Midterm Exam
3/7/2002
The primary question in this exam is based on a study by Bremner, Shobe, and Kihlstrom (2000), who studied false memories in women who reported that they had been sexual abuse suffered from childhood sexual abuse. They wanted to determine whether women who had been abused were more likely to report false memories in a domain that had nothing to do with abuse.
Bremner et al. used a technique called the Deese technique. This technique involves asking participants to read, and then recall (with a recognition task) words that are part of a common category. For example, subjects might be read a list of 15 words including eye, thread, sew and other words that are commonly associated with the word needle in a free association task. However the word needle would not be in the list of words that were presented. After being read six lists like this, participants were presented with a list of words, which included some of the words that they heard, words like needle that they didn't hear, and other unrelated words that they didn't hear. They were to indicate which words they had heard. The authors referred to words like needle as a "lure," and words that had been presented as "studied" words. The idea was to see if abused participants were more likely to (erroneously) report that the lures were among the words they heard.
This study had four groups. (Be careful, because these are not necessarily the four groups that you might be expecting.) One group consisted of women who had been abused and who suffered from PTSD. Another group consisted of women who had been abused but did not suffer from PTSD. A third group was nonabused women who did not suffer from PTSD, and a fourth was nonabused men who did not suffer from PTSD.
The data can be found at Bremner.sav. (Remember, you are safer downloading it with Internet Explorer than with Netscape.)
The dependent variable is the proportion of items recalled relative to the number possible. (For example, there were 6 lures (one for each category), and someone who recalled 5 of them would have a score of .83.)
The means, standard deviations, and sample sizes are shown below. (I have changed the data somewhat because the data that they presented would not come close to producing the statistical results they reported. They happens more often than you might expect.)
| Abused/PTSD n = 23 |
Abused/ NonPTSD n = 13 |
NonAbused/ NonPTSD n = 16 |
NonAbused/ NonPTSD/Men n = 11 |
|
| Lure | 0.95 (0.11) | 0.78 (0.21) | 0.79 (0.20) | 0.66 (0.35) |
| Studied | 0.93 (0.13) | 0.90 (0.155) | 0.99 (0.05) | 0.93 (0.135) |
Answer all of the questions below.
*********************************************************************
1. How would you characterize this study in terms of its experimental design?
2. Why do you suppose that the investigators designed it this way rather than the way you might at first expect?
3. What do you suppose that the experimenters expected to find?
4. What would you expect might be the correlation between Lure and Studied scores, breaking those correlations down separately by groups?
5. Now go ahead and calculate the correlations referred to in question 4. How do they compare to what you expected?
6. Can you come up with a rationale of why the results came out the way they did? (The values you found are close to what the authors found.)
7. Twice I spoke in class about a paper by Bradley and others that related to the power of repeated measures designs. What would that paper, together with the answer to question 5, suggest about power in this experiment? (You are free to go back and look at your notes.) The answer is not the answer that I would typically expect in such a question.)
8. Run the analysis of variance that is appropriate for the data in the table above, using the Bremner.sav data set. (For the moment ignore the variable called "missptsd."
9. The authors of the paper used Duncan's multiple range test to compare groups. I haven't seen that test used in years--everyone thinks that it is not a reasonable test to be using. Use the REGWQ test to compare the group means, and tell me what you found.
10. Just by looking at the means of Lure and Studied, I suspect that what you did in question 9 was probably not a very useful activity. Why might I think that?
11. How does the fact that we found a significant interaction relate to your answer to question 10?
12. Plot the interaction in the analysis in question 8.
13. The real questions in this study hinge on simple effects. What simple effects would you look at, and why?
14. Compute and present the simple effects you referred to in question 13. Interpret the results.
15. You should recall that I "corrected" myself in class in terms of what I had said about effect size, and suggested that there were more appropriate ways of looking at effect size. Review your notes on what I said in class. Then use what you just rediscovered (or knew all along) to report meaningful effect sizes for the data in this study. (You don't have to report every conceivable effect, just ones that you think are relevant for the conclusions you would like to draw.)
16. In the data set you will see a variable called missptsd. This is the Mississippi Scale for PTSD, which measures PTSD symptomatology. Why is this variable relevant to this study, and what does it tell us. (This will require another analysis of variance.)
17. I haven't said one word in class about the analysis of covariance, so this is probably an unfair question. An analysis of covariance adjusts the dependent variable for any relationship between the dependent variable and the covariate, and then runs an analysis of variance on the adjusted scores. You can easily figure out how to tell SPSS to run an analysis of covariance--just tell it that missptsd is the covariate. Run that analysis, give me the results, and take a stab at interpreting those results. (I know that this is new to you, so my expectations are low. I am interested in what you can do with something that we have not covered.)
18. What would you conclude from this study?
19. What did you think of this study? (If you were the editor, what would your editorial opinion be?)
20. I needed 20 questions to count them each worth 5 points. This is the 20th question. What do you think that you would do to make this a better study?
Return