We explore how organizations can use system dynamics as the core analytical decision technology to achieve mission-critical goals.
In the The Basics of System Dynamics course we build and apply several basic System Dynamics models in the first 5 weeks of the course. In the last two weeks, 6 and 7, we will apply our inventory of skills and models to the problem of simulating a B (Benefits) Corporation in the context of a practical Business Canvas approach. For this portion of our studies we will use paper, pencil, and the VensimPLE platform for all of our work.
In the System Dynamics Inference course we apply Bayesian data analytics to calibrate, fit, simulate, and infer conclusions consistent with the causal model of the decisions we are studying. The causal model will be built in VensimPLE, read into R, where we can mash the model with data and use Stan to make probabilistic inferences. During the first 5 of 7 weeks we will build basic system dynamics causal models of various business decision, estimate probability models of decisions based on the causal models, and infer probabilistic results using Machine Learning and information criteria. During weeks 6 and 7 we will apply our knowledge and models to decisions in the global energy and associated commodities supply chain.
[UNDER CONSTRUCTION] The Informing Decisions course will help us formulate and solve problems to inform decision-makers within organizations using simulation and optimization, all deployed with spreadsheets. We will develop the skills and practice the techniques to structure and analyze a wide range of complex business problems to inform and support managerial decision-making in functional business application areas such as finance (e.g., capital budgeting, cash planning, portfolio optimization, valuing options, hedging investments), marketing (e.g., pricing, sales force allocation, planning advertising budgets) and operations (e.g., production planning, workforce scheduling, facility location, project management). Spreadsheets are used to assist in modeling, analysis, and communication of results and findings.
For these courses we will cross four computing platforms.
VensimPLE will help us build and simulate generative causal models, visualize results, and develop scenarios for decision makers.
The R programming language (with R Studio - Posit), the tidyverse of data, optimization, numerical integration, and visualization packages will provide a platform for analysis, inference, and visualization of results.
The Stan (for Stanislaus Ulam) probabilistic programming library with its ability to estimate systems of differential equations (the underlying mathematics of system dynamics) using Hamiltonian Monte Carlo simulation will allow us to estimate the uncertainty within the causal models we have built.
Lastly, spreadsheets? Yes, the ubiquitous, immediate satisfaction of near instantaneous results spreadsheet environment is used by millions. As a prohtotype it surpasses most other environments. But beware its use in production! We will use spreadsheet engineering practices to improve on modeling hygiene and model deployment for decision makers on the run.
(During the Spring 2025 session from January 13th to March 9th, the Basic System Dynamics course is offered as MBA 645 Special Topics: Strategic Management Science by the Manhattan College MBA Program. Please contact Dr. Marc Waldman, Program Director, at marc.waldman@manhattan.edu for more information about the program.)1
The (next to the) last mile …
Try to attend our Sicth Live Session tomorrow Saturday, April 19, 2025 from 10am-noon (ET, UTC-5) on Zoom: https://us06web.zoom.us/j/9177353014. As usual featured will be questions and answers and not a few solutions as we crank up the mechanics of this online course. While mechanics might annoy us from time to time, the purpose of modeling is to enable insightful analysis and interpretation. Sensitivity analysis will dominate much of the discussion. The session may be video’d for posterity and deposited on a Youtube playlist dedicated to this terms’s course experience.
This week we will add a two-decision-2 state model to the mix from Acting on Bayes. We may even perform this feat on a spreadsheet (again!).
Check out the video from an article by John Sterman in 1986. - EXPECTATION FORMATION IN BEHAVIORAL SIMULATION MODELS, [Sterman (1986),](https://systemdynamics101.com/notes/sterman-1986-behavioral-expectations-formation.pdf John Sterman develops the model as well in Business Dynamics, p. 634-643, with case studies following the baseline model. Here is an implementation of the model to highlight the character of initial conditions, and a switch to modify the initial present perceived present condition to change with changing inputs.
Now that we have filed our returns …
We continue with our spreadsheet implementation of Bayesian (probabilistic) decision analysis. This time around we apply our inferential analysis of probabilities of hypotheses (now called decision states) using evidence we have gathered (perhaps simulated) to two decision alternatives, across low and high risk scenarios for customer acquisition.
