Date: Tue, 20 Mar 2007 17:01:22 -0400
Reply-To: Peter Flom <flom@NDRI.ORG>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Peter Flom <flom@NDRI.ORG>
Subject: Re: my confusion between multicolinearity and mediated effect
Content-Type: text/plain; charset=US-ASCII
Dear Wensui
Multicollinearity happens when one independent variable is (very close to) a linear combination of other IVs. Exact multicollinearity is usually the result of a silly error like including height in inches and height in cm as IVs. Near multicollinearity is trickier. An example would be predicting a person's weight based on, say
length of femur + length of tibia + length of humerus + height
or some such.
Mediation is different. Mediation occurs when the effect of one IV on the DV is due to another IV. For example - the amount of damage caused by a fire is positively related to the number of firefighters who show up. But this is mediated by the size of the fire.
My favorite mediation example is that, in young children, astrological sign is related to IQ. This relationship attenuates with age, and disappears in late childhood. hmmmm What's going on? Well, children enter school based on birthday, so some 5 year olds will have had a lot more school than other 5 year olds. This relationship is mediated by amount of time in school.
Mediation may CAUSE collinearity: If you included BOTH number of firefighters AND size of fire, they would likely be colinear. But the idea is different. If you identify the mediating variable, the colinearity goes away.
Does that help?
Peter
>>> Wensui Liu <liuwensui@GMAIL.COM> 03/20/07 11:22 AM >>>
Dear Listers,
Sorry that It is rather a statistical question than a SAS question.
Can someone on the list help me distinct between multicolinearity and
mediated effect? I think I am lost about mediated effect and mediated
model, which are created in psychological world.
Thank you so much!
--
WenSui Liu
A lousy statistician who happens to know a little programming
(http://spaces.msn.com/statcompute/blog)
