sample size in determining a and b-basis allowables

huangyhg

sample size in determining a and b-basis allowables
i seem to remember from my school days that there were different smaple sizes required if one was determining a vs. b-basis allowables.  however, the latest data that i see indicates that a minimum sample size is 100 (assuming a standard distributions) from a minimum number of lots.  this is irrespective of whether a or b-basis is desired as they both require the same confidence level (i.e. 95%).
can someone tell me if this was different at one time, and if so, what the minimum sample size was for each basis.
thanks,
jetmaker.
see mil hdbk 5 or ar-mmpds-01 (chapter 9).  a relevant quote ...
"for example, assume that available data for a relatively new alloy comprise 50 observations of tus
in the specified testing direction. this sample is not considered large enough to determine the distribution
form and reliable estimates of population mean and standard deviation. since only direct computation is
permitted in this instance, determination of t99 and t90 values must be postponed until a larger sample is
available. however, these properties may be considered for presentation on the s basis at the discretion of
the mmpds coordination group, contingent on availability of an acceptable procurement specification for
the material.
if the number of observations increases to 100, this quantity may be adequate to allow determination
of t99 and t90 values, provided data can be described by a pearson type iii (gamma) (subsequently referred
to as simply 鈥淧earson鈥? or weibull distribution. if the distribution cannot be described parametrically, at least
299 observations are required so that computation can proceed without knowledge of the distributional form.
if the above example involved observations of sus instead of tus, the same criteria would apply
for direct computation. however, fsu could be determined by indirect computation with as few as ten paired
observations of sus and tus (representing at least ten lots and three heats), provided ftu has been
established."
this seems to support your thought about 100 samples.
just to clarify the difference between a- and b-basis ...
"a-basis.鈥擳he lower of either a statistically calculated number, or the specification minimum (s-basis).
the statistically calculated number indicates that at least 99 percent of the population of values is expected
to equal or exceed the a-basis mechanical design property, with a confidence of 95 percent.
...
b-basis.鈥擜t least 90 percent of the population of values is expected to equal or exceed the b-basis
mechanical property allowable, with a confidence of 95 percent."
there are no universal "requirements" for sample sizes for basis values (allowables).  the calculations can be performed on samples sizes as small as 3.  the issue is what is acceptable to the organization or agency approving the basis values.  to get metal materail data approved for mil-5/mmpds, a minimum of 100 samples from 10 lots are required (though only for tensile strength, there are smaller sample size requirements for other properties), as quoted above by rp1957.  for composite material data to be fully approved for mil-17, a minimum of 30 samples from 5 batches is required for b-basis values and a minimum of 75 samples from 10 batches for a-basis values.  in reality, the number of material batches in the dataset for calculating an allowable is much more important than the total number of samples, as the basis value is supposed to represent a defined point on the total material population distribution, and you want to be sure to account for batch-to-batch material scatter as well as test variability.  the number of samples for an allowable also can vary with the particular approving agency's (faa, air force, navy, etc.) requirements and whims.  do you have a particular situation for which you are trying to develop a test plan?
regards,
steve