4m ago

27 Views

0 Downloads

499.10 KB

17 Pages

Transcription

Solutions to Homework 5Statistics 302 Professor LargetTextbook Exercises4.79 Divorce Opinions and Gender In Data 4.4 on page 227, we introduce the results of aMay 2010 Gallup poll of 1029 US adults. When asked if they view divorce as “morally acceptable”,71% of the men and 67% of the women in the sample responded yes. In the test for a difference inproportions, a randomization distribution gives a p-value of 0.165. Does this indicate a significantdifference between men and women in how they view divorce?SolutionIf we use a 5% significance level, the p-value of 0.165 is not less than α 0.05 so we would notreject H0 : pf pm . This means the data do not show significant evidence of a difference in theproportions of men and women that view divorce as “morally acceptable”.4.82 Sleep or Caffeine for Memory? The consumption of caffeine to benefit alternatenessis a common activity practiced by 90% of adults in North America. Often caffeine is used in orderto replace the need for sleep. One recent study compares students’ ability to recall memorizedinformation after either the consumption of caffeine or a brief sleep. A random sample of 35 adults(between the ages of 18 and 39) were randomly divided into three groups and verbally given a listof 24 words to memorize. During a break, one of the groups takes a nap for an hour and a half,another group is kept awake and then given a caffeine pill an hour prior to testing, and the thirdgroup is given a placebo. The response variable of interest is the number of words participants areable to recall following the break. The summary statistics for the three groups are shown belowin the table. We are interested in testing whether there is evidence of difference in average recallability between any two of the treatments. Thus we have three possible tests between differentpairs of groups: Sleep vs Caffeine, Sleep vs Placebo, and Caffeine vs Placebo.GroupSleepCaffeinePlaceboSample Size121211Mean15.2512.2513.70Standard Deviation3.33.53.0(a) In the test comparing the sleep group to the caffeine group, the p-value is 0.003. What isthe conclusion of the test? In the sample, which group had better recall ability? According tothe rest results, do you think sleep is really better than caffeine for recall ability?(b) In the test comparing the sleep group to the placebo group, the p-value is 0.06. What is theconclusion of the test using a 5% significance level? Using a 10% significance level? How strongis the evidence of a difference in mean recall ability between these two treatments?(c) In the test comparing the caffeine group to the placebo group, the p-value is 0.22. What isthe conclusion of the test? In the sample, which group had better recall ability? According tothe test results, would we be justified in concluding that caffeine impairs recall ability?(d) According to this study, what should you do before an exam that asks you to recallinformation?Solution(a) The p-value (0.003) is small so the decision is to reject H0 and conclude that the mean recallfor sleep (x̄s 15.25) is different from the mean recall for caffeine (x̄c 12.25). Since the meanfor the sleep group is higher than the mean for the caffeine group, we have sufficient evidence to1

conclude that mean recall after sleep is in fact better than after caffeine. Yes, sleep is really betterfor you than caffeine for enhancing recall ability.(b) The p-value (0.06) is not less than 0.05 so we would not reject H0 at a 5% level, but it isless than 0.10 so we would reject H0 at a 10% level. There is some moderate evidence of a difference in mean recall ability between sleep and a placebo, but not very strong evidence.(c) The p-value (0.22) is larger than any common significance level, so do not reject H0 . Theplacebo group had a better mean recall in this sample (x̄p 13.70 compared to x̄c 12.25), butthere is not enough evidence to conclude that the mean for the population would be different for aplacebo than the mean recall for caffeine.(d) Get a good night’s sleep!4.86 Radiation from Cell Phones and Brain Activity Does heavy cell phone use affect brainactivity? There is some concern about possible negative effects of radiofrequency signals deliveredto the brain. In a randomized matched-pairs study, 47 healthy participants had cell phones placedon the left and right ears. Brain glucose metabolism (a measure of brain activity) was measuredfor all participants under two conditions: with one cell phone turned on for 50 minutes (the “on”condition) and with both cell phones off (the “off” condition). The amplitude of radio frequencywaves emitted by the cell phones during the “on” condition was also measured.(a) Is this an experiment or an observational study? Explain what it means to say that thiswas a “matched-pairs” study.(b) How was randomization likely used in the study? Why did participants have cell phones ontheir ears during the “off” condition?(c) The investigators were interested in seeing whether average brain glucose metabolism wasdifferent based on whether the cell phones were turned on or off. State the null and alternativehypotheses for this test.(d) The p-value for the test in part (c) is 0.004. State the conclusion of this test in context.(e) The investigators were also interested in seeing if brain glucose metabolism was significantlycorrelated with the amplitude of the radio frequency waves. What graph might we use tovisualize this relationship?(f) State the null and alternative hypotheses for the test in part (e).(g) The article states that the p-value for the test in part (e) satisfies p 0.001. State theconclusion of this test in context.Solution(a) This is an experiment since the explanatory factor (cell phone “on” or “off”) was controlled.The design is matched pairs, since all 47 participants were tested under both conditions. For eachparticipant, we find the difference in brain activity between the two conditions.(b) Randomization in this case means that the order of the conditions (“on” and “off”) was randomized for all the participants. Cell phones were on the ears for both conditions to control forany lurking variables and to make the treatments as similar as possible except for the variable ofinterest (the radiofrequency waves).(c) Using µon to represent average brain glucose metabolism when the cell phones are on andµof f to represent average brain glucose metabolism when the cell phones are off, the hypotheses2

