Do networks contribute to the sorting of workers into firms?

Lucas Sage


In the wake of (Abowd et al. 1999) (AKM), an extensive literature has explored the role of sorting of workers into firms and its consequences on absolute level of wage inequality (Card et al. 2018). Based on a statistical decomposition of wage distribution into worker and firm fixed effect (FE), this literature disentangles the respective contributions of person and firm to wage inequality. One major result is that “high wage workers” and “high wage firms” match more often. In other words, the sorting of workers into firms is not random but tends to accentuate pre-existing worker’s ability differences. These results are descriptively interesting but ask for further explanations: what are the mechanisms responsible for these sorting patterns? In this paper, I explore the role of social networks. Even though workers massively resort to their personal contacts to search for jobs (Pellizzari 2010), finding a job through a contact compared to alternative ways does not systematically leads to a higher wage. Such mixed empirical evidence would suggest that networks do not contribute to the sorting of workers into firms. However, Montgomery (1992) demonstrated that this, very likely, is an erroneous conclusion resulting from a confusion between job search and job finding methods. When workers combine informal (networks) and formal (public offers) job search methods, only looking at whether a particular job was found through a personal contact is misleading. Instead what matters is the worker’s network structure and how it shapes her opportunities. With a formal model, he showed that under certain conditions, networks could increase job opportunities and wages, even though a statistical regression would find the contrary. To overcome empirical strategies’ limitations, I build upon Montgomery’s work and develop an Agent Based Model (ABM) to explore the contribution of networks to the sorting of “high wage workers” into “high wage firms”. Two mechanisms can play a role, each of which can differ between formal and informal search methods: the job arrival rate, and the distribution of wage-offers. To increase the realism of the model and reduce the number of free parameters, I calibrate the distributions of individual’s and firm’s characteristics on a large French administrative employer-employee dataset. I take into account the role of worker and firm observed and unobserved heterogeneity by first running an AKM econometric model to extract the empirical distributions of workers’ and firms’ FE, which I then insert in the model. Workers are embedded in an artificial network which properties can be manipulated. In particular, I vary the degree of homophily based on observed characteristics (education, gender, age) in the data. While Montgomery modeled the job arrival rate and the wage-offer distribution’s mechanisms at an aggregate level, the flexibility of ABM allows to explicitly model the job offers arriving to each individual through their contacts. I explore the conditions under which the model generates the sorting of “high wage workers” into “high wage firms”, and whether networks contribute to generating more sorting than what would arise in a world with only formal search strategies.

← Schedule