Good morning AllStatters,
As an independent consultant, I don't have colleagues that I can always turn
to in order to sense check my way of thinking hence this email. I would
really appreciate some feedback on the following issue I have with a client
by 10am Monday (UK time) when I have to give them a call.
The client is a food manufacturer who wants to achieve better control of
their sterilisation process. They measure the effectiveness of the
steriliser at the start and end of each production run (it is not practical
to measure during a run). Initially, my work was to recalculate the target
value for these measurements to ensure their targets were better correlated
with established food safety standards. Once I had the targets, I
recalculated the control limits using standard SPC methods. Crucially, I
based these limits on the known within-run variance of the measures. I knew
from past data that these limits would almost certainly be breached since
the observed between-run variance was about 10 times the within-run
variance. The point of doing this was to demonstrate to the client the
necessity to carry out improvements to their sterilisation process since by
using the within-run variance the process would be constantly out of
control. I warned my client at the time that this would happen and at the
time they appeared to accept the necessity to improve the process.
The client has now come back and asked why I didn't use the between-run
variance. They claim (correctly) that if the limits had been based on this,
the control chart would show a process in control. I feel they are trying
to solve the problem by rewriting the rules rather than tackle what I see as
an unacceptably high ratio of between-run to within-run variance which tells
me there are underlying problems with the process. Before, I speak to them
on Monday, I would like to ask the following questions.
1. What ratio of between-run to within-run variance have you observed
in processes that are deemed to be in control even though this ratio might
be statistically significant?
2. Are there situations when basing control limits on between-run
variance makes sense even though the within-run variance is significantly
less?
I have my own views on these questions but I really do need some other
viewpoints in order to check that I am approaching this problem correctly.
Regards
Nigel Marriott
Chartered Statistician
<http://www.marriott-stats.com/> www.marriott-stats.com
Ground Floor, 21 Marlborough Buildings, Bath BA1 2LY, United Kingdom
Tel (mobile) +44 (0)773 4069997
Tel (office) +44 (0)1225 489033
Marriott Statistical Consulting Limited, Company No. 5577275, VAT No.
883304029
Registered in England, Registered Office - Equity House, 4-6 School Road,
Tilehurst, Reading, RG31 5AL
|