Model Validation as an Integrated Social Process George P. Richardson Rockefeller College of Public Affairs and Policy University at Albany - State University of New York GPR@Albany.edu Rockefeller College of Public Affairs and Policy 1 University at Albany State University of New York What do we mean by ‘validation’? • No model has ever been or ever will be thoroughly validated. …‘Useful,’ ‘illuminating,’ or ‘inspiring confidence’ are more apt descriptors applying to models than ‘valid’ (Greenberger et al. 1976). • Validation is a process of establishing confidence in the soundness and usefulness of a model. (Forrester 1973, Forrester and Senge 1980). Rockefeller College of Public Affairs and Policy 2 University at Albany State University of New York The classic questions • Not ‘Is the model valid,’ but • Is the model suitable for its purposes and the problem it addresses? • Is the model consistent with the slice of reality it tries to capture? (Richardson & Pugh 1981) Rockefeller College of Public Affairs and Policy 3 University at Albany State University of New York The system dynamics modeling process Perceptions of System Structure Comparison and Reconcilation Representation of Model Structure System Conceptualization Model Formulation Empirical and Inferred Time Series Comparison and Reconciliation. Deduction Of Model Behavior Adapted from Saeed 1992 Rockefeller College of Public Affairs and Policy 4 University at Albany State University of New York Processes focusing on system structure Mental Models, Experience, Literature Perceptions of System Structure Comparison and Reconcilation Representation of Model Structure Empirical Evidence System Conceptualization Model Formulation Diagramming and Description Tools Rockefeller College of Public Affairs and Policy 5 University at Albany State University of New York Processes focusing on system behavior Empirical Evidence System Conceptualization Model Formulation Rockefeller College of Public Affairs and Policy 6 Literature, Experience Empirical and Inferred Time Series Comparison and Reconciliation. Deduction Of Model Behavior Computing Aids University at Albany State University of New York Two kinds of validating processes Mental Models, Experience, Literature Perceptions of System Structure Empirical Evidence Literature, Experience Empirical and Inferred Time Series System Conceptualization Comparison and Comparison and Structure Behavior Reconciliation. Reconcilation Validating Validating Processes Processes Model Formulation Representation of Deduction Of Model Structure Model Behavior Computing Diagramming and Aids Description Tools Rockefeller College of Public Affairs and Policy 7 University at Albany State University of New York The classic tests Focusing on STRUCTURE Focusing on BEHAVIOR Testing SUITABILITY for PURPOSES • Dimensional consistency • Extreme conditions • Boundary adequacy • Parameter insensitivity • Structure insensitivity Testing CONSISTENCY with REALITY • Face validity • Parameter values • Replication of behavior • Surprise behavior • Statistical tests Contributing to UTILITY & EFFECTIVENESS • Appropriateness for audience • Counterintuitive behavior • Generation of insights Forrester 1973, Forrester & Senge 1980, Richardson and Pugh 1981 Rockefeller College of Public Affairs and Policy 8 University at Albany State University of New York Validation is present at every step • Conceptualizing: • Do we have the right people? • The right dynamic problem definition? • The right level of aggregation? • • • • • Mapping: Developing promising dynamic hypotheses Formulating: Clarity, logic, and extremes Simulating: Right behavior for right reasons Deciding: Implementable conclusions Implementing: Requires conviction! Rockefeller College of Public Affairs and Policy 9 University at Albany State University of New York Do we have the right people? Rockefeller College of Public Affairs and Policy 10 University at Albany State University of New York Opposition High Weak opponents Strong opponents Weak supporters Strong supporters Low Low Support Problem Frame Problem frame stakeholder map High Weak Strong Stakeholder Power Bryson, Strategic Planning for Public and Nonprofit Organizations Rockefeller College of Public Affairs and Policy 11 University at Albany State University of New York Power versus Interest grid High Players Crowd Context setters Interest Subjects Low Weak Strong Power Eden & Ackerman 1998 Rockefeller College of Public Affairs and Policy 12 University at Albany State University of New York Pursuing validity in mapping • Think causally, not correlationally • Think stocks and flows, even if you don’t draw them • Use units to make the causal logic plausible, even if you don’t write them down • Be able to tell a story for every link and loop • Move progressively from less precise to more precise -- from informal map to formal map Rockefeller College of Public Affairs and Policy 13 University at Albany State University of New York The standard cautions Understandings of the system Understandings of the model Carbon in carnivores System conceptualization Model formulation & testing Carbon in herbivores Carbon in soil Carbon in algae, plants & trees Prejudice Achievements of the minority Carbon in atmosphere Discrimination Opportunities for the minority Rockefeller College of Public Affairs and Policy 14 University at Albany State University of New York