Capturing Unemployment Trends with Social and Mobile Data
New methods, when combined with old, may help paint a better picture of overall economic health.
Unemployment is perhaps one of the more telling signs of labor market health—inspiring agony when on the rise and jubilation when declining. But how accurately do unemployment rates capture the state of the labor market and of the economy in general?
According to Jameson Toole, senior data scientist at Jana, the methodology behind current employment data collection has limitations both in accuracy and application. Unemployment statistics are generally sourced from labor force sample surveys, social insurance and employment office reporting. Toole comments: “These surveys are prohibitively expensive to conduct on larger populations and can take weeks or months to process, limiting how detailed they can be in their estimates and how fast they can provide policy makers with useful information.” He adds, “For decades, unemployment estimates have been limited to the national or state level and reported once a quarter or once per month at best. Making things worse, the response rates to these surveys have been dropping in recent years, bringing about new potential biases.”
In recent years, research has suggested that social media can provide a more accurate basis for capturing unemployment trends in real time by tracking, for example, the incidence of key terms on Twitter like “axed,” “canned” and “lost job.” This offers a potentially lower-cost, large-scale alternative to survey reporting, although it may not be without a sample bias of its own—that of a population of Twitter users who may skew younger and more digitally active.
Toole is one of six co-authors of a recent study that looks at how a different kind of digital data can help predict unemployment: cell phone use. Based on anonymized cell data from a Spanish town where an automotive plant was closed in 2006, the study showed the total number of calls made by laid-off individuals decreased by 51%, compared with working residents, and that the unemployed also received fewer incoming calls—evidence of how job loss affects social interactions. “Our results suggest that more accurate estimates of unemployment can be made much faster using these new data sources,” says Toole. He continues: “Moreover, this is data that already exists. It is collected passively as people use the nearly seven billion mobile phones on the planet. Tapping into this data could be a tremendous help to policy makers who are making decisions based on delayed and/or inaccurate estimates of critical economic indicators. This potential grows even more when one considers adapting these methods to places in the world where institutional resources are so limited even the most basic surveys are out of reach.”
So, does this mean that we’re just around the corner from a total “disruption” of traditional means of measuring unemployment? Not exactly. Michael Clemens, senior fellow at the Center for Global Development, points out “[countries] have strong incentives not to change their survey concept of unemployment, since the whole point of the survey is to capture changes from period to period.” Still, Clemens acknowledges that new avenues for collecting unemployment data can make an important contribution: strengthening international comparisons. “In the absence of a grand global treaty obliging countries to ask the same questions on their national surveys and just accept the time-break in comparability, the only real alternative is to create new channels of acquiring unemployment data that are not nationally run surveys,” suggests Clemens.
In the end, it may be that improving measures and interpretation of unemployment isn’t so much a matter of abandoning old for new, but instead focusing on thinking about how old and new methods can complement each other as part of an updated labor market analysis toolkit. It’s a view Toole is keen to stress, emphasizing that the insights of his study were achieved by combining new big data resources with traditional surveys, not replacing them. “Both types of data have pros and cons and should be thought of as complements.” As new social and mobile data raise the potential to supply regional- and international-level indicators in real time, official statistics are still indispensable, as they offer the benchmarks necessary to temper the volatility of social media practices and lack of long-run trends.