Why does it require Condor?

A run of the model is not as taxing as many classical supercomputing tasks. Running a set of parameter values to equilibrium for all 20 cities requires only about 20 minutes on one of the machines in the University of Wisconsin's Condor flock, which pools the power of desktop workstations and Linux clusters across Wisconsin's campus.

But Ferreyra, with the help of Condor team members including Peter Keller and Zach Miller, ran the model about 50,000 times to find the optimal parameter values, using as many as 200 processors at the same time. This computation took about a week, and, by the time Ferreyra was finished testing the model and working the bugs out, she ran the model about 15 times. Working on a single-processor desktop computer, the computations would have taken about 24 years, according to Keller.

"This is a classical Condor application—a very large number of independent computations that explore a parameter space. Condor provides the resources and management capabilities needed to support studies like this," says Miron Livny, a computer science professor at Wisconsin, head of the Condor project, and leader of the Alliance's Partnerships for Advanced Computational Services efforts.

After estimating the 20-city model, Ferreyra simulated the impact of vouchers in metro Chicago. She found that families using vouchers migrated to neighborhoods with lower housing quality, that vouchers usable at both Catholic and nonreligious private schools increased enrollment in both types of schools, that vouchers restricted to nonreligious private schools caused a smaller increase in private school enrollment, and that, in systems where use was restricted to nonreligious schools, enrollment in Catholic schools declined as the dollar amount of vouchers increased. She also found that low voucher levels of around $2,000 per year helped mostly the middle class, but higher voucher levels aided both the poorer and the richer as well. Finally, she discovered that public schools can be positively or negatively affected, depending upon the school district they are in.

These early insights are not the only upside to this extensive modeling project, though.

"A problem like this is very hard to analyze without a general equilibrium model like this one," says Nechyba, "and without computing resources [like Condor]. As we have successes like Maria's, a larger number of problems are recognized as benefiting from a general equilibrium approach, and we can expand the approach to new ideas. Projects like this help persuade others that this is one very useful way of thinking about the world."

This research is supported by the University of Wisconsin Department of Economics and the Robock Award in Empirical Economics.