Statistical Education for Engineers: An Initial Task Force Report Author(s): Robert V. Hogg Source: The American Statistician, Vol. 39, No. 3 (Aug., 1985), pp. 168-175 Published by: Taylor & Francis, Ltd. on behalf of the American Statistical Association Stable URL: http://www.jstor.org/stable/2683924 Accessed: 24-11-2015 10:16 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Taylor & Francis, Ltd. and American Statistical Association are collaborating with JSTOR to digitize, preserve and extend access to The American Statistician. http://www.jstor.org

This content downloaded from 163.117.20.121 on Tue, 24 Nov 2015 10:16:11 UTC All use subject to JSTOR Terms and Conditions StatisticalEducation forEngineers: An Initial Task Force Report ROBERT V. HOGG et al.*

1. Identifythe customersand theirspecific statistical A three-dayworking conference to developrecommenda- needs; tionsfor the statistical education of engineerstook place at 2. Outlineindustrial training programs for practicing theUniversity of Iowa in July1984. Althoughother areas, engineersand a reasonablefirst course for undergraduate suchas business,chemistry, and biology, are also important engineeringstudents to satisfythese needs, based on the to improvingthe productsof industry,we restrictedour clear identificationof theneed forapplied ; considerationto engineering.This reportof the findings 3. Identifyresource materials and definepedagogical addressesthe following questions: Who are thecustomers? approaches,as possible,to makethese programs effective. What are theirstatistical needs? What typesof courses, Of course,given thatthe audiencefor these programs given by the industryfor practicing engineers (called in- consistsof engineersand appliedscientists, statistical ap- house courses in thisreport), are desirable?How do we plicationsshould complementand reinforceengineering approachstatistical education for engineering management? conceptsand methods,particularly in laboratorypractice, Whattypes of resourcesare desirablefor instructional pur- developmentand assessmentof engineeringmodels, col- poses? Whatis the best firstcourse in statisticsfor engi- lectionand interpretationof data, productand processde- neeringstudents? How can we convincethe engineering sign,and productionand processcontrol. professionthat statistical ideas and methodscan help im- To utilizethe training that occurs effectively, engineering provequality and productivityin Americanindustry and managementmust understand and supportthe use of statis- thusits competitive position in theinternational market? ticaltools. Withoutthis understanding, there is a real pos- sibilitythat the statistical techniques will not be used.Although the conferencewas not able to focusextensive efforts on 1. INTRODUCTION howto buildsuch management awareness, there was a clear Therehas beenmuch publicity in recent years concerning consensusthat statistical education for engineers must also Americanindustry's inability to competesuccessfully in the includesome educationfor their managers. internationalmarketplace. Many of theseconcerns focus on One surprisingaspect of thework of theconference was improvingquality and productivity and thus the competitive thedegree to whichall participantsagreed-despite diverse positionof Americanindustry. To addressthese concerns experienceand occupations-on the basic statisticalcon- effectivelyrequires contributions from many disciplines. ceptsthat all engineersneed to appreciate.These include AppliedStatistics is one such discipline.It providesengi- 1. The omnipresenceof variability neerswith such usefultools as statisticalquality control, 2. The use of simple graphicaltools, such as histo- designof , graphical data analysis, and powerful grams,scatterplots, probability plots, residualplots, and buteasy-to-use statistical computer packages. Yet manyof controlcharts thesetools are notbeing used in appropriateplaces. How 3. The basic conceptsof statisticalinference can thisbe in an age whentough worldwide competition 4. The importanceand essentialsof carefullyplanned leavesus littlechoice but to takeadvantage of all technical experimentaldesigns tools available?The answeris, in part,that American en- 5. The philosophiesof Shewhart,Deming, and other gineershave not been taughtabout these tools-and this practitionersconcerning the manufacturing of quality prod- situationneeds to be changed. ucts Withthis conviction, 42 statisticiansand engineerspar- ticipatedin a three-dayconference on statisticaleducation For manystatisticians and engineeringeducators, this forengineers at theUniversity of Iowa, Iowa City,July 23- reportreinforces the realizationthat an interdisciplinary, 25, 1984. cooperativeapproach is necessaryto respondto the chal- Our goals in thisconference were to lengesthat industry faces. It is imperativethat a dialogue betweenthe engineering and statisticalprofessions take place if majorgains are to be made in qualityand productivity. *RobertV. Hogg is Professor,Department of Statisticsand Actuarial Someof us atuniversities believe that there is littlechance Science,University of Iowa, Iowa City,Iowa 52242. The authorwants it madevery clear, however, that the other 41 participantsare also coauthors of persuadingacademic engineers to includemore statistics of thisreport; they and theiraffiliations are listedin the Appendix.In in theircurricula unless industry unmistakably demands it. particular,the team leaders H. T. David, WilliamGolomski, Bert Gunter, It is thereforeessential that influential industrial leaders and WilliamParr were extremelyhelpful, as was RichardGunst, who promotethe importanceof statisticalideas and methods. helpedprepare a preliminarydraft of thisreport. In addition,two other Certainlysuccessful applications provide our most con- statisticians,who did notattend, should be mentioned:Ron Snee, forhis suggestionsin theplanning stages of theconference, and GerryHahn, for vincingargument. This reportprovides some suggestions his recommendationsthroughout the entireprocess, particularly in the on what should be includedin these programs.It is an preparationof themanuscript. importantfirststep, and everyopportunity should be taken

