1 00:00:08,000 --> 00:00:12,500 So, what do System Dynamics (SD) models look like? 2 00:00:12,500 --> 00:00:14,440 Let's have a look at the technique. 3 00:00:14,440 --> 00:00:18,090 This might be a difficult part but if you get confused, 4 00:00:18,090 --> 00:00:22,860 just have a look at the additional resources we included in this MOOC. 5 00:00:22,860 --> 00:00:25,770 Two types of diagrams are often used in SD. 6 00:00:25,770 --> 00:00:32,540 The first type, Stock-Flow diagrams, are mainly used to build simulation models. 7 00:00:32,540 --> 00:00:35,000 Stock-Flow diagrams focus, as suggested by the name, 8 00:00:35,000 --> 00:00:40,040 on stocks and flows, which are two types of variables. 9 00:00:40,040 --> 00:00:46,310 Stock variables are accumulations of all inflows minus all outflows over time. 10 00:00:46,310 --> 00:00:50,760 And te net flow actually determines the behavior of the stock over time. 11 00:00:50,760 --> 00:00:54,230 Or, mathematically speaking, stock-flow structures are differential or 12 00:00:54,230 --> 00:00:56,829 integral equations. 13 00:00:56,829 --> 00:01:01,070 Apart from stocks and flows, constants and parameters are also included 14 00:01:01,070 --> 00:01:02,220 in these models. 15 00:01:02,220 --> 00:01:07,360 And auxiliary variables are included to build models that closely correspond to the real 16 00:01:07,360 --> 00:01:09,720 world system. 17 00:01:09,720 --> 00:01:14,150 Auxiliary and flow variables often contain very specific functions, 18 00:01:14,150 --> 00:01:19,380 such as non-linear graphical functions or delay functions. 19 00:01:19,380 --> 00:01:25,630 Direct causal relationships between these variables are indicated with (blue) causal 20 00:01:25,630 --> 00:01:26,650 links. 21 00:01:26,650 --> 00:01:29,270 Note that the flows are causal links too. 22 00:01:29,270 --> 00:01:37,100 And also note that variables need to have real-world meaning and that all units need 23 00:01:37,180 --> 00:01:38,690 to be consistent. 24 00:01:38,690 --> 00:01:45,240 Then, if a model is fully specified, it can simulate behavior over time. 25 00:01:45,240 --> 00:01:51,860 This type of representation is good for building models and for communicating stock-flow structures, 26 00:01:51,860 --> 00:01:57,280 but it is not really appropriate for thinking in terms of feedback loops or communicating 27 00:01:57,280 --> 00:02:00,759 important feedback effects. 28 00:02:00,759 --> 00:02:04,520 For this purpose, System Dynamicists use Causal Loop Diagrams. 29 00:02:04,520 --> 00:02:10,729 These show causal links between the main variables, the polarity of these causal links, 30 00:02:10,729 --> 00:02:14,560 the feedback loops, and the polarity of these feedback loops. 31 00:02:14,560 --> 00:02:21,560 A positive causal link from A to B either means that A adds to B (if B is a stock variable), 32 00:02:22,200 --> 00:02:28,290 or that a change in A causes a change in B in the same direction. 33 00:02:28,290 --> 00:02:33,109 And for a negative causal link from A to B, one says that either A subtracts from B if 34 00:02:33,109 --> 00:02:38,939 B is a stock variable, or that a change in A causes a change in B 35 00:02:38,939 --> 00:02:41,379 in the opposite direction. 36 00:02:41,379 --> 00:02:45,680 A feedback loop consists of two or more causal links between elements that are connected 37 00:02:45,680 --> 00:02:50,959 in such a way that if one follows the causality starting at any element in the loop, 38 00:02:50,959 --> 00:02:53,340 one eventually returns to the first element. 39 00:02:53,340 --> 00:02:56,730 There are actually two types of loops. 40 00:02:56,730 --> 00:03:02,239 A feedback loop is called a balancing loop if an initial increase in variable A leads 41 00:03:02,239 --> 00:03:09,239 after some time to a decrease in A, but also if an initial decrease in A leads 42 00:03:10,199 --> 00:03:13,689 to an increase in A. 43 00:03:13,689 --> 00:03:19,549 In isolation, such feedback loops generate balancing or goal-seeking behavior. 44 00:03:19,549 --> 00:03:25,889 A feedback loop is called a reinforcing loop if an initial increase in variable A leads 45 00:03:25,889 --> 00:03:30,639 after some time to an additional increase in A and so on. 46 00:03:30,639 --> 00:03:36,489 And if an initial decrease in A leads to an additional decrease in A and so on. 47 00:03:36,489 --> 00:03:42,319 In isolation, such feedback loops generate reinforcing behavior. 48 00:03:42,319 --> 00:03:44,089 Exponential behavior for instance. 49 00:03:44,089 --> 00:03:47,669 But feedback loops hardly ever exist in isolation. 50 00:03:47,669 --> 00:03:54,669 Feedback loops are often strongly connected, and their relative strength changes over time. 51 00:03:55,469 --> 00:03:59,829 Complex system behaviors often arise due to such shifts in dominance between different 52 00:03:59,829 --> 00:04:02,400 feedback loops in the same system. 53 00:04:02,400 --> 00:04:06,150 When dealing with feedback loop systems consisting of multiple loops, 54 00:04:06,150 --> 00:04:10,919 it is hard to derive the behavior of the system without simulation. 55 00:04:10,919 --> 00:04:17,690 So, let's gradually transform the previous Stock-Flow Diagram into a Causal-Loop Diagram. 56 00:04:17,690 --> 00:04:23,600 The first step would be to turn the flows into causal links: an inflow into a stock 57 00:04:23,600 --> 00:04:29,130 is a positive causal link from the flow variable to the stock variable. 58 00:04:29,130 --> 00:04:33,820 But an outflow out of a stock is a negative causal link from the outflow variable to the 59 00:04:33,820 --> 00:04:40,780 stock variable: the higher the outflow is, the lower the stock variable will become. 60 00:04:40,780 --> 00:04:44,820 If we further transform the Stock-Flow Diagram into a Causal-Loop Diagram, 61 00:04:44,820 --> 00:04:49,220 we need to add the link polarities, identify the feedback loops, 62 00:04:49,220 --> 00:04:52,500 and derive their loop polarities. 63 00:04:52,500 --> 00:04:59,380 In this case we would end up with a reinforcing loop with a delay and a balancing loop. 64 00:04:59,380 --> 00:05:03,300 Depending on which loop is dominant and whether dominance shifts, 65 00:05:03,300 --> 00:05:07,070 this feedback loop system could generate several modes of behavior, 66 00:05:07,070 --> 00:05:13,650 for example exponential growth if the reinforcing loop remains dominant or S-shaped growth if 67 00:05:13,650 --> 00:05:17,550 the balancing loop takes over after some time. 68 00:05:17,550 --> 00:05:21,310 So, what's so specific about System Dynamics? 69 00:05:21,310 --> 00:05:28,310 1) SD models are largely endogenous theories, that is: model boundaries are chosen such 70 00:05:29,330 --> 00:05:36,330 that all important feedback loops are within these boundaries 2) SD models are rather aggregated: 71 00:05:38,500 --> 00:05:43,120 stock variables are often used to group rather homogenous individuals or items. 72 00:05:43,120 --> 00:05:49,170 This also means that we need to be sure the aggregation assumption holds in order to be 73 00:05:49,170 --> 00:05:53,120 able to use SD modeling. 74 00:05:53,120 --> 00:05:58,350 If the aggregation assumption holds, then SD has an advantage over less aggregated 75 00:05:58,350 --> 00:05:59,770 methods. 76 00:05:59,770 --> 00:06:06,430 SD is therefore most useful for studying the long term big picture. 77 00:06:06,430 --> 00:06:13,200 As already mentioned: SD models are integrated numerically to generate the behavior over time. 78 00:06:14,060 --> 00:06:19,420 System Dynamicists assess and interpret the resulting trajectories over time 79 00:06:19,420 --> 00:06:22,450 as general mode of behavior. 80 00:06:22,450 --> 00:06:26,120 For example, oscillatory behavior or S-shaped growth, 81 00:06:26,120 --> 00:06:29,230 or overshoot and collapse. 82 00:06:29,230 --> 00:06:34,170 In case of undesirable modes of behavior, system dynamicists actually analyze which 83 00:06:34,170 --> 00:06:40,020 structures need to be changed or added to change the undesirable modes of behavior, 84 00:06:40,020 --> 00:06:44,110 for example exponential growth, into more desirable modes of behavior, 85 00:06:44,110 --> 00:06:46,780 for instance an S-shape curve. 