These arrows mean ‘and then’ Understandings of the system Understandings of the model • We start with some understandings of the Carbon in problem and its systemic context, and carnivores Carbon in then we conceptualize (map) the system. System conceptualization Model formulation & testing • Carbon in Then we build herbivores the beginnings of a model, Carbon in soil which we then test to understand it. Carbon in algae, reformulate, plants & treesor • Then we reconceptualize, or revise our understandings, or do some of all three, and then continue… Prejudice Achievements of the minority atmosphere Discrimination Opportunities for the minority Rockefeller College of Public Affairs and Policy 15 University at Albany State University of New York Arrows here are flows of material Understandings of the system The words here Understandings represent stocks. System of the model conceptualization This is not a Model formulation & testing causal diagram. Carbon in carnivores Carbon in herbivores Carbon in soil Carbon in algae, plants & trees Prejudice Achievements of the minority Carbon in atmosphere Discrimination Opportunities for the minority Rockefeller College of Public Affairs and Policy 16 University at Albany State University of New York Only this one is a causal loop Understandings of the system No explicit stocks or flows, no clear units, but it tells a Understandings System– It’s a Carbon in of the model compelling story conceptualization herbivores good start. Model formulation & testing Carbon in atmosphere Carbon in soil Carbon in algae, plants & trees Prejudice Achievements of the minority Carbon in carnivores Discrimination Opportunities for the minority Rockefeller College of Public Affairs and Policy 17 University at Albany State University of New York Project modeling core structure Work to be done Work in process beginning work completing work doing work incorrectly starting rework Known rework Rockefeller College of Public Affairs and Policy Work really done rework discovery 18 Undiscovered rework University at Albany State University of New York Identical structure without explicit stocks and flows completing work beginning work Work in process Work to be done Work really done starting rework doing work incorrectly Known rework Undiscovered rework rework discovery Rockefeller College of Public Affairs and Policy 19 University at Albany State University of New York Pursuing validity writing equations • • • • Recognizable parameters Robust equation forms Phase relations Richardson’s Rule: Every complicated, ugly, excessively mathematical equation and every equation flaw saps confidence in the model. Rockefeller College of Public Affairs and Policy 20 University at Albany State University of New York Modeling conflict within & between nations Adaptation Potential for international conflict International conflict + Consequences of conflict + - + Population growth rate Technology growth rate of Public Affairs and Policy + + Population Technology - + Rockefeller College + + Domestic conflict Potential for conflict Lateral pressure Internal stress + Domestic adaptation 21 University at Albany State University of New York Complexity & flaws destroy confidence • P of int'l conflict = DELAY FIXED ((Lateral pressure/10*Military force effect/Trade and bargaining leverage + International conflict)/Lateral conflict break point, 1 , 0) • Flaws Complexity, discreteness, units confusion and disagreement, disembodied parameter, confusion of the effect of a concept [leverage] with the concept itself, and the wonder what keeps this probability between 0 and 1? Rockefeller College of Public Affairs and Policy 22 University at Albany State University of New York Robust equation forms Progress Rockefeller College of Public Affairs and Policy 23 Cumulative progress University at Albany State University of New York Causal mish-mash Hours per person per day Workers Workweek (days) Normal effectiveness (tasks/hour) Progress Cumulative progress Effect of schedule pressure Effect of motivation Rockefeller College of Public Affairs and Policy Effect of ... 24 University at Albany State University of New York Robust equation formulations Effort (hours/month) Progress Cumulative progress Effectiveness (tasks/hour) Rockefeller College of Public Affairs and Policy 25 University at Albany State University of New York Robust equation formulations Hours per person per day Workweek (days) Workers Effort (hours/month) Progress Cumulative progress Effectiveness (tasks/hour) Rockefeller College of Public Affairs and Policy 26 University at Albany State University of New York Robust equation formulations Effort (hours/month) Progress Normal effectiveness (tasks/hour) Cumulative progress Effectiveness (tasks/hour) Effect of motivation Rockefeller College of Public Affairs and Policy Effect of schedule pressure Effect of ... 27 University at Albany State University of New York Robust equation formulations Hours per person per day Workweek (days) Workers Effort (hours/month) Progress Normal effectiveness (tasks/hour) Cumulative progress Effectiveness (tasks/hour) Effect of motivation Rockefeller College of Public Affairs and Policy Effect of schedule pressure Effect of ... 