168 The American Statistician, August 1985, Vol. 39, No. 3 ?) 1985 American Statistical Association This content downloaded from 163.117.20.121 on Tue, 24 Nov 2015 10:16:11 UTC All use subject to JSTOR Terms and Conditions to spreadthe word among practicing professional engineers ing,and aid in communication;Process flow diagrams; and and engineeringeducators. We invitereaders to sharetheir Paretodiagrams that quantify the priorities and list the "vital commentsand experienceswith us by contactingHogg. It few" keysof effectiveefforts. is clearthat the same type of conference concerning business 2. Fundamentals(Fund.) of data variabilityand basic statisticswould also be worthwhilefor the improvement of statisticalmethods for process understanding, control, and qualityand productivity.In anycase, progressin statistical improvementare absolutelynecessary. The engineermust educationshould be an ongoingeffort. understandwhat is observed:the actual value plus some randomvariable (noise). To improveproductivity and qual- 2. STATISTICAL AREAS IMPORTANT ity,reduction in the variabilityof theprocess is required. TO INDUSTRY Accordingly,there is a greatneed to have theprocess gen- eratedata that can be usedto monitor,control, and improve thatthe conference addressed was, Who Thefirst question theprocess. are thecustomers and whatare theirstatistical needs? The 3. Designof Experiments (DOX) indicateshow to probe answersresulted in twolists that then formed the rows and a processto getthe most information at minimalcost. We thecolumns of a matrix,the general element of whichre- shouldstress the interactive nature of experimentationand flectsthe amountof knowledgeneeded by thattype of howthis approach fits into the strategy of ongoing improve- engineerin thatparticular statistical area. Of course,the mentand process learning. That is, itshould be emphasized specificsof these classifications vary from company to com- thatstatistical methods are consistentwith and essentialin pany; but the participantsagreed that these-as generic the applicationof the scientificmethod. Topics to be in- distinctions-arebroadly applicable even though some spe- cludedin DOX includemeasurement error, independence, cificclassifications have been omitted. replicationversus duplication, randomization, , main effects,full and fractionalfactorials (as comparedto the 2.1 Customers poorerone-at-a-time approach), , , Customerswere categorized along the following lines: responsesurface design, product robustness, nested designs and variancecomponents, evolutionary operation/plant ex- 1. Engineeringmanagement is responsiblefor the work perimentation,mixture designs, and acceleratedtesting. performedby thestatistically trained engineers. 4. StatisticalInference (S. Inf.)concerns the error struc- 2. Qualityengineers are responsiblefor product, ma- tureof an estimate.That is, engineersshould know the basic terial,and processquality and forongoing monitoring of conceptsassociated with distributions that explain reliabilityof themanufactured product. why intervalestimation and testsof hypothesesare suc- 3. Manufacturinglprocessengineers (Mfg/Proc) are re- cessful. sponsiblefor the manufacturingprocess design,improve- 5. GraphicalTechniques (Graph) are extremelyimpor- mentin process productivity and performance, process control, tant,not only for the engineer's use, butin presentingma- troubleshooting,and so on. terialsto others. 4. Designengineers are responsible for product designs and settingtolerances of newproducts. a. Single-samplemethods consist of stem-and-leafar- 5. Productengineers are responsible for design changes rays,box plots,, dot plots,and modelfitting. in responseto costs,quality, and manufacturingneeds as- How wellsome of these probability models fit can be checked sociatedwith existing product lines. by using normalprobability paper (more generally,q-q 6. Reliabilityengineers are responsiblefor the reliabil- plotting).All of thesemeasures of sampledistributions can ityengineering of newproducts and processes. be used to checkprocess capabilities against specifications. 7. Researchand Developmentengineers (R&D) are in- Othereffective tools to describethe processsignature are volvedin researchfor new conceptsthat will lead to new classical controlcharts (xT, R, p, c), as well as moving productsand processes. averages,CUSUMS, and statisticsfrom newer multivariate 8. Test engineersare responsiblefor development of techniques.In regardto the latter,more researchcould testmethods and testequipment for both new and existing probablybe done to developnew robustand multivariate products. methodsfor helping find assignable causes and unnatural patterns. 2.2 StatisticalAreas b. Relationshipsamong multiple samples can be found In listingthe statisticalneeds, we recognizedthat there by usingsome of themethods listed in single-samplemeth- aresome overlapping areas. In orderto clarifythese issues, ods, particularlyhistogram types of comparisonsand q-q we will makesome fairlydetailed remarks about some of plotting. thesetopics. c. Multivariatemethods, such as scatterplotsand mul- tiple box plots, are extremelyhelpful and easy to do if 1. ProblemSolving (P. Sol.) involvesan orderlystep- appropriatecomputer software is available. by-step"engineering" approach to improvingprocess un- The lastfour areas are a littlemore specialized. Although derstanding.Structured, cooperative efforts unlock many not as importantfor all engineers,they are extremelyim- resourcesand allow persons with diverse viewpoints to work portantin certainindustries. constructivelytogether using the scientific method. Possible toolsthat can be usedare Fishbone or Ishikawacause-effect 6. Modeling(Model) includes linear and nonlinear diagramsthat organize information, structure understand- regression.

The American Statistician,August 1985, Vol. 39, No. 3 169 This content downloaded from 163.117.20.121 on Tue, 24 Nov 2015 10:16:11 UTC All use subject to JSTOR Terms and Conditions 7. Life Testingand Reliability(Life) includesacceler- variationand quantitativeassessment of qualityto theen- ated lifetesting. gineer.The topicsto be coveredare 8. SurveySampling (Surv.) involvesefficient collection 1. Motivation.Why should we care? Providemotivat- of data thatcan be used to describethe characteristics of a ing examplesthat are constructedin the workcontext of population. thosebeing educated, indicating the large number of prob- lemsamenable to attackby statisticalmethods. This partis 9. AcceptanceSampling (Acc.) includesnarrow limit crucial. gauging. 2. The measurementprocess, including questions about The matrixthat gives our evaluationof the knowledge howto decidewhat to measure,levels of measurement,and neededin each ofthese areas by different types of engineers so on. Properplanning to obtainvalid data is extremely in industryis givenin Table 1. Thismatrix, with its column important. totals,provides a reasonableordering, in our opinion,of 3. Cause-effectdiagrams, including one suchdiagram theimportance of thesestatistical areas for engineers work- producedby each personin theclass ing in industry.There is a distinctbreak in thescore totals 4. Paretodiagrams indicating the ordering of priorities betweenthe first five statistical areas and the last four. Again 5. Histograms,stem-and-leaf arrays, and box-and- we emphasize,however, that even thoselast four could be whiskerplots extremelyimportant to some engineersand certainindus- 6. Variationand itsimplications for making inferences tries. withfinite information. The differencebetween a parameter and itsestimate is stressed. 3. POSSIBLE IN-HOUSE COURSES 7. Basic xi,R, andp chartsand theirinterpretation 8. Introductionto conceptsof statisticalinference and Thereis no singlecourse or set of coursesthat will be blockingthrough independent and dependentt tests,in- suitablefor all engineersand all industries.Hence it was cludingfurther discussion of independence impossiblefor us to constructa genericprogram. But there were some pointsthat the participantscould easily agree To deemphasizecalculations, we stronglyurge the use upon: of a computerin thecourse, particularly to get convenient graphicaldisplays. 1. The instructionshould be drivenby data and case We suggestroughly six half-daysessions for this course, studies,not topics. There is a greatmotivation for engineers whichincludes roughly one-third of the timebeing spent in takingreal data, particularlyfrom their industry, and on studentactivity. These sessionsare to be spaced about weavingin the statistical ideas. Computers,with appropriate one week apart,and involvementin the sessionsby the software,should be available. immediatesupervisor is highlydesirable. The requiredout- 2. The statisticalconcepts should be motivatedby real side workwould be thecreation of a cause-effectdiagram. problemsrather than taught in a logicalmathematical order. Anothershort project, if possible, would be desirable. 3. Because the amountof usefulinformation is large, Homeworkof a traditionalsort could be providedfor those itis clearthat the in-house statistical education for engineers who wantit and thenused forexamples in succeedingses- shouldbe ongoingand modular.These moduleswill be sionsas review. determinedby theneeds of thevarious industries.

