Keywords: confirmation bias, media market, media bias
This paper develops a theoretical model where individuals with limited attention select a portfolio of news sources to learn about the true state of the world. Confirmation bias naturally arises as individuals favor sources reinforcing their prior beliefs; however, I identify conditions where they instead seek neutral or contradictory sources. This occurs among those with weak priors, for whom a mix of the contradictory source with the neutral source is optimal. The model further examines the relationship between source error and polarization, showing an inverse-U pattern. However, endogenous switching to biased sources can disrupt this pattern, influencing overall polarization dynamics.
This paper examines how sentencing range design influences sentencing decisions in the Czech legal system, leveraging a recent reform affecting theft and property offenses. Using differences-in-differences and regression discontinuity, I identify causal effects of sentencing ranges on outcomes. I find evidence of a severity effect, where harsher sentences result from placement in higher ranges, and a reference effect, where cases within the same range serve as benchmarks. These findings provide court-based evidence for phenomena previously studied only experimentally. The results contribute to the discussion on optimal sentencing range design by shedding light on the mechanisms shaping judicial decisions.
This paper is an extended version of my Master's thesis submitted to Charles University, Faculty of Law in September 2024 under the supervision of Michal Šoltés.
This paper addresses the challenge of heterogeneity in risk exposure when estimating the relative risk (RR) of causing road traffic crashes (RTCs) for different driver types. Quasi-induced exposure is a well-established alternative to direct data collection for exposure estimation. We investigate biases imposed on RR estimates caused by errors in fault assignment and unequal driver mix. Simulations reveal the directions and magnitudes of the possible biases and empirical tests of these biases are performed on a Czech dataset (1.2 million RTCs). Results show that errors in responsibility assignment work in opposite directions, with magnitude of bias depending on the size of the error and the target group proportion. Bias caused by unequal mixing depends on the target group proportion and the extent of the heterogeneity of not-at-fault drivers. Empirical tests confirm the discussed biases and underline their importance while interpreting RR estimates, so far mostly ignored by the literature.
Grant projects
New Generation of Traffic Accidents Statistics for Police CR
team member with Peter Bolcha (PI), Josef Montag and Matúš Šucha
01/2022 – 12/2023
Main responsibilities: Data analysis, development of a software tool computing the new statistics
