Hi, I am currently working on a research in the family business field. I have 4 latent variables and also included moderator effects. Using SMART PLS, the results say that the r-squared value for both dependent variables are 0,847 and 0,568 respectively. The theory that I've read indicates that r-squared values above 0,3 are great (Other authors are way more demanding and suggest values above 0,6 or 0,75 depending on the field).
Understanding that this value is just one parameter to be analyzed while using PLS SEM, I checked the path coefficients and external loads and came upon values that are relatively low. The independent variables -> dependent variables path coefficients are mostly lower than 0,1 and a lot of the loads are lower than 0,5. The exception are the moderator variables which have path coefficients of 0,975 and -0,475.
Does my model work great at explaining the variance of the dependent variables but has poor strength if analyzed by construct? What problems could I possibly have and how can I solve them ?
Thanks in advance
You need to check the dimensions of your latent variable first. If the reliability analysis is greater then 0.70 for all the latent variable then ok.
Secondly, you need to check the correlation of dimension for each latent variable separately. The correlation for each dimension should be significant and positive.
Thirdly, check the outer loadings, if the dimensional values are showing in red, remove them.
It will enhance the significance of your latent variable with your dependent variable as well the correct value of r-square.@ Benjamin Aguilera
Hi! If there are low loadings especially below 0.5 then removing them can improve convergent validity and can have positive impact on the path coefficient. The R2 depends on the field of Discipline. If certain R2 value is justified by supportive research evidences should be fine. Thanks.
Dear Benjamin
a couple of things here:
- as far as I understand it, PLS is principal-component based. SEM is a "true" factor analysis. If applicable, turn your attention towards SEM using R or Mplus.
- explained variance of R² = .568 (=57 %), resp. R² = .847 (=85 %) is extremely high in the Social Sciences
- what do the fit-indices (RMSEA, CFI, TLI etc.) tell you about your specifications? Not sure if they show with PLS-SEM, though.
- I would not consider regression coeficients below .1.
- yes, do check dimensionality of your latent variables as well.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling