## How to do factor analysis

Frequent Contributor
Posts: 89

# How to do factor analysis

I was working on linear regression where I have approx 140 variables data set out of which 93 are categorical variable.

I am not able to get how we do factor analysis for these variables.

Im using proc Factor for numerical variables  and categorical variables

For age I have age varlable i want to categorise 1-15 as 1

16-30 as 2,30-rest as 3.
But like this i cannot do  for all 93 categorical variables?

How to approach  to this problem?

SAS Employee
Posts: 340

## Re: How to do factor analysis

Are all those categorical variables ordinal? Perhaps if you have a table, that describes the order of the levels, you could do the recoding by creating a format from the table, then use the format for the recoding.
Factor analysis works best with continuous, normally distributed variables. These new variables are unlikely to be normally distributed.
What is the purpose of your study?
Posts: 3,066

## Re: How to do factor analysis

[ Edited ]

I think you might want to consider performing Partial Least Squares Regression on this data (PROC PLS).

This is, in a certain manner of speaking, similar to Principal Components Regression (not Factor Analysis regression), but has better mathematical properties than Principal Components Regression (and probably better mathematical properties than Factor Analysis regression). PLS has no difficulty handling ordinal or categorical predictor variables.

For age I have age varlable i want to categorise 1-15 as 1 16-30 as 2,30-rest as 3.

Please note that you are creating your own problems by making a continuous variable AGE into a categorical variable, and your life would be so much easier if you treat AGE as a continuous variable. Normally, turning continuous variables into categories is not recommended at all since you are losing information; for example, age 15 and 16 are very close together on a continuous scale but if you create categories, 16 is very different than 15. While I don't know your problem or what types of results you are trying to achieve, in most cases, I wouldn't do this (yes there are exceptions).

--
Paige Miller
Discussion stats
• 2 replies
• 300 views
• 1 like
• 3 in conversation