April 18, 2013
For about two decades voluntary pollution abatement programs have shifted in and out of popularity with national — popular in the George W. Bush administration, not so much under President Obama — state and even international policymakers.
The programs are designed to coax rather than compel businesses, industries and in some cases individuals to reduce, for example, airborne emissions, agricultural runoff or the use of toxic chemicals with incentives ranging from expert advice and public recognition to an easing of some regulatory restrictions and market advantages.
Despite considerable experience with them, however, questions remain about how effective the voluntary programs are, and under what circumstances. Michael Delgado, assistant professor of agricultural economics, is using statistical models and Purdue’s community cluster supercomputers to generate answers that could help guide policymakers in the future.
"Recently, environmental issues are at the forefront of policy at the state level, the national level, the global level," Delgado says. "I think it's a great environment to apply econometric models to get a really rich set of results that we can take to policymakers and say here's what we're finding and here's how you might use our results."
"If these programs don't actually work, then we're investing a lot of money and we're not actually reducing pollution," Delgado adds. "Using the kind of statistics that I use we can see not only did this program work but for what types of companies. We can say, OK, if it worked for firms in this industry then we should be targeting these voluntary programs to this industry and maybe not somewhere else where it doesn't seem to be effective."
Delgado also uses modeling to look at questions involving economic development, for instance the effects of education on growth rates in developing countries, and his research has involved issues such as consumer demand for hybrid cars as well.
In a broad sense his research, falls into the category of econometrics, the application of mathematics, statistical methods — and more recently computer science — to economic data. He works with large data sets on global, country and state scales and data covering individual industries and companies as well as individual pollutants.
His models tend to be nonparametric and semiparametric, which means they make fewer assumptions about the data, as is done in parametric models to simplify a problem, and demand more computational muscle to get timely solutions. The payoff is more robust results. He also employs computationally demanding nonlinear optimization problems to set parameter values for his models.
"The standard approach with a lot of simplifying assumptions can be run on any laptop or any desktop," Delgado says. "What I do can take a standard computer weeks. Even if you want to do it that way, maybe you can run one model. But to get a robust set of results we want to run a model multiple times, or we want to change some parameter values and run it a little bit differently and we want to do that multiple times."
The Carter community cluster, in which Delgado is a faculty partner, lets him run multiple jobs in parallel on a plethora of processors, both from his own piece of Carter and from nodes of other faculty partners when they’re idle.
"That's why I turn to a cluster computing environment where I can run things in parallel," Delgado says. "I just send the job to the cluster and it will do the job much faster and send me back the results."
Delgado also uses Purdue's DiaGrid distributed computing system, which can make thousands of processors available, and SubmitR, a Web-based interface for the R statistical computing environment he employs in his research. SubmitR is available free to Purdue researchers and their students on the DiaGrid hub at diagrid.org.
Delgado says the processing power and long run times available with SubmitR and DiaGrid have been a good fit for large Monte Carlo simulations on which he and colleagues work.
"SubmitR is particularly suited for running large simulations that require thousands of independent computations, at least for my purposes," Delgado says.