Predict M/P ratio (Multiple regression analysis (Why: create a linear…
Predict M/P ratio
Multiple regression analysis
Why: create a linear regression model between one dependent variable and multiple independet variables. How: analyze; regression; linear
Check assumptions. all variables have an interval/ratio scale (categorical variables can be used as independent variable (dummies)), relationship is causal, model is linear, homogeneity of variances, normal distribution of residuals, no multicollinearity.
Check normal distribution. How: analyze; regression; linear; plots; normal probabilty plot
Residuals on/around diagonal
Residuals not on/around diagonal
No normal distribution
Check for multicollinearity. How: see corelate, univariate analysis
Correlation with |r| >= 0,9
Discard one of the correlating variables
No correlation with |r|>=0,9
Check linearity/homogeneity of variances. How: analyze; regression; linear; plots; scatter with y=ZRESID and x=ZPRED.
Residuals not equal around baseline
No obvious pattern of residuals
Residuals equal around baseline
Obvious pattern of residuals
Choose between standard or stepwise method
Look at R-square determination coefficient, variance analysis, coefficients for the multiple regression equation and variables entered/removed (in case of stepwise method)
Translate categorical variables into dummies (dichotomous variable with value 0 and 1. How: translate; recode into different variables.
For example R cubic spline function