Midterm 2 - Chapt. 12 CONTINUED (From Nov 5th Lecture) Flashcards
REVIEW
What does r² represent?
Amount of SHARED VARIANCE in the DV that’s predictable from the IV
Example: A simple linear regression that predicts students’ exam scores (dependent variable) from their study time (independent variable) has an R² of . 71. From this R² value, we know that: 71% of the variance in students’ exam scores is predicted by their study time.
What is Multiple Regression?
Extension of correlation
* Both measure relationships among variables
* Correlation does not imply causation
What’s different about regression?
- Uses language of prediction
- Can use 1 or more variables to predict changes in another variable (still not causation!)
How do regressions model predictive relationships
- Use score on one variable (“the Predictor”) to predict changes in another variable (“the Criterion”)
- Should use when we have a good reason that one of the variables could change the other (e.g., temporal precedence)
When are regression/correlation methods used?
- Instead of experiments (typically)
- When you can’t manipulate variables, can only measure them
E.g., ethical or practical problems - Common in personality, health, & longitudinal/ developmental research
What does b represent?
Y = a + bX
- Slope - Rise over Run
What does a represent?
Y = a + bX
Intercept - when predictor (x) = 0, what’s the value of criterion (y)?
What do Y and X represent?
Y = a + bX
- Y = criterion
- X = predictor
Benefits of Regression Framework
Models predictive relationships
* When most appropriate - ie if you have temporal precedence, when you can’t use an experiment
Can investigate the effect of multiple predictors on the criterion at the same time.
* Ie Multiple Predictor Regression
Multiple Regression Enables…
…investigation of how well many predictors simultaneously predict the criterion
- What’s the unique contribution of each predictor to the prediction of criterion?
- = what part of X1 overlaps with Y, which is UNIQUE and differentiates from the others
Features of Multiple Correlation (R
- Can have any number of predictors
- Need to collect data on each predictor
- Calculate the unique contributions of each predictor individually on predicting criterion (b’s)
- Calculate the contribution of all predictors combined for predicting criterion
Called the Multiple Correlation (R) but discussed as R 2
R2 is the proportion of variance in the criterion that can
be explained by all predictors combined
Partial Correlation & the 3rd Variable Problem
- Statistically “control for” 3 rd variable
- Try to remove the effects of a variable we know likely influences both variables of interest
- EX: how much of the relationship between depression and anxiety is explained by anxiety? Once calculated, we can removed this aspect (the third variable)
- Correlation can decrease due to this removal - but STILL EXIST
Measures of Relationship (R²)
- Correlation (r): Standardized index of how much two variables relate to each other
- Squared Correlation (r²): Proportion of variance shared by 2 variables
- Multiple Regression: A technique used when we want to test the unique contribution of one or more variables in predicting the criterion Y.
- Multiple Correlation (R/R²): A type of correlation coefficient that indexes how much a set of variables, when combined, is related to the criterion Y.
- Used with multiple regression and expressed as R 2 - Partial Correlation: A correlation between X and Y that statistically removes the effect of Z
What is the difference between multiple correlation and multiple regression?
Multiple Regression
* How much does each predictor uniquely contribute to explaining scores on the criterion?
* Calculation for each predictor separately
Multiple Correlation (R) – discussed as R2
* How much do all of the predictors combined explain scores on the criterion?
* One calculation for all of the predictors combined
* Can square R to identify the proportion of variance in the criterion that can be explained by all predictors combined.
What is the difference between multiple correlation and partial correlation?
Multiple Correlation
How much do all of the predictors combined explain
scores on the criterion?
One calculation for all of the predictors combined
Partial Correlation
How much is variable X related to variable Y, with the
effect of variable Z removed?
How much is variable X related to variable Y,
controlling for the effect of variable Z?