Вы находитесь на странице: 1из 80
vo EN Al RightsReserved . One or wrt prison Ks he pubis SPC Press Twos Fos (5) 889010 tse: nan The ane Dist Jacket Photograph otha te ae ‘Sotnch telescope itor st of Yechnagy and Gadi permascn CONTENTS, Introduction Preface tothe Second Edition Data Are Random and Miscellaneous Presenting Data in Content ‘Time Series Graphs Tally Plows Comparing Tie Sess [Numeric Summaries of Ones Aveopes [Nemerial Summaries of Dae: Range “The First Principle for Understanding Data Summary Knowledge Is Orderly and Cumulative Compartons to Specitions CCompanzons 9 Averges Shenlat’s Approach o Iaerpeting Date Understanding the Trade Defiste The Second Principle for Underanding Dats Summary The Purpose af Analy Ie sight Char for Indvidl Vals and Moving Ranges Finding the Moving Ranges The Two Graph inant Char (Comping Limite Foran ink Chast Tmverpreding she Xin Char A Char for On-Time Shipmcas The Premium Freighe Dat Se Whats The Difference Monthly Receipes Summaty 4 15 B » Ps C Four: ‘The Best Analysis I the Simplest Analysis renin Feige Revisit Production Volumes Fit Time Approval Ross ‘There it Nothing New Unc the Sun Pound Scrapped (OSHA Repowables The Rest of the Story Summary But You Have to Use the (On-Time Closings of Aesounss Hot Merl Material Cone Spill Sumtingy ight Data Sie Look What You've Been Missing Less What Should You Do Now The Transformation Whar Should You Do Sang Tomovion? Bur fe Cane Be Thus Eay, Can I? Appendices “These Types af Action. A Chanttersaton of Predicate Proce ‘A Charsctezation of n Unprdicable Proves For Further Reding ink Chars Using the Mian Moving ange init Chars Using he Average Movieg Range Why Not Ure p-Char? A Modest Propo We Ais Broke, Dons File Index o st as 0s 107 na ua 7 23 135 16 130 134 136 be. 133 INTRODUCTION We live in the “laformation Age,” and much of tht information comes t0 us in the form of numbers. Indeed, wherever we look we ae surrounded by ever increasing mountains of dati—indexee and polls marker epores and leading indicator, government sepors, balnees of| ‘his and imbalances ofthat, data showing that food enuses cancer and ‘cures smoking Yer in sive of the increate in database and spreadsheets in spite of| the speed with which one ean be tied into nerworls to acces lly and repore data, wear old that our productive filing, our manufactur ng base is eroding, and our economy it geting werker. While we hive cater and lager number evunching machines, and incseasing aces 10 hese machines, and while we are making grea strides in many areas of| bse science, we are told chat our technological edge i slipping and tha others ae taking the lead in trnilecrng scientific knowledge into applied products for scien’s use Closer ro home, businessmen ae Gnding that while sey have move ets than ever befor, they sll do not know what these numbers mean. If che numbers changed for che beter compared eo last mooth, then just waic—they will change for the worse soon enough, If the numbers actully changed forthe wortecompated to lat month, chen he apocalype ie at hand and all re doomed! ‘The bos isin despait— ‘Don'ejurt stand thee, do something!” You have to come up with an explanation of why che numbers were 2 bady or eee find & scapegoat, by 0:30 tomorrow motning. Moreover—how ate you going to keep these bad numbers feom happening agsin? How are you going to get C Understanding Variation / Managing Chas the workers to work harder! “And Pharaoh said “You ate ley! You will be given no sts. bur you mast produce the same tally of bricks cach day." And 10 ic goes, month afier month, world wiehoue end. From the dawn of time until the present there is nothing new under the sun—jurt mote of, De DR mw Te problem with our information age was sucinedy ated by Daniel Boorstin when he std: “Information is random and miscela- ‘ous, bur knowledge is orderly and cumslaive.” Before information ‘an be useful it must be analyzed, interpreted, and assimilated. In shor. raw data have co be digested before they ean be well This proces of digesting daca has been widely neglected all levls ‘of our educational system. Managers and workers, educators and st amen, financial analysts and bankers, doctors and nurses, and especially lswyess and jouenaiss all have onc thing in common. They come out of heir educational experience knowing how co add, subtract, mulcipl, and divide, er they have mo sundentancding of how to digest nusnbrs to extract the Rnowledge thar may te lacked up inside the data. In fat. this shorcoaning i also seen ta lesser extent, among engincers and scientists, This deficiency has been called "sum: ‘naive is noe a failure with arithmetic, but cs nstead a lure wo know hhow to wie the basic tols of sithmeic co understand data. Numeral dents, accountants and busin al aiveté” Numerical naiveré is not addresed by the traditional courses in the primary or secondary schools, nor isi addeesed by advanced courses in mathe. matics, Thi is shy even highly educated individuals cam be numerical Fortunately, the cure for numerical naiveé is very simple. T Principles are easy to grasp and the vchniques are very easy «0 imp! ¢ iment. This book was writen 10 help organizations overcome the elects of numerical naveré The use of the echniques presented in this book can have profound consequences both for individuals and for organizations, These tech rigues have been thoroughly proven in over 70 years of practice in all walks of life, However. until these techniques are actually put ince nothing wll happen. Namerislnaiveré can only be overcome theough practice. This book will tell you how to go about it. “The ral world wil give you the opportunities for practic. sinderstonding Variation / Managing Choos ACKNOWLEDGMENTS | would like o thank those who have helped and encousaged me in binging this book to completion. First among these is Frances ‘Wheeler who faithfully read and re-read the mateil, catching mis takes, making suggestions, and asking for clarifications at those points where the prose was obscure. Other readers include Prony Ward and Wade Meyercord. In addition, thete is Steve Cab who encouraged sme to bogin this work, and Tom Sands and Bob Kaspriyk who made ‘many helpful suggestions in the early drafts. Finaly, shee is Bob LLucke who kindly checked all of the computations, The contsbutions ofthese individuals have made this hook «beter book Introduction PREFACE TO THE SECOND EDITION This edition includes thece new appendices and an index that were notin the orignal version. In addition I have eaken thie opportunity co update the nomenclature throughout the book. The orginal version sed the tadtional labels which have boon used during the past 50 ysars However, since this book i ut dhe nontraditional applications of the proces be ranageral snd ad ch ‘appropriate 1 we words chat ae more descrip hereore throughout this book thé following changes in wording have beer made “Control chars" have become “process behavior chars A “in-contol proces has become “predictable process.” An “out-of-concol proce" has become an “unpredictable process.” ‘Aa “ouroF-contel point” has Become « “point outside the limi” ‘An “in-contolpoins” has become a “point inde the limits.” [And vatiation is described as being either “routine” or “exeeptionl F what i intended. Thus, jargon has been replaced by ordinary wouds and phrases having the intended connotat The intent of these changes is ro remove the baties of inappropriate terminology, and hereby to make this powerful approach more scce sible o those wh are noe familiar with the tational jargon Those zeaders who would ike more infoumtion about the rationale this change in somenclacure will find the appendix tiled “A by Modes: Proposal” to be of incre. c Understanding Varin / Managing Chaos ONE DATA ARE RANDOM AND MISCELLANEOUS Recently the U.S. Trade Balance for February showed a deficit of 1.4 billion dollars. Asa deseriptive valu, shit deficit meine chat the US irmporeed goods worth 11.4 billion dollars more thas it exported uring this parucule month, Clea, 2 trade balance deficit ofthis sie iebad. bur whae dos it mean about che age picture? Ave hing gesting, ewer ot worse? Comparisons are needed in onde ro make sete ofthis, ov eal, his ei ebay was A iin dg ache et for ayo theme year Ti cmpaion wos sige ‘ings pesing oe —bewre Tower: the Febuary dei of 14 bilon wat 1.6 lion wale than the defor Febrny ofthe prev yea This competion oud sgps hae ings egg ete Ina econ yea you wold ado bth comprions © being ade eating te ld concson ht theater "The problem with both of the comparisons above is chit limired caste Whe wiv ped tot and plc hes a Ea a “Sgr come and whe ah he compro above ae oe tee conde, No eomation boaten eo vue be gba ‘simple compaton beresen che caren gue and ome peo va canioc fill ips ad camigaheTekaior oF any time sete, Yet na Understanding Variation Managing Chace Figure 12: The World According othe Monthly Report In the words of Myron Tribus: - epee opoyycs monl rp | Meme scraper bree phaser ** "Neverticles, managers use these mooshlyeepors to any Explanations ars egies for any fires a they “shoul be. Repous have oe ween Accon pln for deling ‘be robin values have to Be ine ad then che values had beer improve hae! that mean! 7 All of the preceding examples show a common tre to provide some \ contextual background forthe inerpreaion of any gen umber This ‘urges prope and corece—you simply cannot make sense of any number without contextual bats. Yer the waditons temp te provide ie contextual basis ate often faved in-thircxcuton. ‘The most coma that of med comparison C PRESENTING DATA IN. CONTEXT _ _ While iis simple and exey to compare ons aumber with another Tay ace limited umber, such comparions atc limiced ang RK" They a i because ofthe amoune of dat sed, and they are weak because bah of ambers se subject wo te varion thai ineviaDly pind i real e both the Gretta ad the ealrsalue ate aubiet 5s vaiton, wil day be diffcl odetmine just how much of the difrence heweea the alr de how much, any of he ete shis author's hunch ta {coche monthly epat by sione who gored of having to expan the [Up and down movement of the comparison berween monly vals ~The yeatodate values will, athe yer progres, show fest variation Terence doe I changes in the process, youro-date Figures wee ist added such comparisons seal i din scope beat they donot us he pat data in ane ‘Well ifTimied comparisons ace not enough, tien how abut snply (Cpreseatng all ofthe values in a able? While this is useful, iis noe Enough to provides complete analy Jin F& Mar Apr Nay Jon Jl Aug Sep Om Nov De oer 187 GO 1A HS SM MA ML 126 160 117 106 Isis 100 14 7995 40 M8 105 2 92 11 104 105 1988 ($ itions) Figure 13: Monthly U.S. Trade Deficits 19 ‘While ables of values may be wed co presen al ofthe relevant dats they sie not ensy surearine. Tee dificale to digest the inf coniainc i GRIEG purere NTtcove, that which c Understanding Varian / Managing Chaos ited is generally hard to communicate ro othet. ‘The human mind just des not doa very good job of abrrbing lage amos of dt. Bik aa Saas jeuon 18 19 20 2122 23 26-25-26 27 28 29 0 91 32 38 0 fnmeruas tt 29 54 59°35 dt Ae 8 MG ae aH ‘etn ELE RES ESOA nmenee 14 38 16 5 hs, the traditional ways of communicating with numbers—limited ‘comparisons and tables of data—both have severe dawbacks. The ual ‘compatitons are narrowly focused and weak, while ables present an abundance of extrancous details, C,,, The tlaton to shee problems es in si from a digi eepresen “ation of data oa form that ie more fienly co humansgraphs Graphs can remove che exeancous deal given by tables of numbers hile they focus on the