[Paper Review] Promotion through Connections: Favors or Information?

Yann Bramoullé, Kenan Huremović|arXiv (Cornell University)|Aug 23, 2017

Names, Identity, and Discrimination Research18 references3 citations

TL;DR

This paper proposes a novel econometric method to disentangle the effects of favoritism and information in promotion decisions using heteroscedastic probit models with exogenous variation in connections. It finds that information—not favoritism—drives the majority of the positive impact of connections on academic promotions in Spain, with strong connections significantly improving evaluation accuracy.

ABSTRACT

Connections appear to be helpful in many contexts, such as obtaining a job, a promotion, a grant, a loan, or publishing a paper. This may be due either to favoritism or to information conveyed by connections. Attempts at identifying both effects have relied on measures of true quality, generally built from data collected long after promotion. Building on earlier work on discrimination, we propose a new method to identify favors and information from data collected at the time of promotion. Under weak assumptions, we show that promotion decisions for connected candidates look more random to the econometrician due to the information channel. We derive new identification results and estimate the strength of the two effects. We adapt the control function approach to address the issue of the selection into connections. Applying our methodology to academic promotions in Spain and Italy, as well as political advancements in China, we find evidence that connections may both convey information and attract favors.

Motivation & Objective

• To address the limitation of existing methods that rely on imperfect proxies for true quality when studying the role of connections in promotions.
• To develop a method that identifies the relative contributions of information and favoritism without requiring measures of true quality.
• To estimate the separate magnitudes of information and favoritism effects using observable data at the time of application.
• To apply the method to real data on Spanish academic promotions to assess the underlying mechanisms of connection effects.

Proposed method

• Uses heteroscedastic probit models to detect excess variance in the error term of connected candidates, which signals the presence of information effects.
• Identifies favoritism through differences in promotion thresholds between connected and unconnected candidates, assuming systematic bias in evaluation.
• Employs exogenous variation in connections (randomly assigned jury members) to satisfy exclusion restrictions and enable identification.
• Applies the method to data from Spanish academic promotion exams (2002–2006), distinguishing between strong and weak ties.
• Uses marginal effects and likelihood ratio tests to assess significance and robustness across subgroups (Associate and Full Professors).
• Relies on clustering standard errors at the exam level to account for dependence in the data.

Experimental results

Research questions

• RQ1 To what extent do connections improve promotion outcomes through information about candidate quality?
• RQ2 To what extent do connections improve outcomes through favoritism or systematic bias in evaluation?
• RQ3 Can the relative strengths of information and favoritism be estimated without relying on external measures of true quality?
• RQ4 How do the effects of strong versus weak connections differ in their informational and favoritism components?
• RQ5 Is the identification of both effects valid under plausible assumptions about how connections relate to observable characteristics?

Key findings

• The information effect from strong connections accounts for approximately 61% of the total positive impact on promotion, while favoritism explains only 39%.
• For strong ties, the estimated information effect is 0.157 (p < 0.01) in the preferred model, significantly larger than the favoritism effect of 0.123 (p < 0.01).
• For weak ties, the information effect is 0.077 (p < 0.1), while the favoritism effect is negligible and even negative in some subgroups.
• The marginal effect of strong connections on promotion probability is 0.061 (p < 0.01) when accounting for both information and favoritism, with information contributing 0.046 and favoritism 0.016.
• The likelihood ratio test confirms that the heteroscedastic model with both effects is significantly better than the homoscedastic baseline (p < 0.01).
• Robustness checks across different model specifications and subgroups (AP and FP) confirm consistent dominance of information over favoritism.