Julia Tuppat¹; Tim Sawert²

¹ Universität Leipzig; ² Universität Mainz

The digital transformation affects key dimensions of social inequality, including health. Socioeconomic status (SES) remains a strong predictor of health outcomes and behaviors, with lower-SES individuals engaging less in health-promoting activities such as physical activity. Mobile health (mHealth) technologies — such as fitness apps and wearables — are often seen as tools to foster healthier lifestyles. However, their effects on health inequalities remain contested: while they may promote behavior change across groups, theories of the digital divide suggest that such innovations often benefit more advantaged populations to a greater extent.

This study draws on newly collected longitudinal data from the GESIS Panel (2023–2024) to examine whether mHealth use in the area of physical activity is socially stratified (digital divide level 2), whether it is associated with increased physical activity (behavioral effect), and whether this effect varies by SES (digital divide level 3). The analytical sample comprises over 12,000 observations from more than 5,000 respondents aged 18–70 in Germany.

Findings show a clear educational gradient in mHealth use, particularly among younger cohorts. Fixed effects regressions reveal that adopting mHealth technologies is associated with a significant increase in physical activity. However, this positive effect does not vary by SES: both high- and low-SES users benefit equally.

While mHealth technologies appear to support healthier behavior regardless of social position, their unequal uptake may still contribute to widening health disparities. Sociological research should further explore the structural and subjective barriers to mHealth adoption to inform inclusive digital health strategies.

Andri Rutschmann¹; Sascha Grehl²

¹ Department of Politics and Public Administration, Constance University, Germany; ² Institute for Sociology, Leipzig University, Germany

Social media platforms like X (formerly Twitter) play a central role in contemporary communication, not only facilitating interactions but also amplifying societal issues such as sexism. Given that harmful online behaviors can have real-world consequences, this study explores factors influencing individuals’ decisions to post sexist content. Specifically, we investigate how political orientation and social embeddedness mediate sexist behavior on social media.

Using over 5 million tweets from more than 50,000 unique German X accounts, we assigned users political orientation scores based on their following of politicians, influencers, and media outlets. We then classified their recent tweets as sexist or non-sexist using supervised machine learning techniques—including Lasso Linear Regression, Support Vector Machines, Random Forest, XGBoost, and a state-of-the-art large language model (LLM).

Our findings reveal three main insights: First, individuals with stronger right-wing orientations posted sexist tweets at higher rates, consistent with conservative ideological associations with traditional gender roles. Second, social embeddedness significantly deterred sexist posting; users with larger followings were less likely to share sexist content, likely due to reputational concerns. Finally, we found an interaction effect where the deterrent impact of having more followers was weaker for right-leaning users, suggesting reduced normative pressure within their networks.

This study contributes to understanding how political ideologies and social structures shape online behavior. By highlighting these dynamics, our findings provide practical insights for targeted interventions aimed at reducing sexist and harmful content online.

Barbara Binder; Nora Müller

GESIS — Leibniz Institute for the Social Sciences

This study examines whether cryptocurrency investors differ meaningfully from conventional asset holders or replicate familiar patterns of financial participation. Drawing on an online survey of adults in Germany (N = 1,207), we investigate the socio-demographic, attitudinal, and ideological determinants of cryptocurrency ownership. Using confirmatory factor analysis (CFA), we derive latent constructs for general crypto attitudes, antisemitic beliefs, conspiracy thinking, and institutional trust. These dimensions—alongside risk tolerance, political self-placement, views on social inequality, and IHS-transformed wealth proxies—are included in logistic regression models predicting crypto ownership.

Our results confirm findings on sociodemographic determinants of cryptocurrency ownership and expand the state of research on other key predictors, particularly regarding different values and ideological attitudes. Gender and risk tolerance are robust determinants of crypto ownership, even when taking into account a wide array of attitudes and ideological beliefs. Antisemitic attitudes, beliefs in conspiracy theories and a lack of trust in institutions are strong predictors of cryptocurrency ownership. Traditional wealth indicators, by contrast, show little explanatory power.

Crypto ownership appears to be shaped less by traditional economic resources and more by a combination of demographic factors, values and ideological attitudes. This reinforces the notion of cryptocurrencies as a socially meaningful and ideologically charged financial behavior — not just an investment strategy. This supports the idea of a heterogeneous investor base. Our findings also suggest that crypto assets capture dimensions of wealth not reflected in conventional asset categories and should be explicitly included in future wealth surveys.

Ansgar Hudde¹; Shannon Taflinger²

¹ University of Cologne; ² Department für Soziologie und Sozialpsychologie

Open-ended questions in quantitative surveys offer a cost-effective method to gather rich data from large, representative samples. While collecting such data is straightforward, analysis traditionally presents a dilemma: employ costly qualitative methods requiring extensive human coding, or use computational approaches that may only scratch the surface. Large Language Models (LLMs) provide a promising alternative that could potentially combine qualitative depth with computational efficiency.

We examine young Americans’ narratives about dating across party-political lines from a quota-representative survey of approximately 1,400 US Americans aged 20-32. This dataset provides an ideal test case for LLMs, requiring nuanced coding that captures subtle distinctions in attitudes, reasoning, and emotional responses.

Our research aims to: (1) Evaluate LLM accuracy in coding complex survey responses compared to human coders, (2) Assess whether misclassification patterns differ between LLMs and humans, (3) Determine if research conclusions differ based on LLMs versus human-coded data.

We find that advanced LLMs slightly outperform human coders and that research conclusions in an applied example are identical, whether we use a benchmark-dataset or datasets coded by either student assistants or advanced LLMs.

While our findings may interest diverse groups, our target audience is applied social science researchers who may have open-text data or are considering collecting such data but hesitate due to analysis costs. We particularly aim to inform qualitative researchers wanting to leverage larger samples and quantitative researchers recognizing the value of qualitative insights. Our guidance focuses on approaches implementable with standard computers and widespread technical and statistical skills.