intresting bits of information which may be com tained within che data. Graghs provide de conte for interpeting the ‘euzent vale because they eI indude all ofthe relevant they do tht vial ia digital manner which snes te inf Fe are two basic graphe which have proven thei worth in ehis respect—the time seis graph andthe ally plow. ffi rmeearhe) TIME SERIES GRAPHS Time sever raph (or renning record) typically have months (or yas) ‘marked off on the horizontal axis and posible values marked off. om the vertical axis, As you move from left o righe there is a pasageof time: C (One Data are Rand and Micsonene Changes in he ine series een a you she ap om hee 0 ther By ily comparing th cen wa he pled wee fr he posing monte you can lyse i the care ve “hein ere gph forthe US, Trade Dein 1967 ad 1988s shown in Pre 13. The poph in Figure 13 uggs the computed wah 1987 thre ight eben igh imponemen in 198. Figure 15: Monthly US. Trade Defies 1997-1998 The time series graph for che home suns pet season for both Babe Ruth and Roger Matis are shown in Figure 1.6, These graphs show the ecords of 60 aad 61 home rns in the context ofthe eaters ofthese 190 baseball players, Time ste plots communicate the content ofa data st more guicldy and completely than do cables of values, Understanding Variation / Managing Chaos 2, comparisons ofthe curent vale with ancher vale ate the most co tmon pp of comparions encountered or evample ial al government fgues ae subjected to tit ‘ype of compaion when they se tepored in the press. The ie or {his month is compared wo the vale rom the receding mon IF this ‘mont valu has changed forthe beer ching are ooking up TF this ton vale has changed forthe morse, watch cu! Ta similar veins automobie sales re led and reported every 10 days These values lay compared othe sls ies for dhe sme 10-day pid from the previous ynt, Do you emerber how ses were ding during this 10-day period lst year? OF couse you don So what isthe bass or making uch» comparison? Why wold hs com pation provide addiional insighe Inco she cult sles gure? Comptons of Tissot are balton the implict assumption hats {ets notial” You do el how normals yer was do ur” ode tend 6g bse da fie ale we Jo en lve readily ith spots da\Conade the record forthe numberof home Set cee emarineseets eee ee ee tala ec ES veces cr ler sw Rose Mtns hover tee oe Nene bake a Een rena ee aoe ds conmcand ort be he er Mant eed Samp tare ncon wh be aes Lever eee Tsefupatcn tral fone a Se ila he nee tice ie cc Se ase pe ee Sina eooel'n [hr poe Boe aS cipayocmt wee ion was perfectly appropriate. However, 1927 sexony From 1961 10 1990, (One Dats are Rana and Misenens Finally, think about the tational way in which business data are packaged in the monthly reports—eeveral pages of table: of numbers, tbuained fiom computer printouts, and reduced down to 2 size that 90 tone over 45 can read without cei bifocals. A typieal monthly report igh look ike dhe one shown in Figue 1.1 ny Repay 2p Meee ec my on Sas a = 2s. ees BE Gomwspeimee as ae 0 Ce te vio henmsamiaet) tae Be ‘ar ee 88 ‘A-Typial Monthly Management Report For each line, this monthly La) gives dhe curren vale, — (2) lists plan or average vale, (2) compares the carene valu otis plan or average value, (4) compares the current vale othe val forthe sine month as year, ves the current year-to-date value (6) compares this year-o-date value with «plano average valu, and (7) compass the current and previous year-to-date values. “The four comparisons in his list are the ones most commonly used, CCllecively they try to provide some sre af content for interpresing the current values, However, since each of hese four comparison is limited, hey may provide contradictory messages C LUndrstonng Variation Managing Chaos CTAUY PLOTS A tally pls (oe histogram) issimaply an accusation ofthe difeent ‘lcs hey occur without yng to dipay the time order sequsice A mark is placed beside a possible value each cime cat value 8 observed in the sequence. If the posible valucs are witten on the horizon oxin shen che vertical axis wil represent the fequencis for the diferent values ‘observed. A simple tally plor ofthe number of home runs per season for ‘Babe Ruth and Roger Matis is shown in Figue 17. Figure 1.7: Histograms, or Tally Pols of Home Runs Fer Season gorenine ‘The ally plow Simpre he dasa shown onthe dime ste In this cate the compiesion mak i ide by ee comparison dee’ Babe Ruth's 60 homerun season was simply ic high point ofan oussaading ‘acer. Roger Maris’ 61 home sun seion was, by any measure erst sional. Sil; Roger did it 61 home rns in 15GL. But shoul shi be stcsibuted solely to his personal skill?” This question can only be answered by considering an even broader conte c (One Date are Rad and Mines COMPARING TIME SERIES Addivionalinsighc may sometimes be pained by using two time sete raps together. Figure 1.8 shows the home runs per season for both Roger Maris and Mickey Mantle, Figuce 18: Home Runs for Roger Matis nd Mickey Mantle Roger Maris and Mickey Mane weteeamnnes oi 1960 01966. The way shes ro running record move up and down together sugges ‘hac the number of home runs Ric by each was subject to some extemal fluence. One possible explanation is that they Faced essentially the ‘amet of pitches each aaron Perhaps the nuber of home ans ie by these erelaggeswene up and down a the quality ofthe prching wal vatiod CORTES 0 ex Whaieer heSlind fr pull, che values ate beer unde ‘ood when chy ae placed it a broader context c Understanding Variation Managing Coe NUMERICAL SUMMARIES OF DATA: AVERAGES In addition to ime sevice graphs and lly plo i is vecasonaly wss- fal to compure numerical summaries of ast of data. OF couse the basic ‘numerical summary isthe avithmetic average. This average is simply the sum ofthe values divided by the numberof values. A genetic