Hi,
I'm struggling quite a bit, Hope someone can help.
My research question is roughly to what extent (and what) characteristics of murder and manslaughter cases are related. The data consists of 350 murder and manslaughter cases, information on 35 variables, all categorical.
The approach is inductive. I would like to let the data speak for itself - by "comparing the categories of all variables" without expectations, without independent and dependent variable constructions - and gain insight into which characteristics are closely or most closely related. I would like to arrive at results that describe, for example, the following: homicides involving a young victim are often committed outside on the street, often also have a younger perpetrator, are often committed on the basis of physical violence or use of a knife and in connection with a argument that got out of hand. In this way I hope to be able to map out what scenarios arise.
I studied multivariate analyzes for categorical data and came up with Multiple Correspondence Analysis (MCA). I really like the way MCA is visualized. I’m aware of the fact the interpretation of an MCA is not that easy. After producing a first output, I run into the problem that the first 2 dimensions explain only 20% variance.
Q1: Can I interpret results on this model?
Adding more dimensions to the model increases the explained variance per dimension by an average of 8 percent. In this way I think I could arrive at an acceptable percentage. But then I no longer know how to interpret the results. So either the result is that the model is not suitable or not with so many variables.
Q2: Suppose I want to continue with MCA, but with fewer variables. How do I find out which variables to omit first?
Q3: how would you answer my research question? With which analysis methods and in what order?
I checked whether cluster analysis is suitable, but I am not quite sure.
All input is more than welcome!
Kind regards,
Rob
more information:
I use SPSS. I have already compressed the variables. The categories vary from 2 (for example gender) to 6 (types of murder weapons). Some variables have a lot of missing values. With those variables, however, I am only interested in the positive responses and I hope to be able to discover relationships based on a smaller N.
In addition to Rob's question, What is magic about 25 per cent that 24.85 per cent doesn't possess.? Thanks, David Booth
first perform kmo and sphericity test, then exploratory factor analysisis to find the number of latent factors.
Hi, thanks a lot for your respons.
SPSS gives me a non positive warning and closes factor analysis without output. After some cleaning I come to 273 cases and 30 variables, but this data still is not good enough to deliver a KMO outcome. And yes, when I drop some more variables I'm getting KMO's above .5, but this i did randomly. Wonder if there is an indicator that makes it clear which variables I should drop first in order to get a better KMO outcome.
on top of that: I still wonder whether factor analysis is the right method. Isn't an MCA a better approach? Or is the factor analysis necessary to find out which variables I should include in the MCA?
Why don't you revise your experiment by choosing some dependent variables to test and run multivariate or ordinary regression to see what explains what in you data. Set up your hypotheses before hand of course. Just a suggestion D. Booth
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation