multicolinearity

This forum is closed, and read-only.
Locked
needpls
PLS Expert User
Posts: 119
Joined: Mon Jun 11, 2012 4:41 pm
Real name and title:

multicolinearity

Post by needpls »

hiii


1)how can i understand that there are multicolinearity between exogenous
LVs???

2) if there was any multicolinearity so PLS result is wrong ? in handbook i saw that show path coefficient is not wrong.

3) if in second order construct there was any multicolinearity between first orders , so PLS result is wrong??

i will wait for ur help

best regards
needpls
PLS Expert User
Posts: 119
Joined: Mon Jun 11, 2012 4:41 pm
Real name and title:

Post by needpls »

no person can not help me???
plzzzzzzzzzzzzzzzzzz it is so emergency
User avatar
Diogenes
PLS Super-Expert
Posts: 899
Joined: Sat Oct 15, 2005 5:13 pm
Real name and title:
Location: São Paulo - BRAZIL
Contact:

Post by Diogenes »


1)how can i understand that there are multicolinearity between exogenous LVs???

To measure the multicollinearity:
- copy the standardized LV scores
- paste them to SPSS (or other software that has multiple regression estimation)
- run multiple regression analysis, using all exogenous LV as independent variable and other variable (could be anything) as dependent.
- in the procedure click to have de collinearity diagnosis.
- see the VIF values.


2) if there was any multicolinearity so PLS result is wrong ? in handbook i saw that show path coefficient is not wrong.
- the LV scores of the endogenous LV will be correct (not biased), but the path coefficients will be biased (bigger bias with bigger multicollinearity), like this:
Small VIF --> Some path will be lower than expected
Little bigger VIF --> some paths will be nonsignificant
Bigger VIF --> some paths will have signal opposite the correlation (between endogenous and exogenous LV), what is known as “suppression”.

3) if in second order construct there was any multicolinearity between first orders , so PLS result is wrong??
The scores will be correct, but the you could have paths positive, negative, and nonsignificant as result of the multicollinearity.
Then I suggest to you not to try to interpret these coefficients as importance of the 1st order LVs.

best regards

Bido
needpls
PLS Expert User
Posts: 119
Joined: Mon Jun 11, 2012 4:41 pm
Real name and title:

Post by needpls »

wow thx so much

i saw my VIF it is about 1.3.

i saw that lower than 3 or lower than 10 is good VIF.

so i dont have multicolinearty? correct??


best regards
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov »

Hi,
Yes, VIF < 5 or < 10 is good.
Regards,
Hengky
juanclondono
PLS Junior User
Posts: 9
Joined: Mon Jul 16, 2012 8:30 am
Real name and title:

Multicollinearity diagnostics

Post by juanclondono »

The data you copy from PLS to SPSS is the result of the bootstrap report, where it says Measurement model specification, right???

In the structural model some variables are constant, is this wrong??

Thanks,

Juan Carlos.
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov »

Hi Juan,

Copy LV score after run algorithm PLS or compute VIF with formula 1/1-R2.
What your mention constant?

Regards,
Hengky
juanclondono
PLS Junior User
Posts: 9
Joined: Mon Jul 16, 2012 8:30 am
Real name and title:

Post by juanclondono »

Hi Hengky,

Thank you VERY, VERY MUCH!! Found it!

Juan
juanclondono
PLS Junior User
Posts: 9
Joined: Mon Jul 16, 2012 8:30 am
Real name and title:

Formative LV's included???

Post by juanclondono »

I have a mixed model, should I analyse formative and reflective LV's together in SPSS to calculate multicollinearity statistics?
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov »

Hello,

No problem, check among construct.

Regards,
Hengky
juanclondono
PLS Junior User
Posts: 9
Joined: Mon Jul 16, 2012 8:30 am
Real name and title:

Thanks Again!!

Post by juanclondono »

Hi Hengky, thanks again for your help!!
Locked