symbol for theaverage of sof values wil be g ‘This symbol is commonly zead as "X-bas” oy The average is said to be a metre ofthe lction ofthe set of values Babe Ruch averaged 41.1 home eune pr seston between 1918 and 1934. Roger Maris averaged 22.9 home runs per season between 1957 and 1968. And Mickey Mantle averaged 29.7 home runs pe season benween 1951 and 1968. ate tan a eoo m 6 fo boo Figure 19: Averages for Theee Histograms werage is a measure of location, its common to use s¥er- see310 compe ove data seu The et with the greater average Toph (0 “exceed” the other. While such comparisons may be helpful, hey ‘mutt be used with cation, After all, for any given data Se, ose ofthe value wall nor i ——— ro thenverge € (One Duta ae Random and Micelles NUMERICAL SUMMARIES OF DATA: RANGES. A measure ofthe dtpersion ofa se of vals the cange. ‘The range it efined asthe maximum value minus the minimum value. ‘The genetic symbol fo she ange is R ‘The range of the number of home runs pet sean for Babe Ruth i = 49 for these 17 seasons. ‘The ange of che numberof home uns pt seston for Roger Maris is R = 56 for these 12 scsons. The range foe Mickey Mandle's 18 seasons i R= 41. (Here the eange does lil besides remind the eader dha the numbers of home runs pr scison dil deviate fiom che avenge) ae Rah Q soe mo 6 fo Bop St Figere Lite Ranges for The Histograms While other messures of location and dispersion exist, the average © and che range wil be sufficient forthe applications inthis book, Undertaning Variation / Managing Chios THE FIRST PRINCIPLE FOR UNDERSTANDING DATA Many yeas ago Dr. Wer Shewaze gave ovo rules for the presenta tion of daa, They are the basis for hones statistics and shouldbe wed as 2 guide in every presentation of data. His two rule are paraphrased and ‘iscussed below. ‘Shewhart’s Rule One forthe Presentation of Data: Data should alse be presented in rch away ‘har prserse the evidente in the date Jor all he prediction hat might be made fom thse data ‘This cule suggests several chings. First ofall a table of the values should accompany most graphs. Second, 1 tbleof values is no sufiient “i coy che big piccure. We are visually oriented and tables of data are visually boring. graph (such as a sime series graph or a histogram) should accompany any table, Third, the context for the daca shouldbe ‘completely and fully described. This would include the answers to: Who collected the das? How were the dats collected? When were the data collec? Where were che data colleced? What do these values tepte- sen And if the daca ate computed values, how were the values com ted From the raw inputs? Has there een a change ia the formula over a graph is woich’2thousand words, then an annotated graph with this supporting information is easly worth ten thousand words, While shore summaries of asec of value may be equied, they should always be backed up with annotated graphs. Data cannot be divorced from theit ‘context withou the danger of disoction, ~ ¢ (One / Data are Random end Misiones Shona’ ale Two for he Presenatin of Da When nigra hiram is wed amar dt the sen al nll te min ting ey ein the hee wouldnt taf date ere preted atin te Averages cages, andhsorams all cic the ime order for the data Ihe timer forthe dts shows woe wo of dein perm, ther the cbscurngof this pera by the use of aerags ranges othe ogame cm min the wer Since all dna ocala di will have aimeorder Insome css his imeode the ena coment wich en be prod in the proentatin : “These les of Dr. Shewharts can be summa in one owing principe which the ft pinpe for sadetanding ar and commu wing wih da HE FIRST PRINCIPLE FOR UNDERSTANDING DATA at ave meaning ‘er fim ti ome Thice consequences ofthis fst principle are + Thatta ne who cannot, o- will not, provide the content or thi gurs. + Stop reporting compartons besween pit of values ctenp as part of « brondercomparion, 1 Stare using graphs co present current values in context, SUMMARY Understanding Variation / Managing Chaos No comparzon berween two valu can be pabal ‘Management epors ae full of limited comparisons Graphs make data more accesible the human and ‘than do table, ‘Numerical summaries of data may supplement ge bur they can never teplace them No data have meaning apar from ther context, TWO KNOWLEDGE IS ORDERLY AND CUMULATIVE Many yeas ago, David Chambers found the fllowing time series ‘on the wall ofthe ofice ofthe president ofa shoe company. Here was a simple and powerful presentation of dat in concert. The eat nthe ‘etc ans was “Daily Pereentage of Defective Pats.” Joop ne ela nu, Dai sd he pent wy he ad he gah on he wall THe pds condregy pod thas heb ea one ioe kel we panes ng is ning pn pa BALES sh AG Evident SRR EU RE cebiy t askche pede this, beaut he pened aoe the chat on wl td ten awl Someta be hohe 6s ¢ Understanding Variation / Managing Chans = Sp Data Analysis, Inerpreation > Y pu Transformation oxap igare2: Making Sens of Data Ia Process Even though the preside had his data diplayed in wttable graphic fens nd ag be the date ingore ee ‘se then psig ah day hed nat gts a ear toe ‘xd intent then Or tes he didn havea orm wap of wang shen —he just uied a seatoFche-panu analy [No mae whit se da and nose how the ve se aang [| and presented, you must aways use some method of anajni wo sence LL wichan interpretation ofthe dae " ° Data have tobe fie ip some manne make them nigile This ir ils person's experience pls hr prep. 5 more formalized and Tes subjec. fe but there will aleys be Some hod SF ann” Wexpetones {he bas For iniopeting hed, shen the intpation ie re 2 she manage pat eipttivie. rahe canentseuton woeeade ion oF Re da may wll he manager’ expevience, then his iateap inconee, Tkewise Uewed)ssrampions or awed preruppestion cre abe rl in awed inccions. Howeves, in the sake of slate aprons dt ge se che sei-fshe-pane approach