Does it always require AI?

No. Descriptive statistics, qualitative work and well-designed surveys can be central; AI is one possible tool.

Is it the same as computational social science?

The fields overlap substantially, and terminology varies across institutions. Neither has a single mandatory toolkit.

Can public online data be used without considering people?

Public visibility does not eliminate contextual, ethical or disclosure concerns. The intended use and possible effects still need examination.

AI Glossary / Data science and analytics / Intermediate

Social data science

Definition

Social data science combines computational and statistical methods with social-science theory and research design to study people, relationships, institutions and collective behavior. It asks both what patterns data reveal and how those data were produced, measured and connected to the social question being investigated.

What it means

The field is broader than analyzing social-media posts. Its evidence can include surveys, administrative records, experiments, interviews, digital activity and networks. Its questions might concern participation, inequality, communication, work or the effects of technology.

Data science supplies tools for handling complex information, but a technically accurate calculation can still measure the wrong concept. Counting messages, for example, is not automatically a valid measurement of friendship, collaboration or influence.

Social data science therefore connects measurement and interpretation. It considers whose behavior is visible, which actions a platform encourages, what important experiences are missing and whether a result describes an association, predicts an outcome or supports a causal explanation. These are different research aims.

How it works

A project begins with a question and a definition of the social concept involved. Researchers translate that concept into observable measures, explaining the gap between the concept and each indicator. They then select a study design and data sources suited to the intended claim.

Computational methods can include text analysis, network analysis, simulation and machine learning. Surveys, interviews and experiments can complement those methods by testing interpretations or measuring information absent from digital traces.

The analysis needs to account for sampling, missingness, dependence between people and changes in data collection. A platform redesign can change recorded activity even when the underlying social behavior changes little.

Researchers assess alternative explanations and communicate uncertainty. Predicting which users leave a service is not, by itself, evidence that changing a correlated feature will retain them. Ethical decisions about collection, representation and possible consequences belong throughout the project.

Three practical examples

A learning organization combines platform activity with participant interviews to investigate barriers to course completion, distinguishing access problems from lack of interest.
A research team studies how information travels through a community using interaction networks, while checking what a recorded link actually represents.
A business examines adoption of an AI assistant across departments, connecting usage records with work roles and survey evidence instead of assuming low usage means resistance.

Think of it this way

Think of studying a city through several maps: transport routes, housing, population and residents' accounts. Each reveals something, but none is the city itself. Combining them requires knowing what each map measures and leaves out.

The analogy has limits because people respond to measurement and policy. A dashboard can influence behavior, and a category can affect how people are treated. Social data are produced within social relationships rather than being neutral marks waiting to be discovered.

Worked example

Separating activity from participation

Imagine a workshop with 100 enrolled people. Ten participants each post 20 messages, while another 40 each post one message. Fifty people post nothing. The platform records 240 messages from 50 distinct contributors.

A message-based summary gives the ten highly active participants 200/240, or about 83.3%, of the observed activity. A person-based summary says that 50% of enrolled people posted at least once. Neither statistic establishes how many people felt included or learned something.

A social data science study would define the intended concept and gather suitable additional evidence. Quiet participants may read attentively, lack access, prefer another channel or feel excluded. Those possibilities cannot be distinguished from the message count alone. Interviews or a carefully designed survey can help interpret the traces, with their own response and measurement limitations.

Illustrative case study

Fictional case: Willow Learning sees that users who interact with an AI tutor complete more lessons. A dashboard labels this an effect of the tutor.

The research team points out that motivated learners may both use the tutor and complete lessons. It also finds that tutor access differs across devices. The team revises the claim to an observed association, investigates the access issue and plans a study capable of addressing the causal question.

Learner interviews reveal uses that the activity logs classify poorly, so the measurement scheme is refined before further analysis. The case illustrates how social understanding can change the data model and research design, rather than merely decorate a completed predictive analysis.

Common misconceptions and mistakes

Social data science means social-media analytics. Social-media data are only one possible source.
A very large dataset removes the need for social theory. Size does not explain what a variable means or resolve competing mechanisms.
Digital traces record behavior without shaping it. Interface choices and incentives influence which actions occur and are logged.
Accurate prediction proves a causal explanation. Predictive and causal claims require different evidence.

Business use

The approach is useful when a business decision concerns how people behave within a social and organisational setting. It can improve the interpretation of adoption, service use, communication and participation measures.

Ask teams to state the concept, observable measure, population and decision before presenting model accuracy. Include people who understand the setting and, where appropriate, those represented by the data. A smaller study with valid measurement may answer the business question better than a much larger collection of convenient but ambiguous traces.

Check your understanding

An organization counts internal messages and calls the result a collaboration score. What should be checked before interpreting a higher score as better collaboration?

Answer: Check whether message volume validly measures collaboration, how roles and tools affect recorded activity, whose work is missing and what independent evidence supports the interpretation. More messages can have several explanations.

Related terms

Sampling bias
Causal inference
Computational social science
measurement validity
research design
social network analysis
digital trace data

Learn next

Read measurement validity and causal inference. Choose a familiar workplace concept, list two possible indicators and explain what each misses. Then propose a complementary observation or interview question that could test your interpretation.

Causal inference

Sources

Oxford Internet Institute, Research Design for Social Data Science: https://www.oii.ox.ac.uk/study/courses/research-design-for-social-data-science/. Bernie Hogan, social measurement presentation description: https://www.oii.ox.ac.uk/news-events/videos/defining-social-data-science-as-the-science-of-social-measurement/. Matthew Salganik, Bit by Bit, observing-behaviour conclusion: https://www.bitbybitbook.com/en/1st-ed/observing-behavior/observing-conclusion/. The activity calculation and Willow Learning case are original teaching illustrations.

Previous: Social cognitive optimization Next: Social laws