Abstract The method of decreasing abstraction (MDA) has been proposed as a strategy for developing sociological models that balance parsimony and descriptive adequacy. This paper reconstructs Lindenberg’s version of MDA and situates it within broader debates on unrealistic assumptions and the tension between instrumentalism and realism. Lindenberg’s MDA can be viewed as a special case of a more general class of theory-development strategies. It combines explicit heuristic rules for the context of discovery with substantive auxiliary theories that guide the construction of more specialized models. Empirically, however, MDA has been taken up only selectively and rarely in the demanding form envisaged by Lindenberg. Drawing on structuralist reconstructions in the philosophy of science, we analyze hierarchically nested model variants as a general structural feature of research programs, arising from theory-building strategies akin to implicit forms of MDA. We further argue, first, that Lindenberg’s own version of MDA, with its strong commitments to specific auxiliary theories, is too narrow to serve as a general method, and second, that it may often be more promising to use MDA as a strategy that begins with careful reconstructions of existing research programs. We conclude with reflections on the implications for theoretical progress and on the role and balance of theoretical quality criteria such as parsimony and coherence with background knowledge when choosing between competing models.

Zusammenfassung Der Beitrag untersucht, ob und wie stark die Lebensführung sozialer Gruppen mit ihren Einstellungen zu zentralen politischen Konfliktthemen zusammenhängt. In den sozialwissenschaftlichen Diskussionen zu neuen gesellschaftlichen Spaltungslinien wird postuliert, dass in Konflikten um Migration, Klimaschutz oder sexuelle Diversität gesellschaftliche Gruppen mit grundverschiedenen Identitäten, Werten und Lebensweisen aufeinanderprallen. Obwohl derlei Konzepte theoretisch bedeutsam sind, wurden sie in der Polarisierungsforschung bisher kaum empirisch berücksichtigt. Auf Basis aktueller deutscher Umfragedaten und einer etablierten Lebensführungstypologie zeigen unsere Analysen, dass zwischen Lebensführungstypen in der Tat substanzielle Einstellungsunterschiede bestehen, ohne dass diese Unterschiede das Ausmaß einer fundamentalen Opposition erreichen. Modernisierte Gruppen, deren Lebensführung sich durch hohe biografische Offenheit auszeichnet, zeigen tendenziell progressivere Einstellungen zu Themen der Migration, sexuellen Diversität und des Klimaschutzes als traditionalistisch orientierte Gruppen. Bei Migrations- und Klimafragen spielt zudem das Ausstattungsniveau der Lebensführung eine Rolle: Progressive Haltungen treten hier verstärkt bei jenen auf, deren Lebensführung von einem gehobenen Niveau an ökonomischem und kulturellem Kapital geprägt ist. Bei Einstellungen zu materieller Umverteilung erweisen sich Unterschiede in der Lebensführung hingegen als kaum relevant. Die Lebensführung ist damit für jene Themen bedeutsam, die in der öffentlichen Diskussion als „Kulturkämpfe“ rubriziert werden. Multivariate Analysen weisen darauf hin, dass Lebensführungsunterschiede einen eigenständigen Bedingungsfaktor der gesellschaftspolitischen Einstellungen darstellen, der von ähnlicher Relevanz ist wie Klassen- und Bildungsunterschiede.

Abstract The paper exemplifies a contrastive research design for a comprehensive comparison of related middle-range theories. Six adaptations of Hirschman’s exit, voice, loyalty (EVL) model for authoritarian states are formalized using directed acyclic graphs (DAGs). From the DAGs, a set of 112 pivotal observable implications is derived. These implications are evaluated using new data from the German Democratic Republic (GDR), covering the period from 1970 to 1989. The evaluation employs nonparametric linear kernel regression. Results indicate that none of the six EVL models can fully explain the EVL dynamics in the GDR. The paper discusses how the proposed research design complements X‑centered designs and structural equation models.

Abstract Identifying the causes of social inequality is crucial for designing effective policy interventions. Experimental methods, in which variables are manipulated and participants are randomly assigned to conditions, are widely regarded as the gold standard for causal inference. However, it seems impossible to randomly assign social categories such as gender or race. Does that mean that, for example, gender cannot cause outcomes? We propose a conceptually grounded strategy for the causal treatment of such complex social categories. Gender should be viewed as a multidimensional construct that shapes inequality through different mechanisms. To understand these mechanisms, we can decompose gender into its components (such as perceived social roles or personality traits) and study how each one affects outcomes such as hiring decisions. By manipulating these components in experiments, researchers can identify which aspects of gender actually produce unequal treatment and why. Although the approach itself is not novel per se, its implications are, in our opinion, often insufficiently discussed or theoretically integrated in studies that advance causal claims. We illustrate the utility of this approach through studies of gendered hiring discrimination, showing how gender can be theoretically decomposed into distinct components, which in turn can be operationalized in experimental designs to uncover underlying mechanisms. Beyond gender, the approach applies to other seemingly nonmanipulable characteristics such as race and class, providing a generalizable strategy for mechanism-based causal inference and more targeted policy interventions aimed at reducing social inequality.

Abstract Directed acyclic graphs (DAGs) have become popular as graphical representations of causal relationships. In practice, DAGs have proven to be particularly helpful for selecting appropriate control variables in causally oriented analytical models. While confirming such usefulness, this paper also aims to highlight another, often neglected, aspect: the potential for causal diagrams to support the formulation of theories and corresponding hypotheses. This is particularly the case when diagrams have certain graphical properties, and suggestions are offered regarding how this can be achieved. Examples are drawn from the field of life-course research, with the intention of better integrating the visual techniques prevalent in life-course research with DAG-style causal diagrams. While standard causal diagrams may not pay sufficient attention to certain relevant aspects, graphically enhanced causal diagrams can be quite productive for theory development and the analysis of existing life-course data. They are also useful for conceptualising new causally oriented studies. This paper illustrates suitable approaches with original and adapted visualisations.

Abstract Causal explanation is central to sociological research, shaping both theoretical development and empirical inquiry. This paper argues that causal machine learning—which integrates deductive identification strategies with inductive estimation techniques—offers an analytical approach for modeling complex, nonlinear social processes within the potential outcomes framework. We argue that causal machine learning operates through an iterative feedback loop: Theoretical assumptions guide flexible estimation, which inductively uncovers complex heterogeneities and nonlinearities, and these discoveries subsequently refine and expand sociological knowledge. Drawing on a systematic review of recent sociological research (2014–2024), we highlight how causal machine learning is advancing work in three key areas: causal effect heterogeneity, causal mediation analysis, and time-varying causal inference. These developments expand the methodological tool kit available to sociologists and strengthen the discipline’s ability to test, refine, and extend theories of social explanation. We conclude by outlining emerging directions, including high-dimensional causal inference and generative artificial intelligence, that are opening new methodological frontiers in causal machine learning for sociology.

Abstract In recent years, digital behavioral data (DBD) have emerged as a powerful resource in social science research. Their ubiquity, granularity, complexity, and continuous collection provide new opportunities for examining social processes in great detail. However, because DBD are diverse in type and often constitute found data—not generated for research purposes—their potential for causal analysis is commonly underestimated. To address this issue, this paper outlines key considerations for developing a methodological framework for valid causal inference using DBD. The discussion focuses on how design limitations can be (i) ruled out a priori when generating designed DBD or (ii) compensated through theoretical and temporal information, the specification of structural causal models, a posteriori design considerations, and the application of appropriate analytical tools, making found DBD fit for the purpose of causal effect estimation.