and, inthe end, soit ali oy ‘ome dys appear obebecer than okey ‘Thi foeus OF his chpicr wil lupe he vaious ways people ana- {edie Fin, oro of he move cominon methods of imerpreung C Too / Knowledge Orderly end Cumulative daa—compations to specifications, and comparons to averages be outlined: Nex, the procs behavior char appraach Wo the inais and interpretation of data will be presented and concasted with the common approaches. Finally the Fundamental diffecnce between the common approaches and che proets behavior chart appcoch wil be highighed by the econd pine fr understanding date (E7020 DE COMPARISONS TO SPECIFICATIONS soa, budgets, and tages at ll specifications. The ides of ‘ompating sianagement data to plan, goal, or budgets was teansfered che manufacturing pracie of comparing product mes This ype of eomparson defines the Aiecly Fr suremens with speciation limi, postion of the current value relative ro some value (ponsibly arbitrary), with the outcome being a judgment that the cutrene value is either acceptable ot unaccepable (iherimspoc or outof spec). The fact chat ‘hit approach co the analysis of data wil alway esl ina favorable or an unfavorable outcome wil inevitably lead 2 binary world view “Those i favorable figures get 3 pat on the back, and those with unfvorable se with favorable figures are Figures ger kicked litle lower down, ‘doing okay” while the others are “in wouble Ungoraie Favorable Re way weet a oe SS | ote pln Figure 23: The Binary World View ofthe Specification Approach Understoning Variation / Managing Chaos A natural consequence ofthis specification approach tothe interpre tation of management data is theddenis ith which you can change from a sae of bik to'a sate of torment. As long as you sre “doing okay” ther is no reason t9 worry, so st back, ean and le things cake cate of themselves. However! thénéver you se in wouble, “dont just stand there—do something!” Thus, the specification approach tothe interpretation of daca will nevcably resule alecaang with perinds of nen pate This apie athens approach is completely “shihecial'o continual Timprovement “The ipecion spponh a bk gud ses nad, In onder te duinguh kets dee ei will so think of specestone he management dea ling is edifice ~ First there ate specifications which are facts of life—does the bottom line show a prof ora la? he er gwing w lc eae ores Ths Ei eay wo ndenand Of ee einer 2 "co ie only iis knoen to bee ahh bie ee | fmt biog somes sp Nexc thee are he psifcons needed fr planing. Preisions and badges wosld spay lm hs enon Sot igo ts neve be abr. Fgucs eed for pasty shold ae beng saponins of pan dat pls the peanesenon ih kl fare condone. Moreover, pcos fed bade n e Setken at nges. The uncrsininsofenendg he ne bow a into the Furure makes chese values ‘afatiable as targets. Predictions need \bnowisdge, and they should be wed for gues an eet \adgrent and blame Foal, thee ne pccions which we aia nme es That ac uly difien fiom the speifeacnns ee hn ae fetes While che valerie otto encpr ae ae tanh rea necesay, arbitrary numerical eargts ae neither helpful nor neces sary. Infact they ae often detsimenal 2 Ts dinictinn war maby Hey Nee ah Din Diem in periods of beni elec C Tw / Koulaige is Ondey a Camative LO, This Far and No Further! epost Figure 24: The Origin of Many Specifications Specifications ae the Voice ofthe Customer When the specification isa fat of life, or a figure needed for planning icis useful know how the current values compare tothe speifiatons, When the specification ivan asbitary numerical goal, then i can be very dangerous to compate it sagunse che eutrent value. For example consider an atbitrary production ‘acget. If this month's value is above the target the foreman may be rempred ro stockpife he excess and use i aginst next months qua. TF this months value is low, chen he may be tempted to ship marginal of incomplete items in order to make up the deficit a In one plant wher i became clese cht they would not meet the arrest month's production quot, the foreman would send s Yorklif to the warchouse to bring back skids of finished product. There unite ‘would then be unpacked, londed on che conveyor, and sent down the “Patna Bsn Jets pang ne hice emt ee C Understontng Variation Managing Chaos Backingline, As thse unis pase down seine the tomate counts ‘would count chem as finished unite As rol ofthis cxsene the {depureent would fave another “ppod” muah andthe taco eed Snot have to “explana bad vale.” ‘Of course atthe end ofthe yo the warehouse inventory was shore by aboot «milion dlls worth of hn ished product. The showage 5 he plane manag MA" Waurly te new plat manager was nervous, He monitored the outine pant dies very cael, seeking explanation forall values which were unfavorable relative fo he targets. Since he pruetopecorm was maintained, he Production foreman eontaued as before, nat the endo the foe yo the wachoute wat agua about milion dels short on inventory. The second plant manager ed The new plant manage was sry nervous. He Wt nvenony air only three months OF eure the warehouse was about t suanes of milion shor. At shi poin the manager ook action he bal § foie? sound the plant sie and placed guards a the gate, Ae another tee ‘months he had anather inventory done--now they were» halfsailon shor fo the year In derprstion the manager Dui» ee aroued che varchoute ise and placed + guard on the gue. While this lly stopped the Siiskage inthe warehouse che prauction figure tok 2 comidenbledve ces Novice how the emphasis upon the production age was she ‘xgin ofall ofthe euro in his case. People were fred ond hace, smoney was spent all ease the production foreman didnt te ae £© exphin, month afte month, why they hed nt tnt the roducron quota, When peopl ate prenzed co mesa age luther thoes says hey cn proce ae They can work o improve