A Markov Model of Referral-Based Hiring and Workplace Segregation

ABSTRACT

I construct a Markov model of referral hiring to look at two issues: inequality between groups and workplace segregation. The model differs from most models of referral hiring in that it explicitly considers a firm structure where employment opportunities arise. The model suggests that referral hiring does not directly produce inequality between groups at the population level, even though firms have a bias toward hiring workers from the same group they already employ. The result highlights a difference between outcomes of referral hiring at the population and individual levels in that individual firm biases balance out when aggregated. However, referral hiring does produce segregation of groups across firms in a given industry, and the degree of segregation monotonically increases in the amount of referral hiring.

Keywords:

• referral hiring
• segregation
• inequality
• social networks

I thank Mary Beth Combs, Sharon Harrison, George Neumann, Scott Page, two anonymous referees, and conference participants at the Midwest Economics Association, and Missouri Valley Economics Association annual meetings for helpful suggestions and comments. I am grateful to the James S. McDonnell Foundation for financial support.

Notes

¹See the appendix to Granovetter (1995) for a comprehensive literature review through 1995. Some of the better known papers include Bridges and Villemez (1986), Campbell and Marsden (1990), Granovetter (1973), Holzer (1988), Montgomery (1991), Montgomery (1992), and Mortensen and Vishwanath (1994). More recent work includes: Arrow and Borzekowski (2000), Calvo-Armengol (2004), Calvo-Armengol and Jackson (2004), Mouw (1999), and Mouw (2003).

²One exception is Mouw (1999). Since large amounts of job information travel through social networks, friends with similar social networks may tend to receive job offers from the same sources and thereby work in the same firms or industries. Thus referral hiring may produce firm level segregation by race or gender since such segregation exists in social networks.

³Mailath, Samuelson, and Shaked (2000) provide an excellent example of the influence of referral hiring on education choices. In their model they find an example of an equilibrium where search intensity of firms aimed at a particular group (a proxy for network connections) may result in low education choices for the group who is not the target of firm search efforts. The investigation of these interactions is important for understanding the effects of public policy related to equal opportunity. I more fully discuss these issues in section 5.

⁴Note that this assumption is relaxed in the extension in which firms are arranged in a hierarchy. In this extension workers prefer to work for firms higher in the hierarchy. Thus firms can hire workers from other firms below them in the hierarchy.

⁵While there are many other measures that may be chosen that would yield different absolute levels of segregation I am interested in relative changes in segregation as a function of s. Thus limiting the analysis to this one measure is sufficient for my purpose.

⁶Even though there is no bias toward referral hiring the Duncan and Duncan segregation measure is still greater than 0 due to random placement of workers according to a binomial distribution. Therefore the segregation results should be compared to the base case of s = 1 and not to a segregation level of 0.

⁷A better interpretation of this model may be that there are M types of jobs (instead of firms) at each level in the hierarchy but that any worker in job type m at level h is a perfect substitute for another worker at job type m′ at level h.

⁸I thank two thoughtful referees for this suggestion.

⁹For an example see Hong and Page (2001).