Build a latent formativconstruct with demographic indicators

[mikethegerman]

Hello,

is it possible to build a latent formative construct out of indicators such as age, gender, marital status, education and income?
I would like to see if demographic charactreristics such as the above are influence my dependent reflective constrcut Purchase intention. or is it a better idea to test demographics with SPSS in an isolated model?

I would like to integrate demographic data into my structural model. Besides the demographics iam estimating the influence of only psychographic constructs/indicators on purchase intention!

Thanks in advancee

[christian.nitzl]

Hey Michael,

It seems to me that you want to work with control variables. The easiest way for modeling control variables (age, gender etc.) in PLS is to connect them as single item constructs with your focal variable ‘purchase intention’ separately. A good example for doing that you can find in following paper:

Liang, H./Saraf, N./Hu, Q./Xue, Y. (2007): Assimilation of Enterprise Systems: The Effect of Institutional Pressures and the Mediating Role of Top Management, in: Management Information System Quarterly, Vol. 31, No. 1, S. 59-87.

Best,

Christian

[mikethegerman]

Thank you very much Christian!

when i enter my demographic data as 5 single control variables and directly connect them to purhcase intention, the results indicate that they do not have big influence (all path coefficients are between -0.041 and 0.072, neighter is any of them significant). That is good, since theory also says htat demographics are only of limited value when predicitng purchase intention, and its rather the psychographic variables tht have explanatory power.

I still have a few questions:

1.Can i test the whole model with all including all latent constructs at once or do i have to run the pls algortihm 5 times (for all single demographic LV) ? My smple size is 200
2. What exactly is the difference between entering the demographic indicators as single LV, meaning as single control variables and entering them alltogether combined in one LV?
3. All other LV and path coefficients changed only slightly but R² of my dependent variable increased! so eventhough demographics have almost no influence and are not significant they help to improve the overall GoF of the model...is this possible?

Thank you in advance for your help.

Regards
Michi

[christian.nitzl]

Hey Michi,

You’re welcome!

ad 1) Your sample size of 200 should be enough to test your model with all your control variables at once. I know this was a problem for Liang et al. but as far as I know there sample size was much smaller than 200.

ad 2) I would not test the influence of the control variable as a combined formativ variable. The combined variable will give you only an average weighted influence of all your control variables at once. Therefore you get an information loss.

ad 3) Sure, this is possible. If you want you can check also if your control variable have an important effect size on your focal variable. How to calculate the effect size you can find in my downloadable working paper on page 34-35 (http://www.ibl-unihh.de/ap21_Stand_Juni2010.pdf).

Greetings,

Christian

[ruchi]

Dear Sir

If i have 4 independent variables(all reflective) and 1 dependent variable (all reflective) and 1 variable Age.

I have connected all my variables - 4 independent variables(IV)
and Age to my dependent variable (DV) . Then what does result signify

IV 1 ------------->DV
IV 2 ------------->DV
IV 3------------->DV
I V 4------------->DV
AGE ------------->DV

All this is in one model connected . Then what does the path coeffcients of each signify?
Then in this case whar does the path coefficient of Age---->DV tells? is it a effect of Age on DV when all variables are zero
(like simple regression result of two variables)
only or it is effect of Age on DV keeping all other IV1,IV2,IV3, IV 4 constant.(like multiple regression result)

Now in this case, what am i controlling.

Please provide me some insights on this.

Thanks
Ruchi

[ruchi]

Dear Prof Christian

In my PLS model I have 4 independent variables and 1 dependent variable.
They are kinked as
IV1 --->D
IV 2--->D
IV3---->D
IV4---->D

IV3-->IV1
IV3-->IV2

Questions--

1) Then in case, If i want to see the effect of age , then should i connect my latent construct( with single item as Age (continuous variable)) to all the variables behaving as dependent variables(like - IV1, IV2 and D) or Only to D.

2 )In case , I connect to all my dependent variables(like - IV1, IV2 and D), then what is the interpretation of the link Age-->IV1;
Age-->IV2 ; Age---> D

My overall model looks like
Age-->IV1
Age-->IV2
Age---> D
IV1 --->D
IV 2--->D
IV3---->D
IV4---->D
IV3-->IV1
IV3-->IV2

Does it tell the direct effect of Age on these variables keeping all other variables constant. Is the direct effect same as the term "Main effect"

3) In case I connect to only variable D i..e Age-->D


My overall model in this case looks like

Age---> D
IV1 --->D
IV 2--->D
IV3---->D
IV4---->D
IV3-->IV1
IV3-->IV2

Please guide me for this issue. I am bit confused

Thanks
Warm Regards
Ruchi