1 LE SMED LE SMED Outil d’amélioration Outil d’amélioration Silvia OLIVIERI Décembre 2009
Dec 21, 2015
33
The Number of O-Ring Failures Per Launch
0
0.5
1
1.5
2
2.5
3
3.5
30 40 50 60 70 80 90
Temperature, F
Nu
mb
er
44
Number of Failed O-Rings Per Launch Vs. Temperature; No zeros
y = -0.0254x + 3.0465
R2 = 0.0693
0
0.5
1
1.5
2
2.5
3
3.5
0 10 20 30 40 50 60 70 80
Temperature, F
Nu
mb
er
Never Throw Away Data
66
Number of O-Ring Failures Per Launch Vs. Temperature
y = -0.0608x + 4.675
R2 = 0.321
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
0 10 20 30 40 50 60 70 80 90
Temperature, F
Nu
mb
er
Which is the greater sin, (1) throwing away data or running a biased regression?
88
Two Regression ApproachesTwo Regression Approaches Probability Models: Qualitative dependentProbability Models: Qualitative dependent
Linear probability modelLinear probability model Non-linearNon-linear
ProbitProbit logitlogit
Number Models: Quantitative dependentNumber Models: Quantitative dependent OLS, biasedOLS, biased Tobit: extension of probitTobit: extension of probit Count modelsCount models
PoissonPoisson
99
Probability ModelsProbability Models
One or more failures per launch coded as One or more failures per launch coded as 11
Zero failures per launch coded as 0Zero failures per launch coded as 0
1010
Low Temperature: 6 launches with failures out of 12 cases
High Temperature: 1 launch with failure out of12 cases
5.012/6ˆ Lp
083.012/1ˆ HP
53.2165.0/)083.05.0(
]}12/)ˆ1(ˆ[]12/)ˆ1(ˆ{[ˆ
ˆ/)]()ˆˆ[(
0.:
0.,:0
z
pppp
ppppz
pepippH
pepippH
HHLL
HLHL
HLHLA
HLHL
1111
Density Function for the Standardized Normal Variate
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
-5 -4 -3 -2 -1 0 1 2 3 4 5
Standard Deviations
Den
sity
2]1/)0[(2/1*]2/1[)( zezf
1.645-34.7
5%
1212
k n prob0 12 0.1121571 12 0.2691762 12 0.2960943 12 0.1973964 12 0.0888285 12 0.028425 0.0284249836 12 0.006632 0.0350574797 12 0.001137 0.0361944788 12 0.000142 0.0363366039 12 1.26E-05 0.036349236
10 12 7.58E-0711 12 2.76E-0812 12 4.59E-10
H0: p(low temp) = p(high temp) Binomial Prob(k≥5) in 12 Trials, Given p = 2/12
Power 10
1313
The Probability that O-Rings Fail In a Low Temperature Launch Given Probability of Failure At High Temperature =1/6
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0 1 2 3 4 5 6 7 8 9 10 11 12
Number of Launches with Failures
Pro
bab
ility
Power 10
Probability 5 or more fail = 0.036
1414
O-Ring Failures?: Yes or No Vs. Temerature
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 10 20 30 40 50 60 70 80 90
Temperature
Pro
ba
bili
ty
Ex Post: the event either happens (code 1) or does not (code 2)
1515
Probabilty of One Or More O-Rings Failing Per Launch Vs. Temperature
y = -0.0367x + 2.8583
R2 = 0.3254
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
0 10 20 30 40 50 60 70 80 90
Temperature
Pro
ba
bili
ty
Bernoulli
OLS: LPM
Linear (Bernoulli)
LinearApproximation
Ex Ante: what is the probability the event will happen?
1616
Probability Model of O-Ring Failure Per Launch, Yes-No Vs. Temerature
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
20 30 40 50 60 70 80 90
Temperature, F
Pro
ba
bili
ty
Bern
Logit
2121
Who Me Worry?Who Me Worry?The Culture at NASA: getting away with The Culture at NASA: getting away with
problems that should be fixedproblems that should be fixedO-ringsO-ringsFoamFoam
NASA may not have been so worried NASA may not have been so worried about o-ring failures. They had about o-ring failures. They had experienced successful returns of the experienced successful returns of the shuttle with as many as three o-ring shuttle with as many as three o-ring failuresfailures
2323
The Number of O-Ring Failures Per Launch Vs. Temperature
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
0 10 20 30 40 50 60 70 80 90
Temperature, F
Nu
mb
er
o-rings
OLS
Tobit
Linear (OLS)
2 obs.
1 obs.
2424
Number of Failed O-Rings Per Launch Vs. Temperature
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
0 10 20 30 40 50 60 70 80 90
Temperature
Nu
mb
er
O-Rings
OLS
Poisson
Linear (OLS)
2525
Tobit with Dummy Variable for Three Observations Vs. Temperature
0
0.5
1
1.5
2
2.5
3
3.5
0 10 20 30 40 50 60 70 80 90
Temperature, F
Nu
mb
er
O-Rings
Tobit
Tobit with Dummy
2727
Check ListCheck ListDataDataPlotsPlots
7 launches with failures7 launches with failures24 launches24 launches
Linear probability model, logit or probitLinear probability model, logit or probit# of O-Ring failures per launch, OLS, Tobit, Poisson# of O-Ring failures per launch, OLS, Tobit, Poisson
Estimation Results, LabeledEstimation Results, LabeledGoodness of fitGoodness of fitSignificance: t-stat, F-statSignificance: t-stat, F-stat
2929
Probabilty of One Or More O-Rings Failing Per Launch Vs. Temperature
y = -0.0367x + 2.8583
R2 = 0.3254
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
0 10 20 30 40 50 60 70 80 90
Temperature
Pro
ba
bili
ty
Bernoulli
OLS: LPM
Linear (Bernoulli)
LinearApproximation