An Introduction to System Dynamics

Pre-Conference Workshop at the Society for Prevention Research: Systems Science Methodologies for Prevention Research An Introduction to System Dynamics Kristen Hassmiller Lich, Ph.D., MHSA Assistant Professor Department of Health Policy and Management UNC Chapel Hill, Gillings School of Global Public Health klich@unc.edu With thanks to: Andrew Jones, at the Sustainability Institute, who provided a bunch of these slides (most of the entertaining ones!!!); www.sustainabilityinstitute.org 1

Agenda Complex problems (what and why) System Dynamics (what and why) The general approach Key tools for building models Causal loop diagrams Stock and flow models Conclusions and resources 2 GOAL: To share my perspective on the value of System Dynamics and to teach you enough so that you can read more!

Complex problems Many of the problems we deal with today in public health and prevention are complex, not owned by a single stakeholder (each with differing perspectives and/or objectives), and can often seem completely overwhelming: How best can we alleviate the hypertension problem in Durham county, North Carolina? How should we invest as a nation to address diabetes? What are reasonable policy targets? Why can t we get tuberculosis under control in India despite substantial investment and improvement in control efforts? How can current resources be used more effectively to provide mental health services locally? How can we reengineer our hospital system to better handle surges in demand following a disaster? 3

Complex problems There is a difference between detail complexity and dynamic complexity Dynamic complexity arises because of: Dynamic behavior (things change over time) Time delays between cause and effect 4

How Do You Cover Problems That Show Up So Gradually??? 5 Say, Thag... Wall of ice closer today?

Complex problems There is a difference between detail complexity and dynamic complexity Dynamic complexity arises because of: Dynamics (things change over time) Time delays between cause and effect Nonlinear relationships Interactions Feedback loops (x increases y, which increases x ) Emergence of often counterintuitive system behavior 6

Tendency: Decompose and study (Industrial Revolution to Today) All complex systems, like the human body, the Earth s atmosphere, Mozart s Jupiter symphony, and Japan s foreign policy, are better understood by slicing big things into small things. Five Star Mind, by Tom Wujec, 1995 Thanks to Linda Booth Sweeney for pointing out this quote. 7

8 The result

The result Band-aid solutions Parts thrive and whole suffers Silo thinking Policy resistance 9

Tendency: Linear Thinking John Sterman, System dynamics Group Leader at MIT's Sloan School of Management, in Business Dynamics (page 5-14) 10 http://www.stewardshipmodeling.com/policy_resistance.htm

Tendency: Linear Thinking 11 Hirsch GB, Levine R, and Miller RL (2007). Using system dynamics modeling to understand the impact of social change initiatives. Am J Community Psychol ; 39:239-253.

Tendency: Linear Thinking 12 Hirsch GB, Levine R, and Miller RL (2007). Using system dynamics modeling to understand the impact of social change initiatives. Am J Community Psychol ; 39:239-253.

13 The result Sometimes the Cause of Problems Is Solutions

We Humans Have A Long History of Unintended Consequences Low tar and low nicotine cigarettes actually increase intake of carcinogens, CO, etc. Manufacturing -- Having employees who actively used the product made it more difficult to improve product quality. Paving dirt roads in mountain areas leads to decrease in safety; similarly, anti-lock brakes create safety decrease for some Fourth highest cause of death in U.S. is medical treatments Despite widespread use of labor-saving devices, Americans have less leisure today than 50 years ago US policy of fire suppression has increased the size and strength forest fires in many areas Road building programs designed to reduce congestion have increased traffic, delays, and pollution. 14

15 But We Have Begun to See That We Are the Source of Many of Our Own Problems

What Really Often Happens Sterman, cont d 16

What Really Often Happens Sterman, cont d 17

Complex problems How can you get anywhere with so much complexity? How do you build a shared understanding? How do you get anyone to agree on a course of action? How do we improve anything in a timely manner? 18

System Dynamics Methods can help! System: A functional whole, composed of a set of components, coupled together to function in a way that might not be apparent from the functioning of the separate component parts. Levine and Fitzgerald, 1992 System Dynamics is a set of methods to help us map and model dynamically complex systems -- to learn about why they behave the way they do and how to improve them. Encourages a different way of thinking about system behavior Two key tools -- causal loop diagrams and stock and flow models Rich, standardized language to describe and conceptualize systems 50 years of work improving methods to involve stakeholders in model building and utilization 19

General Approach 1. Define Problem What is happening over time that we are concerned about? First be sure to identify your client Clearly articulate the problem you would like to focus on What is the purpose of building and studying a model? DO NOT focus on a symptom! DO NOT model a whole system, just because you can! 20

General Approach 1. Define Problem What is happening over time that we are concerned about? 2. List Factors What are important drivers? Write as variables. Indicate any known direct causal connections. Of particular interest is your reference mode -- variables you will focus on over time to characterize your problem Consider developing a model boundary chart: 21

