Shakked Noy


Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence (with Whitney Zhang)
Science, 2023, 381(6654), p.187-192
Online Appendix
Replication Package

Click to expand abstract We examined the productivity effects of a generative artificial intelligence 10 technology—the assistive chatbot ChatGPT—in the context of mid-level professional writing tasks. In a preregistered online experiment, we assigned occupation-specific, incentivized writing tasks to 453 college-educated professionals, and randomly exposed half of them to ChatGPT. Our results show that ChatGPT substantially raised productivity: average time taken decreased by 40% and output quality rose by 18%. Inequality between workers decreased, and concern and 15 excitement about AI temporarily rose. Workers exposed to ChatGPT during the experiment were 2x as likely to report using it in their real job two weeks after the experiment, and 1.6x as likely two months after the experiment.

The German Model of Industrial Relations: Balancing Flexibility and Collective Action (with Simon Jäger and Benjamin Schoefer)
Journal of Economic Perspectives, 2022, 36(4), p.53-80
Published Version (open-access)
Replication Package

Click to expand abstract We give an overview of the two pillars of the “German model” of industrial relations: sectoral collective bargaining and firm-level codetermination. Relative to the United States, Germany outsources collective bargaining to the sectoral level, resulting in higher coverage and the avoidance of firm-level distributional conflict. Relative to other European countries, Germany makes it easy for employers to avoid coverage or use flexibility provisions to deviate downwards from collective agreements. The greater flexibility of the German system may reduce unemployment, but may also erode bargaining coverage and increase inequality. Meanwhile, firm-level codetermination through worker board representation and works councils creates cooperative dialogue between employers and workers. Board representation has few direct impacts owing to worker representatives’ minority vote share, but works councils, which hold a range of substantive powers, may be more impactful. Overall, the German model highlights tensions between efficiency-enhancing flexibility and equity-enhancing collective action.

What Does Codetermination Do? (with Simon Jäger and Benjamin Schoefer)
ILR Review, 2022, 75(4), p.857-890
Summary for the Harvard Law School Forum on Corporate Governance
Replication Package

Click to expand abstract We provide a comprehensive overview of codetermination, i.e., worker representation in firms’ governance and management. The available micro evidence points to zero or small positive effects of codetermination on worker and firm outcomes, and leaves room for moderate positive effects on productivity, wages, and job stability. We also present new country-level, general-equilibrium event studies of codetermination reforms between the 1960s and 2010s, finding no effects on aggregate economic outcomes or the quality of industrial relations. We offer three explanations of the institution’s limited impact. First, existing codetermination laws convey little authority to workers. Second, countries with codetermination laws have high baseline levels of informal worker voice. Third, codetermination laws may interact with other labor market institutions, such as union representation and collective bargaining. We close by discussing implications for recent codetermination proposals in the United States.

The Effects of Neighborhood and Workplace Income Comparisons on Subjective Wellbeing (with Isabelle Sin)
Journal of Economic Behavior and Organization, 2021, 185, p.918-945
Master’s Thesis version

Click to expand abstract We investigate how a person’s happiness is affected by the incomes of her neighbors and coworkers. Using an unprecedentedly rich combination of administrative and survey data, we establish two central results. First, a person’s happiness is sensitive to her ordinal rank within her peer income distribution: people are happier the higher their income rank. Second, workplace rank matters much more than neighborhood rank. We confirm that our results reflect a causal effect of peer income by implementing sensitivity analyses, identifying off changes in peer income over time for immobile people, exploiting plausibly exogenous moves between workplaces triggered by mass layoffs, and testing for the effects of unobservable group-level confounders.

Policy and Other Research

Codetermination and Power in the Workplace (with Simon Jäger and Benjamin Schoefer)
Journal of Law and Political Economy, 2022, 3(1)
Economic Policy Institute Working Paper, prepared as part of the EPI’s Unequal Power Series

Click to expand abstract How does codetermination—entitling workers to participate in firm governance, either through membership on company boards or the formation of works councils—affect worker welfare and corporate decision making? We critically discuss the history and contemporary operation of European codetermination arrangements and review empirical evidence on their effects on firms and workers. Our review suggests that these arrangements are unlikely to significantly shift power in the workplace, but may mildly improve worker welfare and firm performance, in part by boosting information-sharing and cooperation and in part by slightly increasing worker influence.

