We are examining a 200-variable database for factors affecting a discrete dependent variable. What methods are recommended to sift through the potential covariates before running our regression? More specifically, we are interested in re-admission to the hospital following various types of pediatric surgery. Many of the potential independent variables are correlated with one another and I am not sure which ones to exclude.
It is usual to exclude highest co variates first but there needs to be some indicator of the effect. Of removal
this is is something that is undertaken in AMOS to refine a model - by removing factors with the largest covariates to start you can observe indexes such as AGFI and GFI to see if the indexes increase towards a value of 1. Quite simply the removal of factors with large covariates can stop once the indexes show no further improvement (in movement towards 1
One option is using principle component analysis (PCA). This analyze rescues n variables to k component; each component is a linear combination of related variables.
Hi,
Factor analysis is one of the method to do the reduction of some variables.
Kindly go through the following links. Then you will get an idea to reduce the no. of covariates.
http://ohttp://www.ncbi.nlm.nih.gov/pmc/articles/PMC3144483/
rdination.okstate.edu/reduce.htm
http://www.statsoft.com/Textbook/Principal-Components-Factor-Analysis
http://www.ipr.northwestern.edu/qeworkshops/stage2/812_1_Cook/day4.ppt
Many thanks to everyone for their suggestions. I'll be looking into PCA with the help of the links some of you noted.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis