ANNOUNCEMENTS: Reminder from when we started Chapter 9 We will be covering chapters out of order for the rest of the quarter. Five situations we will cover for the rest of this quarter: Today Sections 12.1 to 12.3, except 12.2 Lesson 3, then (see website): Parameter name and description Population parameter Sample statistic • More of Chapter 12 (hypothesis tests for proportions) For Categorical Variables: • Finish Chapter 9 (sampling distributions for means) One population proportion (or probability) p pˆ • Chapter 11 (confidence intervals for means) Difference in two population proportions p1 – p2 pˆ1 − pˆ 2 • Start Chapter 13 (hypothesis tests for means) For Quantitative Variables: One population mean µ x • Section 15.3 (chi-square hypothesis test for goodness of fit) Population mean of paired differences µ • Sections 16.1 and 16.2 (“analysis of variance”)* (dependent samples, paired) d d Difference in two population means • Finish Chapters 12 and 13 and cover Chapter 17* µ − µ x − x (independent samples) 1 2 1 2 • Last day: Review for final exam* *No clicker, quiz or homework due this week. • No discussion on Monday (Nov 15); 2 more for credit after that. For each situation will we: √ Learn about the sampling distribution for the sample statistic √ Learn how to find a confidence interval for the true value of the parameter HOMEWORK: (Due Friday, Nov 19) • Test hypotheses about the true value of the parameter Chapter 12: #15b and 16b (counts as 1), 19, 85 (counts double) Review: Standardized Statistics Chapter 12: Significance testing = hypothesis testing. sample statistic - population parameter Saw this already in Chapter 6. z = s.d.(sample statistic) Today: Hypothesis tests for one proportion Example: Replacing s.d. with s.e. (standard error) in denominator: Gallup poll of adult Canadians on same-sex marriage Standardized statistic is still z for situations with proportions. n = 1003; 481 of the 1003 = 48% oppose same-sex marriage. For means, standardized statistic is t, not z (more Mon). News story said “For the first time in the 7 years Gallup has asked Summary of sampling distributions for the 5 parameters (p. 382): the question, less than half (48%) of Canadians oppose same-sex Parameter Statistic Standard Deviation Standard Error Standardized of the Statistic of the Statistic Statistic with s.e. marriages.” One p(1− p) pˆ(1− pˆ) proportion p pˆ n n z Difference Question: Is the media quote correct about the population? p1 (1 − p1 ) p2 (1− p2 ) pˆ (1− pˆ ) pˆ (1− pˆ ) Between + 1 1 + 2 2 p1− p2 pˆ − pˆ n n n n Proportions 1 2 1 2 1 2 z pˆ • The sample proportion based on n = 1003 was = .48. One Mean σ s µ x • How unlikely is that if the population proportion p is .50 or more, i.e. n n t Mean σ d sd if the true p is ≥ .50? Difference, µd d n t Paired Data n • In other words, is this convincing evidence that less than half of the Difference 2 2 2 2 σ σ s s Between µ µ x − x 1 + 2 1 + 2 1− 2 1 2 n n t Canadian population opposed same-sex marriage? I.e. that p < .50? Means 1 2 n1 n2 Five Steps to Hypothesis Testing, and how they apply in general (for our 5 situations), for one proportion, and for our example. EXAMPLE: Want to know if less than half of all Canadians oppose same-sex STEP 1: Determine the null and alternative hypotheses. marriage, as media claimed. That is the alternative hypothesis. General: Alternative hypothesis is usually what researchers hope H0: p = 0.5 (or could be written as p ≥ 0.5; data do not support to show. “Null is dull.” Null is often status quo, a specific value. media claim) For two proportions or two means, null is usually “no difference.” Ha: p < 0.5 (the media quote is correct, less than half of all Canadians oppose same sex marriages) For One Proportion: p = population proportion The null value is p0 = 0.5. This is a one-sided test. Null hypothesis is H0: p = p0 (a specified value) The “equal sign” always goes in the null hypothesis. Remember how hypothesis testing works: Alternative hypothesis is one of these, based on context: • Assume the null hypothesis is true about the population Ha: p ≠ p0 (a two-sided hypothesis) • Compute what we expect if that’s the case Ha: p > p0 (a one-sided hypothesis) • Compare it to what we observed in the sample Ha: p < p0 (a one-sided hypothesis) • For our 5 situations, we do this using standardized statistics STEP 2: Rationale: Compare observed to expected if null is true Verify data conditions. If met, summarize data into a test statistic. If p0 really is the true population proportion, where does the pˆ General: observed value of fall in the sampling distribution of For our five situations, test statistic will be z (for situations with possibilities? proportions) or t (for situations with means) = Sampling distribution of p-hat when null hypothesis is true sample statistic − null value (null) standard error For One Proportion: Density Data conditions: np0 and n(1 – p0) are both at least 10. For the test statistic the null value is p0 and null standard error is the usual s.d.