CTL.

SC0xSupplyChainAnalytics

KeyConceptsDocument
V1.1

ThisdocumentcontainstheKeyConceptsdocumentsforweek4,lessons1and2within
theSC0xcourse.

Thesearemeanttocomplement,notreplace,thelessonvideosandslides.Theyare
intendedtobereferencesforyoutousegoingforwardandarebasedontheassumption
thatyouhavelearnedtheconceptsandcompletedthepracticeproblems.

ThefirstdraftwascreatedbyDr.AlexisBatemanintheFallof2016.

Thisisadraftofthematerial,sopleasepostanysuggestions,corrections,or
recommendationstotheDiscussionForumunderthetopicthreadKeyConcept
DocumentsImprovements.

Thanks,
ChrisCaplice,EvaPonceandtheSC0xTeachingCommunity
Fall2016v1

v1.1Fall2016|ThisworkislicensedunderaCreativeCommonsAttributionNonCommercialShareAlike4.0InternationalLicense.

Week4:ManagingUncertainty
LearningObjectives

Understandprobabilities,importanceandapplicationindailyoperationsandextreme
circumstances.
Understandandapplydescriptivestatistics.
Understanddifferencebetweencontinuousvs.discreterandomvariabledistributions.
Reviewmajordistributions:Uniform(discreteandcontinuous),Poisson,Normaland
Triangle.
Understandthedifferencebetweendiscretevs.continuousdistributions.
Recognizeandapplyprobabilitymassfunctions(pmf),probabilitydensityfunctions
(pdf),andcumulativedensityfunctions(cdf).

SummaryofLesson
Thefirstlessonreviewstwoveryimportanttopicsinsupplychainmanagement:probabilityand
distributions.Probabilityisanoftenreoccurringthemeinsupplychainmanagementduetothe
commonconditionsofuncertainty.Onagivenday,astoremightsell2unitsofaproduct,on
another,50.Toexplorethis,theprobabilityreviewincludesanoverviewofprobabilitytheory,
probabilitylaws,andpropernotation.Summaryordescriptivestatisticsareshownfor
capturingcentraltendencyandthedispersionofadistribution.Finally,weintroducetwo
theoreticaldiscretedistributions:UniformandPoisson.

Thesecondlessonreviewsthreecommoncontinuousdistributions:Uniform,Normal,and
Triangle.Animportantnotethatismostcommonlyinsupplychain,wearedealingwitha
samplepopulation,notthetotalpopulation.Thelessonthengoesthroughthedifference
betweendiscretevs.continuousdistributionsandhowtorecognizethesedifferences.The
remainderofthelessonisadeepdiveintoeachtypeofdistribution,whattheylooklike
graphicallyandwhataretheprobabilitydensityfunctionandcumulativedensityfunctionof
each.

CTL.SC0xSupplyChainAnalytics

KeyConcepts
Probability
Probabilitydefinestheextenttowhichsomethingisprobable,orthelikelihoodofanevent
happening.Itismeasuredbytheratioofthecasetothetotalnumberofcasespossible.

ProbabilityTheory
Mathematicalframeworkforanalyzingrandomeventsorexperiments.
Experimentsareeventswecannotpredictwithcertainty(e.g.,weeklysalesatastore,
flippingacoin,drawingacardfromadeck,etc.).
Eventsareaspecificoutcomefromanexperiment(e.g.,sellinglessthan10itemsina
week,getting3headsinarow,drawingaredcard,etc.)

Notation
P(A)theprobabilitythateventAoccurs
P(A)=complementofP(A)probabilitysomeothereventthatisnotAoccurs.Thisis
alsotheprobabilitythatsomethingotherthanAhappens.

Union of sets (OR)

Intersection of sets (AND)

2.IfAandBaremutuallyexclusiveevents,thenP(AorB)=P(AUB)=P(A)+P(B)

3.IfAandBareanytwoevents,then
P(A and B) P A B

P(A | B)

P(B)
P(B)
whereP(AIB)istheconditionalprobabilityofAoccurringgivenBhasalreadyoccurred.

4.IfAandBareindependentevents,then
P(A | B) P(A)

P(A and B) P(A B) P A | B P(B) P A P B

WhereeventsAandBareindependentifknowingthatBoccurreddoesnotinfluencethe
probabilityofAoccurring.

CTL.SC0xSupplyChainAnalytics

Summarystatistics
Descriptiveorsummarystatisticsplayasignificantroleintheinterpretation,presentation,and
organizationofdata.Itcharacterizesasetofdata.Therearemanywaysthatwecan
characterizeadataset,wefocusedontwo:CentralTendencyandDispersionorSpread.
CentralTendency
Thisis,inroughterms,themostlikelyvalueofthedistribution.Itcanbeformallymeasured
inanumberofdifferentwaystoinclude:
Modethespecificvaluethatappearsmostfrequently
Medianthevalueinthemiddleofadistributionthatseparatesthelowerfromthe
higherhalf.Thisisalsocalledthe50thpercentilevalue.
Mean()thesumofvaluesmultipliedbytheirprobability(calledtheexpectedvalue).
Thisisalsothesumofvaluesdividedbythetotalnumberofobservations(calledthe
average).

E[X] x pi xi
n

i1

DispersionorSpread
Thiscapturesthedegreetowhichtheobservationsdifferfromeachother.Themore
commondispersionmetricsare:
Rangethemaximumvalueminustheminimumvalue.
InnerQuartiles75thpercentilevalueminusthe25thpercentilevalue capturesthe
centralhalfoftheentiredistribution.
Variance(2)theexpectedvalueofthesquareddeviationaroundthemean;also
calledtheSecondMomentaroundthemean
2

Var[X] 2 pi xi x pi xi
n

i1

i1

StandardDeviation()thesquarerootofthevariance.Thisputsitinthesameunits
astheexpectedvalueormean.
CoefficientofVariation(CV)theratioofthestandarddeviationoverthemean=/.
Thisisacommoncomparablemetricofdispersionacrossdifferentdistributions.Asa
generalrule:
o 0CV0.75,lowvariability
o 0.75CV1.33,moderatevariability
o CV>1.33,highvariability

PopulationversusSampleVariance
Inpractice,weusuallydonotknowthetruemeanofapopulation.Instead,weneedto
estimatethemeanfromasampleofdatapulledfromthepopulation.Whencalculatingthe
variance,itisimportanttoknowwhetherweareusingallofthedatafromtheentire
populationorjustusingasampleofthepopulationsdata.Inthefirstcasewewanttofind
thepopulationvariancewhileinthesecondcasewewanttofindthesamplevariance.

CTL.SC0xSupplyChainAnalytics

Theonlydifferencesbetweencalculatingthepopulationversusthesamplevariances(and
thustheircorrespondingstandarddeviations)isthatforthepopulationvariance,2,we
dividethesumoftheobservationsbyn(thenumberofobservations)whileforthesample
variance,s2,wedividebyn1.

n
i1

xi

s2

n
i1

xi x

n 1

Notethatthesamplevariancewillbeslightlylargerthanthepopulationvarianceforsmall
valuesofn.Asngetslarger,thisdifferenceessentiallydisappears.Thereasonfortheuse
n1isduetohavingtouseadegreeoffreedomincalculatingtheaverage(xbar)fromthe
samesamplethatweareestimatingthevariance.Itleadstoanunbiasedestimateofthe
populationvariance.Inpractice,youshouldjustusethesamplevarianceandstandard
deviationunlessyouaredealingwithspecificprobabilities,likeflippingacoin.
SpreadsheetFunctionsforSummaryStatistics
Allofthesesummarystatisticscanbecalculatedquiteeasilyinanyspreadsheettool.Thetable
belowsummarizesthefunctionsforthreewidelyusedpackages.
Function

Microsoft Excel

GoogleSheets

LibreOffice>Calc

Minimum

=MIN(array)

=MINA(array)

=MIN(array)

Median

=MEDIAN(array)

=MEDIAN(array)

=MEDIAN(array)

Mode

=MODE(array)

=MODE(array)

=MODE(array)

Mean()

=AVERAGE(array)

=AVERAGE(array)

=AVERAGE(array)

Maximum

=MAX(array)

=MAX(array)

=MAX(array)

Percentile

=PERCENTILE.INC(array,k)

=PERCENTILE(array,
percentile)

=PERCENTILE.INC(array,
alpha)

