Do you have to run a correlation analysis before running a multiple regression?

IK31

New Member
#1
Is it a prerequisite that you must do a correlation analysis before doing a hierarchical multiple linear regression? I am short of words (only 2,500) for a paper so I was wondering if I needed to include a correlation analysis when I'm really interested in the regression.

Thanks!
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Model building should be based on context knowledge, so you shouldn't run correlations or report them. Correlations can miss confounding, collider bias, mediation, and reverse causality.
 
Last edited:

Miner

TS Contributor
#3
If you look back to the days before computers or even calculators existed, it made sense to do a correlation to screen variables before the more rigorous math required for regression. Today, it would be redundant.
 

noetsi

No cake for spunky
#6
Here is why you don't do this.

(this is real not made up)

Ice Cream Sales -> murder rates. If you did correlation analysis you might believe the lizards are adding drugs to ice scream.

In reality Hot weather -> murder rates and Hot Weather -> ice cream sales.