It’s really hard for an OR analyst to do good OR modeling to solve real problems. First, the analyst needs to know a lot. He or she needs to know about the situation on the ground (the facts underlying the model and the data that will be available to support the model), the organizational background (who will use the model, and how much support their will be), and also needs to know a lot about OR models in general. Then the analyst needs to make careful choices in modeling, knowing full well that the mathematical model will not be “true” or “complete”. At best, the model will be useful. And this is just the start… So, how can we teach this difficult skill?

I think it is impossible to fully convey the art of modeling during lectures. Fortunately, we can teach classes that will be useful.

A commonly used method is to help students develop a repertoire of different models. One gives scenarios that are sophisticated word problems. “What is the best plan for manufacturing widgets over time at ABC widget factory, given the data in the spreadsheet?” This approach gives students a good sense for the types of models that they can develop and solve. Moreover, instructors use sensitivity analysis to give students a sense for how one can manipulate models to learn about the situation at hand.

Another approach is case-based. It is very similar to the first approach, except that the scenario is more subtle and complex.

But neither approach gives students a strong sense for structuring an unstructured problem. And it does not give a sense for the importance of making plausible assumptions that one may want to revisit later on.

Recently, I’ve been thinking about a different modeling approach that may be useful in class. It can be called “Back-of-the-envelope spreadsheet modeling.” (Please forgive the mixed metaphor.) The goal is to take a large open-ended problem, and then guide the students to come up with useful analysis. It would be best if the question asked was inherently interesting. “Should AIDS testing be required of all married couples?” “Should clean needles be provided to addicts?” The discussion in class is the key aspect, and so it would not be acceptable for someone to look up a paper on the web and just repeat its conclusions.

I’d be very interested in knowing whether it has been used in OR classes, and how to make it work. Please let me know. I’ll try to come up with some workable examples in future posts, either ones that I find or are sent to me or ones that I develop.

Tags: back-of-the-envelope, OR Education, spreadsheet models

March 9, 2009 at 9:49 pm |

Jim, I took a course for my MS in Operations Research at SMU taught by Dr. Dick Barr. The course was called Optimization Models for Decision Support.

http://faculty.smu.edu/barr/models/

This class was a case-study approach using OR methods to create and model decisions. The most useful class in OR I have taken to date. The class mimics almost exactly what you are describing. I would be Dick Barr would love to talk shop with you on this subject. Dick Barr is the Engineering Managment and Information Systems department head at SMU. He is also active in the Dallas INFORMS chapter.

Create topic of discussion by the way. Exactly what I’m trying to do in my blog.

March 10, 2009 at 2:31 am |

Jim, you should also look at Steve Powell and Robert Batt’s text Modeling for Insight.

http://mba.tuck.dartmouth.edu/pages/faculty/steve.powell/modelingforinsight/

I believe his problems (and approach) very closely mirrors what you are advocating. He has written several related Interfaces articles you might enjoy. Steve Pollock has also taken a similar approach in a graduate modeling course at Michigan.

http://www-personal.umich.edu/~pollock/IOE640/

They both advocate teaching OR modeling as done in an art studio where the focus is not on finding the “answer.” Students focus on creating and discerning between “good” and “bad” models. I have tried to implement some of these approaches at the undergraduate level within a decision sciences course. It has gone well, but you have to be willing to sacrifice topics. Ideally an additional course would be offered solely on business modeling.