You should cross off 15 points worth of questions. If you forget to cross them off, I’ll delete that final 15.

The following exam is based on a single study by Kathleen Carrol et al. (1998) Treatment of cocaine and alcohol dependence with psychotherapy and disulfiram. Addiction, 93, 713-728. You may not look this paper up until after you turn in your answers, but that probably wouldn’t make much difference anyway.

There has been a good deal of evidence that some kinds of therapy, including Cognitive Behavior Therapy (CBT), are at least partially effective in dealing with cocaine addiction. With one exception, there has not been any evidence that specific drug therapies are useful in this case. You probably know that antibuse (also known as disulfiram) has been shown to work with alcoholics, because it makes them sick if they have a drink while on antibuse. But there is no pharmacological reason to think that antibuse will work with cocaine.

But psychologists are smart people!! We know a few things about learning and about behavior change. One of the things we know is that there are context effects. For example, just being in the context (physical or otherwise) in which you have previously taken cocaine is likely to increase your desire for cocaine. Well, cocaine addicts are generally (85%) also abusers of alcohol, so it’s a good bet that alcohol provides, in part, a context for cocaine. So keep the folks away from alcohol, and they may be less inclined to use cocaine. Similarly we know, though only second hand, I’m sure, that people are not at their best when they drink. Good intentions, such as not to drive, not to molest your date, etc., tend to go out the window when you drink. And not using cocaine may be one of those good intentions. Again, keep people from drinking and they may do better at not using cocaine. And how do we keep people from drinking?  Well, fill them with antibuse, of course!

Carroll et al. set out to investigate this possibility. They divided 117 cocaine users into 5 groups. All of these groups got some sort of therapy, but it ranged from Cognitive Behavior Therapy (CBT), to 12-Step Facilitation (TSF) (similar to Alcoholics Anonymous), to Clinical Management (CM), which I take to be the equivalent of Foa’s supportive counseling. In addition, some of the clients got disulfiram (antibuse), and others did not. Ideally this would have led to a factorial design with all combinations of therapy and drug, but they did not have a CM/No Drug condition.

The design of the study, as well as the means, standard deviations, and sample size are:

These are the data only for the abstinence from cocaine, where the dependent variable is the number of weeks continuously abstinent. In the data file you will also find comparable data for number of weeks continuously abstinent from alcohol. (I have left out the number of weeks continuously abstinent from both.) The data can be found in a file called Carroll2.sav. The groups are ordered in the way given here, and the value labels have been applied. (There are some cases where the weeks are negative, which is impossible. That’s just the way that the data were generated, and you will have to be tolerant. Just pretend you didn’t notice.)

Points

1.  An easy question:  Without looking at the data, what do you expect that the distributions would look like? I am not asking about group differences, but about the kinds of numbers you would really expect to see in such a study. (Such things as maximum, minimum, shape, mode, whatever.) Why would you expect whatever you expect?

2.  Now draw a histogram of each variable using SPSS. Does it look like the results you expected—it’s not fair to go back and change your answer to #1.

3.  If you are going to run an analysis of variance on these data, what kinds of assumptions would you need to make?

4.  Are you satisfied with whether or not you have met these assumptions? If not, what might you do about that? (I’m not asking you to actually do this—in fact, I don’t want you to.)

5.  The following is a quotation from Carroll’s paper.

"The primary contrasts were: (1) medication (disulfiram versus no medication), (2) Active psychotherapy versus psychotherapy control (CBT and TSF versus CM), (3) active psychotherapy type (CBT versus TSF), and (4) psychotherapy/pharmacotherapy interaction (interaction of medication and psychotherapy type)." (Don't worry about the last one.)

a)  Tell me how you would translate these contrasts into actual groups to be compared. I don’t mean that I want you to tell me what coefficients you should use (yet), but I want you to tell me which of the five groups you would compare with which other groups? (hint: This is not obvious—you’ll do best if you pretend that the CM/Disulf group wasn’t there for the first and third contrast.)

