Skip to content

Can marriage increase gender equality by estimating the causal effect of marriage vs cohabitation on labour market trajectories of new parents.

Traditionally, a "marriage surplus" was created through specialization of household activities, but in modern times gains from a more egalitarian marriage can be through increased coordination.

We ask for the first time whether marriage can increase gender equality by estimating the causal effect of marriage vs cohabitation on labour market trajectories of new parents. Applying a Marginal Treatment Effects framework, the average treatment effect of marriage is consistent with specialization - marriage causes women to work less and men more. This average effect hides treatment effect heterogeneity across unobservables, whereby the couples "more resistant" to marry - i.e. the more modern couples, exhibit coordination of labour market activities.

There is no longer a marriage penalty to women and the coordinating men earn less if married than if cohabiting. Given this, we ask whether increased gender equality for the coordinators lowers or raises household welfare, finding no effect of marriage on children for specializers or coordinators, and a reduction in separation from marriage for coordinators - suggesting that moving away from masculine male breadwinner norms can improve relationship contentedness.

This seminar is about understanding how individuals form expectations about future wages.

Understanding how individuals form expectations about future wages and how these prospects evolve over time is crucial for assessing the acceptability of existing wage disparities. Providing a novel metric of wage prospects, we propose using probit regressions to assess individual ex-ante expectations of earning a wage from a specific quintile of the wage distribution and relate these estimates to realized (i.e., ex-post) wage outcomes.

Utilizing a large dataset of almost 250,000 observations from the German Socio-Economic Panel, covering the period from 1992 to 2020, our study reveals strong segmentations in the German labor market by gender and region, with women and East German workers more likely to earn lower wages. The 2003-2005 labor market reforms increased the probability of earning lower wages overall but slightly reduced gender and regional segmentation. For the bottom income quintiles, the reforms worsened wage prospects, particularly affecting part-time workers, minijobbers, and workers in West Germany.

Conversely, higher income earners saw improved wage prospects post-reforms. Finally, the new measure of wage prospects is predictive of actual wage transitions and exhibits theory-conform correlation with life satisfaction, which underscores its importance for understanding individual fairness considerations.

This paper quantifies the share of dismissals distorted by conflict and identifies the drivers.

Dismissal costs are shaped by firm and worker behavior. While they might coordinate to minimize costs, adversarial separations may also entail cost-seeking actions ("conflict").

This paper quantifies the share of dismissals distorted by conflict and identifies the drivers. Our strategy exploits the choice between two modes of separation in France: personal dismissals and ''separations by mutual agreement'' (SMAs). Since SMAs waive dismissal red tape costs and enable severance pay bargaining, they should always be preferred to dismissals in an efficient bargaining model. In contrast, we find that only 12% of potential dismissals are resolved through SMAs. We then identify the sources of conflict that lead to the choice of the costlier separation mode in 88% of dismissals.

Our survey of HR directors reveals three crucial drivers, which account for 63% of the failures to convert dismissals into SMAs: (i) hostility between the employer and the employee, (ii) employers using dismissals as a discipline device, and (iii) asymmetric beliefs about labor court outcomes following a dismissal.

Advancements in Artificial Intelligence (AI) such as Large Language Models (LLM) promise to open unprecedented possibilities in applications with the potential to fundamentally change social research methods. Simultaneously, its development is currently dominated by venture-backed 'BigTech' companies, and it's once again a winner-take-all race. The scientific community is mostly only consumer and not driver of changes, especially from the domains of social science.
These developments do not go without critique. A key concern with AI for social scientists is how to evade methodological black boxes and resist the illusion of Emily Bender's stochastic parrots with their potential risks of hidden biases. The dangers of systems that are not understood sufficiently and that potentially breach privacy, data protection and research ethics principles are crucial.

Some scholars suggest, therefore, that this might be avoidable by gaining a profound understanding of the AI systems' training models and data. However, our understanding of LLM and their training data, for instance, is often limited by proprietary information and business secrets. Hence, we can almost only trust companies with what happens to our data.
Questions on the effectiveness of new laws on the digital frontier (e.g. EU AI Act) to make the principles within these programs understandable by obliging transparency arise, as well as the difficulties of grasping (arithmetic) biases and inequalities, despite their well-documented existence.

Discussing awareness of (arithmetic) biases questions knowledge and resources. Who can really understand LLM? Where do sociologists find resources and time for such an endeavour on top of their research? Should sociology trust computer science with this issue to bridge epistemological differences with another discipline?

This presentation will cover potentials and challenges of AI in societal analysis, especially with qualitative methods, questions of trust and knowledge and will explore possible solutions for feasible sociological AI application.

This study is about the return intentions of Ukrainians and how these intentions may depend on the feeling of national identity and pride.

We conducted a poll of 1139 Ukrainians who currently live in Poland. 937 of them took refuge after the start of the full-scale russian invasion on Feb. 24, 2022, and the other 202 migrated earlier. Our major focus was on the return intentions of Ukrainians and how these intentions may depend on the feeling of national identity and pride. In the pre-registered hypotheses, we stated that a stronger national identity was positively associated with the willingness to return home and that we could amplify this willingness by making the identity more salient.

In the survey, we randomly exposed individuals to three priming settings. Two of them primed subjects towards enhancing identity and pride feelings, while the third group contained neutral questions for a control group. Then, we measured the key variables of interest (return intentions) and found strong support for the first hypothesis and unexpected results for the other.