are:H0 : µon µof fHa : µon 6 µof fNotice that since this is a matched pairs study, we could also write the hypotheses in terms of theaverage difference µD between the two conditions, with H0 : µD 0 vs Ha : µ 6 0.(d) Since the p-value is quite small (less than a significance level of 0.01), we reject the nullhypothesis. There is significant evidence that brain activity is affected by cell phones.(e) Both of these variables (brain glucose metabolism and amplitude of radiofrequency) are quantitative, so we use a scatterplot to graph the relationship.(f) We are testing to see if the correlation ρ between these two variables is significantly different from zero, so the hypotheses areH0 : ρ 0Ha : ρ 6 0where ρ is the correlation between brain glucose metabolism and amplitude of radiofrequency.(g) This p-value is very small so we reject H0 . There is strong evidence that brain activity iscorrelated with the amplitude of the radiofrequency waves emitted by the cell phone.For 4.94 and 4.96, indicate whether it makes more sense to use a relatively large significancelevel (such as α 0.10) or a relatively small significance level (such as α 0.01).4.94 Using your statistics class as a sample to see if there is evidence of a difference betweenmale and female students in how many hours are spent watching television per week.SolutionA Type I error (saying there’s a difference in TV habits by gender for the class, when actually thereisn’t) is not very serious, so a large significance level such as α 0.10 will make it easier to see anydifference.4.96 Testing to see if a well-known company is lying in its advertising. If there is evidence thatthe company is lying, the Federal Trade Commission will file a lawsuit against them.SolutionA Type I error (suing the company when they are not lying) is quite serious so it makes sense touse a small significance level such as α 0.01.For 4.100 and 4.102, describe what it means in that context to make a Type I and Type II error. Personally, which do you feel is a worse error to make in the given situation?4.100 The situation described in Exercise 4.94.SolutionType I error: Conclude there’s a difference in TV habits by gender for the class, when actually3

