During this month's working group telecon, the topic of
confidence intervals for aero data came up. I think I misspoke then
about using a single value for "confidence," and wanted to
clarify my thoughts.
The current draft AIAA standard for simulation data
exchange
<http://dcb.larc.nasa.gov/utils/fltsim/DAVE/AIAA_stds/SimDataExchange_Jan2003.pdf>
recommends carrying, for any data table so desired, a set of
points showing the 95% confidence bound (with the table values
themselves being the nominal value to use if not performing a Monte
Carlo or stress test).
I take this to mean, for a gaussian or normal distribution, to be
about 1.645 standard distributions.
A more common statistic would, I think, to give the +/- 3 sigma
value. This corresponds to 99.86% confidence that the data lie within
the bounds given. Alternatively, we could adopt a single sigma
boundary, which works out to 84.1% confidence interval.
Someone check me on this - my statistics background is a bit
dusty.
I think we should also support a standard with a selectable
gaussian (bell curve) or uniform (flat line) distribution, where the
actual value of the parameter is believed to lie in equal probability
anywhere between the two boundary values.
These two changes to the draft simulation exchange standard
would, I think, move us closer to industry practice for Monte Carlo
simulation results, in my limited experience.
Any opinions welcomed.
--Bruce Jackson
--
Bruce Jackson mailto:[EMAIL PROTECTED] Dynamics and Control
Branch
18C West Taylor Street MS 132 Airborne Systems Competency
NASA Langley Research Center Building 1192C, Room 149
Hampton, VA 23681-0001 scud://N 37 05'31.7" W 76 22'55.1"
http://dcb.larc.nasa.gov/DCBStaff/ebj/ebj.html "There is no try"
18C West Taylor Street MS 132 Airborne Systems Competency
NASA Langley Research Center Building 1192C, Room 149
Hampton, VA 23681-0001 scud://N 37 05'31.7" W 76 22'55.1"
http://dcb.larc.nasa.gov/DCBStaff/ebj/ebj.html "There is no try"
