Data & Modeling Decisions

1. Data & Modeling Decisions MSC 636 Winter 2003

2. Problem Solving • Relies on problem definition • Difference between current and desired state of affairs • Problem statement • Current state • Desired state • Key objective • Problem scope to match resources & priorities

3. Evolution toward DSS • Intersection of data processing and management science • Database management systems • large amount of data • data independence • flexible, easy access by non-programmers • Modeling • Sopisticated, accurate depictions of reality • Go beyond simple, static models • Use computers to make modeling more accessible, interactive, and meaningful

4. Components Interface Network Data Models Knowledge

5. Internal data Shared sources TPS/operational data Financial Accounting Production HRM Data warehouses/data marts Private/personal data External data Particularly for EIS Commercial/ subscription databases Industry data Governmental/agency/public access data Primary data collection Sources of Data

6. Timeliness Sufficiency Detail/aggregation Understandability Bias Relevance Comparability Reliability Redundancy Cost efficiency Quantifiability Format Data/Information Quality Issues

7. Models/Modeling • Simplified representation of reality • Model building (modeling) • Study environment • Define problem structure • Choose or create models • Identify important variables & relationships • Express in mathematically, logically, or graphically

8. Problem Structure • Choices • Multiple alternatives • Objectives • Goal of decision making • Objective criteria of success • Uncertainties • Different than choice • Characteristic of all good problems

9. Components of Influence Diagrams and Decision Trees Represents relationships & sequence of context Decision Uncertainty Objective

10. B A B A Event A outcome is relevant to probability of Event B outcome Outcome of Event A is known when making Decision B B B A A Decision A is necessary to estimate probability of Event B Decision A is made prior to Decision B Relevance Arrows (arcs) in Influence Diagrams Uncertainties become events after they happen

11. WinContest Win large return on wager EnterContest Lose wager Lose Contest Lose/Gain nothing Do Not EnterContest Simple Decision Tree • More decision detail; predicted outcomes • Rules • Singular • Mutually exclusive & exhaustive • Complete, sequential event paths

12. Decision Table • Tabular way of representing decision logic (rather than sequencing) • Multiple criteria • May be used in conjunction with decision trees & influence diagrams

13. Uncertainty Price goes up Gain Buy Stock Loss Price goes down Lose/Gain nothing Decision Objective Do Not Buy Stock Basic Risky Decision

14. Basic Risky Policy Uncertainty Decision Objective Basic Risky Decision with Multiple Objectives Objective 1 Uncertainty Total Satisfaction Objective 2 Decision Objective n Variations of Basic Risky Decision Model

15. Objective 1 Objective 2 Total Satisfaction Decision Objective n Multiple Objective, No-Risk Decision

16. Uncertainty t3 Uncertainty t1 Uncertainty t2 Uncertainty tn Total Satisfaction Decision t1 Decision t2 Decision t3 Decision tn Objective t1 Objective t2 Objective t3 Objective tn Multiple-Period Sequential Decision

17. Model Characteristics • Time Dimension • Static • Dynamic • Methodology (type) • Representational • Abstract • Conceptual

18. Representational • Iconic/scale • Replicas • Model airplanes, bridges, production lines • Measured drawings • Maps, photographs • Analog • Symbolizes or acts like reality • Organization chart, histograms, flow chart, time lines • EKG readouts, lie detector results • Blueprints

19. Abstract Models • Deterministic • Static, algorithmic models • Optimization • Stochastic • Probabilistic statistics • Estimated parameters • Regression, time series, game theory, etc.

20. Abstract Models (cont) • Simulation • Descriptive rather than normative • Predicts outcomes of interaction between environment & series of decision alternatives • Effective communication • Risks, environmental impacts, interactions, training, problem identification • Costly & specialized

21. Simulation Example • Lining up for Santa • Using random numbers & probabilities (Monte Carlo) • Observe environment • Generate random number to mimic nature • Outcome: Long lines are possible, but not certain • Observations • No optimal solution • Repetition increases confidence • Better than trial & error with reality

22. Conceptual Models • Mathematical models • Too costly • Too simple • Not available • Drawing from experience & analogies • Structuring tools useful • Decomposing • Organizing relationships • Considering probabilities associated with uncertainties

23. Multiple Criteria Decision Table • Predict grade • Complete decision table • Are the predictions different? • Why? • Which grade do you want to bet on?

24. Probabilities • Methods for determining • Long-run frequency • Observation of large numbers • Subjective • Degree of belief (expert based?) • Subjective approaches • Direct • Odds/comparison forecasting

25. Calibration, Sensitivity, Value • Accuracy of probability estimates • Self-awareness of ability to make accurate predictions • Tendency toward over-confidence • Highly sensitivecareful of error • Insensitivedrop for parsimony • ValueHow much time & money should you spend?

26. Factors Affecting the Selection of a Forecasting Technique • Problem Structure • Complexity & size • Precision required • Availability • DSS designer preference

27. Models in DSS • Essential in most • Model-based vs. data-based • Simple or complex • Standard or custom-made • Single or multiple

28. Model Management • Spreadsheet as model mgt tool • User interaction with models • Easy & flexible input • Understandable output (vary by user) • Use system intelligence to prompt (passive/active) or automate actions • Balance flexibility/control/ease of use • Default with option to change • Default by user profile, dept., level