Hi everyone,
I am currently examining results for a measurement model and following structural equation model using SMARTpls.
When I did the CFA (measurement model) aspect of analysis, I examined 3 facets within an overarching construct all of which have been previously validated and concluded as reliable in various studies. However, when examining the factor loadings there was one indicator/item in each facet that did not meet the loading cut off of .7. They certainly weren't horrible as they ranged from .579-.629-.691.
The validity and reliability tests for this specific study (during the measurement model analysis) were also satisfied and there were no concerns.
Does anyone have any references stating that these items with loadings below .7 can be retained, or does anyone have any feedback and/or advice?
Many thanks,
James
I apologize, I do not have information because the subject is far from my specialty
@James Phillip White
As far as factor loadings are concerned, an observed item should load with loading equal or greater than .70 as it accounts for average variance extracted (AVE = Sum of Loadings^/no of observed items) that should be above 50% i.e., .50. As you stated your case that apart from one item in each latent factor loading below .70, others are loading well enough and meeting the criteria of AVE, then you can go ahead with items with low loading if they are necessary for the conceptuality else you can drop them as well.
Hello James,
There's no immutable threshold for what a variable-factor loading must be; at best, there are guidelines which can be helpful. Among the PLS-SEM folks, the 0.7 threshold is frequently used as an approximate cut-off. As Imran Anwar 's answer correctly notes, that's likely because if all loadings are 0.71 or better, then average variance accounted for will equal or exceed 50%. That's a guideline, not a commandment!
I agree with you that values such as .58, .63, and .69 are not "awful." If these variables are useful in the identification of the factor (and theoretically), then keep them. Consider that one possible consequence of jettisoning variables too freely for not meeting the your chosen threshold is that you may qualitatively alter what the factor actually represents. In exploratory factor analysis, data analysts would almost always call loadings of .58 or higher salient for the model.
Good luck with your work.
As for as factor loading concern 0.7 is good if the goodness of fit, validity and reliability are satisfied. When the observed items are less, you may have good validity and reliability report with this factor 0.7 factor loadings but if you have more number of observed items the factor loading is better to be more than 0.7.
Check this paper, it will try to answer your maximum questions that may arise while using smartPLS (reflective measurements, formative measurements, single-factor constructs, Higher-order constructs, path difference
https://www.researchgate.net/publication/350310378
- Badges
- Science topic
- Similar topics
- Education