Downloaded from https://www.everyspec.com AMCP 70196 TABLE 2-21 MCoet'd) (D) Modified and wawsoet controversill) AMtdsi Calculations *n Allocetlonl #1 #2 #3 #4 #1 #6 #7 F approx. Allkeuimon SS P F % e-slgnw, a-if. % t 70.0863 1 70.1 190 12.3 v 33.500 2 16.8 45.4 0.55 h 0.0633 1 0.0833 0.226 12.3 03 all interactions 2.534 7 0.3001 - - - Now let us analyze the results of Example shown In Table 2-21(D). It declares, more No. 13: thtan ever, that temperature and vibration ame most important, and that humidity is 1. One of the first things to note is that negligible.
Such remanlpulations must be the low v for the reference si (PrA - 2) treated with caution; consult a competent causes very poor ability to distinguish very statistician before basing any Important high from ordinary ratios of the as decisions on them. 2. Thr; temperature effect Is undoubted- 6. The basic data themselves are very ly important (for the 2 temperatures tested) coarse. It is unlikely that they come from a the vibration is likely to be important; but s-normal distribution. A lognormal would none of the other effects or Interactions is perhaps be more appropriate. The reference likely to be Important, population standard deviation (from Con- clusion No.
5) Is about 0.61 minIon, By 3. This is a "fixed effects" analysis. The going back now and looking at Table levels are presumed "fixed", not to be a 2-21(B) it is not unreasonAble that the random sample from all possible levels of humidity effect is small. In 3 cells, there the factor. See Refs. 31 and 32 or a com- was no effect, and in the other cells there petent statistician for a fuller discussion of was at most I mission difference. this point. 7. After running such an expensive test, 4. If more tests are to be run, they the small extra cost of several analyses is ought to be on temperature and vibration not unreasonable, The only unreasonable separately.
The separatp tests are easier/ thing would be to place much importance cheaper to run. on results at, say, the 5% or 10% s-signifi- cance level, because these will occur In 5% 5. Since tU' .re was no replication, It was to 10% of the calculations. presumed that there was no 3-way inter- action. From the looks of the results, It 2.8 REGRESSION AND CORRELATION is easy to assume (see conclusion No. 2) ANALYSIS that none of the 2-way interactions is im. portant.
Some (but not all, or even most) Variations in component part values in- statisticians would argue that there is now troduced by manufacturing processes and justification for lumping all SS's for all variations in environmental conditions caus. interactions tojether and to estimate the changes In circuit or equipment character- reference s as (1.1667 + 0,1667 + 0.0833 istics which may affect reliability (Refs. 2 ' +1 1,1667)(2 + 2 + I + 2) a 0.3691 with 7 and 6), Numerical relationships that relate l,, degrees of freedom. The new analysis is reliability to design variables can be derived L !i ' ,' | ,t,, ! , ,, ,