Comprehensive notes on the Scientific Method and Research Methods (Sociology)

Method as a rule structure: a framework of rules, penalties, consequences, and rewards that guide behavior across activities (sports, academics, research).
Children naturally create methods during play: rule structure, penalties, bonuses, rewards; brains are wired to seek order, preventing chaos.
Core takeaway: in any investigation, the scientific method provides a structured way to study and understand phenomena.

Definition: a procedural framework for conducting research that yields reliable, empirical results.
If you follow the scientific method, you reduce issues with methodology and improve the credibility of findings; skipping steps risks worthless data.
Core idea: we observe the world to form questions, hypotheses, and theories, then test predictions derived from those theories.

Observation: start with noticing something interesting, odd, or unusual.
Theory: a systematic, generalized model of how some aspect of the world works (a reasoned explanation based on observations).
Hypothesis: a testable prediction derived from the theory; not interchangeable with theory.
Null hypothesis (H0): the default position that there is no effect or relationship; testing proceeds with the assumption that H0 is true until evidence disproves it.
Process: observation → theory → hypothesis → test/experiment → data → revision or support for the theory.
Important caveat: real-world research involves many variables; revisions are almost always necessary.

The phrase “the science is settled” is misleading; science is an ongoing, iterative process.
Example: classical physics’ law of non-occupancy of the same space by matter was challenged by quantum mechanics; knowledge evolves with new evidence.
This caveat is especially important in social sciences where policies and programs are based on best available evidence, not final truth.

Theory: a generalized model of how some aspect of the world works; explains a set of observations and makes testable predictions.
The goal is grounding empirical work in a theoretical framework so findings are interpretable and scalable.

Quantitative: numerical data, statistical analysis; often used for hard statistics and large samples.
Qualitative: non-numeric data (e.g., interviews, transcripts) focusing on meanings, themes, and context; often requires coding and thematic analysis.
Mixed methods: a combination of qualitative and quantitative approaches; converts qualitative data into quantitative measures when needed.
Context: the chosen method depends on the research question and the kind of data available.

Longitudinal study (quantitative example): ADD longitudinal health study
- Description: longest-running longitudinal study with data collected over time.
- Size and scope: roughly N = 6500 respondents; about V = 3000 variables per respondent; duration > 40 years.
- Data handling: analyzed with software like Stata or SPSS to identify trends.
Qualitative example: thematic reviews and code books
- Process: conduct interviews, transcribe, develop a code book, assign themes, analyze qualitative data for connotations (positive/negative).
- Challenge: qualitative analysis is time-consuming and requires careful interpretation.
Mixed methods example: combine qualitative themes with quantitative data from large datasets to create measurable indicators and compare trends.

Quantitative data and analysis
- Data type: numerical, statistical, empirical data.
- Typical usage: measuring, testing, and modeling with numerical fields and variables.
Qualitative data and analysis
- Data type: non-numeric, textual or narrative data.
- Typical usage: extracting themes, patterns, and meanings; requires coding and thematic development.
Mixed methods
- Rationale: leverage strengths of both approaches; quantify qualitative insights or qualify quantitative results.
- Example workflow: conduct qualitative interviews, code and quantify responses, then merge with quantitative indicators from a large dataset.

Correlation: two variables co-vary (positive or negative) without implying causation.
- Example: ice cream sales and drownings both rise in summer (positive correlation), but one does not cause the other.
- Third factor often drives both: Summer/heat increases swimming and ice cream consumption.
Causality: a change in one factor (the cause) leads to a change in another (the effect); requires time order and a demonstrated mechanism.
- Time order: the cause must occur before the effect.
- Example structure: X → Y, with X preceding Y in time.
- Important: correlation alone does not prove causation; there must be a plausible mechanism and ruling out alternative explanations.
Negative correlation example: higher education → lower likelihood of criminal activity (not deterministic, but a trend).
Complex causal diagrams: arrows denote causal influence; third factors (confounders) can influence both X and Y.

Time order: establishing that X precedes Y is essential for arguing causality.
Reverse causality: sometimes researchers think X causes Y, but Y may cause X; must test directions over time.
Ruling out alternatives: identify and test other plausible explanations (confounders) to isolate the true causal pathway.
Practical guidance: begin with correlation, test time order, and consider multiple potential causal pathways; if a single factor remains the most plausible cause, you may infer causality with caution.

Natural experiments: events in the real world that affect groups in a way that approximates random assignment to treatment/control, useful when randomized experiments aren’t feasible.
Treatment and control groups: used to compare outcomes where one group receives an intervention and the other does not.
Placebo effect: participants’ belief they receive a treatment can influence outcomes, even if the treatment is inert (e.g., sugar pill).
Relevance: helps distinguish genuine treatment effects from expectations or perceptual biases.

Independent variable (IV): the hypothesized cause; can have multiple IVs in a study.
Dependent variable (DV): the outcome; typically only one DV in standard regression models.
Relationship: DV = f(IV1, IV2, …, IVk) + error, represented in regression as
$$Y = eta0 + eta1 X1 + eta2 X_2 + \