Notes on Experimental Design, Operational Definitions, and Correlation in Psychology

Experimental Design: Key Concepts

  • Experiments in psychology manipulate the independent variable (IV) and measure the dependent variable (DV) while controlling other variables as much as possible.
  • A study is an experiment when it:
    • Manipulates the IV,
    • Measures the DV,
    • Controls extraneous variables (random assignment is the hallmark method of control).
  • Correlational designs examine relationships between variables without manipulation; they cannot establish causation due to directionality and third-variable problems.
  • Mixed designs combine manipulated variables with non-manipulated variables (e.g., a treatment amount plus a participant characteristic like alcoholism status).
  • Operational definitions specify exactly how you will measure your variables (e.g., how you quantify health, happiness, aggression).
  • Random assignment ensures groups are approximately equivalent at baseline, so differences in DV can be attributed to the IV rather than preexisting differences.
  • Control conditions (e.g., wait-list, placebo) help isolate the effect of the IV and account for expectancy effects.
  • Ethical considerations are essential: do no harm, IRB oversight, deception considerations, and ensuring participants are not harmed or unduly distressed.

Independent and Dependent Variables; Operational Definitions

  • Independent variable (IV): what the experimenter manipulates. Examples:
    • AZT dosage levels in an HIV treatment study:
    • Dosages could include a zero-dose (control), 25 mg, 50 mg, 75 mg, etc. The exact number of conditions can vary (two to ten is a practical range).
    • Amount of alcohol: zero, placebo, 2 beers, 4 beers, 6 beers, etc.
    • Amount of television violence: high violence vs. low violence.
    • Presence of blood in a bystander scenario (blood visible vs not visible) as part of a helping study.
    • Number of phone calls in the foot-in-the-door study: one vs two calls (and control with no first call).
  • Dependent variable (DV): what you measure to assess the effect of the IV. Examples:
    • HIV health outcome: viral load, or white blood cell count; sometimes CD4 count.
    • Health-related outcome: general health indicators (operationalized via tests or clinical metrics).
    • Happiness: score on a happiness inventory (self-report scale).
    • Aggressiveness: count of aggressive acts observed during free play (e.g., punches, slaps, kicks, etc.) via a coded checklist.
    • Dexterity/coordination: performance measures (e.g., dexterity scores).
    • Helping behavior: time to help or whether help occurs; proportion of participants who help.
  • Operational definitions (examples):
    • Viral load: measured via blood tests (primary DV for AZT study).
    • Happiness: score on the Happiness Inventory (self-report scale).
    • Aggressive acts: total number of aggressive gestures counted by trained raters using a predefined checklist.
    • Dexterity: a standardized dexterity score from a coordination test.
    • Helping: whether the subject intervenes or the time to intervene in an emergency scenario.

Experimental Examples: Design, IVs, DVs, and Controls

  • AZT and HIV treatment study

    • IV: treatment level (dosage) of AZT; possible conditions include 0 mg (control/wait-list), 25 mg, 50 mg, 75 mg (example), with up to 10 levels considered in some designs.
    • DV: health outcome measured by viral load (primary) and/or white blood cell count; may also reference CD4 counts.
    • Control: random assignment to conditions; placebo group (placebo pill) to control for expectancy effects; wait-list control where participants know they are in the study but do not receive treatment immediately.
    • Notes: random assignment helps ensure equivalence across groups; ethical considerations include minimizing harm and using blinded assessments where possible.

  • Alcohol and happiness study

    • IV: amount of alcohol consumed (including zero, placebo, 2, 4, 6 beers etc.).
    • DV: happiness score from an inventory (self-report scale).
    • Control/experimental nuances: include a placebo condition (non-alcoholic beverage presented as beer) to control expectancy; zero-alcohol condition as a baseline control; mixed designs may compare alcoholics vs. non-alcoholics (non-randomly assigned group status).
    • Key idea: if happiness scores increase with alcohol dose under controlled conditions, one can claim an effect of alcohol on happiness within the design limits.

  • Watching television violence and aggressiveness

    • IV: amount of television violence (high vs. low) – two conditions.
    • DV: observed aggressive acts during a 30-minute free-play period; scored via a checklist by raters.
    • Subjects: approximately N = 100 children, with roughly 50 in each condition.
    • Control: random assignment to high vs low violence groups; age-matching or keeping age ranges similar to minimize age effects.
    • Important point: despite finding a relationship, you cannot claim causation without proper control; random assignment and other controls are necessary to support causal claims.

  • Bystander/Helping behavior experiments (classic social psychology studies)

    • Core idea: several studies examined how likely people are to help under various conditions. Common elements:
    • Use of staged emergencies (e.g., a worker in distress, a confederate acting injured).
    • IVs often include group size (alone vs in groups), presence of danger (e.g., holding live wires vs not), and context (e.g., whether the event is obvious as an emergency).
    • DV: helping behavior (whether help is offered, how quickly, and how forcefully).
    • Example with electric shock scenario:
    • IVs: whether the subject is alone vs in a group; whether the “electrician” is holding live wires vs not.
    • DV: time to offer help, or whether help is offered (and perhaps unintended consequences like attempting to intervene with a broom).
    • Notes: ethical concerns led to modifications (e.g., staged, no actual harm; debriefing; IRB oversight). The researcher Bib Latane’s work is cited as a foundational but ethically strict approach; later adjustments were made to avoid harm.

