By Brigid Sweeney
For decades, policymakers have debated whether unemployment insurance provides a critical safety net during tough times, or whether it prolongs joblessness by reducing people’s incentive to find new work.
But, to answer this question, researchers need a clear measure of how much effort people are actually making to search for jobs. And this activity has proven difficult to track. That is why Scott R. Baker, an associate professor of finance at the Kellogg School, turned to Google—specifically, to its data on search traffic.
Baker and his collaborator, Andrey Fradkin of Boston University, used data from Google Trends, the free site that analyzes the popularity of various search queries across regions and languages. By tracking queries containing the word “jobs,” they were able to create a new system for measuring this activity, which they have dubbed the Google Job Search Index.
“We realized we could look at people’s job-search habits in a way that traditional government datasets had a hard time doing,” Baker says.
Analyzing job-search data
Pulling job-search data out of Google Trends is not as simple as, well, Googling it.
Since Google keeps the exact volume of searches for any term confidential, Google Trends results represent only the relative frequency of Google searches, set on scale of 1 to 100.
In order to estimate the raw numbers of job-related searches and ensure they were adequately large to use as the basis of their study, Baker and Fradkin tapped other Google tools. For example, Baker says, Google’s online advertising tool, Adwords, showed there were 68 million monthly searches for the term “jobs” in the year preceding April 2013, which was when they started this project.
Having validated the heft of the dataset, the researchers started analyzing it. In doing so, they knew that the GJSI offered several advantages over previous methods for job-search tracking.
First, the most commonly used survey to track time spent job hunting is the American Time Use Survey, a report released once a year by the U.S. Department of Labor. Though painstakingly thorough, the survey—which interviews about 26,400 households via phone throughout the year—doesn’t focus on out-of-work individuals, so it often contains fewer than five unemployed respondents from each state per month. Google, meanwhile, provides real-time access to many millions of searches that can be aggregated across geographies. This big-data oomph lets policymakers drill down to measure changes in local labor market conditions in a way not previously possible.
Second, the search engine’s ability to track searches for more specific queries—e.g., “marketing jobs,” “New York jobs,” or “Walmart jobs”— means researchers could use it to study how people are searching across sectors, regions, and employers.
And third, while other online platforms such as CareerBuilder offer proprietary data that can be useful, this information remains far more difficult to access and smaller in volume than an open data source like Google Trends.
Assessing the Google Job Search Index
But, clearly, the GJSI’s attractive qualities would not matter if the researchers could not prove the index was a true reflection of the volume of people looking for work.
“We have this measure of how often people are searching for something containing the word ‘jobs,’ but does that actually say anything about how people are searching for jobs in the economy?” Baker says.
To find out, the researchers compared its results against those of several other data sources. This included comScore, which tracks the web-browsing habits of 100,000 consenting Americans. They found the GJSI to be a good proxy for overall online job-search effort.
Moreover, benchmarking their results against those of the American Time Use Survey and other data, they found that Google job searches fluctuated in the same way as those reported in the survey, and that a higher unemployment rate corresponded to a higher GJSI.
“That showed us the GJSI is telling us something important about the amount of time people are spending searching for jobs,” Baker says.
With its significance validated, the GJSI could now be used to address a fundamental economic question—one that was hotly debated in the wake of the Great Recession.
Do unemployment benefits present a moral hazard?
Economists have argued the costs and benefits of unemployment insurance since its inception during the Great Depression as part of the Social Security Act of 1935.
“There’s always a question in economic policy that boils down to the discussion of two competing forces,” Baker says. “One is the liquidity effect, which serves to make the unemployed better off by redistributing money to people at a time when they need it. The other is moral hazard. It’s the idea that the more you support these programs, the more people will take advantage of them and not find new work.”
“There is a moral hazard story to be told here, but … the aggregate effect seems to be relatively small.”
After the Great Recession, most states saw an increase in their unemployment benefits from a maximum of 26 weeks to a maximum of 99 weeks. Did these more generous benefits deter people from working?
Through the Texas Workforce Commission, the researchers obtained detailed weekly records of the number of out-of-work Texans in each metropolitan area between 2005 and 2014, as well as the number of weeks of benefits people had left. They then compared weekly GJSI results for each of these metropolitan areas against the percentage of unemployed residents at each level of benefit. Their results reaffirmed earlier research showing that people search more for jobs when they have fewer weeks of benefits remaining.
In fact, they found that in areas where the average unemployed person had fewer than 10 weeks of unemployment insurance remaining, the index revealed 66 percent more search activity than in nearby regions where people averaged 10 to 20 weeks remaining—and a whopping 108 percent more than in metro areas in which the unemployed averaged 30 weeks left.
“This makes sense: if you’re getting close to the expiration of your unemployment benefits, you’d feel more pressure to search for a job,” Baker says.
The duo also crunched data across all 50 states, relying on the Current Population Survey, produced monthly by the Bureau of Labor Statistics, to track the duration of residents’ unemployment benefits between 2008 and 2014. Then, by using the same method they did in Texas, the researchers found that every 10-week extension reduced total search activity—by both the employed and unemployed—by only 1.5 to 3 percent. (After all, people with jobs sometimes look for a new one, too.)
Translated, that means “there is a moral hazard story to be told here, but we also show that the aggregate effect seems to be relatively small,” Baker explains. “If you’re looking to compare how much the overall increase in joblessness [during the Great Recession] could be driven by the increase in unemployment benefits, the answer is very little.”
In addition to shedding light on this economics question, Baker says the value of the GJSI extends well beyond job-search tracking. In fact, since the research was published, other academics have cited it “who aren’t looking at job-search data at all,” he says.
Instead, they are looking to use the same tool kit to develop other Google-based economic indices, including new ways to forecast consumer sentiment and economic uncertainty by tracking search terms such as “recession” and “bankruptcy.”
“This approach can be really useful for policymakers because it can be much more local and granular,” Baker notes, “and they can access it without a six-month or year-long lag.”