System Dynamics – 1ZM65/1ZS24 Lecture 2 September 8, 2014 Dr.ir. Bob Walrave Agenda • Recap of Lecture 1 • Structure and behavior of dynamic systems (chapter 4) • Causal loop diagramming (chapter 5) 13-4-2015 PAGE 1 Recap of Lecture 1 • Dilbert contemplating policy resistance. “The behavior of a system arises from its structure!” 13-4-2015 PAGE 2 Recap of this lecture (take-aways) • System Dynamics is: • About understanding why and how things usually work out differently than you expected (policy resistance) • About counterintuitive behavior of systems • Looking at the world differently: Feedback view • Seeing the bigger picture (bird’s-eye view) • A methodology to analyze problems • Learning System Dynamics requires a lot of practice! • 5 ECTS = 140 hours • 18 weeks until due date 7.8 hrs/wk 13-4-2015 PAGE 3 Reference: Business Dynamics 13-4-2015 PAGE 4 Recap of Lecture 1 • Characteristics of dynamic complex systems: • The modeling process: your decisions 1. Problem Articulation (Boundary Selection) unintended side effects your goals situation goals of others 5. Policy Formulation & Evaluation 4. Testing 2. Dynamic Hypothesis 3. Formulation decisions by others 13-4-2015 PAGE 5 Example of Test Question (Lecture 1) • This figure describes the learning cycle. However, there are many barriers to learning that cause this cycle to fail. • Which of the following examples is NOT a way to break this cycle? (which is NOT a barrier to learning) A) Time delay B) Dynamic complexity C) Misperception of feedback D) Policy formulation 13-4-2015 PAGE 6 Agenda • Recap of Lecture 1 • Structure and behavior of dynamic systems • Causal loop diagramming Sterman, Chapter 4 13-4-2015 PAGE 7 Interactions of Structure and Behavior • The behavior of a system arises from its structure. • Behavior = output of a simulation model • Structure = input of a simulation model, consisting of • • • • Feedback loops (causal loop diagram) Stocks and flows Time delays Nonlinearities structure of the system Simulation Model behavior of the system 13-4-2015 PAGE 8 Interactions of structure and behavior • Structure + + eggs chicken road crossing s - + • Behavior over time chickens time chickens chickens + = time ??? time 13-4-2015 PAGE 9 Fundamental modes of behavior of dynamic systems • Basis modes of dynamic behavior • Exponential growth • Goal seeking • Oscillation • Other forms of behavior • Equilibrium • Random • Chaos • Interactions of basic modes • S-shaped growth • Growth & overshoot • Overshoot & collapse 13-4-2015 PAGE 10 Exponential growth Besides growth, we can also have exponential decline! Behavior Structure money on bank account interest per year 13-4-2015 PAGE 11 Goal-seeking Behavior actual number of customers desired number of customers of a product Structure This system is characterized by exponential decay (i.e., a half-life). attract more customers by advertising in supermarket 13-4-2015 PAGE 12 Oscillation desired temperature in the room is 20ºC Behavior Structure actual temperature in the room is 15 ºC increase heater 13-4-2015 PAGE 13 S-shaped growth The interaction between positive and negative loops is non-linear. Behavior actual number of people in a bar Structure is there still room in the bar? people allowed 13-4-2015 in the PAGE bar14 Growth & Overshoot number of rabbits in a meadow Behavior Structure Food available 13-4-2015 PAGE 15 Overshoot & Collapse Model these fundamental modes of behavior in VENSIM PLE Behavior actual number of Structure people infected with a certain type of flu The more infected people, the less people may be infected in the future number of people that may be 13-4-2015 PAGE 16 infected Vensim illustration • State of the system Stock • Net increase rate Flow 13-4-2015 PAGE 17 Systems archetypes • Use these fundamental modes – by combining them – when you are modeling! • Besides these fundamental modes there are generic templates that can be used to … • Interpret a Causal Loop Diagram’s overall nature • Refine a preliminary CLD • See the leverage in the CLD’s structure 13-4-2015 PAGE 18 Four well-known archetypes are: 1. 2. 3. 4. Fixes that fail Shifting the burden Limits to growth Tragedy of the commons For more information on archetypes, see: Wolstenholme, E.F., 2003, Towards the definition and use of a core set of archetypal structures in system dynamics, 13-4-2015 PAGE 19 System Dynamics Review, 19(1): 7-26 “Fixes that fail” Increase of crime + Problem symptom Fix Recruitment of police to reduce reported crime - + delay + Unintended Consequence Shortage of prison cells, leading to shortening prison sentences 13-4-2015 PAGE 20 “Shifting the burden” Expediting orders for important customers Quick Fi x + Lead time is too long in a production plant + - Si de Effect Probl em symptom delay + Fundamental Sol uti on Lead time of all other orders becomes even longer. 13-4-2015 PAGE 21 “Limits to growth” C onstraint + + Effort + Li miting Action Perform ance + Increase of sales effort leads to increase of orders - Customers are unsatisfied with long delays, leading to decrease of future orders Production capacity cannot keep up with increased demand, leading to longer production delays 13-4-2015 PAGE 22 “Tragedy of the Commons” + A's Acti vity N et gain for A R esource li mit + + + + Total Activity delay + Gai n per i ndividu al activi ty + + B 's Acti vity N et gain for B + Increase number of visitors to a national park, which increases profit, but also increases the damage to the park (disturb wildlife) 13-4-2015 PAGE 23 Agenda • Recap of Lecture 1 • Structure and behavior of dynamic systems • Causal loop diagramming Sterman, Chapter 5 5.1-5.4 13-4-2015 PAGE 24 Causal Loop Diagrams (CLDs) • Capture hypotheses about the causes of dynamics • Elicit and capture mental models • Communicate the important feedbacks that are responsible for a problem • See a CLD as a illustration of a story; a narrative that explain how a certain problem came to be • The loops within a CLD a characterized by ‘loop dominance’, which facilitate your story 13-4-2015 PAGE 25 Elements of a Causal Loop Diagram • Basic elements: • Variables • Arrows (causal link) • Link polarity (+ or -) • Loop identifiers • Additional elements: • Loop Polarity: − Balancing − Reinforcing • System Delays actual work pressure + coaching effort by experienced employees + delay + total number of productive employees perceived work pressure - + delay quality of work hiring new employees - delay 13-4-2015 PAGE 26 Link polarity: definitions and examples • Positive link: The higher the actual work pressure, the higher the perceived work pressure will be, after a certain (time) delay or: lower actual work pressure leads to lower perceived work pressure • Negative link: The higher the perceived work pressure, the lower the quality of work (more errors are made) or: lower perceived work pressure leads to higher quality of work actual work pressure perceived work pressure + - perceived work pressure quality of work 13-4-2015 PAGE 27 CLD Represent causation, not correlation Figure 5-2, Sterman: Causal diagrams must include only (what you believe to be) genuine causal relationships. 13-4-2015 PAGE 28 Label link and loop polarities ? Figure 5-3, Sterman ? 13-4-2015 PAGE 29 “Challenge” p.143: assign polarity, add loops • Consider the attractiveness of a product to customers as it depends on various attributes of the product. • Assign link polarities. Quality + Price Product Attractiveness • What feedback loops might be created as product attractiveness changes the demand for the firm’s product? Delivery Delay + Functionality + + Demand Figure 5-4, Sterman: The attractiveness of a product as it depends on various attributes. 13-4-2015 PAGE 30 Challenge p.145: Identify and label the polarity of links and loops Attractiveness of Market + Number of Competitors Profits + + Price - - Bank Cash Reserves + Perceived Solvency of Bank Net Withdrawals - 13-4-2015 PAGE 31 Three ways to determine loop polarity • The fast way: count the number of negative links • A better way: trace the effect of change around a loop • The formal way: do the mathematics Attractiveness of Market + Number of Competitors Profits + + Price - 13-4-2015 PAGE 32 Assign unambiguous polarities Incorrect Correct ? (+ or -) Price Revenue + ? Price - Revenue + Sales Figure 5-7, Sterman: Causal links must have unambiguous polarity 13-4-2015 PAGE 33 Name and number your loops Figure 5-9, Sterman: Name and number your loops to increase diagram clarity and provide labels. 13-4-2015 PAGE 34 Indicate important delays in links Price Supply Delay + Figure 5-10 Representing delays in causal diagrams 13-4-2015 PAGE 35 Variable names should be nouns or noun phrases Incorrect Correct + Costs Rise Price Rises + Costs Price ? Figure 5-12: Sterman 13-4-2015 PAGE 36 Variable names should have a clear sense of direction ? Figure 5-13, Sterman 13-4-2015 PAGE 37 Choose variables whose normal sense of direction is positive Incorrect Correct - + Costs Losses Criticism + Unhappiness Costs Profit Criticism Happiness Figure 5-14, Sterman 13-4-2015 PAGE 38 Make intermediate links explicit Figure 5-15, Sterman: Make intermediate links explicit to clarify a causal relationship. 13-4-2015 PAGE 39 Make goals of negative loops explicit Figure 5-16, Sterman: Make the goals of negative loops explicit. 13-4-2015 PAGE 40 Distinguish between actual and perceived conditions Figure 5-17, Sterman 13-4-2015 PAGE 41 Using Causal Diagrams to Model an Issue examples Using Causal Diagrams to Model an Issue • The majority of our problems can be captured by a combination of two loops: balancing & reinforcing • The balancing loop describes what our goal is, what we try to accomplish • The reinforcing loop describes what our main obstacle is or could be 13-4-2015 PAGE 43 Structure & Behavior actual weight (kg) 72 71 70 69 68 67 66 65 64 63 DOMINANT LOOP: BALANCING 0 4 8 12 16 20 24 28 32 DOMINANT LOOP: REINFORCING 36 40 44 48 52 week - First the balancing loop is driving the behavior, because the actual weight is always close to the desired weight - Around week 46 something happens, because the actual weight is not going back to its desired value - The reinforcing loop is now dominant 56 desired weight actual weight + + food intake B + deviation propensity to eat - + propensity to snack (between meals) + mood R - PAGE 44 13-4-2015 need for comfort food What do the loops tell us? Balancing loop: • The loop is selfstabilizing • The manager’s primary role is to monitor • Act only when there is an exception or likelihood of one, i.e., gap between desired & actual weight Reinforcing loop: •The loop is not selfstabilizing •The manager’s primary role is to act •If the cycle is vicious, find a way to break it PAGE 45 13-4-2015 Guest Satisfaction Index in a hotel • The GSI in the hotel is measured on a monthly basis • When the GSI is above 8.2, the customers are happy, as well as the hotel staff • However, the GSI recently dropped below 8.2, and now management is worried • During a staff meeting, hotel management tells everyone that customer focus should be increased • It is expected that after a while customer happiness returns and that the GSI will rise accordingly PAGE 46 13-4-2015 Where to start? • Start with a narrative, a story • Underlying a specific problem • What is the goal that you try to reach • What happens when you deviate from the goal PAGE 47 13-4-2015 Drawing the first loop Identify the goal seeking behavior in question •Goal: desired GSI Starting point: indication of drift from goal •GSI Shortfall (desired – actual GSI) focus on customer + GSI shortfall + customer happiness B + GOAL: desired GSI actual GSI + Decision: focus on customer •Effects: customer happiness, GSI Link variables & qualify links, insert delays Determine type of loop – balancing or reinforcing PAGE 48 13-4-2015 The problem is not as easy as it looks • The increased focus on the customer leads to extra work for most of the team • After a while, the extra work starts to exhaust the team • As a result of this exhaustion, some team members start making mistakes • Although the customer is still happy with the increased focus, the mistakes made by the staff have a negative effect on the GSI PAGE 49 13-4-2015 Drawing the second loop focus on customer + extra work + GSI shortfall + GOAL: desired GSI + customer happiness B actual GSI R + exhaustion + + - mistakes Starting point: what could be another (side-) effect of increasing the focus on customers Other variables: • More focus on customer can lead to extra work • Doing more work can lead to exhaustion • Exhaustion can lead to making mistakes, which reduces GSI Link variables & qualify links, include delays Determine type of loop – balancing or reinforcing PAGE 50 13-4-2015 What can we learn from the loops? + focus on customer extra work + GSI shortfall + GOAL: desired GSI + customer happiness B actual GSI R + exhaustion + + - mistakes • As long as GSI increases while the focus on customer increases, the BALANCING loop is in charge • Apparently the hotel staff can handle the extra work • When GSI decreases while the focus increases, you are in danger. The REINFORCING loop takes over • Act now! PAGE 51 13-4-2015 Using Causal Diagrams to Model an Issue • But CLDs can become far more complex! 13-4-2015 PAGE 52 Causal loop diagramming (NRC 28 April 2010) PAGE 53 13-4-2015 Using Causal Diagrams to Model an Issue • Problem definition: managing workload • Identifying key variables: • • • • • • Assignment rate Work completion rate Assignment backlog Grades/Performance Workweek Energy level • Developing the reference mode (next slide) • Developing the causal diagrams 13-4-2015 PAGE 54 Reference mode - example • When a person manages his workload according to the “grasshopper strategy”, work will not be done until the deadline is very near. • As a consequence the assignment backlog builds up • And when the deadline is near, working overtime (or even at night) is no exception. Figure 5-20, Sterman: Reference mode for the grasshopper strategy 13-4-2015 PAGE 55 Develop causal diagrams – managing workload + Assignment Rate Calendar Time + - Work Completion + Rate + - Assignment Backlog - B2 Time Remaining Due Date - Corner Cutting + Work Pressure - Effort Devoted to Assignments Productivity B1 Midnight Oil Figure 5-21: Sterman: Basic control loops for the assignment backlog + Workweek 13-4-2015 PAGE 56 Adding the burnout-loop + Assignment Rate Calendar Time + - Work Completion + Rate + - Assignment Backlog - B2 Time Remaining Due Date - Corner Cutting + Work Pressure - Effort Devoted to Assignments Productivity + B1 Midnight Oil R1 Burnout Figure 5-22: Sterman: The burnout loop + Workweek Delay Energy Level 13-4-2015 PAGE 57 Adding the “too tired to think”-loop + Assignment Rate Calendar Time + - Work Completion + Rate + - Assignment Backlog - B2 Time Remaining Due Date - Corner Cutting + Work Pressure - Effort Devoted to Assignments B3 Grades + Productivity + R1 Burnout B1 Midnight Oil Quality Control + Figure 5-23: Sterman: The “too tired to think” loop R2 Quality of Work + Too Tired to Think + Delay Workweek Energy Level 13-4-2015 PAGE 58 Adding “my dog ate my homework”-loop + Assignment Rate Calendar Time + Work Completion + Rate + - Assignment Backlog - B2 Time Remaining B4 Due Date - - Corner Cutting + Work Pressure - Effort Devoted to Assignments - My Dog Ate My Homework + Requests for Extensions + B3 Grades + Productivity + R1 Burnout Quality Control B1 Midnight Oil + Figure 5-24: Sterman: My dog ate my homework—Parkinson’s Law R2 Quality of Work + Too Tired to Think + Delay Workweek Energy Level 13-4-2015 PAGE 59 From practice: what is the best team size? • Linear view of the world: Small teams low communication overhead high efficiency But is this true? • Lets examine the feedback view of the world 13-4-2015 PAGE 60 From practice: what is the best team size? communication overhead + increase focus - knowledge sharing + - + fatigue + - dependency on tacit knowledge overtime priority change + efficiency + need to increase progress + team size + + progress So, what is better: a large or a small team? By the way: what is wrong with this CLD? + + vulnerability to disturbances/deviations delays + + + interruptions, complications, etc. Info/resources not available + 13-4-2015 PAGE 61 The beer game • https://www.youtube.com/watch?v=qxpgM8paegQ • https://www.youtube.com/watch?v=Aqi5-KzQZWc • Play the game here: • (http://www.beergame.lim.ethz.ch) 13-4-2015 PAGE 62 The beer game 13-4-2015 PAGE 63 Example of test question (Lecture 2) • In the Netherlands, in the spring, when the sun is shining for the first time since months and the temperature is rising, many people want to go out and enjoy the lovely weather, instead of staying at the office. As a consequence, as soon as the temperature starts rising, many people take a day off (or a short holiday). Below, a causal structure of this phenomenon is given. • What is the polarity of the links? A) Link 1 is positive, Link 2 is negative B) Link 1 is positive, Link 2 is positive C) Link 1 is negative, Link 2 is negative D) Link 1 is negative, Link 2 is positive 13-4-2015 PAGE 64 Example of test question (Lecture 2) • When people go on holiday, their colleagues have to take over some of their work. So, the workload for these colleagues is increasing. Most likely, these colleagues have to work overtime to get all the work done. Working overtime has a positive effect, because it increases productivity and reduces the remaining workload. But, it also has a negative effect, because it reduces motivation. The causal loops below describe this behavior. • What is the polarity of the loops? A) Loop 1 is positive, Loop 2 is negative B) Loop 1 is positive, Loop 2 is positive C) Loop 1 is negative, Loop 2 is negative D) Loop 1 is negative, Loop 2 is positive 13-4-2015 PAGE 65 Questions? 13-4-2015 PAGE 66