Niklas M. Witzig

profile.png

I am a PhD student at the Johannes Gutenberg University in Mainz, Germany at the chair of Daniel Schunk.

I mostly work in Behavioral Economics with a strong interest in how computational and machine learning tools can benefit (behavioral) economic analysis and how (behavioral) economics can inform current debates in discussions on fair and explainable AI.

   

Ongoing Projects

 

Cognitive Imprecision and Altruistic Preferences

[draft in preparation]

I study altruistic choices --- trading-off own vs. another persons's payoff --- through the lens of a cognitively noisy Bayesian decision-maker. I propose a simple choice framework where cognitive noise can bias altruistic decisions. I run an experiment featuring a between-subject manipulation of cognitive noise, where the altruistic trade-off is cognitively more difficult. The treatment group shows a flatter association between changes in payoffs and choices and decides significantly more often for the other person's payment, i.e., is more altruistic. I explore the origins of this effect both with Bayesian hierarchical modelling and a number-comparison task, mirroring the "arithmetics" of the altruism choices absent any preference dimension. I find comparable treatment effects, which hints at an adaption of the perception of numerical magnitudes to the statistics of the experiment trials as the driver of the treatment effect. The structural estimations support this interpretation. I further explore the implications of a "cognitive lens" to altruism decisions and, e.g., find positive associations between measures of cognitive ability and choice variability.

Trust in Fair Algorithms

with Mattia Cerrato, Marius Köppel and Alesia Vallenas
[pilot completed]

We study how humans incorporate advice by an algorithm in a setting where gender biases are known to play a large role: hiring decisions and labor market outcomes. In an online experiment, we investigate if humans, when tasked to predict e.g., the employment status of another person, rely more strongly on predictions by *fair* algorithms, i.e., whose accuracy is the same regardless of gender. Crucially, we omit the sensitive attribute of the person at hand for both the human and machine. We furthermore investigate the role of prior beliefs of group differences and are able to capture the optimal Bayesian action and can compare if human behavior favors fair predictions beyond what is statistically optimal.

Strategic Risk-Taking and Time Pressure in Professional Chess

with Johannes Carow

[draft available upon request]

We study the impact of time pressure on strategic risk-taking of professional chess players. We propose a novel machine-learning-based measure for the degree of strategic risk of a single chess move and apply this measure to games of the 2013-2021 FIDE Chess World Cups. Crucially, this tournament allows for an arguably exogenous variation in available thinking time to a chess player. We find that time pressure consistently leads chess players to opt for risk-averse moves more frequently. In addition, we find correlational evidence for strategic loss aversion, a tendency to play more risky moves after a mistake or when in a disadvantageous position. Our results suggest that high-proficiency decision- makers in high stake settings react to context factors such as time pressure.

Behavioral Time Choices in Speed-Accuracy Trade-offs

with Alexander Dzionara
[pre-registration] [draft available upon request]

In many economic contexts, people need to solve trade-offs between doing an activity (e.g., solving a task at work) faster and doing it better. While time choices in speed-accuracy trade-offs have been extensively studied in cognitive science for motor-response and perception tasks, little evidence is available for economic decision-making. Furthermore, the impact of behavioral biases on time choices in such environments has yet to be explored. In this paper, we present a theoretical model linking time choices in speed-accuracy trade-offs to agents' abilities, subjective beliefs, and uncertainty attitudes. We test the predictions of the model in an experiment for two distinct, but mathematically identical, environments: prospective time choices before solving a task and simultaneous time choices while solving a task. We find that the behavioral model better captures time choices in the prospective but not in the simultaneous environment, where a rational model is more performant. This is in line with a literature in psychology and economics documenting differences between determinants of planned and actual actions.