- capture the effect of a binary independent variables
- use one less dummy variable than the number of categories
- required methods other than ordianry least squares (e.g. probit, logit (using max. likelihood), or discriminant analysis (using a linear function similar to ordinary least squares))
Conditional Heteroskedasticity
What is it? Residual variance related to the level of independent variable
What it does? Too many Type I errors
How to detect? Breusch-Pagan Chi Square Test BP = n x R^{2}_{resid}
How to Fix: Used White-corrected (robust) standard errors
Serial Correlation
Can be positive or negative (+ive S.C. is more common)
What is it? Residuals are correlated
What it does? Too many Type I errors (too many type II errors for -ive S.C.)
How to detect? Durbin Watson Test = 2(1-r)
How to Fix: Use Hansen method to adjust standard errors
If C.H. and +ive S.C. are present together, use Hansen method to adjust standard errors
Multicollinearity
What is it? Two or more independent variables are correlated
What it does? Too many Type II errors
How to detect? t-Test implies insignificance of individual variables but F-Test implies significance
How to Fix: remove one of the correlated variables