One of the most unexpected findings is the negative average effect of “pride priming” on return intentions of forced displaced persons, with a positive gradient along the levels of national pride. In fact, people with initially low levels of pride express strictly negative return intentions, whereas people with high pride are more likely to return. At the same time, pre-war migrants have not been affected by our priming experiment, which suggests more stable staying preferences in this group.

This study is about moving an establishment survey from telephone administration to online administration.

The European Company Survey (ECS) 2019 – commissioned by two European Agencies, Eurofound and Cedefop, and carried out by Ipsos – was the first large-scale, cross-national survey of establishments to use a push-to-web approach. Establishments across all EU Member States were contacted via telephone to identify a management respondent, and, where possible, an employee representative respondent. Respondents are then asked to fill out the 20-25 minute survey questionnaire online. The questionnaire captured a wide range of practices and strategies implemented by European companies in terms of work organisation, human resource management, skills use and skills development, and employee voice. Fieldwork for the survey took place in the first half of 2019, in all EU Member States.

Around a quarter of respondents to the ECS 2019 consented to being re-contacted for follow-up research. In November 2020 Eurofound and Cedefop approached these respondents, inviting them to complete a 10-15-minute follow-up questionnaire on the impact of COVID-19 on workplace practices.
In my presentation I will discuss the survey design, fieldwork outcomes, and the lessons we have drawn from conducting the ECS 2019 and the ECS 2020 follow-up – including the results from an experiment we ran as part of the ECS 2020 follow-up with offering customised reports to entice survey respondents. I will also briefly reflect on our plans for the next ECS which is scheduled for 2028.

This paper estimates the medium-run effect of duration of residence in reception centers of asylum seekers.

After arrival, asylum seekers are often housed in reception centers. The type, quality and duration of stay in such centers varies considerably across or within countries. In the context of the so-called “EU refugee crisis” in 2014-2016, reports emerged that some asylum seekers remained in reception centers for several years due to limited capacity of municipalities, lengthy asylum procedures and tight housing markets. It is often argued that reception centers have a detrimental effect on integration processes of asylum seekers and refugees, yet empirical, inferential evidence is still lacking. This paper estimates the medium-run effect of duration of residence in reception centers on language skills, contacts to the host population, and employment status. We use high-quality panel data on refugees living in Germany and apply inverse-probability-weighting (IPW). The results suggest that a quick transition from reception centers into private housing modestly increases refugees’ interactions with the host population and their language proficiency. We find no effects on labor market participation. Using additional analyses, we find that moving into private housing is often associated with a shift to more precarious neighborhoods, potentially hindering a stronger realization of the benefits linked to independent living in general.

This study is about the influence of family members, neighbors and coworkers on retirement behavior.

We study the influence of family members, neighbors and coworkers on retirement behavior. To estimate causal retirement spillovers between individuals, we exploit a pension reform in the Netherlands that creates exogenous variation in peers' retirement ages, and we use administrative data on the full Dutch population.

We find large spillovers in couples, primarily due to women reacting to their husband's retirement choices. Consistent with homophily in social interactions, the influence of the average sibling, neighbor and coworker is modest, but sizable spillovers emerge between similar individuals in these groups.

Additional evidence suggests both leisure complementarities and the transmission of social norms as mechanisms behind retirement spillovers. Our findings imply that pension reforms have a large social multiplier, amplifying their overall impact on retirement behavior by 40%.

The principal effects on the probability of engagement in the criminal justice system are much larger for Black than for non-Black males.

Using rich Texas administrative data, we estimate the impact of middle school principals on post-secondary schooling, employment, and criminal justice outcomes. The results highlight the importance of school leadership, though striking differences emerge in the relative importance of different skill dimensions to different outcomes. The estimates reveal large and highly significant effects of principal value-added to cognitive skills on the productive activities of schooling and work but much weaker effects of value-added to noncognitive skills on these outcomes.

In contrast, there is little or no evidence that middle school principals affect the probability a male is arrested and has a guilty disposition by raising cognitive skills but strong evidence that they affect these outcomes through their impacts on noncognitive skills, especially those related to the probability of an out-of-school suspension. In addition, the principal effects on the probability of engagement in the criminal justice system are much larger for Black than for non-Black males, corresponding to race differences in engagement with the criminal justice system.

Stressing strict privacy policies and changing the location of the survey URL have no response-enhancing effect.

Researchers collect data in most experiments not all at once but sequentially over a period of time. This allows to observe outcomes early and to adapt the treatment assignment to reduce the costs of inferior treatments. This talk discusses multi-armed-bandit-type adaptive experimental designs and algorithms for balancing exploration of treatment effects and exploitation of better treatments. By design, bandits break usual asymptotic and make inference difficult. We show how a batched bandit design allows for valid confidence intervals and compare coverage of the batched bandit estimator in Monte Carlo simulations. In a real-world application, we investigate elements of a survey invitation message targeted to businesses. We implement a full factorial experiment with five elements adaptively.

Our results indicate that personalizing the message, emphasizing the authority of the sender, and pleading for help increase survey starting rates, while stressing strict privacy policies and changing the location of the survey URL have no response-enhancing effect. As a tool for researchers, we introduce bandits in Stata, which facilitates running Monte Carlo simulations to assist the design and implementation of experiments before data are collected, interactively running own bandit experiments, and analyzing adaptively collected data. Bandits implement three popular treatment assignment algorithms: ε-first, ε-greedy, and Thompson sampling. Bandits facilitates estimation, inference, and visualization.