# Not sure which statistical method to use to analyze results

#### kylemcat

##### New Member
I am trying to figure out which specific bone fractures (out of a total of 16 different types of fractures) are associated with two types of bladder injuries (a severe one (867.1) and a less severe one (867.0)). Below is an imgur link that shows a picture of a simplified version of my data set. I used a "TRUE/FALSE" nomenclature to describe when an event occurred. So a "TRUE" value under the "Pelvic fracture 1" columns means that that patient suffered from a type 1 pelvic fracture (for example).
I have no clue which SPSS tools that I need to use to figure out which specific pelvic fracture are associated with 867 and 867.1 bladder injuries.

Thank You,

#### obh

##### Member
Hi Kylemcat,

So patient #1 has 2 pelvic fractures #1 and #2 and suffers from less severe bladder injury?
I assume the dependent variable is the bladder injury?
I assume you have more than 2 patients?

#### kylemcat

##### New Member
Hi Kylemcat,

So patient #1 has 2 pelvic fractures #1 and #2 and suffers from less severe bladder injury?
I assume the dependent variable is the bladder injury?
I assume you have more than 2 patients?
Hi! Yes, that is correct. And yes, there are many more patients. The total is around 20K. I just provided a fake example of the rules just to illustrate how my data is structured.

#### obh

##### Member
I assume you want to be able to predict what bladder injury base on the fractures?

Or more accurately to determine the probability for

Otherwise it is only analysing the data. (example for bladder 1. 20% fracture 1 30% fracture 2 and 50% fracture 3

Last edited:

#### GretaGarbo

##### Human
Isn’t it about just doing a cross table with one variable (pelvic fracture 1 or other) as rows and type of bladder injury as columns? And possibly do a chi-square test.

A confusing thing is that, from the example, it seems to be possible to have both pelvic fracture 1 and pelvic fracture 2.

Maybe do a logistic regression (which is close to the chi-squarred test) or maybe a classification, like Fisher’s linear discrimination analysis.

#### obh

##### Member
Hi Greta

I'm not sure why do you want to do the chi-square test in this case? just comparing between Bladder 1 and bladder 2 may not give too much?

I was leading to Logistic Regression if Kylemcat wants to predict the probabilities of Bladder 1 and bladder 2 based on the fractures.
Also, In this case, having more than one Fracture is not a problem.

#### GretaGarbo

##### Human
Maybe (or probably!) I have not understood the situation. I thought he had a bladder-variable (with two levels) and a fracture-variable (also with two level) so that you can make a two-by-two table. The Pearson chi-squarred test for that will be very similar to the likelihood based (likelihood ratio test) in logistic regression.

But since the OP has 20k observations I guess that estimation (simply doing the table or parameter estimates in logit) is more important than testing.

But maybe I have misunderstood what it is about.