Social media, mobile communications, Internet searches: The amount of data produced by mankind is growing by the minute. The so-called NowCasting method promises to provide forecasts of future developments through real-time data analysis. It is used in a wide variety of areas, including attempts to predict migration movements. How exactly does it work?
The impact of the coronavirus lockdowns in Germany could also be observed with the aid of Google data: Search queries for "puzzles" suddenly shot up, "vacation in Germany" was as popular a search term as information about "unemployment benefits.” The Google Trends application provides some of the most up-to-date data on life on the web, taking the pulse of users on an hourly basis and providing evidence of how the world may change in the seconds or weeks to come.
This principle is called Nowcasting, or "prediction of the present." Instead of making forecasts based on past data, Nowcasting analyzes data in real time. In meteorology, this methodology is routine. During the pandemic, it has come to determine policy as well: Predictions of the pandemic were based on daily updated infection figures, making it possible to take appropriate measures.
What else can the Nowcasting method be used for? Worldwide, we are seeing first attempts to apply Nowcasting in the area of migration as well. One example: The Berlin-based Minor project office wants to use social media data to find out the level of interest among users in different countries in living and working in Germany – and whether this can be used to calculate potential immigration from abroad. The project is supported by the German Federal Ministry of Labor and Social Affairs, which wants to learn more about such migrants’ interests because of the shortage of skilled workers. "Migration data published by the Federal Statistical Office, for example, often lags far behind," explains Laura Spitaleri, a research associate at the Minor project office, a collaborator of the Robert Bosch Stiftung on various digital projects. "We wanted to test whether social media data can provide faster results."
For their model study, Spitaleri and her colleague Paul Becker used Facebook's advertising platform, where you can retrieve data for predefined groups. The selection parameters can be defined in great detail: country, place of residence, educational background, occupation, hobbies, political attitudes, etc.
The NowCasters searched for Facebook users in the top 20 sending countries to Germany who, according to Facebook, showed an interest in "Germany," "German," and, in the case of non-EU states, also "visa." In a first step, the researchers over the course of one year extracted the relevant data from Facebook using special software.
In a second step, the data was sorted and published in a working paper. It was found, for example, that Croatian Facebook users showed the highest interest in Germany: Almost ten percent of Croatian Facebook users searched for the above-mentioned terms. Another interesting observation: Interest in "Germany" and "visa" suddenly increased in the United States in 2021. "Triggered by Covid, there was a realignment of interests among workers and employees in the U.S.," is how Becker interprets the numbers. "In the process, some may have become more interested in moving to Europe."
"Migration data published by the Federal Statistical Office often lags far behind. We wanted to test whether social media data can provide faster results."
In a third step, it would be necessary to do a time-lag statistical study of whether and if this recorded migration interest is reflected in migration movements to Germany – this would also involve analyses of social developments in the potential sender countries. There is one important insight already: monitoring the interest in Germany among different occupational groups could prove to be a competitive advantage in recruiting skilled workers: "People could be targeted via advertising and provided with information about immigration opportunities and professional recognition in Germany," says Becker.
How Facebook calculates exactly which people are interested in a visa for Germany is a trade secret, however. Facebook's algorithms use likes or page visits by individual users to calculate whether there is an interest in "Germany" or "visa." But what kind of data does Facebook collect? And how do these automated categorizations work? The researchers do not receive an answer to this question.
Still, according to Paul Becker, the analysis of Facebook ads is a frequently used tool in the social sciences. That's because their user data seems to be quite accurate.
In 2021, the German computer scientist Ingmar Weber, together with the demographic researcher Emilio Zagheni, conducted an online survey of 137,000 Facebook users and found that the users’ information on their age, gender and country largely matched Facebook's calculations – depending on the country, there was a 86 to 99 percent match rate. "With Nowcasting, we've seen time and time again that it works well with Facebook," says Ingmar Weber, Research Director for Social Computing at the Qatar Computing Research Institute. For example, there have been attempts to predict election results based on social media data. And the RAND Institute did a study to predict migration movements to Europe.
One drawback of the method, however, is that access to social networks is not always easy or reliable. "Sometimes, Facebook shares data with NGOs and researchers, for example; at other times, it again restricts access," says Weber.
In the field of NowCasting, Weber currently sees a trend toward combining different data. "This can be Google Trends, news, Twitter, Facebook, but also satellite data," he says, "for example, to track crop failures, which in turn can lead to migration." Satellite data is becoming cheaper and is processed automatically using so-called computer vision. That's a great gain for us, Weber says. But the increase in data volume and computing power does not mean that every data project will automatically be a success. "You always get a signal or a number, of course, but there's the danger of reading too much into the results", Weber says. One should not be seduced by exciting fluctuations, correlations and anecdotal evidence, he adds. "The question is always whether predictions are sound and stable."
In a new project, the Minor researchers now want to find out whether Facebook data can be used to calculate the number of Ukrainian refugees in Germany at a given time. To do this, they will compare the number of Ukraine-related users active on German Facebook with the official reports. Will a reliable prediction model emerge? Paul Becker says: "We won’t dare make a forecast."