Withthese points in mindand knowingthat we cannot 3.2 StatisticalProcess Control constructone set of coursesfor all engineers,we outlined threeshort courses, possibly of greatestinterest. The courseon some elementarytopics in statisticalpro- cess control(SPC) has thefollowing outline: 3.1 IntroductionTo Applied Statistics 1. Basic controlcharts: ix and R, x-and s, p, u, c, and The firstcourse, designed to be a prerequisitefor the so on. Thereis a tendencyto concentrateonly on theme- others,has as its goal the introductionof the conceptsof chanics;these should be minimizedbecause few will ac- tuallyfind the mechanics difficult. Emphasize interpretation, set-upissues (especially subgroup determination), and con- siderationof assignableversus random causes. 2. Processcapability assessment Table 1. DesirableStatistical Knowledge 3. Statisticaltolerancing StatisticalArea 4. Operatingcharacteristic curves for control charts Engineer P. Sol. Fund DOX S. Inf. Graph Model Life Surv. Assoc. 5. CUSUM charts 6. A briefintroduction to the of Management 3 2 1 1 3 0 0 1 1 concepts acceptance Quality 3 3 3 3 3 2 2 3 3 sampling Mfg/Proc 3 3 3 3 2 2 1 0 1 Design 3 3 3 3 2 1 3 1 1 7. Analysisof is suggestedto bridgethe gap Product 3 3 3 3 3 1 2 2 1 fromSPC to classicalinference. Reliability 3 3 3 3 2 3 3 2 2 R&D 3 3 3 3 3 3 1 1 0 Roughlyeight half-day sessions are needed.There is a Test 3 3 3 3 2 0 0 0 1 Total 24 23 22 22 20 12 12 10 10 requirementof a job-relatedproject that is to be completed by each student.Other optional homework can be assigned NOTE: 0 = no knowledgenecessary; 1 = understandingconcepts, not mechanics; 2 = conceptsand some mechanics;3 = conceptsand technicalcompetence. and used as reviewfrom one sessionto thenext.

170 The AmericanStatistician, August 1985, Vol. 39, No. 3 This content downloaded from 163.117.20.121 on Tue, 24 Nov 2015 10:16:11 UTC All use subject to JSTOR Terms and Conditions 3.3 PlanningEffective Experiments Table 2. Levels of Awareness

This courseshould include issues of hypothesestesting Management Level and confidenceintervals as relatedto the analysisof de- Upperl First-Line Engineers signedexperiments. Some basic designs,along withcon- Concepts Top Middle Supervision (Consultants) cepts such as interactions,are considered.Consumers of Decision the course will not necessarilybe able to plan effective strategy 3 3 3 3 experimentsbeforehand and thencarry out theanalysis of Planningand theirown without statistical assistance. Suggested topics are implementation 3 3 3 3 Naturalvariation thefollowing: of data 1 2 2 3 Design 1. A quick surveyof thebasic ideas of confidencein- strategy 1 2 2 2 tervalsand testsof hypotheses Modeling 1 1 2 3 Response 2. Variabilityof experimentalresults and samplingdis- surfaces 1 2 2 3 tributions Graphics 2 2 3 3 3. One-wayanalysis of varianceand multiplecompar- Viewpoint Whyto Whyto What,When, How to Where isons,including graphical displays Total exposure 4. Fixed and randomeffects, including components of time (hrs.) 2 8 48 120+ variancemodels (nested designs) Typical contact time (hrs.) 1/2-1 4-8 1-8 1-2 5. Two-waydesigns, including randomized complete = = blockdesigns and theconcept of interaction NOTE: 3 Highest/Critical;2 Medium/SomeDepth; 1 = Lowest/Surface. 6. Full and fractionaldesigns, emphasizing that all var- iablesare notcreated equal 7. Responsesurfaces 2. Planningand Implementation.The definitionof the At least eighthalf-day sessions are needed.In a longer objectiveor targetsto be met,nature of themeasurements, the courseor in a subsequentcourse, we couldconsider incom- selectionof appropriatediagnostic methods, the action plete block designs,regression models, Taguchi-type de- plan, and follow-up signs,and so on. Otherswill find extensions to the Statistical 3. NaturalVariation of Data. Real lifedata are shrouded ProcessControl course to be of greatinterest. Courses or by a "sea of fog." Numerousuncontrolled and unknown factors observation programsin specialtopics such as evolutionaryoperation, influencethe process. multipleregression, reliability, and acceptancesampling 4. DesignStrategy. The use ofrandomization, blocking, shouldbe providedif appropriateto the demandsof the and othertechniques to isolatesources of variation particularindustry. 5. Modeling. Approximatingthe real worldby using simplemathematical descriptions 6. ResponseSurfaces. Contour map representationof a 4. GUIDELINES FOR model MANAGEMENT EDUCATION 7. Graphics.Visual displayof information We thoughtit appropriateto say somethingspecial about We feel thatboth management and engineersshould be engineeringmanagement because we wantto stressthe im- exposedsimultaneously to thistraining. There may be some pact of statisticson everythingfrom corporate strategy to benefitin havingmanagement start before the engineers to evaluationof experimentaldata. We recognizeat leastthree giventhis entire process a top-downflavor. By and large, levelsof managementthat have differentlevels of control, trainingshould be done in one-or two-hourdoses. We also spheresof influence,and decision-timehorizons. They re- see a crucialneed forindustrial case studies,across com- quiredifferent