Hi everyone.
There are 5 latent variables forming a 2nd order formative latent construct (L6). The problem is, one of the 1st order constructs (L1) is reflective and the rest are formative. SmartPLS requires that all indicator level variables be connected to the higher order construct but naturally I can't connect some indicators as formative and the rest as reflective. I have decided to impute L1 in order to go around this problem. Does anyone know of a better way? I appreciate your attention.
@Rouhollah If you assume a formative relationship between the construct and its indicators, then the second-order factor would not be a factor anymore, Rather, you would treat it as an observed variable that you are using to explain other variables. So, it would be a regression model rather than a measurement model at the end. You can take a look at Dr. Kenny's website and Wilson and Henseler paper for more details.
http://davidakenny.net/cm/mvar.htm
https://www.researchgate.net/publication/241562978_Modeling_Reflective_Higher-Order_Constructs_using_Three_Approaches_with_PLS_Path_Modeling_A_Monte_Carlo_Comparison
Dear Imran Anwar , I appreciate your response.
In my case the 2nd order variable is caused by a combination of reflective and formative 1st order variables and thus cannot be modelled purely as a regression model which brings me to my predicament: Is it even possible to treat L6 as a single composite construct without imputing L1?
Best regards.
- Badges
- Science method
- Similar topics
- Methodology
- Research Methodology
- Qualitative Research