the system, They ean distor the system, Or they cen disoo the dat ae 5 nie ame p wi ch ee es pm (i C Too / Knoulale Ondery at Cumulative Before you can improve ay ase you mut ten 1 the oie of thease the Vio the Pcs). Then you mus undewtand how the inputs affect the opus ofthe system” aly you mat be able to Chung she inputs and posy the em) i onder to ahi the deated ree This wil requ suseained fo, contany of pp, nd an environment mre continual improvenen he spering pls. Comparing ube secon il at ad the inp tment of the pose. Spelicons tthe Vo ofthe Carma a th Vie ofthe Proc The psa apponch dos ot eel sy inh ine how the pos Soifouonly compare dso he econ ten ou wl e nse improe the sc, ad wl ene ee with le at so was of meting yout ga. hen erent le i cnpacdtan :Peary ame trge h binay word ew which seomqueee ofthe pecixton spptsch wil avays cent tempaon to mk the tae lok forbes And dono asus ear than woking to ingmreteems eapre ‘Thercre wlile i pecan approach will you whe you sree atl} ¥u ow you porte adit wil nel you fo f+ out ofthe nin whch Jou find ourcle Whi the setfeoon :pprouch may be wei when the pecans efoto Ge ‘ered for panning the aprsch becomes merges hen thee atone aby george COMPARISONS TO AVERAGES fe There ate some figures for which the only posible speciation is perfection. We want no accident. We want no spill, We want no workers injured. In such ege is imponible to compute a percent di ference fom earget—even computers cannot divide by zero. Soin oder Understanding Variation Managing Chas to have a contextual comparzon, you will often find the eutent values compared to an average val, . ‘When values are compared oan averags itt ot uncommon to heat 2 manager as for an explanation than, ay five percent fom the average, OF course, this is just sophia ‘ated way of saying tht one should expece all ofthe values to fll at, oF very close to, che average. I enough prestue i applied, chose responsible for coning’ip with the “explanations” wil tke ep to avid having co do so inthe future, They will begin so practice one ofthe wo simple “cover your anatomy” techniques” tir he stem or distort the dit Also like the specification approach, the average value approach has te outcomes. You wil find the cuzren value tobe either “above aver hier the erent value varies more 182" or “below average.” Ofcourse snce the average i genecally neat the -mid-poine of asx of data, you should expect to be above average about half the time, and to be below average about half the cme, A flue «0 appreciate this simple fact often leads co headlines familiar coals “Half the — in th ete are below average.” Whee tis the reading sors for seudents, the reine: office deparkment, or the brakes om dump tmucks, half of them are bound to be below average on any given day of week Ava consequence, the averge value approach will make people feel good abouc half the time, and i wll make them fel bat sbout half the time. So, the average value approach ro interpreting data has about the same consequences as the specification approach it has a binay ou ‘come and it yields some knowledge of where the curren value sand fel- ative toa fixed reference point. While che speciation approach com- pares she current value with the Voie of the Customer, the average vl ‘approach compares che curtent valu with «value from the proces ite Unfortunately, che average vale is only par of the Voive of the Process Thies why che average value approach cannot flly convey the ifr, tion concained inthe cutent value C Ts / Kowlee Onder and Cuma SHEWHART'S APPROACH TO INTERPRETING DATA ‘We anlyae numbers in det to kaow when a change as acute in cour proceses of systems. We want to know about such changes in 4 tidy manners hat we can respond appropsnely. While this sounds rather saighfrbrd thee is complication the numbers an change ven when our proces does ot, Soin our anal of tubers, we nd tohave away to diinguishchowe changes inthe nrnbers hat present Changer in out proces from hoe that a exnill oie. Te this end Walter Shewhacr nade 3 cucldinction Deween to ype of vation inthe numbers. Some vration is routine run-of che tl, and wo be expected even when the proc ha nt changed Other ‘arson it exepional, ouside thé bound of routine, and therfore ro beinerpreted ara signal of proces lange. And inorder fo separate variation nt thee eo components he created dhe consol chan (wich 1s T noted in che peice, i now bing ile a proces behavior cha). The proces behavior chan begos with the data plowed ina ie series. A cent lines added a «visual reference for detecting shifts ot trends, and limits are computed from the data. These limits ate placed symmetaly on either side of the ental line that ditance which wll dow them fle ou vial all of th oie variation, : eye ir tte : Senter byte Rowen reenter Figure 25: Process Behavior Char for Diy Percentage of Defective Pass B - C Understanding Veriton / Managing Chaos ‘The key to the effectiveness ofthe process behavior chasis coneined in the way in which dhe limite ae computed from the data. (These eae lations will be described inthe nex chapter) By characterizing the extent ‘of routine variation, the limits on a process behavior chat allow you to diferenciateberween routine variation and exceptional variation. I, over a reasonably long period of ime, ll of the point fill within the limite of a process behavoe chart, and ifthe point ate well-bchaved, chen the pro= cess can be said ro display nothing but rouine variation, When his hap pens the process canbe thought ofa being predictable within those limm= its, and i is reasonable vo expect shat, unless something is changed, ic wll ‘continue 0 operate this way in the Future. Thu, the limite om a process bbchavior chart allow you to characterize the behavior of your process 4¢ predictable or unpredictable, and define how much