Dear professors,
I want to make a model which measures the impact on loyalty towards a brand.
I have a formative model with 4 latent variables, "image", "product", "service" and "loyalty". Each of these concist of several measurement variables. Image, Product and service all have a "impact" Path Coefficients to loyalty. How do I determine the the impact of the measurement variables in each of the LV (image, product and service).
Is this by looking at the cross loading? will this show me how much an increase of 1 in "image a." will have on the loyalty LV?
Or should the be calculated by the outer weights and the path Coefficients?
Best regards
Cross loading or Outer weights impact on latent variable
- Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi Wilke,
This kind of interpretation is not usual.
Usually we assess the measurement model in the first step and the structural model in the second step as it is recommended by:
Anderson, J. C. & Gerbing, D. W. (1988). Structural Equation Modeling in Practice: A Review and Recommended Two-Step Approach. Psychological Bulletin, Vol. 103, No. 3, pp.411-423.
This is done in this way because the constructs are in the hypothetical (or conceptual) level and the indicators are in the operational level. We could have others indicators to measure the same construct (or latent variable).
See (p.13-16):
LOHMÖLLER, Jan-Bernd. Latent Variable Path Modeling With Partial Least Squares. Heidelberger: Physica-Verlag, 1989.
I could not find a way to interpret the results as you want (directly from the output).
Maybe it will be something like this: For each indicator --> outer weight * path coefficient
But the outer weight tell us how much the indicator weights in the score of its LV, but these scores are standardized AFTER the computations that use these weights and BEFORE the structural model is estimated (for this reason is “Partial”).
I hope this help.
Best regards,
Bido
This kind of interpretation is not usual.
Usually we assess the measurement model in the first step and the structural model in the second step as it is recommended by:
Anderson, J. C. & Gerbing, D. W. (1988). Structural Equation Modeling in Practice: A Review and Recommended Two-Step Approach. Psychological Bulletin, Vol. 103, No. 3, pp.411-423.
This is done in this way because the constructs are in the hypothetical (or conceptual) level and the indicators are in the operational level. We could have others indicators to measure the same construct (or latent variable).
See (p.13-16):
LOHMÖLLER, Jan-Bernd. Latent Variable Path Modeling With Partial Least Squares. Heidelberger: Physica-Verlag, 1989.
I could not find a way to interpret the results as you want (directly from the output).
Maybe it will be something like this: For each indicator --> outer weight * path coefficient
But the outer weight tell us how much the indicator weights in the score of its LV, but these scores are standardized AFTER the computations that use these weights and BEFORE the structural model is estimated (for this reason is “Partial”).
I hope this help.
Best regards,
Bido
- Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi Wilke,
I have thought a little more and:
- If the exogenous LV is formative (mode B), we could interpret (outer weight * path coefficient) as a beta in a multiple regression where this indicator were a predictor of the endogenous LV. Results comparable with canonical correlation if the model has two formative LV.
- If the exogenous LV is reflective, we haven't a way of interpretation directly, because the result will be between OLS regression and PLS-R (like a PLS-R with just one component).
Best regards,
Bido
I have thought a little more and:
- If the exogenous LV is formative (mode B), we could interpret (outer weight * path coefficient) as a beta in a multiple regression where this indicator were a predictor of the endogenous LV. Results comparable with canonical correlation if the model has two formative LV.
- If the exogenous LV is reflective, we haven't a way of interpretation directly, because the result will be between OLS regression and PLS-R (like a PLS-R with just one component).
Best regards,
Bido