  • Foot-in-the-door technique study (social influence)

    • IV: number of phone calls (one or two calls) or presence of initial contact.
    • DV: compliance rate (e.g., willingness to allow strangers into the house to search cabinets).
    • Result: subjects who received the first call were more likely to comply after a second call; those who received no initial contact complied at zero rate for the second phase in the described scenario.

  • Ethics and IRB considerations discussed in these examples

    • Do no harm: minimize potential distress; ensure participants are debriefed.
    • Deception and staged scenarios: often used historically; modern practice requires IRB review and ethical justification.
    • In some classic studies (e.g., bystander experiments), participants were not fully aware they were part of a study, raising ethical concerns; contemporary standards require informed consent or thorough debriefing to mitigate potential harm.

Correlational Designs and the Correlation Coefficient

  • Correlational design purpose: determine whether a relationship exists between two variables without manipulation.
  • Correlation coefficient r:
    • Range: −1 ≤ r ≤ 1-1 \, \leq \, r \, \leq \, 1
    • Direction: sign indicates direction of the relationship (positive vs. negative).
    • Strength: absolute value |r| indicates strength; larger |r| means stronger association; sign does not affect strength.
    • Example: ACT scores and college GPA may be positively correlated, e.g., r=0.42r = 0.42, indicating a moderate positive relationship.
    • Interpretation nuances: correlation does not imply causation due to two main problems:
    • Directionality problem: the direction of causality cannot be determined from correlation alone (e.g., does TV violence cause aggression, or are more aggressive individuals drawn to violent TV?).
    • Third-variable problem: a third variable may influence both variables, creating a spurious relationship (e.g., teacher quality, family environment, or socioeconomic status).
  • Predictive use of correlation
    • Correlation can be used to predict outcomes, but prediction quality depends on the strength of the relationship.
    • Regression line concept: a line that minimizes the squared differences between observed values and the line (least squares).
    • Regression equation example: y^=β<em>0+β</em>1x\hat{y} = \beta<em>0 + \beta</em>1 x where y^\hat{y} is the predicted DV value from the IV value xx.
  • Visual representation: scatter plots show individual data points; a regression line illustrates the overall trend; the tighter the points cluster around the line, the stronger the relationship.
  • Practical example in context: a scatter plot of years of experience (x) vs. pay (y) shows a positive trend; regression line can illustrate direction and strength of the relationship.

Key Differences: Causality, Directionality, and Third Variables

  • Why correlations do not prove causation:
    • Directionality problem: the direction of influence between two variables is ambiguous in correlational data.
    • Third-variable problem: a separate variable may cause changes in both variables, creating a noncausal association.
  • When can we infer causality?
    • In well-controlled experiments with random assignment, manipulation of the IV and control of extraneous variables, causality claims are more warranted.
  • How to handle mixed designs?
    • Combine manipulated IVs with non-manipulated factors (e.g., alcoholism status) to study interactions, while recognizing that such designs may not be fully classical experiments and can require specialized statistical analysis.

Practice Focus: Experimental Rigor and Interpretation

  • Remember the three core features of an experiment:
    • A manipulated independent variable,
    • A measured dependent variable,
    • Random assignment and/or tight control of extraneous variables (to enable causal inference).
  • Always consider potential confounds and extraneous variables that could threaten internal validity;
    • Examples: participant characteristics (age, gender), situational differences, experimenter effects, demand characteristics.
  • In correlational analyses, be prepared to discuss:
    • The directionality problem, the third-variable problem,
    • The meaning of the correlation coefficient and what it can and cannot tell you about future predictions.
  • Ethical considerations to mention:
    • Do no harm, informed consent, debriefing, minimization of deception, safeguarding participant well-being, and IRB involvement.

Quick Reference: Summary of Notable Points from the Transcript

  • IVs and DVs were illustrated via concrete cases: AZT dosage, alcohol amount, violence exposure, helping behaviors, and foot-in-the-door techniques.
  • Control methods emphasized: zero-dose control, placebo, wait-list control, and random assignment to reduce confounds.
  • Operational definitions were underscored as essential for objective measurement (e.g., viral load, happiness inventory scores, counts of aggressive acts).
  • The two main barriers to causal claims in nonexperimental designs were highlighted: directionality and third-variable problems.
  • The ethical dimension of social psychology experiments was stressed, including the use of deception and the responsibility to minimize harm.
  • The discussion included how to interpret and visualize data: scatter plots, regression lines, and the meaning of r values in terms of strength and direction.

Notation and Formulas (recap)

  • Range of correlation coefficient: −1≤r≤1-1 \le r \le 1
  • Example correlation: r=0.42r = 0.42 (moderate positive)
  • Significance indicator example: p < 0.05
  • Regression line representation: y^=β<em>0+β</em>1x\hat{y} = \beta<em>0 + \beta</em>1 x
  • Sample size examples from scenarios: N=100,n<em>1=n</em>2=50N=100,\quad n<em>1=n</em>2=50
  • Conceptual cue: a regression line minimizes squared residuals (least squares).