Evaluating Gender Wage Inequality in Academia using Causal Inference Methods for Observational Data
Zihan Zhang, Jan Hannig · 2025
Read Evaluating Gender Wage Inequality in Academia using Causal Inference Methods for Observational Data by Zihan Zhang, Jan Hannig (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-24078.txt.
Introduction
Zihan Zhang and Jan Hannig apply modern cause-and-effect methods to the pay gap between men and women in higher education. Their 2025 paper uses salary records for more than twelve thousand tenure-track faculty in the University of North Carolina system, matched with data on field, rank, and writing output.
The authors matched similar faculty and used a forest-based method to adjust for rank, subject area, output, and years on the job. After those adjustments, women still earned about six percent less than men with like profiles. The size of the gap shifted with career stage and output level. The study shows how such tools can give fairer readings of pay gaps in large records.
Catalog record
| CREATOR | Zihan Zhang, Jan Hannig |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 83 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-24078
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Evaluating Gender Wage Inequality in Academia using Causal Inference Methods for Observational Data Zihan Zhang, Jan Hannig Abstract Observational studies often present challenges for causal inference due to confounding and heterogeneity. In this paper, we illustrate how modern causal inference methods can be applied to large-scale academic salary data. Using records from 12,039 tenure-track faculty in the University of North Carolina system, linked with bibliometric indicators and institutional classifications, we estimate the causal effect of gender on faculty salaries. Our analysis combines propensity score matching with causal forests to adjust for rank, discipline, research productivity, and career experience. Results indicate that female faculty earn approximately 6% less than comparable male colleagues, with variation in the gap across career stages and levels of research productivity. This case study demonstrates how causal inference methods for observational data can provide insight into structural disparities in complex social systems. Keywords: Causal inference, Observational data, Propensity score matching, Causal forests, Gender wage inequality, Higher education 1 Introduction Understanding and explaining the structure and persistence of the gender wage gap remains a central question in labor economics and the social sciences. Since the foundational work of Blinder (1973) and Oaxaca (1973) , the literature has sought to distinguish disparities attributable to productive characteristics from those arising through labor-market institutions. The academic labor market, which prides itself on meritocracy and performance-based advancement, presents a particularly puzzling case: despite controlling for rank, field, and research output, standard wage equations consistently reveal lower earnings for women faculty relative to their male counterparts ( Ginther and Kahn, 2004 ; Ceci et al., 2014 ; Monroe et al., 2008 ) , yet this gap has rarely been examined using causal estimators. These disparities persist despite the growing representation of women in academia and widespread institutional efforts to promote equity in hiring and advancement ( Blau and Kahn, 2017 ; U.S. Bureau of Labor Statistics, 2023 ) . Two main perspectives have shaped scholarly understanding of gender pay gaps in academia. One attributes disparities to structural factors: women are more likely to occupy lower-paid positions—such as assistant professorships, non-tenure-track roles, or teaching-intensive appointments—which naturally depresses their average earnings ( Ginther and Kahn, 2004 ; Monroe et al., 2008 ) . Another perspective emphasizes differences in human capital, suggesting that salary disparities reflect variation in research productivity, often measured through publication counts or citation impact ( Ceci et al., 2014 ; Kim et al., 2024 ) . While some studies—such as Ceci et al. (2014) —report relatively small pay differences after accounting for productivity, their work continues to treat scholarly output as the primary metric of academic value. In practice, however, it is difficult to disentangle these explanations, as rank, research output, and compensation typically evolve in tandem over the course of a faculty member’s career. Much of the existing empirical literature relies on associational methods, such as Oaxaca–Blinder decompositions or pooled regressions based on cross-sectional data. These approaches are limited in that they cannot estimate counterfactual salary outcomes, nor can they fully account for unobserved confounding—such as career interruptions or uneven service burdens—that may influence both productivity and compensation. While prior work has documented the existence of gender pay disparities in academia, there remains limited causal evidence on how these gaps vary by institutional type, career stage, or research performance. These gaps are central to identifying equity interventions. In this study, we address these limitations by applying modern causal inference tools to examine the gender wage gap among professors in the North Carolina public university system. We merge faculty salary records from all sixteen campuses in the University of North Carolina system—which are legally required to disclose annual salaries—with academic performance data from Google Scholar and gender labels inferred via genderize.io. Our dataset covers salary information from 2022 and includes 12,039 faculty, with detailed information on academic rank, department, institutional affiliation, years of experience, and research productivity, measured using the log-transformed i10-index. This rich dataset enables us to move beyond descriptive associations and address core challenges in observational causal inference, such as confounding, covariate imbalance, and treatment effect heterogeneity, by estimating the causal effect of gender on log-transformed salary outcomes. We also explore treatment effect heterogeneity by examining how the gender wage gap varies across career stages and levels of research performance. Our study contributes to the literature on gender inequality and academic labor markets in three key ways. First, we provide a causal estimate of the gender wage gap across a comprehensive public university system, offering a more generalizable view of pay disparities than analyses limited to a single institution or field. Second, we move beyond average treatment effects by applying causal forests to uncover patterns of heterogeneity in the gender wage gap. This method reveals how the magnitude of pay disparities varies across career stages and levels of research productivity, offering new insights into where interventions may be most needed. Third, we release a reproducible workflow that integrates public salary disclosures, bibliometric data from Google Scholar, and modern causal inference techniques. This framework can be readily adapted for equity audits in other academic systems. The remainder of the paper proceeds as follows. Section 2 describes the dataset and variable construction. Section 3 outlines the methodological framework, including both parametric and non-parametric approaches. Section 4 presents the empirical findings. Section 5 concludes with implications and future directions. 2 Data and Variables To investigate the relationship between gender and faculty salary, we compiled a large database of employee salaries from 2022 of all sixteen institutions in North Carolina public university system. These sixteen universities within the one university system share the same governance structure and operate under the same policy orientation including faculty hiring and promotion ( The University of North Carolina, 2023 ) . The data used in this study was obtained from the University of North Carolina Digital Measures website ( The University of North Carolina, 2022 ) . The dataset contains 47,405 entries, with information on full names, age, initial hired date, rank, department, and the salaries of all employees, from faculty members to administrative and support staff, affiliated with one of the institutions. We removed non-faculty employees (such as department staff, facilities employees, athletic coaches, etc.) from the data based on the employee home department and a column that indicates job functions. Also, the professors with less than 27,000 yearly salary were eliminated ( Lu and Hannig, 2024 ) as these are likely clerical errors. In total, seven observations were dropped from the dataset. The resulting dataset contains 12,039 entries, focusing on tenure track faculty (assistant professor, associate professor, and full professor) in all institutions from North Carolina public university system. Our study focuses on examining differences in faculty salary across gender. The primary outcome variable is annual faculty salary. This variable was directly generated from the raw dataset. In our cleaned dataset, the average yearly salary is $128,000. We use a log 10 transformation for our analysis because salary distributions tend to be right-skewed, see Figure 2 , and the logarithmic transformation normalizes the data and reduces the influence of extreme values, Figure 2 . Additionally, this transformation allows us to interpret the results in terms of percentage change rather than absolute dollar amounts, which is more