Question 1

A recent study collected data from 5,038 participants and found that the average body temperature was 97.7 degrees F \((SD = 0.72)\). Is this significantly different from our previous benchmark for normal body temperature, 98.6?

Question 2

Calculate the effect size for the previous question.

Question 3

At a conference, you attend a talk about the effectiveness of a DBT therapy intervention on anxiety. The presenter puts up the figure below. The presenter claims that these groups are significantly different from one another, because the experimental group shows significantly lower anxiety than the clinical threshold, but the control group does not. Do you agree with thee presenter? Explain your answer.

Question 4

Daylight in Eugene winter is a precious commodity. You believe that undergraduates who live on the top floor of their dorm or apartment building are getting more sun (and more of that sweet, sweet Vitamin D) than students living on the bottom floors. You measure the amount of vitamin D in undergraduates blood streams and record the averages by where students live.

\(\hat{\mu}_{\text{TopFloor}}\) = 31 ng/mL

\(\hat{\mu}_{\text{BottomFloor}}\) = 26 ng/mL

\(\hat{\sigma}_{\text{TopFloor}}\) = 7.4 ng/mL

\(\hat{\sigma}_{\text{BottomFloor}}\) = 8.3 ng/mL

\(n_{\text{TopFloor}}\) = 20

\(n_{\text{BottomFloor}}\) = 23

• Calculate the t-statistic for the difference between the two means. Assume the variances of the two groups are equal.

• Is the difference between the two means significant? Use \(\alpha = .05\) and a two-tailed test. (Don’t forget to calculate the degrees of freedom.)

Question 5

Your collaborator thinks there’s a better way to test this question. He conducts a longitudinal study on bottom-floor residents, measuring blood Vitamin D levels one year, and then measures them again if they move to a the top floor the next year. Overall, he manages to measure Vitamin D twice in a group of 16 bottom-then-top floor residents.

\(\hat{\mu}_{\text{TopFloor}}\) = 31 ng/mL

\(\hat{\mu}_{\text{BottomFloor}}\) = 26 ng/mL

\(\hat{\sigma}_{\text{TopFloor}}\) = 7.4 ng/mL

\(\hat{\sigma}_{\text{BottomFloor}}\) = 8.3 ng/mL

\(n_{\text{TopFloor}}\) = 16

\(r_{\text{TopFloor,BottomFloor}}\) = 0.78

• Calculate the t-statistic for the difference between the two means.

• Is the difference between the two means significant? Use \(\alpha = .05\) and a two-tailed test. (Don’t forget to calculate the degrees of freedom.)

• Do you agree that this way is better? What are the advantages and disadvantages of your collaborator’s method?

Question 1

The Census at School database includes the following question:

Which would you prefer to be? Select one: Rich, Happy, Famous, Healthy

Use the homework-school dataset as your sample. The relevant variable in this dataset is called Preferred_Status. Use this data set to answer the following questions.

• How many students preferred each of these categories?

• Nationally, 6% of students would prefer to be famous, 30% would prefer to be happy, 35% would prefer to be healthy, and 29% would prefer to be rich. You decide to use a statistical test to compare the Oregon student sample to the USA proportions. What kind of test will you use? Justify your choice.

• What are your null and alternative hypotheses?

• What is/are the parameter(s) that define your sampling distribution? What is the critical value on this distribution?

• Calculate your test statistic “by hand” (meaning, don’t use a function that ends in .test).

• How does your test statistic compare to your critical value? What do you conclude?

• Test your calculation by using the appropriate test statistic function in R. Check to make sure you calculated the correct expected values and test statistic.

Question 2

Use the dataset homework-nls for the following question.

• Create the following four new variables:

  • content_2000: average of calm_2000 and happy_2000
  • sad_2000: average of blue_2000 and depressed_2000
  • content_2010: average of calm_2010 and happy_2010
  • sad_2010: average of blue_2010 and depressed_2010

• Variables should not be combined into composites or averages unless they are measuring the same thing. Report the correlations between the measures being averaged above and comment on whether you think there is sufficient evidence to support these averaging operations.

• Conduct the appropriate t-tests to determine if contentedness and sadness changed over time. Report the t-value, p-value, confidence interval, and an effect size for each test.

• Create a graphic or two to visualize these effects.

Question 3

Use the dataset homework-nls for the following question.

• Create the following two new variables:

  • diff_content = content_2010-content_2000
  • diff_sad = sad_2010-sad_2000

• Conduct independent samples t-tests for each measure; use sex as the grouping variable. Report the t-value, p-value, confidence interval, and effect size for each test.

• Conceptually, what kind of effect is being examined with the t-tests in this question?

