Please find the paper and short summaries on my website: felixrusche.github.io
We are happy about any feedback.

We decompose this 20-point difference by simulating news reporting with and without negative reporting bias. We find that even when muting the negative reporting bias half of the gap remains.

A case in point is Germany’s most-watched nightly news – the ZDF heute Journal. The nightly news include a 1:30 min segment reporting economic and financial news from the stock market in Frankfurt.

To motivate: From 2017 to 2024, the main national stock market indices rose in the US and the five largest European economies. Yet, the average daily performance of all six indices turns from positive to negative when weighted by coverage of the countries’ ten most-read outlets.

Do the media focus on the negative? If so, why? We study this in the context of stock market reporting and identify and quantify an overlooked bias. A thread:

I find strong effects of community radio on women's empowerment. Exposure to community radio
📚 Increases affected women’s education by 0.3 years
👰🏽‍♀️ Reduces child marriage by 22%
👶🏽 Lowers young women’s fertility by ~10%.
💪🏽 Boosts women’s autonomy in household decisions by 11pp

I build on the displacement algorithm and external data to compute the PDF of original locations, conditional on observing a displaced location. I then compute the probability mass on the coverage area. This allows for clean identification of parameters despite the displacement.

I show that education and women's empowerment are key themes in radio programming. Using ChatGPT, I further show that radios argue against child marriage, in favour of girls' education, etc.

By 2020 more than 250 radio stations launched, covering a population of >300mln individuals.

Gender norms are extremely persistent and constrain women's life opportunities, especially so in poor countries. In my Job Market Paper, I show that grassroots media are an effective policy instrument to address gender norms at scale. #EconJMP #EconSky

In the second stage, our profiles ask their networks for job-related information or carrier advice via a direct message. The response rate is 21%. Most replies provide valuable information, like details about the company’s application process and even referrals.

The resulting networks of White profiles have 13% more connections (26 vs. 23) than those of Black profiles. We find that virtually all groups of users discriminate, but surprisingly, we document less discrimination among men and older website users.

Half of the profiles represent Black and half White male users. We signal race exclusively through AI-generated profile pictures, while keeping first and last names racially neutral. The profiles were otherwise comparable in other characteristics.

Ever wondered how discrimination affects job networks? In our two-stage experiment on LinkedIn, we provide insights into the question, causally studying the effect of discrimination on Black individuals' job networks in the U.S.