Hypothesis Testing: P-Value, Steps, and Interpretation

Administrative Information for Course Review

Course Review Deadline: The process to review or opt-out from the course must be completed by September 29. Students should follow the instructions provided in an email sent earlier today.
Cost: While the exact charge is uncertain, it is estimated to be around $20 or potentially more, but not excessively high. Students who do not need the review should opt out.</li></ul><h3 collapsed="false" seolevelmigrated="true">P-Value Determination and Interpretation</h3><ul><li>P-Value Location: The location of the p-value is always determined by the directional sign of the alternative hypothesis ($ H_A $).<ul><li>Example: If$ H_A $uses a "less than" ($ < $) sign, the p-value is the area to the left of the test statistic.</li><li>Concept: The p-value represents the probability of observing a test statistic as extreme as, or more extreme than, what was observed in the sample, assuming the null hypothesis ($ H_0 $) is true.</li></ul></li><li>Conceptualizing Hypothesis Tests (The 10,000 Lives Analogy):<ul><li>Imagine that every hypothesis test you perform has, in theory, been conducted 10,000 times by someone else before you.</li><li>Illustration with Coin Flip Applet:<ul><li>Null Hypothesis: Probability of getting heads (p) from a fair coin is$ P=0.5 $.</li><li>Sample Size:$ n=60 $people in a classroom.</li><li>Experiment: Each of the 60 people flips a coin once, and the number of heads (successes) is counted.</li><li>Simulation: The applet simulates taking samples of size 60 from a population where$ p=0.5 $and calculating the proportion of heads ($ p_{hat} $) 10,000 times.</li><li>Expected Outcome: If you repeat this experiment many times, the distribution of$ p_{hat} $values will form a bell-shaped (normal) curve centered around$ P=0.5 $.</li><li>Likelihood: The taller a specific$ p{hat} $value is on this curve, the more likely it is to occur if$ H0 $is true. Conversely, the shorter a value is, the less likely it is.</li><li>Classroom Example:<ul><li>Getting 30 out of 60 heads (i.e.,$ p_{hat} = 0.5 $) is very likely and would be at the peak of the curve.</li><li>Getting 12 out of 60 heads is very unlikely and would be a short point on the curve.</li><li>Getting 60 out of 60 heads would be extremely unlikely.</li></ul></li></ul></li></ul></li><li>Interpreting a Low P-Value: If your observed sample result (e.g., getting only 22 heads out of 60, resulting in a very low p-value like$ P-value = 0.0192 $) is very unlikely under the assumption that the null is true, then you reject the null hypothesis. It suggests that your sample result is too extreme to be explained by chance alone.</li><li>Crucial P-Value Interpretation Rule: When interpreting a p-value, it always done under the assumption that the null hypothesis ($ H0 $) is, in reality, true. This holds true regardless of whether you reject$ H0 $or fail to reject$ H_0 $.</li></ul><h3 collapsed="false" seolevelmigrated="true">Six Steps of a Hypothesis Test</h3><h4 collapsed="false" seolevelmigrated="true">Step 1: Draft a Research Question</h4><ul><li>Purpose: To clearly state the question the hypothesis test aims to answer.</li><li>Strategy: Paraphrase the relevant sentence from the problem statement, usually the one preceding the request to "perform a hypothesis test."</li><li>Example 1 (Bank Accounts in Arrears):<ul><li>Problem Statement Snippet: "…determine if less than 1% of all accounts held by the bank are in arrears."</li><li>Research Question: "Is less than 1% of all accounts held by the bank in arrears?"</li></ul></li><li>Example 2 (Diet Soda Taste):<ul><li>Problem Statement Snippet: "…determine if the majority of all diet soda drinkers now like the taste of their soda."</li><li>Research Question: "Do the majority (i.e., more than half) of all diet soda drinkers now like the taste of their soda?"</li></ul></li></ul><h4 collapsed="false" seolevelmigrated="true">Step 2: Formulate Hypotheses ($ H0 $and$ HA $)</h4><ul><li>Alternative Hypothesis ($ H_A $):<ul><li>Must always align with the directional sign implied by the research question (e.g., "less than," "more than," "different from").</li><li>Example 1 ($ p < 0.01 $): If the research question is "less than 1%," then$ H_A: p < 0.01 $.</li><li>Example 2 ($ p > 0.5 $): If the research question is "majority (more than half)," then$ H_A: p > 0.5 $.</li></ul></li><li>Null Hypothesis ($ H_0 $):<ul><li>The parameter symbol (e.g.,$ p $for proportion) must be the same as in$ H_A $.</li><li>The value of the parameter (e.g.,$ 0.01 $or$ 0.5 $) must be the same as in$ H_A $.</li><li>The directional symbol for$ H_0 $must always include an equal sign.<ul><li>Options for$ H_0 $:<ul><li>$ H_0: p = ext{value} $(most common and always acceptable)</li><li>$ H0: p ext{ (greater than or equal to) value} $(if$ HA $is less than)</li><li>$ H0: p ext{ (less than or equal to) value} $(if$ HA $is greater than)</li></ul></li></ul></li></ul></li><li>Example 1 (Bank Accounts):<ul><li>$ H_0: p = 0.01 $</li><li>$ H_A: p < 0.01 $</li></ul></li><li>Example 2 (Diet Soda):<ul><li>$ H_0: p = 0.5 $</li><li>$ H_A: p > 0.5 $</li></ul></li></ul><h4 collapsed="false" seolevelmigrated="true">Step 3: Check Conditions</h4><ul><li>Type of Test: Identify if the problem involves proportions or means. (Discussion focuses on proportions here, indicated by percentages or "majority").</li><li>Conditions for Proportions (Two Methods):<ol><li>Method 1 (Using Population Proportion under$ H_0 $):<ul><li>$ np ext{ (greater than or equal to) } 10 $</li><li>$ n(1-p) ext{ (greater than or equal to) } 10 $</li><li>Where$ p $is the value from the null hypothesis ($ P_0).