People Judge Discrimination Against Women More Harshly Than Discrimination Against Men – Does Statistical Fairness Discrimination Explain Why? (with Jan Feld and Eberhard Feess)
Frontiers in Psychology, 2021, 12, p.3504
Replication Package

Click to expand abstract Previous research has shown that people care less about men than about women who are left behind. We show that this finding extends to the domain of labor market discrimination: In identical scenarios, people judge discrimination against women more morally bad than discrimination against men. This result holds in a representative sample of the US population and in a larger but not representative sample of Amazon Mechanical Turk (Mturk) respondents. We test if this gender gap is driven by statistical fairness discrimination, a process in which people use the gender of the victim to draw inferences about other characteristics which matter for their fairness judgments. We test this explanation with a survey experiment in which we explicitly hold information about the victim of discrimination constant. Our results provide only mixed support for the statistical fairness discrimination explanation. In our representative sample, we see no meaningful or significant effect of the information treatments. By contrast, in our Mturk sample, we see that providing additional information partly reduces the effect of the victim’s gender on judgment of the discriminator. While people may engage in statistical fairness discrimination, this process is unlikely to be an exhaustive explanation for why discrimination against women is judged as worse.

The Drivers of Mothers’ Parental Leave Decisions: Evidence from the Growing Up in New Zealand Survey (with Isabelle Sin)
Motu Working Paper 21-08, 2021
Coverage on Radio New Zealand

Click to expand abstract In this paper we compare mothers’ preferred leave, anticipated leave, and realised leave to shed light on how well different types of mothers are able to predict the parental leave they will take, and the factors that drive them to deviate from their plans. We use data from the Growing Up in New Zealand longitudinal survey on mothers’ preferred and anticipated leave reported antenatally, their realised leave, and the reasons they give for their leave-related choices to better understand the drivers of mothers’ leave decisions. We find mothers tend to anticipate substantially less leave than they prefer, but end up taking more leave on average than they anticipate. They have a moderate ability to take their preferred leave up to a year, but very little ability to take more than a year of leave. The 52 weeks of job-protected leave specified by law may play a role in this. Financial constraints are the most important factor driving mothers back to work. Certain types of mothers, such those with low income, are particularly prone to shocks that cause them to return to work earlier than anticipated, whereas first-time mothers who plan a longer period of leave are vulnerable to shocks that cause them to delay their return to work.

Involuntary Job Loss: Welfare Effects, Earnings Impacts, and Policy Options (with Dean Hyslop, Dave Maré, and Isabelle Sin)
Motu Working Paper 21-06, 2021

Click to expand abstract Workers who experience involuntary job loss suffer from deep and persistent negative consequences. In this paper, we first summarise the evidence on the effects of involuntary job loss on displaced workers’ wellbeing. We conclude that displacement harms workers’ mental health and economic security in the short term and negatively affects their earnings and mortality risk in the long term. We then extrapolate the estimates of Hyslop and Townsend (2017) to estimate the economy-wide net-present value of wages lost as a result of displacement by the workers displaced in New Zealand in a representative year. Our estimates suggest that this value is likely between $3.3 billion (in a year of economic upswing) and $15.4 billion (in a year of very severe economic downswing). Finally, we survey the policy options available for dealing with involuntary displacement. We conclude that unemployment insurance or unemployment benefits can effectively mitigate the immediate negative effects of displacement and have only small downsides. By contrast, training and job placement programs are typically ineffective, but in some circumstances might have high potential upside.

Explaining the Reliability of our Mathematical Beliefs
Philosophy Honors Thesis, 2019

Click to expand abstract The Benacerraf-Field Problem is the challenge, for mathematical realists, of explaining why our mathematical beliefs are reliable, given that we can’t interact with or observe mathematical entities. On one interpretation, the Benacerraf-Field Problem threatens the counterfactual sensitivity of our mathematical beliefs: the Problem suggests that, if the mathematical facts were different, our mathematical beliefs would not be correspondingly different. I argue that, on this interpretation, the Benacerraf-Field Problem can be overcome. If the mathematical facts were different, the physical world would have to be correspondingly different. For example, if the solution to a set of differential equations predicts the equilibrium behaviour of a physical system, then if the solution to those equations were different, the equilibrium behaviour of that system would have to be correspondingly different. Moreover, our mathematical beliefs are informed by the physical world, since our mathematical beliefs are influenced by our physical intuitions and because mathematics often develops in tandem with science. So if the mathematical facts were different, the physical world would have to be correspondingly different; and if the physical world were different, our mathematical beliefs would be correspondingly different. It follows that our mathematical beliefs are counterfactually sensitive, and the Benacerraf-Field Problem can be overcome.