Job-devouring robots or faithful droids?
A decade of underfunding the Bureau of Labor Statistics means we don’t know what’s shaping the future of work in the U.S. economy.
May 28, 2019
Photo credit: ipopba for istockphoto.com
As Congress considers increasing the budget of the Bureau of Labor Statistics for the first time in a decade, observers say there are some very clear gaps in the current statistical picture to fill.
Currently, nobody knows exactly what Americans do in their jobs because of outdated occupation categories and scant information on daily tasks in existing surveys. Lack of data on capital expenditure makes it unclear to what extent companies are adopting robots and artificial-intelligence programs. Nor has there been a survey of employer-provided training in more than 20 years, a gaping hole in briefing papers for policymakers promising to help workers who compete with robots or AI.
Alternative data sources such as stores’ back-office software could be tapped to make consumer-price data more current. New contingent-employment surveys have to be devised with questions geared towards capturing the emergence of the “gig economy.”
Since its formation in 1884, the BLS has produced the data on which the Federal Reserve and Congress rely to make strategic decisions affecting all aspects of U.S. economic life.
Now, statisticians say, years of under-funding means that the BLS cannot answer policymakers’ questions about critical tech trends.
“The BLS has had essentially flat nominal funding for the past decade,” said Erica Groshen, who was commissioner of the bureau from 2013 to 2017 and who now chairs the independent advocacy group Friends of the BLS. Taking into account rising wage and data-security costs, funding has effectively shrunk, Groshen said.“Right now, because of budgetary constraints we’re under, it doesn’t have the capacity to design and put out a new survey from scratch in a reasonable amount of time that would give you the gold standard answer to a burning question.”
While there remains confusion over how much of the $65 million proposed increase could be applied to a potential office move, the House of Representatives markup on the 2020 budget would likely provide the $30 million increase recommended by the Friends of the BLS for modernization of the bureau. Below are some of the critical information gaps that money could fill.
The last BLS survey targeting work-place training happened in 1995, when Americans were getting to grips with things like fax cover sheets and Word Perfect
What’s your job?
How many electrical engineers are there in the USA? It may sound like a simple question for a government statistics agency, but the most comprehensive data set—a mandated filing from employers used to calculate unemployment-insurance taxes—does not even ask for employees’ job titles. In other surveys, said Ken Poole, chief executive of the Center for Regional Economic Competitiveness, chemical, electrical and mechanical engineers are often lumped in together. “Limited resources means that it takes BLS’ Occupational Employment Statistics survey at least three-to-five years to add new occupations,” Poole said in an e-mail.
But the best place to get at specific job titles, according to Groshen, would be in individual worker records available from the same employers that file the generalized information for unemployment-insurance purposes. “You could supplement this with surveys that would ask the employers about their adoption of technology and the impacts of it.Then you’d be able to track when Cupcake Ninja emerges as a job, and how many people are doing it, and when DVD store clerk starts to disappear.”
Demythologizing the Effects of Automation on the Economy
Analyses of automation swing between apocalyptic visions of job-devouring robots and a newly empowered work force liberated from humdrum tasks by faithful droids.
Until a national statistics agency steps in, it will be impossible to know which side of the argument to believe.
“These techno panic fears around mass job displacement—are they really true?” said Rob Atkinson, President of the Information Technology and Innovation Foundation, a non-profit group that provides research on new technology to policymakers. “You’d really have a better sense how workers are faring there if you could match occupational data with deeper tech data from establishments.”
Atkinson concurred with Groshen on the need for supplemental surveys of work establishments. The most straightforward approach would be a questionnaire on capital expenditure.
“’What percentage of companies are using AI? How are companies engaged in ecommerce?’” said Atkinson. [These are] “simple things you would imagine we would know by now.”
The Paris-based international agency, the Organization for Economic Co-operation and Development, has this kind of information for many countries, so the U.S. has no excuse, he added. “Franky, it’s a joke.”
Even if more granular data on job titles was gathered, automation will not be understood until workers’ daily tasks and training are taken into account.