treatments. Top-level management, like the panies,industries, and so forth,to supporteach concept.A divisionpresident and the generalmanagers, need to un- book such as Statistics.A Guide to the Unknown(Tanur derstandthe impact of theuse of statistics.At the next level et al. 1972) wouldbe mostappropriate. of management,the managersneed selectedexposure to whatboth top managementand first-linesupervision have 5. RESOURCES FOR IN-HOUSE PROGRAMS received.First-line supervision has the actual functional The widespreadimplementation ofquality in-house train- responsibilityfor the workperformed by the statistically ingprograms in statisticsfor professional engineers requires trainedengineers reporting to them.In all cases, we rec- the availabilityof a varietyof resources.As used in this ommenddealing with concepts and issues first. report,the term "resources" encompasses the means whereby We displayin Table 2 therelationship among manage- statisticaltraining is providedto practicingengineers through mentlevel, concepts,viewpoint, and level of awareness. anyof severalmodes of instruction.The mostcritical need Awarenessexpresses the depth of understandingof a con- is fora well-trainedstatistician, ideally one who is expe- cept froma particularviewpoint. To help read thistable, riencedin the specificengineering focus of the company we definein the broadestof termseach of the statistical and able to communicateeffectively in a classroomsetting. conceptsthat we feelmanagement must grasp. No attemptis made to discuss instructorsor theirqualifi- 1. Decision Strategy.Trade-offs between costs of data cations;rather, the primary focus is on writtenand visual and risks materialsthat can be used to enhanceinstruction.

The American Statistician,August 1985, Vol. 39, No. 3 171 This content downloaded from 163.117.20.121 on Tue, 24 Nov 2015 10:16:11 UTC All use subject to JSTOR Terms and Conditions 5.1 Formal InstructionalMedia of articlesthat distinguish statistical principles of experi- mentationand processcontrol from the deterministic train- Based on the principlethat there is no singlemode of ing given engineersin colleges and universities.Product instructionthat will sufficein all industrialenvironments, developmentand processengineers should have references threemajor modes of instructionhave been identifiedfor thatinclude a detailedstatement of theproblem under in- special consideration:traditional (but possiblymodified) vestigation,discussion of the experimental design, data list- classroominstruction, videotape instruction, and special- ingsand displays,and carefuldiscussion of theconclusions purposereference materials. In orderto be effectivein an downfrom the data analysis. industrialsetting, each of thesemodes of instructionmust be speciallytailored to itsintended audience. In generalwe recommendthat all formalinstructional media containa 5.2 Hands-On Materials clear statementthat designates (a) the course objectives, The use of hands-onmaterials is intendedto allow en- (b) theintended audience, (c) thespecific subject matter to gineersto gain experience with statistical techniques in those be covered,including the engineering specialization at which instanceswhen a workingknowledge of thesetechniques is thematerial is directed,and (d) thecompetence level that needed. Threepossible methods of achievingthat experi- will be achievedat theend of the instruction.In addition ence are listedhere. to thesegeneral recommendations that apply to all three modes of instruction,specific recommendations are pro- 1. In-HouseExperimentation. Instructor involvement in posed foreach . actualin-house experiments is often the most desirable means of ensuringrelevance and achievinga satisfyingexperience 1. Textbooksand CourseSyllabi. Textbooks and course withstatistical methods. The reportingof past successful syllabi for trainingprograms should emphasize practical uses of statisticalmethodology is an importantmotivational approachesto the solutionof engineeringproblems. For factorin the acceptanceof statisticalprocedures by prac- example,topics on experimentaldesign and the natural var- ticing engineers. iationof experimentalresults should be introducedearly in 2. Case Studies. Case studiesshould be realisticbut the instruction.Formal discussion of probabilitydistribu- should not always containall of the problemsof actual tions,usually presented in earlylectures of mathematically experimentation.The realismis neededfor relevance, but orientedstatistics courses for engineers, should be intro- the inclusionof all of theproblems that one mightbe re- duced onlyas needed,with theoretical derivations kept to quiredto face can obscurethe intentof theinstruction. A a minimum.Actual engineering applications should be stressed case studyshould focus on onlya fewconcepts and attempt throughoutthe course. Case studiesshould be an integral to stressonly a smallnumber of statisticalissues. componentof thistype of formalinstruction and shouldtie 3. Computer-AidedInstruction. The increasingpromi- together,where possible, differenttopics covered in the nenceof microcomputersin industrial research opens a new course. avenuefor instructional materials. Data