there is no difference. Type II error: Find no significant difference in TV habits by gender, whenactually there is a difference. Personal opinions will vary on which is worse.4.102 The situation described in Exercise 4.96.SolutionType I error: Sue the company when they are not lying. Type II error: Let the company off thehook, when they are actually lying in their advertising. Personal opinions will vary on which isworse.4.123 Paul the Octopus In the 2010 World Cup, Paul the Octopus (in a German aquarium)became famous for being correct in all eight of the predictions it made, including predicting Spainover Germany in a semifinal match. Before each game, two containers of food (mussels) were lowered into the octopus’s tank. The containers were identical, expect for country flags of the opposingteams, one on each container. Whichever container Paul opened was deemed his predicted winner.Does Paul have psychic powers? In other words, is an 8-for-8 record significantly better than justguessing?(a) State the null and alternative hypotheses.(b) Simulate one point in the randomization distribution by flipping a coin eight times andcounting the number of heads. Do this five times. Did you get any results as extreme as Paulthe Octopus?(c) Why is flipping a coin consistent with assuming the null hypothesis is true?Solution(a) The hypotheses are H0 : p 0.5 vs Ha : p 0.5, where p is the proportion of all games Paulthe Octopus picks correctly.(b) Answers vary, but 8 out of 8 heads should rarely occur.(c) The proportion of heads in flipping a coin is p 0.5 which matches the null hypothesis.4.124 How Unlikely Is Paul the Octopus’s Success? For the Paul the Octopus data inExercise 4.123, use StatKey or other technology to create a randomization distribution. Calculatea p-value. How unlikely is his success rate if Paul the Octopus is really not psychic?SolutionWe use technology to simulate many samples of size 8 from a population that has an equal numberof “successes” and “failures”, i.e. one where p 0.5. For each sample we count the number ofsuccesses out of the 8 trials to obtain a randomization distribution such as the one shown below(or find the proportion of successes in each sample). We then count the number of samples forwhich all 8 trials are successes, and divide by the total number of samples to get a p-value. Forthe distribution below, only 4 of the 1000 samples gave 8 correct guesses in 8 trials, so we estimatethe p-value 0.004. Answers will vary for other randomizations but the p-value will always besmall, indicating that it is very unlikely to predict all eight games correctly when just guessing atrandom.4

4.126 Finger Tapping and Caffeine In Data 4.6 we look at finger-tapping rates to see if ingesting caffeine increases average tap rate. The sample data for the 20 subjects (10 randomly gettingcaffeine and 10 with no-caffeine) are given in Table 4.4 on page 241. To create a randomizationdistribution for this test, we assume the null hypothesis µc µn is true, that is, there is no difference in average tap rate between the caffeine and no-caffeine groups.(a) Create one randomization sample by randomly separating the 20 data values into two groups.(One way to do this is to write the 20 tap rate values on index cards, shuffle, and deal theminto two groups of 10.)(b) Find the sample mean of each group and calculate the difference x̄c x̄n , in the simulatedsample means.(c) The difference in sample means found in part (b) is one data point in a randomizationdistribution. Make a rough sketch of the randomization distribution shown in Figure 4.11 onpage 242 and locate your randomization statistic on the sketch.Solution(a) Answers vary. For example, one possible randomization sample is shown below.caffeineno 247245248242246248mean 246.8mean 246.3(b) Answers vary. For the randomization sample above, x̄c x̄nc 246.8 246.3 0.5.(c) The sample difference of 0.5 for the randomization above would fall a bit to the right of thecenter of the randomization distribution.4.130 Effect of Sleep and Caffeine on Memory Exercise 4.82 on page 261 describes a studyin which a sample of 24 adult are randomly divided equally into two groups and given a list of 24words to memorize. During a break, one group takes a 90-minute nap while another group is givena caffeine pill. The response variable of interest is the number of words participants are able torecall following the break. We are testing to see if there is a difference in the average number ofwords a person can recall depending on whether the person slept or ingested caffeine. The data areshown in the table below and are available in SleepCaffeine.5

14151510Mean 15.25Mean 12.25(a) Define any relevant parameter(s) and state the null and alternative hypotheses.(b) What assumption do we make in creating the randomization distribution?(c) What statistic will we record for each of the simulated samples to create the randomizationdistribution? What is the value of the statistic for the observed sample?(d) Where will the randomization distribution be centered?(e) Find one point on the randomization distribution by randomly dividing the 24 data valuesinto two groups. Describe how you divide the data into two groups and show the values ineach group for the simulated sample. Compute the sample mean in each group and computethe difference in the sample means for this simulated result.(f) Use StatKey or other technology to create a randomization distribution. Estimate thep-value for the observed difference in means given in part (c).(g) At a significance level of α 0.01, what is the conclusion of the test? Interpret the resultsin context.Solution(a) The hypotheses are H0 : µs µc vs Ha : µs 6 µc , where µs and µc are the mean number ofwords recalled after sleep and caffeine, respectively.(b) The number of words recalled would be the same regardless of whether the subject was put inthe sleep or the caffeine group.(c) The sample statistic is x̄s x̄c . For the original sample the value is 15.25 12.25 3.0.(d) Under H0 we have µs µc 0 so the randomization distribution should be centered at zero.(e) We randomly divide the 24 sample word recall values into two groups of 12 (one for “sleep”group, the other for “caffeine”) and find the difference in sample means. One such randomizationis shown below where xs xc 14.75 12.75 2.0. Answers 9131421121516mean 14.75mean 12.75(f) A randomization distribution for 1000 differences in means is shown below. Since this is atwo-tailed test so we double the count in one tail (25 out of 1000 values at or beyond xs xc 3.0)in order to account for both tails. We see that the p-value is 2 0.025 0.05.6

