notes on correlation.
Lecture Supplement on Correlation
Overview of Correlation
This lecture serves as a replacement for the missed class focused on correlation.
Assumes familiarity among psychology students with correlation concepts.
Recap of Previous Topics
Recent classes covered analysis of experience.
Discussed logic behind ANOVA and calculations involved.
Reviewed post hoc tests following significant F tests to identify which groups differ.
Correlation as a New Concept
Correlation distinguishes itself from previous analyses, focusing on the strength of association between two quantitative variables which can be on an interval or ratio scale.
Key Characteristics:
Deviates from t-tests and ANOVA which analyze categorical predictor variables against quantitative outcome variables.
Understanding Correlation
Definition of Correlation
The term "correlation" combines Latin roots:
Cor or Co: Meaning together.
Relatio: Meaning relation.
A correlation quantifies the degree of relationship between two variables (bivariate correlation).
Focused on two quantitative variables at a time.
Types of Correlation
Bivariate Correlations: Relationships between two variables.
Multiple Correlations: Involving more than two variables, not discussed in detail here.
Example Correlations
Children’s weight and height.
Generally, taller children are expected to weigh more, but it doesn't guarantee it.
SAT scores and college freshman GPA.
SAT scores predict freshman GPA, helping forecast college retention.
Correlation vs. Causation
The Importance of Distinction
Causation is not implied by correlation.
Concurrent relationships do not establish causal links.
Experimental designs and testing are necessary for causal conclusions.
Reasons for Misconceptions:
Spurious Correlations: Illusory correlations that occur by chance. Example: Ice cream sales correlate with crime rates, both driven by temperature.
Causality Limitations:
Many variables in psychology cannot be manipulated ethically or practically.
Examples of Spurious Correlations:
Age of Miss America correlating with murders by steam and hot objects.
Cheese consumption correlating with deaths through bed sheet tangling.
Interpreting Correlation
Range of Correlation Values
Values range from -1 to +1.
Notations:
Lowercase italicized r for sample correlations.
Greek letter ρ (rho) for population parameter correlations.
Pearson correlation: Most common type, measuring linear relationships.
Sensitive to whether as one variable increases, the other does too (positive correlation) or decreases (negative correlation).
Zero Correlation: Indicates no linear relationship; non-linear relationships (curvilinear) can exist.
Curvilinear Relationships:
Not detected by Pearson correlation; relationships often exist outside linear models.
Covariance and its Limitations:
Covariance measures relationship strength but is scale-dependent, making interpretation difficult.
Calculation involves:
Differences between individual scores and means of each variable, multiplied and summed up, then divided by sample size minus one to form covariance value.
Calculating Correlation from Covariance:
Correlation is derived by dividing covariance by the product of standard deviations of both variables.
Produces a value between -1 and 1, indicating strength and direction of correlation.
Practical Application of Correlation
Calculating Correlation in Statistical Software (e.g., SPSS):
Use the correlation matrix to analyze relationships between variables (e.g., trait anxiety pre and post treatment).
Null Hypothesis: $
ho = 0$, suggesting no correlation in the population.Alternative Hypothesis: $
ho \neq 0$, suggesting a correlation exists.
Significance Levels:
Example: A correlation statistically significant at $p < 0.05$ leads to rejecting the null hypothesis.
Credible intervals indicate the range of expected values for population correlation; intervals not containing zero point to significant results.
Visualizing Correlations
Scatter Plots:
Effective for visual correlation understanding, with each dot representing paired scores across variables.
Higher correlation strength produces tighter clusters along the trend line, whether positive or negative.
Effect Size Considerations in Correlation
Understanding Effect Sizes:
Correlations serve as effect sizes, indicating the strength of relationships in a more practical manner.
Guidelines for interpretation of correlation strength:
< 0.1 = very small
0.1 - 0.29 = small
0.3 - 0.49 = moderate
> 0.5 = strong
Conclusion and Future Topics
Closing emphasis on credible intervals in correlation interpretation.
Understanding statistical significance is crucial for research outcomes.
Next class will address bivariate regression as an extension of correlation concepts.