Second order factor

ATDIAS · Post by **ATDIAS** » Fri Nov 25, 2005 6:22 pm

How could I model a second order factor (latent variable) in Smartpls?

cringle · Post by **cringle** » Sat Nov 26, 2005 6:16 pm

Hi,

second order factor: if this is supposed to be a latent variable in the structural model that has no indicators, it is not possible to model and calculate such models in SmartPLS.

Best
Christian

stefanbehrens · Post by **stefanbehrens** » Fri Dec 09, 2005 2:27 pm

CHIN recommends the following procedure:
Simply assign to the 2nd order LV all the indicators of the 1st order LVs belonging to that 2nd order LV.
Maybe that helps...
Cheers,
Stefan

Note: LV - Latent Variable

joerghenseler · Post by **joerghenseler** » Sat Dec 10, 2005 8:36 pm

True.

However, I think it we should leave the honor to the person who first suggested this approach. Actually it was Wold (1982), who suggested to re-use the indicators. He called it "hierarchical structure".

ATDIAS · Post by **ATDIAS** » Fri Dec 23, 2005 4:55 pm

Please, in order to improve my knowledge, what is the reference of Wold(1982) and how can I get it?

awill · Post by **awill** » Sat Dec 24, 2005 10:25 am

Hi!

I suppose Wold (1982) means:

viewtopic.php?p=360#360

panhans · Post by **panhans** » Wed Jan 11, 2006 2:12 pm

... just for clarification: Does that mean for example that my two 1st-order LVs (each formatively measured by two sets of indicators) have only paths towards a 2nd-order LV (that is simultaneously formatively measured by the combined set of the same indicators) that in turn has paths towards endogeneous LVs?

rmerz · Post by **rmerz** » Mon Feb 06, 2006 10:59 pm

I have run second order models in PLS using a slightly different approach. Try merging the first order LV scores back into the original data and then create a new model using the first order LVs as indicators of a new second order LV. Construct the rest of the model as was done in the original first order model--of course, leaving out the two first order LVs that are now indicators.

So, for example, if you have two first order LVs that are more highly correlated with each other than with a common target LV, combining them as indicators of a second order LV will help control the resulting multicollinearity and stabilize the estimates. If necessary, the weights can then be considered the impacts of the first order LVs onto the second order LV.

Russ

panhans · Post by **panhans** » Tue Feb 07, 2006 7:49 am

Dear Russ, many thanks for your help. Your recommendation intuitively makes sense to me. Do you know whether I can also quote this approach from some article? Best, Dirk

rmerz · Post by **rmerz** » Wed Feb 08, 2006 4:31 am

Dirk,

I am aware of two articles that have mentioned this approach (there are probably others):

Chin, Wynn and Abhijit Gopal (1995) "Adoption Intention in GSS: Relative Importance of Beliefs," Data Base Advances, Vol 26, Nos, 2&3 (May/August) 42-63.

Bagozzi, Richard (1985) "Expectancy-Value Attitude Models: An Analysis of Critical Theoretical Issues," International Journal of Marketing Research, Vol 2, 43-60.

Also--the correct way to calculate the path coefficient of the first order LV onto the second order LV is to multiply the first order path coefficient times the second order weight.

I hope this is helpful.

Russ

ghozali · Post by **ghozali** » Thu Feb 09, 2006 11:45 am

Hi,
Second order factor can be estimated using the repeated indicators approach, also known as the hierarchical component model suggested by Wold (cf. Lohmoller, 1989).

Let say the second order latent construct is measured with two first order component measures, C1 and C2. The component measure are further measured with two item measures each - component C1 is measured with item I1 and I2 whereas component C2 is measured with item I3 and I4.

In the repeated indicators approach, the item measures I1,I2,I3 and I4 are used twice: once for measuring the first order component measures C1 and C2, and then again for measuring the second order latent construct that is also measured by the first order component. I have done this using Smart PLS and the result is fine.

Best regards,

Prof. Dr. Imam Ghozali
Faculty of Economics
Diponegoro University, Indonesia

nielsvanq · Post by **nielsvanq** » Wed Mar 22, 2006 3:51 am

Hi there

I tried to reconstruct the repeated indicators approach but I cannot make it work in SmartPLS. The program refuses to draw the connection line between the second order LV and the MVs which are already connected to the first order LV.

How can one do it in SmartPLS?

Thx

Regards
Niels

panhans · Post by **panhans** » Wed Mar 22, 2006 6:58 am

... I just added the same MVs to the model a second time and connected this second set to the 2nd-order LVs.

Best,
Dirk

stefan0603 · Post by **stefan0603** » Wed Mar 29, 2006 9:43 am

--- and then: I am just wondering if the path coefficients from the first-order LVs to the second-order LV may be neglected? And the only path coefficient which counts is the one from the second-order LV to another exog. LV?

Am I right or do the path coefficients have to be somehow multiplied?

Many thanks,
Stefan

rpelz · Post by **rpelz** » Sat Jun 17, 2006 5:17 pm

How about using summated scales of the 1st-order LV as indicators for the 2nd-order LV (usually employed in regression or ANOVA anlysis)?

forum.smartpls.com

Second order factor

Second order factor

An alternative way for second order factors

Hierarquical structure

How to implement repeated manifest variable method in sPLS?

second order models

second order models

second order factor