I need statistics help! :confused:

[Finella]

Hi, everyone!

I am analyzing data for a music therapy study, and I am having problems finding the right kinds of statistical tests to apply to my data. I figure there's got to be someone here who knows something about this stuff... and on my day off I am using all the resources I got... statistics book, e-mail to the prof, and here. So I'm setting out the problem and I will wait for bites while I try to make sense of the stats book over there.

My research is comparing substance-addicted people's scores on rhythm tests with non-substance addicted ("normal" for our purposes) people's scores on the same tests. I have the tests analyzed, plus I have data on the addicted Ss drugs used, how many years they used each type of drug, and how long ago they last used each drug.

My first problem:

I have the addicted Ss broken down into ten different drug types (Cocaine/Crack, Hallucinogenics, Barbituates, Amphetamines, Alcohol, etc.). I want to compare all the scores by these types at one time to see if there's any significant differences between them. Most of the data is in numerical form (# of seconds a person played in a steady pulse, # of beats synchronized with a provided drum beat, # beats/minute).

The problem with the factorial ANOVA is that one source I have says it's "preferable" to keep the number of compared groups to "two or three". I have ten.

Additionally, some of the data is categorical: some subjects, when asked to play with a provided drum beat, played with a completely different pulse, maintaining this pulse but without regard to the one played to them. So I have Y/N data to answer the question "did the subject match the provided pulse?" I've looked at Chi-Square stuff to do this kind of data, but I have no "expected" scores to enter in here (there's never been this kind of research done before on anyone).


Okay, don't know who on earth would have the time to think on this, but I'm guessing someone out there really knows statistics and might be able to at least point me in a direction....

Thank you.

---,---'--{@

 

[TruthSeeker]

for a start, how many subjects do you have in each group (and each drug type)?

Did you recruit by (or stratify by) drug type or did you take all comers?

How many tests are there? Any repeated measures?

I guess what I'm asking for is a more detailed and precise description of your design.

Also, I'm just curious, are these real data you collected or is this a stats assignment?

 

[Finella]

Hi, TruthSeeker...

yeah, you do need more info... here it is.

for a start, how many subjects do you have in each group (and each drug type)?

20 in each group of Normal and Addicted. So there's smaller numbers in each group of drug used in the addicted category... and of course some addicts used more than one drug, so that complicates things.

Did you recruit by (or stratify by) drug type or did you take all comers?

I took all comers I could get for the addicted category, so there's no limits or boundaries on the drug types. For instance, I have three MDMA users, but 17 pot users.

How many tests are there? Any repeated measures?

Five tasks, no repeated measures. I would probably analyze each task seperately with each statistical test.

Also, I'm just curious, are these real data you collected or is this a stats assignment?

Oh, this is real data -- this is my master's research.

thanks!
---,---'--{@

 

[TruthSeeker]

Hi, TruthSeeker...

yeah, you do need more info... here it is.


20 in each group of Normal and Addicted. So there's smaller numbers in each group of drug used in the addicted category... and of course some addicts used more than one drug, so that complicates things.


[/B]

With an n of 20 in the addicted group, you probably do not have enough power to do any real analyses by drug type. 20/group with 5 DVs....what is your power here?


I took all comers I could get for the addicted category, so there's no limits or boundaries on the drug types. For instance, I have three MDMA users, but 17 pot users.

The overlap and low n's probably mean you can't do the analyses you want. You might be able to do a covariate analysis but I doubt it. Again, what is the power?

Five tasks, no repeated measures. I would probably analyze each task seperately with each statistical test.

why are you assuming these tasks are independent? They don't sound independent to me. I bet you'd find they are highly correlated with each other. I'm sure you checked this when you did your prelim analysis and assumption checks, right?


Oh, this is real data -- this is my master's research.

B]Well, then, you must have a data analysis plan written for your proposal...post it and we'll go from there.

I have to run to a meeting soon so I'll leave this to my fellow posters and will check in with you tomorrow.
[/B]

 

[Finella]

Hi, Truth Seeker...

thanks for your help. My prof is suggesting I wait on this particular analysis for later when I publish this study (!). But to clarify a couple things you asked about...

This is a pilot study, hence why the n is so low for both groups. I wasn't certain if there would be enough data to compare each drug type used, and so when I do get to it I might just compare users of groups of drugs against each other rather than the individual drugs themselves.

The 5 tasks are different in nature, and therefore scored completely differently and so, while I may see patterns in the results between a group's performance of all the tasks, I don't think there's a way to compare all of them at once. I wasn't planning on comparing them altogether in my proposal, anyway. My purpose was to see in what specific ways addicted people may differ from normal people in their rhythmic skills, not to say that overall they are different.

My analysis plan was pretty basic. I've been doing t-tests for two independent means to compare numeric scores between the addicted and normal groups. I was then going to do correlations between age and scores and, for the addicted subjects, number of years using drugs/alcohol and scores. I have here in my proposal that I would do a discriminant function analysis on the addicted scores in order to possibly make predictions based on drug used. But, as I said, I will likely leave that to later as I've recently decided.

It's been years since I've done this kind of math, and it's stretching my brain a bit. But I'm enjoying it! Again, thanks for the help. I think I'm on the home stretch of this part for now, but I may be back for the journal preparations in a few months.
---,---'--{@

 

[TruthSeeker]

The t tests you have planned sound fine and with an n of 20 is probably about all you'd want to do. Also, when you do run the analysis for duration of drug you, you might want to use age as a covariate.

But, just to carry forward my thought that these skills are in fact related...you could very easily generate a correlation matrix to get a sense of the magnitude of the relationship. Another possiblity is to run a MANOVA. More sophisticated would to run a factor analysis (40 observations of 5 IVs should be ok, not great but ok for a pilot) on the 5 scores and see if they hang together as one greater measure of "skill" and then compare the two groups on this overall factor score. Anyway, that's probably way to fancy for a master's degree. Just some random thoughts on a rainy and overcast Saturday afternoon.