[ | Next | Previous | Up ]

Low "n" - low risk field test sampling plan needed

From: jbent@execpc.com
Date: 02 Feb 2000
Time: 16:08:46

Comments

We're interested in knowing if a proposed design will improve customer satisfaction for certain types of complaints (type-x). We'd like to minimize sample number by targeting specific groups of consumers (as opposed to random dispersement) because of the extremely low incident rate (approx. 0.1% of our installed base of ~50 million). Would a Baysian plan be appropriate?

Objective #1: Is the new design better than the old one in reducing type-x complaints?

Proposed Scheme: As type-x problems occur, offer consumers a new unit at no cost in exchange for information on product performance. Half of the target group would receive the new design, the other half would get the current design. Questions: 1. How many respondants do we need from each group? 2. Assuming the results will be different between the two groups, how do we know if it's statistically significant?

Objective #2: Confirm lab testing that shows new design does not introduce any new malfunctions that cause consumer dissatifaction.

Proposed Scheme: Randomly disperse a larger sample group and compare complaint data to current design.

Objective: Same as above.

Question: Same as above.


Last changed: November 20, 2007