:Jan 199.1:ABI/NFORM Global Survey of Software with Mixture Experiment Capabilities GREGORY F.PIEPEL Battelle.Pacific Northwest Laboratories.Richland.WA 99352 total of 5ack litie ture ex mer analysis capa os)are smmarized in table form.Mailing addr and faxnumbers of the software vendors are also provide Introduction The dor in three taband were provided to the A two of the s paper.The three tables and their contents are dis Othe amount of the mixtur cussed in the next section. or pr rocess variables (nonmix It was b nd the of my ec software package.or to loo nd gather inform software having mixture ex. rience with several of the sofware packages.andI maires and asked to indicat whether or not their software had 【or des1gI rative mixtur mation about types of li problems with the Performing compar ative review are er e After receiving the vendors'respo es to the ques that ntere the vend or contact was th capahilities they need. To make intormatio Journal of Quality Technology Vol.29.No.1.January 1997 Reproduced with permission of the copyright owner.Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Survey of software with mixture experiment capabilities Piepel, Gregory F Journal of Quality Technology; Jan 1997; 29, 1; ABI/INFORM Global pg. 76
SURVEY OF SOFTWARE WITH MIXTURE EXPERIMENT CAPABILITIES n over the whole simplex (>0=1.2. Quality Technology. Infor accommodate in designing experiments over irregu- pab tware of th Tables 1.2,and 3 contain summaries of the in- linear constrints of the general form A1+A22++Axg+A0>0 (1 able ins inf imental design capabilities,while Table 2 contain are the most common forms of constraints on th mixture component proportions (Cornell (1990. inforntioneraionenodeigandotherdaa ber,hardware thesoferface,and A11+A22+.+Agg+A0=0 e blank.where: lity cor feature n Table ailable but obtain the software checs the con ming language consistency.and provides blank feature not available The component Transformations g However.some entries in the tables are ubers cates the types of transformations that can be im while others are references to footnotes that provide while U-pseudoco -1).(See Cornell (1990. Brief descriptions of the information provided in of ter The Constrained Region Points group indi references to pages of the book by Cornell (1990)o rguielougereaonrtemmobgthaLol cates whether or not each software package can gen- potnts (e.g.,extreme ve boundary of or inside a constrained mixture region an Such points s are often used s candidate points to der this group indicate the te the software can handle in generating points for the recoTnpoeh are can ac m number of comp computer and its available memory. The Mixture Designs group indicates the type (1990.pp.21-24)).simpler centroid (Cornell (1990 which of the have sig Vol.29,No.1.January 1997 Joumal of Quality Technology Reproduced with permission of the copyright owner.Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
GREGORY F.PIEPEL TABLE 1.Mixture Experiment Design Features Software Package Features ACED CARD COPS DXI CHIP SET JMP TAB SOFT M/D2 MODDE RS/D SAS S/G STGS eampoiaeabe (a)101524(a)(a850(a)420a)a1447 Simplex De 之8。 xx B 8 38065000(a (a) 12000k8 K2 >100 3 idat (a)6000ss000国aa (a) 12000k)G Blocking of Designs xx o ponent pr ple ned by sand do not r ndidate point Design-Expert;Mixture-Design;RS/Discover:STATGRAPHICS;Strategy Quality Technology Vol.9,No.1.January 1997 Reproduced with permission of the copyright owner.Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
SURVEY OF SOFTWARE WITH MIXTURE EXPERIMENT CAPABILITIES 79 Mixture Ex eriment Analysis Features only a few more design points than the number of SET focus completel on experimental design.while mixturecom eing screene Hence This centroids of a constrained mixture the blank spaces in Table 2 for these packages. are applica nd a tions the responsefs)of interest (see Cornell (1990.section ity to transform the response variable(s)before mod- eling or other data analyses.Some software packages while othe s well as the mixture components (sce Cornell (1990 tions (usually via a built-in programming language). omponent designs The Compon mixture of two r more components (see Cornell mixture components