General Approach 1. Define Problem What is happening over time that we are concerned about? 2. List Factors What are important drivers? Write as variables. Indicate any known direct causal connections. 3. Draw Reference Mode Graph behavior over time. Any other factors needed to explain trends? Does the pattern suggest any familiar structures? 22

General Approach 1. Define Problem What is happening over time that we are concerned about? 2. List Factors What are important drivers? Write as variables. Indicate any known direct causal connections. 3. Draw Reference Mode Graph behavior over time. Any other factors needed to explain trends? Does the pattern suggest any familiar structures? 4. Build A Dynamic Hypothesis and System Map This could be a causal loop diagram, a stock and flow model (or a combination of the two). In any case, it is a causal hypothesis about system behavior. 23

Causal Loop Diagrams Give a Language to Talk about Feedback and Make Our Assumptions Explicit We start by looking for important causal relationships Change in the SAME direction A B When A increases, B will tend To increase, all else equal. Or when A decreases B will tend to decrease, all else equal. They change in the same direction. Change in the Opposite direction A - B When A increases, B will tend to decrease, all else equal. Or when A decreases B will tend to increase, all else equal They move in the opposite direction. Houses Residents Traffic 24 - Reported New Traffic Desirably of Community Construction

Parts of a Feedback Loop Diagram 2. Arrow -- Means one variable affects the next one in some direction, all else being equal. 1. Variable -- Important factors in the systems. Can go up or down. Shared support for innovation Investment in innovation R Word of mouth Awareness of positive results Positive results 3. Sign - S or means the second variable changes in the Same direction as the first. O or - would mean the Opposite direction 4. Type of loop -- R for reinforcing. B for balancing. 5. Name for the loop 25

A Feedback Loop That Builds On Itself Is Called a Reinforcing Loop They are also called positive feedback loops, virtuous cycles, vicious cycles, bandwagon effects, snowball effects Changing a variable in one direction produces a response in the same direction of that variable. Results Figure 1 in Repenning Positive loop of reinforcement 26 Effort allocated R reinforcement Commitment to the innovation Repenning, N. P., A Simulation-Based Approach to Understanding the Dynamics of Innovation Implementation. Organization Science. 13(2):109-127.

Behavior Over Time: Reinforcing Loops Are the Engines Behind Exponential Growth Reinforcing loops create growth, usually exponential growth. They give a system the potential for growth. Reinvestment Sawmill capacity R Econ Growth Engine Revenue Lumber production The bigger something is, the faster it grows MMBF 2500 2000 1500 1000 500 0 Lumber Production 1970 1980 1990 2000 Data: U.S. Census -- Lumber Prod. and Mill Stocks 27

More Than One Loop Can Intersect Observation of the effortresults linkage by others R2 Results Diffusion Figure 2 in Repenning The diffusion process Effort allocated R1 Reinforcement Commitment to the innovation 28

Balancing Loops Seek Balance or Equilibrium Balancing loops are created when there are an odd number of negative links. Balancing loops move the system towards a goal. They counteract change. Commitment to innovation - B1 Normative pressures Commitment gap Managers goal for commitment Commitmen to innovation Desired Actual 29 Normative pressure from managers Time

and Balancing Loops Can Calibrate the System Observation of the effortresults linkage by others R2 Figure 3 in Repenning External sources of commitment as a balancing loop Managers goal for commitment Results Diffusion - Commitment gap 30 Effort allocated R1 Reinforcement Commitment to the innovation B1 Normative pressures Normative pressure from managers

Delays Can Have Profound Effects on Feedback Loops Dampen feedback by weakening signal Effort allocated Results R reinforcement Commitment to the innovation Modified Figure 1 in Repenning Positive loop of reinforcement 31

Hints On Choosing Variables There is generally no final solution no right number of variables How to pick which to include Would the same basic dynamic exist if I took this out? Would people understand the sequence better if I left it in? Is this something I might be able to change? 32

Hints On Choosing Good Variable Names Use nouns or noun phrases They can clearly go up or down -- it would work as an indicator on a behavior over time graph For example, interest in surgery not people became more interested in surgery (which is a verb phrase) annual number of surgeries not more people got the surgery Use phrases with a clear sense of direction E.g., interest in surgery not attitude towards surgery Positive word of mouth not word of mouth If you have picked good names an observer will assign the same - and signs that you do. 33

Stock and Flow Diagrams Stocks represent accumulations and are generally measured in units (gallons, people, tons, etc.,) Flows change the level of stocks and must be measured in units per time (gallons/day, people/month, tons/year, etc.,) They often are verbs. Clouds represent factors outside our consideration at this point 34

ACTIVITY: Sketch Out What Is Happening To The Stock Of People Over This 30-Minute Period 40 35 30 People/Minute 25 20 15 10 5 Entering Leaving 35 0 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 M inute

Examples of Stock/Flow Diagrams People who are obese who do not do vigorous physical activity Adopting physical activity routine People who are obese who do vigorous physical activity Dropping routine 36

