This month, I had the pleasure of speaking with Alice Wu, the UC Berkeley graduate who, in her thesis, exposed the gendered and sexist language used to describe women on EJMR, a website used widely by economists to disseminate information on PhD job market candidates. During our discussion, Wu and I discussed her intellectual motivation as well as the material consequences of her work. To Wu, her thesis was an exercise in producing thorough economic research and exploring how to employ modern machine learning methods to examine economically interesting labor market interactions. However, with the explosion of her work in the news and into the consciousness of the economic field, Wu recognizes the value of her work in spurring an important conversation among economists regarding how we can make our field more inclusive.
The transcript of our conversation has been lightly edited for readability.
EE: What led you to study EJMR?
AW: Honestly, it was a bit unexpected. I was picking a research topic for my senior thesis. I had worked for David Card for a semester before and he was going to be my advisor. In a conversation while I was debating research topics, he pointed out that there was a controversy going on at the time regarding EJMR and asked whether I would be interested in studying that platform. I had taken a machine learning course where I had gotten some experience with text analysis, so I decided to see if I could employ those skills while writing my thesis.
EE: Were you familiar with EJMR as an undergrad, before coming up with your research project?
AW: I first heard about EJMR during my junior year of college at UC Berkeley. I had some friends who were also undergrads and were interested in reading online reviews of their professors, and so they found the website. These were male friends and they knew I wanted to go to graduate school in economics, so I remember them showing me and saying “look, these [are] going to be your future colleagues!” I was shocked when I saw the explicit sexual content of posts and the racism and sexism on the website. When I asked my friends who were graduate students at Cal if they knew of the website they seemed really embarrassed.
EE: What has been the most challenging part of this research project for you?
AW: The most challenging part was definitely the beginning, after I had the raw text data and had done the cleaning of the dataset. I was thinking about what to look at and the first thing that I tried was sentiment analysis, which I had used in my machine learning course. However, many of the online open-source packages that perform sentiment analysis on text are built for specific styles of text or types of documents. For example, the sentiment analysis picked out the comments about women as beautiful as “positive” comments. When I saw that the algorithm I was using picked out the comments about women’s appearance as “positive” I realized the sentiments assumed by the algorithm didn’t match the purpose of the forum. It didn’t feel right or appropriate to me to discuss people’s physical appearance on EJMR, so it seemed strange to label it as positive. At that point, I had to take a step out of the pre-existing packages and think about what I wanted to get out of the data and what my actual questions were. That was the most challenging part of the research project.
EE: It’s exciting that you were able to apply machine learning and text analysis techniques in your thesis. What was your first experience using machine learning techniques and how do you see them contributing to economics?
AW: The project I did in my machine learning class at UC Berkeley was a sentiment analysis of tweets on twitter. I looked at whether tweets were positive or negative using LASSO to pick up positive or negative terms.
The machine learning literature is very different from the economics literature in terms of what type of analysis they are performing. It seems to me that overall, people who study machine learning only care about the prediction power of a model, rather than estimating and studying the relationship between two variables of interest. For example, there is a method called the “ensemble” method in machine learning where an iterative process over multiple models is used to fit data. In the end, the ensemble method aggregates results from the different models to get the best prediction. In economics, we can’t really use this method because we wouldn’t be able to make inference. If we want to use machine learning techniques in economics, we need to be able to perform some sort of causal inference. In my project, I had to think hard about what I was measuring and what economic interpretation I could infer from the data I was looking at.
EE: Turning back to the question of sexism in economics, why do you think women are underrepresented? Do you think that the online conversations on EJMR reflect people’s day-to-day experience?
AW: There are multiple components to answering this question. The stereotype literature (which is more psychology than economics) says that young girls often think they will be more successful in humanities than in quantitative fields. That belief can guide what courses women and girls take in high school and college. Since economics requires a lot of math, many women are unprepared to actually take economics courses or go on in the field. This is an entry level issue.
However, in addition to the entry level issue, there is a “leaky pipeline” in economics (higher rate of attrition for women than for men). This poses another problem and is more a workplace issue. There is an issue in the culture of economics that might turn people away, and my paper exposes a part of that culture.
I hope that the people who post these crude and objectifying comments are not representative of the profession. But overall, the whole atmosphere and culture of a profession can be shaped by a select few people. When you have a few people standing out in a certain way, that has a big impact on the culture of the field. It doesn’t really matter who these people are but just their existence and participation is disturbing.
EE: What do you see as the main contribution of your paper?
AW: I’m really happy to see that this study has triggered an open discussion about the culture of economics and how it may impact women’s representation in the field. People are acknowledging that there’s an issue and that’s the first step to addressing it.
EE: How do you see the relationship between academic research and politics/change?
AW: The honest answer is that I didn’t think about politics at all when I was writing this originally. I was working on my senior thesis and focused on the research question itself and I didn’t see it as a political tool in any way. However I’m glad that it has sparked conversation and may induce positive change.
EE: Your paper focused on the difference in language used to describe men and women on EJMR. However, there was clearly evidence of homophobia and racism in your results. What do you make of this evidence?
AW: I see very problematic language with respect to race and homophobia but the way I designed my research question I was looking at the simple difference between language used to describe men and women. The homophobia definitely shows up in my analysis with respect to how the LGBT community are described online. However, there should be more careful analysis on these topics.
Author: Emily Eisner