Dynamic Complexity 50th Annual Meeting of the International Society for the Systems Sciences George P.
Download ReportTranscript Dynamic Complexity 50th Annual Meeting of the International Society for the Systems Sciences George P.
Dynamic Complexity 50th Annual Meeting of the International Society for the Systems Sciences George P. Richardson Rockefeller College of Public Affairs and Policy University at Albany, SUNY Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 1 Learning in and about Complex Systems Sterman (1994) Real World Unknown structure Dynamic complexity Time Delays Impossible experiments Virtual World Implementation Game playing Decisions Inconsistency Short term Known structure Variable Complexity Controlled Experiments Selected Missing Information Delayed Feedback Biased Ambiguous Strategy, Structure, Inability to infer dynamics from Decision Rules mental models Rockefeller College of Public Affairs and Policy University at Albany Misperceptions Mental Unscientific Biases Models Defensiveness G. P. Richardson July 2006 2 Dynamic Complexity arises because systems are… • • • • Changing over time Tightly coupled Governed by feedback Nonlinear: changing dominant structure • History-dependent Rockefeller College of Public Affairs and Policy University at Albany • • • • • Self-organizing Adaptive Counterintuitive Policy resistant Characterized by tradeoffs G. P. Richardson July 2006 3 System Dynamics Contributions • Thinking dynamically • Move from events and decisions to patterns of continuous behavior over time and policy structure • Thinking in circular causal / feedback patterns • Self-reinforcing and self-balancing processes • Compensating feedback structures and policy resistance • Communicating complex nonlinear system structure Rockefeller College of Public Affairs and Policy University at Albany • Thinking in stocks and flows • Accumulations are the resources and the pressures on policy • Policies influence flows • Modeling and simulation • Accumulating (and remembering) complexity • Rigorous (daunting) model evaluation processes • Controlled experiments • Reflection G. P. Richardson July 2006 4 The system dynamics modeling process Perceptions of System Structure Comparison and Reconcilation System Conceptualization Representation of Model Structure 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 University at Albany G. P. Richardson July 2006 5 Processes focusing on system structure Mental Models, Experience, Literature Empirical Evidence Perceptions of System Structure Comparison and Reconcilation System Conceptualization Representation of Model Structure Model Formulation Diagramming and Description Tools Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 6 Processes focusing on system behavior Empirical Evidence System Conceptualization Model Formulation Rockefeller College of Public Affairs and Policy University at Albany Literature, Experience Empirical and Inferred Time Series Comparison and Reconciliation. Deduction Of Model Behavior Computing Aids G. P. Richardson July 2006 7 Two kinds of validating processes Mental Models, Experience, Literature Literature, Experience Empirical Evidence Perceptions of System Structure System Conceptualization Comparison and Structure Reconcilation Validating Processes Representation of Model Structure Diagramming and Description Tools Rockefeller College of Public Affairs and Policy University at Albany Model Formulation Empirical and Inferred Time Series Behavior Validating Processes Comparison and Reconciliation. Deduction Of Model Behavior Computing Aids G. P. Richardson July 2006 8 Six Traditions Contributing to the Evolution of Feedback Thought •Biology: math models •Econometrics •Engineering •Social Sciences •Biology: homeostasis •Logic QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture. Two Threads of Feedback Thought •System dynamics arises in the servomechanisms thread (the first four in this list) Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 9 Forrester’s Hierarchy of System Structure • Closed boundary around the system • Feedback loops as the basic structural elements within the boundary • Level [stock] variables representing accumulations within the feedback loops • Rate [flow] variables representing activity within the feedback loops • • • • Goal Observed condition Detection of discrepancy Action based on discrepancy Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 10 The Endogenous Point of View • The closed causal boundary takes top billing • Dynamics arise from interactions within that boundary • Systems thinking is the mental effort to uncover endogenous sources of system behavior. Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 11 Dynamics Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 12 New York City Population, 1900-2000 Bronx Brooklyn Manhattan Queens Staten Island 9000000 8000000 7000000 6000000 5000000 4000000 3000000 2000000 1000000 0 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 13 Global Atmospheric Methane (1860-1994) Global Atmospheric Methane 400 350 Teragrams 300 250 200 150 100 50 1980 1950 1920 1890 1860 0 Year Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 14 Global Average Temperature (Reconstruction 1400-1980; Data 1902-1998) Quic kTime™ and a TIFF ( LZW) dec ompres s or are needed to s ee this pic ture. Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 15 Stocks and Flows Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 16 Stocks and flows help to explain self-reported drug use data Have ever used Potential users Frequent users Occasional users Past users Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 17 ‘Challenging the clouds’ in a study of leasing in the automobile industry ? New vehicle inventory Production Purchase or lease “We’re not in the used car business!” Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 18 Stocks and flows in new car purchase and leasing Relative attractiveness of leased cars Used inventory Trade cycle Purchase used Vehicles being driven New vehicle inventory Production Sell or trade in Purchase or lease Rockefeller College of Public Affairs and Policy University at Albany Scrapping G. P. Richardson July 2006 19 Intuitive view of effect of leasing on auto sales: Leased car pipeline Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 20 Stocks and Flows in Global Warming Thought experiment: Capital stock capital investment CO2 annual production Atmospheric CO2 Breakdown of atmopheric CO2 Economic activity incoming solar heat energy Rockefeller College of Public Affairs and Policy University at Albany Global heat energy outgoing global heat energy G. P. Richardson July 2006 21 Life of aerosols in atm Life of GH gases in atm Aer osols i n atm Aer osol br eakdown Aer osol pr oduction But although the stock-and-flow insight holds, global climate is of course much more complex than that. GH g ases in atm GH g as pr oducti on GH g as br eakdown Aer osol al bedo efct Aer osol concentr ati on Atm volume Atm volume ~ GH g as concentr ation ~ Ice and cloud cover Total albedo ~ Ear th Ocean and Atm heat Cloud cover GH g as reten efct Ear th ar ea Solar heat reachi ng ear th Ear th heat radi ati on Ear th ar ea ~ Ice cover Cloud reten efct Ice thi ckness ~ Solar heat incoming Water vap reten efct Sur face temp Ice ar ea Cloud thickness Cloud ar ea CO2 r eten efct Atm temp ~ Ice density Atm volume Net thawing factor Evap factor Ice volume Ear th ice Ear th water Net thawing And still much more complex than this simple global climate model, as well! Cloud volume Water vap conc Water vapor in atm Cloud density Water in clouds Evaporation Condensation Water density Precipitati on Ear th water vol ume Atm volume CO2 atm conc CO2 water conc Ear th photosynthesi s Ocean photosynthesis CO2 i n atm CO2 i n oceans Car bon in ocean biomass Ocean CO2 br eakdown CO2 ocean r el ease Atm CO2 breakdown Ear th biomass decay Ocean biomass decay Rockefeller College of Public Affairs and Policy University at Albany Car bon in earth biomass Ocean CO2 pr oduction CO2 ocean uptake G. P. Richardson July 2006 Atm CO2 production 22 Feedback Thinking “For one good deed leads to another good deed, and one transgression leads to another transgression.” (Pirke Avot) Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 23 The Classic Cybernetic Balancing Loop Goal Perceived gap Perceived state Planned action to reduce gap Implicit, unstated goals Implemente d action Actual state of the system Changes in the State of the system Rockefeller College of Public Affairs and Policy University at Albany Intended actions G. P. Richardson July 2006 24 The Cybernetic Loop with Complications Goal Perceived gap Perceived state Autonomous changes in the state of the system Actual state of the system Changes in the State of the system Rockefeller College of Public Affairs and Policy University at Albany Planned action to reduce gap Implicit, unstated goals Implemente d action Intended actions G. P. Richardson July 2006 25 The Cybernetic Loop with Complications Goal Perceived gap Perceived state Autonomous changes in the state of the system Actual state of the system Changes in the State of the system Planned action to reduce gap Implicit, unstated goals Implemente d action Intended actions Unintended actions Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 26 The Cybernetic Loop with Complications Goal Perceived gap Perceived state Autonomous changes in the state of the system Actual state of the system Changes in the State of the system Planned action to reduce gap Implicit, unstated goals Implemente d action Intended actions Unintended actions Ramifying effects Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 27 A Classic Reinforcing Loop (Myrdal 1944, Merton 1948) Prejudice against the minority group Majority’s perception of the inferiority of the minority (R) Economic and educational discrimination against the minority Achievements of the minority group Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 28 Structure and