Lecture_4Testing Hypotheses about Proportions 生物统计课件.pptx

IV. Testing Hypotheses about ProportionsNow we study categorical, rather than continuous outcomes.Let X be a dichotomous categorical variable - two outcomes (e.g., success vs. failure, improve vs. non-improve, male vs. female).Parameter p = proportion of the population with one of the two outcomes. (Note: 1-p = proportion with the other outcome). A. Inference for a Population Proportion (p.46)Point Estimate and Confidence Interval for a Population Proportion (p. 46-47)Data: x1, x2 ,…, xn where xi = 1 if subject has the outcome of interest, otherwise = 0.Estimate of p:Approximate 100% (1-?) confidence interval for p: Research Quesiton: “Does the antihypertensive drug, Minoxidil, given topically to balding men result in hair growth?”Dichotomous Variable: Xi = 1 if man had hair growth (success); Xi = 0, otherwise (failure).In sample of 95 men, 43 had new hair growth. What is the estimate (and 95% confidence interval) for p, the proportion of all balding men who will respond favorably to Minoxidil?Example 1: (p. 47-48)Estimate of p:95% confidence interval for p:95% confident that true proportion of bald men who would respond (ie, grow hair) to Minoxidil is in range of 0.353 to 0.553Example 1: (continued)X = dichotomous variable with parameter p. Compare p to some known value p0. Forms of hypotheses: H0: p = p0 H1: p p0 (one sided alternative) H1: p p0 (one sided alternative) H1: p ? p0 (two sided alternative)Test statistic which has an approximate normal distribution. P-value obtained from Table A.1.One-Sample Test for a Proportion (p. 48)Research Question: “Nationwide, President Obama has a 53% approval rating (p0 = 0.53). Is his approval rating different among Los Angeles citizens?”H0: p = 0.53 vs. H1: p ? 0.53Data: n = 200 residents (randomly selected), 131 approved, the remainder disapproved. = 131/200 = 0.655Example: (p. 49-50)Test statisticP-value: P 0.001These data indicate that President Obama’s approval rating (i.e, the proportion of voters that ap