=VAR.P(array)

=VARP(array)

=VAR.P(array)

SampleVariance(2)

=VAR.S(array)

=VAR(array)

=VAR.S(array)

Pop. Std Deviation()

=STDEV.P(array)

=STDEVP(array)

=STDEV.P(array)

SampleStd Deviation()

=STDEV(array)

=STDEV(array)

=STDEV.S(array)

PopulationVariance(2)

Table1SpreadsheetFunctionsforDescriptiveStatistics

CTL.SC0xSupplyChainAnalytics

ProbabilityDistributions
Probabilitydistributionscaneitherbeempirical(basedonactualdata)ortheoretical(basedon
amathematicalform).Determiningwhichisbestdependsontheobjectiveoftheanalysis.
Empiricaldistributionsfollowpasthistorywhiletheoreticaldistributionsfollowanunderlying
mathematicalfunction.Theoreticaldistributionsdotendtoallowformorerobustmodeling
sincetheempiricaldistributionscanbethoughtofasasamplingofthepopulationdata.The
theoreticaldistributioncanbeseenasbetterdescribingtheassumedpopulationdistribution.
Typically,welookforthetheoreticaldistributionthatbestfitsthedata

Wepresentedfivedistributions.Twoarediscrete(UniformandPoisson)andthreeare
continuous(Uniform,Normal,andTriangle).Eachissummarizedinturn.
DiscreteUniformDistribution~U(a,b)
Finitenumber(N)ofvaluesobservedwithaminimumvalueofaandamaximumvalueofb.
Theprobabilityofeachpossiblevalueis1/NwhereN=ba+1

ProbabilityMassFunction(pmf):
1

for a x b
P X x f (x | a,b) n
0
otherwise

SummaryMetrics
Mean=(a+b)/2
Median=(a+b)/2
ModeN/A(allvaluesareequallylikely)
Variance=((ba+1)21)/12

PoissonDistribution~P()
Discretefrequencydistributionthatgivestheprobabilityofanumberofindependentevents
occurringinafixedtimewheretheparameter =mean=variance.Widelyusedtomodel
arrivals,slowmovinginventory,etc.Notethatthedistributiononlycontainsnonnegative
integersandcancapturenonsymmetricdistributions.Asthenumberofobservationsincrease,
thedistributionbecomesbelllikeandapproximatestheNormalDistribution.

ProbabilityMassFunction(pmf):
x
e
for x 0,1,2,...

P X x f (x | ) x!

0
otherwise

where
e=Eulersnumber~2.71828...
=meanvalue(parameter)
x!=factorialofx,e.g.,3!=321=6and0!=1
CTL.SC0xSupplyChainAnalytics

SummaryMetrics
Mean=
Median ( + 1/3 0.02/)
Mode=
Variance=

Spreadsheet

Function

Prob 1

Prob 2

Microsoft Excel

=POISSON.DIST(x,mean,cumulative)

=POISSON.DIST(0,2.2,0)

=POISSON.DIST(2,2.2,1)

GoogleSheets

=POISSON(x,mean,cumulative)

=POISSON(0,2.2,0)

=POISSON(2,2.2,1)

LibreOffice>Calc

=POISSON(Number;Mean;C)

=POISSON(0;2.2;0)

=POISSON(2;2.2;1)

Table2SpreadsheetFunctionsforPoissondistribution

ContinuousUniformDistribution~U(a,b)
Sometimesalsocalledarectangulardistribution
Xifuniformlydistributedovertherangeatob,orX~U(a,b).
1

if a t b
pdf: f (t | a,b) b a

0
otherwise

cdf:

0 if t a

ta
F(t | a,b)
if a t b
b

1 if t b

SummaryMetrics

Mean=(a+b)/2
Median= (a+b)/2
ModeN/Aallvaluesequallylikely
Variance=(ba)2/12

CTL.SC0xSupplyChainAnalytics

NormalDistribution~N(,)
Widelyusedbellshaped,symmetriccontinuousdistributionwithmeanandstandard
deviation.Mostcommonlyuseddistributioninpractice.