[Spreadsheet here](models/spreadsheets/(models/spreadsheets/mooseco-wolfco-limits-capacity-synthetic-waic-decision.xlsm)
We also walked back to Week 3 and Acting on Bayes for some further examples of decision optimization. We are bringing all of our experiences and models to the fore now. Interpretation is our next big hurdle.
When April Showers … Again
Try to attend our Fifth Live Session tomorrow Saturday, April 12, 2025 from 10am-noon (ET, UTC-5) on Zoom: https://us06web.zoom.us/j/9177353014. As usual featured will be questions and answers and not a few solutions as we crank up the mechanics of this online course. While mechanics might annoy us from time to time, the purpose of modeling is to enable insightful analysis and interpretation. Sensitivity analysis will dominate much of the discussion. The session may be video’d for posterity and deposited on a Youtube playlist dedicated to this terms’s course experience.
This week we will add a two-decision-2 state model to the mix from Acting on Bayes. We may even perform this feat on a spreadsheet.
Thermodynamics, Entropy, Model Comparison: WAICing through a comparison of models
In this episode we compare two models based on two data sets of MooseCo customers. Each is sampled at the same 7 MooseCo sites but one counts customers who took less than an hour to decide to transact and the other, well, more than an hour. We use a simple grid approximation model to mash together data with hypothesis about the average intensity of customer transactions under the two regimes of low and high touch sales experience. We use log predictive probabilitiles (log odds really: lppd, for short) and a volatility penalty (pWAIC = variance(sum(log(probability of each observation, given the model)) to measure information uncertainty (Wide Area Information Criterion (WAIC = -2(lppd - pWAIC)); entropy and the 2nd Law of Thermodynamics at work) and forecast predictability. We then compare the uncertainty and volatility difference between the two regimes and their Wide Area Information Criterion. Watch me fumble through a scatterplot - yikes, the spreadsheet platform was not very forgiving! A reboot of the platform allowed for a simple yet fairly decisive view of two distinct customer decision regimes.
For those of you on a spreadsheet track, try to replicate the inference worksheet for the model you have been building. For those on the R track, just continue to work the regular assignment with Nettle’s model. Or if you like, replicate the spreadsheet in R? That would be a useful exercise. Let me know if you have any concerns!
Thanks for your patience. Bill
When April Showers …
Try to attend our Fourth Live Session tomorrow Saturday, April 4, 2025 from 10am-noon (ET, UTC-5) on Zoom: https://us06web.zoom.us/j/9177353014. As usual featured will be questions and answers and not a few solutions as we crank up the mechanics of this online course. While mechanics might annoy us from time to time, the purpose of modeling is to enable insightful analysis and interpretation. Sensitivity analysis will dominate much of the discussion. The session may be video’d for posterity and deposited on a Youtube playlist dedicated to this terms’s course experience.
Here is yet another model two of your mates are working on. Cobb-Douglas Supply and Demand with Price Formation.
Yet another model with synthesizer.
Again, these spreadsheet models do not run very quickly, but they are pleasing in so many other ways.
Spring has Sprung
Here is a spreadsheet version of the R synthetic sampler model from Week 2.
Spreadsheet synthesizer: MooseCo data
in this semester’s Bayesian Decision Analysis - 2025 playlist.
We demonstrate a workflow in a spreadsheet to sample data from our limits to predation endogenized (partly) model of predator WolfCo and prey MooseCo model. We sneakily (week 3 material) use a hierarchical Bayesian prior-to- posterior generative model for synthesizing MooseCo customer count data.
We sample the initial customer market parameter using the Poisson distribution with Gaussian distributed lambda intensities. This will stand in for our synthesized and sampled observational model of customer base interactions in this model.
We then pull monthly (month 1, 2, …, 24) data from the 12,000 simulations using INDEX( MATCH()) to build a calculation region. Every time the spreadsheet recalculates, this region changes.
We feed, in this case, sMoose recalculated monthly sampled Poisson realizations into an interface region.
A VBA subroutine recalculates the calculation region, copies the interface region’s values, and pastes the values into a simulatioin region across 100 consecutive runs using OFFSET() to advance the pasting cell positions. A simple statusbar display monitors our sampling process.