(g) The p-value is more than α 0.01 so we do not reject H0 . There is not sufficient evidence (ata 1% level) to show a difference in mean number of words recalled after taking a nap or ingestingcaffeine.4.132 Does Massage Help Heal Muscles Strained by Exercise? After exercise, massageis often used to relieve pain, and a recent study shows that it also may relieve inflammation andhelp muscles heal. In the study, 11 male participants who had just strenuously exercised had 10minutes of massage on one quadricep and no treatment on the other, with treatment randomlyassigned. After 2.5 hours, muscle biopsies were taken and production of the inflammatory cytokineinterleukin-6 was measured relate to the resting level. The differences (control minus massage) aregiven in the table below.(a) Is this an experiment or an observational study? Why is it not double blind?(b) What is the sample mean difference in inflammation between no massage and massage?(c) We want to test to see if the population mean difference µD is greater than zero, meaningmuscle with no treatment has more inflammation than muscle that has been massaged. Statethe null and alternative hypotheses.(d) Use StatKey or other technology to find the p-value from a randomization distribution.(e) Are the results significant at a 5% level? At a 1% level? State the conclusion of the test ifwe assume a 5% significance level (as the authors of the study a) This is an experiment since a treatment was actively manipulated. In fact, it is a matched pairsexperiment. The experiment cannot be blind to the subject since s/he will know which muscle isbeing massaged. However the person measuring the level of inflammation should not know whichis which.(b) The mean difference is x̄D 1.30.(c) We define µD to be the mean difference in inflammation in muscle between a muscle thathas not been massaged and a muscle that has (using control level minus massage level). The nullhypothesis is no difference from a massage and the alternative is that levels are lower in a muscle7

that has been massaged. The hypotheses are:H0 : µD 0Ha : µD 0(d) Using StatKey or other technology, we create a randomization distribution of mean differencesunder the null hypothesis that µD 0. One such distribution is shown below. Since the alternativehypothesis is Ha : µD 0, this is a right-tail test. Using our sample mean x̄D 1.30, we see that24 of the 1000 simulated samples were as extreme so the p-value is 0.024.(e) Based on the randomization distribution and p-value in (d), the results are significant at a 5%level but not at a 1% level. Using a 5% level, we reject H0 and find evidence that massage doesreduce inflammation in muscles after exercise.4.141 Quiz vs Lecture Pulse Rate Do you think that students undergo physiological changeswhen in potentially stressful situations such as taking a quiz or exam? A sample of statistics students were interrupted in the middle of quiz and asked to record their pulse rates (beats for 1-minuteperiod). Ten of the students had also measured their pulse rate while siting in class listening to alecture, and these values were matched with their quiz pulse rates. The data appear in the tablebelow and are stored in QuizPulse10. Note that this is paired data since we have two values, aquiz and a lecture pulse rate, for each student in the sample. The question of interest is whetherquiz pulse rates tend to be higher, on average, then lecture pulse rates. (Hint: Since this is paireddata, we work with the differences in pulse rate for each student between quiz and lecture. If thedifference are D quiz pulse rate-lecture pulse rate, the question of interest is whether muD isgreater than 0.)(a) Define the parameter(s) of interest and state the null and alternative hypotheses.(b) Determine an appropriate statistic to measure and compute its value for the original sample.(c) Describe a method to generate randomization samples that is consistent with the nullhypothesis and reflects the paired nature of the data There are several viable methods. Youmight use shuffled index cards, a coin, or some other randomization procedure.(d) Carry out your procedure to generate one randomization sample and compute the statisticyou chose in part (b) for this sample.(e) Is the statistic for your randomization sample more extreme (in the direction of thealternative) than the original sample?8

