Transcription
ANP Sensitivity AnalysisOrrin Cooper
3rd Party Logistics Provider Selection0.110.1050.10.0950.09Third Party A0.085Third Party B0.08Third Party 2390.06Third Party AThird Party BThird Party C
Value of Sensitivity AnalysisuHave you ever “known” which alternative would be the most preferred, onlyto synthesize the model and get a different result?uHow robust is my decision?uWhich criteria are most important or influential?uWhat would happen if “this” or “this” criteria had been weighted differently?uuHow much can this priority change before the decision changes? (Limits: gas priceexample)Would it even make a difference in the final outcome? (Stakeholders)
LiteratureuMu, E., Cooper, O., & Peasley, M. (2020). Best Practices In Analytic NetworkProcess Studies. Expert Systems with ApplicationsuSava, M. G., Vargas, L. G., May, J. H., & Dolan, J. G. (2020). Multi-dimensionalstability analysis for Analytic Network Process models. Annals of OperationsResearch, 1-24.uChen, Y., Yu, J., & Khan, S. (2013). The spatial framework for weight sensitivityanalysis in AHP-based multi-criteria decision making. Environmental modelling &software, 48, 129-140.uAdams, B. (2011, May). ANP Row Sensitivity. In International Symposium on TheAnalytical Hierarchy Process.uSearch: ANP Sensitivity Analysis – many methodsuLinear Programming, Optimization, Machine Learning Sensitivity AnalysisuOther MCDA tools also have/use Sensitivity Analysis
Value of AHP/ANP?Competing Arguments:uThe method itself and the FINAL ANSWER (Hypothesis Testing/Post HocAnalysis)uuThe PROCESS and how it helps us understand, design, and frame the problem.And then thoughtfully evaluate and make trade-offs.uu“If you torture the data enough, it’s going to ultimately tell you what you want it to tellyou.” (Dr. Fauci, 2020)What we learn in the PROCESS is as important as the FINAL ANSWERPROCESS and FINAL ANSWER and SENSITIVITY ANALYSISuFINAL ANSWER of course, what we learn in the PROCESS, AND what we LEARN fromthe SENSITIVITY ANALYSIS (LP Baking Model: Sugar)
Understand the problem and theSupermatrixuBeauty of ANP - don’t need to know scales or too expensiveuuuMust pay attention to units of measurementMust know what is going on in your model or else:uTamales Sales (regression)- increase price and sell more uLabrador dogs life expectancy – bleach my dog’s hair uPivot Tables – hard to get errors but students won’t read the data key, no doublechecking – again the process as much as final decisionMust know your model and how it works to do good / meaningful sensitivityanalysis
OutlineuUnderstanding the Supermatrix via an ANP Fruit Baskets modeluExplore sensitivity analysisuExceluSuper Decisions
Understanding the SupermatrixANP – Fruit Baskets ModelHomogenous ClustersSmall Fruit Cherry Tomato0.07Small Green Tomato0.28Lime0.65Medium Fruit Lime0.08Grapefruit0.22Honey Dew0.70Large Fruit Honey Dew0.10Sugar Baby Watermelon0.30Oblong Watermelon0.60
Understanding the SupermatrixANP – Fruit Baskets ModelAlternative 1Alternative 2Alternative 3Variety BowlGo Green BowlWarm Bowl20020Small Green Tomato050Lime5105Grapefruit205Honey Dew141Sugar Baby melon001Oblong Watermelon100Cherry Tomato
Understanding the SupermatrixANP – Fruit Baskets ModeluUnweighted SupermatrixuWeighted SupermatrixuLimit MatrixuFinal PrioritiesuuAlternativesuCriteriaWhat Sensitivity to do?uControversial elements (whether or not “important”)uSize of Influenceu“Weakest” Comparisons
Understanding the SupermatrixANP – Fruit Baskets ModeluExceluSuper Decisions
Super Decisions Relocation of CorporateHeadquarters2 – ANP BOCR ModeluLocated under: Super Decisions Help/Sample Models/BOCR ModelsuExceluSuper Decisionsu12 SubnetworksuChanges in Subnetworks WILL LIKELY Impact the Ratings Modelu“If a node changes in 1 subnetwork should it change in another?”uPotentially Complicated but still some meaningful analysis
ReportinguGood Quality GraphsuConcerns not resolved in the literature: about how to reapportion weights,Many opinions but each need more research, is the analysis meaningful (orjust complicated).uPotentially different valid arguments depending on reasonsuImportant to be transparent and defend
Questions
u Explore sensitivity analysis uExcel uSuper Decisions. Understanding the Supermatrix ANP -Fruit Baskets Model Homogenous Clusters Small Fruit Cherry Tomato 0.07 Small Green Tomato 0.28 Lime 0.65 Medium Fruit Lime 0.08 Grapefruit 0.22 Honey Dew 0.70 Large Fruit Honey Dew 0.10 Sugar Baby Watermelon 0.30
