Hi
I am investigating customer satisfaction for a large complex organisation within the framework of a survey study. Since only a fraction of customers have placed complaints and a slightly larger fraction has been in direct contact with customer service (both subpopulations identified with filter questions) some items/indicators remain unanswered for the wast majority of customers.
I strive for a single coherent model, which do not exclude customers or important items/indicators and catches all important aspects of customer satisfaction. That is, I need to include answers related to customer service and complaints.
What is the most up-to-date PLS approach to this problem - can it be solved using moderator/subgroup PLS? I cannot figure it out, I am new in this field, and to me it seems that all items/indicators by default should be answered by all respondents. Is there a smartPLS approach?
Please provide as many details as possible in your answer and/or direct me to a tutorial or paper on this problem.
Best wishes
Customer satisfaction, filter questions, subpopulations
A possible solution?
Suppose filter questions only have an affect on the number of well defined indicators and do not affect the design of the structural model.
Suppose the filter question has two levels (yes/no) and is always answered, furthermore suppose the i'th of a total of n filter subquestions has levels l_1, ... l_k, if the answer to the filter question was yes.
Construct n new nomial variables such that the i'th variable has levels
no, l_1, ..., l_k
Use the newly constructed variable as indicator.
Suppose the filter question has two levels (yes/no) and is always answered, furthermore suppose the i'th of a total of n filter subquestions has levels l_1, ... l_k, if the answer to the filter question was yes.
Construct n new nomial variables such that the i'th variable has levels
no, l_1, ..., l_k
Use the newly constructed variable as indicator.
Correction and new question
According to the following post and the smartPLS algorithms, the suggested variables above should actually be recoded into dummyvariables.
viewtopic.php?t=479&highlight=multi+group
By the way: Any references to 'SEM models' with ANOVA (since my variables are all nominal/ordinal)
Best wishes
viewtopic.php?t=479&highlight=multi+group
By the way: Any references to 'SEM models' with ANOVA (since my variables are all nominal/ordinal)
Best wishes