1)  Contrast 1  

2)  Contrast 2

3)  Contrast 3

b)  This one is hard. What do you think they meant by their 4^th contrast?

c)  Now tell me what contrast coefficients you would use when you go to SPSS to run the anova. You can give me those coefficients as if the groups were of equal size. (I’ll give you the last one.)

1)

2)

3)

4)  1    -1    0    -1    1

d)  Is there some other analysis that you would think is particularly important? I am not saying there needs to be, but perhaps you have a favorite. If so, why; if not, why not?

6)  Before you run the analysis, what are you going to do if the overall F is not significant? And why did you say that?

7)  Now run the analysis and paste in the results below.  At this point I only want the overall summary table.

8)  What would you conclude from what you have here?

9)  The important part of the results focuses on the contrasts that they wanted to run. First write a convincing argument to tell the editor to drop dead when he complains that you should not have done what you did because you have a non-significant F.  (It is not good policy to tell an editor to drop dead, so phrase it more tactfully.)

10)  Now run the contrasts and paste those results below.

11)  What would you conclude from these contrasts. Break your answers down by the hypotheses outlined above by Carroll.

12)  Carroll reported that “the effect sizes (d) for disulfiram compared with no medication on duration of abstinence from cocaine, … were 0.42, … . In addition, compared with the control condition (CM) the two active psychotherapies (CBT and TSF)…(had effect size of) … 0.16.” Show me how you would calculate these results. (Hint: There is room for rounding error here, so don’t fall on your sword if you don’t get exactly the same result.)

13.  Now we come to power. As you can tell, the overall F was pretty borderline. However the contrasts are certainly interesting, and it really would be neat if someone would go out and replicate these results. I could ask you to calculate the power for one of the specific contrasts, but you wouldn’t think that was fair. So tell me how many subjects you would need, assuming equal sample sizes, if you wanted power to be .80 in a replication of the overall study. (Hint: To get power of .80, you are going to need a value of f of approximately 1.70. I gave you that because you don’t have tables.)

14.  Now that you have calculated effect size and then power, you might just as well give me another measure of effect size—i.e. a measure of magnitude of effect. I don’t care which one you give me, but explain what the other one is, (we discussed two) and when you would use it.

15.  Good studies generally come at the same experimental hypothesis from more than one direction, and this paper is no exception. Let’s assume that my original hypothesis about drinking being both a context for cocaine and something that makes it hard to resist temptation. We ought to be able to address this problem in a correlational sense. We know how well each person abstained from cocaine, and how much they abstained from alcohol. Frame a statistical hypothesis about the role of alcohol in cocaine abuse, and test it with the data that you have here. What will you conclude?

16.  We have been worrying about the number of weeks each individual subject was free from cocaine use. What if we look at the groups themselves. Carroll reports that the five groups (ordered as above) had the following percentage of subjects who were cocaine free for at least 3 weeks: TSF = 30%, CBT = 28%, CM/dis = 30%, TSF/dis = 52%, and CBT/dis = 57%. You know the sample sizes, so you can create a contingency table showing the numbers of subjects in the “free” and “not free” category for each group. You can then run a chi-square, either by hand or using SPSS, to test an hypothesis. What hypothesis are you testing?  (No, I’m not cheating by taking something from before the last exam, because I am going to turn this into a question we covered in this half of the semester.)

17.  Now that you know chi-square and the overall sample size, you should be able to get yet another measure of how well we are doing (sort of an effect size measure). From the measures using chi-square in Chapter 10, what can you conclude here.

18.  You will have noticed that I had you study some stuff on post-hoc tests, but didn’t ask you to use them here. Why are the tests that we ran here better at asking the kinds of questions we want to ask?

19.  What would you conclude from this whole study if you were suddenly given your degree and sent off to work in the area of drug abuse treatment and prevention? 

That’s all, folks!