Hi everybody,
I was running a model in SmartPLS 3.
1. 40 indicators' outer loadings (out of 101 indicators) are between 0.4-0.7. Not all of them are too low.
2. AVE results for 5 variables (out of 11) are below 0.5.
3. Fornell-Larcker criterion for two correlations is not established.
4. All HTMT results are below 0.85.
According to Hair (2011): Generally, indicators with outer loadings between 0.40 and 0.70 should be considered for removal from the scale only when deleting the indicator leads to an increase in the composite reliability (or the average variance extracted;) above the suggested threshold value.
Here are the results after eliminating 13 indicators with the lowest outer loadings:
1. All the AVE becomes more than 0.5.
2. Fornell-Larcker criterion isn't still established for two correlations.
4. On the other hand, the results for HTMT changes, and the HTMT for one correlation is upper than 0.9.
I am aware that the elimination of items purely on statistical grounds can have adverse consequences for the construct measures’ content validity (e.g., Hair et al. 2014). Therefore, researchers should carefully scrutinize the scales (either based on prior research results or on those from a pretest in case of the newly developed measures) and determine whether all the construct domain facets have been captured. At least two expert coders should conduct this judgment independently to ensure a high degree of objectivity (Diamantopoulos et al. 2012).
I do not know about the theoretical support for eliminating some indicators. I would be really thankful if you'd help me what I can do in this situation.
Many thanks
Therefore, researchers should carefully scrutinize the scales (either based on prior research results or on those from a pretest in case of the newly developed measures) and determine whether all the construct domain facets have been captured.
Yes, this is an important point because statistical grounds alone may not be sufficient. You can justify retaining constructs based on theoretical grounds even though it is not statistically significant.
The theoretical grounds will depend on your research questions and the literature review of other people's study and what their findings are like. This is not provided above.
Therefore, researchers should carefully scrutinize the scales (either based on prior research results or on those from a pretest in case of the newly developed measures) and determine whether all the construct domain facets have been captured.
Yes, this is an important point because statistical grounds alone may not be sufficient. You can justify retaining constructs based on theoretical grounds even though it is not statistically significant.
The theoretical grounds will depend on your research questions and the literature review of other people's study and what their findings are like. This is not provided above.
Pardon if I a incorrect but according to my opinion , you should use other model fit indices such as RAMSEA, SRMR, CFI, NFI etc along with CVE to check whether the your instrument has enough evidence convergent and discriminant validity.
You should consider the evaluation of the dimensionality of each construction before making the general model, it may present some low loads that you may consider withdrawing considering that the measures have not yet been adapted to your context and your specific sample.
Thanks, Cristian Ramos-Vera But even eliminating some low loads need a highly supported theoretical basis.
Thanks Shamas Arshad . Since SmartPLS is a covariance-based model, I can check the fir Indexes like GoF.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling