You are on page 1of 7

Lesson4:PrinterFriendly

MeasurePartI:TheMofDMAIC
Lessons Lesson4 PrinterFriendly

Chapter1
Introduction
HelloandwelcometoLesson4.Ihopeyou'reenjoyingthecoursesofar.Inournexttwolessons,we'lltacklethemeasureaspectofDMAIC.
Afteryoudefineaprocess,problem,oropportunity,ifyoudon'tmeasureit,youcan'timproveit.You'llneedyourcalculatorsasweexcavate
deeperintotheworldofstatistics.
Today,we'lllookatafewdifferentformsofdataandseveraltypesofstatistics.Then,we'lllookatsamplingandseehowit'susedtoanalyze
andmakeobservationsaboutpopulations.We'llwrapthingsupbyexaminingthreegraphicalmethodstomeasureapopulation:histograms,
stemandleafdiagrams,andboxandwhiskersplots.
Idon'tseeanyreasonwhyweshouldn'tgetstarted,solet'steeoffintoLesson4.

Chapter2
FormsofData
Whenyoumeasuresomething,thevastmajorityofthetimeyoulookatdata.Datacomesintwoprimaryforms:variablesdataandattributes
data.
Allofusencountervariablesdataeveryday.They'rehardtoescapebecausetheyrepresentthingswemeasurebyusinglength,height,
weight,strength,thickness,diameter,density,pressure,temperature,andtime.Thatprettymuchcoversjustabouteverything,doesn'tit?
Acomponentofusingvariablesdataisitsdegreeofaccuracy.Whenyoureceiveashipmentandplaceitonascale,youcanmeasureitto
thenearestgram,centigram,ormilligram.Thesamegoesfortheothermembersofthevariablesdataclass.Anothercharacteristicof
variablesdataisthattheyoccuronacontinuum.Temperaturerunsfromfreezingtoboilingdensityrangesfromlighttoheavyandsoon.
Variablesdataareusuallyfoundinpairsofcharts.WhenwegettoourdiscussionofcontrolchartsinLesson11,you'llseeanexample.One
chartshowsameasureofcentrallocationandtheothershowsspread.
Attributesdataisalowerformofdata.It'sbinaryinnature,whichmeansthatit'sonewayoranother.Examplesincludeyesorno,defectiveor
acceptable,orinthecaseofalight,eitheronoroff.Thinkofattributesdataascountdata.Bydefinition,attributesdatadoesnotofferas
muchopportunityforimprovementasvariablesdata.
Nowthatwe'veintroducedthetypesofdata,let'sresumethestudyofstatisticswebeganinLesson1.
Enumerative,Analytical,Descriptive,andInferentialStatistics
Iapologizeforthemilliondollarwords.Ididn'tmakethemup.But,Iwilldomybesttoclarifythem.
Enumerativestatisticsprovideabasisforactiononasituationasitcurrentlyexists.Theydealwithaconcreteconditionthatyoucanreadily
seefirsthand.Thereisnothoughtgiventowhatcausedasituation.Forexample,supposethatacoffeeexpertcomestoyourneighborhoodto
sampletheespressoatyourlocalcoffeeshop.Iftheexperttakesafewsamplesandanalyzesthebeans,theircolor,aroma,thickness,and
taste,heisperformingastudyusingenumerativestatistics.
Analyticalstatistics,asyoumaysurmise,dealswithaprocessthatgivesrisetoasituation.So,itlooksattherootcause.Theoretically,a
studyinvolvinganalyticalstatisticscouldgoonforaninfiniteamountoftime.Aslongasasituationcontinues,itsunderlyingprocessesalso
continue.Thecoffeeexpertcouldgoasfarasinvestigatingthegrowingconditionsforthebeans,thetypeofwaterused,andtheperformance

characteristicsoftheespressomachine.YoumaywanttobuyhimaplanetickettoSouthAmericasohecouldsamplethesoilorarrangea
visittothemanufactureroftheespressomachine.
Regardlessifyouuseenumerativeoranalyticalstatistics,bothtypesprovidearationalbasisfortakingactiontosolveaproblem.
Descriptivestatisticsdealwiththeclassificationofdata.Whenyoumakeahistogramthatcorrespondstoadatafrequencydistribution,orto
calculatethemean,variance,orstandarddeviation,youusedescriptivestatistics.Mostofthestatisticsworkdoneinthe19thandearly20th
centurywasdescriptiveinnature.
Inferentialstatisticsinvolvesforminganopinion,makingadecision,andreachingaconclusionfromapopulationsample.Onthebasisofa
randomsample,youinferthingsaboutapopulation.Commonsamplingplansmakeuseofinferentialstatistics.
Supposethatabicyclemanufacturermakesanagreementwithasupplierthatnomorethanonepercentofboltscanbedefective.It's
inefficientforthemanufacturertocheckeachboltbeforethey'reused.But,themanufacturerdoesn'twanttofindoutaboltisdefectiveafterit
isused.Themanufacturerneedsaquickandinexpensivewaytodetermineifashipmentexceedsthepercentdefectivelimit.Sosheobtains
arandomsamplefromtheshipmentofbolts,andonthebasisofthepercentageofdefectivesinthesample,shemakesadecisionaboutthe
percentageofdefectivesintheshipment.
Wehaveonemorestatisticalconcepttodiscussinthischapter:thecentrallimittheorem.
TheCentralLimitTheorem
Thecentrallimittheoremisveryhandybecauseinmanysituations,thedistributionofapopulationcan'tbeidentified.Statisticianshavebeen
abletodetermine(usingmethodswellbeyondthescopeofthiscourse)thatasapopulationsamplesizeincreases,thesamplingdistribution
approachesanormaldistributionregardlessofthetypeofpopulation.So,inplainlanguage,whenyoupickalargesamplesize,the
distributionaboutthemean(Xbar)isnormal.
Youmightbewonderinghowbigasamplesizeyouneed.Italldependsonhownormalthepopulationdistributionistobeginwith.Butin
general,thesamplingdistributionofXbarisnormalforsamplesizeslargerthan30.
Doesthecentrallimittheoremmakesensetoyouintuitively?Considerthis.Inalargerandomsample,wearelikelytogetarepresentative
setofdata,containingvaluesbothaboveandbelowthepopulationmean.Soforalargeenoughsamplesize,thedistributionofpossible
outcomesofXbarwillbesymmetricalandunimodal(singlepeaked).
Sincewe'veusedthewordsampleafewtimes,let'sleaveourdiscussionofstatisticsandexaminehowsamplingcanbeusedeffectively.

Chapter3
Sampling
Inrecentyears,samplinghasdeclinedinimportanceasstatisticalprocesscontrol(SPC)hastakenonalargerrole.Nevertheless,sampling
stillhasaplaceinquality.
Sampling,betterknownasacceptancesampling,isaformofinspectionappliedtoproductionbatches,ratherthantoitemsthatarepartofan
ongoingprocess.Whenyouusesampling,youusesomeoftheitemsinagrouporprocesstorepresentallmembers.Inthemajorityof
cases,batches(alsocalledlots)representpurchasedmaterialsorfinalproductsawaitingshipment.Thepurposeofsamplingistodecideifa
lotmeetsastandard,usuallycalledaspecification.Youacceptlotsthatmeetaspecificationhencethetermacceptancesamplingand
scrap,return,or100%inspectlotsthatdon't.
Acceptancesamplingproceduresareusefulwhenyoumakealargenumberofitemsinashortperiodoftime,whenthecostofpassing
defectivesislow,orwhenit'simpracticaltoinspectlargenumbersofitems.Samplingisbeneficialbecauseitrequiresfewerinspections,
appliesarationalapproach,andrejectsanentirelot,therebygivingstrongermotivationforimprovement.Thedownsideofsamplingistherisk
ofacceptinganonconforminglotandviceversa.Samplingisnotanexactscience,butitisuseful.
Lotbylotacceptancesamplingisthemostcommontype.Whenyouusethistype,youobtainapredefinednumberofunitsfromalotand
inspectitbyexaminingitscharacteristics.Youcanuseacceptancesamplingforthenumberofnonconformingunitsinapopulationorthe
numberofdefectsperunit.
Beforegettingstartedwithsampling,youneedtomeettwoessentialrequirements.Thefirstismakingsureyoupickarandomsampleandthe
secondistomakesureyoursamplerepresentstheentirepopulation.
Whenyouuserandomsampling,youensurethelikelihoodthateachunitinalotcanbeselected.Ifyouviolatethisrequirement,yoursample
isbiasedandtheconclusionsyoumakeareinvalid.Assigningnumbersfromatableofrandomnumbersassurestherandomnessofyour
sample.Ifyou'venotworkedwithatableofrandomnumbersbefore,let'slookatanexample.

Fig.4.1.Tableofrandomnumbers

Supposethatyoureceiveanorderof66partsthatrequiresasamplesizeof7.Eachunithasasequentialserialnumber.Yourollaseven
sideddie,eachcontainingthenumbers0to6,andcomeupwiththenumber62.Thisisyourstartingpointonthetableofrandomnumbers
(it'sthefirsttwodigitsforthenumber62349atthetopofcolumntwointhefirstsectionofrandomnumbers).Topickyoursevensamples,
youbeginwith62andcountdownward.Thenextnumberis35050soyoupick35asyoursecondsample.Wecan'tusethethirdnumber,
71571,because71istoolarge.So,wekeepmoving.Afterwe'redonewiththesecondcolumn,wejumpuptothethirdcolumn.Eventually,
oursamplesare62,35,40,43,27,24,and42.
Sometimesit'snotfeasibletoassignnumberstounits.Ifyoubuyacratewith10,000washers,thetimetoassignnumbersisprobablyworth
morethanthewasher.Analternativeistodrawsamplesfromlayersbyusingstratifiedsampling.Afteryoucreatelayers,youcandivideand
pullsamplesfromsubsectionsineachlayer.
Whenyoursampleaccuratelyreflectsthelargerpopulation,youhavearepresentativesample.Ifyousamplethefirst10piecesfromalotof
100,isthissamplerepresentative?Amachinemayoperatefineduringthefirstpartofaproductionrun,butmayrungooutoftolerancenear
theendoftheorder.Obtainingavalidsamplecan'tbeanarbitraryact.Fortunately,therearesoundmethodstohelpus.
Lotformationinfluencestheeffectivenessofyoursamplingplan.Toincreaseyoursuccesswithsampling,lotsshouldbehomogeneous,
whichmeansthatthesamemachine,operators,materials,andmethodsproducealot.Changinganyofthesevariablesduringaproduction
runcanimpactyoursamplingplanandreduceyourabilitytotakemeaningfulcorrectiveactioniftherearedefectsinalot.Largelotsalso
havelowersamplingcostsbecausesamplesizesdon'tincreaseasfastaslotsizes.
Samplingplansspecifythelotsize,N,thesamplesize,n,thenumberofsamples,andtheacceptance/rejectioncriteria.Theletterc
representstheacceptablenumberofdefectiveunits.ThisisoftencalledtheacceptablequalitylevelorAQL.Youmayuseasinglesampling
planwhereyoudrawonerandomsamplefromeachlot,andclassifyeveryiteminthesampleaspassorfail.Ifanysamplecontainsmore
thanc,yourejecttheentirelot.Whentheresultsofasinglesamplingplanareborderline,it'swisetotakeanothersamplebyusingdouble
sampling.Thecriteriafordoublesamplingcouldbethefollowing:
Ifyoufindtwoorlessdefects,acceptthelot.Ifyoufindsixormoredefects,rejectthelot.Ifyoufindthree,four,orfivedefects,takeanother
sample.Rejectthelotifthesecondsampleproducesfiveormoredefects.
Animportantfeatureofasamplingplanishowwellitdistinguishesbetweenlotsofhighqualityandlotsoflowquality.Asamplingplan's
operatingcharacteristic(OC)curveidentifiesitsabilitytomakethisdistinction.AnOCshowstheprobabilitythattheuseofasamplingplan
willresultinlotswithdefectiveunitsbeingaccepted.Let'slookatatypicalOCcurve.

Fig.4.2.Operatingcharacteristic(OC)curve

At3.7%(.037),wehaveaboutan80%probabilityofacceptingalotwith3.7%defectiveunits.Thepercentageishighbecausethesampling
planmaynotcatchthislowdefectrate.AswemovefromlefttorightontheOC,ifwehavea7.4%defectrate,there'sonlya20%chanceof
acceptingareject.
It'simportanttonotethatmanymanufacturingoperationsdon'tcreatelotsastheyuseacontinuousprocess,suchasaconveyorsystem.
Theseoperationsusuallybeginwith100%inspection,andiftheyfindastatednumberofunitsthatarefreeofdefects,theyswitchto
sampling.Thistypeofplan,createdbyH.F.Dodge,iscalledCSP1.Also,samplingplansexistforvariablesdata,althoughtheyareless
commonthanforattributesdata.
Let'sswitchgearsandexplorethreegraphicalwaystomeasuredata.Thefirstoneisprobablyfamiliartoyoutheothertwohaveunusual
names.Together,thelasttwosoundlikeacatprowlinginthegarden.

Chapter4
Histogram,StemandLeaf,andBoxandWhiskers
Histogram
Histogramsgraphicallypresentinformationthatyoucollectasrawdata.Itshowshowinformationfromaprocessvariesfrommachineto
machineorpersontoperson.Agreatuseofahistogramistoallowyoutocomparewhereyouareinrelationtocustomer'sexpectations.
Whenyoucollectdatatoconstructahistogram,trytocollectatleast40observations.Ifyoucancollect100,thatisevenbetter.Thisallows
foramorethoroughanalysis.Determinetherange(thedifference)betweenthelargestandsmallestvalues.Next,setthenumberofclasses
numberofbarsinyourchartusingthefollowingguidelines:
Lessthan50:35classes
5099:610classes
100250:712classes
Morethan250:1020classes
Afteryoucalculatetherange,divideitbythenumberofclassestodeterminethewidthofeachbar.
Supposewewanttodeterminethenumberofsecondsittookagroupof63peopletoswimouttoabuoyandbacklastSaturdaymorning.We
useRtosignifythedatarange.Sincetheslowesttimewas98secondsandthefastestwas49,ourrangeis49.We'lluse7classesforour63
observations.Eachbarrepresents7unitsastherange(49)dividedbythenumberofclasses(7)equals7.Wesetupbaronebytakingthe
lowestvalueandaddingseven,soitrepresentsswimtimesfrom49and55seconds.Bartwois56to62andsoon.
Let'stakeapeekatwhatwecameupwith.

Fig.4.3.Histogram

Thehistogramisverynormal,meaningthatislookslikeabellcurve.Hadthe56to62barbeenthelargest,theshapewouldbereferredtoas
skewedright.Likewise,if84to90hadbeenthelargest,itwouldbeskewedleft.If56to62and84to90wereboththelargest,wewouldhave
abimodalhistogram.
Wecanvaryourhistogrambydevelopingastemandleafdiagram.Let'sseewhatit'sallabout.
StemandLeafDiagram
Touseastemandleafdiagram,weneedtorearrangeandmodifyourswimtimedata.Let'sgobacktoourtestscoresanddojustthat.

Fig.4.4.Stemandleafdiagram

Weplacethefirstdigitsfromourswimtimes(4,5,6,7,8,and9)inourfirstcolumn.Theninthesecondcolumn,weplacetheseconddigits
fromtheswimtimes.Forthe4incolumnone,weadd9incolumntwoforthetimeof49seconds.Andforthe5incolumnone,weadd2,4,6,
7,7,8,and9inthesecondcolumnforthetimesof52,54,56,57,57,58,and59.Wecontinueuntilallthetimesareinthediagram.
Thistechniqueisconsideredalowbudgethistogram.It'sidealforpeoplewhodon'thaveaccesstomakinggraphsorwhodon'thavethetime.
Thewordstemcomesfromthehighestdigit(forexampletens,hundreds,orthousands).Inourexample,wehavetens.So,ourstemsare40,
50,60,70,80,and90.Theleavesaretheunitsthatgowitheachstem49is4forthestemand9fortheleaf,52is5and2,and60is6and
0.
Thepurposeofthisistocreateapatternbasedonthestemsandleafs.Ifyouturnthestemandleafdiagramonitsside,youcanclearlysee
thatthe70sarethelargestcategoryandthe60sandthe80sareequal.It'ssimilartothehistogram,butdiffersslightlybecausewehaveonly
sixclassesinthestemandleafdiagraminsteadoftheseventhatwehadinthehistogram.
Ihopeyouenjoyedconstructingthestemandlifediagram.Let'smoveontotheboxandwhiskersplot.
BoxandWhiskersPlot
Icouldmakeabadpunandsaythatthismethodisaplottocaptureacatinabox,butIwon't.
Let'sstartoutwithanexampletoillustratehowboxandwhiskersworks.
Assumethattheswimmerswerenothappywiththeirtimesandwanttogiveitanothergo.Thistimewedividetheswimmersintotwogroups
often.Herearethetimesforeachgroup:
GroupA:42,56,58,66,68,75,80,92,96,97
GroupB:53,59,61,72,78,85,85,87,91,91
Whichgroupdidbetterinthesecondroundtriptothebuoy?It'sdifficulttotell.ThemeanforgroupAis73andforgroupBitis76.Butwhat
elsedoweknow?Let'susetheboxandwhiskersplotforhelp.
Wehavetenstudentsineachclass.ThemedianforgroupAis(68+75)2=71.5andforgroupBitis(78+85)2=81.5.Ournextstepis
tofindthemedianforthefirsthalfofthevaluesandthemedianforthesecondhalfofthevalues.ThemedianforthefirsthalfofAis58and
forthesecondhalfit's92.AndthemedianforfirsthalfofBis61andforthesecondhalfit's87.Thesearecalledtheupperhingesandthe
lowerhinges.
Next,weneedtolocatetheHspread,whichisthedifferencebetweenthehinges.TheHspreadforAis9258=34andforBit's8761=
26.Wethenneedtoconstructinnerandouterfences.Why?Sowecankeepthecatscontained.Seriously,ithelpsusdeterminetherange.
Herearetheformulas:
upperinnerfence(UIF)=upperhinge+(1.5Hspread)
lowerinnerfence(LIF)=lowerhinge(1.5Hspread)
upperouterfence(UOF)=upperinnerfence+(1.5Hspread)
lowerouterfence(LOF)=lowerinnerfence(1.5Hspread)
UIFA=92+(1.534)=143
UIFB=87+(1.526)=126
LIFA=58(1.534)=7
LIFB=61(1.526)=22
UOFA=143+(1.534)=194
UOFB=126+(1.526)=165
LOFA=7(1.534)=44
LOFB=22(1.526)=17
We'renowreadytousethisdatatomakeourboxandwhiskersplot.Let'sseehowitlooks.

