The first part of this exam is based on a study by Anderson and Dill (2000) on the influence of violent video games on behavior. This is part of a long-running series of studies on this topic by Anderson. The study had two parts, the first involved data collected on undergraduates, and consisted of self-report data on aggressive behavior, video game playing, and other variables. I have reproduced the data fairly faithfully, and the conclusions that you should draw will parallel those drawn by the authors.
Anderson and Dill collected data on 227 respondents on (1) Aggressive delinquent behavior (AggBeh), (2) Nonaggressive delinquent behavior (NoAggBeh), (3) exposure to Video Game Violence (VGV), (4) Aggressive Personality (AP), (5) Gender (1 = Female, 2 = Male), (6) Time spent playing video games (Time), and (7) Grade Point Average (GPA). Their goal was to predict the level of aggressive behavior from variables related to video games. The data are available in AndersonDillRevised.sav.
1. Produce a correlation matrix among these variables, and concisely describe what you learn from this matrix.
2. Anderson and Dill point out that some of those correlations are point biserial correlations. What ones are those?
3. Several of the measures that they used were computed by collecting data on several questions and then averaging the answers on these questions to obtain a measure on a theoretical scale. For example, the Aggressive Personality (AP) score is the average of an irritability scale (CIS) and an aggressiveness questionnaire (AQ). Comment on whether this would be a good or a bad thing, and what effect would it have on the variance of AP scores? (You probably want to think about this a bit, but you should be able to come up with some sort of answer. You might even try averaging some of the variables here, just to see what happens with the resulting variable--but this is certainly not required.)
4. Using a simple additive model (i.e. no interaction terms), how well can you predict aggressive behavior from these variables? (Think about this question first, because you would probably not want to include all of the variables in the regression.) Explain why you used the variables you did.
5. I would have used Gender as one of those variables, because it is certainly likely to be a predictor of delinquent aggressive behavior. But when I did, it was not significant in the model that I used. Why might that be?
6. The authors wrote "When testing for interaction and main effects simultaneously in regression models with correlated predictors, it is recommended that continuous independent variables be centered to reduce multicollinearity problems (Aiken and West, 1991). We standardized all three continuous independent variables used in the various regression analyses to follow (...) to facilitate comparisons among them."
Comment on the reason for centering variables.
Comment on the fact that they standardized the variables [z = (X - mean)/st.. dev.] rather than simply centering them. What effect would this have?
7. In their abstract they wrote "Study 1 found that aggressive behavior and delinquency was positively related to real-life violent video game play . The relation was stronger for individuals who are characteristically aggressive ...)." What is the term that describes the kind of relationship that they are looking at? (I found their original wording slightly confusing, so I reversed the wording a bit.)
8. How would you go about demonstrating the truth of the statement in the previous question on the basis of the data that you have? (Tell me how, don't do it yet.)
9. Now do what you said that you would do in the previous question. (I'm a bit worried about this question, because if you can't answer #7, you may not be able to answer #8 and #9. if you are really stuck, send me a message.)
10. Editors have recently started asking authors to present a plot showing what it means to say that the relationship between aggressive behavior and video game play is "stronger for individuals who are characteristically aggressive." One way to do this would be to create a variable that identifies which quartile a person is in on Aggressive Personality (AP). First you have to find the quartiles (which you can do using the frequencies procedure under descriptive statistics). Then you have to recode your data into a new variable (Quartile) that indicates on a 1 to 4 scale which quartile the subject is in. And then you have to plot the overall relationship using a scatterplot with the quartile variable entered into the "fit markers by" box. Lastly, where you would normally tell it to insert a regression line, you instead tell it to "fit line by subgroups." You may have to play a bit, but you can certainly figure this out. Do this and tell me what you see.
11. The authors state "The positive association between violent video games and aggressive personality is consistent with a developmental model in which extensive exposure to violent video games (...) contributes to the creation of an aggressive personality. Can you think of an alternative explanation. Why can't we compare your explanation and theirs here?
12. The authors point out in their introduction that "Three of the studies (...) yielded reliable positive correlations between video game playing and aggression. The fourth correlation (...) did not differ from zero." [To make this question even more appropriate to your circumstances, Maggie Boyer's dissertation (which I am reading on a beautiful Saturday afternoon, and has noting to do with Anderson's work) says "When children's patterns of bias were compared to a neutral attentional stance, children with RAP demonstrated selective attention to subliminal pain words [t(46) = 2.53, p < .05], and exhibited trends of selective atten6tion to subliminal social threat words (p < .09) and selective avoidance of supraliminal social threat words (p < .07).] Comment on these conclusions from what you know about statistical power and research design. Does Anderson's conclusion mean that there is a 3/4 chance that video games are related to aggression?
13. In the future I suspect that you will frequently see SPSS, or some other program printing out a statistic (e.g. a coefficient) and following that statistic with a bootstrapped confidence interval on that statistic.
a) What would you do with the confidence interval that is printed out?
b) Why do you suppose that the programmers might have chosen to generate that interval by bootstrapping?
c) Very briefly and informally tell me how they would have gone about obtaining that bootstrapped interval.
14. I am stretching our example so as to sneak in the analysis of covariance. One thing that you could do is to compare males and females on Aggressive Delinquent Behavior. (Run that analysis.) But you might also think that it makes sense to use NonAggressive delinquent behavior as a covariate. (Run that analysis too.)
a) Briefly tell me what kind of question the analysis of covariance is answering here, and how does it differ from the question addressed by the analysis of variance.
b) I have some reservations about the validity of this analysis of covariance. Why might that be?
Last revised: 05/04/2002