to L-or U-pseudocomponents (1990,Pp.205-220) In some package this ca in programming language The Optimal Mixture Experiment Designs group indicates the capabilities of each software The Regression Model Forms group indicates or g cring sta that sa the types of re on mode ify.for experiments invoving mixture components nonmixture variable 465) optimal desie and software are model-driven,and require the user listed.See Cornell ()for to specify a model form Hence,there pp,24-27) 290296.and m(i.e.lists/menus of models users may pic log contrast models (pp.325-332 The next two om)while others allow for us in the group indicate whet in which the total amount of the mixture quire specifying a model form.Rathe .space-filling erperiments)or the levels of process designs try to fill he experimenta (1990. 407-41d.Not that models for a mixture experiment with one pro son,Moore,and Ylvisaker (1990)).The final item ir cess variabl an be used mixture-amoun e group indicates th side ating an optimal experimental design. zero value (Piepel (1988)). The last two items in he group indicate e whether provides fo ities for blocking experimental designs (see Cornell (1990.pp.436-451) Vol.29.No.1,January 1997 Journal of Quality Technology Reproduced with permission of the copyright owner.Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
GREGORY F.PIEPEL TABLE 2.Mixture Experiment Analysis Feature Software Packag Features M/D2 MODDE RS/D SAS/TG Gpnm28asmtTranfarmoioio xXxx XX XX al quartic or quarti H2,H3 H2 mll B rergreio Xx Output xxxx x (e) 0 XX 0。 e (effects)plots (¥ eoptimi7atio n erimental design module.but not for no-intercent mixture models fitted einclude teting component effect and stepwise pooling of (g)Produces input data file for plot:must be plotted using someother oftware product Design-Expert:Mixture-Deign:RS/Discover:STATGRAPHICS;Strateg Joumal of Quality Technology Vol.29.No.1.January1997 Reproduced with permission of the copyright owner.Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
SURVEY OF SOFTWARE WITH MIXTURE EXPERIMENT CAPABILITIES 81 nell (1990,pp.250-256 and 270-274)).The next item indicates whether the software can produce respons Types of s(s Cornell (1990.pp 20920) can perform.The types of regression included in plots (see Piepel (192)).Component effects andre (par the c used and need no explanation. Ridge regression yields biased estimates of the mo mponents hxed.Ihe I paran last three itemsi with smal nodel parame estimates in cases where multicollinearity causes or for different responses on a single plot),andD dinary least squares to In particular,an points which might oth roduced for a simplex region or for a constrained subregion of the simplex. Montgomery and The Optimization group indicates the mathe matical optim capabi ties of ea softwar components and process variables,if any)and the atithat is.determining the levels of mix- ture components (and other r experimental variables any,such as or the The Regression Output and Diagnostics target,or maximm yalue of a single response vari group indicates the type of regre andd able. The second item addresses multiple respons ermining the the correct analysis of variance(ANOVA)table for a if any)that yield an optimum (minimum,target.or 76 naximum)value or some sort of compromise ame putes rsiduals (ordinary. the op s oI tw more standardized,single-point-deleted.etc.),collie ity digno (a)optimizing one response given constraints on the value of other responses,or (b)optimizing a d ger and Suich (1980) Box and Draper (1987,pp.373-375)).The desirabil be ty function approac converts a multipl dicate whether there are one or more exact or nearly optimization el terms for data hother the Miscellaneous Information highly influence the estimates of one or more of the able model coefficients. 3 summarize miscellaneous The Cor nd plots ovides four items of information package.The first item gives the for computing component effe s and producing var eac lous types of nd the co user interface responding standard errors (see Piepel (1982)or Cor Vol.29.No.1,January 1997 Joural of Quality Technology Reproduced with permission of the copyright owner.Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