86 00:06:46,780 --> 00:06:56,500 So, outcomes are not interpreted as precise predictions; outcomes are general foresights at most. 87 00:06:56,550 --> 00:07:01,830 System Dynamicists are far more interested in improving our understanding and changing 88 00:07:01,830 --> 00:07:06,970 faulty mental models, and generating general policy insights than 89 00:07:06,970 --> 00:07:08,840 generating predictions. 90 00:07:08,840 --> 00:07:12,840 As such, SD models are essentially tools for thought. 91 00:07:12,840 --> 00:07:19,840 More, reflection beyond the model is also hugely important in SD. 92 00:07:21,310 --> 00:07:26,930 Let me give you two examples now: a very simple example first and then a more advanced. 93 00:07:26,930 --> 00:07:33,030 Let's make a very simple model about the electrification of the European car fleet between the year 94 00:07:33,030 --> 00:07:35,730 2000 and the year 2100. 95 00:07:35,730 --> 00:07:39,430 Say, at the start, there are nearly 100 million conventional 96 00:07:39,430 --> 00:07:43,210 vehicles and only 2000 electric vehicles. 97 00:07:43,210 --> 00:07:47,510 Owners of electric vehicles stick to their electric vehicles. 98 00:07:47,510 --> 00:07:50,570 At the end of the average lifetime of a conventional vehicle of, 99 00:07:50,570 --> 00:07:54,470 say, 10 years, conventional vehicle owners may consider buying 100 00:07:54,470 --> 00:07:59,560 a conventional vehicle or an electric vehicle. 101 00:07:59,560 --> 00:08:03,389 Suppose that the electrification process depends on some sort of incentive. 102 00:08:03,389 --> 00:08:08,350 Say, this incentive makes electric vehicles 5 times more attractive than without this 103 00:08:08,350 --> 00:08:12,630 incentive between 2005-2015. 104 00:08:12,630 --> 00:08:19,500 Simulating this vey simplistic model, we obtain the S-shaped growth of electric vehicles; 105 00:08:20,500 --> 00:08:25,500 following the electrification process, displayed below. 106 00:08:26,520 --> 00:08:32,019 The second example relates to material scarcity, mining and recycling infrastructure. 107 00:08:32,019 --> 00:08:37,709 The model displayed here is a bit bigger than the previous model. 108 00:08:37,709 --> 00:08:42,870 It consists of a demand sub model, a supply sub model, 109 00:08:42,870 --> 00:08:48,069 an extraction infrastructure sub model and a recycling infrastructure sub model. 110 00:08:48,069 --> 00:08:52,879 All sub models are linked although this does not show in this picture. 111 00:08:52,879 --> 00:08:59,879 The model was made with mineral/metal scarcity experts and used to develop scarcity scenarios. 112 00:08:59,889 --> 00:09:02,730 The model generates very interesting dynamics. 113 00:09:02,730 --> 00:09:09,200 The blue lines on the left and the right give an indication of mineral/metal abundance and scarcity. 114 00:09:09,200 --> 00:09:14,709 In this simulation, there are first some temporary periods of scarcity, 115 00:09:14,709 --> 00:09:20,499 after that scarcity becomes chronic beyond 2030. 116 00:09:20,600 --> 00:09:26,529 Changing a few assumptions, leads to even more interesting dynamics. 117 00:09:26,529 --> 00:09:30,920 These and other scenarios can then be used to design and test policies and strategies 118 00:09:30,920 --> 00:09:32,709 for different stakeholders. 119 00:09:32,709 --> 00:09:39,050 So, System Dynamics is useful for modeling complex social-technical systems and simulating 120 00:09:39,050 --> 00:09:41,709 dynamic complex behavior over time. 121 00:09:41,709 --> 00:09:46,930 Without SD or other simulation techniques, this is almost impossible. 122 00:09:46,930 --> 00:09:52,180 Here, we briefly looked at the basics of SD. 123 00:09:52,180 --> 00:09:58,540 From here on we recommend you to work through various e-books, books, online resources, 124 00:09:58,540 --> 00:10:02,449 do supervised projects, and take some advanced courses. 125 00:10:02,449 --> 00:10:09,100 After that, you are ready to make and use your own SD models and combine SD with other methods. 126 00:10:09,500 --> 00:10:11,900 And then the fun really starts. Enjoy!