Correlation vs. Experiment: Key Concepts and the Correlation Coefficient

Correlation vs Experiments

• Core idea: a score on variable A is used to predict a score on variable B. The goal is to understand what factors can help predict B from A.

• Example scenario from transcript:

  • Hypothesis: engaging in more social media usage (A) will lead to recognizing more celebrities (B).

  • Prediction: knowing how much someone uses social media should allow us to predict how many celebrities they can recognize on a test.

• Major difference between correlation designs and experiments:

  • Experiments focus on causation.

  • Correlation designs focus on prediction.

• In an experiment:

  • The independent variable (IV) is manipulated.

  • The dependent variable (DV) is measured.

  • There is usually random assignment and groups for comparison.

• In a correlational design:

  • No manipulation occurs.

  • Both variables are measured, not controlled by the experimenter.

  • There are no groups or random assignments; each participant provides data for both variables.

• Measurement example in the transcript:

  • Variable A (social media usage): measured by hours per day used.

  • Variable B (celebrities recognized): measured by a test of celebrity faces.

  • Both variables are obtained from the participants; neither is controlled by the experimenter.

• Predictive hypothesis in correlation:

  • Hypothesize that Variable A provides an indication or prediction of Variable B.

  • After measurement, test whether A predicts B via a correlational analysis.

• Testing in correlational design differs from testing in an experiment:

  • In correlation, there are no groups to compare (no experimental vs control groups).

  • The test focuses on the relationship between A and B rather than group differences.

• Statistical procedure used: correlation analysis.

  • Output is the correlation coefficient, denoted by $r$.

  • Range: $r \,\in\, [-1, \; 1]$.

• What the correlation coefficient tells us (two main aspects):

  • Strength of the relationship: how well A predicts B. The farther from zero, the stronger the relationship.

  • Direction of the relationship: whether the association is positive or negative.

• Interpretation of strength relative to zero:

  • The stronger the correlation, the closer the data points lie to a line of best fit.

  • The closer data points cluster to a straight line, the stronger the prediction.

• Interpretation of the range:

  • Strongest possible correlations: $r = -1\text{ or }r = 1$.

  • If $r = 1$, A perfectly predicts B with a positive relationship.

  • If $r = -1$, A perfectly predicts B with a negative relationship.

  • Zero correlation: $r = 0$, indicating no relationship between A and B.

• Practical note about typical data:

  • In psychology, correlations close to -1, 0, or 1 are uncommon; real data often show weaker, intermediate correlations.

• Graphical interpretation (scatter plots):

  • A strong correlation yields data points that lie close to the line of best fit.

  • A weaker correlation yields data points that spread more around the line.

  • In the examples given, a positive correlation is shown when increases in A correspond to increases in B.

• Direction of relationship details:

  • Positive correlation: As Variable A increases, Variable B increases as well.

  • Negative correlation: As Variable A increases, Variable B decreases.

• Examples from the transcript:

  • Positive correlation example: more social media usage (A) associated with recognizing more celebrities (B).

  • Negative correlation example (hypothetical in the transcript): more social media usage (A) associated with recognizing fewer celebrities (B).

• Summary of the approach:

  • In correlational studies, we measure A and B for each participant.

  • We test whether A predicts B using the correlation coefficient $r.$

  • The results inform us about prediction strength and direction, not causation.

• Connecting to broader principles:

  • Correlation designs are used for prediction and association testing before or alongside causal investigations.

  • They rely on observational data rather than manipulated conditions.

• Key formulas to remember:

  • Pearson correlation coefficient:
    $r = \frac{\mathrm{cov}(X,Y)}{\sigma<em>X \sigma</em>Y}$

  • Interpretation guide:

  • If $|r| \approx 1$, strong predictive relationship.

  • If $|r| \approx 0$, little to no predictive relationship.

• Terminology recap:

  • Variable A: predictor/independent variable in the context of the correlation (though not manipulated).

  • Variable B: outcome/dependent variable in the context of the correlation (though not manipulated).

  • Line of best fit: the best straight-line approximation through the data points in a scatter plot.

• Final takeaway:

  • Correlation helps us understand whether there is a predictable relationship between two measured variables and in which direction that relationship goes, but it does not establish causation or imply that changing A will cause B to change.