The Twitter data your company isn’t using…yet

Let’s say you are a behemoth book/food/cloud/everything retail company looking for a place to build a second HQ, wouldn’t it be helpful to have a dynamic, targeted view of labor markets in different localities? We think it would. Our interdisciplinary team of data scientists and behavioral scientists are finding promising new ways to translate social networking data into information organizations of all sizes use to make business decisions.

Recently, we have been spending time analyzing how publicly available data sets enable us to understand real world applications and specifically, how public interactions and social commentary influence and help us to understand our everyday lives. There is still much research that can be done to leverage social media data and the science of human behavior to better understand how these types of datasets can be used to the benefit of individuals, organizations, and indeed society at large. Through our understanding of I/O Psychology and modeling of diverse data, HCW continues to seek opportunities such as this to gain insight into the world of work and its effect on our everyday lives.

Research has shown the power of social media data when added to existing models to better identify potential health related issues (1). Building off some of the work that has been done surrounding social media data and health care, we began our most recent study into the world of work by collecting data from the CDC (health related features like Diabetes, obesity, smoking), American Community Survey (education, income), BLS (unemployment data) US Census (regional demographics) and Twitter. Tweets were categorized into six measures of sentiment including such emotions as anger, engagement and anxiety. The data were selected to give us an understanding of the regional health and demographics of the population and the Twitter sentiment provides access to an individual’s view of the world. We grouped the data by FIPS code (regional identifier) and modeled the data to explain hospital survey measures of care and efficiency like mortality rates and patient response to treatment. Grouping the data by county allowed us to account for regional trends and cultural effects that likely influence the measures being observed in the hospitals.

The most striking findings were the correlations between the social media features of our data set and its potential to explain regional employment trends. Specifically, Twitter sentiment showed high correlation and potential predictability towards understanding regional unemployment trends across various demographics. For example, in areas of the country exhibiting high levels of disengagement in their tweets, there was strong correlation with unemployment levels. Combining the twitter sentiment with health-related features and demographic data we started to see even greater explanatory power for regional unemployment levels.

We are continuing to study the relationships between diverse data sets and labor markets not only to feed our intellectual curiosity but also to further understand human capital models and their use in everyday business-related decisions. For businesses with extensive human capital needs, a near real-time awareness of the health of targeted labor markets can have profound impacts on both strategy and operation — something that could augment data from your current Census, university recruiting, salary benchmarking, and quality of life-based models.

But even if you are not looking to invest a few billion in a new campus, real time awareness of the labor market can be a powerful tool for directing your organization’s human capital programs. Recruiting is the most obvious, and indeed most direct. Recruiters could use this awareness to better focus recruiting efforts and more accurately target compensation in employment offers. Other Talent Management functions such as workforce planning and training and development could also use this kind of visibility into the labor market to understand the environment that they operate in and make better decisions about whether to build (train), buy (hire), or borrow (contract) the knowledge, skills and abilities they need.

References

(1) Eichstaedt, J. C., Schwartz, H. A., Kern, M. L., Park, G., Labarthe, D. R., Merchant, R. M.,et al. (2015). Psychological language on Twitter predicts county-level heart disease mortality, Psychological Science, 26(2), 159–169.