forcase studiescan 2. VideotapeModules. Videotapemodules can be ef- be suppliedon diskettes,perhaps along withappropriate fectivesubstitutes for formal classroom instruction when softwareand graphical displays. This is especiallyimportant qualifiedinstructions are notavailable. Videotape modules whencase studiescontain so muchdata thatan engineer are not to be regardedas preferableto instructionby an would be reluctantto spendthe timeneeded to enterthe experiencedprofessional statistician, but when properly used data intothe computer. Workbook exercises can sometimes theycan providebasic instructionon manystatistical topics. be moreexpeditiously solved via microcomputersthan by We recommendthat several guidelines be followedin the usingcalculators, thereby allowing more time to concentrate creationand use of videotapemodules. First, each com- on thestatistical concepts contained in the instruction. Like- ponentof instruction should not exceed 30 minutes.Second, wise, individualpractice exercise can be computer-gener- printedmaterials should accompany the videotape.These ated along withnumerical solutions. printedmaterials should include copies of all displaysused in thevideotape, a synopsisof thetopics discussed in the 5.3 StatisticalSoftware module,and study guide containing problems that reinforce Statisticalsoftware should be an integralcomponent in thekey ideas of themodule. Finally, and mostimportant, thestatistical training of professionalengineers. Three rec- videotapesshould be viewedin groups,preferably with a ommendationsare made concerningthe use of statistical resourceperson who can clarifydiscussion points and an- softwarein trainingprograms for engineers: swerquestions about the presentation. A good exampleof video materialis thatproduced by Westinghouse. 1. Instructorsshould have suitablesoftware available 3. Short,Directed Reference Materials. Documents that or have theirown prepared. have limited,focused objectives should be availableas ref- 2. Evaluationsof statisticalsoftware should be pub- erencematerials. These reference materials should motivate lishedperiodically so thatinstructors and engineeringusers by discussingrealistic engineering problems and provide can be guidedto efficient,numerically accurate programs. guidanceof a practicalnature, rather than that of thetra- 3. A suitablemechanism should be foundwhereby an ditionalmathematical statistics texts. AmericanStatistical Association (ASA) sectionor standing Descriptivereferences similar to Statistics:A Guideto committeewould be chargedwith setting criteria for statis- theUnknown (Tanur et al. 1972) mightbe proposedfor tical software.Among the criteria for acceptable software managers.Of importancein this type of reference is a variety mustbe a higherlevel of documentationthan is available

172 The AmericanStatistician, August 1985, Vol. 39, No. 3 This content downloaded from 163.117.20.121 on Tue, 24 Nov 2015 10:16:11 UTC All use subject to JSTOR Terms and Conditions today.Minimum criteria should require appendixes that de- mentation,are best presentedin, say, a one-semesterEn- scribethe computing formulas and algorithmsused to make gineeringStatistics course. Again, wheresuch a semester thecomputations and a detailedexplanation of all termsthat coursecannot, despite its value, fitinto an existingcurric- appearas partof theoutput. ulum,our synopsismay well serveto suggestor reinforce selectedwork units in existingcourses or laboratorysec- These recommendationsare intendedto stressthe fact tions.That is, laboratoryexercises can be usedto introduce thatpracticing engineers, like practicingstatisticians, are statisticalpractices to theengineering students. dependenton statisticalsoftware for much of the design and Because studentsdo not have the same motivationthat analysisof engineeringexperiments. The responsibilityfor thepracticing engineers have, we suggestthat whatever the identifyingquality statistical software should not be shirked formator amountof statisticalmaterial presented, students by thestatistical . oughtto be giventhe opportunity to experience,early and firsthand,the need forand value of statisticalideas. Two 5.4 Interchangeof Information testedways of securingsuch involvement are "black box" Informationon instructionalmaterials is an importantre- experimentsand calibrationexercises. source that warrantsgreater attention by the statistical 1. In the first,the instructorprograms a hypothetical profession.Two approachesare recommended. process,say a concaveresponse to fourfactors plus a noise component.The studentis thenasked to findthe optimum 1. BetweenStatistical and EngineeringSocieties. We operatingcondition in a fixednumber of runs. Students will encouragea greaterexchange of informationbetween ASA value the preferredstatistical experimental plan in direct and themany engineering societies on meetingsand sym- proportionto how badlythey will have flounderedon their posia of mutualinterest. The obvious mechanismwithin own. ASA is an announcementinAMSTAT News. Pressreleases 2. The calibrationexercises illustrate the creationof to engineeringnewsletters on importantspecial-purpose empiricalrelationships