We view the medians of the sampled runs in a graph.
We can access the spreadsheet synthesizer model here.
On my Lenovo, tricked out with several gig of RAM, it still took over 5 seconds for each run summing to over 8 minutes for this one-factor demonstration model. The https://systemdynamics101.com/add-inference site has the same model in R, which will run many more samplings much faster.
Try to attend our third Live Session tomorrow Saturday, March 28, 2025 from 10am-noon (ET, UTC-5) on Zoom: https://us06web.zoom.us/j/9177353014. Featured will be questions and answers and not a few solutions as we crank up the mechanics of this online course. While mechanics might annoy us from time to time, the purpose of modeling is to enable insightful analysis and interpretation. Sensitivity analysis will dominate much of the discussion. The session may be video’d for posterity and deposited on a Youtube playlist dedicated to this terms’s course experience.
Spring is here!
Try to attend our second Live Session tomorrow Saturday, March 22, 2025 from 10am-noon (ET, UTC-5) on Zoom: https://us06web.zoom.us/j/9177353014. Featured will be questions and answers and not a few solutions as we crank up the mechanics of this online course. While mechanics might annoy us from time to time, the purpose of modeling is to enable insightful analysis and interpretation. Sensitivity analysis will dominate much of the discussion. The session may be video’d for posterity and deposited on a Youtube playlist dedicated to this terms’s course experience.
You may access the Spring 2025 playlist here.
This week we force our model to curtail the predatory activity and reactive decisions of two interacting organizations. Both want to acquire, retain, and reduce switching of customers. But we can also interpret these models as interactions between technological components, humans and machines, humans and humans sharing work (and thus rework). In this note we produce a model in a spreadsheet, again. But this time we add a potential market of customers, an allocation of potential customers (a very naive one at that) to the predator and prey customer bases, all to limit the growth of the market.
Here is a video, and supporting spreadsheet model, for us to peruse.
The Ides of March Await Us!*
Try to attend our first Live Session tomorrow Saturday, March 15, 2025 from 10am-noon (ET, UTC-5) on Zoom: https://us06web.zoom.us/j/9177353014. Featured will be questions and answers and not a few solutions as we crank up the mechanics of this online course. While mechanics might annoy us from time to time, the purpose of modeling is to enable insightful analysis and interpretation. Sensitivity analysis will dominate much of the discussion. The session will be video’d for posterity and deposited on a Youtube playlist dedicated to this terms’s course experience.
Some housekeeping notes:
For those registered in a current Manhattan University MBA course, access the WALL course blog-site on the course Learning Management System (Moodle) and post your response there for credit. Examples of responses from other participants in the course are located at this public site https://systemdynamics101.blogspot.com/.
For those registered in a current Manhattan University MBA course, access the weekly grade assignment activity on the course Learning Management System (Moodle) and post your response there for credit. Answer the questions, and upload your first model.
Due dates are not deadlines. The only deadline in the course is at its completion when grades must be posted to the Registrar for credit. But the due dates are there to help us pace ourselves, keep up with readings, and simply digestion of the complex of ideas we are attempting to conform to the reality of what we are modeling and ultimately interpreting for decision
Et tu? Brute?
Welcome to our first week together as we explore strategic decision intelligence with Bayesian System Dynamics and highly interactive models of business decisions.
Here is a video, and supporting spreadsheet model, for us to peruse.
Video: spreadsheet predator-prey model.
We bring a Vensim model through its equation documentation into a spreadsheet. We then simulate the model and plot results. We find that this development might help us peer into the model mechanics of input and output flows as well as the accumulation (yes, a simple Euler integration) of state values in stock variables. All is System Dynamics of a complex predator-prey interaction. We might consider reusing this model as we deepen our understanding of more complex models.
Enjoy!
We are about to begin our first week together in MBA 645 Strategic Decision Intelligence at Manhattan University with the module Add Inference. Welcome to all who enrolled. A welcome to those who might begin a self-study course for your own edification.
Enjoy the ride!
You can text me on my mobile (917-767-7980) anytime. Please let me know who you are and give me 24 hours to respond. I’m usually a bit quicker than that. We will have live sessions on zoom every Saturdays from 10am-noon.
Thanks, Bill
Enjoy, and always encourage one another daily, while it is still today!
