Notes on Psychology Research Methods: Observation, Ethics, Correlation, and Experimental Design

Observing and Natural Behavior in Research

• People change their behavior when they know they are being observed (observer effect).
• Solutions to reduce this include:
  • unobtrusive/actualistic observation when feasible, though often interaction with participants is required.
  • selecting measures that are hard to fake (e.g., heart rate, blinking), though many research questions require participant interaction and self-report.
  • assuring confidentiality/anonymity to reduce demand characteristics:
  • name not linked to data; data collected anonymously or with coding; codes stored separately and destroyed at the end so only numbers remain.
  • repeated interaction with participants to habituate them to the study and reduce behavior shaped by the experimental setting (initial try-to-impress phase fades over time).
  • deception or placebo use to control for demand characteristics (see details below).
• Real-world analogy: first day of class vs two weeks later; people start behaving more naturally once they’re used to the setting.

Placebos and Demand Characteristics

• Placebos can help rule out demand characteristics when participants expect a treatment but receive something inert.
• Example: in an alcohol study, participants may think they drank alcohol when they received a non-alcoholic substitute with similar taste/smell.
• Deception can be used to prevent participants from guessing the true hypothesis and changing behavior accordingly, but it must be ethically justified.

Informed Consent and Deception

• Informed consent requires telling participants the general nature of the research and any risks involved.
• Specific hypotheses should not be disclosed because revealing them could bias behavior and contaminate results.
• Deception is discussed and approved only when there is no other viable method and the deception does not increase risk beyond minimal risk; participants are debriefed after the study.
• Institutional Review Board (IRB) approvals are required for all studies involving human participants; this applies to psychology and biomedical research funded by state or federal sources.
• Deception examples and constraints:
  • Example where the study is framed as memory-focused, but the true purpose is different; the deception is allowed only if no reasonable alternative exists and risks are minimal.
  • If deception would increase risk, it cannot be used; debriefing explains the true purpose after participation.

Deception in Practice: An Illustrative Scenario

• A classroom example where the study is framed about memory, but the setup actually tests a different variable.
• The decision to use deception hinges on the absence of viable alternatives and risk assessment by the IRB.
• Debriefing clarifies what was truly studied and why deception was necessary.

Correlation vs Causation: Core Concepts

• Many studies rely on survey data where participants respond to questions and numerical values are assigned to responses.
• Scatter plots illustrate the relationship between two variables: each dot represents a participant.
• Positive correlation: as scores on variable A increase, scores on variable B tend to increase.
• Negative correlation: as scores on variable A increase, scores on variable B tend to decrease.
• No correlation: random scatter around the origin; r ≈ 0.
• Strength of a correlation is given by the magnitude of the Pearson correlation coefficient r (0 ≤ |r| ≤ 1). The closer |r| is to 1, the stronger the relationship.
• Sign of r indicates direction (positive vs negative); magnitude indicates strength, not value judgement about good/bad.
• Key caution: correlation does not imply causation. Even strong correlations do not prove that one variable causes the other.
• Important questions to consider:
  • Directionality problem: does X cause Y or does Y cause X?
  • Third-variable problem: might a third factor Z cause both X and Y?
• Classic example: Alcohol consumption and GPA.
  • Possible interpretations include bidirectional relations or a third variable (e.g., stress) influencing both.
  • Without temporal sequencing or manipulation, causal inferences are limited.
• Another example: violent video games and aggression.
  • Correlation observed, but alternative explanations include preexisting aggression (reverse causation) or a third variable (parental supervision).
  • Partial correlations can statistically control for some third variables (e.g., parental supervision) to see if the X–Y association persists.
• Diet soda and diabetes example (illustrative, cautions about causality and confounds): associations may persist after controlling for obesity; causality cannot be established from correlation alone.

The Architecture of an Experiment: Establishing Causality

• To claim that X causes Y, you must conduct an experiment that manipulates X (the independent variable, IV) and measures Y (the dependent variable, DV), while holding all else constant (extraneous variables).
• Basic terms:
  • Independent Variable (IV): the manipulated variable with at least two levels/groups (e.g., violent vs nonviolent video game).
  • Dependent Variable (DV): the outcome measured (e.g., aggression, or a proxy like observed behavior).
  • Extraneous Variables: other factors that must be controlled or held constant.
• Example: Do violent video games cause aggression in children?
  • Sample: a convenience sample of children (random sampling is often impractical).
  • Random assignment: participants are randomly assigned to two groups to ensure equivalence on non-IV factors (e.g., age, IQ, SES, baseline aggression).
  • Levels of IV: two levels—violent vs nonviolent video games (could include more levels like low/medium/high violence).
  • Experimental groups: experimental (violent game) vs control (nonviolent game).
  • Operationalization: aggression must be defined and measured in observable terms (e.g., number of pushes, hits, screams, or other coded behaviors).
  • Random assignment rationale: ensures groups are comparable on everything except the IV, mitigating third-variable confounds like baseline aggression, parental supervision, or SES.
  • If a difference in DV emerges, causal attribution is strengthened, assuming the groups remained equivalent and there were no confounds.