6618717597061106678Solution(a) We are interested in whether pulse rates are higher on average than lecture pulse rates, so ourhypotheses areH0 : µQ µLHa : µQ µLwhere µQ represents the mean pulse rate of students taking a quiz in a statistics class and µLrepresents the mean pulse rate of students sitting in a lecture in a statistics class. We could alsoword hypotheses in terms of the mean difference D Lecturepulse Quiz pulse in which case thehypotheses would be H0 : µD 0 vs Ha : µD 0.(b) We are interested in the difference between the two pulse rates, so an appropriate statisticis x̄D , the differences (D Lecture Quiz) for the sample. For the original sample the differencesare: 2, 1, 5, 8, 1, 20, 15, 4, 9, 12and the mean of the differences is x̄D 2.7.(c) Since the data were collected as pairs, our method of randomization needs to keep the data inpairs, so the first person is labeled with (75, 73), with a difference in pulse rate of 2. As long as wekeep the data in pairs, there are many ways to conduct the randomization. In every case, of course,we need to make sure the null hypothesis (no difference) is met and we need to keep the data in pairs.One way to do this is to sample from the pairs with replacement, but randomly determine theorder of the pair (perhaps by flipping a coin), so that the first pair might be (75,73) with a difference of 2 or might be (73,75) with a difference of -2. Notice that we match the null hypothesisthat the quiz/lecture situation has no effect by assuming that it doesn’t matter - the two valuescould have come from either situation. Proceeding this way, we collect 10 differences with randomlyassigned signs (positive/negative) and compute the average of these differences. That gives us onesimulated statistic.A second possible method is to focus exclusively on the differences as a single sample. Sincethe randomization distribution needs to assume the null hypothesis, that the mean difference is 0,we can subtract 2.7 (the mean of the original sample of differences) from each of the 10 differences,giving the values 0.7, 3.7, 2.3, 10.7, 1.7, 17.3, 12.3, 6.7, 6.3, 14.7.Notice that these values have a mean of zero, as required by the null hypothesis. We then selectsamples of size ten (with replacement) from the adjusted set of differences (perhaps by putting the10 values on cards or using technology) and compute the average difference for each sample.There are other possible methods, but be sure to use the paired data values and be sure to forcethe null hypothesis to be true in the method you create!9

(d) Here is one sample if we randomly assign /- signs to each difference: 2, 1, 5, 8, 1, 20, 15, 4, 9, 12 x̄D 1.9Here is one sample drawn with replacement after shifting the differences. 6.7, 1.7, 6.3, 2.3, 3.7, 1.7, 0.7, 17.3, 2.3, 0.7 x̄D 1.3(e) Neither of the statistics for the randomization samples in (d) exceed the value of x̄D 2.7 fromthe original sample, but your answers will vary for other randomizations.4.144 Exercise Hours Introductory statistics students fill out a survey on the first day of class.One of the questions asked is “How many hours of exercise do you typically get each week?” Responses for a sample of 50 students are introduced in Example 3.25 on page 207 and stored inthe file Exercise Hours. The summary statistics are shown in the computer output. The meanhours of exercise for the combined sample of 50 students is 10.6 hours per week and the standarddeviation is 8.04. We are interested in whether these sample data provide evidence that the meannumber of hours of exercise per week is difference between male and female statistics uss whether or not the methods described below would be appropriate ways to generate randomization samples that are consistent with H0 : µf µm vs HA : µf 6 µm . Explain yourreasoning in each case.(a) Randomly label 30 of the actual exercise values with “F” for the female group and theremaining 20 exercise value with “M” for the males. Compute the difference in the samplemeans x̄f x̄m .(b) Add 1.2 to every female exercise value to give a new mean of 10.6 and subtract 1.8 fromeach male exercise value to move their mean to 10.6 (and match the females). Sample 30 values(with replacement) from the shifted female values and 20 values (with replacement) from theshifted male values. Compute the difference in the sample means, x̄f x̄m .(c) Combine all 50 sample values into one set of data having a mean amount of 10.6 hours.Select 30 values (with replacement) to represent a sample of female exercise hours and 20 values(also with replacement) for a sample of male exercise values. Compute the difference in thesample means, x̄f x̄m .Solution(a) This method is appropriate. Under H0 the exercise amounts should be unrelated to the genders,so each exercise value could have just as likely come from either gender.(b) This method is appropriate. Adjusting the values so that the male and female exercise meansare the same produces a “population” to sample from that agrees with the null hypothesis of equalmeans. Note that adding 3.0 to all of the female exercise values or subtracting 3.0 from all the maleexercise values would also accomplish this goal. Adjusting both, as suggested in part (b) has theadvantage of keeping the combined mean at 10.6. Any of these methods give equivalent results.10

