Week 10 - Multiple Linear Regression Flashcards
Bivariate statistics
Relationship between 2+ variables
Correlation coefficient
Degree of association, scatterplot
Pearson’s R
Linear, 2 scale variables, sample size 40+
Spearman’s Rho
1 variable must be ordinal, sample size <40
Restriction of range effect
May find correlation due to limited range of 1 IV
R-squared
How much of a score increasing is explained by other variable. Correlation-coefficient squared!
Mahalanobis distance
Remove outliers
Cook’s distance
Remove outliers for regressions!
Partial correlation
Remove variance from 3rd variable
Regression
Predicts values of Y based on X
Least squares
Finds line that minimises squared distances between data and line
MLR
If difference in r-squared values of simple linear and MLR is ‘big’ and p-value is ‘small’ them MLR is worth it!
Dummy variables
Use if predictor is nominal eg. Welly gets 1 for Welly but 0 for CHCH.
Hierarchical method (MLR)
Predictors and order are based on prior knowledge - you decide order
Stepwise method (MLR)
Order based on math criteria
- forward: put small predictors first
- backward: includes all variables then excludes as necessary (most common)
step-wise: used for new models/explanatory studies (no prior research)
