A Text Analytics Question

How many survey questions are needed?

Yesterday’s Q&A with BRB prompted an interesting discussions on one of the research related LinkedIn groups I belong to. As a result today I have a semi hypothetical question I’d like to put out to the marketing research community.

Let’s assume that a customer satisfaction survey has these four quite common questions:

1. What is your likelihood to recommend the brand/product?
   (10 point scale)
2. What is your likelihood to try the brand/product again?
   (10 point scale)
3. What is your overall satisfaction with the brand/product?
   (10 point scale)
4. Please explain why you are satisfied/dissatisfied with the brand/product?
   (text comment)

If you could predict the average of questions 1-3 (with say 80% accuracy), by analyzing just question Q4, would you bother asking Q1-Q3?

What if, together with question 4 and any single one of the other likert scale questions, you could predict the other two questions with >90% accuracy, would you still ask the other two questions?

@TomHCAnderson
@OdinText