in the face of measurementerror, conferenceswould also heightenthe awareness of profes- and theways in whichsuch measurement error is takeninto sionalengineers of developmentsin statisticsthat might be account. of importanceto variousgroups within their profession. An emphasiswithin ASA shouldbe initiatedto approachen- A thirdproven way of involvingthe student, especially gineeringcolleagues who are membersof engineeringac- in thecurrent climate of urgencysurrounding statistics-re- creditationboards about making statisticaltraining a latedcountermeasures to America'seroding trade position, recommended,if notrequired, component in theuniversity is to bringcurrent periodicals, both popular and technical, trainingof all engineers.To help withthis interchange, intothe classroom,as well as otherpress and media ma- statisticiansshould attend meetings of engineeringsocieties terials.In anycase, thestudent must be providedwith proper and presentpapers to make theirpresence more widely motivationearly in thecourse. known. We listsome ideal objectives and methods associated with 2. WithinASA. A clearinghouseshould be established such a course. Clearlywe are consideringclasses of rela- forinstructional materials. Some effortof a similarnature tivelysmall size, say approximately25-30 students,with forgraphical aids and general-purposeteaching materials is thenecessary computer backup available. From experience, alreadyin place in the reviewsections of The American we knowthat the ideal does notalways exist, but at least Statistician,but there is no concertedfocus on engineering we can strivefor these optimal conditions. education.The ASA nationaloffice should investigate fund- 6.1 Objectives ingor thesecuring of fundingfor ASA sponsorshipof con- tinuingeducation programsin statisticsfor practicing Objectivesof the coursewould includegiving students engineers.Two benefitswould accrueto the Association theability to fromsuch an effort.First, as an ASA continuingeducation 1. Plan ,turn data intoinformation, and program,the quality and competency of theinstruction could achieve action be monitoredby thestatistical profession. Second, income 2. Applythe methods taught in real-lifesituations fromthe program could be used to fundthe development 3. Communicatestatistical information inoral and writ- of otherinstructional materials, such as case studiesand tenform videotapes. 4. Use computerand graphicaltechniques 5. Plan, analyze, and interpretthe resultsof experi- 6. ACADEMIC PROGRAM ments Withthe backgroundand the knowledgeof industry's 6. Understandthe scientific method needs, it is our convictionthat there is a body of ideas concerningexperimentation and measurementthat are es- 6.2 Methodsof Instruction sentialcomponents of a modemnengineering curriculum. The methodsof instructionshould Some of theseconcepts arise incidentallyin the present engineeringcurricula. But manyare notrepresented at all 1. Emphasizecase studiesand workshop and laboratory orare not headlined in their own right. Many of these clearly experience. belongto thefield of statistics. 2. Make use of suitable,advanced computer hardware These statisticalconcepts, plus detailsof theirimple- and software.

The AmericanStatistician, August 1985, Vol. 39, No. 3 173 This content downloaded from 163.117.20.121 on Tue, 24 Nov 2015 10:16:11 UTC All use subject to JSTOR Terms and Conditions 3. Requirecompletion of an individualproject, possibly Thisis a veryambitious course, but some of us havetried a designof an experimentalong witha suitablereport. coursessimilar to thisone withsome success. The use of thecomputer the We now describea possiblecourse, having three semes- throughout coursepermits more time for the interpretationof the statistics.We ter-hoursof credit,with about 40 periodsfor lecture and resulting hope that thestudent will appreciatethe of discussionplus fouror fiveperiods for examinations and importance good descrip- tions,modeling, basic statistical and review.Although some theory is givento explainwhy cer- inference, good ex- perimentaldesigns upon the of thecourse. tain statisticalmethods are successful,this is not a theo- completion In addition,we would also expecta reticalcourse, and mostteaching is doneby examples. Here book of case studiesof some of thesemethods to we hopeto be boldand fashion something radically different be extremelyuseful to the in- structorsand students. fromexisting courses, recognizing that the computerwill play a majorrole in statisticaleducation in thefuture. 