MIT OpenCourseWare — searchable YouTube transcripts & AI Q&A

Session 10: Student Projects with Nelson Repenning

Nelson Repenning argues that system dynamics models should be built to answer a specific puzzle or policy question, not to model an entire system. He advises starting with a simple model that captures a core dynamic hypothesis, then adding complexity incrementally, and warns that including too many feedback loops makes it hard to understand the model's behavior. He illustrates this with student projects, such as a model of book sales versus reading rates, where the key is to formulate a clear dynamic hypothesis and test it with a minimal model.

Session 12: Student Projects (cont’d)

The instructor argues that the most valuable next step for students is to apply their modeling skills to real problems they care about, rather than seeking further textbooks. He emphasizes the importance of capturing the data-generating process in models, using examples like delays in reporting COVID cases or product registrations, and advises starting with simple, iterative models to avoid overwhelming complexity. He also stresses the need to understand the existing literature and speak the language of the field when making theoretical contributions.

Session 4: Predator Prey with Carrying Capacity

In this session, the instructor critiques common modeling pitfalls in predator-prey systems with carrying capacity, emphasizing that consumption rates must be constrained by both predator appetite and prey availability to avoid unrealistic outcomes like infinite consumption. He demonstrates how to formulate robust outflows using min functions or softmax alternatives, and shows how to identify feedback loops driving oscillations by tracing causal paths and testing hypotheses. The session concludes with an assignment to model a drug development pipeline as a stock management problem with failure rates and desired revenue growth.

Session 7: Market Growth with Boom and Bust

The market growth model's core challenge is that its sectors are tightly coupled, making iterative formulation difficult. The instructor recommends isolating production, ordering, and customer sectors with exogenous inputs to build and test each chunk independently. He also critiques common formulation errors, such as conflating time constants, missing replacement flows in capacity management, and using non-standard capacity units, which can cause steady-state errors or unrealistic behavior.

Session 5: Managing Drug Development Pipeline

The speaker argues that modeling a drug development pipeline requires accounting for success rates and growth targets to avoid oscillations and steady-state errors. He demonstrates that a cascade formulation, where desired stocks are calculated backward from the desired market stock, is necessary for equilibrium, and introduces an expansion factor (1 + growth rate × stage duration) to handle exponential growth. He also notes that ignoring supply-line adjustments or using current patent expirations instead of desired rates can cause instability.