Power And Confidence Intervals

1. Iowa GPA: H 0 : = 2.80

Ha :

2.80

In the above hypotheses,

H 0 : is the notation for the null hypothesis, H a : is the notation for the

alternative hypothesis, and is the parameter, the mean GPA of the population, about which were hypothesizing. 2. P-value: a) This P-value does not give strong evidence against the null hypothesis. b) This extreme P-value does give strong evidence against the null hypothesis. 3. Another test of therapeutic touch: p = proportion of trials guessed correctly H 0 : p = 0.50 and H a : p > 0.50

a) b)

= 53/130 = 0.4077 p
se =

p1 p / n .501 .50 / 130 0.0439

z = (0.4077 0.5)/0.0439 = - 2.10; the sample proportion is a bit more than 2 standard errors less than would be expected if the null hypothesis were true. c) The P-value is 0.9821 (rounds to 0.98). Because the P-value is so much greater than the significance level of 0.05, we do not reject the null hypothesis. The probability would be 0.98 of getting a test statistic at least as extreme as the value observed if the null hypothesis were true, and the population proportion were 0.50. d) np = (130)(0.5) = 65= n(1-p); the sample size was large enough to make the inference in (c). We also would need to assume randomly selected practitioners and subjects for this to apply to all practitioners and subjects. 4. Garlic to repel ticks: a) The relevant variable is whether garlic or placebo is more effective, and the parameter is the population proportion, p = those for whom garlic is more effective than placebo. b) H 0 : p = 0.50 and H a : p 0.50; the sample size is adequate because there are at least 15 c) successes (garlic more effective) and failures (placebo more effective). = 37/66 = 0.561 p se =

p1 p / n .501 .50 / 66 0.062

z = (0.561 0.5)/0.062 = 0.984 d) The P-value is 0.325 (rounds to 0.33). This P-value is not that extreme. If the null hypothesis were true, the probability would be 0.33 of getting a test statistic at least as extreme as the value observed. It is plausible that the null hypothesis is correct. Because the probability is 0.33 that we would observe our test statistic or one more extreme due to random variation, there is insufficient evidence that the population proportion which would have fewer tick bites with garlic vs. placebo is not different from 0.50.

5. Which cola?: a) The test statistic (Z-Value) is calculated by taking the difference between the sample proportion and the null proportion and dividing it by the standard error. b) We get the P-value by looking up the Z-value in Table A or using software. We have to determine the two-tail probability from the standard normal distribution below -1.286 and

above 1.286. The P-value of 0.20 tells us that if the null hypothesis were true, a proportion of 0.1985 (rounds to 0.20) of samples would fall at least this far from the null hypothesis proportion of 0.50. This is not very extreme; it is plausible that the null hypothesis is correct, and that Coke is not preferred to Pepsi. c) It does not make sense to accept the null hypothesis. It is possible that there is a real difference in the population that we are not detecting in our test, and we can never accept a null hypothesis. d) The 95% confidence interval tells us the range of plausible values, whereas the test merely tells us that 0.50 is plausible. 6. Lake pollution: a) From software, the mean of the four observations is 2000; the standard deviation is 816.5; and the standard error is 408.25. b) Software gives a t-score of 2.45. c) The P-value is 0.046 for a one-sided test. This is smaller than 0.05, so we have enough evidence to reject the null hypothesis at a significance level of 0.05. We have strong evidence that the wastewater limit is begin exceeded. d) The one-sided analysis in (b) implicitly tests the broader null hypothesis that 1000. We know this because if it would be unusual to get a sample mean of 2000 if the population mean were 1000, we know that it would be even more unusual to get this sample mean if the population mean were less than 1000.

7. Selling a burger: 1) Assumptions: The data are quantitative (dollar differences). The data might have been produced randomly. 2) Hypotheses: H 0 : = 0; H a : 0 3) Test statistic:

3000 0 4000 / 10

2.372

4) P-value: 0.04177 (rounds to 0.042) 5) Conclusion: If the null hypothesis were true, the probability would be 0.042 of getting a test statistic at least as extreme as the value observed. Because the P-value is 0.04, which is < 0.05, there is sufficient evidence that the coupons led to higher sales than did the outside posters.

8. Decision errors in medical diagnostic testing: a) A Type I error is a false positive because we have rejected the null hypothesis that there is no disease, but we were wrong. The woman in fact does not have breast cancer. The consequence would be that the woman would have treatment, or at least further testing, when she did not need any. b) A Type II error is a false negative because we have failed to reject the null hypothesis that there is no disease, but we were wrong. The woman in fact does have breast cancer. The consequence would be failing to detect cancer and treat the cancer when it actually exists. c) The disadvantage of this tactic is that more women who do have breast cancer will have false negative tests and not receive necessary treatment. 9. Misleading summaries?: a) Researcher A: se =

p1 p / n .501 .50 / 400 0.025

z = (0.55 0.50)/0.025 = 2.00 P-value =2P(z >2.0) =0.046. . b) Researcher B: z = (0.5475 0.50)/0.025 = 1.90

P-value =2P(z >1.90)= 0.057. c) Researcher As result has a P-value less than 0.05; thus, it is statistically significant. Researcher Bs P-value is not less than 0.05, and is not, therefore, statistically significant. Results that are not different from one another in practical terms might lead to different conclusions if based on statistical significance alone. d) If we do not see these two P-values, but merely know that one is statistically significant and one is not, we are not able to see that the P-values are so similar. e) We can add and subtract the result of (1.96)(0.025) the z-score associated with a 95% confidence interval multiplied by the standard error to the mean for each of these samples to get confidence intervals of (0.501, 0.599) for Researcher A and (0.499, 0.596) for Researcher B. This method shows the enormous amount of overlap between the two confidence intervals. The plausible values for the population proportions are very similar in the two cases, which we would not realize by merely reporting whether the null was rejected in a test.