Random Assignment and Group Equivalence

• Random assignment distributes individual differences (IQ, SES, personality) across groups, increasing the likelihood that groups are equivalent on these factors.
• Monte Carlo simulations suggest that with group sizes around n ≈ 30 per group, groups tend to be similar on most variables on average.
• Key principle: any observed difference in DV should be attributable to the IV if groups are equivalent on all other variables.
• Ethical and practical note: random assignment requires careful ethical consideration and IRB approval; not all studies can or should randomize participants to potentially harmful conditions.

Operational Definitions and Measurement

• Operational definitions specify exactly how a concept will be measured (e.g., aggression via observed behavior in a playground setting after gaming exposure).
• Clear operational definitions are essential for reliability and for replicability of findings.

Confounds, Extraneous Variables, and Experimental Integrity

• Confounds: variables that systematically differ between groups and could account for DV differences (e.g., room temperature differences during sessions).
  • If a confound affects only one group, causal claims are invalid.
  • If a confound affects both groups equally, it adds noise and reduces the ability to detect true effects but does not necessarily invalidate the causal claim.
• Extraneous variables: factors other than the IV that could influence the DV; these must be controlled or randomized to reduce their impact.
• Design integrity examples:
  • Air conditioning failure in one condition but not the other can introduce a confound if it differentially affects groups.
  • If failures occur equally across groups, the confounding effect is reduced, though it still adds variability to the data.
• Attrition (differential dropout): when participants drop out unevenly across groups, threatening equivalence and internal validity.
  • Example: dropout rates differ between therapies A and B, biasing results.
  • Preventative strategy: maintain similar dropout rates across groups; track reasons for dropout; use intent-to-treat analyses where appropriate.

Internal vs External Validity

• Internal validity: confidence that observed DV differences are caused by the IV and not by confounds; strengthened by random assignment, control of extraneous variables, and replication.
• External validity: generalizability of results beyond the study sample and context (e.g., different ages, cultures, or real-world settings).
• Researchers often trade off some external validity to maximize internal validity, then test generalizability via replication in varied contexts.

Real-World Relevance and Ethical/Philosophical Considerations

• Ethical considerations underline the design of human-subject research: informed consent, risk disclosure, confidentiality, debriefing, and minimization of harm.
• Philosophical takeaway: correlational data can point to relationships and help generate hypotheses, but a well-controlled experiment is needed to establish causality, at least for the specific context and variables studied.
• Practical takeaway: always consider the possibility of unmeasured third variables and the directionality problem when interpreting correlations; use experimental designs or longitudinal methods when causal conclusions are essential.

Quick Reference: Key Equations and Concepts

• Pearson correlation coefficient (linear relationship between X and Y):
  r = rac{ ext{Cov}(X,Y)}{\sigmaX \sigmaY} = rac{rac{1}{n-1}
  abla igl((xi-ar{x})(yi-ar{y})igr)}{rac{1}{n-1}
  abla igl(sX^2igr)^{1/2}rac{1}{n-1} abla igl(sY^2igr)^{1/2}}

  In simplified sample form:
  r = rac{ extstyle rac{ ext{sum}{i=1}^n (xi - ar{x})(yi - ar{y})}{n-1}}{ extstyle igl[sX^2igr]^{1/2}igl[s_Y^2igr]^{1/2}}

  • Range: $$r \u2264 1 ext{ and } r d -1

• Interpreting r:

  • Magnitude (|r|) indicates strength; closer to 1 means stronger association.
  • Sign indicates direction: positive or negative correlation.
  • Note: the magnitude does not imply causation.

• Experimental design primitives:

  • IV with at least two levels (e.g., violent vs nonviolent game).
  • DV to be measured.
  • Random assignment to groups to control for third variables.
  • Operational definitions for all measured behaviors.
  • Debriefing after the study when deception is used.

Connections to Earlier Lectures and Real-World Implications

• The discussion links observer effects, confidentiality, placebo use, and deception to experimental control and ethical practice.
• It connects basic statistical concepts (correlation, r) to methodological questions about causality, directionality, and third-variable threats.
• The material emphasizes how careful design (randomization, control, operational definitions) underpins credible claims about cause and effect in psychology and related fields.

Key Takeaways

• Observing people alters behavior; use strategies to minimize this bias, including confidentiality and habituation.
• Placebos and deception can be ethical tools to control for demand characteristics, but require IRB approval and debriefing; informed consent governs what participants are told upfront.
• Correlation reveals relationships but does not establish causation due to directionality and third-variable problems.
• Experiments manipulate the IV, measure the DV, and hold extraneous variables constant to establish causality; random assignment is essential to create equivalent groups.
• Operational definitions and careful control of confounds are needed for reliable and valid conclusions; consider both internal and external validity and watch for attrition.
• Real-world data often require replication and cross-context testing to build robust causal inferences.