Policy discussions around online anonymity have intensified in academic and professional communities. While anonymity facilitates free expression, it also liberates offensive conduct. This column identifies links between the Economics Job Market Rumors forum and patterns of user behaviour. Initially created to support job-market candidates in economics, the site has become a platform for sexist, racist, and defamatory posts, raising questions about the balance between protecting free speech, addressing the harms associated with anonymous online behaviour, and efforts to make economics a safer and more welcoming discipline.
In the context of ongoing debates about online anonymity and its potential harms (Müller et al. 2022), the role of anonymous forums like Economics Job Market Rumors (EJMR) has come under increasing scrutiny. Initially created to support job market candidates in economics, the site has become a controversial platform, notorious for toxic content including sexist, racist, and defamatory posts. The persistence of such content raises questions about the extent to which anonymity contributes to harmful online behaviour and whether platforms like EJMR reflect broader problems in economics (Ayarza and Iriberri 2024) and its academic culture (Advani et al. 2021).
The policy debate
The policy discussion around anonymity online has intensified, especially in academic and professional communities. While anonymity can facilitate free expression, it also enables users to engage in harmful and offensive behaviour without accountability (Blanchard 2017, Romer 2017). The debate often pits the values of privacy and free speech against the need to protect individuals and communities from harassment and abuse. Existing research on toxic behaviour online provides evidence that anonymity can amplify hate speech and misogyny (Wu 2018, 2020). Our study examines EJMR, a forum that is influential in the economics profession and a source of significant controversy. We provide new evidence that posting activity and toxic speech are pervasive across locations and institutions, including and especially elite universities. Our evidence suggests that instead of serving as a platform to elevate posts from less well-known institutions, EJMR perpetuates existing inequalities in the economics profession (Bayer and Rouse 2016, Antecol et al. 2018, Lundberg and Stearns 2019, Dupas et al. 2021, Hengel 2022).
Data and methodology
Our study relies on publicly accessible data from the EJMR forum, focusing on its unique system of generating anonymous usernames. EJMR assigns four-character usernames based on a combination of a user’s IP address and the topic ID. By analysing the statistical properties of this username scheme, we reverse-engineered the underlying algorithm, allowing us to identify the IP addresses associated with a substantial portion of posts made on the platform. This process was made possible by examining the structure of SHA-1 hashes, which EJMR used to create its usernames. We leveraged cryptographic techniques and utilised high-performance GPU computing to efficiently handle the vast scale of data involved.
To decode the username scheme, we employed a multi-step approach. First, we hypothesised that EJMR’s usernames were derived from SHA-1 hash functions based on the combination of the topic ID and the user’s IP address. This hypothesis was confirmed by testing known usernames against a variety of IP addresses and topics. We then developed software that computed almost nine quadrillion SHA-1 hashes, allowing us to enumerate every possible IP address for each topic and dramatically narrow down the candidate IPs for each username observed on EJMR. The use of graphical processing units (GPUs) enabled us to complete this computationally intensive task in a fraction of the time it would have taken with traditional CPU-based methods.
The next phase involved statistical analysis to match posts with their originating IP addresses. The uniform distribution of SHA-1 hash values allowed us to isolate IP addresses that consistently generated posts across multiple topics. By using stringent p-value thresholds (on the order of 10^-11), we minimised the risk of false positives. Our method ensured that we only attribute posts to IP addresses with very high confidence.
Figure 1 plots the cumulative distribution functions of the minimum p-values of posts for different hash positions. The orange line plots the CDF of the minimum p-values for posts with an incorrect (position-11) hash that contains only noise. Comparing this line with the approximate p-value p* ≈ 10^−11, represented by the dashed vertical line, it is evident that not one of the roughly seven million EJMR posts would be assigned to an IP address because no post would have a low enough minimum p-value. In contrast, the green line shows the CDF under the correct (position-10) hash. Around two thirds of posts have minimum p-values below p*, and for almost 30% of posts the minimum p-values are smaller than 10^−50.