Method 2 (Using Sample Counts):

Number of successes $\ge 10$
Number of failures $\ge 10$
Note: Both methods yield the same pass/fail result. You can use either one.

Example (Diet Soda):

Sample Size: n=75 $</li><li>Null Population Proportion:$ P_0 = 0.5 $</li><li>Successes: 39 people liked the taste.</li><li>Failures: 36 people disliked the taste ($ 75 - 39 = 36 $).</li><li>Checking Conditions:<ul><li>Method 1:<ul><li>$ n P_0 = 75 imes 0.5 = 37.5 ext{ (greater than or equal to) } 10 $(Passes)</li><li>$ n (1-P_0) = 75 imes (1-0.5) = 75 imes 0.5 = 37.5 ext{ (greater than or equal to) } 10 $(Passes)</li></ul></li><li>Method 2:<ul><li>Successes:$ 39 ext{ (greater than or equal to) } 10 $(Passes)</li><li>Failures:$ 36 ext{ (greater than or equal to) } 10 $(Passes)</li></ul></li></ul></li><li>Conclusion for Conditions: Conditions are met.</li></ul></li><li>Random Sample: The problem must state that a random sample was taken. (e.g., "random sample of 75 diet soda drinkers").</li></ul><h4 collapsed="false" seolevelmigrated="true">Step 4: Calculate Test Statistic (Z-score) and P-value</h4><ul><li>Key Consideration for$ p{hat} $: The sample proportion ($ p{hat} $) must refer to the same concept as the population proportion ($ p $) in the hypotheses. If$ HA $is about liking the taste,$ p{hat} $must be the proportion who liked the taste.</li><li>Example (Diet Soda):<ul><li>$ P0 = 0.5 $(from$ H0 $)</li><li>$ n = 75 $</li><li>Observed Successes: 39 people liked the taste.</li><li>Calculate$ p{hat} $:$ p{hat} = 39/75 $</li><li>Test Statistic (Z-score): (Details for calculation not provided in transcript, but it was stated to be$ Z = 0.35 $).</li></ul></li><li>Finding the P-value from Z-Table:<ul><li>The Z-table always provides the area to the left of the Z-score.</li><li>Example: For$ Z = 0.35 $:<ul><li>Look up$ Z=0.35 $in the Z-table: yields$ 0.6368 $</li><li>This value ($ 0.6368 $) is the area to the left of$ Z=0.35 $.</li></ul></li><li>Adjusting for$ H_A $Direction:<ul><li>Recall that the p-value location is dictated by$ HA $. In the diet soda example,$ HA: p > 0.5 $(greater than sign).</li><li>Therefore, the p-value is the area to the right of the test statistic.</li><li>P-value Calculation:$ 1 - ( ext{Area to the left}) = 1 - 0.6368 = 0.3632 $.</li><li>Result: P-value$ = 0.3632 $.</li></ul></li><li>Interpretation of P-value ($ 0.3632 $): This p-value is relatively large, indicating that observing a sample proportion like$ 39/75 $(or more extreme) is quite likely if the null hypothesis ($ p=0.5 $) were true. It corresponds to a relatively "tall" point on the normal curve.</li></ul></li></ul><h4 collapsed="false" seolevelmigrated="true">Step 5: Decision</h4><ul><li>Decision Rule: Compare the p-value to the significance level (alpha,$ \alpha).
- If P-value $\le \alpha$, then Reject H_0.
- If P-value $> \alpha$, then Fail to Reject H_0 $.</li></ul></li><li>Alpha Level: Alpha ($ \alpha $) is always provided in the problem statement (e.g., "perform a hypothesis test at alpha$ 0.05 $").<ul><li>Example (Diet Soda):$ \alpha = 0.05 $</li></ul></li><li>Applying the Rule:<ul><li>P-value$ = 0.3632 $</li><li>$ \alpha = 0.05 $</li><li>Since$ 0.3632 > 0.05 $(P-value is greater than alpha), the decision is to Fail to Reject$ H_0 $.</li></ul></li></ul><h4 collapsed="false" seolevelmigrated="true">Step 6: Conclusion (Not fully detailed in transcript but implied after decision)</h4><ul><li>State the decision in terms of the research question and the context of the problem. If you fail to reject$ H0 $, it means there is not enough evidence to support$ HA $. If you reject$ H0 $, it means there is sufficient evidence to support$ HA $.</li></ul>Example (Concluding for Diet Soda Problem): (Based on FTR$ H_0 $)<ul><li>There is not sufficient evidence at the$ \alpha = 0.05$$ significance level to conclude that the majority of all diet soda drinkers now like the taste of their soda.