Too many variables - Multivariate analysis of covariance - New to this

[griffin.625]

Hi there.

This will be my first post on the forums, I'm a Psychology student currently conducting research at an internship. We treat IDD (Intellectual Developmental Disorder) patients with different types of therapy, and I've been assigned to study the dosage components of DBT (Dialectic Behavioral Therapy).*

I've never done legitimate statistical research, and my only background in the area is reading studies in my private time, a basic course on Research Methods, and a basic course on Data Analysis.*

Here's a quick background on the study:

From a pool of about 50 subjects, the correlational data I'm expected to produce accounts for a very large number of variables, both with multiple predictor IV's, dosage component IV's, and multiple behavioral outcome DV's. All of this data is pulled from client records - it is not an ongoing experiment with trials.*

We have 6 qualitative, categorical IV's (Race, Diagnoses, Medication(s), Trauma, Co-occurring Treatment, and whether DBT was individual or group therapy).*

4 quantitative, predictor IV's (Age, SES, ACE, IQ).

9 quantitative, dosage component IV's (Weeks in Treatment, Total Sessions, 6 DBT component dosages, and Out of session practice)

15 quantitative, behavioral outcome DV's (Pre/Post measures of behaviors measured on diary cards).


I'm writing the Protocol for the study, and have coded the DV's for # of incidents, as well as a 0-2 Likert scale for the diary cards in protocol.*


Here's my issue:

1). I attempted setting this up in excel, and when I ran a test of regressional analysis, my data tables and accompanying graphs do not make sense or even appear as readable data.

2). My variables are all existing on different scales:

IQ level scale
0-2 scale [none, a little, a lot] and 0-1 scale [no yes] from diary cards
0-10 scale [ACE]
0-7 scale [SES]
Unlimited scale [number of behavioral incidences]
categorical IV's

3). I don't have an equation set up for such a maddening cluster of variables, nor do I possess the know-how of creating one.*



My apologies if this is a lot for a first post, or a post in general. I care a lot about this project as it is the cornerstone of my professional crusade - I'm going to make it great.*

Please let me know if you have any insight or guidance into such a situation - any help is greatly appreciated.*


Alex

 

[noetsi]

I've never done legitimate statistical research...

Have you done illegitimate statistical research? Well there are plenty who do this in honesty or do it wrong. Including "experts."

That seems like way to many variables for that number of cases in honesty. I don't think using excel for regression is ideal (it has not way to test regression assumptions for example). You should try to run it in one of the software's created for statistics. Why do you data tables and graphs not make sense? It is not possible to tell from your comments.

I am not really sure what you are asking. Are you asking how you reduce the number of variables used? Theory is the best way. Trying what makes sense to you is probably second best removing variables that are not significant. Your problem is your power will probably be low so something may not be statistically significant that is. Doing a power test, determining what your power is, is helpful.

It is always a good idea to look at the existing literature before you run any type of test.

 

[CB]

From a pool of about 50 subjects, the correlational data I'm expected to produce accounts for a very large number of variables, both with multiple predictor IV's, dosage component IV's, and multiple behavioral outcome DV's. All of this data is pulled from client records - it is not an ongoing experiment with trials.*

We have 6 qualitative, categorical IV's (Race, Diagnoses, Medication(s), Trauma, Co-occurring Treatment, and whether DBT was individual or group therapy).*

4 quantitative, predictor IV's (Age, SES, ACE, IQ).

9 quantitative, dosage component IV's (Weeks in Treatment, Total Sessions, 6 DBT component dosages, and Out of session practice)

15 quantitative, behavioral outcome DV's (Pre/Post measures of behaviors measured on diary cards).

I'm sorry to say this, but this simply isn't a feasible research study. 50 participants isn't really enough to study the effect of a single dichotomous IV on a single quantitative DV: You have 19 IVs and 15 DVs. It's really just out of the question.

I would suggest that you go back to your supervisor and suggest that you need to have a meeting between those designing the project and someone with expertise in research design and statistical methods to re-design the project. Essentially this will need to comprise both a massive reduction of the number of variables involved - focus on your key hypotheses - and a substantial increase in the number of participants.