Context #3 - Evans

Definition: uses new tools and data sources to expand the scope and scale of sociological inquiry; expands the sociological imagination.
Data windfall: internet as a massive digital observatory; sensors in cell phones and social media; born-digital observations plus digitization (e.g., Google Books) enable unprecedented scales.
Big data for free: computation provides lenses to see and analyze large-scale social data; can reveal patterns that challenge or require new theories.
Platform for active inquiry: web-scale surveys and experiments to test theories and causal relationships; online investigations surpass offline in scale, duration, and complexity.
Not a passing fad: computational sociology is a reactor that can transform the study of many problems (e.g., polarization, segregation, diffusion of ideas).
Methods and training: extends familiar methods and demands new skills; social researchers will need to write code, wrangle data, and think computationally as well as sociologically.

Traditional social theory uses natural language and rich nuance, which can invite ambiguity and overgeneralization.
Computational sociologists often express theory in mathematics or algorithms; formal models can be empirically validated or explored via simulation.
Example: a simple formal model of how scientists choose phenomena to study; using data from millions of articles/patents showed unfolding knowledge is often generated by a conservative strategy linking popular phenomena to nearby less prominent ones.
Simulation on a Cray revealed more efficient strategies could have accelerated discovery; verbal theory is essential but not sufficient for exploration.
Takeaway: verbal theory inspires formalizations, but combining both enables deeper exploration.

Ethnography surfaces novel practices and rich, nuanced detail for theory building; tests ideas against lived experience.
Digital traces offer massive-scale data but are often low-dimensional “cartoons” of social life (categories, clusters, scores).
Digital cartoons can reveal novel behaviors (e.g., online dating biases, cross-race messaging effects) and allow testing ethnographic insight at scale (e.g., police officers’ respectful behavior across stops).
Limitations: they cannot capture full multimodal nuance or sequence/tone; collaboration with ethnographers remains essential.
Ethnographers should guide trace collection and construct development; qualitative data can justify assumptions used by computational methods.

Traditional surveys underpin quantitative sociology: designed sampling, randomization, and generalizable findings.
Computational approaches predict attitudes/behaviors from digital breadcrumbs; machine learning imputes signals (e.g., demographics from images/text) and online traces can help select respondents.
Simple surveys mimic real-world platforms (e.g., rating/selection tasks) via crowdsourcing (Mechanical Turk, CrowdFlower) or marketplaces; online responses can reach larger samples with fewer questions.
Advantages: broader reach, faster data collection, responses reflect actual online behavior rather than stated beliefs.
Caveats: many online traces are collected for commercial purposes; potential algorithmic confounding; consent/ethics/privacy concerns; interpreting data requires understanding the predictive models that generated actions.

Laboratory/field experiments test causal claims and theory in controlled settings; online experiments extend reach to scales and designs impractical offline.
Classic field experiments (e.g., Pager) demonstrate causal effects (e.g., racial discrimination in hiring).
Online experiments offer scalability and complexity (e.g., Salganik, Dodds, Watts): multiple “worlds” show how social information affects market outcomes.
Limitations: online modalities constrain sensory input; participant environments uncontrolled; sample or ecological validity concerns; some designs on pre-existing networks require advanced randomization.
Future strength: higher bandwidth devices, augmented/virtual reality, and AI agents enabling more sophisticated, ecologically valid experiments; ethical considerations must scale with capability.

Statistical inference underpins traditional social science: estimating causal/associational relationships.
Machine learning emphasizes prediction over parametric estimation; supervised learning maps features to labels; unsupervised learning discovers structure (clusters, dimensionality reduction).
ML benefits: handles diverse data types; can yield surprising findings (e.g., migration clustering, “fuzzy boundaries” of ethnic conflict); large feature spaces (e.g., 12,000+ features) may offer limited gains for certain predictions, highlighting the need for theory.
Deep neural networks offer strong predictive power but can be hard to interpret and may lack causal clarity for policy relevance.
Sociology has a special vocation: to integrate ML insights with causal explanation and theory testing.
Curriculum implications: teach languages like Python, R, Julia alongside Stata; boot camps for data gathering/cleaning; introduce discrete math and linear algebra early; prioritize collaborative work with ML experts.

Sociology should maintain its core mission while embracing computational methods.
A broadened imagination links micro and macro scales, momentary and historical perspectives, and new digital experiences.
Departments should adapt training and culture to include data science fluency, collaboration, and access to modern computational tools.
Computation enables modeling, measuring, and modifying social structures and experiences across scales.
The sociological imagination has never been more expansive; the digital age demands new imagination and applied ethics.

Mills’ idea of connecting individual experience to large-scale social currents remains central; computation expands that capacity across scales and contexts.
Sociologists have a unique role in predicting, explaining, and interpreting social data, with a blend of qualitative and quantitative habits.
Collaboration with machine learning and data-science experts will be essential; avoid treating ML as a black box.
Training should frontload practical skills (coding, data handling) and conceptual grounding in theory and methods.
The field envisions formal worlds, digital observatories, intelligent surveys, virtual laboratories, and machine-driven discovery—requiring an empowered, imaginative sociological imagination.