The two-step approach to model 2nd order constructs

[sasaluoluo]

Hi,

I am using smartPLS 3 to test a model with 2 second-order constructs and several first-order constructs. A simple version of my model is like this:

1) Construct S1: 2nd order, reflected by three 1st order constructs A1, B1, C1, all of which are reflective.
2) Construct S2: 2nd order, reflected by three 1st order construct A2, B2, C2, all of which are reflective.
3) Construct S3 and S4, 1st order reflective.
4) S1 -> S4, S2 -> S4, S3 -> S4.

I want to use the two-step approach to model the second-order constructs as my referenced papers did. However, in those papers, all their constructs are second-order, different from mine. So I don't know which of the following two methods is appropriate to evaluate the measurement model:

Method 1:
Step 1,include all 1st order constructs (A1, B1, C1, A2, B2, C2, S2, S3), link them to one another, avoid loops, and perform a CFA to test the reliability of the indicators.
Step 2, use factor scores of A1, B1, C1 as indicators for S1, those of A2, B2, C2 for S2, link S1, S2, S3, S4 to one another, perform another CFA to test the reliability of S1 and S2 (since that of S3 and S4 have already been tested in step 1), and discriminant and convergent validity of S1, S2, S3, S4.

Method 2:
Step1, include only 1st order constructs under the 2nd order constructs, i.e. A1, B1, C1, A2, B2, C2, link them to one another and perform a CFA to test the reliability.
step2, use factor scores of 1st order constructs as indicators for S1 and S2, and link S1, S2, S3, S4 to one another to perform another CFA, test reliability and validity.

The difference is whether to include 1st order constructs S3 and S4 in the first step. Can anybody offer some advice and hopefully some relevant articles for me?

Thanks in advance.

[Hengkov]

Hi,

Methods 1 and 2 will give the same results for the 1st order, only the second method is added validity and reliability results for 2nd order constructs.

Regards,

[semia.barouni]

hi
i'm working on a complex model with several second order latent constructs. i know that there's three methods to modelize latent constructs in SMARTPLS :
-repeated indicator
-two-step approach
-hybrid approach

i'm looking for technical note (screen shots) which illustrate step by step how to apply thoose méthod on SMARTPLS.

kind regards

[MrcJnk]

Dear Ms. Barouni,

you can find path model depictions for each approach in Becker et al. (2012, pp. 380-382). For the two stage approach there is also the possibility to combine it with the repeated indicator approach in the first step (Ringle et al., 2012, pp. 6-7).

Best

MJ

Becker, J.-M., Klein, K., & Wetzels, M. (2012). Hierarchical Latent Variable Models in PLS-SEM: Guidelines for Using Reflective-Formative Type Models. Long Range Planning, 45(5–6), 359–394.

Ringle, C. M., Sarstedt, M., & Straub, D. W. (2012). A critical look at the use of PLS-SEM in MIS Quarterly. MIS Quarterly, 36(1), iii–xiv.

[MRodrigues]

Hi everyone,

Can you explain me how to apply two-stage approach in a formative third-model?
You can find the model in attachment.

Thank you for your help.

[jmbecker]

You will then essentially have a three-stage approach and not a two-stage approach. You need to calculate the latent variable scores for each level of your constructs.

[MRodrigues]

Thank you very much for your help!

For purposes of model identification, I don't need to ensure the formative construct has at least two structural paths to reflective constructs, right?

Regards,
Mariana

[jmbecker]

In PLS you always only need at least one path.

[valeperren]

Hi Mariana

I am currently working on a similar model measuring tourist destination image. It is also a third order construct with formative indicators.
I tried to draw the model on PLS, but PLS doesn't accept my model as the second order doesn't have it's own indicators (because their indicators are latent variables). How did you manage to define a latent variable as indicator for another latent variable?

Could you offer me some advise ?
Kind regards
Valerie

[MRodrigues]

Hi Valerie,

Firstly, you use "two-step approach", in our case, three-step approach. In my model, this approach was performed in three stages: thus, in a first stage, the construct scores of the first-order constructs (power through relationships, political functioning, shaping ideologies, satisfaction of needs and opportunities for involvement, support and emotional connection with peers, support and emotional connection in the community, sense of belonging, opportunities for influence, activism, civic engagement, and online civic engagement) were estimated in a model without the second-order constructs present, which, in the second stage, served as manifest variables in the second-order latent variables (cognitive empowerment, relational empowerment, and behavioral empowerment) measurement model in a separate analysis. The cognitive empowerment, relational empowerment, behavioral empowerment and policy control (representing emotional empowerment) were then used as manifest variables in the third-order latent variable (psychological empowerment). This procedure was conducted using factor weighting scheme as a PLS algorithm with 300 iterations (Tenenhaus, Vinzi, Chatelin, & Lauro, 2005). The estimation of the parameters of the outer and inner model was performed using mode B (formative) and factor weighting scheme as a PLS algorithm with 300 iterations. Additionally, it was used nonparametric bootstrapping procedure (Tenenhaus et al., 2005) with 5000 replications and construct level change option to obtain the standard errors and calculate t statistics.

[rajachilies]

Dear All,
I am using smart-PLS for the first time. I would like to clarify few doubts in this forum

1) My survey instruments are adopted scale which has first order and second order constructs with observed/manifest variables as reflective. how to identify whether my first order to second order construct relationship is reflective or formative?