Earlier this year, economists at the Brookings Institution, drawing on work by the consultancy McKinsey, estimated that one in four U.S. workers are doing jobs at risk of near-full automation. They concluded that the U.S. urgently needed a broad survey of “the extent and pace of task-level job change.”
That’s why Groshen and others said one of the first steps the BLS should take on analyzing the extent of automation in the economy is a study of employer provided training. The last BLS survey targeting work-place training happened in 1995, when Americans were getting to grips with things like fax cover sheets and Word Perfect.As Groshen noted, “nobody can tell you if employers are doing more training now or less training now than in the past.”
Everyone knows that digital development has propagated a universe of side jobs known as the “gig economy,” and yet it’s almost completely invisible in current economic data
Employers may be retraining machine-displaced workers for more complex tasks. If you overlayed the training data, with statistics from the proposed capital expenditure survey, and existing employee tallies then we would be able to answer many of policymakers’ questions on automation.
“If we knew what the scale of automation was and how it was replacing jobs, it would make work-place training and education systems work better for our nation,” said Michael Horrigan, a former associate commissioner of the BLS.
Training data would also help policymakers concerned about professions succumbing to automation to target educational incentives.
“If you get a handle on what the private sector is doing in terms of training, then the government can look at spending their funds into gaps in what private sector isn’t providing,” said John Thompson, former head of the Census Bureau.
Consumer Price Data
Billions of dollars in social-security payments are benchmarked to this measure, and it’s used in setting the poverty line, among other critical decisions. Currently, much of the data collected is done through in-person surveys and consumer diaries.These are updated on a quarterly and weekly basis, respectively. In the age of electronic transactions, that’s price data at the speed of sloth.
A major retailer recently agreed to open up its transaction data to the BLS. The great advantage of this data is that it’s showing how the economy is working in real time.
“Imagine the head of the Fed could wake up every morning and have something on his or her cellphone saying this is what retail sales were yesterday,” said Atkinson.
Other transaction data that could replace the in-person surveys could include individuals’ credit-card bills, according to Groshen.
Web scraping, where algorithms go hunting for data on the Internet, could, said Horrigan, save the Bureau time and money. If an auto maker updated a car model in a way that changed calculations for price indexes, a Web scraping program would be able to digest the updates automatically.
“Big data techniques can save you a lot of money and would be more efficient than the old-fashioned way of having data collectors do it,” said Horrigan.
Thompson, the former head of the Census Bureau, said the use of modern techniques would also address the declining participation in government surveys. Federal statistical surveys are renowned for the response rates, which are essential to accurate and assessment of the state of the economy and reliable prediction, but they are not immune to either suspicion or a jaded environment saturated with SurveyMonkey-style feedback requests.
The Gig Economy
Sometimes, there is no substitute for a survey, and statisticians say that’s the key to an enduring mystery: Everyone knows that digital development has propagated a universe of side jobs known as the “gig economy,” and yet it’s almost completely invisible in current economic data.
The only survey focused on second and third jobs, the Contingent Employment Survey, has only been conducted once in the last 10 years, in May 2017. Funded by the Department of Labor, the survey was designed on a tight deadline, which did not allow for designing optimal questions to get at ride-hailing jobs and other “side hustles,” according to Thompson.
“What the BLS had before was, essentially, questions they’d asked 10 years ago when they were experimental questions,” said Thompson. “I think you really need to look at, and understand how, that segment of the work force would respond in a survey.”
To get at the scale of the gig economy, statisticians say, would require extensive field-testing of questionnaires ahead of their mass distribution. The key to a good survey is finding not only the right question to ask, but to find the right way to ask it so that it gives you the most useful answer. A redesigned contingent-employment survey could become an annual supplement to the existing household survey, Thompson and others say.
Above all, the Bureau of Labor Statistics needs extra funding to be flexible, to have the ability to respond to shifts in the economy with supplemental surveys sent to its huge database of households and employers.
“The economy is an evolutionary organism, but we treat it as if it were some static machine,” said Atkinson. “We need more insight into that evolutionary change that’s going on all around us. We’re really flying blind when it comes to the innovation economy.”