William G. Foote, Ph.D.
Mobile/Text: 917-767-7980
Zoom: https://us06web.zoom.us/j/9177353014
GitHub: https://github.com/wgfoote/
Office hours (MBA 645 Summer 2024):
At the end of these courses students can expect to demonstrate progress in meeting the following goals, proposed here as actions with verbs in the imperative mood.
Pose a researched business question, model the causal influences implicit in the question, simulate potential causal relationships and counterfactual inferences and their sensitivities, and align inferences with decision alternatives and plausible choices for stakeholders.
Deploy analyses which produce interactive analytical products using an industry-grade computational platform engineered according to a tradition of design principles.
Using endogenous generative models, summarize experience and beliefs about stakeholders, their data, and the processes that the generated data used, to infer potential outcomes to answer business questions.
Practice quantitative critical thinking skills through a compound of statistical and normative problem solving which links strategic policies and practices with stakeholders.
Understand the role of the analyst and the analytics process in the decision-making context of complex organizations and their environments.
Communicate analytical decision results to decision makers and other consumers of analytical products effectively using interactive tables and graphs.
For my part this curriculum emanates from over 45 years of learning from and teaching managers system dynamics and statistical inference at Fordham University, Clarkson University, Syracuse University and LeMoyne College. I have used SD techniques and simulations at a variety of financial institutions, high tech, energy, retail, governmental and not-for-profit organizations world-wide. I especially want to acknowledge the many years of working with my son, Andrew Foote, who, with his company Paraclete Risk Solutions LLC, was critical in the development, promotion, and delivery of systems models, strategy, consulting, and services to multiple public and private sector clients over the past 20 years.
I have taken liberally materials and ideas (some might say I curated materials) from several extant courses. They all flow from the avowed discoverer of the systems dynamics methodology, Jay W. Forrester, and his decades of work, and students, at the Sloan School of Management, MIT.
First, the maths: Harry Hochstadt’s Differential equations : a modern approach (1963-4) in the first 84 pages details the math behind system dynamics, namely solving systems of simultaneous differential equations.
Second, the numerics: Joel Ferziger’s Numerical Methods for Engineering Applications, 1978. Yes, FORTRAN. I moved most of these routines to MATLAB and APL2 in the 80’s.
Jay Forrester’s 1998 MIT Introduction to System Dynamics self-study course Everything you will need to know about the formulation and interpretation of System Dynamics models from the inventor.
George Richardson’s 2013 Albany University Public Policy courses.
John Sterman’s 2013 Introduction to System Dynamics MIT-OCW course
Ventana System’s VensimPLE Modeling Guide and Tutorial along with Tom Fiddaman’s MetaSD model library
The premise of this curriculum is that learning is inference. Learning can be reading, understanding, reflecting whether in our heads or with complex computing environments. We begin with the following chain of reasoning:
All events, and data collected from events, have a truth value.
Probability is the strength of plausibility of a truth value.
Inference is a process of attaining justified true belief, otherwise called knowledge; learning is inference.
Justification derives from strength of plausibility, that is, the probability distribution of a hypothesis conditional on the data and any background, prior, and assumptive knowledge.
The amount of surprise, or informativeness, of the probability distribution of data given our experiences, is the criterion for statistical decision making – it is the divergence between what we known to be true and what we find out to be true.
All statistical analysis, and reasoning within analysis, begins from a disturbance in the status quo. The disturbance is the outlier, the error, the lack of understanding, the inattentiveness to experience, the irrationality of actions that is the inconsistency of knowledge and action based on knowledge.
We are surprised when the divergence between what we used to know and what we come to know is wider than we expected, that is, believed. The analysis of surprise is the core tool of this course. In a state of surprise we achieve insight, the aha! moment of discovery, the eureka of innovation.
The course will boil down to the statistics (minimum, maximum, mean, quantiles, deviations, skewness, kurtosis) and the probability that the evidence we have to support any proposition(s) we claim.
The evidence is the strength (for example in decibels, base 10) of our hypothesis or claim. The measure of evidence is the measure of surprise and its complement informativeness of the data, current and underlying, inherent in the claim.
Copyright 2024, William G. Foote, all rights reserved.↩