(c) This method is appropriate. As in part(b) the randomizations samples are taken from a “population” where the mean exercise levels are the same for males and females, so the distribution ofx̄F x̄M should be centered around zero as specified by H0 : µF µM .4.158 Testing for a Home Field Advantage in Soccer In Exercise 3.108 on page 215, wesee that the home team was victorious in 70 games out of a sample of 120 games in the FA premierleague, a football (soccer) league in Great Britain. We wish to investigate the proportion p of allgames wine by the home team in this league.(a) Use StatKey or other technology to find and interpret a 90% confidence interval for theproportion of games won by the home team.(b) State the null and alternative hypotheses for a test to see if there is evidence that theproportion is different from 0.5.(c) Use the confidence interval from part (a) to make a conclusion in the test from part (b).State the confidence level used.(d) Use StatKey or other technology to create a randomization distribution and find the p-valuefor the test in part (b).(e) Clearly interpret the results of the test using the p-value and using a 10% significance level.Does your answer match your answer from part (c)?(f) What information does the confidence interval give that the p-value doesn’t? Whatinformation does the p-value give that the confidence interval doesn’t?(g) What’s the main different between the bootstrap distribution of part (a) and therandomization distribution of part (d)?Solution(a) The proportion of home wins in the sample is p̂ 70/120 0.583. Using StatKey or othertechnology, we construct a bootstrap distribution of sample proportions such as the one below. Wesee that a 90% confidence interval in this case goes from 0.508 to 0.658. We are 90% confident thatthe home team will win between 50.8% and 65.8% of soccer games in this league.(b) To test if the proportion of home wins differs from 0.5, we use H0 : p 0.5 vs Ha : p 6 0.5.(c) Since 0.5 is not in the interval in part (a), we reject H0 at the 10% level. The proportionof home team wins is not 0.5, at a 10% level.(d) We create a randomization distribution (shown below) of sample proportions when n 12011

using p 0.5. Since this is a two-tailed test, we have p-value 2(0.041) 0.082.(e) At a 10% significance level, we reject H0 and find that the proportion of home team wins isdifferent from 0.5. Yes, this does match what we found in part (c).(f) The confidence interval shows an estimate for the proportion of times the home team wins,which the p-value does not give us. The p-value gives a sense for how strong the evidence is thatthe proportion of home wins differs from 0.5 (only significant at a 10% level in this case).(g) The bootstrap and randomizations distributions are similar, except that the bootstrap proportions are centered at the original p̂ 0.583, while the randomization proportions are centeredat the null hypothesis, p 0.5.4.160 How Long Do Mammals Live? Data 2.2 on page 61 includes information on longevity(typical lifespan), in years, for 40 species of mammals.(a) Use the data, available in MammalLongevity, and StatKey or other technology to test tosee if the average lifespan of mammal species is difference from 10 years. Include all details ofthe test: the hypotheses, the p-value, and the conclusion in context.(b) Use the result of the test to determine whether µ 10 would be included as a plausiblevalue in a 95% confidence interval of average mammal lifespan. Explain.Solution(a) We are testing H0 : µ 10 vs Ha : µ 6 10, where µ represents the mean longevity, in years,of all mammal species. The mean longevity in the sample is x̄ 13.15 years. We use StatKeyor other technology to create a randomization distribution such as the one shown below. For thistwo-tailed test, the proportion of samples beyond x̄ 13.15 in this distribution gives a p-value of20.005 0.010. We have strong evidence that mean longevity of mammal species is different from10 years.(b) Since, for a 5% significance level, we reject µ 10 as a plausible value of the population meanin the hypothesis test in part (a), 10 would not be included in a 95% confidence interval for themean longevity of mammals.4.167 Does Massage Really Help Reduce Inflammation in Muscles? In Exercise 4.132 onpage 279, we learn that massage helps reduce levels of the inflammator