“It’s a mansion with 50 rooms, and each one relates to all the others,” says Charles Komanoff, referring to his Balanced Transportation Analyzer, a spreadsheet that models in intricate detail the daily flow of all transit–public, private, wheeled and bipedal–in New York City. Komanoff says his model can show the way to policies that would rid New York of gridlock, if only the city had the political will to implement them (you can download a version of the .xls file here).?Wired has the story. Meanwhile, a new book recounts the consequences of overreliance on computer models in city planning: Joe Flood‘s The Fires explores the connection between a computer model that promised to revolutionize the New York Fire Department in the 1960s and 70s, and a subsequent series of devastating fires.[%comments]
A Shining City on a Spreadsheet
TAGS: data analysis, urban planning
and 
. You can also watch it on 
People so often forget that models are hypotheses that need to be tested extensively with real data. They are not the truth itself or necessarily a close approximation or even a description of it. They may be useful for understanding processes, but before trusting a model, ask yourself if you would go to sea in a scaled up version of that little plastic Revell battleship you assembled as a kid.
Yeah, forget computer or mathematical models – lets just go with our “gut” or whatever our corporate masters tell us – I am sure thngs will work out for the best then. A few bad computer models does not mean we should not try to analyze the consequences of our actions and mathematical models are one of the best ways of doing that.
There is a happy balance between #1 and #2. Models are great for parsimoniously representing a phenomenon so that we can explore alternate choices and boundary conditions. However, the parsimony means that we necessarily make simplifying assumptions. It seems critical to understand and test the robustness of the key underlying assumptions before basing actions on the predictions from the model. Naturally, the greater the cost of error, the greater the need for testing and validation. To either embrace models wholeheartedly without reservations or to discard them in favor of intuition is, in my opinion, not particularly wise.
So when it comes to attacking Keynesian policies the appropriate critique (that I’ve read, or been referred to from this blog) is that they didn’t base their policy prescription on a fully specified stochastic dynamic general equilibrium.
But when a guy comes along with a simple spreadsheet based on observed congestion data and simple traffic flow diagrams (to get from point A to point B the car has to pass through this and that street) to do simple cost-benefit analysis of the impact of simple congestion pricing, then that’s too much theoretical modeling !!
I used to really enjoy this blog but you guys have become too predictably ideological.
re: comment #3 . While I don’t pretend to understand the contents and equations in the spreadsheet, I will say, that, anytime the word ‘parsimony’ occurs two times in one short paragraph, I know I should pay attention.
NMB
PS – I’m glad the U.S. Uses modeling to test our nuclear weapons.
#3 “To either embrace models wholeheartedly without reservations [...] is [...] not particularly wise.”
If you can’t embrace your model results whole-heartedly, you’re doing it wrong.
It’s very easy (in theory) to achive the same result without going through the complexity of congestion pricing, etc.
Just raise the gas tax, so the price is close to the european levels, or even higher. And to make it tax revenue-neutral, lower income and/or sales taxes.
models can be wrong (or at least not entirely right) even if they are done right because many phenomena are not completely understood. models require a simplification of phemomena of necessity. in many instances,the making of the model,and the subsequent comparison of the results to the reality, is how we discover more about the modeled phenomena in question. so you can embrace the results of your correctly constructed model as a learning tool,but if you embrace your model results as the truth, especially when dealing with complex phenomena like the weather, etc., before confirming them as truth,that is hubris and has often led to error.
in sum, models are useful, some more useful than others, but need to be constantly monitored and compared to the actual phenomena which they represent in order that they be useful tools.