( pˆ ) with null value p0 in place of p: p0-3se p0-2se p0-se p0 p0+se p0+2se p0+3se p0 pˆ − p0 z = p0 (1− p0 ) Test statistic is the z-score for the location of the observed pˆ in this n picture. If the null hypothesis is true, test statistic is standard normal. Example (Canadian poll): STEP 3: Data conditions are met: np0 and n(1 – p0) = (1003)(0.5) = 501.5 Assuming the null hypothesis is true, find the p-value. (.5)(.5) General: p-value = the probability of a test statistic as extreme as Null standard error is 1003 = .0158; test statistic is: the one observed or more so, in the direction of Ha, if the null Sample statistic - null value .48 − .50 hypothesis is true. = = −1.27 (null) standard error .0158 One proportion: Depends on the alternative hypothesis. See pictures on board and Sampling distribution of p-hat when n = 1003 and p = 0.5 Standard normal distribution on p. 517 Alternative hypothesis: p-value is: Ha: p > p0 (a one-sided hypothesis) Area above the test statistic z Ha: p < p0 (a one-sided hypothesis) Area below the test statistic z 0.48 0.50 -1.27 0 Ha: p ≠ p0 (a two-sided hypothesis) 2 × the area above |z| = area in possible values of p-hat Values of z tails beyond −z and z Example: STEP 4: Alternative hypothesis is one-sided Decide whether or not the result is statistically significant based Ha: p < 0.5 on the p-value. Two possible conclusions. p-value = Area below the test statistic z = –1.27 From Table A.1, p-value = .1020 If p-value ≤ level of significance (usually .05), these are equivalent ways to state the conclusion: p-value for Example 1 = area below -1.27 • Reject the null hypothesis. Normal, Mean=0, StDev=1 • Accept the alternative hypothesis • The result is statistically significant If p-value > level of significance (usually .05), these are equivalent ways to state the conclusion: • Do not reject the null hypothesis 0.102 • There is not enough evidence to support the alternative hypothesis. -1.27 0 • The result is not statistically significant. z Example: TECHNICAL NOTE p-value = .1020 > .05, so our conclusion can be stated as: • Do not reject the null hypothesis. Often in hypothesis testing the standardized score will be “off the • There is not enough evidence to support the alternative hypothesis. charts.” The bottom of Table A.1 has values for “In the extreme.” • The result is not statistically significant. Here are some of them for the lower tail: Step 5: Report the conclusion in the context of the situation. z area below z -3.09 .001 Example: -4.75 .000001 We could not reject the null hypothesis. There was not enough -6.00 .000000001 evidence to accept the alternative hypothesis. The proportion of adult Canadians who opposed same-sex marriage when poll was Example: Suppose the test statistic is z = –5.00 taken is not statistically significantly less than half. So the media For Ha: p < p0 we would say the p-value < .000001. quote was not justified. (Recall, quote was “For the first time in For Ha: p ≠ p0 we would say the p-value < .000002. the 7 years Gallup has asked the question, less than half (48%) of Canadians oppose same-sex marriages.”) EXAMPLE 2: According to the CDC website, 59% of 18 to 25 STEP 1: Determine the null and alternative hypotheses. year-olds in the United States drink alcohol. For One Proportion: p = population proportion QUESTION: Is the percent different for UC Davis students? Null hypothesis is H0: p = p0 (a specified value) (Note that before looking at data, we have no reason to ask if it’s Alternative hypothesis is one of these, based on context: in one direction or the other, just is it different from national Ha: p ≠ p0 (a two-sided hypothesis) proportion.) Ha: p > p0 (a one-sided hypothesis) Ha: p < p0 (a one-sided hypothesis) Define the population parameter: p = proportion of all UC Davis students who drink alcohol at least Example: once a month. Is UC Davis proportion who drink different from national proportion of 0.59? Data: Sample of n = 405 students in Introductory Statistics.