GREGORY F.PIEPEL TABLE3.Miscellaneous Information Software Packag Features TM/D2 MODDE RS/D SAS S/G4 STG Software Information g2998883&品站89出9 S EEEE S EE SE E EEEE XXXXXX uired (MB A d,A=advised. Mouse (O-optional,OR=required)N/A R O R (a)N/A R O N/A N/A R OOO N/A 61锅e560足99" o STATGRAPHICS Pl)and Version 20 for STATGRAPHICS P he full systen n,qual and multivariate analysis) Design-Exper Mixture-Design;RS/Discover:4 STATGRAPHICS Strategy indica s the cost of a single-cop (or both)are provided. ts of Summary package is available for PCs es,or othe hard disk sr oace reauired for the software files.The vided for 15 software packages next item indicates whether a math coprocessor is op This information The na should prove helpful to those who are interested in gmore ab are It a also be useful to those who already have or use one ties ofthe Journal of Quality Technology Vol.29.No.1.January 1997 Reproduced with permission of the copyright owner.Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
SURVEY OF SOFTWARE WITH MIXTURE EXPERIMENT CAPABILITIES 83 package(s).Finally,the information will inform users COPS(Computer Optimized Process Solutions) with olo r versions of the covered software packages Bapidlnoatipaacep ties for those packages. nd,0H44145 All of the soft packages cov ered in this paper ments whic中 EMA1G-0 an important consideration for those planning to pur- Technical Contact:Harold S.Haller Ho wever,it was beyond on-mixture criment capabilities of the software Suite 191 2021E.He epin Avenue Acknowledgments 0ewN5513-2723 acknowledge the help and patience of the vendo technical contacts who provided information and re 915 EMAIL:74034.1347@compuserve.com Technical Contact:Pat Whitcomb the adito nd the tu ECHIP ECHIP.Ine paper 724 Yorklyn Rd Vendor Information FAX.302.230.622 EMAIL:support@echip.com Technical Contact:Bob Wheeler ware.Several vendors requested that these people GOSSET cept N.J.A.Sloane abilities. -376 ral ar tions ar Ro Labs in A d I support staff(when such staff exist for larger Murray Hill,NJ 07974 005 ACED (Algorithms for the Construction of Experi- 237 arch.att com JMP and Actuarial Science Waterloo,ON,Canada N2L 3G 567ext.5545 CARD (Computer Aided Research and Develop com/jmp Technical Contact:John Sall 317 Third St.Suite 5 Eureka,CA 9550 3081 Enterprise Drive Sta trixaol.com Technical Contact:George Cooney EMAIL:info@minitab.com Vol.29,No.1,January 1997 Journal of Quality Technology Reproduced with permission of the copyright owner. Further ed without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
84 GREGORY F.PIEPEL WWW:http://www.minitab.com VOICE:609-924-9374 Technical Contact:Terry Ziemer FAX.600.0240g54 MIXSOFT EMAIL:sgcasgcorp.com MIXSOFT-Mixture Experiment Softwar WWW:http://www.sgcorp.com 649 Cherrywood Loop Technical Contact:Neil Polhemus Richlan STATGRAPHICS In FAX 2115 F Ioffor WWW:http://members.aol.com/mixsoft Rockville.MD 20852 Technical Contact:Greg Piepe V0ICE:800-592-0050Ext.1125 FAX:301-9845094 MIXTURE-DESIGN EMAIL:answeramanu.com Statistical Programs 9941 Rowlett,Suite 6 STRATEGY 155 FAX.713.047.0604 Anacortes.WA 98221 EMAIL:info@statisticaldesigns.com V0ICE:1-800-732-7381 FAX:360-299-0202 MODDE Umetrics,Ine References VO1CE:617.756.0433 Box .G.E.P.and DRAPER.N.R.(1087.Empirical Model FAX:617-721-2652 EMAIL: CoRNELL J.A.(1990).Erp ts With Mirtu RS/DISCOVER "Si Opti Domain Solutions Corporation 21L.210 0TCE617-873-5000 Joo.M.E. L.M.:and YINISAKER.D.(1990) FAX:617-873-4751 KENNARD.R.W.and STONE L.A.(1969)."Computer Aided I.pp.I SAS Partial Least Squar pp VOICE:919-677-s000 G.F.(1982)."Measuring FAX:91967-8123 EPE C F (10 STATGRAPHICS When the Total .F.and.(1957)."Designs for Princeton. Journal of Quality Technology Vol.29.No.1.January 1997 Reproduced with permission of the copyright owner.Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
SURVEY OF SOFTWARE WITH MIXTURE EXPERIMENT CAPABILITIES .F.and.(1)"Mixture Expe Conditioning.and Ridge amples,Disc and Re Tecknology 16.pp.81-96. ST.JoHN.R.C.(184).Experiments with Mixtures.II Vol.29,No.1,January 1997 Joural of Quality Technology Reproduced with permission of the copyright owner.Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission