Happiness, an inside job?: Turnover prediction using employee likeability, engagement and relative happiness

In this paper, we describe how to rank employees for risk of turnover by using data obtained from a happiness self-reporting app. Two data sources are used: daily happiness and social interactions. The data spans 2.5 years and 4,356 employees of 34 companies based in Barcelona. For each employee, we build features at three levels: individual, company level and social interaction graph level. We develop various turnover risk models and we compare how different features affect performance prediction. The results show that the top three features that explain turnover risk are: ratio of likes received (likeability), posting frequency (engagement), and relative happiness (employee happiness normalized by company mean). Surprisingly, a priori expected explanatory features such as mean happiness level and the ratio of likes (positivity), were not significant. Precision@50 = 80% out of a test set with 116 churns, sample size N=2k.