By SHANNON O. WELLS
Pitt’s Morgan Frank is leading a team of researchers from multiple universities and entities to provide empirically grounded perspectives on U.S. unemployment and job loss based on such contemporary factors as artificial intelligence (AI), transition from fossil fuel energy sources, global pandemics and trade policy.
On the encouraging side, the team is uniquely armed with comprehensive data from a range of governmental sources.
“Nobody else has the type of data that we do,” said Frank, assistant professor in the Department of Informatics and Networking Systems in the School of Computing and Information. “We were, I think, among the first to use the state-level data.”
And nobody else, he emphasized, has county-level data.
“Not even the federal government has that data. It’s siloed in each state’s department of labor. Pulling that together will give us an unprecedented opportunity to study the many different factors shaping unemployment in the U.S.,” he said.
On the frustrating side, this is information that Frank and his fellow researchers believe should be readily available.
“I wish that the federal government had done this for us,” he confessed. “It seems awfully silly that I have to call 50 different people in 50 different state departments of labor to get this data.”
That said, the requests have been quite well received.
“When we do contact these people, they’re so excited that somebody is interested in this data for research. I think the IRS could back out exactly the same data that we’re going to collect directly from labor departments,” he noted, “but the IRS does not share data well, (and) they don’t do that lightly.”
Announced June 20, the collaborative Observatory for U.S. Job Disruption project is based on a $1.6 million gift from the Alfred P. Sloan Foundation. It brings together co-lead institutions Carnegie Mellon University and Massachusetts Institute of Technology (MIT), along with Pitt, Northeastern University, University of Virginia, the California Policy Lab and the U.S. Chamber of Commerce Foundation.
“The Observatory will provide a new perspective on unemployment across the U.S. and support empirically grounded discussions of job loss from, for example, AI, the energy transition, global pandemics or trade policy,” Frank said in announcing the research. “Currently, federal agencies report total unemployment for U.S. states and sub-state regions, but it is difficult to know which sectors, occupations or populations are contributing to those aggregated statistics.”
“We’re part of this big multi-institutional effort to find new data sources that will give us new insights into labor dynamics. (We’ll) work with each state’s unemployment insurance office to get monthly data by county detailing the most recent job and sector of employment for everyone who claims unemployment in that state,” he said. “This gives us a really big advantage in that we can now say, ‘Tell me where you are, tell me what job you have, and I’ll tell you the risk of claiming unemployment, basically a probability that you will claim unemployment.’”
To examine, for example, if fossil fuel industry workers claim more unemployment as government incentives shift to support a growing green-energy sector, the researchers collaborate with each state’s unemployment insurance program to collect the number of claimants including their most-recent employment sector and occupation.
A website provides a demonstration of cases for similar state-level data.
“Combined with employment statistics, this allows us to calculate the workers’ unemployment risk based on their role in their local economy,” Frank explained, adding that the research takes a rather unique approach to examining the relationship between AI and job/career disruption.
“Research is usually not about deploying AI to solve these research questions, (but rather) trying to study how AI or some other thing disrupts workers’ careers and the ability of a firm, or of a city, to adapt to these big changes in the economy.
“Traditional labor economics, usually they’re trying to get at something called causality, and those are usually a very high bar of analysis,” he said. “A lot of conditions need to be met to reach a causal conclusion, especially if you’re just doing data-driven research and not actually running a field experiment.
“And this is what traditional labor economists have been thriving on, and in order to do it, they often make a lot of simplifying assumptions about the economy.”
To avoid those assumptions, the Observatory researchers will instead attempt to study the economy and all its complexity “using all the different data sources you can find.”
Among approaches that would be more difficult in the traditional paradigm are examining career pathways in different cities.
“For example, take each city’s labor market, model it as a network of occupations based on how easy it is for workers to move from one job to another, and use the properties of that network to tell us how quickly that city will recover when there’s a big shock to the system,” Frank explained.
“We have a paper using this approach looking at how cities recovered from the (2008) Great Recession, and we’re working on using the same approach to study how cities recovered during the 2020 COVID recession.”
The Observatory for U.S. Job Disruption researchers aim to move beyond traditional labor economics by finding and incorporating less aggregated sources of data.
“For example, I can tell you monthly the total number of unemployment recipients in Pittsburgh using government data, but traditionally, I can’t tell you how many of those unemployment recipients come from a certain employer or had a certain occupation or came from a certain sector,” he said.
“If you’re a policymaker and you were trying to do something to mitigate unemployment or to help workers find their next job, knowing what sector they came from or what jobs they’ve had in the past is the first thing you would want to know to try to solve that problem.”
Much of the state-level data are available through Department of Labor grants intended to collect more precise unemployment data during the rise of the COVID epidemic.
“We’re profiting from that,” Frank said. “But there’s no federal agency that’s collecting this type of data.”
Ultimately, he sees the Observatory project combining the various data sources into a U.S. Chamber of Commerce resource for studying human resources-related data.
“Employment, unemployment, hiring, careers, promotions, things like that. This is a multi-pronged effort that I think will eventually end up in a government resource housed in the U.S. Chamber of Commerce.”
While earning his doctoral degree in computational science at MIT, Frank became increasingly interested in AI and its impact on careers in labor markets.
“It became clear to me that this question about how skills might be irrelevant with new technology and trying to figure out how different parts of the workforce interact, that problem lent itself to the tools from complex systems. And there’s a lot of labor data out there,” he said of what he called a “complex system and a big-data problem.
“It became clear pretty quickly that the tools and skills I was using to study these other problems were applicable to studying AI in the future of work …
“I was able to take this different perspective, but also collaborate with traditional labor economists who already do work on this topic. And that helped me really connect where my ideas could possibly fill gaps in the way things have been traditionally done in that field of research,” he said. “And we’ve been doing that ever since.”
Shannon O. Wells is a writer for the University Times. Reach him at shannonw@pitt.edu.
Have a story idea or news to share? Share it with the University Times.