Systematically Missing Values in Customer Satisfaction Study

[unterreitmeier]

Hello,

How would you best treat systematically missing data in customer satisfaction surveys (CSS) with PLS? For example, there are some dimensions of customer satisfaction (call center, reaction to complaints, etc) that only a small percentage of the respondents has experienced. Other areas (e.g. prices, quality) can be answered by almost every customer. How can you treat data that is systematically missing? Replacing these by means/EM/etc. would not be an appropriate solution in my opinion.

Is there any possibility to build a model for missing data, for example include a binary moderator variable if data is missing or not? Or can you only look at partial models (e.g. comparing customers who made this experience vs. those who did not make the experience, but this can result in small case numbers)?

Thanks,
Andreas