Examples of Stock/Flow Diagrams Work to be done beginning work Work in process completing work Work really done starting rework doing work incorrectly Known rework rework discovery Undiscovered rework 37

Robust equation forms Progress Cumulative progress 38 Rockefeller College of Public Affairs and Policy George Richardson 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 39 Rockefeller College of Public Affairs and Policy Effect of motivation Effect of... George Richardson University at Albany State University of New York

Robust equation formulations Effort (hours/month) Progress Cumulative progress Effectiveness (tasks/hour) 40 Rockefeller College of Public Affairs and Policy George Richardson University at Albany State University of New York

Robust equation formulations Hours per person per day Workers Workweek (days) Effort (hours/month) Progress Cumulative progress Effectiveness (tasks/hour) 41 Rockefeller College of Public Affairs and Policy George Richardson University at Albany State University of New York

Robust equation formulations Effort (hours/month) Normal effectiveness (tasks/hour) Progress Effectiveness (tasks/hour) Cumulative progress Effect of schedule pressure 42 Rockefeller College of Public Affairs and Policy Effect of motivation Effect of... George Richardson University at Albany State University of New York

Robust equation formulations Hours per person per day Workers Workweek (days) Effort (hours/month) Normal effectiveness (tasks/hour) Progress Effectiveness (tasks/hour) Cumulative progress Effect of schedule pressure 43 Rockefeller College of Public Affairs and Policy Effect of motivation Effect of... George Richardson University at Albany State University of New York

You can integrate CLD and stock and flow models 44 Hovmand PS, Ford DN (2009). Sequence and Timing of Three Community Interventions to Domestic Violence. American Journal of Comuunity Psychology; 44(3-4): 261-272.

General Approach 1. Define Problem What is happening over time that we are concerned about? 2. List Factors What are important drivers? Write as variables. Indicate any known direct causal connections. 3. Draw Reference Mode Graph behavior over time. Any other factors needed to explain trends? Does the pattern suggest any familiar structures? 45 4. Build A Dynamic Hypothesis and System Map This could be a causal loop diagram, a stock and flow model (or a combination of the two). In any case, it is a causal hypothesis about system behavior. 5. ID Leverage Points What are the levers for change (add to map)? What changes would lead to a more desirable behavior? What strategy could you use to achieve these changes?

Sometimes We Find Leverage Points -- Where Small Actions Yield Large Results 46 Maybe we should write that spot down.

We Get to Stop Following Rules That Don't Really Exist 47 Hey! They re lighting their arrows!...can they do that?

How can CLDs motivate intervention? Think about ways to: Reverse causal direction somewhere Change a sign Remove directionality Decouple two variables Tighten or loosen the connection between two variables Alter delays Add a loop whose effect cancels out the original one 48

49 How can CLDs motivate intervention?

50 How can CLDs motivate intervention?

How Can Stock-Flow Models Motivate Intervention? Consider each model flow What variables affect that rate of flow? Are any of these variables (or flows themselves) amenable to change? Draw these leverage points and related interventions right on the stock and flow diagram Simulate impact of intervention scenarios under alternate possible realities about the future 51

Possible interventions 52 Jones, Homer, et al. 2006

General Approach 1. Define Problem What is happening over time that we are concerned about? 2. List Factors What are important drivers? Write as variables. Indicate any known direct causal connections. 3. Draw Reference Mode Graph behavior over time. Any other factors needed to explain trends? Does the pattern suggest any familiar structures? 53 4. Build A Dynamic Hypothesis and System Map This could be a causal loop diagram, a stock and flow model (or a combination of the two). In any case, it is a causal hypothesis about system behavior. 5. ID Leverage Points What are the levers for change (add to map)? What changes would lead to a more desirable behavior? What strategy could you use to achieve these changes? 6. Test & Improve Theory Get feedback from others. Find data. Act and observe real world results. Reflect.

Why Use System Dynamics Methods Help us develop a shared understanding of the system Teach us to think differently about how systems behave (that is, in terms dynamics, circular causal feedbacks, accumulations, etc) Allow stakeholders to view the larger system they are embedded within Provide a framework for integrating what we know, and determining importance of what we don t know Support identification of high impact leverage points Offer a virtual world in which to try out and compare policies 54

How Do I Build CLD and Stock And Flow Models? Diagrams Paper and pencil (or whiteboard) Powerpoint Visio Isee/Ithink, http://www.iseesystems.com/ Vensim, http://www.vensim.com/ Simulation models Excel (with add-ons for simulation, such as Palisade @Risk (http://www.palisade.com/) Software: Stella, http://www.iseesystems.com/; Vensim, http://www.vensim.com/; Simio, http://www.simio.biz/; AnyLogic, http://www.xjtek.com/ 55

Where Can I Learn More? See resource document I have provided (also available from me today) Please do not hesitate to get in touch with questions and comments: Kristen Hassmiller Lich klich@unc.edu 56