Dynamics of Terrorist Cells (R) Peripheral support for terrorists Terrorist funding (R) Interfering with terrorist funding Terrorist actions Efforts to suppress terrorists (B) (R) Terrorist zeal (B) Recruiting (R) terrorists (R) Terrorist group Terrorist martyrs to the cause Rockefeller College of Public Affairs and Policy University at Albany (B) (B) Losing terrorists (R) G. P. Richardson July 2006 29 Insights about building teamwork in a public school Teamwork and Communication are self-reinforcing Quality of communication + Trust + (+) Risk taking + Quality of communication within teams + Teamwork + Resistance to teamwork Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 30 Isolation of teams and punishing risk-taking inhibit the growth of trust + Quality of communication + + Positive responses to experiments ? Trust + (+/-) Individual experiments (+) Risk taking Quality of communication between teams + (-) Quality of communication within teams + Teamwork + - + Resistance to teamwork Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 31 But longterm experience with teamwork can build communication Quality of + communication + between teams + Quality of (-) communication Quality of + (+) communication + + within teams Trust (+) + Cumulative Positive + Risk experience with responses to Teamwork taking teamwork (+/-) experiments + + ? + Individual Resistance to + + experiments Team teamwork effectiveness - Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 32 Risk taking can enhance effectiveness, which can build trust Quality of + communication + between teams + Quality of (-) communication Quality of + (+) communication + + within teams Trust (+) + Cumulative Positive + Risk experience with responses to Teamwork taking teamwork (+/-) experiments + + ? + Individual Resistance to + + experiments Team teamwork effectiveness + + Personal learning Average personal effectiveness + Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 33 A team-player culture is self-reinforcing: an opportunity or a trap Quality of + communication + between teams + Quality of (-) communication Attractiveness Quality of + (+) of the org to communication + + within teams + team players Trust (+) + Cumulative Positive + Risk experience with responses to Teamwork + taking teamwork (+/-) experiments + Fraction of staff who are + ? team players + Individual Resistance to + + experiments Team teamwork + effectiveness + + Personal learning Average personal effectiveness + Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 34 Likely leverage points + Extent of Learning Organization characteristics present Quality of communication + + Trust (+) + + Positive responses to experiments ? + (+/-) Teamwork + - + Resistance to teamwork + Team effectiveness - + Personal learning + Average personal effectiveness Rockefeller College of Public Affairs and Policy University at Albany Dialogue training Attractiveness of the org to + team players Cumulative experience with teamwork + Individual experiments Adm teaching role Risk taking + Quality of communication + + between teams + (-) Quality of (+) communication + within teams + + + + Fraction of staff who are team players + + Understanding stages of community building G. P. Richardson July 2006 35 The Problem: 1996 U.S. welfare reform • Since 1930, a guarantee of lifetime Federal support • 1996 legislation ended that: • Temporary Assistance for Needy Families - TANF • At most five years of Federal support in one’s lifetime • The clock started for everyone on TANF in 1997 • People began timing out in 2002 • Financial burden will begin shifting to the states and counties • A series of facilitated group modeling efforts in three New York State counties tried to help counties cope. Where are the leverage points? Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 36 Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 37 Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 38 Three Policy Mixes • Base run (for comparison) • Flat unemployment rate • Historical client behaviors • Investments in the “Middle” • Additional services to TANF families • Increased TANF assessment & monitoring • Safety net assessment & job services • Investments on the “Edges” • Prevention • Child support enforcement • Self-sufficiency promotion Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 39 Investing in the “Middle” Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 40 Investing on the “Edges” Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 41 Base, “Edges,” and “Middle” Compared: Populations on the Welfare Rolls Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson “Edges” looks better. July 2006 42 Total Job-Finding Flows from TANF Rockefeller College of Public Affairs and Policy University at Albany “Middle” looks better. G. P. Richardson July 2006 43 Program Expenditures G. P. Richardson “Edges” looks worse, then better. July 2006 Rockefeller College of Public Affairs and Policy University at Albany 44 Populations in the Welfare System Rockefeller College of Public Affairs and Policy G. P. Richardson “Middle” looks worse than “Base”! “Edges” looks much better. University at Albany July 2006 45 Total Recidivism Flows (back to TANF) The hint for understanding the puzzling dynamics: recidivism. Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 46 A Stock-and-Flow Archetype at Work Here + (R) Probability of recidivism Recidivism Families on TANF Enter TANF Job finding rate Post TANF employed To mainstream employment (R) Load on TANF support capacity + (R) Time to find first job Rockefeller College of Public Affairs and Policy University at Albany Load on employment support capacity G. P. Richardson July 2006 47 Behavior of the Archetype in response to increased TANF support capacity 6,000 Total families at risk 4,500 Families on TANF 3,000 1,500 Post-TANF employed 0 0 6 12 18 Fam ilies on T ANF : archet ype base P ost T ANF employed : archet ype base T otal families at risk : archet ype base 24 Rockefeller College of Public Affairs and Policy University at Albany 30 36 T im e (M ont h) 42 48 54 60 families families families G. P. Richardson July 2006 48 The Behavior of the Archetype • Families on TANF initially declines, as more support hastens job finding. • Post-TANF families employed initially increases, just as policy makers would predict. • Eventually (it takes a year and a half to begin to see it), … • Families on TANF rises higher to a new high, • Post-TANF Employed declines to a new low, • And Total Families at Risk rises! • …All because of increased TANF support capacity! Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 49 Why? • Increasing TANF support • Speeds job finding, • Swamping downstream Post-TANF jobs and support Time in post TANF employ Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity + - Job finding rate Time to find first job TANF support capacity Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 50 Misattribution? • Desirable rise in Post-TANF employed continues for almost a year and half after the intervention • Families on TANF falls below initial for over a year after increasing TANF support capacity • Very hard (impossible?) to see that the rise in Total Families at Risk is attributable solely to the improvement in TANF support capacity • Dynamics almost certainly to be blamed on a weakening economy, a rise in client pathologies, or other exogenous factors Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 51 A Loop View of the Archetype in Detail Time in post TANF employ Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity + - Job finding rate Time to find first job TANF support capacity + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity Suppose TANF support capacity increases… Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 52 4,000 White bar (left) is the time slice of interest 3,000 B: Employed load controls recidivism 2,000 Red arrows (below) are the dominant influences 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 53 4,000 3,000 B: Employed load controls recidivism 2,000 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 54 4,000 3,000 B: Employed load controls recidivism 2,000 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 55 4,000 3,000 B: Employed load controls recidivism 2,000 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 56 4,000 White bar (left) is the time slice of interest 3,000 B: Employed load controls recidivism 2,000 Red arrows (below) are the dominant influences 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 57 4,000 3,000 B: Employed load controls recidivism 2,000 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 58 4,000 3,000 B: Employed load controls recidivism 2,000 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 59 4,000 3,000 B: Employed load controls recidivism 2,000 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 60 4,000 3,000 B: Employed load controls recidivism 2,000 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 61 4,000 Dynamic complexity even in a structure this aggregate and tiny! 3,000 B: Employed load controls recidivism 2,000 1,000 0 0 30 60 Time (Month) 90 Time in post TANF employ 120 Outflow from post TANF employ Recidivism Enter TANF + Families on TANF Load on TANF support capacity TANF support capacity + - Job finding rate Time to find first job Rockefeller College of Public Affairs and Policy University at Albany + Post TANF employed - To mainstream employment Load on employment support capacity Probability of recidivism Post TANF employment support capacity G. P. Richardson July 2006 62 System Dynamics and Dynamic Complexity • Thinking dynamically moves us beyond separate events and decisions, toward understanding. • Feedback thinking extends traditional causal thinking. • It improves (makes more realistic) how we think about the world and how we think about changing it. • The endogenous point of view is empowering. Rockefeller College of Public Affairs and Policy University at Albany G. P. Richardson July 2006 63