Model Performance

Question 1

Coefficient of Determination

Answer 1

Also known as r-squared (w/ multiple explanatory variables), this measures how well a linear regression line fits the observed values, a larger number indicates a greater fit. More formally, it is defined as the proportion of variance in the response variable that is predictable from the explanatory variable(s). 0 to 1 measure

Question 2

Residual standard error (RSE)

Answer 2

measures the typical size of the residuals; smaller is better

Measures the difference between predicted and observed response variable; this is always measured in the units of the response variable; for example, an RSE of 20 would mean the typical difference b/w predicted and observed response variable is 20 units

Not sure why this is called error. Found disagreement when looking it up and it isn’t always referred to as such. It really should be called mean error of the residuals.

Question 3

Degrees of Freedom

Answer 3

An estimate of the number of independent pieces of information that went into calculating an estimate; how many calculations have the freedom to vary

Question 4

Root-mean-square error (RMSE)

Answer 4

Close to residual squared error (RSE) but doesn’t take into account the # of coefficients

Worse than the RSE for comparisons b/w models; should usually just use RSE

Question 5

Residual

Answer 5

Difference between an observed value and the mean value that a model predicts for that observation