28 University at Albany State University of New York Pursuing validity in equations: Phasing (B) Problems threaten scope generating problems Problems generated integrating info Problems generated per info unit integrated (R ) Problems compound Unintegrated information Unitegrated info within scope Scope of integration effort Ease of integrating info of Public Affairs and Policy (R ) Success enhances resources Effort to integrating info (R ) Perceived value enhances scope Willingness to cede control of info Rockefeller College (B) Low hanging fruit (B) Problems impede progress Pressure to allocate resources elsew here integrating info (B) Problems rob resources subtracting resources to integration 29 Integrated information Perceived value of integrated information Resources devoted to info integration adding to resources to integration University at Albany State University of New York Phase relations Integrated information Constant Perceived Value suggests continually rising Resources, but that doesn’t seem correct Perceived value of integrated information Resources devoted to info integration adding to resources to integration Rockefeller College of Public Affairs and Policy 30 University at Albany State University of New York Phase relations Perceived value of integrated information Here, the Perceived Value of Integrated Information sets a planned level of resources Resources allocated to integration project Resources planned to info integration adding to resources to integration Time to allocate resources Rockefeller College of Public Affairs and Policy 31 University at Albany State University of New York Pursuing validity fitting to data • Generally, a weak test of model validity • Whole-model procedures • Optimization • Partial-model procedures • Reporting results • Graphically • Numerically: Theil statistics Rockefeller College of Public Affairs and Policy 32 University at Albany State University of New York Example of weakness of fitting to data Discovery / production experience & technology • Logistic curve • dx/dt = ax - bx2 (R) • Gompertz curve Cumulative production Petroleum production • dx/dt = ax - bx ln(x) (B) Constraints from the resource remaining Rockefeller College of Public Affairs and Policy 33 University at Albany State University of New York Fitting global petroleum with Logistic Production 40,000 40,000 20,000 20,000 Cum Production 2M 2M 0 0 1880 1902 1924 1946 1968 1990 2012 2034 2056 2078 2100 Time (year) 1M 1M 0 0 1880 1902 1924 1946 1968 1990 2012 Time (year) Rockefeller College of Public Affairs and Policy 34 2034 2056 2078 2100 University at Albany State University of New York Fitting global petroleum with Gompertz Production 40,000 30,000 20,000 Cum Production 10,000 4M 2M 0 1880 1902 1924 1946 1968 1990 2012 2034 2056 2078 2100 Time (year) 2M 1M 0 0 1880 1902 1924 1946 Rockefeller College of Public Affairs and Policy 35 1968 1990 2012 Time (year) 2034 2056 2078 2100 University at Albany State University of New York Presenting model fit visually Rockefeller College of Public Affairs and Policy 36 University at Albany State University of New York Presenting model fit numerically • Theil statistics, for example • Based on a breakdown of the mean squared error: 2 2 2 ( ) (X Y ) s s 1/ n i i (X Y ) x y 2(1 r) sx sy • 1 = Bias + Variation + Covariation Rockefeller College of Public Affairs and Policy 37 University at Albany State University of New York Presenting model fit numerically Bia s Va riati on Covaria tion RMSPE RMSEPM U r As sets 0.001 0.001 0.998 0.111 0.119 0.097 0.986 Li abil itie s 0.136 0.426 0.438 0.208 0.099 0.079 0.996 Prem ium inc 0.376 0.047 0.576 10 66.0 00 0.693 0.562 0.664 Surp lus 0.019 0.640 0.341 0.437 -0.178 0.133 0.994 Case s op en 0.119 0.354 0.526 0.049 0.061 0.051 0.998 To tal premi ums 0.094 0.409 0.497 0.173 0.298 0.256 0.908 Rockefeller College of Public Affairs and Policy 38 University at Albany State University of New York Learning from surprise model behavior • Have clear a priori expectations • Follow up all unanticipated behavior to appropriate resolution • Confirm all behavioral hypotheses through appropriate model tests (Mass 1991/1981) Rockefeller College of Public Affairs and Policy 39 University at Albany State University of New York Tests to reveal and resolve surprise behavior • • • • • • Testing the symmetry of policy response (up and down) Testing large amplitude versus small amplitude response Testing policies entering at different points Testing different patterns of behavior Isolating uniqueness of equilibrium or steady state Understanding forces producing equilibrium positions (Mass 1991/1981) Rockefeller College of Public Affairs and Policy 40 University at Albany State University of New York Summary • Modelers, stakeholders, problem experts, and others in the modeling process pursue validity at every step along the way. • We have rigorous traditions guiding model creation, formulation, exploration, and implications. • We have a powerful, intimidating battery of tests of model structure and behavior. • Model-based conclusions that make it through all this deserve the confidence of everyone in the process. Rockefeller College of Public Affairs and Policy 41 University at Albany State University of New York Epilog • Reason is itself a matter of faith. It is an act of faith to assert that our thoughts have any relation to reality. (G.K. Chesterton) • I have no exquisite reason for’t, but I have reason good enough. (Sir Andrew, Twelfth Night) Rockefeller College of Public Affairs and Policy 42 University at Albany State University of New York