The social sciences are facing a scientific identity crisis that increasingly resembles a combat sport. Researchers compete over theories, methods, datasets and interpretations, while public debates often reward the most confident claim rather than the most reliable one. A provocative new analysis of the field argues that the central problem is not simply disagreement between disciplines. It is the difficulty of studying human societies when the subjects being studied can change their behavior, reinterpret their histories and react to the research itself. From the “Markovian temptation” of treating social life as a predictable sequence of states to the “scalar indifference” that hides crucial differences between individuals and institutions, the analysis identifies several obstacles that make social science fundamentally different from laboratory science. It also examines what it calls the historian’s obsession with the past, warning that historical explanation can become so focused on origins that it loses sight of present mechanisms.
The comparison with combat sports reflects the unusually adversarial nature of scientific disputes about society. In physics or chemistry, competing explanations can often be tested under controlled conditions, with the same variables measured repeatedly. Social researchers rarely enjoy that stability. A policy can alter the behavior it was designed to measure, an election can change public opinion while it is being studied, and a published finding can influence institutions, markets and voters. The social world is therefore both an object of observation and a participant in the production of evidence. This reflexivity creates a difficult technical problem: researchers must distinguish patterns that arise from durable mechanisms from patterns produced by temporary institutions, measurement choices or the expectations of the people involved. A correlation between education and political participation, for example, may reflect income, social networks, civic norms, access to information or selection effects rather than education alone.
One of the most seductive approaches to this problem is the Markov model, a mathematical framework in which the probability of a future state depends primarily on the current state. In a simplified social example, a researcher might estimate the probability that an unemployed person becomes employed next month based on their present employment status, age, education and other observed characteristics. Markovian methods are powerful because they convert complex change into a sequence of transitions that can be calculated, simulated and compared. They are widely used in population studies, epidemiology, economics, political science and machine learning. The danger appears when the model’s convenience is mistaken for a property of reality. Human trajectories often depend on unobserved memories, institutional histories, expectations and cumulative advantages. Two people who occupy the same measured state today may have radically different futures because they arrived there through different pathways.
This is the Markovian temptation: the belief that a sufficiently detailed snapshot can explain what happens next. In technical terms, the assumption of conditional independence may be violated when the past continues to influence outcomes after the variables included in a model have been controlled. A person’s previous job loss, for instance, may affect employer perceptions, mental health and professional networks even if current income and education are identical to those of someone who has never been unemployed. Statistical models can partially address this problem by adding lagged variables, random effects, hidden-state structures or sequence analysis. Yet every additional variable creates new demands for data and new opportunities for error. The more researchers try to reconstruct the past mathematically, the more they confront missing records, inconsistent definitions and feedback loops. Social prediction is therefore not merely a matter of choosing a better algorithm; it requires deciding which histories are causally relevant and which are artifacts of measurement.
A second obstacle is scalar indifference, the tendency to treat phenomena as though their meaning remains constant when the scale of analysis changes. A pattern observed among individuals may disappear among households, neighborhoods or nations, a problem associated with ecological fallacies and the modifiable areal unit problem. Conversely, an institutional pattern may be invisible at the individual level. A city can experience rising employment while many residents remain excluded from the sectors generating growth. A school system can improve its average test scores while widening the gap between students at the top and bottom. These contradictions are not statistical nuisances; they reveal that social processes operate simultaneously across levels. Multilevel models, hierarchical Bayesian methods and network analysis can help separate individual effects from contextual effects, but they cannot eliminate the theoretical question of which scale matters for the claim being made.
The issue becomes especially urgent when scientific findings are translated into headlines. A small effect averaged across millions of people may be politically important, while a dramatic effect in a narrow population may have little relevance beyond that setting. Relative risk can make a modest difference appear enormous, whereas absolute risk may show that the practical consequence is limited. Similarly, national averages can conceal regional inequalities, and global indicators can flatten cultural differences. Social scientists must therefore report uncertainty not only around numerical estimates but also around the level at which those estimates are valid. A model predicting voting behavior in one district cannot automatically explain a country, just as an experiment involving university students cannot be assumed to describe an entire population. Generalization requires explicit evidence, not rhetorical confidence.
The historian’s obsession with the past introduces a different, though related, danger. Historical research is indispensable because institutions, identities and inequalities are often path-dependent: earlier decisions constrain later possibilities. The design of a welfare system, the borders of a state or the legacy of colonial administration may shape behavior long after the original actors have disappeared. Yet an explanation centered too heavily on origins can become deterministic. Knowing how a structure began does not by itself reveal how it operates now, whether it has been modified by later events or which mechanisms could change it. Historians and social scientists increasingly address this challenge through process tracing, counterfactual reasoning and comparative historical analysis. These methods ask not only what happened, but what alternative sequences were possible and which links in the chain were necessary, contingent or merely coincidental.
The emerging solution is not to force the social sciences into the mold of the natural sciences, nor to abandon scientific standards because human behavior is complicated. It is to combine methods while making their assumptions visible. Longitudinal surveys can track individuals over time; administrative records can reveal institutional patterns; interviews can expose meanings that numerical indicators miss; experiments can test causal mechanisms under controlled conditions; and computational models can explore how local decisions generate large-scale outcomes. Triangulation is most useful when methods are not simply placed side by side but used to challenge one another. A statistical association can suggest a mechanism, ethnographic evidence can clarify how that mechanism is experienced, and historical research can show whether it persists across political regimes. The goal is not a single final model of society, but a disciplined account of what each method can and cannot establish.
Seen this way, social science is “combative” not because disagreement signals failure, but because disagreement is built into the object of study. The challenge is to turn intellectual conflict into cumulative knowledge rather than methodological tribalism. Researchers need to resist the seduction of elegant models, the false comfort of averages and the assumption that the past dictates the future. They must also recognize that prediction, explanation and interpretation are different scientific tasks. A model may forecast a transition without explaining its cause; a historical narrative may identify a causal sequence without predicting its recurrence; and a survey may describe attitudes without revealing how those attitudes were formed. The most credible social science will be the work that states these limits openly, tests claims across scales and time periods, and treats uncertainty as information rather than weakness. In a world increasingly governed by data-driven decisions, that intellectual discipline may be the field’s most important defense against both sensationalism and error.
Subject of Research: Obstacles to the scientific study of society, including Markovian modeling, scale-dependent analysis, causal inference and historical explanation.
Article Title: Are the Social Sciences Combat Sports? Reflections on the Obstacles to the Scientific Study of the Social
Image Credits: AI Generated
Keywords: Social sciences, scientific method, Markov models, causal inference, multilevel analysis, historical analysis, social data, predictive modeling, research methodology