Fig.4.5.Boxandwhiskersplot

Interpretingaboxandwhiskersplottakesalittlebitofdoing.Whatdoesallthisstuffmean?
Forstarters,groupAhasamuchwiderspread(from7to143)thanB,eventhoughAhadthehighesttimes.GroupBhasthehighestmedian.
So,overallgroupAisperformingbetter.Theasterisksreflectthelowestandhighesttimesforeachgroup.Wereyouabletoseetheboxes
andwhiskers?
KeepinmindthatastatisticaltechniquesuchasSDismuchmorestraightforwardinitscalculation.Also,knowthatit'simpossibletohavea
negativeswimtime.Manypeopleusetheboxandwhiskersplottomakeavisualdisplaybetweentwopopulationswithouttheuseof
statistics.

Chapter5
Conclusion
IhopethatyouenjoyedpartoneofourstudyofthemeasureelementofDMAIC.Wespenttimewithstatistics,sampling,andschematics(I
hadtopickathirdwordthatbeganwiththeletters).Youlearnedaboutvariablesandattributesdata,andenumerative,analytical,descriptive,
andinferentialstatistics.Youalsofoundoutaboutthenatureofthecentrallimittheorem.Youdiscoveredthefinerpointsofsamplingandthat
youmustmakesureyoursamplesarerandomandrepresentativeofapopulation.Finally,youlearnedhowtousehistograms,stemandleaf
diagrams,andboxandwhiskersplotstomeasuremembersofapopulation.
I'llseeyounexttimeinLesson5wherewe'lldiscussfailuremodeandeffectsanalysis(FMEA),physicalmeasurementmethods,and
probabilitydistributions.Makeitagreatday!

SupplementaryMaterial
Comingupwithvalidsamplesizes
http://www.custominsight.com/articles/randomsampling.asp
Thisisanoutstandingreference.

HistogramsandStemandLeafDisplays
http://davidmlane.com/hyperstat/A28364.html
Yougetabonusatthissite.Bothtoolsarediscussed.Enjoy!

BoxPlotsorBoxandWhiskersPlots
http://ellerbruch.nmu.edu/cs255/jnord/boxplot.html
Theinformationatthissitewillhelpyouunderstandhowtocreateandinterpretaboxandwhiskersplot.

FAQs
Q:AhistogramandaParetochartappeartobeverysimilarinconstruction.Canyouexplainthedifferencesbetweenthetwotools?

A:Whilebothtoolsdisplaypatternsorfrequencyofdata,theParetochartshowssummarizeddatainadescendingmanner,whileahistogram
showssummarizeddatainanordinalmanner(abarforgroup1015,thanabarforgroup1520,andsoon).Paretochartsalsohavetwoy
axes,oneforeachgroupofdata(bars)andacumulativetotal(line).Ihopethishelpedclarifythedifferences.

Assignment

Useeitherahistogramorstemandleaftoconstructadiagram.Selectdatafromaprocessyouknowwell.Also,interviewsomeoneto
determinewhattypeofsamplingplanstheyuse.

Backtotop

Copyright19972015Allrightsreserved.Thematerialonthissitecannotbereproducedorredistributedunlessyouhaveobtainedpriorwrittenpermission
fromCengageLearning.
tqa0

You might also like