routine vacation you should expect inthe furare. Since prediction i the esence of manage ment, this abiley to know what to expect when a proces is behaving predictably is invaluable, However, when point fill outside the limits of proces behavior chart she are interpreted assigns of exceptional vatition. Exceptional variation isatibuted co asignable causes which by definition, dominate the many common cuss of routine variation, Therefore, when a process displays exceptional variation, i will he worchile see to identi the assignable cause of hae exceptional variation, 0 that you ean climinate is effect upon your process. In other words, the presence of exceptional variation i sgaal chat there ate dominane cause-and-effect elationships which affec your process and which you ae no effectively controling Since you would tot Keowgly allow + dominant cause-and-effect rela tionship to exist without attempting 9 control or compensate for iin some manner, is reasonable to say tha exceptional variation isan in cation of assignable causes chat have exaped your arension. ‘Thus, by separating coutine variation from exceptional varatios behavior chart allows you to learn abour dominant cause-and-effect rel tionships chat may have been overlooked inthe past. As you take ation to eliminate the effect of asignable causes from your process you will the process C ‘Tio / Krstlage Ondery ant Cumulative Plt yearn elitr lve Figure 26: Process Behavior Char for Daly Percentage of Defective Pls find chat your proces wil operate more consistently, more prdicably, and more reliably.” Therelore the proces: behavior chart can be wed (0 rain dhe esental informacion needed ro improve your process when it is subject ro exceptional varistion, “The proces behavior char in Figure 2.6 shows a time series which cons of 67 consecutive pointe. The fact that the values remain within ‘he computed limits and the fc that theve i 99 obvious rend, noe any this long requence of points above ot Below the cent line, suggests proce may be considered to be predictable. Based on this chat, it ‘would appear tht wna the proce ir chenged in some fundamental way. the plane will continue go produce anywhere from 7% defective w 30%,” dsfaive ech day, with along-zcem average of about 19% defective ‘Notice how the process behavior chart has helped to interpret the daca Fie the caret used to characterize the behavior ofthe datt—ace they predicrable or not? Second, the pracers behavior chact allows the manager to predict what wo expec in the Fucure—the amount of outing ‘ation defined by the lie the Voie ofthe Paces aq Finally, notice che diflerence bensees the president's inerpretation of shete data and the interpretation based on the proces behavior char. Somedays only appeared ta be beter than other! Thirty-seven daye were “woue than average” (he above 18.7%), and 30 days were "beter than sverige” uc the proce sons na evidence of any changes daring he past 67 doy In wth, both the “good” days snd the “bad” days came fom the a Lnderstonaing Veriton Managing hace ‘same proces. Unles, and uni, chs undesying proces is changed in some fundamental manner, the president will coninge to pls values hich average about 19% defecive. Looking for differences besween the "good" days and the “bad” days wil simply be a wast of time, UNDERSTANDING THE TRADE DEFICITS ‘The USS Trade Defic forthe fee en months of 1987 ae given in Figure 27. In ths peti ehe deficit gor wore (increased) relative tothe preceding month six mes, and ic improved (decreased) only thece times. ‘While the year started with a deficit of 107 billion dalle, by Ocsober this had worsened oa dfc of 16 bilon dels Surly hija «tion for eloom and doom, ed TA ti BG Me ts kd Soe me BS ‘igure 27: Time Series Graph for US. Trade Deficit 1987 ‘As each of ches values was reported they would invariably be accom panied by statements like “the U.S. erade balance defice increased (or decreased) last month to value of — billion dollars” According to the news, the trade balance is always incieaing or decreasing, It arly ever sys the same, But how much of chs churning around is spay and how mach of tis jus nae? ‘We begin by placing the data for 1987 on a process behavior chart. ‘The use of one year's worth of daa is exzentilly arbitery, but we have historically used the calendae to arbiterly subdivide all sors of time 2s Tw / Knowledge Onder ant Camuatioe “4 v Tim RD Mir hoe May eM Aug Sp O8 Nov Om ¢ ‘ Figure 28: Process Behavior Chart for Monthy US. Trade Deficits in 1987 series, and we hal, no dou, continue todo 30 inthe farure. While ic may be habical, cere is nothing magic abou the use of year's worth of Anta The average defce for 1987 was 12.75 bili dollars. Using the technique which is descibed in the next chapter, ic can be sen that, thse on the amoune of month-to-month variation, the dec could vary fiom 8.32 billion o 17.18 billion without representing a real departure 1 the average of 12.75 billion "The chart in Figure 2.8 shows né evidence ofa sustained trend. The efit are noe systematically geting beuet, nor are they systematically ‘grting worse For the year ae whole, his chart shows no clear-cut evi- ‘dence of change. Some monthe appear ro be better than others, bu his indicates that it willbe a waste to analyze any one month to see What is different from preceding months. You shoald wae all the ‘months of 1987 2 if they came from the same system. The daca for 1988 are shown in Figure 2.9, There values could be plowed agsinet the limite shown ia Figure 28 above, This is done in Figures 2.10 and 2.11 he Sp Ov Bw oe jn Fe Mw Aw May J 3968S lions) Figure 29: Monthly U.S. Trade Defi Understanding Variation / Monging Cas TEMAWSJASOND] Fw Figore 210: Process Behavior Chat for U.S, Trade Deficits, 1987-early 1968 Figure 2.10 shows that by March af 1988 shere was definite evidence ofan improvement. The March dfcc is Below the lower limit of 8.32 Before a single month ean be sid to signal a change inthe time series, ‘har single value mast go beyond one ofthe two linia. This happens ia March of 1988. Now that we have definite evidence of a change, how do we interpret the char?’ One method is to look atthe sequence of points adjacent othe point outside che imie which ae aio onthe same side of ‘the central ine as the poine outde the limit. This sequence is shown in Figure 2.11 ‘The incespetation ofthis sequence could be exprese es follows. A change i erly indicated in March of 1988—i may have begun ay caly a8 November of 1987-—and it continued thioughout the test of 1986, Thus, there is definite evidence char che cade deficic improved duting Figure 211: Process Behavior Chart for US. Trade Deficits 1987-1988 C ‘Tuo / Kowal Onder ond Comte 1988, compared with 1987. We could now recompute limits for 1988, and wee dhem Co evaluate Further monthly values foe signs of improve= ‘ent or deterioration, THE SECOND PRINCIPLE FOR UNDERSTANDING DATA The proces behave chac approach ro the aa of data is more owed chan cher the speciation approach or the average value {oproach. Italo is findamenly diferent, Instead of atemping {ich a meaning 10 cach nd every specif vale ofthe ime sre the proces behavior chart concentats onthe behuvior ofthe undedyng proce. Iti therfore, more fondant and more compichentve This is why the proces behavior chat jel ove insight and peer understanding than che specification approach or the average value approach. ka The process behavior char uses the ime series define the Voice of the Process. Ie also gives the user a way 10 know whether iis safe co ‘extrapolate into the near Furuse. Moreoves,whcnéver ii easonable to make this extrapolation, the process behavior char alo defines the ange of values that you are likely to see inthe near ftuce, The specication approach and the average value approach do none of these ching, The process behavior chart does all of chese things Because ic cakes, ‘aration inte scout, Variation is the nandem and mscelancous compo: nent shat inderis the simple and limited comparvons. “The “noise” in ‘roduced by routine variation is what confses and coisa compations bowen single values. nil you can allow fr the nose in atime seis, You cannot fully understand just what may be indicated by a single value, 1s the current value 2 “sgoal” that something has changed, or docs the ‘uttene value differ from the'haroric average by nothing but “nose”? » c The answer to this question isthe exvence of making sense of any value Understanding Varition/ Managing Cacs {roma time series Of course, before you an ever hope to answer this questioa, you wll have to have some historic values. Next you will have to ie thee his- tori values to determine jst wha the effets of “noite” might befor this Particular time seis. Then, and only then can you begin to diferentate berween the noise of routine variation and the signals of exceptional var ation. This whole proces may be summatizd in the second principle of “understanding and communicating with data, ‘THE SECOND PRINCIPLE FOR UNDERSTANDING DATA | wide ery dase conan nae, some det set may conan signal Therefore, before you can detect a signal | wisin any given dataset, | A mt fist fier ot the nase ‘The process behavior chart filters ou the noise of routine variation by the construction ofthe limits. Signal of exception variation are ind ‘ated by poins which fill outside the limits or by obviows non-randony patterns ofvariaon around the ental ine This distinction berwcen signals and not ithe foundation for evesy ‘meaaingful ana of data. Ie also defines the wo mistakes which you fan make when you snale dat The fist mistake is chat of interpreting routine variation as a mean ingfl deparure fom the past—interpreting nite as fi wer «signal ‘Since this mistake wil lead to actions which are, ae best inappoopsace and at wort, completely contrary tothe prope couse of vcton the ms, take is a source of ware and los every day. The second mistake consists of nor recognizing when s change has ‘curred im a processing to detect a igual when iti proent This 0 C Ton / Kalas is Orel and Comaatioe mistake is most often found in conjunction with the specification approach w analysis. The undedyin proces changes, but che values are ssl within the specication limits, 1 no one nocces Cleary, you may avoid the frst mistake by never reacting to any value a fit were a signal, buc this will guarantee many mites of the second kind’ Likewise, you may avoid the second type of mistake by reacting ro evry point as fie wee asgnal, but shir guarantees many mie cakes of the fist kind. yn pa “Those who do not make the disinction between signals and noise wil inevitably be bssed in one dretion othe other. So they wil make snore mistakes of one type oF the other. The process behavior chart swikesabalanee between these rwo eros. ‘The use of the limits to Her cout the noise of touine vrition wll minimize the verurenes of both . This is why those who do nor use process behavior charts to analyze ac wilalvay be ata disadvantage compared to those who do, Unles and until, you make th discon beeen signals and noe you wl temnin unable propely aaa ad inept dats The Scood Pri ciple of Understanding Daa show why every eoive dee analp !in by separating the potential signal rom che probable nae And proces behavior chats ae the simplest method tat hat evs eget vented eparateprenialsigals ftom rouble no Before you can we dat justly any actin, you mt be able wo + potential signal within she data Otherwise you ae key wo be note. Nob ds and lice atic on xx tom 4 interp 0 why should you try co sun your business by listening to, and tempting to incerpet, static?

Вам также может понравиться