents finds $44$ are of Hispanic origin. Does the county differ from the nation?
Hypotheses: - $H_0: p = 0.16$ - $H_a: p \neq 0.16$ (Two-tailed test)
Calculations: - $\hat{p} = \frac{44}{437} = 0.101$ - $SE \approx 0.018$ - $z = \frac{0.101 - 0.16}{0.018} \approx -3.28$
p-value: 2 \times P(Z < -3.28) = 0.001
Conclusion: Reject $H_0$ . The population proportion in this county differs significantly from the national proportion.