Lecture 7- Correlation Flashcards
What is a correlation?
Relationship between DVs
How do we represent correlations?
Through scatterplots we can look at the direction and the strength of the relationship
What do correlations not show?
They are nothing to do with IV, therefore they can only be used to describe and predict not explain or infer causation.
What is the most valuable thing about making a correlation?
Can predict
What is a positive correlation?
As one variable gets bigger, the other variable gets bigger
What is a negative correlation?
As one variable gets bigger, the other variable gets smaller
What will be shown on a scatterplot if there is no correlation?
Points scattered everywhere (no clear consistent trend)
How does a strong correlation vary to a weak correlation? What does this allow?
Points are closely clustered around the trend line (fewer outliers)
The stronger the correlation the better the predictability.
What is the method used for computing a correlation?
Pearson’s R coefficient
What does the range of possible R values mean?
-1= perfect negative correlation 0= no correlation \+1= perfect positive correlation
What are some constraints to r?
- The variables to be correlated must be measured on the same individuals
- Variables must be measured on a ratio or interval scale
- r can detect only linear relationships
What are the possible explanations for r=0?
- No relationship
- The relationship is non-linear (curvilinear)
- There is a restricted range so simply can’t see the correlation
IQ scores predict primary school grades well
SAT scores reflect university grades not too well
GRE scores do a poor job at predicting postgrad performance
What does this reflect?
The range is getting more restricted so not picking up a correlation
What is the aim of doing a cross-lagged-panel correlation procedure? What is a problem with it?
To solve a directionality problem (don’t know whether it is x causing y or the other way around). Problems of interpretation still remain (3rd variable problem).
How do you do a crossed-panel correlation study?
Obtain several correlations over time. Then look at the size and direction of the correlation coefficients to determine what leads to what. Basically investigate every single relationship possible (cross between the studies)