Figure 1
Ultimately, our approach allowed us to attribute approximately two thirds of all EJMR posts to 47,630 distinct IP addresses, covering a time span from December 2010 to May 2023.
Key findings
- Widespread toxicity across the profession. We used several transformer-based machine learning models to classify EJMR posts by sentiment, misogyny, and toxicity. Our analysis reveals that EJMR is a significant hub of toxic speech, with 11.8% of posts classified as toxic, 3.3% as misogynistic, and 3.1% as hate speech. Problematic posts originate from both university and non-university IP addresses. These figures are alarmingly high even compared to other anonymous online forums such as Reddit. EJMR posts mentioning women are more likely to include problematic content than Reddit posts. Across the board, we find that mentions of women on EJMR attract more misogyny and hate speech than on Reddit but provoke a relatively similar level of toxicity.
- Participation from elite institutions. Contrary to the belief that EJMR’s most toxic users are not representative of the economics profession, our analysis shows that posting behaviour is prevalent across all levels of the profession, including elite universities and prominent research institutions. Approximately 15% of posts originated from university networks, including those of top-ranked institutions. Figure 2 reports the share of posts accounted for by a given US university or research institution among all posts originating from IP addresses associated with US universities or research institutions.
Figure 2
- Increased engagement during the COVID-19 pandemic. Posting activity on EJMR surged during the COVID-19 pandemic, with the number of posts tripling in early 2020. This spike was driven primarily by an increase in off-topic discussions, indicating heightened user engagement during the period of widespread social isolation.
- Geographical patterns. The majority of posts on EJMR originated from major urban centres in the US and internationally, including such cities as Chicago, New York, London, and Hong Kong. Smaller cities with prominent research institutions, such as Cambridge and Berkeley, also featured prominently in our analysis. Figure 3 shows the share of posts with an assigned IP address originating from a given city.
Figure 3
- Drivers of posting behaviour. Despite the anonymity of users on EJMR, our analysis suggests that users are driven partly by a desire for attention rather than monetary or professional gain. We employed an empirical design inspired by existing studies on online behaviour (Srinivasan 2023), examining the relationship between initial post engagement and subsequent activity. We found that posts receiving significant attention within the first day of creation led users (strictly speaking, IP addresses) to post more frequently in subsequent threads. This behaviour points to an intrinsic motivation for social validation and engagement, even in the absence of tangible rewards. Additionally, the level of toxicity in posts was not significantly correlated with the rank of the contributing user’s institution, indicating that harmful content was distributed broadly across the profession.
Broader implications
The findings from our study have several important implications for the policy debate around online anonymity. First, they challenge the notion that anonymity online is an absolute shield against identification. As we demonstrate, it is possible to de-anonymise a substantial portion of users even on a platform designed to preserve anonymity. Often, anonymity is only as secure as a platform makes it. That security was significantly below industry practice in this setting, which has significant implications for users posting under the assumption of complete anonymity.
Second, our results highlight the role of elite academic institutions in toxic online environments. The prevalence of harmful content originating from top-ranked universities suggests that issues of harassment, sexism, and racism are deeply ingrained in the culture of the economics profession, and not solely a reflection of dissatisfied loners at lower-ranked schools who feel disfavoured by the profession.
Lastly, the surge in EJMR activity during the COVID-19 pandemic underscores the potential for online platforms to become more toxic during periods of increased social isolation and stress. This finding aligns with broader trends observed on social media, where anonymity and a lack of accountability can exacerbate harmful behaviour.
Conclusion
Our research provides a comprehensive analysis of the toxic content on EJMR and its link to identifiable patterns of user behaviour. The study contributes to the ongoing policy debate about the role of anonymity in online forums, particularly in professional and academic communities. It raises critical questions about the balance between protecting free speech and addressing the harms associated with anonymous online behaviour. We encourage policymakers and academic institutions to consider these findings in their efforts to create safer and more inclusive online spaces as well as a more welcoming academic economics profession (Pereda et al. 2023).
Source : VOXeu
