Factor Analysis Question - Covariance and Correlation matrix

September 13, 2010 5:26 PM

In exploratory factor analysis, why would I want to do a factor analysis on the covariance matrix rather than the correlation matrix? I did some tests, and they clearly assess different things, but I'm unable to find an explanation of the conditions in which you would prefer to use the covariance matrix over the more commonly used correlation matrix.
posted by singingfish to Science & Nature (1 answer total)

Best answer: This is an area of quantitative modeling which is as much art as science, which is why you haven't found any clear-cut criteria for choosing between covariance and correlation. One question to ask yourself is whether the variables are in units that can be meaningfully compared (e.g., are they all lengths expressed in cm). If so, then keeping all the information in the covariance matrix might make sense; your factor model will try to explain the variable with the highest variance. If the variables are not really comparable (e.g., height, weight, and age) the correlation matrix might be preferable because it puts everything in units of standard deviation.
posted by drdanger at 7:02 PM on September 13, 2010 [1 favorite]