Combination of mediation and second order factor

139 views
Skip to first unread message

Christina Freise

unread,
Mar 5, 2014, 1:35:48 AM3/5/14
to lav...@googlegroups.com
Hello everybody,

after overcoming the first obstacles in R I'm wondering if my model is possible as I thought or if I should rethink.
Could I combine a mediation effect with a second order factor?

In my model brand loyalty is splitted in two separate factors.

Thanks in advance!
Christina 



yrosseel

unread,
Mar 5, 2014, 9:45:30 AM3/5/14
to lav...@googlegroups.com
On 03/05/2014 07:35 AM, Christina Freise wrote:
> Hello everybody,
>
> after overcoming the first obstacles in R I'm wondering if my model is
> possible as I thought or if I should rethink.

Maybe it is just me, but your picture did not come through. Could you
convert to it to, say, pdf?

Yves.

Christina Freise

unread,
Mar 5, 2014, 10:18:49 AM3/5/14
to lav...@googlegroups.com
Thanks for the information, Yves.
Here my theoretical model as pdf...

Christina

SEM model.pdf

yrosseel

unread,
Mar 5, 2014, 11:26:15 AM3/5/14
to lav...@googlegroups.com
On 03/05/2014 04:18 PM, Christina Freise wrote:
> Thanks for the information, Yves.
> Here my theoretical model as pdf...

Looks good to me.

Yves.

Christina Freise

unread,
Mar 5, 2014, 2:08:07 PM3/5/14
to lav...@googlegroups.com

Dear Yves,


This was my first try building that model in lavaan.

Why is the second-order- factor "Com" of my first-order "Loy" set to 1?

I would have expected two values describing "Loy".


Thanks Christina



# Regression 

    Tru ~ Exp + Sat

    Sat ~ Exp

    Loy ~ Exp

    Loy ~ Sat

         Loy ~ Tru 



 Loy =~
    Com     1.000
    Rec     0.927    0.087   10.703    0.000

Regressions:
  Tru ~
    Exp         0.124    0.033    3.803    0.000
    Sat          0.769    0.114    6.761    0.000
  Sat ~
    Exp         0.293    0.034    8.651    0.000
  Loy ~
    Exp         0.108    0.033    3.225    0.001
    Sat          0.610    0.180    3.377    0.001
    Tru          0.316    0.134    2.350    0.019

Terrence Jorgensen

unread,
Mar 5, 2014, 3:52:18 PM3/5/14
to lav...@googlegroups.com

Why is the second-order- factor "Com" of my first-order "Loy" set to 1?

I would have expected two values describing "Loy".


Thanks Christina



# Regression 

    Tru ~ Exp + Sat

    Sat ~ Exp

    Loy ~ Exp

    Loy ~ Sat

         Loy ~ Tru 



 Loy =~
    Com     1.000
    Rec     0.927    0.087   10.703    0.000


It's hard to tell exactly what you are modeling without seeing the full syntax.  The output shows that Loy is defined by Com and Rec, yet you describe Com as a second-order factor and Loy as a first-order factor, which might imply the opposite (i.e., if Com is a latent construct that is an indicator of Loy, then Loy must be a higher-order factor).

As far as the factor loading = 1 is concerned, the default setting in sem() is to identify the construct by fixing the first loading to 1, but you can override that in the syntax (e.g., Loy =~ NA*Com) or in the call to sem() by choosing a different method of identification: sem(model, data, std.lv = TRUE).  Again, without seeing your full syntax, it's hard to see the problem.

Terry

Christina Freise

unread,
Mar 5, 2014, 4:08:23 PM3/5/14
to lav...@googlegroups.com
Hi Terry,

thanks a lot for the information so far.

Here's my full syntax:

 

my.model <- '

   # Latente Variablen

     Tru =~ Var29+ Var30+ Var31+ Var32+ Var33+ Var34 

     Exp =~ Var9+ Var10+ Var11+ Var12+ Var15+ Var16+ Var17

     Sat =~ Var18+ Var19+ Var20+ Var21+ Var22+ Var23+ Var24

     Com =~ Var35+ Var36+ Var37+ Var40  

     Rec =~ Var42+ Var43+ Var44+ Var45

     Loy =~ Com + Rec

'


Terrence Jorgensen

unread,
Mar 9, 2014, 1:15:04 PM3/9/14
to lav...@googlegroups.com

I don't see a call to lavaan in your R syntax, so I'm still not sure how you identify your constructs (e.g., std.lv = TRUE?).  But I would think you might have trouble identifying a second-order construct with only 2 first-order constructs.  You don't gain any information by doing this, and it might be necessary to identify Loy either by fixing the latent variance of Loy to 1, and constraining its loadings on Com and Rec to equality.  Or to identify Loy by fixing by loadings to 1 and freely estimating its variance, which would be identical to the estimated (residual) covariance between Com and Rec if you were to exclude Loy from the model.

Terry

Christina Freise

unread,
Mar 9, 2014, 4:22:30 PM3/9/14
to lav...@googlegroups.com
Hi Terry,

thanks for your answer. I'll rethink that.

Best regards
Christina
Reply all
Reply to author
Forward
0 new messages