7. CONCLUSIONS 6.3 A Possible Course forEngineering Students Today's statisticiansknow that our professioncan con- A. Data collection,communication, and presentation(8 periods) tributeto the improved quality and productivity ofAmerican 1. Problemidentification and flowdiagrams industry.Many engineershave not been taught,or have 2. Data collectionand variability been taughtpoorly, the value of statisticalideas and appli- a. Variablesversus attributes cations.The purposeof b. Sourcesof variation(time, product variability vs. measurement, thisconference was to helpcorrect assignablecause vs. randomerror) thissituation by consideringbetter in-house statistical pro- 3. Measurementand error gramsfor practicing engineers and an excitingand more 4. Descriptivestatistics (, , , etc.), relevantfirst course for engineering students. We hopethat notingdifference between the true parameter and itsestimate statisticiansinvolved in theseprograms will startmaking 5. Usefultechniques a. Paretodiagrams thesesuggested changes now. b. Ishikawadiagrams Moreover,to help speed thisprocess along, we need to c. Graphicsand appropriatesoftware have top-levelmanagement support our recommendation d. Histograms,stem-and-leaf arrays, scatterplots, residual plots strongly.Only if employers insist upon these statistical pro- e. Run charts,rational subgroups, control charts gramswill therebe some chanceof persuadingacademic B. Basic probability(7 periods) engineersto includemore statistics in theircurricula. We 1. Models: Deterministicand probabilistic need theirsupport to getour programsmoving-but more 2. Laws of probabilityand theirapplications 3. Reliabilityof productsand systems important,they need our statisticalknowledge to gain a 4. Joint,marginal, and conditionalprobabilities morecompetitive position for American industry. 5. Distributions:Normal, binomial, Poisson, others C. Basic statisticalinference (7 periods) APPENDIX 1. Sampleversus population RobertV. 2. Samplingdistributions Hogg servedas thedirector of theconference, 3. Centrallimit theorem and theother participants were organized into four teams. 4. Reference distributions Team One. H. T. 5. Pointand intervalestimation David, Captain(Iowa State),David 6. Basic comparisons:One mean, two means,one proportion,one W. Bacon (Queens), Daniel Penia(Madrid, Spain), Ron (optional), q-q plotsand normalprobability plots, signal Askin (Iowa), Duane Dietrick(Arizona), Mel Peterson to noise ratios(optional) (Proctor& Gamble),William C. Guenther(Wyoming), David D. Regressionmodels (4 periods) S. Moore(Purdue), James M. Landwehr(AT&T Bell Labs), 1. Simplelinear regression CharlesG. Pfeifer(DuPont). 2. Multipleregression Team Two. WilliamA. Golomski,Captain (Golomski 3. Residualplots 4. Correlation Assoc.),Fred Leone (ASA), StephenVardeman (Iowa State), JohnS. Ramberg(Arizona), John D. Hromi(Rochester Inst. E. Carefullydesigned experiments (10 periods) 1. Objectives(screening, exploration, optimization) Tech.), Ed Mykytka(Auburn), James M. McBride(IBM), 2. Comparisons NeilM. Werner(Corning Glass), A. BlantonGodfrey (AT&T a. Qualitative Bell Labs), Lloyd Knowler(Iowa). b. Classical (one variableat a time) Team Three. BertGunter, Captain (RCA), LynneB. c. Factorialexperiments (interactions and graphicalmethods) Hare (T. J. Lipton),Marilyn M. Hixson(Harris Semicon- d. Fractionalfactorials 3. Conceptsof replications,randomization, blocking reinforced ductor),G. Rex Bryce(Brigham Young), PeterJ. Jacobs 4. Componentsof variance(nested designs with data obtained during (3M), Tom Boardman(Colorado State), E. B. Fowlkes(Bell and afterproduction) Comm. Research),Robert L. Perry(McDonnell Douglas 5. Pitfallsof experimentation(lurking variables, collinearity, inade- Elec.), W. J.Hill (Allied),Betty Stewart (U.S. Steel),Janet quateexperimental ranges) Fiero (Motorola). F. Reliability(4 periods) Team Four. WilliamC. Parr,Captain (Florida), Norm 1. Failurerates Heitkamp(Shell Dev. ), Neil R. Ullman(College of Morris), 2. Failuredistributions EileenBoardman (Colorado State), Bob E. Robertson(Intel), 3. Reliabilitycalculations (series, parallel, k out of n) 4. Graphicalmethods: Weibull plots, probability plots, failure rates CharlesHendrix (Union Carbide), Richard F. Gunst(South- 5. Reliabilitypredictions ern Methodist),Ken Kotnour(3M), Mary Natrella(Nat. 6. Field failureanalysis Bur. of Standards),Steve Zayac (Ford).

174 The AmericanStatistician, August 1985, Vol. 39, No. 3 This content downloaded from 163.117.20.121 on Tue, 24 Nov 2015 10:16:11 UTC All use subject to JSTOR Terms and Conditions The firsttwo teamswere concerned primarily with the (ASQC). The majorfinancial support came fromthe Ellis academicprogram, and theother two concentratedon in- OttFoundation, ASA, theStatistics Division of theASQC, houseindustrial training programs. In addition,two prom- Motorla,RCA, andThomas J. Lipton, Inc. Othercompanies inentconsultants from the engineering societies participated: also helpedwith expenses by payingthe registrationfees Lionel Baldwin,President of NationalTechnological Uni- of theirparticipants. versityand formerDean of Engineering,Colorado State [ReceivedJanuary 1985. RevisedFebruary 1985.] University,and Ed Jonesof Iowa StateUniversity, repre- sentingthe Accreditation Board for Engineering and Tech- nology. REFERENCE The conferencewas cosponsoredby the Universityof Tanur,Judith M., Mosteller,Frederick, Kruskal, William H., Link,Rich- Iowa, theAmerican Statistical Association, and theStatis- ard F., Pieters,Richard S., and Rising,Gerald R. (eds.) (1972), Sta- tics Divisionof the AmericanSociety of QualityControl tistics:A Guide to the Unknown,San Francisco:Holden-Day.

The AmericanStatistician, August 1985, Vol. 39, No. 3 175 This content downloaded from 163.117.20.121 on Tue, 24 Nov 2015 10:16:11 UTC All use subject to JSTOR Terms and Conditions