Desperate urgent help

#1
Hi everybody,

First, let me say how much I appreciate your help here. Its a life saver. I have a presentation tomorrow on my research and I am about to be Embrassed :eek::eek::eek:
My statistical analysis is, well, awful. Here's where I need help.

I am researching the relationship between certain traits of court judgments. What of utmost interest is how certain factors (say, gender of judge, jurisdiction or subject-matter) correlate to the outcome of the case. I am also interested in other inter-correlations.

What I have is an excel sheet with 30 variables. The sample size is ~100, and I estimate that the entire population is ~300 (but it could be as high as 700). Most of my variables are binaries. Outcome is 1 for victory, 0 for loss. Jurisdiction is 1 for California, 0 for nothing. Jurisdiction2 is 1 for NY and 0 for nothing (and so on).

At first I tried my limited knowledge and produced correlations between certain columns. It was o.k., but I didn't know what was the statistical significance of these correlations or how to compute them (I am using excel).

I turned in my hour of need to a friend, who suggested a hyper- geometric test. According to him, this test is the only one which makes sense in the context of binaries. He did the computation and a few interesting results came up. However, after I found the results he told me that he have to correct FDR (for multiple hypotheses). He did it and not much was left of my findings... :(

Then, someone who I talked to said that I should do regression analysis and not test correlations.

I am completely at loss. My presentation is tomorrow and I have no idea how to present the data or how to justify the choice of correlation / hyper-geometric analysis.

I truly look forward to hear you comments and ideas. Please elaborate as possible and try to assume that my statistical intelligence(my SI) is pretty low. :confused:

I truly appreciate your help.

Y.
 

bugman

Super Moderator
#2
What is the question you are trying to answer?

Certain log-linear models will help you here. Are you interested in the interaction of all three catergories?
 
#3
Thanks!

It is not so much a concrete question as just trying to figure out the connection of the outcome of the case to certain properties of it (the value of the deal, the identity of the parties etc.).

Do you think the hyper- geometric test makes sense in this context? what about regression analysis?
 
#4
correlation in excel

So I didn't quite get if you did or did not know how to do correlations in Excel.

If the answer is "no", this site is pretty helpful. http://www.stattutorials.com/EXCEL/EXCEL-pearson-correlation.html

Regarding using a regression, you do have a pretty small sample size and a lot of potential independent variables. Rule of thumb (or so I've been told) is to have at least 10 people per IV. So you can have 10 independent variables at most.

But I guess the real answer/direction would be rather than mining the data, to see what others have done, or to use a theory to guide you. Don't get me wrong... I LOVE to mine data. But I got slapped down for what appeared to be "looking for only significant variables" in a recent journal article submission. Sometimes, non-significant findings are also important if the variables selected have good reason at being selected.