Dear allstat,
I am currently reviewing a simulation model for a client, representing the
behaviour of a large set of components over time. A component may suffer a
defect, which is then immediately repaired. From time to time, components
are replaced for a variety of reasons. An equation for the expected number
of defects per year has been fitted (least squares regression) using data
from the last 9 years. Component age is one of the most important
explanatory variables.
Due to the way the data have been collected, some of the data on component
ages are missing. The age of every component currently in service is known.
In addition, when a defect is repaired, the age of the component has been
measured and recorded. However, in a case where a component has been
replaced less than 9 years ago and where there were no defects prior to
replacement, the age of the original component is not known.
The proportion of components for which the age is unknown ranges from about
1% (last year) to about 10% (9 years ago). These missing data points have
been excluded from the analysis to fit the equation (however, the equation
has been normalised so that the total expected number of defects is correct
for the whole population). The client believes that the missing ages may be
younger than average, and hence some bias may have been introduced.
Any suggestions on ways to approach this problem are gratefully received.
Regards,
Crispin.
Dr Crispin Allard
Senior Consultant
QinetiQ Consulting
Cody Technology Park
Farnborough GU14 0LX
Tel: 01252 395312
Fax: 01252 394109
Mobile: 07813 832526
email: [log in to unmask]
The Information contained in this E-Mail and any subsequent correspondence
is private and is intended solely for the intended recipient(s).
For those other than the recipient any disclosure, copying, distribution,
or any action taken or omitted to be taken in reliance on such information
is prohibited and may be unlawful.
Emails and other electronic communication with QinetiQ may be monitored.
Calls to our Customer Contact Centre may be recorded for quality control,
regulatory and monitoring purposes.
|