SummaryMetrics
Mean=
Median=
Mode=
Variance=2

pdf:

f (x | , )

1
e
(2 )1/2

2

1 x
2

Commondispersionvalues~N(,)
P(Xw/in1around)=0.6826
P(Xw/in2around)=0.9544
P(Xw/in3around)=0.9974
+/1.65around=0.900
+/1.96around=0.950
+/2.81around=0.995

UnitorStandardNormalDistributionZ~N(0,1)
Thetransformationfromany~N(,)totheunitnormaldistribution=Z=(x)/
Zscore(standardscore)givesthenumberofstandarddeviationsawayfromthemean
Allowsforuseofstandardtablesandisusedextensivelyininventorytheoryforsetting
safetystock
Function
cdf ofNormal
Distribution
pdf ofNormal
Distribution
InverseofNormal
cdf
StandardNormalcdf
InverseStandard
Normalcdf

Microsoft Excel

GoogleSheets

LibreOffice>Calc

=NORM.DIST(X,,,1)

=NORMDIST(X,,,1)

=NORM.DIST(X,,,1)

=NORM.DIST(X,,,0)

=NORMDIST(X,,,0)

=NORM.DIST(X,,,,0)

=NORM.INV(Probability,,)

=NORMINV(Probability,,)

=NORM.INV(Probability,,)

=NORM.S.DIST(z,1)
=NORM.S.INV(Probability)

=NORMSDIST(z)
=NORMSINV(Probability)

=NORM.S.DIST(z,1)
=NORM.S.INV(Probability)

Table3SpreadsheetFunctionsforNormalDistribution

CTL.SC0xSupplyChainAnalytics

TriangleDistribution~T(a,b,c)
Thisisacontinuousdistributionwithaminimumvalueofa,maximumvalueofb,andamode
ofc.Itisagooddistributiontousewhendealingwithananecdotalorunknowndistribution.It
canalsohandlenonsymmetricdistributionswithlongtails.
2
(b a)

b x

c
Figure1TriangleDistribution

pdf:

2 xa

ba ca
f (x)
2 b x

ba bc

parameters:

xa

axc

cxb

xb

SummaryMetrics
abc
E x
3
1
Var x a 2 b2 c 2 ab ac bc
18
2

for c d b
Px d
b a b c

d b P x d b a b c

for c d b

CTL.SC0xSupplyChainAnalytics

DifferencesbetweenContinuousandDiscreteDistributions
Justlikevariables,distributionscanbeclassifiedintocontinuous(pdf)anddiscrete(pmf)
probabilitydistributions.Whilediscretedistributionshaveaprobabilityforeachoutcome,the
probabilityforaspecificpointinacontinuousdistributionmakesnosenseandiszero.Instead
forcontinuousdistributionswelookfortheprobabilityofarandomvariablefallingwithina
specificinterval.Continuousdistributionsuseafunctionorformulatodescribethedataand
thusinsteadofsumming(aswedidfordiscretedistributions)tofindtheprobability,we
integrateovertheregion.

DiscreteDistributions

ContinuousDistributions

E( X ) i1 pi xi
n

pi (xi )2
2

i1

b
a

t f (t) dt
b
a

(t ) 2 f (t) dt

ProbabilityDensityFunction(pdf)(seeError!Referencesourcenotfound.below)
Thepdfisfunctionofacontinuousvariable.TheprobabilitythatXliesbetweenvaluesaandb
isequaltoareaunderthecurvebetweenaandb.Totalareaunderthecurveequals1,butthe
P(X=t)=0foranyspecificvalueoft.

Figure2ProbabilityDensityFunction

CumulativeDensityFunction(cdf)
F(t)=P(Xt)ortheprobabilitythatXdoesnotexceedt
0F(t)1
F(b)F(a)ifb>aitisincreasing
Simplerules
P(Xt)=F(t)
P(X>t)=1F(t)
P(cXd)=f(d)F(c)
P(X=t)=0

CTL.SC0xSupplyChainAnalytics

10


References
HillierandLieberman(2012)IntroductiontoOperationsResearch,McGrawHill.

Taha,H.A.(2010).OperationsResearch.Anintroduction.9thedition.PearsonPrenticeHall.

Winston(2003)OperationsResearch:ApplicationsandAlgorithms,CengageLearning.Thereare
manydifferentbooksbyWayneWinstontheyareallprettygood.

CTL.SC0xSupplyChainAnalytics

11