• Create a graphic (two if you must, but it would be better if you could create just one) to visualize these effects. Be sure to include confidence intervals.

Question 1

Using the homework-school dataset, you will test whether adolescent boys or girls (Gender) spend more time using computers (Computer_Use_Hours).

• What is the design of this study? One-sample, independent samples, or paired samples. How can you tell?

• What are your null and alternative hypotheses?

• What is your alpha level? Justify your choice?

• What kind of t-test will you run to test your hypotheses?

• Check the assumptions of the t-test you want to run. (Because independence cannot be tested statistically, I will tell you this assumption, where or if it is required, has been met.)

• If you have met the assumptions of the test, run the test and interpret the result. If you have not met the assumptions, do something that will allow you to test your hypotheses.

Question 2

In this question, you will perform an interpret a \(\chi^2\) test of independence. In the dataset called homework-school there is the following question:

Which would you prefer to be? Select one: Rich, Happy, Famous, Healthy

Answers are coded in the variable called Preferred_Status. Students are also coded by which kind of school they attend, High school or Middle school, in the variable called School.

• Count the number of students who prefer each status by school. Your answer should have the same format as following table (the rows and columns may be in a different order):

• Convert the preferences to proportions within school level and display in a bar graph.

• Conduct a \(\chi^2\) test for independence. Report the \(\chi^2\) statistic, degrees of freedom, and \(p\)-value.

• Interpret this test: What were your null and alternative hypotheses? What was your decision (reject or fail to reject)? How do you interpret this in terms of your variables?

• How are the expected values calculated for a \(\chi^2\) test of independence? Calculate the expected value of the cell High School students who prefer being famous “by hand.”

Question 1

In class we discussed the problem of p-hacking, which can arise in quite a number of ways. You will simulate one form of it for this question that appears on the surface to actually be a sound strategy for saving resources.

Imagine that you have done a power analysis to determine the number of participants you will need in a study in which the null hypothesis mean is 50 and the alternative hypothesis mean is 55. The standard deviation is expected to be 10. Your power analysis indicates you will need around 35 participants to have power of .80 for rejecting the null hypothesis at \(\alpha = .05\) using a two-tailed t-test. But, a power analysis is just an educated guess and the procedure you will be using in the study is very costly. You would like to get by with fewer participants if you can. So, you start collecting data, adding each case to the data file as you go, and when you get to Participant 5, you decide to start hypothesis testing. As you add each additional participant, you conduct the t-test to see if the mean from the sample collected to that point is significantly different from the null value of 50. After all, if the results are significant with less than 35 participants, why waste additional resources collecting more participants than you need? If the t-test is significant, reject the null and stop the study early. On the other hand, if the t-test is not significant, keep collecting data until you get a significant t-test or you hit your target sample size of 35.

Simulate this scenario 10,000 times. Set a seed (using set.seed at the beginning so I can reproduce your results). For each simulated study, the data should be drawn from a normal distribution, with mean of 50 and standard deviation of 10. In other words, the data will be consistent with the null hypothesis. For each study, tally whether the null hypothesis is rejected, the number of hypothesis tests that are conducted, and the effect size when the study was ended (either prematurely because of a significant t-test or because the planned 35 participants were included). Use Cohen’s d: the difference between the sample mean and the null mean (50) divided by the sample standard deviation. The proportion of rejections over the 10,000 studies is the empirical Type I error rate. We are interested in whether this matches closely the significance level chosen for the t-test (i.e., .05). The average effect size across the 10,000 studies should be close to 0.

One strategy would be to use an outer loop to index the 10,000 studies and an inner loop to index the building up of the sample within a study. Some conditionals will be needed to decide if an interim t- test is significant and whether a test is even conducted (“no” if the sample size is less than 5). You will also need a way to stop a given study and move on to the next one if a t-test is significant. You can use the cohensD( ) function from the lsr package to get the effect size or homebrew your own.

• First, determine the empirical Type I error rate for the scenario described above. How does it compare to the significance level of .05 set for the t-test?

• Determine the average effect size for this repeated testing scenario. How does it compare to the expected value given that the null hypothesis is true in this simulation?

• Construct a histogram that shows the number of hypothesis tests conducted for the 10,000 studies.

• Construct a histogram that shows the effect sizes calculated for the 10,000 studies.

• Now modify your code so that just a single t-test is conducted for each study, after all 35 participants are collected. What is the empirical Type I error rate and how does it compare to the significance level chosen for the t-test?

• Comment on the likely consequence of repeatedly testing the hypothesis in the way that was simulated. Can you think of a way that might have the benefits of ending a study early but without the negative impact on the empirical Type I error rate?